| 1 | #define TORCH_ASSERT_ONLY_METHOD_OPERATORS |
|---|---|
| 2 | // @generated by torchgen/gen.py from RegisterFunctionalization.cpp |
| 3 | |
| 4 | #include <ATen/core/LegacyTypeDispatch.h> |
| 5 | #include <ATen/EmptyTensor.h> |
| 6 | #include <ATen/FunctionalTensorWrapper.h> |
| 7 | #include <ATen/FunctionalInverses.h> |
| 8 | #include <torch/library.h> |
| 9 | |
| 10 | #ifndef AT_PER_OPERATOR_HEADERS |
| 11 | #include <ATen/Operators.h> |
| 12 | #include <ATen/NativeFunctions.h> |
| 13 | #else |
| 14 | // needed for the meta tensor calls to get stride info in functionalization |
| 15 | #include <ATen/ops/empty_strided_native.h> |
| 16 | // needed for special handling of copy_(). |
| 17 | // See Note [functionalizating copy_() and not preserving strides] |
| 18 | #include <ATen/ops/to_ops.h> |
| 19 | #include <ATen/ops/expand_copy_ops.h> |
| 20 | |
| 21 | #include <ATen/ops/_cudnn_rnn_flatten_weight_native.h> |
| 22 | #include <ATen/ops/_cudnn_rnn_flatten_weight_ops.h> |
| 23 | #include <ATen/ops/_cudnn_rnn_flatten_weight_native.h> |
| 24 | #include <ATen/ops/_cudnn_rnn_flatten_weight_ops.h> |
| 25 | #include <ATen/ops/abs_native.h> |
| 26 | #include <ATen/ops/abs_ops.h> |
| 27 | #include <ATen/ops/abs_native.h> |
| 28 | #include <ATen/ops/abs_ops.h> |
| 29 | #include <ATen/ops/abs_native.h> |
| 30 | #include <ATen/ops/abs_ops.h> |
| 31 | #include <ATen/ops/absolute_native.h> |
| 32 | #include <ATen/ops/absolute_ops.h> |
| 33 | #include <ATen/ops/absolute_native.h> |
| 34 | #include <ATen/ops/absolute_ops.h> |
| 35 | #include <ATen/ops/absolute_native.h> |
| 36 | #include <ATen/ops/absolute_ops.h> |
| 37 | #include <ATen/ops/_conj_physical_native.h> |
| 38 | #include <ATen/ops/_conj_physical_ops.h> |
| 39 | #include <ATen/ops/_conj_physical_native.h> |
| 40 | #include <ATen/ops/_conj_physical_ops.h> |
| 41 | #include <ATen/ops/acos_native.h> |
| 42 | #include <ATen/ops/acos_ops.h> |
| 43 | #include <ATen/ops/acos_native.h> |
| 44 | #include <ATen/ops/acos_ops.h> |
| 45 | #include <ATen/ops/acos_native.h> |
| 46 | #include <ATen/ops/acos_ops.h> |
| 47 | #include <ATen/ops/arccos_native.h> |
| 48 | #include <ATen/ops/arccos_ops.h> |
| 49 | #include <ATen/ops/arccos_native.h> |
| 50 | #include <ATen/ops/arccos_ops.h> |
| 51 | #include <ATen/ops/arccos_native.h> |
| 52 | #include <ATen/ops/arccos_ops.h> |
| 53 | #include <ATen/ops/affine_grid_generator_native.h> |
| 54 | #include <ATen/ops/affine_grid_generator_ops.h> |
| 55 | #include <ATen/ops/affine_grid_generator_native.h> |
| 56 | #include <ATen/ops/affine_grid_generator_ops.h> |
| 57 | #include <ATen/ops/arange_native.h> |
| 58 | #include <ATen/ops/arange_ops.h> |
| 59 | #include <ATen/ops/arange_native.h> |
| 60 | #include <ATen/ops/arange_ops.h> |
| 61 | #include <ATen/ops/arange_native.h> |
| 62 | #include <ATen/ops/arange_ops.h> |
| 63 | #include <ATen/ops/arange_native.h> |
| 64 | #include <ATen/ops/arange_ops.h> |
| 65 | #include <ATen/ops/arccosh_native.h> |
| 66 | #include <ATen/ops/arccosh_ops.h> |
| 67 | #include <ATen/ops/arccosh_native.h> |
| 68 | #include <ATen/ops/arccosh_ops.h> |
| 69 | #include <ATen/ops/arccosh_native.h> |
| 70 | #include <ATen/ops/arccosh_ops.h> |
| 71 | #include <ATen/ops/arcsinh_native.h> |
| 72 | #include <ATen/ops/arcsinh_ops.h> |
| 73 | #include <ATen/ops/arcsinh_native.h> |
| 74 | #include <ATen/ops/arcsinh_ops.h> |
| 75 | #include <ATen/ops/arcsinh_native.h> |
| 76 | #include <ATen/ops/arcsinh_ops.h> |
| 77 | #include <ATen/ops/atanh_native.h> |
| 78 | #include <ATen/ops/atanh_ops.h> |
| 79 | #include <ATen/ops/atanh_native.h> |
| 80 | #include <ATen/ops/atanh_ops.h> |
| 81 | #include <ATen/ops/atanh_native.h> |
| 82 | #include <ATen/ops/atanh_ops.h> |
| 83 | #include <ATen/ops/asin_native.h> |
| 84 | #include <ATen/ops/asin_ops.h> |
| 85 | #include <ATen/ops/asin_native.h> |
| 86 | #include <ATen/ops/asin_ops.h> |
| 87 | #include <ATen/ops/asin_native.h> |
| 88 | #include <ATen/ops/asin_ops.h> |
| 89 | #include <ATen/ops/binary_cross_entropy_backward_native.h> |
| 90 | #include <ATen/ops/binary_cross_entropy_backward_ops.h> |
| 91 | #include <ATen/ops/binary_cross_entropy_backward_native.h> |
| 92 | #include <ATen/ops/binary_cross_entropy_backward_ops.h> |
| 93 | #include <ATen/ops/binary_cross_entropy_with_logits_native.h> |
| 94 | #include <ATen/ops/binary_cross_entropy_with_logits_ops.h> |
| 95 | #include <ATen/ops/binary_cross_entropy_with_logits_native.h> |
| 96 | #include <ATen/ops/binary_cross_entropy_with_logits_ops.h> |
| 97 | #include <ATen/ops/logical_not_native.h> |
| 98 | #include <ATen/ops/logical_not_ops.h> |
| 99 | #include <ATen/ops/logical_not_native.h> |
| 100 | #include <ATen/ops/logical_not_ops.h> |
| 101 | #include <ATen/ops/logical_not_native.h> |
| 102 | #include <ATen/ops/logical_not_ops.h> |
| 103 | #include <ATen/ops/logical_and_native.h> |
| 104 | #include <ATen/ops/logical_and_ops.h> |
| 105 | #include <ATen/ops/logical_and_native.h> |
| 106 | #include <ATen/ops/logical_and_ops.h> |
| 107 | #include <ATen/ops/logical_and_native.h> |
| 108 | #include <ATen/ops/logical_and_ops.h> |
| 109 | #include <ATen/ops/concatenate_native.h> |
| 110 | #include <ATen/ops/concatenate_ops.h> |
| 111 | #include <ATen/ops/concatenate_native.h> |
| 112 | #include <ATen/ops/concatenate_ops.h> |
| 113 | #include <ATen/ops/concatenate_native.h> |
| 114 | #include <ATen/ops/concatenate_ops.h> |
| 115 | #include <ATen/ops/concatenate_native.h> |
| 116 | #include <ATen/ops/concatenate_ops.h> |
| 117 | #include <ATen/ops/block_diag_native.h> |
| 118 | #include <ATen/ops/block_diag_ops.h> |
| 119 | #include <ATen/ops/block_diag_native.h> |
| 120 | #include <ATen/ops/block_diag_ops.h> |
| 121 | #include <ATen/ops/chain_matmul_native.h> |
| 122 | #include <ATen/ops/chain_matmul_ops.h> |
| 123 | #include <ATen/ops/chain_matmul_native.h> |
| 124 | #include <ATen/ops/chain_matmul_ops.h> |
| 125 | #include <ATen/ops/convolution_backward_native.h> |
| 126 | #include <ATen/ops/convolution_backward_ops.h> |
| 127 | #include <ATen/ops/convolution_backward_native.h> |
| 128 | #include <ATen/ops/convolution_backward_ops.h> |
| 129 | #include <ATen/ops/_copy_from_native.h> |
| 130 | #include <ATen/ops/_copy_from_ops.h> |
| 131 | #include <ATen/ops/_copy_from_native.h> |
| 132 | #include <ATen/ops/_copy_from_ops.h> |
| 133 | #include <ATen/ops/cosh_native.h> |
| 134 | #include <ATen/ops/cosh_ops.h> |
| 135 | #include <ATen/ops/cosh_native.h> |
| 136 | #include <ATen/ops/cosh_ops.h> |
| 137 | #include <ATen/ops/cosh_native.h> |
| 138 | #include <ATen/ops/cosh_ops.h> |
| 139 | #include <ATen/ops/cudnn_convolution_transpose_native.h> |
| 140 | #include <ATen/ops/cudnn_convolution_transpose_ops.h> |
| 141 | #include <ATen/ops/cudnn_convolution_transpose_native.h> |
| 142 | #include <ATen/ops/cudnn_convolution_transpose_ops.h> |
| 143 | #include <ATen/ops/_mps_convolution_transpose_native.h> |
| 144 | #include <ATen/ops/_mps_convolution_transpose_ops.h> |
| 145 | #include <ATen/ops/_mps_convolution_transpose_native.h> |
| 146 | #include <ATen/ops/_mps_convolution_transpose_ops.h> |
| 147 | #include <ATen/ops/cudnn_grid_sampler_native.h> |
| 148 | #include <ATen/ops/cudnn_grid_sampler_ops.h> |
| 149 | #include <ATen/ops/cudnn_grid_sampler_native.h> |
| 150 | #include <ATen/ops/cudnn_grid_sampler_ops.h> |
| 151 | #include <ATen/ops/cumprod_native.h> |
| 152 | #include <ATen/ops/cumprod_ops.h> |
| 153 | #include <ATen/ops/cumprod_native.h> |
| 154 | #include <ATen/ops/cumprod_ops.h> |
| 155 | #include <ATen/ops/cumprod_native.h> |
| 156 | #include <ATen/ops/cumprod_ops.h> |
| 157 | #include <ATen/ops/cumprod_native.h> |
| 158 | #include <ATen/ops/cumprod_ops.h> |
| 159 | #include <ATen/ops/cumprod_native.h> |
| 160 | #include <ATen/ops/cumprod_ops.h> |
| 161 | #include <ATen/ops/cumprod_native.h> |
| 162 | #include <ATen/ops/cumprod_ops.h> |
| 163 | #include <ATen/ops/diag_embed_native.h> |
| 164 | #include <ATen/ops/diag_embed_ops.h> |
| 165 | #include <ATen/ops/diag_embed_native.h> |
| 166 | #include <ATen/ops/diag_embed_ops.h> |
| 167 | #include <ATen/ops/diagonal_backward_native.h> |
| 168 | #include <ATen/ops/diagonal_backward_ops.h> |
| 169 | #include <ATen/ops/diagonal_backward_native.h> |
| 170 | #include <ATen/ops/diagonal_backward_ops.h> |
| 171 | #include <ATen/ops/div_native.h> |
| 172 | #include <ATen/ops/div_ops.h> |
| 173 | #include <ATen/ops/div_native.h> |
| 174 | #include <ATen/ops/div_ops.h> |
| 175 | #include <ATen/ops/div_native.h> |
| 176 | #include <ATen/ops/div_ops.h> |
| 177 | #include <ATen/ops/div_native.h> |
| 178 | #include <ATen/ops/div_ops.h> |
| 179 | #include <ATen/ops/div_native.h> |
| 180 | #include <ATen/ops/div_ops.h> |
| 181 | #include <ATen/ops/div_native.h> |
| 182 | #include <ATen/ops/div_ops.h> |
| 183 | #include <ATen/ops/div_native.h> |
| 184 | #include <ATen/ops/div_ops.h> |
| 185 | #include <ATen/ops/div_native.h> |
| 186 | #include <ATen/ops/div_ops.h> |
| 187 | #include <ATen/ops/div_native.h> |
| 188 | #include <ATen/ops/div_ops.h> |
| 189 | #include <ATen/ops/div_native.h> |
| 190 | #include <ATen/ops/div_ops.h> |
| 191 | #include <ATen/ops/div_native.h> |
| 192 | #include <ATen/ops/div_ops.h> |
| 193 | #include <ATen/ops/div_native.h> |
| 194 | #include <ATen/ops/div_ops.h> |
| 195 | #include <ATen/ops/_embedding_bag_native.h> |
| 196 | #include <ATen/ops/_embedding_bag_ops.h> |
| 197 | #include <ATen/ops/_embedding_bag_native.h> |
| 198 | #include <ATen/ops/_embedding_bag_ops.h> |
| 199 | #include <ATen/ops/_embedding_bag_per_sample_weights_backward_native.h> |
| 200 | #include <ATen/ops/_embedding_bag_per_sample_weights_backward_ops.h> |
| 201 | #include <ATen/ops/_embedding_bag_per_sample_weights_backward_native.h> |
| 202 | #include <ATen/ops/_embedding_bag_per_sample_weights_backward_ops.h> |
| 203 | #include <ATen/ops/new_full_native.h> |
| 204 | #include <ATen/ops/new_full_ops.h> |
| 205 | #include <ATen/ops/new_full_native.h> |
| 206 | #include <ATen/ops/new_full_ops.h> |
| 207 | #include <ATen/ops/empty_quantized_native.h> |
| 208 | #include <ATen/ops/empty_quantized_ops.h> |
| 209 | #include <ATen/ops/empty_quantized_native.h> |
| 210 | #include <ATen/ops/empty_quantized_ops.h> |
| 211 | #include <ATen/ops/empty_strided_native.h> |
| 212 | #include <ATen/ops/empty_strided_ops.h> |
| 213 | #include <ATen/ops/empty_strided_native.h> |
| 214 | #include <ATen/ops/empty_strided_ops.h> |
| 215 | #include <ATen/ops/exp_native.h> |
| 216 | #include <ATen/ops/exp_ops.h> |
| 217 | #include <ATen/ops/exp_native.h> |
| 218 | #include <ATen/ops/exp_ops.h> |
| 219 | #include <ATen/ops/exp_native.h> |
| 220 | #include <ATen/ops/exp_ops.h> |
| 221 | #include <ATen/ops/expm1_native.h> |
| 222 | #include <ATen/ops/expm1_ops.h> |
| 223 | #include <ATen/ops/expm1_native.h> |
| 224 | #include <ATen/ops/expm1_ops.h> |
| 225 | #include <ATen/ops/expm1_native.h> |
| 226 | #include <ATen/ops/expm1_ops.h> |
| 227 | #include <ATen/ops/fill_native.h> |
| 228 | #include <ATen/ops/fill_ops.h> |
| 229 | #include <ATen/ops/fill_native.h> |
| 230 | #include <ATen/ops/fill_ops.h> |
| 231 | #include <ATen/ops/fill_native.h> |
| 232 | #include <ATen/ops/fill_ops.h> |
| 233 | #include <ATen/ops/fill_native.h> |
| 234 | #include <ATen/ops/fill_ops.h> |
| 235 | #include <ATen/ops/fill_native.h> |
| 236 | #include <ATen/ops/fill_ops.h> |
| 237 | #include <ATen/ops/fill_native.h> |
| 238 | #include <ATen/ops/fill_ops.h> |
| 239 | #include <ATen/ops/floor_native.h> |
| 240 | #include <ATen/ops/floor_ops.h> |
| 241 | #include <ATen/ops/floor_native.h> |
| 242 | #include <ATen/ops/floor_ops.h> |
| 243 | #include <ATen/ops/floor_native.h> |
| 244 | #include <ATen/ops/floor_ops.h> |
| 245 | #include <ATen/ops/floor_divide_native.h> |
| 246 | #include <ATen/ops/floor_divide_ops.h> |
| 247 | #include <ATen/ops/floor_divide_native.h> |
| 248 | #include <ATen/ops/floor_divide_ops.h> |
| 249 | #include <ATen/ops/floor_divide_native.h> |
| 250 | #include <ATen/ops/floor_divide_ops.h> |
| 251 | #include <ATen/ops/full_native.h> |
| 252 | #include <ATen/ops/full_ops.h> |
| 253 | #include <ATen/ops/full_native.h> |
| 254 | #include <ATen/ops/full_ops.h> |
| 255 | #include <ATen/ops/full_native.h> |
| 256 | #include <ATen/ops/full_ops.h> |
| 257 | #include <ATen/ops/full_native.h> |
| 258 | #include <ATen/ops/full_ops.h> |
| 259 | #include <ATen/ops/from_file_native.h> |
| 260 | #include <ATen/ops/from_file_ops.h> |
| 261 | #include <ATen/ops/from_file_native.h> |
| 262 | #include <ATen/ops/from_file_ops.h> |
| 263 | #include <ATen/ops/grid_sampler_2d_native.h> |
| 264 | #include <ATen/ops/grid_sampler_2d_ops.h> |
| 265 | #include <ATen/ops/grid_sampler_2d_native.h> |
| 266 | #include <ATen/ops/grid_sampler_2d_ops.h> |
| 267 | #include <ATen/ops/grid_sampler_3d_native.h> |
| 268 | #include <ATen/ops/grid_sampler_3d_ops.h> |
| 269 | #include <ATen/ops/grid_sampler_3d_native.h> |
| 270 | #include <ATen/ops/grid_sampler_3d_ops.h> |
| 271 | #include <ATen/ops/hamming_window_native.h> |
| 272 | #include <ATen/ops/hamming_window_ops.h> |
| 273 | #include <ATen/ops/hamming_window_native.h> |
| 274 | #include <ATen/ops/hamming_window_ops.h> |
| 275 | #include <ATen/ops/hamming_window_native.h> |
| 276 | #include <ATen/ops/hamming_window_ops.h> |
| 277 | #include <ATen/ops/hamming_window_native.h> |
| 278 | #include <ATen/ops/hamming_window_ops.h> |
| 279 | #include <ATen/ops/hamming_window_native.h> |
| 280 | #include <ATen/ops/hamming_window_ops.h> |
| 281 | #include <ATen/ops/hamming_window_native.h> |
| 282 | #include <ATen/ops/hamming_window_ops.h> |
| 283 | #include <ATen/ops/hamming_window_native.h> |
| 284 | #include <ATen/ops/hamming_window_ops.h> |
| 285 | #include <ATen/ops/hamming_window_native.h> |
| 286 | #include <ATen/ops/hamming_window_ops.h> |
| 287 | #include <ATen/ops/native_group_norm_native.h> |
| 288 | #include <ATen/ops/native_group_norm_ops.h> |
| 289 | #include <ATen/ops/native_group_norm_native.h> |
| 290 | #include <ATen/ops/native_group_norm_ops.h> |
| 291 | #include <ATen/ops/_fft_c2r_native.h> |
| 292 | #include <ATen/ops/_fft_c2r_ops.h> |
| 293 | #include <ATen/ops/_fft_c2r_native.h> |
| 294 | #include <ATen/ops/_fft_c2r_ops.h> |
| 295 | #include <ATen/ops/isnan_native.h> |
| 296 | #include <ATen/ops/isnan_ops.h> |
| 297 | #include <ATen/ops/isnan_native.h> |
| 298 | #include <ATen/ops/isnan_ops.h> |
| 299 | #include <ATen/ops/ldexp_native.h> |
| 300 | #include <ATen/ops/ldexp_ops.h> |
| 301 | #include <ATen/ops/ldexp_native.h> |
| 302 | #include <ATen/ops/ldexp_ops.h> |
| 303 | #include <ATen/ops/ldexp_native.h> |
| 304 | #include <ATen/ops/ldexp_ops.h> |
| 305 | #include <ATen/ops/log2_native.h> |
| 306 | #include <ATen/ops/log2_ops.h> |
| 307 | #include <ATen/ops/log2_native.h> |
| 308 | #include <ATen/ops/log2_ops.h> |
| 309 | #include <ATen/ops/log2_native.h> |
| 310 | #include <ATen/ops/log2_ops.h> |
| 311 | #include <ATen/ops/logaddexp2_native.h> |
| 312 | #include <ATen/ops/logaddexp2_ops.h> |
| 313 | #include <ATen/ops/logaddexp2_native.h> |
| 314 | #include <ATen/ops/logaddexp2_ops.h> |
| 315 | #include <ATen/ops/xlogy_native.h> |
| 316 | #include <ATen/ops/xlogy_ops.h> |
| 317 | #include <ATen/ops/xlogy_native.h> |
| 318 | #include <ATen/ops/xlogy_ops.h> |
| 319 | #include <ATen/ops/xlogy_native.h> |
| 320 | #include <ATen/ops/xlogy_ops.h> |
| 321 | #include <ATen/ops/xlogy_native.h> |
| 322 | #include <ATen/ops/xlogy_ops.h> |
| 323 | #include <ATen/ops/xlogy_native.h> |
| 324 | #include <ATen/ops/xlogy_ops.h> |
| 325 | #include <ATen/ops/xlogy_native.h> |
| 326 | #include <ATen/ops/xlogy_ops.h> |
| 327 | #include <ATen/ops/xlogy_native.h> |
| 328 | #include <ATen/ops/xlogy_ops.h> |
| 329 | #include <ATen/ops/xlogy_native.h> |
| 330 | #include <ATen/ops/xlogy_ops.h> |
| 331 | #include <ATen/ops/logspace_native.h> |
| 332 | #include <ATen/ops/logspace_ops.h> |
| 333 | #include <ATen/ops/logspace_native.h> |
| 334 | #include <ATen/ops/logspace_ops.h> |
| 335 | #include <ATen/ops/matrix_power_native.h> |
| 336 | #include <ATen/ops/matrix_power_ops.h> |
| 337 | #include <ATen/ops/matrix_power_native.h> |
| 338 | #include <ATen/ops/matrix_power_ops.h> |
| 339 | #include <ATen/ops/_aminmax_native.h> |
| 340 | #include <ATen/ops/_aminmax_ops.h> |
| 341 | #include <ATen/ops/_aminmax_native.h> |
| 342 | #include <ATen/ops/_aminmax_ops.h> |
| 343 | #include <ATen/ops/_aminmax_native.h> |
| 344 | #include <ATen/ops/_aminmax_ops.h> |
| 345 | #include <ATen/ops/_aminmax_native.h> |
| 346 | #include <ATen/ops/_aminmax_ops.h> |
| 347 | #include <ATen/ops/aminmax_native.h> |
| 348 | #include <ATen/ops/aminmax_ops.h> |
| 349 | #include <ATen/ops/aminmax_native.h> |
| 350 | #include <ATen/ops/aminmax_ops.h> |
| 351 | #include <ATen/ops/_compute_linear_combination_native.h> |
| 352 | #include <ATen/ops/_compute_linear_combination_ops.h> |
| 353 | #include <ATen/ops/_compute_linear_combination_native.h> |
| 354 | #include <ATen/ops/_compute_linear_combination_ops.h> |
| 355 | #include <ATen/ops/_mps_max_pool2d_native.h> |
| 356 | #include <ATen/ops/_mps_max_pool2d_ops.h> |
| 357 | #include <ATen/ops/_mps_max_pool2d_native.h> |
| 358 | #include <ATen/ops/_mps_max_pool2d_ops.h> |
| 359 | #include <ATen/ops/mkldnn_max_pool3d_backward_native.h> |
| 360 | #include <ATen/ops/mkldnn_max_pool3d_backward_ops.h> |
| 361 | #include <ATen/ops/mkldnn_max_pool3d_backward_native.h> |
| 362 | #include <ATen/ops/mkldnn_max_pool3d_backward_ops.h> |
| 363 | #include <ATen/ops/min_native.h> |
| 364 | #include <ATen/ops/min_ops.h> |
| 365 | #include <ATen/ops/min_native.h> |
| 366 | #include <ATen/ops/min_ops.h> |
| 367 | #include <ATen/ops/min_native.h> |
| 368 | #include <ATen/ops/min_ops.h> |
| 369 | #include <ATen/ops/min_native.h> |
| 370 | #include <ATen/ops/min_ops.h> |
| 371 | #include <ATen/ops/mps_convolution_backward_native.h> |
| 372 | #include <ATen/ops/mps_convolution_backward_ops.h> |
| 373 | #include <ATen/ops/mps_convolution_backward_native.h> |
| 374 | #include <ATen/ops/mps_convolution_backward_ops.h> |
| 375 | #include <ATen/ops/miopen_rnn_native.h> |
| 376 | #include <ATen/ops/miopen_rnn_ops.h> |
| 377 | #include <ATen/ops/miopen_rnn_native.h> |
| 378 | #include <ATen/ops/miopen_rnn_ops.h> |
| 379 | #include <ATen/ops/mv_native.h> |
| 380 | #include <ATen/ops/mv_ops.h> |
| 381 | #include <ATen/ops/mv_native.h> |
| 382 | #include <ATen/ops/mv_ops.h> |
| 383 | #include <ATen/ops/_native_batch_norm_legit_native.h> |
| 384 | #include <ATen/ops/_native_batch_norm_legit_ops.h> |
| 385 | #include <ATen/ops/_native_batch_norm_legit_native.h> |
| 386 | #include <ATen/ops/_native_batch_norm_legit_ops.h> |
| 387 | #include <ATen/ops/_native_batch_norm_legit_native.h> |
| 388 | #include <ATen/ops/_native_batch_norm_legit_ops.h> |
| 389 | #include <ATen/ops/_native_batch_norm_legit_native.h> |
| 390 | #include <ATen/ops/_native_batch_norm_legit_ops.h> |
| 391 | #include <ATen/ops/_native_batch_norm_legit_native.h> |
| 392 | #include <ATen/ops/_native_batch_norm_legit_ops.h> |
| 393 | #include <ATen/ops/batch_norm_stats_native.h> |
| 394 | #include <ATen/ops/batch_norm_stats_ops.h> |
| 395 | #include <ATen/ops/batch_norm_stats_native.h> |
| 396 | #include <ATen/ops/batch_norm_stats_ops.h> |
| 397 | #include <ATen/ops/batch_norm_backward_elemt_native.h> |
| 398 | #include <ATen/ops/batch_norm_backward_elemt_ops.h> |
| 399 | #include <ATen/ops/batch_norm_backward_elemt_native.h> |
| 400 | #include <ATen/ops/batch_norm_backward_elemt_ops.h> |
| 401 | #include <ATen/ops/_euclidean_dist_native.h> |
| 402 | #include <ATen/ops/_euclidean_dist_ops.h> |
| 403 | #include <ATen/ops/_euclidean_dist_native.h> |
| 404 | #include <ATen/ops/_euclidean_dist_ops.h> |
| 405 | #include <ATen/ops/_cdist_forward_native.h> |
| 406 | #include <ATen/ops/_cdist_forward_ops.h> |
| 407 | #include <ATen/ops/_cdist_forward_native.h> |
| 408 | #include <ATen/ops/_cdist_forward_ops.h> |
| 409 | #include <ATen/ops/_cdist_backward_native.h> |
| 410 | #include <ATen/ops/_cdist_backward_ops.h> |
| 411 | #include <ATen/ops/_cdist_backward_native.h> |
| 412 | #include <ATen/ops/_cdist_backward_ops.h> |
| 413 | #include <ATen/ops/pixel_unshuffle_native.h> |
| 414 | #include <ATen/ops/pixel_unshuffle_ops.h> |
| 415 | #include <ATen/ops/pixel_unshuffle_native.h> |
| 416 | #include <ATen/ops/pixel_unshuffle_ops.h> |
| 417 | #include <ATen/ops/rad2deg_native.h> |
| 418 | #include <ATen/ops/rad2deg_ops.h> |
| 419 | #include <ATen/ops/rad2deg_native.h> |
| 420 | #include <ATen/ops/rad2deg_ops.h> |
| 421 | #include <ATen/ops/rad2deg_native.h> |
| 422 | #include <ATen/ops/rad2deg_ops.h> |
| 423 | #include <ATen/ops/scalar_tensor_native.h> |
| 424 | #include <ATen/ops/scalar_tensor_ops.h> |
| 425 | #include <ATen/ops/scalar_tensor_native.h> |
| 426 | #include <ATen/ops/scalar_tensor_ops.h> |
| 427 | #include <ATen/ops/rand_native.h> |
| 428 | #include <ATen/ops/rand_ops.h> |
| 429 | #include <ATen/ops/rand_native.h> |
| 430 | #include <ATen/ops/rand_ops.h> |
| 431 | #include <ATen/ops/rand_native.h> |
| 432 | #include <ATen/ops/rand_ops.h> |
| 433 | #include <ATen/ops/rand_native.h> |
| 434 | #include <ATen/ops/rand_ops.h> |
| 435 | #include <ATen/ops/rand_native.h> |
| 436 | #include <ATen/ops/rand_ops.h> |
| 437 | #include <ATen/ops/rand_native.h> |
| 438 | #include <ATen/ops/rand_ops.h> |
| 439 | #include <ATen/ops/rand_native.h> |
| 440 | #include <ATen/ops/rand_ops.h> |
| 441 | #include <ATen/ops/rand_native.h> |
| 442 | #include <ATen/ops/rand_ops.h> |
| 443 | #include <ATen/ops/rand_like_native.h> |
| 444 | #include <ATen/ops/rand_like_ops.h> |
| 445 | #include <ATen/ops/rand_like_native.h> |
| 446 | #include <ATen/ops/rand_like_ops.h> |
| 447 | #include <ATen/ops/relu_native.h> |
| 448 | #include <ATen/ops/relu_ops.h> |
| 449 | #include <ATen/ops/relu_native.h> |
| 450 | #include <ATen/ops/relu_ops.h> |
| 451 | #include <ATen/ops/relu_native.h> |
| 452 | #include <ATen/ops/relu_ops.h> |
| 453 | #include <ATen/ops/logit_native.h> |
| 454 | #include <ATen/ops/logit_ops.h> |
| 455 | #include <ATen/ops/logit_native.h> |
| 456 | #include <ATen/ops/logit_ops.h> |
| 457 | #include <ATen/ops/logit_native.h> |
| 458 | #include <ATen/ops/logit_ops.h> |
| 459 | #include <ATen/ops/select_scatter_native.h> |
| 460 | #include <ATen/ops/select_scatter_ops.h> |
| 461 | #include <ATen/ops/select_scatter_native.h> |
| 462 | #include <ATen/ops/select_scatter_ops.h> |
| 463 | #include <ATen/ops/softmax_native.h> |
| 464 | #include <ATen/ops/softmax_ops.h> |
| 465 | #include <ATen/ops/softmax_native.h> |
| 466 | #include <ATen/ops/softmax_ops.h> |
| 467 | #include <ATen/ops/stack_native.h> |
| 468 | #include <ATen/ops/stack_ops.h> |
| 469 | #include <ATen/ops/stack_native.h> |
| 470 | #include <ATen/ops/stack_ops.h> |
| 471 | #include <ATen/ops/vstack_native.h> |
| 472 | #include <ATen/ops/vstack_ops.h> |
| 473 | #include <ATen/ops/vstack_native.h> |
| 474 | #include <ATen/ops/vstack_ops.h> |
| 475 | #include <ATen/ops/nansum_native.h> |
| 476 | #include <ATen/ops/nansum_ops.h> |
| 477 | #include <ATen/ops/nansum_native.h> |
| 478 | #include <ATen/ops/nansum_ops.h> |
| 479 | #include <ATen/ops/sqrt_native.h> |
| 480 | #include <ATen/ops/sqrt_ops.h> |
| 481 | #include <ATen/ops/sqrt_native.h> |
| 482 | #include <ATen/ops/sqrt_ops.h> |
| 483 | #include <ATen/ops/sqrt_native.h> |
| 484 | #include <ATen/ops/sqrt_ops.h> |
| 485 | #include <ATen/ops/prod_native.h> |
| 486 | #include <ATen/ops/prod_ops.h> |
| 487 | #include <ATen/ops/prod_native.h> |
| 488 | #include <ATen/ops/prod_ops.h> |
| 489 | #include <ATen/ops/prod_native.h> |
| 490 | #include <ATen/ops/prod_ops.h> |
| 491 | #include <ATen/ops/prod_native.h> |
| 492 | #include <ATen/ops/prod_ops.h> |
| 493 | #include <ATen/ops/prod_native.h> |
| 494 | #include <ATen/ops/prod_ops.h> |
| 495 | #include <ATen/ops/prod_native.h> |
| 496 | #include <ATen/ops/prod_ops.h> |
| 497 | #include <ATen/ops/threshold_backward_native.h> |
| 498 | #include <ATen/ops/threshold_backward_ops.h> |
| 499 | #include <ATen/ops/threshold_backward_native.h> |
| 500 | #include <ATen/ops/threshold_backward_ops.h> |
| 501 | #include <ATen/ops/_transform_bias_rescale_qkv_native.h> |
| 502 | #include <ATen/ops/_transform_bias_rescale_qkv_ops.h> |
| 503 | #include <ATen/ops/_transform_bias_rescale_qkv_native.h> |
| 504 | #include <ATen/ops/_transform_bias_rescale_qkv_ops.h> |
| 505 | #include <ATen/ops/_nested_from_padded_native.h> |
| 506 | #include <ATen/ops/_nested_from_padded_ops.h> |
| 507 | #include <ATen/ops/_nested_from_padded_native.h> |
| 508 | #include <ATen/ops/_nested_from_padded_ops.h> |
| 509 | #include <ATen/ops/_nested_tensor_size_native.h> |
| 510 | #include <ATen/ops/_nested_tensor_size_ops.h> |
| 511 | #include <ATen/ops/_nested_tensor_size_native.h> |
| 512 | #include <ATen/ops/_nested_tensor_size_ops.h> |
| 513 | #include <ATen/ops/_nested_view_from_buffer_copy_native.h> |
| 514 | #include <ATen/ops/_nested_view_from_buffer_copy_ops.h> |
| 515 | #include <ATen/ops/_nested_view_from_buffer_copy_native.h> |
| 516 | #include <ATen/ops/_nested_view_from_buffer_copy_ops.h> |
| 517 | #include <ATen/ops/trunc_native.h> |
| 518 | #include <ATen/ops/trunc_ops.h> |
| 519 | #include <ATen/ops/trunc_native.h> |
| 520 | #include <ATen/ops/trunc_ops.h> |
| 521 | #include <ATen/ops/trunc_native.h> |
| 522 | #include <ATen/ops/trunc_ops.h> |
| 523 | #include <ATen/ops/unique_dim_consecutive_native.h> |
| 524 | #include <ATen/ops/unique_dim_consecutive_ops.h> |
| 525 | #include <ATen/ops/unique_dim_consecutive_native.h> |
| 526 | #include <ATen/ops/unique_dim_consecutive_ops.h> |
| 527 | #include <ATen/ops/where_native.h> |
| 528 | #include <ATen/ops/where_ops.h> |
| 529 | #include <ATen/ops/where_native.h> |
| 530 | #include <ATen/ops/where_ops.h> |
| 531 | #include <ATen/ops/_weight_norm_interface_backward_native.h> |
| 532 | #include <ATen/ops/_weight_norm_interface_backward_ops.h> |
| 533 | #include <ATen/ops/_weight_norm_interface_backward_native.h> |
| 534 | #include <ATen/ops/_weight_norm_interface_backward_ops.h> |
| 535 | #include <ATen/ops/_sample_dirichlet_native.h> |
| 536 | #include <ATen/ops/_sample_dirichlet_ops.h> |
| 537 | #include <ATen/ops/_sample_dirichlet_native.h> |
| 538 | #include <ATen/ops/_sample_dirichlet_ops.h> |
| 539 | #include <ATen/ops/binomial_native.h> |
| 540 | #include <ATen/ops/binomial_ops.h> |
| 541 | #include <ATen/ops/binomial_native.h> |
| 542 | #include <ATen/ops/binomial_ops.h> |
| 543 | #include <ATen/ops/native_norm_native.h> |
| 544 | #include <ATen/ops/native_norm_ops.h> |
| 545 | #include <ATen/ops/native_norm_native.h> |
| 546 | #include <ATen/ops/native_norm_ops.h> |
| 547 | #include <ATen/ops/native_norm_native.h> |
| 548 | #include <ATen/ops/native_norm_ops.h> |
| 549 | #include <ATen/ops/native_norm_native.h> |
| 550 | #include <ATen/ops/native_norm_ops.h> |
| 551 | #include <ATen/ops/_sparse_sum_native.h> |
| 552 | #include <ATen/ops/_sparse_sum_ops.h> |
| 553 | #include <ATen/ops/_sparse_sum_native.h> |
| 554 | #include <ATen/ops/_sparse_sum_ops.h> |
| 555 | #include <ATen/ops/_sparse_sum_backward_native.h> |
| 556 | #include <ATen/ops/_sparse_sum_backward_ops.h> |
| 557 | #include <ATen/ops/_sparse_sum_backward_native.h> |
| 558 | #include <ATen/ops/_sparse_sum_backward_ops.h> |
| 559 | #include <ATen/ops/_sparse_softmax_native.h> |
| 560 | #include <ATen/ops/_sparse_softmax_ops.h> |
| 561 | #include <ATen/ops/_sparse_softmax_native.h> |
| 562 | #include <ATen/ops/_sparse_softmax_ops.h> |
| 563 | #include <ATen/ops/clone_native.h> |
| 564 | #include <ATen/ops/clone_ops.h> |
| 565 | #include <ATen/ops/clone_native.h> |
| 566 | #include <ATen/ops/clone_ops.h> |
| 567 | #include <ATen/ops/resize_as_native.h> |
| 568 | #include <ATen/ops/resize_as_ops.h> |
| 569 | #include <ATen/ops/resize_as_native.h> |
| 570 | #include <ATen/ops/resize_as_ops.h> |
| 571 | #include <ATen/ops/resize_as_native.h> |
| 572 | #include <ATen/ops/resize_as_ops.h> |
| 573 | #include <ATen/ops/zero_native.h> |
| 574 | #include <ATen/ops/zero_ops.h> |
| 575 | #include <ATen/ops/zero_native.h> |
| 576 | #include <ATen/ops/zero_ops.h> |
| 577 | #include <ATen/ops/zero_native.h> |
| 578 | #include <ATen/ops/zero_ops.h> |
| 579 | #include <ATen/ops/heaviside_native.h> |
| 580 | #include <ATen/ops/heaviside_ops.h> |
| 581 | #include <ATen/ops/heaviside_native.h> |
| 582 | #include <ATen/ops/heaviside_ops.h> |
| 583 | #include <ATen/ops/heaviside_native.h> |
| 584 | #include <ATen/ops/heaviside_ops.h> |
| 585 | #include <ATen/ops/addmm_native.h> |
| 586 | #include <ATen/ops/addmm_ops.h> |
| 587 | #include <ATen/ops/addmm_native.h> |
| 588 | #include <ATen/ops/addmm_ops.h> |
| 589 | #include <ATen/ops/addmm_native.h> |
| 590 | #include <ATen/ops/addmm_ops.h> |
| 591 | #include <ATen/ops/_sparse_coo_tensor_with_dims_native.h> |
| 592 | #include <ATen/ops/_sparse_coo_tensor_with_dims_ops.h> |
| 593 | #include <ATen/ops/_sparse_coo_tensor_with_dims_native.h> |
| 594 | #include <ATen/ops/_sparse_coo_tensor_with_dims_ops.h> |
| 595 | #include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors_native.h> |
| 596 | #include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors_ops.h> |
| 597 | #include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors_native.h> |
| 598 | #include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors_ops.h> |
| 599 | #include <ATen/ops/sparse_resize_and_clear_native.h> |
| 600 | #include <ATen/ops/sparse_resize_and_clear_ops.h> |
| 601 | #include <ATen/ops/sparse_resize_and_clear_native.h> |
| 602 | #include <ATen/ops/sparse_resize_and_clear_ops.h> |
| 603 | #include <ATen/ops/sparse_resize_and_clear_native.h> |
| 604 | #include <ATen/ops/sparse_resize_and_clear_ops.h> |
| 605 | #include <ATen/ops/hspmm_native.h> |
| 606 | #include <ATen/ops/hspmm_ops.h> |
| 607 | #include <ATen/ops/hspmm_native.h> |
| 608 | #include <ATen/ops/hspmm_ops.h> |
| 609 | #include <ATen/ops/to_sparse_native.h> |
| 610 | #include <ATen/ops/to_sparse_ops.h> |
| 611 | #include <ATen/ops/to_sparse_native.h> |
| 612 | #include <ATen/ops/to_sparse_ops.h> |
| 613 | #include <ATen/ops/to_sparse_native.h> |
| 614 | #include <ATen/ops/to_sparse_ops.h> |
| 615 | #include <ATen/ops/to_sparse_native.h> |
| 616 | #include <ATen/ops/to_sparse_ops.h> |
| 617 | #include <ATen/ops/to_sparse_bsr_native.h> |
| 618 | #include <ATen/ops/to_sparse_bsr_ops.h> |
| 619 | #include <ATen/ops/to_sparse_bsr_native.h> |
| 620 | #include <ATen/ops/to_sparse_bsr_ops.h> |
| 621 | #include <ATen/ops/to_mkldnn_native.h> |
| 622 | #include <ATen/ops/to_mkldnn_ops.h> |
| 623 | #include <ATen/ops/to_mkldnn_native.h> |
| 624 | #include <ATen/ops/to_mkldnn_ops.h> |
| 625 | #include <ATen/ops/mkldnn_reorder_conv3d_weight_native.h> |
| 626 | #include <ATen/ops/mkldnn_reorder_conv3d_weight_ops.h> |
| 627 | #include <ATen/ops/mkldnn_reorder_conv3d_weight_native.h> |
| 628 | #include <ATen/ops/mkldnn_reorder_conv3d_weight_ops.h> |
| 629 | #include <ATen/ops/q_per_channel_scales_native.h> |
| 630 | #include <ATen/ops/q_per_channel_scales_ops.h> |
| 631 | #include <ATen/ops/q_per_channel_scales_native.h> |
| 632 | #include <ATen/ops/q_per_channel_scales_ops.h> |
| 633 | #include <ATen/ops/int_repr_native.h> |
| 634 | #include <ATen/ops/int_repr_ops.h> |
| 635 | #include <ATen/ops/int_repr_native.h> |
| 636 | #include <ATen/ops/int_repr_ops.h> |
| 637 | #include <ATen/ops/_make_per_tensor_quantized_tensor_native.h> |
| 638 | #include <ATen/ops/_make_per_tensor_quantized_tensor_ops.h> |
| 639 | #include <ATen/ops/_make_per_tensor_quantized_tensor_native.h> |
| 640 | #include <ATen/ops/_make_per_tensor_quantized_tensor_ops.h> |
| 641 | #include <ATen/ops/_thnn_fused_lstm_cell_native.h> |
| 642 | #include <ATen/ops/_thnn_fused_lstm_cell_ops.h> |
| 643 | #include <ATen/ops/_thnn_fused_lstm_cell_native.h> |
| 644 | #include <ATen/ops/_thnn_fused_lstm_cell_ops.h> |
| 645 | #include <ATen/ops/_thnn_fused_lstm_cell_backward_impl_native.h> |
| 646 | #include <ATen/ops/_thnn_fused_lstm_cell_backward_impl_ops.h> |
| 647 | #include <ATen/ops/_thnn_fused_lstm_cell_backward_impl_native.h> |
| 648 | #include <ATen/ops/_thnn_fused_lstm_cell_backward_impl_ops.h> |
| 649 | #include <ATen/ops/masked_fill_native.h> |
| 650 | #include <ATen/ops/masked_fill_ops.h> |
| 651 | #include <ATen/ops/masked_fill_native.h> |
| 652 | #include <ATen/ops/masked_fill_ops.h> |
| 653 | #include <ATen/ops/masked_fill_native.h> |
| 654 | #include <ATen/ops/masked_fill_ops.h> |
| 655 | #include <ATen/ops/masked_fill_native.h> |
| 656 | #include <ATen/ops/masked_fill_ops.h> |
| 657 | #include <ATen/ops/masked_fill_native.h> |
| 658 | #include <ATen/ops/masked_fill_ops.h> |
| 659 | #include <ATen/ops/masked_fill_native.h> |
| 660 | #include <ATen/ops/masked_fill_ops.h> |
| 661 | #include <ATen/ops/_masked_softmax_native.h> |
| 662 | #include <ATen/ops/_masked_softmax_ops.h> |
| 663 | #include <ATen/ops/_masked_softmax_native.h> |
| 664 | #include <ATen/ops/_masked_softmax_ops.h> |
| 665 | #include <ATen/ops/bitwise_right_shift_native.h> |
| 666 | #include <ATen/ops/bitwise_right_shift_ops.h> |
| 667 | #include <ATen/ops/bitwise_right_shift_native.h> |
| 668 | #include <ATen/ops/bitwise_right_shift_ops.h> |
| 669 | #include <ATen/ops/bitwise_right_shift_native.h> |
| 670 | #include <ATen/ops/bitwise_right_shift_ops.h> |
| 671 | #include <ATen/ops/bitwise_right_shift_native.h> |
| 672 | #include <ATen/ops/bitwise_right_shift_ops.h> |
| 673 | #include <ATen/ops/bitwise_right_shift_native.h> |
| 674 | #include <ATen/ops/bitwise_right_shift_ops.h> |
| 675 | #include <ATen/ops/bitwise_right_shift_native.h> |
| 676 | #include <ATen/ops/bitwise_right_shift_ops.h> |
| 677 | #include <ATen/ops/bitwise_right_shift_native.h> |
| 678 | #include <ATen/ops/bitwise_right_shift_ops.h> |
| 679 | #include <ATen/ops/bitwise_right_shift_native.h> |
| 680 | #include <ATen/ops/bitwise_right_shift_ops.h> |
| 681 | #include <ATen/ops/cauchy_native.h> |
| 682 | #include <ATen/ops/cauchy_ops.h> |
| 683 | #include <ATen/ops/cauchy_native.h> |
| 684 | #include <ATen/ops/cauchy_ops.h> |
| 685 | #include <ATen/ops/cauchy_native.h> |
| 686 | #include <ATen/ops/cauchy_ops.h> |
| 687 | #include <ATen/ops/log_normal_native.h> |
| 688 | #include <ATen/ops/log_normal_ops.h> |
| 689 | #include <ATen/ops/log_normal_native.h> |
| 690 | #include <ATen/ops/log_normal_ops.h> |
| 691 | #include <ATen/ops/log_normal_native.h> |
| 692 | #include <ATen/ops/log_normal_ops.h> |
| 693 | #include <ATen/ops/diag_native.h> |
| 694 | #include <ATen/ops/diag_ops.h> |
| 695 | #include <ATen/ops/diag_native.h> |
| 696 | #include <ATen/ops/diag_ops.h> |
| 697 | #include <ATen/ops/ne_native.h> |
| 698 | #include <ATen/ops/ne_ops.h> |
| 699 | #include <ATen/ops/ne_native.h> |
| 700 | #include <ATen/ops/ne_ops.h> |
| 701 | #include <ATen/ops/ne_native.h> |
| 702 | #include <ATen/ops/ne_ops.h> |
| 703 | #include <ATen/ops/ne_native.h> |
| 704 | #include <ATen/ops/ne_ops.h> |
| 705 | #include <ATen/ops/ne_native.h> |
| 706 | #include <ATen/ops/ne_ops.h> |
| 707 | #include <ATen/ops/ne_native.h> |
| 708 | #include <ATen/ops/ne_ops.h> |
| 709 | #include <ATen/ops/not_equal_native.h> |
| 710 | #include <ATen/ops/not_equal_ops.h> |
| 711 | #include <ATen/ops/not_equal_native.h> |
| 712 | #include <ATen/ops/not_equal_ops.h> |
| 713 | #include <ATen/ops/not_equal_native.h> |
| 714 | #include <ATen/ops/not_equal_ops.h> |
| 715 | #include <ATen/ops/not_equal_native.h> |
| 716 | #include <ATen/ops/not_equal_ops.h> |
| 717 | #include <ATen/ops/not_equal_native.h> |
| 718 | #include <ATen/ops/not_equal_ops.h> |
| 719 | #include <ATen/ops/not_equal_native.h> |
| 720 | #include <ATen/ops/not_equal_ops.h> |
| 721 | #include <ATen/ops/addcmul_native.h> |
| 722 | #include <ATen/ops/addcmul_ops.h> |
| 723 | #include <ATen/ops/addcmul_native.h> |
| 724 | #include <ATen/ops/addcmul_ops.h> |
| 725 | #include <ATen/ops/addcmul_native.h> |
| 726 | #include <ATen/ops/addcmul_ops.h> |
| 727 | #include <ATen/ops/ormqr_native.h> |
| 728 | #include <ATen/ops/ormqr_ops.h> |
| 729 | #include <ATen/ops/ormqr_native.h> |
| 730 | #include <ATen/ops/ormqr_ops.h> |
| 731 | #include <ATen/ops/lu_unpack_native.h> |
| 732 | #include <ATen/ops/lu_unpack_ops.h> |
| 733 | #include <ATen/ops/lu_unpack_native.h> |
| 734 | #include <ATen/ops/lu_unpack_ops.h> |
| 735 | #include <ATen/ops/dist_native.h> |
| 736 | #include <ATen/ops/dist_ops.h> |
| 737 | #include <ATen/ops/dist_native.h> |
| 738 | #include <ATen/ops/dist_ops.h> |
| 739 | #include <ATen/ops/arctan2_native.h> |
| 740 | #include <ATen/ops/arctan2_ops.h> |
| 741 | #include <ATen/ops/arctan2_native.h> |
| 742 | #include <ATen/ops/arctan2_ops.h> |
| 743 | #include <ATen/ops/arctan2_native.h> |
| 744 | #include <ATen/ops/arctan2_ops.h> |
| 745 | #include <ATen/ops/histc_native.h> |
| 746 | #include <ATen/ops/histc_ops.h> |
| 747 | #include <ATen/ops/histc_native.h> |
| 748 | #include <ATen/ops/histc_ops.h> |
| 749 | #include <ATen/ops/_histogramdd_from_bin_cts_native.h> |
| 750 | #include <ATen/ops/_histogramdd_from_bin_cts_ops.h> |
| 751 | #include <ATen/ops/_histogramdd_from_bin_cts_native.h> |
| 752 | #include <ATen/ops/_histogramdd_from_bin_cts_ops.h> |
| 753 | #include <ATen/ops/hypot_native.h> |
| 754 | #include <ATen/ops/hypot_ops.h> |
| 755 | #include <ATen/ops/hypot_native.h> |
| 756 | #include <ATen/ops/hypot_ops.h> |
| 757 | #include <ATen/ops/hypot_native.h> |
| 758 | #include <ATen/ops/hypot_ops.h> |
| 759 | #include <ATen/ops/min_native.h> |
| 760 | #include <ATen/ops/min_ops.h> |
| 761 | #include <ATen/ops/min_native.h> |
| 762 | #include <ATen/ops/min_ops.h> |
| 763 | #include <ATen/ops/msort_native.h> |
| 764 | #include <ATen/ops/msort_ops.h> |
| 765 | #include <ATen/ops/msort_native.h> |
| 766 | #include <ATen/ops/msort_ops.h> |
| 767 | #include <ATen/ops/_foreach_add_native.h> |
| 768 | #include <ATen/ops/_foreach_add_ops.h> |
| 769 | #include <ATen/ops/_foreach_add_native.h> |
| 770 | #include <ATen/ops/_foreach_add_ops.h> |
| 771 | #include <ATen/ops/_foreach_add_native.h> |
| 772 | #include <ATen/ops/_foreach_add_ops.h> |
| 773 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 774 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 775 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 776 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 777 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 778 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 779 | #include <ATen/ops/_foreach_add_native.h> |
| 780 | #include <ATen/ops/_foreach_add_ops.h> |
| 781 | #include <ATen/ops/_foreach_add_native.h> |
| 782 | #include <ATen/ops/_foreach_add_ops.h> |
| 783 | #include <ATen/ops/_foreach_add_native.h> |
| 784 | #include <ATen/ops/_foreach_add_ops.h> |
| 785 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 786 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 787 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 788 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 789 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 790 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 791 | #include <ATen/ops/_foreach_add_native.h> |
| 792 | #include <ATen/ops/_foreach_add_ops.h> |
| 793 | #include <ATen/ops/_foreach_add_native.h> |
| 794 | #include <ATen/ops/_foreach_add_ops.h> |
| 795 | #include <ATen/ops/_foreach_add_native.h> |
| 796 | #include <ATen/ops/_foreach_add_ops.h> |
| 797 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 798 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 799 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 800 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 801 | #include <ATen/ops/_foreach_clamp_min_native.h> |
| 802 | #include <ATen/ops/_foreach_clamp_min_ops.h> |
| 803 | #include <ATen/ops/_foreach_zero_native.h> |
| 804 | #include <ATen/ops/_foreach_zero_ops.h> |
| 805 | #include <ATen/ops/_foreach_zero_native.h> |
| 806 | #include <ATen/ops/_foreach_zero_ops.h> |
| 807 | #include <ATen/ops/_foreach_zero_native.h> |
| 808 | #include <ATen/ops/_foreach_zero_ops.h> |
| 809 | #include <ATen/ops/_foreach_asin_native.h> |
| 810 | #include <ATen/ops/_foreach_asin_ops.h> |
| 811 | #include <ATen/ops/_foreach_asin_native.h> |
| 812 | #include <ATen/ops/_foreach_asin_ops.h> |
| 813 | #include <ATen/ops/_foreach_asin_native.h> |
| 814 | #include <ATen/ops/_foreach_asin_ops.h> |
| 815 | #include <ATen/ops/_foreach_ceil_native.h> |
| 816 | #include <ATen/ops/_foreach_ceil_ops.h> |
| 817 | #include <ATen/ops/_foreach_ceil_native.h> |
| 818 | #include <ATen/ops/_foreach_ceil_ops.h> |
| 819 | #include <ATen/ops/_foreach_ceil_native.h> |
| 820 | #include <ATen/ops/_foreach_ceil_ops.h> |
| 821 | #include <ATen/ops/_foreach_cosh_native.h> |
| 822 | #include <ATen/ops/_foreach_cosh_ops.h> |
| 823 | #include <ATen/ops/_foreach_cosh_native.h> |
| 824 | #include <ATen/ops/_foreach_cosh_ops.h> |
| 825 | #include <ATen/ops/_foreach_cosh_native.h> |
| 826 | #include <ATen/ops/_foreach_cosh_ops.h> |
| 827 | #include <ATen/ops/_foreach_log1p_native.h> |
| 828 | #include <ATen/ops/_foreach_log1p_ops.h> |
| 829 | #include <ATen/ops/_foreach_log1p_native.h> |
| 830 | #include <ATen/ops/_foreach_log1p_ops.h> |
| 831 | #include <ATen/ops/_foreach_log1p_native.h> |
| 832 | #include <ATen/ops/_foreach_log1p_ops.h> |
| 833 | #include <ATen/ops/_foreach_log2_native.h> |
| 834 | #include <ATen/ops/_foreach_log2_ops.h> |
| 835 | #include <ATen/ops/_foreach_log2_native.h> |
| 836 | #include <ATen/ops/_foreach_log2_ops.h> |
| 837 | #include <ATen/ops/_foreach_log2_native.h> |
| 838 | #include <ATen/ops/_foreach_log2_ops.h> |
| 839 | #include <ATen/ops/_foreach_round_native.h> |
| 840 | #include <ATen/ops/_foreach_round_ops.h> |
| 841 | #include <ATen/ops/_foreach_round_native.h> |
| 842 | #include <ATen/ops/_foreach_round_ops.h> |
| 843 | #include <ATen/ops/_foreach_round_native.h> |
| 844 | #include <ATen/ops/_foreach_round_ops.h> |
| 845 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 846 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 847 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 848 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 849 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 850 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 851 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 852 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 853 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 854 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 855 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 856 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 857 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 858 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 859 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 860 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 861 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 862 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 863 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 864 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 865 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 866 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 867 | #include <ATen/ops/_foreach_addcdiv_native.h> |
| 868 | #include <ATen/ops/_foreach_addcdiv_ops.h> |
| 869 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 870 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 871 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 872 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 873 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 874 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 875 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 876 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 877 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 878 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 879 | #include <ATen/ops/_foreach_addcmul_native.h> |
| 880 | #include <ATen/ops/_foreach_addcmul_ops.h> |
| 881 | #include <ATen/ops/bucketize_native.h> |
| 882 | #include <ATen/ops/bucketize_ops.h> |
| 883 | #include <ATen/ops/bucketize_native.h> |
| 884 | #include <ATen/ops/bucketize_ops.h> |
| 885 | #include <ATen/ops/bucketize_native.h> |
| 886 | #include <ATen/ops/bucketize_ops.h> |
| 887 | #include <ATen/ops/bucketize_native.h> |
| 888 | #include <ATen/ops/bucketize_ops.h> |
| 889 | #include <ATen/ops/mse_loss_native.h> |
| 890 | #include <ATen/ops/mse_loss_ops.h> |
| 891 | #include <ATen/ops/mse_loss_native.h> |
| 892 | #include <ATen/ops/mse_loss_ops.h> |
| 893 | #include <ATen/ops/multi_margin_loss_native.h> |
| 894 | #include <ATen/ops/multi_margin_loss_ops.h> |
| 895 | #include <ATen/ops/multi_margin_loss_native.h> |
| 896 | #include <ATen/ops/multi_margin_loss_ops.h> |
| 897 | #include <ATen/ops/multilabel_margin_loss_native.h> |
| 898 | #include <ATen/ops/multilabel_margin_loss_ops.h> |
| 899 | #include <ATen/ops/multilabel_margin_loss_native.h> |
| 900 | #include <ATen/ops/multilabel_margin_loss_ops.h> |
| 901 | #include <ATen/ops/multilabel_margin_loss_forward_native.h> |
| 902 | #include <ATen/ops/multilabel_margin_loss_forward_ops.h> |
| 903 | #include <ATen/ops/multilabel_margin_loss_forward_native.h> |
| 904 | #include <ATen/ops/multilabel_margin_loss_forward_ops.h> |
| 905 | #include <ATen/ops/nll_loss2d_forward_native.h> |
| 906 | #include <ATen/ops/nll_loss2d_forward_ops.h> |
| 907 | #include <ATen/ops/nll_loss2d_forward_native.h> |
| 908 | #include <ATen/ops/nll_loss2d_forward_ops.h> |
| 909 | #include <ATen/ops/soft_margin_loss_native.h> |
| 910 | #include <ATen/ops/soft_margin_loss_ops.h> |
| 911 | #include <ATen/ops/soft_margin_loss_native.h> |
| 912 | #include <ATen/ops/soft_margin_loss_ops.h> |
| 913 | #include <ATen/ops/glu_backward_native.h> |
| 914 | #include <ATen/ops/glu_backward_ops.h> |
| 915 | #include <ATen/ops/glu_backward_native.h> |
| 916 | #include <ATen/ops/glu_backward_ops.h> |
| 917 | #include <ATen/ops/glu_backward_jvp_native.h> |
| 918 | #include <ATen/ops/glu_backward_jvp_ops.h> |
| 919 | #include <ATen/ops/glu_backward_jvp_native.h> |
| 920 | #include <ATen/ops/glu_backward_jvp_ops.h> |
| 921 | #include <ATen/ops/hardtanh_native.h> |
| 922 | #include <ATen/ops/hardtanh_ops.h> |
| 923 | #include <ATen/ops/hardtanh_native.h> |
| 924 | #include <ATen/ops/hardtanh_ops.h> |
| 925 | #include <ATen/ops/hardtanh_native.h> |
| 926 | #include <ATen/ops/hardtanh_ops.h> |
| 927 | #include <ATen/ops/hardtanh_backward_native.h> |
| 928 | #include <ATen/ops/hardtanh_backward_ops.h> |
| 929 | #include <ATen/ops/hardtanh_backward_native.h> |
| 930 | #include <ATen/ops/hardtanh_backward_ops.h> |
| 931 | #include <ATen/ops/leaky_relu_native.h> |
| 932 | #include <ATen/ops/leaky_relu_ops.h> |
| 933 | #include <ATen/ops/leaky_relu_native.h> |
| 934 | #include <ATen/ops/leaky_relu_ops.h> |
| 935 | #include <ATen/ops/leaky_relu_native.h> |
| 936 | #include <ATen/ops/leaky_relu_ops.h> |
| 937 | #include <ATen/ops/log_sigmoid_backward_native.h> |
| 938 | #include <ATen/ops/log_sigmoid_backward_ops.h> |
| 939 | #include <ATen/ops/log_sigmoid_backward_native.h> |
| 940 | #include <ATen/ops/log_sigmoid_backward_ops.h> |
| 941 | #include <ATen/ops/softplus_native.h> |
| 942 | #include <ATen/ops/softplus_ops.h> |
| 943 | #include <ATen/ops/softplus_native.h> |
| 944 | #include <ATen/ops/softplus_ops.h> |
| 945 | #include <ATen/ops/adaptive_avg_pool2d_native.h> |
| 946 | #include <ATen/ops/adaptive_avg_pool2d_ops.h> |
| 947 | #include <ATen/ops/adaptive_avg_pool2d_native.h> |
| 948 | #include <ATen/ops/adaptive_avg_pool2d_ops.h> |
| 949 | #include <ATen/ops/_adaptive_avg_pool3d_backward_native.h> |
| 950 | #include <ATen/ops/_adaptive_avg_pool3d_backward_ops.h> |
| 951 | #include <ATen/ops/_adaptive_avg_pool3d_backward_native.h> |
| 952 | #include <ATen/ops/_adaptive_avg_pool3d_backward_ops.h> |
| 953 | #include <ATen/ops/adaptive_max_pool2d_backward_native.h> |
| 954 | #include <ATen/ops/adaptive_max_pool2d_backward_ops.h> |
| 955 | #include <ATen/ops/adaptive_max_pool2d_backward_native.h> |
| 956 | #include <ATen/ops/adaptive_max_pool2d_backward_ops.h> |
| 957 | #include <ATen/ops/fractional_max_pool2d_backward_native.h> |
| 958 | #include <ATen/ops/fractional_max_pool2d_backward_ops.h> |
| 959 | #include <ATen/ops/fractional_max_pool2d_backward_native.h> |
| 960 | #include <ATen/ops/fractional_max_pool2d_backward_ops.h> |
| 961 | #include <ATen/ops/fractional_max_pool3d_backward_native.h> |
| 962 | #include <ATen/ops/fractional_max_pool3d_backward_ops.h> |
| 963 | #include <ATen/ops/fractional_max_pool3d_backward_native.h> |
| 964 | #include <ATen/ops/fractional_max_pool3d_backward_ops.h> |
| 965 | #include <ATen/ops/max_pool3d_with_indices_native.h> |
| 966 | #include <ATen/ops/max_pool3d_with_indices_ops.h> |
| 967 | #include <ATen/ops/max_pool3d_with_indices_native.h> |
| 968 | #include <ATen/ops/max_pool3d_with_indices_ops.h> |
| 969 | #include <ATen/ops/max_pool3d_with_indices_backward_native.h> |
| 970 | #include <ATen/ops/max_pool3d_with_indices_backward_ops.h> |
| 971 | #include <ATen/ops/max_pool3d_with_indices_backward_native.h> |
| 972 | #include <ATen/ops/max_pool3d_with_indices_backward_ops.h> |
| 973 | #include <ATen/ops/max_unpool2d_native.h> |
| 974 | #include <ATen/ops/max_unpool2d_ops.h> |
| 975 | #include <ATen/ops/max_unpool2d_native.h> |
| 976 | #include <ATen/ops/max_unpool2d_ops.h> |
| 977 | #include <ATen/ops/reflection_pad2d_backward_native.h> |
| 978 | #include <ATen/ops/reflection_pad2d_backward_ops.h> |
| 979 | #include <ATen/ops/reflection_pad2d_backward_native.h> |
| 980 | #include <ATen/ops/reflection_pad2d_backward_ops.h> |
| 981 | #include <ATen/ops/upsample_bilinear2d_native.h> |
| 982 | #include <ATen/ops/upsample_bilinear2d_ops.h> |
| 983 | #include <ATen/ops/upsample_bilinear2d_native.h> |
| 984 | #include <ATen/ops/upsample_bilinear2d_ops.h> |
| 985 | #include <ATen/ops/upsample_bilinear2d_backward_native.h> |
| 986 | #include <ATen/ops/upsample_bilinear2d_backward_ops.h> |
| 987 | #include <ATen/ops/upsample_bilinear2d_backward_native.h> |
| 988 | #include <ATen/ops/upsample_bilinear2d_backward_ops.h> |
| 989 | #include <ATen/ops/_upsample_bilinear2d_aa_native.h> |
| 990 | #include <ATen/ops/_upsample_bilinear2d_aa_ops.h> |
| 991 | #include <ATen/ops/_upsample_bilinear2d_aa_native.h> |
| 992 | #include <ATen/ops/_upsample_bilinear2d_aa_ops.h> |
| 993 | #include <ATen/ops/upsample_trilinear3d_backward_native.h> |
| 994 | #include <ATen/ops/upsample_trilinear3d_backward_ops.h> |
| 995 | #include <ATen/ops/upsample_trilinear3d_backward_native.h> |
| 996 | #include <ATen/ops/upsample_trilinear3d_backward_ops.h> |
| 997 | #include <ATen/ops/_upsample_nearest_exact3d_native.h> |
| 998 | #include <ATen/ops/_upsample_nearest_exact3d_ops.h> |
| 999 | #include <ATen/ops/_upsample_nearest_exact3d_native.h> |
| 1000 | #include <ATen/ops/_upsample_nearest_exact3d_ops.h> |
| 1001 | #include <ATen/ops/upsample_nearest3d_backward_native.h> |
| 1002 | #include <ATen/ops/upsample_nearest3d_backward_ops.h> |
| 1003 | #include <ATen/ops/upsample_nearest3d_backward_native.h> |
| 1004 | #include <ATen/ops/upsample_nearest3d_backward_ops.h> |
| 1005 | #include <ATen/ops/logit_backward_native.h> |
| 1006 | #include <ATen/ops/logit_backward_ops.h> |
| 1007 | #include <ATen/ops/logit_backward_native.h> |
| 1008 | #include <ATen/ops/logit_backward_ops.h> |
| 1009 | #include <ATen/ops/thnn_conv2d_native.h> |
| 1010 | #include <ATen/ops/thnn_conv2d_ops.h> |
| 1011 | #include <ATen/ops/thnn_conv2d_native.h> |
| 1012 | #include <ATen/ops/thnn_conv2d_ops.h> |
| 1013 | #include <ATen/ops/_slow_conv2d_backward_native.h> |
| 1014 | #include <ATen/ops/_slow_conv2d_backward_ops.h> |
| 1015 | #include <ATen/ops/_slow_conv2d_backward_native.h> |
| 1016 | #include <ATen/ops/_slow_conv2d_backward_ops.h> |
| 1017 | #include <ATen/ops/slow_conv3d_native.h> |
| 1018 | #include <ATen/ops/slow_conv3d_ops.h> |
| 1019 | #include <ATen/ops/slow_conv3d_native.h> |
| 1020 | #include <ATen/ops/slow_conv3d_ops.h> |
| 1021 | #include <ATen/ops/slow_conv3d_forward_native.h> |
| 1022 | #include <ATen/ops/slow_conv3d_forward_ops.h> |
| 1023 | #include <ATen/ops/slow_conv3d_forward_native.h> |
| 1024 | #include <ATen/ops/slow_conv3d_forward_ops.h> |
| 1025 | #include <ATen/ops/slow_conv_dilated3d_native.h> |
| 1026 | #include <ATen/ops/slow_conv_dilated3d_ops.h> |
| 1027 | #include <ATen/ops/slow_conv_dilated3d_native.h> |
| 1028 | #include <ATen/ops/slow_conv_dilated3d_ops.h> |
| 1029 | #include <ATen/ops/special_log_ndtr_native.h> |
| 1030 | #include <ATen/ops/special_log_ndtr_ops.h> |
| 1031 | #include <ATen/ops/special_log_ndtr_native.h> |
| 1032 | #include <ATen/ops/special_log_ndtr_ops.h> |
| 1033 | #include <ATen/ops/special_exp2_native.h> |
| 1034 | #include <ATen/ops/special_exp2_ops.h> |
| 1035 | #include <ATen/ops/special_exp2_native.h> |
| 1036 | #include <ATen/ops/special_exp2_ops.h> |
| 1037 | #include <ATen/ops/special_digamma_native.h> |
| 1038 | #include <ATen/ops/special_digamma_ops.h> |
| 1039 | #include <ATen/ops/special_digamma_native.h> |
| 1040 | #include <ATen/ops/special_digamma_ops.h> |
| 1041 | #include <ATen/ops/special_gammaln_native.h> |
| 1042 | #include <ATen/ops/special_gammaln_ops.h> |
| 1043 | #include <ATen/ops/special_gammaln_native.h> |
| 1044 | #include <ATen/ops/special_gammaln_ops.h> |
| 1045 | #include <ATen/ops/special_erfcx_native.h> |
| 1046 | #include <ATen/ops/special_erfcx_ops.h> |
| 1047 | #include <ATen/ops/special_erfcx_native.h> |
| 1048 | #include <ATen/ops/special_erfcx_ops.h> |
| 1049 | #include <ATen/ops/special_xlog1py_native.h> |
| 1050 | #include <ATen/ops/special_xlog1py_ops.h> |
| 1051 | #include <ATen/ops/special_xlog1py_native.h> |
| 1052 | #include <ATen/ops/special_xlog1py_ops.h> |
| 1053 | #include <ATen/ops/special_xlog1py_native.h> |
| 1054 | #include <ATen/ops/special_xlog1py_ops.h> |
| 1055 | #include <ATen/ops/special_xlog1py_native.h> |
| 1056 | #include <ATen/ops/special_xlog1py_ops.h> |
| 1057 | #include <ATen/ops/special_xlog1py_native.h> |
| 1058 | #include <ATen/ops/special_xlog1py_ops.h> |
| 1059 | #include <ATen/ops/special_xlog1py_native.h> |
| 1060 | #include <ATen/ops/special_xlog1py_ops.h> |
| 1061 | #include <ATen/ops/special_i1_native.h> |
| 1062 | #include <ATen/ops/special_i1_ops.h> |
| 1063 | #include <ATen/ops/special_i1_native.h> |
| 1064 | #include <ATen/ops/special_i1_ops.h> |
| 1065 | #include <ATen/ops/special_i1e_native.h> |
| 1066 | #include <ATen/ops/special_i1e_ops.h> |
| 1067 | #include <ATen/ops/special_i1e_native.h> |
| 1068 | #include <ATen/ops/special_i1e_ops.h> |
| 1069 | #include <ATen/ops/fft_fft_native.h> |
| 1070 | #include <ATen/ops/fft_fft_ops.h> |
| 1071 | #include <ATen/ops/fft_fft_native.h> |
| 1072 | #include <ATen/ops/fft_fft_ops.h> |
| 1073 | #include <ATen/ops/fft_rfft_native.h> |
| 1074 | #include <ATen/ops/fft_rfft_ops.h> |
| 1075 | #include <ATen/ops/fft_rfft_native.h> |
| 1076 | #include <ATen/ops/fft_rfft_ops.h> |
| 1077 | #include <ATen/ops/fft_hfft_native.h> |
| 1078 | #include <ATen/ops/fft_hfft_ops.h> |
| 1079 | #include <ATen/ops/fft_hfft_native.h> |
| 1080 | #include <ATen/ops/fft_hfft_ops.h> |
| 1081 | #include <ATen/ops/fft_hfft2_native.h> |
| 1082 | #include <ATen/ops/fft_hfft2_ops.h> |
| 1083 | #include <ATen/ops/fft_hfft2_native.h> |
| 1084 | #include <ATen/ops/fft_hfft2_ops.h> |
| 1085 | #include <ATen/ops/fft_ifftn_native.h> |
| 1086 | #include <ATen/ops/fft_ifftn_ops.h> |
| 1087 | #include <ATen/ops/fft_ifftn_native.h> |
| 1088 | #include <ATen/ops/fft_ifftn_ops.h> |
| 1089 | #include <ATen/ops/fft_rfftn_native.h> |
| 1090 | #include <ATen/ops/fft_rfftn_ops.h> |
| 1091 | #include <ATen/ops/fft_rfftn_native.h> |
| 1092 | #include <ATen/ops/fft_rfftn_ops.h> |
| 1093 | #include <ATen/ops/fft_hfftn_native.h> |
| 1094 | #include <ATen/ops/fft_hfftn_ops.h> |
| 1095 | #include <ATen/ops/fft_hfftn_native.h> |
| 1096 | #include <ATen/ops/fft_hfftn_ops.h> |
| 1097 | #include <ATen/ops/fft_fftfreq_native.h> |
| 1098 | #include <ATen/ops/fft_fftfreq_ops.h> |
| 1099 | #include <ATen/ops/fft_fftfreq_native.h> |
| 1100 | #include <ATen/ops/fft_fftfreq_ops.h> |
| 1101 | #include <ATen/ops/linalg_det_native.h> |
| 1102 | #include <ATen/ops/linalg_det_ops.h> |
| 1103 | #include <ATen/ops/linalg_det_native.h> |
| 1104 | #include <ATen/ops/linalg_det_ops.h> |
| 1105 | #include <ATen/ops/linalg_ldl_factor_ex_native.h> |
| 1106 | #include <ATen/ops/linalg_ldl_factor_ex_ops.h> |
| 1107 | #include <ATen/ops/linalg_ldl_factor_ex_native.h> |
| 1108 | #include <ATen/ops/linalg_ldl_factor_ex_ops.h> |
| 1109 | #include <ATen/ops/linalg_lstsq_native.h> |
| 1110 | #include <ATen/ops/linalg_lstsq_ops.h> |
| 1111 | #include <ATen/ops/linalg_lstsq_native.h> |
| 1112 | #include <ATen/ops/linalg_lstsq_ops.h> |
| 1113 | #include <ATen/ops/linalg_matrix_exp_native.h> |
| 1114 | #include <ATen/ops/linalg_matrix_exp_ops.h> |
| 1115 | #include <ATen/ops/linalg_matrix_exp_native.h> |
| 1116 | #include <ATen/ops/linalg_matrix_exp_ops.h> |
| 1117 | #include <ATen/ops/linalg_slogdet_native.h> |
| 1118 | #include <ATen/ops/linalg_slogdet_ops.h> |
| 1119 | #include <ATen/ops/linalg_slogdet_native.h> |
| 1120 | #include <ATen/ops/linalg_slogdet_ops.h> |
| 1121 | #include <ATen/ops/_linalg_eigh_native.h> |
| 1122 | #include <ATen/ops/_linalg_eigh_ops.h> |
| 1123 | #include <ATen/ops/_linalg_eigh_native.h> |
| 1124 | #include <ATen/ops/_linalg_eigh_ops.h> |
| 1125 | #include <ATen/ops/inner_native.h> |
| 1126 | #include <ATen/ops/inner_ops.h> |
| 1127 | #include <ATen/ops/inner_native.h> |
| 1128 | #include <ATen/ops/inner_ops.h> |
| 1129 | #include <ATen/ops/linalg_matrix_norm_native.h> |
| 1130 | #include <ATen/ops/linalg_matrix_norm_ops.h> |
| 1131 | #include <ATen/ops/linalg_matrix_norm_native.h> |
| 1132 | #include <ATen/ops/linalg_matrix_norm_ops.h> |
| 1133 | #include <ATen/ops/linalg_matrix_norm_native.h> |
| 1134 | #include <ATen/ops/linalg_matrix_norm_ops.h> |
| 1135 | #include <ATen/ops/linalg_matrix_norm_native.h> |
| 1136 | #include <ATen/ops/linalg_matrix_norm_ops.h> |
| 1137 | #include <ATen/ops/linalg_tensorinv_native.h> |
| 1138 | #include <ATen/ops/linalg_tensorinv_ops.h> |
| 1139 | #include <ATen/ops/linalg_tensorinv_native.h> |
| 1140 | #include <ATen/ops/linalg_tensorinv_ops.h> |
| 1141 | #include <ATen/ops/linalg_matrix_power_native.h> |
| 1142 | #include <ATen/ops/linalg_matrix_power_ops.h> |
| 1143 | #include <ATen/ops/linalg_matrix_power_native.h> |
| 1144 | #include <ATen/ops/linalg_matrix_power_ops.h> |
| 1145 | #include <ATen/ops/_make_dual_copy_native.h> |
| 1146 | #include <ATen/ops/_make_dual_copy_ops.h> |
| 1147 | #include <ATen/ops/_make_dual_copy_native.h> |
| 1148 | #include <ATen/ops/_make_dual_copy_ops.h> |
| 1149 | #include <ATen/ops/view_as_real_copy_native.h> |
| 1150 | #include <ATen/ops/view_as_real_copy_ops.h> |
| 1151 | #include <ATen/ops/view_as_real_copy_native.h> |
| 1152 | #include <ATen/ops/view_as_real_copy_ops.h> |
| 1153 | #include <ATen/ops/view_as_complex_copy_native.h> |
| 1154 | #include <ATen/ops/view_as_complex_copy_ops.h> |
| 1155 | #include <ATen/ops/view_as_complex_copy_native.h> |
| 1156 | #include <ATen/ops/view_as_complex_copy_ops.h> |
| 1157 | #include <ATen/ops/_conj_copy_native.h> |
| 1158 | #include <ATen/ops/_conj_copy_ops.h> |
| 1159 | #include <ATen/ops/_conj_copy_native.h> |
| 1160 | #include <ATen/ops/_conj_copy_ops.h> |
| 1161 | #include <ATen/ops/_neg_view_copy_native.h> |
| 1162 | #include <ATen/ops/_neg_view_copy_ops.h> |
| 1163 | #include <ATen/ops/_neg_view_copy_native.h> |
| 1164 | #include <ATen/ops/_neg_view_copy_ops.h> |
| 1165 | #include <ATen/ops/permute_copy_native.h> |
| 1166 | #include <ATen/ops/permute_copy_ops.h> |
| 1167 | #include <ATen/ops/permute_copy_native.h> |
| 1168 | #include <ATen/ops/permute_copy_ops.h> |
| 1169 | #include <ATen/ops/split_copy_native.h> |
| 1170 | #include <ATen/ops/split_copy_ops.h> |
| 1171 | #include <ATen/ops/split_copy_native.h> |
| 1172 | #include <ATen/ops/split_copy_ops.h> |
| 1173 | #include <ATen/ops/t_copy_native.h> |
| 1174 | #include <ATen/ops/t_copy_ops.h> |
| 1175 | #include <ATen/ops/t_copy_native.h> |
| 1176 | #include <ATen/ops/t_copy_ops.h> |
| 1177 | #include <ATen/ops/ccol_indices_copy_native.h> |
| 1178 | #include <ATen/ops/ccol_indices_copy_ops.h> |
| 1179 | #include <ATen/ops/ccol_indices_copy_native.h> |
| 1180 | #include <ATen/ops/ccol_indices_copy_ops.h> |
| 1181 | #include <ATen/ops/unfold_copy_native.h> |
| 1182 | #include <ATen/ops/unfold_copy_ops.h> |
| 1183 | #include <ATen/ops/unfold_copy_native.h> |
| 1184 | #include <ATen/ops/unfold_copy_ops.h> |
| 1185 | #include <ATen/ops/_transformer_encoder_layer_fwd_native.h> |
| 1186 | #include <ATen/ops/_transformer_encoder_layer_fwd_ops.h> |
| 1187 | #include <ATen/ops/_transformer_encoder_layer_fwd_native.h> |
| 1188 | #include <ATen/ops/_transformer_encoder_layer_fwd_ops.h> |
| 1189 | #include <ATen/ops/_native_multi_head_attention_native.h> |
| 1190 | #include <ATen/ops/_native_multi_head_attention_ops.h> |
| 1191 | #include <ATen/ops/_native_multi_head_attention_native.h> |
| 1192 | #include <ATen/ops/_native_multi_head_attention_ops.h> |
| 1193 | #include <ATen/ops/_transformer_decoder_only_layer_fwd_native.h> |
| 1194 | #include <ATen/ops/_transformer_decoder_only_layer_fwd_ops.h> |
| 1195 | #include <ATen/ops/_transformer_decoder_only_layer_fwd_native.h> |
| 1196 | #include <ATen/ops/_transformer_decoder_only_layer_fwd_ops.h> |
| 1197 | #include <ATen/ops/_native_decoder_only_multi_head_attention_native.h> |
| 1198 | #include <ATen/ops/_native_decoder_only_multi_head_attention_ops.h> |
| 1199 | #include <ATen/ops/_native_decoder_only_multi_head_attention_native.h> |
| 1200 | #include <ATen/ops/_native_decoder_only_multi_head_attention_ops.h> |
| 1201 | #include <ATen/ops/special_modified_bessel_i1_native.h> |
| 1202 | #include <ATen/ops/special_modified_bessel_i1_ops.h> |
| 1203 | #include <ATen/ops/special_modified_bessel_i1_native.h> |
| 1204 | #include <ATen/ops/special_modified_bessel_i1_ops.h> |
| 1205 | #include <ATen/ops/special_modified_bessel_k0_native.h> |
| 1206 | #include <ATen/ops/special_modified_bessel_k0_ops.h> |
| 1207 | #include <ATen/ops/special_modified_bessel_k0_native.h> |
| 1208 | #include <ATen/ops/special_modified_bessel_k0_ops.h> |
| 1209 | #include <ATen/ops/_foobar_native.h> |
| 1210 | #include <ATen/ops/_foobar_ops.h> |
| 1211 | #include <ATen/ops/_foobar_native.h> |
| 1212 | #include <ATen/ops/_foobar_ops.h> |
| 1213 | #include <ATen/ops/view_as_real_native.h> |
| 1214 | #include <ATen/ops/view_as_real_ops.h> |
| 1215 | #include <ATen/ops/view_as_real_copy_native.h> |
| 1216 | #include <ATen/ops/view_as_real_copy_ops.h> |
| 1217 | #include <ATen/ops/broadcast_to_native.h> |
| 1218 | #include <ATen/ops/broadcast_to_ops.h> |
| 1219 | #include <ATen/ops/contiguous_native.h> |
| 1220 | #include <ATen/ops/contiguous_ops.h> |
| 1221 | #include <ATen/ops/flatten_native.h> |
| 1222 | #include <ATen/ops/flatten_ops.h> |
| 1223 | #include <ATen/ops/flatten_native.h> |
| 1224 | #include <ATen/ops/flatten_ops.h> |
| 1225 | #include <ATen/ops/flatten_native.h> |
| 1226 | #include <ATen/ops/flatten_ops.h> |
| 1227 | #include <ATen/ops/flatten_native.h> |
| 1228 | #include <ATen/ops/flatten_ops.h> |
| 1229 | #include <ATen/ops/moveaxis_native.h> |
| 1230 | #include <ATen/ops/moveaxis_ops.h> |
| 1231 | #include <ATen/ops/moveaxis_native.h> |
| 1232 | #include <ATen/ops/moveaxis_ops.h> |
| 1233 | #include <ATen/ops/mT_native.h> |
| 1234 | #include <ATen/ops/mT_ops.h> |
| 1235 | #include <ATen/ops/ravel_native.h> |
| 1236 | #include <ATen/ops/ravel_ops.h> |
| 1237 | #include <ATen/ops/reshape_native.h> |
| 1238 | #include <ATen/ops/reshape_ops.h> |
| 1239 | #include <ATen/ops/slice_native.h> |
| 1240 | #include <ATen/ops/slice_ops.h> |
| 1241 | #include <ATen/ops/slice_copy_native.h> |
| 1242 | #include <ATen/ops/slice_copy_ops.h> |
| 1243 | #include <ATen/ops/hsplit_native.h> |
| 1244 | #include <ATen/ops/hsplit_ops.h> |
| 1245 | #include <ATen/ops/hsplit_native.h> |
| 1246 | #include <ATen/ops/hsplit_ops.h> |
| 1247 | #include <ATen/ops/t_native.h> |
| 1248 | #include <ATen/ops/t_ops.h> |
| 1249 | #include <ATen/ops/t_copy_native.h> |
| 1250 | #include <ATen/ops/t_copy_ops.h> |
| 1251 | #include <ATen/ops/coalesce_native.h> |
| 1252 | #include <ATen/ops/coalesce_ops.h> |
| 1253 | #include <ATen/ops/_indices_native.h> |
| 1254 | #include <ATen/ops/_indices_ops.h> |
| 1255 | #include <ATen/ops/_indices_copy_native.h> |
| 1256 | #include <ATen/ops/_indices_copy_ops.h> |
| 1257 | #include <ATen/ops/indices_native.h> |
| 1258 | #include <ATen/ops/indices_ops.h> |
| 1259 | #include <ATen/ops/indices_copy_native.h> |
| 1260 | #include <ATen/ops/indices_copy_ops.h> |
| 1261 | #include <ATen/ops/crow_indices_native.h> |
| 1262 | #include <ATen/ops/crow_indices_ops.h> |
| 1263 | #include <ATen/ops/crow_indices_copy_native.h> |
| 1264 | #include <ATen/ops/crow_indices_copy_ops.h> |
| 1265 | #include <ATen/ops/col_indices_native.h> |
| 1266 | #include <ATen/ops/col_indices_ops.h> |
| 1267 | #include <ATen/ops/col_indices_copy_native.h> |
| 1268 | #include <ATen/ops/col_indices_copy_ops.h> |
| 1269 | #include <ATen/ops/_autocast_to_reduced_precision_native.h> |
| 1270 | #include <ATen/ops/_autocast_to_reduced_precision_ops.h> |
| 1271 | #include <ATen/ops/swapaxes_native.h> |
| 1272 | #include <ATen/ops/swapaxes_ops.h> |
| 1273 | #include <ATen/ops/unfold_native.h> |
| 1274 | #include <ATen/ops/unfold_ops.h> |
| 1275 | #include <ATen/ops/unfold_copy_native.h> |
| 1276 | #include <ATen/ops/unfold_copy_ops.h> |
| 1277 | #include <ATen/ops/_cast_Byte_native.h> |
| 1278 | #include <ATen/ops/_cast_Byte_ops.h> |
| 1279 | #include <ATen/ops/_cast_Float_native.h> |
| 1280 | #include <ATen/ops/_cast_Float_ops.h> |
| 1281 | #include <ATen/ops/is_leaf_native.h> |
| 1282 | #include <ATen/ops/is_leaf_ops.h> |
| 1283 | #include <ATen/ops/_assert_async_native.h> |
| 1284 | #include <ATen/ops/_assert_async_ops.h> |
| 1285 | #include <ATen/ops/_use_cudnn_ctc_loss_native.h> |
| 1286 | #include <ATen/ops/_use_cudnn_ctc_loss_ops.h> |
| 1287 | #include <ATen/ops/_use_cudnn_ctc_loss_native.h> |
| 1288 | #include <ATen/ops/_use_cudnn_ctc_loss_ops.h> |
| 1289 | #include <ATen/ops/_reshape_from_tensor_native.h> |
| 1290 | #include <ATen/ops/_reshape_from_tensor_ops.h> |
| 1291 | #include <ATen/ops/dropout_native.h> |
| 1292 | #include <ATen/ops/dropout_ops.h> |
| 1293 | #include <ATen/ops/dropout_native.h> |
| 1294 | #include <ATen/ops/dropout_ops.h> |
| 1295 | #include <ATen/ops/adaptive_max_pool1d_native.h> |
| 1296 | #include <ATen/ops/adaptive_max_pool1d_ops.h> |
| 1297 | #include <ATen/ops/_is_any_true_native.h> |
| 1298 | #include <ATen/ops/_is_any_true_ops.h> |
| 1299 | #include <ATen/ops/arange_native.h> |
| 1300 | #include <ATen/ops/arange_ops.h> |
| 1301 | #include <ATen/ops/atleast_2d_native.h> |
| 1302 | #include <ATen/ops/atleast_2d_ops.h> |
| 1303 | #include <ATen/ops/atleast_2d_native.h> |
| 1304 | #include <ATen/ops/atleast_2d_ops.h> |
| 1305 | #include <ATen/ops/_batch_norm_impl_index_backward_native.h> |
| 1306 | #include <ATen/ops/_batch_norm_impl_index_backward_ops.h> |
| 1307 | #include <ATen/ops/_convolution_double_backward_native.h> |
| 1308 | #include <ATen/ops/_convolution_double_backward_ops.h> |
| 1309 | #include <ATen/ops/conv1d_native.h> |
| 1310 | #include <ATen/ops/conv1d_ops.h> |
| 1311 | #include <ATen/ops/conv1d_native.h> |
| 1312 | #include <ATen/ops/conv1d_ops.h> |
| 1313 | #include <ATen/ops/corrcoef_native.h> |
| 1314 | #include <ATen/ops/corrcoef_ops.h> |
| 1315 | #include <ATen/ops/cummaxmin_backward_native.h> |
| 1316 | #include <ATen/ops/cummaxmin_backward_ops.h> |
| 1317 | #include <ATen/ops/cumprod_backward_native.h> |
| 1318 | #include <ATen/ops/cumprod_backward_ops.h> |
| 1319 | #include <ATen/ops/embedding_sparse_backward_native.h> |
| 1320 | #include <ATen/ops/embedding_sparse_backward_ops.h> |
| 1321 | #include <ATen/ops/_rowwise_prune_native.h> |
| 1322 | #include <ATen/ops/_rowwise_prune_ops.h> |
| 1323 | #include <ATen/ops/_embedding_bag_sparse_backward_native.h> |
| 1324 | #include <ATen/ops/_embedding_bag_sparse_backward_ops.h> |
| 1325 | #include <ATen/ops/floor_divide_native.h> |
| 1326 | #include <ATen/ops/floor_divide_ops.h> |
| 1327 | #include <ATen/ops/floor_divide_native.h> |
| 1328 | #include <ATen/ops/floor_divide_ops.h> |
| 1329 | #include <ATen/ops/group_norm_native.h> |
| 1330 | #include <ATen/ops/group_norm_ops.h> |
| 1331 | #include <ATen/ops/_validate_compressed_sparse_indices_native.h> |
| 1332 | #include <ATen/ops/_validate_compressed_sparse_indices_ops.h> |
| 1333 | #include <ATen/ops/is_distributed_native.h> |
| 1334 | #include <ATen/ops/is_distributed_ops.h> |
| 1335 | #include <ATen/ops/is_neg_native.h> |
| 1336 | #include <ATen/ops/is_neg_ops.h> |
| 1337 | #include <ATen/ops/is_same_size_native.h> |
| 1338 | #include <ATen/ops/is_same_size_ops.h> |
| 1339 | #include <ATen/ops/is_signed_native.h> |
| 1340 | #include <ATen/ops/is_signed_ops.h> |
| 1341 | #include <ATen/ops/fbgemm_linear_fp16_weight_fp32_activation_native.h> |
| 1342 | #include <ATen/ops/fbgemm_linear_fp16_weight_fp32_activation_ops.h> |
| 1343 | #include <ATen/ops/matrix_exp_backward_native.h> |
| 1344 | #include <ATen/ops/matrix_exp_backward_ops.h> |
| 1345 | #include <ATen/ops/max_pool2d_native.h> |
| 1346 | #include <ATen/ops/max_pool2d_ops.h> |
| 1347 | #include <ATen/ops/miopen_convolution_add_relu_native.h> |
| 1348 | #include <ATen/ops/miopen_convolution_add_relu_ops.h> |
| 1349 | #include <ATen/ops/cosine_similarity_native.h> |
| 1350 | #include <ATen/ops/cosine_similarity_ops.h> |
| 1351 | #include <ATen/ops/native_channel_shuffle_native.h> |
| 1352 | #include <ATen/ops/native_channel_shuffle_ops.h> |
| 1353 | #include <ATen/ops/pinverse_native.h> |
| 1354 | #include <ATen/ops/pinverse_ops.h> |
| 1355 | #include <ATen/ops/rrelu_native.h> |
| 1356 | #include <ATen/ops/rrelu_ops.h> |
| 1357 | #include <ATen/ops/rrelu_native.h> |
| 1358 | #include <ATen/ops/rrelu_ops.h> |
| 1359 | #include <ATen/ops/prelu_native.h> |
| 1360 | #include <ATen/ops/prelu_ops.h> |
| 1361 | #include <ATen/ops/softmax_native.h> |
| 1362 | #include <ATen/ops/softmax_ops.h> |
| 1363 | #include <ATen/ops/stft_native.h> |
| 1364 | #include <ATen/ops/stft_ops.h> |
| 1365 | #include <ATen/ops/stft_native.h> |
| 1366 | #include <ATen/ops/stft_ops.h> |
| 1367 | #include <ATen/ops/tile_native.h> |
| 1368 | #include <ATen/ops/tile_ops.h> |
| 1369 | #include <ATen/ops/trapezoid_native.h> |
| 1370 | #include <ATen/ops/trapezoid_ops.h> |
| 1371 | #include <ATen/ops/trapezoid_native.h> |
| 1372 | #include <ATen/ops/trapezoid_ops.h> |
| 1373 | #include <ATen/ops/trapz_native.h> |
| 1374 | #include <ATen/ops/trapz_ops.h> |
| 1375 | #include <ATen/ops/trapz_native.h> |
| 1376 | #include <ATen/ops/trapz_ops.h> |
| 1377 | #include <ATen/ops/_has_compatible_shallow_copy_type_native.h> |
| 1378 | #include <ATen/ops/_has_compatible_shallow_copy_type_ops.h> |
| 1379 | #include <ATen/ops/vander_native.h> |
| 1380 | #include <ATen/ops/vander_ops.h> |
| 1381 | #include <ATen/ops/where_native.h> |
| 1382 | #include <ATen/ops/where_ops.h> |
| 1383 | #include <ATen/ops/where_native.h> |
| 1384 | #include <ATen/ops/where_ops.h> |
| 1385 | #include <ATen/ops/where_native.h> |
| 1386 | #include <ATen/ops/where_ops.h> |
| 1387 | #include <ATen/ops/where_native.h> |
| 1388 | #include <ATen/ops/where_ops.h> |
| 1389 | #include <ATen/ops/_weight_norm_native.h> |
| 1390 | #include <ATen/ops/_weight_norm_ops.h> |
| 1391 | #include <ATen/ops/_weight_norm_differentiable_backward_native.h> |
| 1392 | #include <ATen/ops/_weight_norm_differentiable_backward_ops.h> |
| 1393 | #include <ATen/ops/_sparse_sum_native.h> |
| 1394 | #include <ATen/ops/_sparse_sum_ops.h> |
| 1395 | #include <ATen/ops/_sparse_sum_native.h> |
| 1396 | #include <ATen/ops/_sparse_sum_ops.h> |
| 1397 | #include <ATen/ops/_sparse_sum_native.h> |
| 1398 | #include <ATen/ops/_sparse_sum_ops.h> |
| 1399 | #include <ATen/ops/_sparse_softmax_native.h> |
| 1400 | #include <ATen/ops/_sparse_softmax_ops.h> |
| 1401 | #include <ATen/ops/_sparse_softmax_native.h> |
| 1402 | #include <ATen/ops/_sparse_softmax_ops.h> |
| 1403 | #include <ATen/ops/_sparse_mm_reduce_impl_native.h> |
| 1404 | #include <ATen/ops/_sparse_mm_reduce_impl_ops.h> |
| 1405 | #include <ATen/ops/sparse_csr_tensor_native.h> |
| 1406 | #include <ATen/ops/sparse_csr_tensor_ops.h> |
| 1407 | #include <ATen/ops/sparse_bsc_tensor_native.h> |
| 1408 | #include <ATen/ops/sparse_bsc_tensor_ops.h> |
| 1409 | #include <ATen/ops/sparse_csr_tensor_native.h> |
| 1410 | #include <ATen/ops/sparse_csr_tensor_ops.h> |
| 1411 | #include <ATen/ops/sparse_bsc_tensor_native.h> |
| 1412 | #include <ATen/ops/sparse_bsc_tensor_ops.h> |
| 1413 | #include <ATen/ops/_sparse_csr_tensor_unsafe_native.h> |
| 1414 | #include <ATen/ops/_sparse_csr_tensor_unsafe_ops.h> |
| 1415 | #include <ATen/ops/_sparse_bsc_tensor_unsafe_native.h> |
| 1416 | #include <ATen/ops/_sparse_bsc_tensor_unsafe_ops.h> |
| 1417 | #include <ATen/ops/_validate_sparse_bsr_tensor_args_native.h> |
| 1418 | #include <ATen/ops/_validate_sparse_bsr_tensor_args_ops.h> |
| 1419 | #include <ATen/ops/sparse_dim_native.h> |
| 1420 | #include <ATen/ops/sparse_dim_ops.h> |
| 1421 | #include <ATen/ops/is_coalesced_native.h> |
| 1422 | #include <ATen/ops/is_coalesced_ops.h> |
| 1423 | #include <ATen/ops/q_scale_native.h> |
| 1424 | #include <ATen/ops/q_scale_ops.h> |
| 1425 | #include <ATen/ops/_saturate_weight_to_fp16_native.h> |
| 1426 | #include <ATen/ops/_saturate_weight_to_fp16_ops.h> |
| 1427 | #include <ATen/ops/item_native.h> |
| 1428 | #include <ATen/ops/item_ops.h> |
| 1429 | #include <ATen/ops/_thnn_fused_lstm_cell_backward_native.h> |
| 1430 | #include <ATen/ops/_thnn_fused_lstm_cell_backward_ops.h> |
| 1431 | #include <ATen/ops/_thnn_differentiable_lstm_cell_backward_native.h> |
| 1432 | #include <ATen/ops/_thnn_differentiable_lstm_cell_backward_ops.h> |
| 1433 | #include <ATen/ops/rnn_relu_native.h> |
| 1434 | #include <ATen/ops/rnn_relu_ops.h> |
| 1435 | #include <ATen/ops/rnn_relu_native.h> |
| 1436 | #include <ATen/ops/rnn_relu_ops.h> |
| 1437 | #include <ATen/ops/quantized_lstm_cell_native.h> |
| 1438 | #include <ATen/ops/quantized_lstm_cell_ops.h> |
| 1439 | #include <ATen/ops/quantized_rnn_relu_cell_native.h> |
| 1440 | #include <ATen/ops/quantized_rnn_relu_cell_ops.h> |
| 1441 | #include <ATen/ops/_pack_padded_sequence_backward_native.h> |
| 1442 | #include <ATen/ops/_pack_padded_sequence_backward_ops.h> |
| 1443 | #include <ATen/ops/histogramdd_native.h> |
| 1444 | #include <ATen/ops/histogramdd_ops.h> |
| 1445 | #include <ATen/ops/histogramdd_native.h> |
| 1446 | #include <ATen/ops/histogramdd_ops.h> |
| 1447 | #include <ATen/ops/histogramdd_native.h> |
| 1448 | #include <ATen/ops/histogramdd_ops.h> |
| 1449 | #include <ATen/ops/min_native.h> |
| 1450 | #include <ATen/ops/min_ops.h> |
| 1451 | #include <ATen/ops/l1_loss_native.h> |
| 1452 | #include <ATen/ops/l1_loss_ops.h> |
| 1453 | #include <ATen/ops/upsample_bilinear2d_native.h> |
| 1454 | #include <ATen/ops/upsample_bilinear2d_ops.h> |
| 1455 | #include <ATen/ops/_upsample_bilinear2d_aa_native.h> |
| 1456 | #include <ATen/ops/_upsample_bilinear2d_aa_ops.h> |
| 1457 | #include <ATen/ops/_upsample_nearest_exact3d_native.h> |
| 1458 | #include <ATen/ops/_upsample_nearest_exact3d_ops.h> |
| 1459 | #include <ATen/ops/_slow_conv2d_backward_native.h> |
| 1460 | #include <ATen/ops/_slow_conv2d_backward_ops.h> |
| 1461 | #include <ATen/ops/isfinite_native.h> |
| 1462 | #include <ATen/ops/isfinite_ops.h> |
| 1463 | #include <ATen/ops/special_log_softmax_native.h> |
| 1464 | #include <ATen/ops/special_log_softmax_ops.h> |
| 1465 | #include <ATen/ops/_nested_tensor_softmax_with_shape_native.h> |
| 1466 | #include <ATen/ops/_nested_tensor_softmax_with_shape_ops.h> |
| 1467 | #include <ATen/ops/scaled_dot_product_attention_native.h> |
| 1468 | #include <ATen/ops/scaled_dot_product_attention_ops.h> |
| 1469 | #include <ATen/ops/_fused_sdp_choice_native.h> |
| 1470 | #include <ATen/ops/_fused_sdp_choice_ops.h> |
| 1471 | #include <ATen/ops/_scaled_dot_product_attention_math_native.h> |
| 1472 | #include <ATen/ops/_scaled_dot_product_attention_math_ops.h> |
| 1473 | #endif |
| 1474 | |
| 1475 | namespace at { |
| 1476 | namespace functionalization { |
| 1477 | |
| 1478 | // This keyset is used by functionalization when it calls into meta kernels |
| 1479 | // to accurately propagate stride metadata. |
| 1480 | // Exclude any modes: the purpose of calling into meta kernels is only as an implementation |
| 1481 | // detail to perform shape inference, and we don't want any modal keys to run. |
| 1482 | // Specifically, we want to prevent functionalization and Python modes from running. |
| 1483 | constexpr auto exclude_keys_for_meta_dispatch = |
| 1484 | c10::functorch_transforms_ks | |
| 1485 | c10::DispatchKeySet({ |
| 1486 | c10::DispatchKey::FuncTorchDynamicLayerBackMode, |
| 1487 | c10::DispatchKey::FuncTorchDynamicLayerFrontMode, |
| 1488 | c10::DispatchKey::Python |
| 1489 | }); |
| 1490 | |
| 1491 | |
| 1492 | inline Tensor to_meta(const Tensor& t) { |
| 1493 | if (!t.defined()) return t; |
| 1494 | return at::native::empty_strided_meta_symint(t.sym_sizes(), t.sym_strides(), |
| 1495 | /*dtype=*/c10::make_optional(t.scalar_type()), /*layout=*/c10::make_optional(t.layout()), |
| 1496 | /*device=*/c10::make_optional(c10::Device(kMeta)), /*pin_memory=*/c10::nullopt); |
| 1497 | } |
| 1498 | |
| 1499 | inline c10::optional<Tensor> to_meta(const c10::optional<Tensor>& t) { |
| 1500 | if (t.has_value()) { |
| 1501 | return c10::make_optional<Tensor>(to_meta(*t)); |
| 1502 | } |
| 1503 | return c10::nullopt; |
| 1504 | } |
| 1505 | |
| 1506 | inline std::vector<Tensor> to_meta(at::ITensorListRef t_list) { |
| 1507 | std::vector<Tensor> outputs; |
| 1508 | outputs.reserve(t_list.size()); |
| 1509 | for (const auto& tensor : t_list) { |
| 1510 | outputs.push_back(to_meta(tensor)); |
| 1511 | } |
| 1512 | return outputs; |
| 1513 | } |
| 1514 | |
| 1515 | inline c10::List<Tensor> to_meta(const c10::List<Tensor>& t_list) { |
| 1516 | c10::List<Tensor> outputs; |
| 1517 | outputs.reserve(t_list.size()); |
| 1518 | for (const auto i : c10::irange(t_list.size())) { |
| 1519 | outputs.push_back(to_meta(t_list[i])); |
| 1520 | } |
| 1521 | return outputs; |
| 1522 | } |
| 1523 | |
| 1524 | inline c10::List<c10::optional<Tensor>> to_meta(const c10::List<c10::optional<Tensor>>& t_list) { |
| 1525 | c10::List<c10::optional<Tensor>> outputs; |
| 1526 | outputs.reserve(t_list.size()); |
| 1527 | for (const auto i : c10::irange(t_list.size())) { |
| 1528 | outputs.push_back(to_meta(t_list[i])); |
| 1529 | } |
| 1530 | return outputs; |
| 1531 | } |
| 1532 | |
| 1533 | |
| 1534 | |
| 1535 | at::Tensor & _cudnn_rnn_flatten_weight_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList weight_arr, int64_t weight_stride0, c10::SymInt input_size, int64_t mode, c10::SymInt hidden_size, c10::SymInt proj_size, int64_t num_layers, bool batch_first, bool bidirectional, at::Tensor & out) { |
| 1536 | if (false) { |
| 1537 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1538 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1539 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1540 | auto weight_arr_meta = to_meta(weight_arr); |
| 1541 | auto out_meta = to_meta(out); |
| 1542 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1543 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1544 | at::_ops::_cudnn_rnn_flatten_weight_out::call(weight_arr_meta, weight_stride0, input_size, mode, hidden_size, proj_size, num_layers, batch_first, bidirectional, out_meta); |
| 1545 | } |
| 1546 | |
| 1547 | ::std::vector<at::Tensor> weight_arr_; |
| 1548 | if (at::functionalization::impl::isFunctionalTensor(weight_arr)) { |
| 1549 | at::functionalization::impl::sync(weight_arr); |
| 1550 | weight_arr_ = at::functionalization::impl::from_functional_tensor(weight_arr); |
| 1551 | } else { |
| 1552 | weight_arr_ = weight_arr.vec(); |
| 1553 | } |
| 1554 | |
| 1555 | at::Tensor out_; |
| 1556 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 1557 | at::functionalization::impl::sync(out); |
| 1558 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 1559 | } else { |
| 1560 | out_ = out; |
| 1561 | } |
| 1562 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 1563 | if ((false || at::functionalization::impl::isFunctionalTensor(weight_arr))) { |
| 1564 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1565 | TORCH_INTERNAL_ASSERT(false, |
| 1566 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1567 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1568 | } else { |
| 1569 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1570 | at::AutoDispatchSkipFunctionalize guard; |
| 1571 | at::Tensor tmp_output = at::_ops::_cudnn_rnn_flatten_weight_out::call(weight_arr_, weight_stride0, input_size, mode, hidden_size, proj_size, num_layers, batch_first, bidirectional, out_); |
| 1572 | return out;; |
| 1573 | } |
| 1574 | } else { |
| 1575 | at::Tensor tmp_output; |
| 1576 | { |
| 1577 | at::AutoDispatchSkipFunctionalize guard; |
| 1578 | tmp_output = at::_ops::_cudnn_rnn_flatten_weight::call(weight_arr_, weight_stride0, input_size, mode, hidden_size, proj_size, num_layers, batch_first, bidirectional); |
| 1579 | } |
| 1580 | at::functionalization::impl::replace_(out, tmp_output); |
| 1581 | at::functionalization::impl::commit_update(out); |
| 1582 | at::functionalization::impl::sync(out); |
| 1583 | return out; |
| 1584 | } |
| 1585 | } |
| 1586 | |
| 1587 | at::Tensor & abs_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 1588 | if (false) { |
| 1589 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1590 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1591 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1592 | auto self_meta = to_meta(self); |
| 1593 | auto out_meta = to_meta(out); |
| 1594 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1595 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1596 | at::_ops::abs_out::call(self_meta, out_meta); |
| 1597 | } |
| 1598 | |
| 1599 | at::Tensor self_; |
| 1600 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1601 | at::functionalization::impl::sync(self); |
| 1602 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1603 | } else { |
| 1604 | self_ = self; |
| 1605 | } |
| 1606 | |
| 1607 | at::Tensor out_; |
| 1608 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 1609 | at::functionalization::impl::sync(out); |
| 1610 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 1611 | } else { |
| 1612 | out_ = out; |
| 1613 | } |
| 1614 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 1615 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 1616 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1617 | TORCH_INTERNAL_ASSERT(false, |
| 1618 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1619 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1620 | } else { |
| 1621 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1622 | at::AutoDispatchSkipFunctionalize guard; |
| 1623 | at::Tensor tmp_output = at::_ops::abs_out::call(self_, out_); |
| 1624 | return out;; |
| 1625 | } |
| 1626 | } else { |
| 1627 | at::Tensor tmp_output; |
| 1628 | { |
| 1629 | at::AutoDispatchSkipFunctionalize guard; |
| 1630 | tmp_output = at::_ops::abs::call(self_); |
| 1631 | } |
| 1632 | at::functionalization::impl::replace_(out, tmp_output); |
| 1633 | at::functionalization::impl::commit_update(out); |
| 1634 | at::functionalization::impl::sync(out); |
| 1635 | return out; |
| 1636 | } |
| 1637 | } |
| 1638 | |
| 1639 | at::Tensor & abs_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 1640 | if (true) { |
| 1641 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1642 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1643 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1644 | auto self_meta = to_meta(self); |
| 1645 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1646 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1647 | at::_ops::abs_::call(self_meta); |
| 1648 | } |
| 1649 | |
| 1650 | at::Tensor self_; |
| 1651 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1652 | at::functionalization::impl::sync(self); |
| 1653 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1654 | } else { |
| 1655 | self_ = self; |
| 1656 | } |
| 1657 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 1658 | if ((false)) { |
| 1659 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1660 | TORCH_INTERNAL_ASSERT(false, |
| 1661 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1662 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1663 | } else { |
| 1664 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1665 | at::AutoDispatchSkipFunctionalize guard; |
| 1666 | at::Tensor tmp_output = at::_ops::abs_::call(self_); |
| 1667 | return self;; |
| 1668 | } |
| 1669 | } else { |
| 1670 | at::Tensor tmp_output; |
| 1671 | { |
| 1672 | at::AutoDispatchSkipFunctionalize guard; |
| 1673 | tmp_output = at::_ops::abs::call(self_); |
| 1674 | } |
| 1675 | at::functionalization::impl::replace_(self, tmp_output); |
| 1676 | at::functionalization::impl::commit_update(self); |
| 1677 | at::functionalization::impl::sync(self); |
| 1678 | return self; |
| 1679 | } |
| 1680 | } |
| 1681 | |
| 1682 | at::Tensor & absolute_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 1683 | if (false) { |
| 1684 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1685 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1686 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1687 | auto self_meta = to_meta(self); |
| 1688 | auto out_meta = to_meta(out); |
| 1689 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1690 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1691 | at::_ops::absolute_out::call(self_meta, out_meta); |
| 1692 | } |
| 1693 | |
| 1694 | at::Tensor self_; |
| 1695 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1696 | at::functionalization::impl::sync(self); |
| 1697 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1698 | } else { |
| 1699 | self_ = self; |
| 1700 | } |
| 1701 | |
| 1702 | at::Tensor out_; |
| 1703 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 1704 | at::functionalization::impl::sync(out); |
| 1705 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 1706 | } else { |
| 1707 | out_ = out; |
| 1708 | } |
| 1709 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 1710 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 1711 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1712 | TORCH_INTERNAL_ASSERT(false, |
| 1713 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1714 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1715 | } else { |
| 1716 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1717 | at::AutoDispatchSkipFunctionalize guard; |
| 1718 | at::Tensor tmp_output = at::_ops::absolute_out::call(self_, out_); |
| 1719 | return out;; |
| 1720 | } |
| 1721 | } else { |
| 1722 | at::Tensor tmp_output; |
| 1723 | { |
| 1724 | at::AutoDispatchSkipFunctionalize guard; |
| 1725 | tmp_output = at::_ops::absolute::call(self_); |
| 1726 | } |
| 1727 | at::functionalization::impl::replace_(out, tmp_output); |
| 1728 | at::functionalization::impl::commit_update(out); |
| 1729 | at::functionalization::impl::sync(out); |
| 1730 | return out; |
| 1731 | } |
| 1732 | } |
| 1733 | |
| 1734 | at::Tensor & absolute_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 1735 | if (true) { |
| 1736 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1737 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1738 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1739 | auto self_meta = to_meta(self); |
| 1740 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1741 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1742 | at::_ops::absolute_::call(self_meta); |
| 1743 | } |
| 1744 | |
| 1745 | at::Tensor self_; |
| 1746 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1747 | at::functionalization::impl::sync(self); |
| 1748 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1749 | } else { |
| 1750 | self_ = self; |
| 1751 | } |
| 1752 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 1753 | if ((false)) { |
| 1754 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1755 | TORCH_INTERNAL_ASSERT(false, |
| 1756 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1757 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1758 | } else { |
| 1759 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1760 | at::AutoDispatchSkipFunctionalize guard; |
| 1761 | at::Tensor tmp_output = at::_ops::absolute_::call(self_); |
| 1762 | return self;; |
| 1763 | } |
| 1764 | } else { |
| 1765 | at::Tensor tmp_output; |
| 1766 | { |
| 1767 | at::AutoDispatchSkipFunctionalize guard; |
| 1768 | tmp_output = at::_ops::absolute::call(self_); |
| 1769 | } |
| 1770 | at::functionalization::impl::replace_(self, tmp_output); |
| 1771 | at::functionalization::impl::commit_update(self); |
| 1772 | at::functionalization::impl::sync(self); |
| 1773 | return self; |
| 1774 | } |
| 1775 | } |
| 1776 | |
| 1777 | at::Tensor & _conj_physical_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 1778 | if (false) { |
| 1779 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1780 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1781 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1782 | auto self_meta = to_meta(self); |
| 1783 | auto out_meta = to_meta(out); |
| 1784 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1785 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1786 | at::_ops::_conj_physical_out::call(self_meta, out_meta); |
| 1787 | } |
| 1788 | |
| 1789 | at::Tensor self_; |
| 1790 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1791 | at::functionalization::impl::sync(self); |
| 1792 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1793 | } else { |
| 1794 | self_ = self; |
| 1795 | } |
| 1796 | |
| 1797 | at::Tensor out_; |
| 1798 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 1799 | at::functionalization::impl::sync(out); |
| 1800 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 1801 | } else { |
| 1802 | out_ = out; |
| 1803 | } |
| 1804 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 1805 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 1806 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1807 | TORCH_INTERNAL_ASSERT(false, |
| 1808 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1809 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1810 | } else { |
| 1811 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1812 | at::AutoDispatchSkipFunctionalize guard; |
| 1813 | at::Tensor tmp_output = at::_ops::_conj_physical_out::call(self_, out_); |
| 1814 | return out;; |
| 1815 | } |
| 1816 | } else { |
| 1817 | at::Tensor tmp_output; |
| 1818 | { |
| 1819 | at::AutoDispatchSkipFunctionalize guard; |
| 1820 | tmp_output = at::_ops::_conj_physical::call(self_); |
| 1821 | } |
| 1822 | at::functionalization::impl::replace_(out, tmp_output); |
| 1823 | at::functionalization::impl::commit_update(out); |
| 1824 | at::functionalization::impl::sync(out); |
| 1825 | return out; |
| 1826 | } |
| 1827 | } |
| 1828 | |
| 1829 | at::Tensor & acos_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 1830 | if (false) { |
| 1831 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1832 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1833 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1834 | auto self_meta = to_meta(self); |
| 1835 | auto out_meta = to_meta(out); |
| 1836 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1837 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1838 | at::_ops::acos_out::call(self_meta, out_meta); |
| 1839 | } |
| 1840 | |
| 1841 | at::Tensor self_; |
| 1842 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1843 | at::functionalization::impl::sync(self); |
| 1844 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1845 | } else { |
| 1846 | self_ = self; |
| 1847 | } |
| 1848 | |
| 1849 | at::Tensor out_; |
| 1850 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 1851 | at::functionalization::impl::sync(out); |
| 1852 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 1853 | } else { |
| 1854 | out_ = out; |
| 1855 | } |
| 1856 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 1857 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 1858 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1859 | TORCH_INTERNAL_ASSERT(false, |
| 1860 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1861 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1862 | } else { |
| 1863 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1864 | at::AutoDispatchSkipFunctionalize guard; |
| 1865 | at::Tensor tmp_output = at::_ops::acos_out::call(self_, out_); |
| 1866 | return out;; |
| 1867 | } |
| 1868 | } else { |
| 1869 | at::Tensor tmp_output; |
| 1870 | { |
| 1871 | at::AutoDispatchSkipFunctionalize guard; |
| 1872 | tmp_output = at::_ops::acos::call(self_); |
| 1873 | } |
| 1874 | at::functionalization::impl::replace_(out, tmp_output); |
| 1875 | at::functionalization::impl::commit_update(out); |
| 1876 | at::functionalization::impl::sync(out); |
| 1877 | return out; |
| 1878 | } |
| 1879 | } |
| 1880 | |
| 1881 | at::Tensor & acos_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 1882 | if (true) { |
| 1883 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1884 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1885 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1886 | auto self_meta = to_meta(self); |
| 1887 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1888 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1889 | at::_ops::acos_::call(self_meta); |
| 1890 | } |
| 1891 | |
| 1892 | at::Tensor self_; |
| 1893 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1894 | at::functionalization::impl::sync(self); |
| 1895 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1896 | } else { |
| 1897 | self_ = self; |
| 1898 | } |
| 1899 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 1900 | if ((false)) { |
| 1901 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1902 | TORCH_INTERNAL_ASSERT(false, |
| 1903 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1904 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1905 | } else { |
| 1906 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1907 | at::AutoDispatchSkipFunctionalize guard; |
| 1908 | at::Tensor tmp_output = at::_ops::acos_::call(self_); |
| 1909 | return self;; |
| 1910 | } |
| 1911 | } else { |
| 1912 | at::Tensor tmp_output; |
| 1913 | { |
| 1914 | at::AutoDispatchSkipFunctionalize guard; |
| 1915 | tmp_output = at::_ops::acos::call(self_); |
| 1916 | } |
| 1917 | at::functionalization::impl::replace_(self, tmp_output); |
| 1918 | at::functionalization::impl::commit_update(self); |
| 1919 | at::functionalization::impl::sync(self); |
| 1920 | return self; |
| 1921 | } |
| 1922 | } |
| 1923 | |
| 1924 | at::Tensor & arccos_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 1925 | if (false) { |
| 1926 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1927 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1928 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1929 | auto self_meta = to_meta(self); |
| 1930 | auto out_meta = to_meta(out); |
| 1931 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1932 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1933 | at::_ops::arccos_out::call(self_meta, out_meta); |
| 1934 | } |
| 1935 | |
| 1936 | at::Tensor self_; |
| 1937 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1938 | at::functionalization::impl::sync(self); |
| 1939 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1940 | } else { |
| 1941 | self_ = self; |
| 1942 | } |
| 1943 | |
| 1944 | at::Tensor out_; |
| 1945 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 1946 | at::functionalization::impl::sync(out); |
| 1947 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 1948 | } else { |
| 1949 | out_ = out; |
| 1950 | } |
| 1951 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 1952 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 1953 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1954 | TORCH_INTERNAL_ASSERT(false, |
| 1955 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1956 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 1957 | } else { |
| 1958 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 1959 | at::AutoDispatchSkipFunctionalize guard; |
| 1960 | at::Tensor tmp_output = at::_ops::arccos_out::call(self_, out_); |
| 1961 | return out;; |
| 1962 | } |
| 1963 | } else { |
| 1964 | at::Tensor tmp_output; |
| 1965 | { |
| 1966 | at::AutoDispatchSkipFunctionalize guard; |
| 1967 | tmp_output = at::_ops::arccos::call(self_); |
| 1968 | } |
| 1969 | at::functionalization::impl::replace_(out, tmp_output); |
| 1970 | at::functionalization::impl::commit_update(out); |
| 1971 | at::functionalization::impl::sync(out); |
| 1972 | return out; |
| 1973 | } |
| 1974 | } |
| 1975 | |
| 1976 | at::Tensor & arccos_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 1977 | if (true) { |
| 1978 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 1979 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 1980 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 1981 | auto self_meta = to_meta(self); |
| 1982 | at::AutoDispatchSkipFunctionalize func_guard; |
| 1983 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 1984 | at::_ops::arccos_::call(self_meta); |
| 1985 | } |
| 1986 | |
| 1987 | at::Tensor self_; |
| 1988 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 1989 | at::functionalization::impl::sync(self); |
| 1990 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 1991 | } else { |
| 1992 | self_ = self; |
| 1993 | } |
| 1994 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 1995 | if ((false)) { |
| 1996 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 1997 | TORCH_INTERNAL_ASSERT(false, |
| 1998 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 1999 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2000 | } else { |
| 2001 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2002 | at::AutoDispatchSkipFunctionalize guard; |
| 2003 | at::Tensor tmp_output = at::_ops::arccos_::call(self_); |
| 2004 | return self;; |
| 2005 | } |
| 2006 | } else { |
| 2007 | at::Tensor tmp_output; |
| 2008 | { |
| 2009 | at::AutoDispatchSkipFunctionalize guard; |
| 2010 | tmp_output = at::_ops::arccos::call(self_); |
| 2011 | } |
| 2012 | at::functionalization::impl::replace_(self, tmp_output); |
| 2013 | at::functionalization::impl::commit_update(self); |
| 2014 | at::functionalization::impl::sync(self); |
| 2015 | return self; |
| 2016 | } |
| 2017 | } |
| 2018 | |
| 2019 | at::Tensor & affine_grid_generator_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & theta, at::IntArrayRef size, bool align_corners, at::Tensor & out) { |
| 2020 | if (false) { |
| 2021 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2022 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2023 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2024 | auto theta_meta = to_meta(theta); |
| 2025 | auto out_meta = to_meta(out); |
| 2026 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2027 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2028 | at::_ops::affine_grid_generator_out::call(theta_meta, size, align_corners, out_meta); |
| 2029 | } |
| 2030 | |
| 2031 | at::Tensor theta_; |
| 2032 | if (at::functionalization::impl::isFunctionalTensor(theta)) { |
| 2033 | at::functionalization::impl::sync(theta); |
| 2034 | theta_ = at::functionalization::impl::from_functional_tensor(theta); |
| 2035 | } else { |
| 2036 | theta_ = theta; |
| 2037 | } |
| 2038 | |
| 2039 | at::Tensor out_; |
| 2040 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2041 | at::functionalization::impl::sync(out); |
| 2042 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2043 | } else { |
| 2044 | out_ = out; |
| 2045 | } |
| 2046 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2047 | if ((false || at::functionalization::impl::isFunctionalTensor(theta))) { |
| 2048 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2049 | TORCH_INTERNAL_ASSERT(false, |
| 2050 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2051 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2052 | } else { |
| 2053 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2054 | at::AutoDispatchSkipFunctionalize guard; |
| 2055 | at::Tensor tmp_output = at::_ops::affine_grid_generator_out::call(theta_, size, align_corners, out_); |
| 2056 | return out;; |
| 2057 | } |
| 2058 | } else { |
| 2059 | at::Tensor tmp_output; |
| 2060 | { |
| 2061 | at::AutoDispatchSkipFunctionalize guard; |
| 2062 | tmp_output = at::_ops::affine_grid_generator::call(theta_, size, align_corners); |
| 2063 | } |
| 2064 | at::functionalization::impl::replace_(out, tmp_output); |
| 2065 | at::functionalization::impl::commit_update(out); |
| 2066 | at::functionalization::impl::sync(out); |
| 2067 | return out; |
| 2068 | } |
| 2069 | } |
| 2070 | |
| 2071 | at::Tensor & arange_out_out(c10::DispatchKeySet dispatchKeySet, const at::Scalar & end, at::Tensor & out) { |
| 2072 | if (false) { |
| 2073 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2074 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2075 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2076 | auto out_meta = to_meta(out); |
| 2077 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2078 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2079 | at::_ops::arange_out::call(end, out_meta); |
| 2080 | } |
| 2081 | |
| 2082 | at::Tensor out_; |
| 2083 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2084 | at::functionalization::impl::sync(out); |
| 2085 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2086 | } else { |
| 2087 | out_ = out; |
| 2088 | } |
| 2089 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2090 | if ((false)) { |
| 2091 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2092 | TORCH_INTERNAL_ASSERT(false, |
| 2093 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2094 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2095 | } else { |
| 2096 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2097 | at::AutoDispatchSkipFunctionalize guard; |
| 2098 | at::Tensor tmp_output = at::_ops::arange_out::call(end, out_); |
| 2099 | return out;; |
| 2100 | } |
| 2101 | } else { |
| 2102 | at::Tensor tmp_output; |
| 2103 | { |
| 2104 | at::AutoDispatchSkipFunctionalize guard; |
| 2105 | tmp_output = at::_ops::arange::call(end, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 2106 | } |
| 2107 | at::functionalization::impl::replace_(out, tmp_output); |
| 2108 | at::functionalization::impl::commit_update(out); |
| 2109 | at::functionalization::impl::sync(out); |
| 2110 | return out; |
| 2111 | } |
| 2112 | } |
| 2113 | |
| 2114 | at::Tensor & arange_out_start_out(c10::DispatchKeySet dispatchKeySet, const at::Scalar & start, const at::Scalar & end, const at::Scalar & step, at::Tensor & out) { |
| 2115 | if (false) { |
| 2116 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2117 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2118 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2119 | auto out_meta = to_meta(out); |
| 2120 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2121 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2122 | at::_ops::arange_start_out::call(start, end, step, out_meta); |
| 2123 | } |
| 2124 | |
| 2125 | at::Tensor out_; |
| 2126 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2127 | at::functionalization::impl::sync(out); |
| 2128 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2129 | } else { |
| 2130 | out_ = out; |
| 2131 | } |
| 2132 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2133 | if ((false)) { |
| 2134 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2135 | TORCH_INTERNAL_ASSERT(false, |
| 2136 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2137 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2138 | } else { |
| 2139 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2140 | at::AutoDispatchSkipFunctionalize guard; |
| 2141 | at::Tensor tmp_output = at::_ops::arange_start_out::call(start, end, step, out_); |
| 2142 | return out;; |
| 2143 | } |
| 2144 | } else { |
| 2145 | at::Tensor tmp_output; |
| 2146 | { |
| 2147 | at::AutoDispatchSkipFunctionalize guard; |
| 2148 | tmp_output = at::_ops::arange_start_step::call(start, end, step, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 2149 | } |
| 2150 | at::functionalization::impl::replace_(out, tmp_output); |
| 2151 | at::functionalization::impl::commit_update(out); |
| 2152 | at::functionalization::impl::sync(out); |
| 2153 | return out; |
| 2154 | } |
| 2155 | } |
| 2156 | |
| 2157 | at::Tensor & arccosh_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 2158 | if (false) { |
| 2159 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2160 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2161 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2162 | auto self_meta = to_meta(self); |
| 2163 | auto out_meta = to_meta(out); |
| 2164 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2165 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2166 | at::_ops::arccosh_out::call(self_meta, out_meta); |
| 2167 | } |
| 2168 | |
| 2169 | at::Tensor self_; |
| 2170 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2171 | at::functionalization::impl::sync(self); |
| 2172 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2173 | } else { |
| 2174 | self_ = self; |
| 2175 | } |
| 2176 | |
| 2177 | at::Tensor out_; |
| 2178 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2179 | at::functionalization::impl::sync(out); |
| 2180 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2181 | } else { |
| 2182 | out_ = out; |
| 2183 | } |
| 2184 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2185 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 2186 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2187 | TORCH_INTERNAL_ASSERT(false, |
| 2188 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2189 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2190 | } else { |
| 2191 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2192 | at::AutoDispatchSkipFunctionalize guard; |
| 2193 | at::Tensor tmp_output = at::_ops::arccosh_out::call(self_, out_); |
| 2194 | return out;; |
| 2195 | } |
| 2196 | } else { |
| 2197 | at::Tensor tmp_output; |
| 2198 | { |
| 2199 | at::AutoDispatchSkipFunctionalize guard; |
| 2200 | tmp_output = at::_ops::arccosh::call(self_); |
| 2201 | } |
| 2202 | at::functionalization::impl::replace_(out, tmp_output); |
| 2203 | at::functionalization::impl::commit_update(out); |
| 2204 | at::functionalization::impl::sync(out); |
| 2205 | return out; |
| 2206 | } |
| 2207 | } |
| 2208 | |
| 2209 | at::Tensor & arccosh_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 2210 | if (true) { |
| 2211 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2212 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2213 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2214 | auto self_meta = to_meta(self); |
| 2215 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2216 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2217 | at::_ops::arccosh_::call(self_meta); |
| 2218 | } |
| 2219 | |
| 2220 | at::Tensor self_; |
| 2221 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2222 | at::functionalization::impl::sync(self); |
| 2223 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2224 | } else { |
| 2225 | self_ = self; |
| 2226 | } |
| 2227 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 2228 | if ((false)) { |
| 2229 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2230 | TORCH_INTERNAL_ASSERT(false, |
| 2231 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2232 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2233 | } else { |
| 2234 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2235 | at::AutoDispatchSkipFunctionalize guard; |
| 2236 | at::Tensor tmp_output = at::_ops::arccosh_::call(self_); |
| 2237 | return self;; |
| 2238 | } |
| 2239 | } else { |
| 2240 | at::Tensor tmp_output; |
| 2241 | { |
| 2242 | at::AutoDispatchSkipFunctionalize guard; |
| 2243 | tmp_output = at::_ops::arccosh::call(self_); |
| 2244 | } |
| 2245 | at::functionalization::impl::replace_(self, tmp_output); |
| 2246 | at::functionalization::impl::commit_update(self); |
| 2247 | at::functionalization::impl::sync(self); |
| 2248 | return self; |
| 2249 | } |
| 2250 | } |
| 2251 | |
| 2252 | at::Tensor & arcsinh_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 2253 | if (false) { |
| 2254 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2255 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2256 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2257 | auto self_meta = to_meta(self); |
| 2258 | auto out_meta = to_meta(out); |
| 2259 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2260 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2261 | at::_ops::arcsinh_out::call(self_meta, out_meta); |
| 2262 | } |
| 2263 | |
| 2264 | at::Tensor self_; |
| 2265 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2266 | at::functionalization::impl::sync(self); |
| 2267 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2268 | } else { |
| 2269 | self_ = self; |
| 2270 | } |
| 2271 | |
| 2272 | at::Tensor out_; |
| 2273 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2274 | at::functionalization::impl::sync(out); |
| 2275 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2276 | } else { |
| 2277 | out_ = out; |
| 2278 | } |
| 2279 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2280 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 2281 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2282 | TORCH_INTERNAL_ASSERT(false, |
| 2283 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2284 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2285 | } else { |
| 2286 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2287 | at::AutoDispatchSkipFunctionalize guard; |
| 2288 | at::Tensor tmp_output = at::_ops::arcsinh_out::call(self_, out_); |
| 2289 | return out;; |
| 2290 | } |
| 2291 | } else { |
| 2292 | at::Tensor tmp_output; |
| 2293 | { |
| 2294 | at::AutoDispatchSkipFunctionalize guard; |
| 2295 | tmp_output = at::_ops::arcsinh::call(self_); |
| 2296 | } |
| 2297 | at::functionalization::impl::replace_(out, tmp_output); |
| 2298 | at::functionalization::impl::commit_update(out); |
| 2299 | at::functionalization::impl::sync(out); |
| 2300 | return out; |
| 2301 | } |
| 2302 | } |
| 2303 | |
| 2304 | at::Tensor & arcsinh_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 2305 | if (true) { |
| 2306 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2307 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2308 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2309 | auto self_meta = to_meta(self); |
| 2310 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2311 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2312 | at::_ops::arcsinh_::call(self_meta); |
| 2313 | } |
| 2314 | |
| 2315 | at::Tensor self_; |
| 2316 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2317 | at::functionalization::impl::sync(self); |
| 2318 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2319 | } else { |
| 2320 | self_ = self; |
| 2321 | } |
| 2322 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 2323 | if ((false)) { |
| 2324 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2325 | TORCH_INTERNAL_ASSERT(false, |
| 2326 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2327 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2328 | } else { |
| 2329 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2330 | at::AutoDispatchSkipFunctionalize guard; |
| 2331 | at::Tensor tmp_output = at::_ops::arcsinh_::call(self_); |
| 2332 | return self;; |
| 2333 | } |
| 2334 | } else { |
| 2335 | at::Tensor tmp_output; |
| 2336 | { |
| 2337 | at::AutoDispatchSkipFunctionalize guard; |
| 2338 | tmp_output = at::_ops::arcsinh::call(self_); |
| 2339 | } |
| 2340 | at::functionalization::impl::replace_(self, tmp_output); |
| 2341 | at::functionalization::impl::commit_update(self); |
| 2342 | at::functionalization::impl::sync(self); |
| 2343 | return self; |
| 2344 | } |
| 2345 | } |
| 2346 | |
| 2347 | at::Tensor & atanh_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 2348 | if (false) { |
| 2349 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2350 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2351 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2352 | auto self_meta = to_meta(self); |
| 2353 | auto out_meta = to_meta(out); |
| 2354 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2355 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2356 | at::_ops::atanh_out::call(self_meta, out_meta); |
| 2357 | } |
| 2358 | |
| 2359 | at::Tensor self_; |
| 2360 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2361 | at::functionalization::impl::sync(self); |
| 2362 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2363 | } else { |
| 2364 | self_ = self; |
| 2365 | } |
| 2366 | |
| 2367 | at::Tensor out_; |
| 2368 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2369 | at::functionalization::impl::sync(out); |
| 2370 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2371 | } else { |
| 2372 | out_ = out; |
| 2373 | } |
| 2374 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2375 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 2376 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2377 | TORCH_INTERNAL_ASSERT(false, |
| 2378 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2379 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2380 | } else { |
| 2381 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2382 | at::AutoDispatchSkipFunctionalize guard; |
| 2383 | at::Tensor tmp_output = at::_ops::atanh_out::call(self_, out_); |
| 2384 | return out;; |
| 2385 | } |
| 2386 | } else { |
| 2387 | at::Tensor tmp_output; |
| 2388 | { |
| 2389 | at::AutoDispatchSkipFunctionalize guard; |
| 2390 | tmp_output = at::_ops::atanh::call(self_); |
| 2391 | } |
| 2392 | at::functionalization::impl::replace_(out, tmp_output); |
| 2393 | at::functionalization::impl::commit_update(out); |
| 2394 | at::functionalization::impl::sync(out); |
| 2395 | return out; |
| 2396 | } |
| 2397 | } |
| 2398 | |
| 2399 | at::Tensor & atanh_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 2400 | if (true) { |
| 2401 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2402 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2403 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2404 | auto self_meta = to_meta(self); |
| 2405 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2406 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2407 | at::_ops::atanh_::call(self_meta); |
| 2408 | } |
| 2409 | |
| 2410 | at::Tensor self_; |
| 2411 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2412 | at::functionalization::impl::sync(self); |
| 2413 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2414 | } else { |
| 2415 | self_ = self; |
| 2416 | } |
| 2417 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 2418 | if ((false)) { |
| 2419 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2420 | TORCH_INTERNAL_ASSERT(false, |
| 2421 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2422 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2423 | } else { |
| 2424 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2425 | at::AutoDispatchSkipFunctionalize guard; |
| 2426 | at::Tensor tmp_output = at::_ops::atanh_::call(self_); |
| 2427 | return self;; |
| 2428 | } |
| 2429 | } else { |
| 2430 | at::Tensor tmp_output; |
| 2431 | { |
| 2432 | at::AutoDispatchSkipFunctionalize guard; |
| 2433 | tmp_output = at::_ops::atanh::call(self_); |
| 2434 | } |
| 2435 | at::functionalization::impl::replace_(self, tmp_output); |
| 2436 | at::functionalization::impl::commit_update(self); |
| 2437 | at::functionalization::impl::sync(self); |
| 2438 | return self; |
| 2439 | } |
| 2440 | } |
| 2441 | |
| 2442 | at::Tensor & asin_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 2443 | if (false) { |
| 2444 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2445 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2446 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2447 | auto self_meta = to_meta(self); |
| 2448 | auto out_meta = to_meta(out); |
| 2449 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2450 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2451 | at::_ops::asin_out::call(self_meta, out_meta); |
| 2452 | } |
| 2453 | |
| 2454 | at::Tensor self_; |
| 2455 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2456 | at::functionalization::impl::sync(self); |
| 2457 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2458 | } else { |
| 2459 | self_ = self; |
| 2460 | } |
| 2461 | |
| 2462 | at::Tensor out_; |
| 2463 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2464 | at::functionalization::impl::sync(out); |
| 2465 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2466 | } else { |
| 2467 | out_ = out; |
| 2468 | } |
| 2469 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2470 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 2471 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2472 | TORCH_INTERNAL_ASSERT(false, |
| 2473 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2474 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2475 | } else { |
| 2476 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2477 | at::AutoDispatchSkipFunctionalize guard; |
| 2478 | at::Tensor tmp_output = at::_ops::asin_out::call(self_, out_); |
| 2479 | return out;; |
| 2480 | } |
| 2481 | } else { |
| 2482 | at::Tensor tmp_output; |
| 2483 | { |
| 2484 | at::AutoDispatchSkipFunctionalize guard; |
| 2485 | tmp_output = at::_ops::asin::call(self_); |
| 2486 | } |
| 2487 | at::functionalization::impl::replace_(out, tmp_output); |
| 2488 | at::functionalization::impl::commit_update(out); |
| 2489 | at::functionalization::impl::sync(out); |
| 2490 | return out; |
| 2491 | } |
| 2492 | } |
| 2493 | |
| 2494 | at::Tensor & asin_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 2495 | if (true) { |
| 2496 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2497 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2498 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2499 | auto self_meta = to_meta(self); |
| 2500 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2501 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2502 | at::_ops::asin_::call(self_meta); |
| 2503 | } |
| 2504 | |
| 2505 | at::Tensor self_; |
| 2506 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2507 | at::functionalization::impl::sync(self); |
| 2508 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2509 | } else { |
| 2510 | self_ = self; |
| 2511 | } |
| 2512 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 2513 | if ((false)) { |
| 2514 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2515 | TORCH_INTERNAL_ASSERT(false, |
| 2516 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2517 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2518 | } else { |
| 2519 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2520 | at::AutoDispatchSkipFunctionalize guard; |
| 2521 | at::Tensor tmp_output = at::_ops::asin_::call(self_); |
| 2522 | return self;; |
| 2523 | } |
| 2524 | } else { |
| 2525 | at::Tensor tmp_output; |
| 2526 | { |
| 2527 | at::AutoDispatchSkipFunctionalize guard; |
| 2528 | tmp_output = at::_ops::asin::call(self_); |
| 2529 | } |
| 2530 | at::functionalization::impl::replace_(self, tmp_output); |
| 2531 | at::functionalization::impl::commit_update(self); |
| 2532 | at::functionalization::impl::sync(self); |
| 2533 | return self; |
| 2534 | } |
| 2535 | } |
| 2536 | |
| 2537 | at::Tensor & binary_cross_entropy_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, const at::Tensor & target, const c10::optional<at::Tensor> & weight, int64_t reduction, at::Tensor & grad_input) { |
| 2538 | if (false) { |
| 2539 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2540 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2541 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2542 | auto grad_output_meta = to_meta(grad_output); |
| 2543 | auto self_meta = to_meta(self); |
| 2544 | auto target_meta = to_meta(target); |
| 2545 | auto weight_meta = to_meta(weight); |
| 2546 | auto grad_input_meta = to_meta(grad_input); |
| 2547 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2548 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2549 | at::_ops::binary_cross_entropy_backward_grad_input::call(grad_output_meta, self_meta, target_meta, weight_meta, reduction, grad_input_meta); |
| 2550 | } |
| 2551 | |
| 2552 | at::Tensor grad_output_; |
| 2553 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 2554 | at::functionalization::impl::sync(grad_output); |
| 2555 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 2556 | } else { |
| 2557 | grad_output_ = grad_output; |
| 2558 | } |
| 2559 | |
| 2560 | at::Tensor self_; |
| 2561 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2562 | at::functionalization::impl::sync(self); |
| 2563 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2564 | } else { |
| 2565 | self_ = self; |
| 2566 | } |
| 2567 | |
| 2568 | at::Tensor target_; |
| 2569 | if (at::functionalization::impl::isFunctionalTensor(target)) { |
| 2570 | at::functionalization::impl::sync(target); |
| 2571 | target_ = at::functionalization::impl::from_functional_tensor(target); |
| 2572 | } else { |
| 2573 | target_ = target; |
| 2574 | } |
| 2575 | |
| 2576 | c10::optional<at::Tensor> weight_; |
| 2577 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 2578 | at::functionalization::impl::sync(weight); |
| 2579 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 2580 | } else { |
| 2581 | weight_ = weight; |
| 2582 | } |
| 2583 | |
| 2584 | at::Tensor grad_input_; |
| 2585 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 2586 | at::functionalization::impl::sync(grad_input); |
| 2587 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 2588 | } else { |
| 2589 | grad_input_ = grad_input; |
| 2590 | } |
| 2591 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 2592 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(target) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 2593 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2594 | TORCH_INTERNAL_ASSERT(false, |
| 2595 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2596 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2597 | } else { |
| 2598 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2599 | at::AutoDispatchSkipFunctionalize guard; |
| 2600 | at::Tensor tmp_output = at::_ops::binary_cross_entropy_backward_grad_input::call(grad_output_, self_, target_, weight_, reduction, grad_input_); |
| 2601 | return grad_input;; |
| 2602 | } |
| 2603 | } else { |
| 2604 | at::Tensor tmp_output; |
| 2605 | { |
| 2606 | at::AutoDispatchSkipFunctionalize guard; |
| 2607 | tmp_output = at::_ops::binary_cross_entropy_backward::call(grad_output_, self_, target_, weight_, reduction); |
| 2608 | } |
| 2609 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 2610 | at::functionalization::impl::commit_update(grad_input); |
| 2611 | at::functionalization::impl::sync(grad_input); |
| 2612 | return grad_input; |
| 2613 | } |
| 2614 | } |
| 2615 | |
| 2616 | at::Tensor & binary_cross_entropy_with_logits_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & target, const c10::optional<at::Tensor> & weight, const c10::optional<at::Tensor> & pos_weight, int64_t reduction, at::Tensor & out) { |
| 2617 | if (false) { |
| 2618 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2619 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2620 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2621 | auto self_meta = to_meta(self); |
| 2622 | auto target_meta = to_meta(target); |
| 2623 | auto weight_meta = to_meta(weight); |
| 2624 | auto pos_weight_meta = to_meta(pos_weight); |
| 2625 | auto out_meta = to_meta(out); |
| 2626 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2627 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2628 | at::_ops::binary_cross_entropy_with_logits_out::call(self_meta, target_meta, weight_meta, pos_weight_meta, reduction, out_meta); |
| 2629 | } |
| 2630 | |
| 2631 | at::Tensor self_; |
| 2632 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2633 | at::functionalization::impl::sync(self); |
| 2634 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2635 | } else { |
| 2636 | self_ = self; |
| 2637 | } |
| 2638 | |
| 2639 | at::Tensor target_; |
| 2640 | if (at::functionalization::impl::isFunctionalTensor(target)) { |
| 2641 | at::functionalization::impl::sync(target); |
| 2642 | target_ = at::functionalization::impl::from_functional_tensor(target); |
| 2643 | } else { |
| 2644 | target_ = target; |
| 2645 | } |
| 2646 | |
| 2647 | c10::optional<at::Tensor> weight_; |
| 2648 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 2649 | at::functionalization::impl::sync(weight); |
| 2650 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 2651 | } else { |
| 2652 | weight_ = weight; |
| 2653 | } |
| 2654 | |
| 2655 | c10::optional<at::Tensor> pos_weight_; |
| 2656 | if (at::functionalization::impl::isFunctionalTensor(pos_weight)) { |
| 2657 | at::functionalization::impl::sync(pos_weight); |
| 2658 | pos_weight_ = at::functionalization::impl::from_functional_tensor(pos_weight); |
| 2659 | } else { |
| 2660 | pos_weight_ = pos_weight; |
| 2661 | } |
| 2662 | |
| 2663 | at::Tensor out_; |
| 2664 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2665 | at::functionalization::impl::sync(out); |
| 2666 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2667 | } else { |
| 2668 | out_ = out; |
| 2669 | } |
| 2670 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2671 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(target) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(pos_weight))) { |
| 2672 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2673 | TORCH_INTERNAL_ASSERT(false, |
| 2674 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2675 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2676 | } else { |
| 2677 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2678 | at::AutoDispatchSkipFunctionalize guard; |
| 2679 | at::Tensor tmp_output = at::_ops::binary_cross_entropy_with_logits_out::call(self_, target_, weight_, pos_weight_, reduction, out_); |
| 2680 | return out;; |
| 2681 | } |
| 2682 | } else { |
| 2683 | at::Tensor tmp_output; |
| 2684 | { |
| 2685 | at::AutoDispatchSkipFunctionalize guard; |
| 2686 | tmp_output = at::_ops::binary_cross_entropy_with_logits::call(self_, target_, weight_, pos_weight_, reduction); |
| 2687 | } |
| 2688 | at::functionalization::impl::replace_(out, tmp_output); |
| 2689 | at::functionalization::impl::commit_update(out); |
| 2690 | at::functionalization::impl::sync(out); |
| 2691 | return out; |
| 2692 | } |
| 2693 | } |
| 2694 | |
| 2695 | at::Tensor & logical_not_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 2696 | if (false) { |
| 2697 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2698 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2699 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2700 | auto self_meta = to_meta(self); |
| 2701 | auto out_meta = to_meta(out); |
| 2702 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2703 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2704 | at::_ops::logical_not_out::call(self_meta, out_meta); |
| 2705 | } |
| 2706 | |
| 2707 | at::Tensor self_; |
| 2708 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2709 | at::functionalization::impl::sync(self); |
| 2710 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2711 | } else { |
| 2712 | self_ = self; |
| 2713 | } |
| 2714 | |
| 2715 | at::Tensor out_; |
| 2716 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2717 | at::functionalization::impl::sync(out); |
| 2718 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2719 | } else { |
| 2720 | out_ = out; |
| 2721 | } |
| 2722 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2723 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 2724 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2725 | TORCH_INTERNAL_ASSERT(false, |
| 2726 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2727 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2728 | } else { |
| 2729 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2730 | at::AutoDispatchSkipFunctionalize guard; |
| 2731 | at::Tensor tmp_output = at::_ops::logical_not_out::call(self_, out_); |
| 2732 | return out;; |
| 2733 | } |
| 2734 | } else { |
| 2735 | at::Tensor tmp_output; |
| 2736 | { |
| 2737 | at::AutoDispatchSkipFunctionalize guard; |
| 2738 | tmp_output = at::_ops::logical_not::call(self_); |
| 2739 | } |
| 2740 | at::functionalization::impl::replace_(out, tmp_output); |
| 2741 | at::functionalization::impl::commit_update(out); |
| 2742 | at::functionalization::impl::sync(out); |
| 2743 | return out; |
| 2744 | } |
| 2745 | } |
| 2746 | |
| 2747 | at::Tensor & logical_not_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 2748 | if (true) { |
| 2749 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2750 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2751 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2752 | auto self_meta = to_meta(self); |
| 2753 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2754 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2755 | at::_ops::logical_not_::call(self_meta); |
| 2756 | } |
| 2757 | |
| 2758 | at::Tensor self_; |
| 2759 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2760 | at::functionalization::impl::sync(self); |
| 2761 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2762 | } else { |
| 2763 | self_ = self; |
| 2764 | } |
| 2765 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 2766 | if ((false)) { |
| 2767 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2768 | TORCH_INTERNAL_ASSERT(false, |
| 2769 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2770 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2771 | } else { |
| 2772 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2773 | at::AutoDispatchSkipFunctionalize guard; |
| 2774 | at::Tensor tmp_output = at::_ops::logical_not_::call(self_); |
| 2775 | return self;; |
| 2776 | } |
| 2777 | } else { |
| 2778 | at::Tensor tmp_output; |
| 2779 | { |
| 2780 | at::AutoDispatchSkipFunctionalize guard; |
| 2781 | tmp_output = at::_ops::logical_not::call(self_); |
| 2782 | } |
| 2783 | at::functionalization::impl::replace_(self, tmp_output); |
| 2784 | at::functionalization::impl::commit_update(self); |
| 2785 | at::functionalization::impl::sync(self); |
| 2786 | return self; |
| 2787 | } |
| 2788 | } |
| 2789 | |
| 2790 | at::Tensor & logical_and_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 2791 | if (false) { |
| 2792 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2793 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2794 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2795 | auto self_meta = to_meta(self); |
| 2796 | auto other_meta = to_meta(other); |
| 2797 | auto out_meta = to_meta(out); |
| 2798 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2799 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2800 | at::_ops::logical_and_out::call(self_meta, other_meta, out_meta); |
| 2801 | } |
| 2802 | |
| 2803 | at::Tensor self_; |
| 2804 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2805 | at::functionalization::impl::sync(self); |
| 2806 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2807 | } else { |
| 2808 | self_ = self; |
| 2809 | } |
| 2810 | |
| 2811 | at::Tensor other_; |
| 2812 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 2813 | at::functionalization::impl::sync(other); |
| 2814 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 2815 | } else { |
| 2816 | other_ = other; |
| 2817 | } |
| 2818 | |
| 2819 | at::Tensor out_; |
| 2820 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2821 | at::functionalization::impl::sync(out); |
| 2822 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2823 | } else { |
| 2824 | out_ = out; |
| 2825 | } |
| 2826 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2827 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 2828 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2829 | TORCH_INTERNAL_ASSERT(false, |
| 2830 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2831 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2832 | } else { |
| 2833 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2834 | at::AutoDispatchSkipFunctionalize guard; |
| 2835 | at::Tensor tmp_output = at::_ops::logical_and_out::call(self_, other_, out_); |
| 2836 | return out;; |
| 2837 | } |
| 2838 | } else { |
| 2839 | at::Tensor tmp_output; |
| 2840 | { |
| 2841 | at::AutoDispatchSkipFunctionalize guard; |
| 2842 | tmp_output = at::_ops::logical_and::call(self_, other_); |
| 2843 | } |
| 2844 | at::functionalization::impl::replace_(out, tmp_output); |
| 2845 | at::functionalization::impl::commit_update(out); |
| 2846 | at::functionalization::impl::sync(out); |
| 2847 | return out; |
| 2848 | } |
| 2849 | } |
| 2850 | |
| 2851 | at::Tensor & logical_and_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 2852 | if (true) { |
| 2853 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2854 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2855 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2856 | auto self_meta = to_meta(self); |
| 2857 | auto other_meta = to_meta(other); |
| 2858 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2859 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2860 | at::_ops::logical_and_::call(self_meta, other_meta); |
| 2861 | } |
| 2862 | |
| 2863 | at::Tensor self_; |
| 2864 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 2865 | at::functionalization::impl::sync(self); |
| 2866 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 2867 | } else { |
| 2868 | self_ = self; |
| 2869 | } |
| 2870 | |
| 2871 | at::Tensor other_; |
| 2872 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 2873 | at::functionalization::impl::sync(other); |
| 2874 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 2875 | } else { |
| 2876 | other_ = other; |
| 2877 | } |
| 2878 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 2879 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 2880 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2881 | TORCH_INTERNAL_ASSERT(false, |
| 2882 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2883 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2884 | } else { |
| 2885 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2886 | at::AutoDispatchSkipFunctionalize guard; |
| 2887 | at::Tensor tmp_output = at::_ops::logical_and_::call(self_, other_); |
| 2888 | return self;; |
| 2889 | } |
| 2890 | } else { |
| 2891 | at::Tensor tmp_output; |
| 2892 | { |
| 2893 | at::AutoDispatchSkipFunctionalize guard; |
| 2894 | tmp_output = at::_ops::logical_and::call(self_, other_); |
| 2895 | } |
| 2896 | at::functionalization::impl::replace_(self, tmp_output); |
| 2897 | at::functionalization::impl::commit_update(self); |
| 2898 | at::functionalization::impl::sync(self); |
| 2899 | return self; |
| 2900 | } |
| 2901 | } |
| 2902 | |
| 2903 | at::Tensor & concatenate_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList tensors, int64_t dim, at::Tensor & out) { |
| 2904 | if (false) { |
| 2905 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2906 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2907 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2908 | auto tensors_meta = to_meta(tensors); |
| 2909 | auto out_meta = to_meta(out); |
| 2910 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2911 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2912 | at::_ops::concatenate_out::call(tensors_meta, dim, out_meta); |
| 2913 | } |
| 2914 | |
| 2915 | ::std::vector<at::Tensor> tensors_; |
| 2916 | if (at::functionalization::impl::isFunctionalTensor(tensors)) { |
| 2917 | at::functionalization::impl::sync(tensors); |
| 2918 | tensors_ = at::functionalization::impl::from_functional_tensor(tensors); |
| 2919 | } else { |
| 2920 | tensors_ = tensors.vec(); |
| 2921 | } |
| 2922 | |
| 2923 | at::Tensor out_; |
| 2924 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2925 | at::functionalization::impl::sync(out); |
| 2926 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2927 | } else { |
| 2928 | out_ = out; |
| 2929 | } |
| 2930 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2931 | if ((false || at::functionalization::impl::isFunctionalTensor(tensors))) { |
| 2932 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2933 | TORCH_INTERNAL_ASSERT(false, |
| 2934 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2935 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2936 | } else { |
| 2937 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2938 | at::AutoDispatchSkipFunctionalize guard; |
| 2939 | at::Tensor tmp_output = at::_ops::concatenate_out::call(tensors_, dim, out_); |
| 2940 | return out;; |
| 2941 | } |
| 2942 | } else { |
| 2943 | at::Tensor tmp_output; |
| 2944 | { |
| 2945 | at::AutoDispatchSkipFunctionalize guard; |
| 2946 | tmp_output = at::_ops::concatenate::call(tensors_, dim); |
| 2947 | } |
| 2948 | at::functionalization::impl::replace_(out, tmp_output); |
| 2949 | at::functionalization::impl::commit_update(out); |
| 2950 | at::functionalization::impl::sync(out); |
| 2951 | return out; |
| 2952 | } |
| 2953 | } |
| 2954 | |
| 2955 | at::Tensor & concatenate_out_names_out(c10::DispatchKeySet dispatchKeySet, at::TensorList tensors, at::Dimname dim, at::Tensor & out) { |
| 2956 | if (false) { |
| 2957 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 2958 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 2959 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 2960 | auto tensors_meta = to_meta(tensors); |
| 2961 | auto out_meta = to_meta(out); |
| 2962 | at::AutoDispatchSkipFunctionalize func_guard; |
| 2963 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 2964 | at::_ops::concatenate_names_out::call(tensors_meta, dim, out_meta); |
| 2965 | } |
| 2966 | |
| 2967 | ::std::vector<at::Tensor> tensors_; |
| 2968 | if (at::functionalization::impl::isFunctionalTensor(tensors)) { |
| 2969 | at::functionalization::impl::sync(tensors); |
| 2970 | tensors_ = at::functionalization::impl::from_functional_tensor(tensors); |
| 2971 | } else { |
| 2972 | tensors_ = tensors.vec(); |
| 2973 | } |
| 2974 | |
| 2975 | at::Tensor out_; |
| 2976 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 2977 | at::functionalization::impl::sync(out); |
| 2978 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 2979 | } else { |
| 2980 | out_ = out; |
| 2981 | } |
| 2982 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 2983 | if ((false || at::functionalization::impl::isFunctionalTensor(tensors))) { |
| 2984 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 2985 | TORCH_INTERNAL_ASSERT(false, |
| 2986 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 2987 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 2988 | } else { |
| 2989 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 2990 | at::AutoDispatchSkipFunctionalize guard; |
| 2991 | at::Tensor tmp_output = at::_ops::concatenate_names_out::call(tensors_, dim, out_); |
| 2992 | return out;; |
| 2993 | } |
| 2994 | } else { |
| 2995 | at::Tensor tmp_output; |
| 2996 | { |
| 2997 | at::AutoDispatchSkipFunctionalize guard; |
| 2998 | tmp_output = at::_ops::concatenate_names::call(tensors_, dim); |
| 2999 | } |
| 3000 | at::functionalization::impl::replace_(out, tmp_output); |
| 3001 | at::functionalization::impl::commit_update(out); |
| 3002 | at::functionalization::impl::sync(out); |
| 3003 | return out; |
| 3004 | } |
| 3005 | } |
| 3006 | |
| 3007 | at::Tensor & block_diag_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList tensors, at::Tensor & out) { |
| 3008 | if (false) { |
| 3009 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3010 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3011 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3012 | auto tensors_meta = to_meta(tensors); |
| 3013 | auto out_meta = to_meta(out); |
| 3014 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3015 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3016 | at::_ops::block_diag_out::call(tensors_meta, out_meta); |
| 3017 | } |
| 3018 | |
| 3019 | ::std::vector<at::Tensor> tensors_; |
| 3020 | if (at::functionalization::impl::isFunctionalTensor(tensors)) { |
| 3021 | at::functionalization::impl::sync(tensors); |
| 3022 | tensors_ = at::functionalization::impl::from_functional_tensor(tensors); |
| 3023 | } else { |
| 3024 | tensors_ = tensors.vec(); |
| 3025 | } |
| 3026 | |
| 3027 | at::Tensor out_; |
| 3028 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3029 | at::functionalization::impl::sync(out); |
| 3030 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3031 | } else { |
| 3032 | out_ = out; |
| 3033 | } |
| 3034 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3035 | if ((false || at::functionalization::impl::isFunctionalTensor(tensors))) { |
| 3036 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3037 | TORCH_INTERNAL_ASSERT(false, |
| 3038 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3039 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3040 | } else { |
| 3041 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3042 | at::AutoDispatchSkipFunctionalize guard; |
| 3043 | at::Tensor tmp_output = at::_ops::block_diag_out::call(tensors_, out_); |
| 3044 | return out;; |
| 3045 | } |
| 3046 | } else { |
| 3047 | at::Tensor tmp_output; |
| 3048 | { |
| 3049 | at::AutoDispatchSkipFunctionalize guard; |
| 3050 | tmp_output = at::_ops::block_diag::call(tensors_); |
| 3051 | } |
| 3052 | at::functionalization::impl::replace_(out, tmp_output); |
| 3053 | at::functionalization::impl::commit_update(out); |
| 3054 | at::functionalization::impl::sync(out); |
| 3055 | return out; |
| 3056 | } |
| 3057 | } |
| 3058 | |
| 3059 | at::Tensor & chain_matmul_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList matrices, at::Tensor & out) { |
| 3060 | if (false) { |
| 3061 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3062 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3063 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3064 | auto matrices_meta = to_meta(matrices); |
| 3065 | auto out_meta = to_meta(out); |
| 3066 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3067 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3068 | at::_ops::chain_matmul_out::call(matrices_meta, out_meta); |
| 3069 | } |
| 3070 | |
| 3071 | ::std::vector<at::Tensor> matrices_; |
| 3072 | if (at::functionalization::impl::isFunctionalTensor(matrices)) { |
| 3073 | at::functionalization::impl::sync(matrices); |
| 3074 | matrices_ = at::functionalization::impl::from_functional_tensor(matrices); |
| 3075 | } else { |
| 3076 | matrices_ = matrices.vec(); |
| 3077 | } |
| 3078 | |
| 3079 | at::Tensor out_; |
| 3080 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3081 | at::functionalization::impl::sync(out); |
| 3082 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3083 | } else { |
| 3084 | out_ = out; |
| 3085 | } |
| 3086 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3087 | if ((false || at::functionalization::impl::isFunctionalTensor(matrices))) { |
| 3088 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3089 | TORCH_INTERNAL_ASSERT(false, |
| 3090 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3091 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3092 | } else { |
| 3093 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3094 | at::AutoDispatchSkipFunctionalize guard; |
| 3095 | at::Tensor tmp_output = at::_ops::chain_matmul_out::call(matrices_, out_); |
| 3096 | return out;; |
| 3097 | } |
| 3098 | } else { |
| 3099 | at::Tensor tmp_output; |
| 3100 | { |
| 3101 | at::AutoDispatchSkipFunctionalize guard; |
| 3102 | tmp_output = at::_ops::chain_matmul::call(matrices_); |
| 3103 | } |
| 3104 | at::functionalization::impl::replace_(out, tmp_output); |
| 3105 | at::functionalization::impl::commit_update(out); |
| 3106 | at::functionalization::impl::sync(out); |
| 3107 | return out; |
| 3108 | } |
| 3109 | } |
| 3110 | |
| 3111 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> convolution_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & input, const at::Tensor & weight, at::OptionalSymIntArrayRef bias_sizes, at::IntArrayRef stride, c10::SymIntArrayRef padding, at::IntArrayRef dilation, bool transposed, c10::SymIntArrayRef output_padding, int64_t groups, ::std::array<bool,3> output_mask, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 3112 | if (false) { |
| 3113 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3114 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3115 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3116 | auto grad_output_meta = to_meta(grad_output); |
| 3117 | auto input_meta = to_meta(input); |
| 3118 | auto weight_meta = to_meta(weight); |
| 3119 | auto out0_meta = to_meta(out0); |
| 3120 | auto out1_meta = to_meta(out1); |
| 3121 | auto out2_meta = to_meta(out2); |
| 3122 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3123 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3124 | at::_ops::convolution_backward_out::call(grad_output_meta, input_meta, weight_meta, bias_sizes, stride, padding, dilation, transposed, output_padding, groups, output_mask, out0_meta, out1_meta, out2_meta); |
| 3125 | } |
| 3126 | |
| 3127 | at::Tensor grad_output_; |
| 3128 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 3129 | at::functionalization::impl::sync(grad_output); |
| 3130 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 3131 | } else { |
| 3132 | grad_output_ = grad_output; |
| 3133 | } |
| 3134 | |
| 3135 | at::Tensor input_; |
| 3136 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 3137 | at::functionalization::impl::sync(input); |
| 3138 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 3139 | } else { |
| 3140 | input_ = input; |
| 3141 | } |
| 3142 | |
| 3143 | at::Tensor weight_; |
| 3144 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 3145 | at::functionalization::impl::sync(weight); |
| 3146 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 3147 | } else { |
| 3148 | weight_ = weight; |
| 3149 | } |
| 3150 | |
| 3151 | at::Tensor out0_; |
| 3152 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 3153 | at::functionalization::impl::sync(out0); |
| 3154 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 3155 | } else { |
| 3156 | out0_ = out0; |
| 3157 | } |
| 3158 | |
| 3159 | at::Tensor out1_; |
| 3160 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 3161 | at::functionalization::impl::sync(out1); |
| 3162 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 3163 | } else { |
| 3164 | out1_ = out1; |
| 3165 | } |
| 3166 | |
| 3167 | at::Tensor out2_; |
| 3168 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 3169 | at::functionalization::impl::sync(out2); |
| 3170 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 3171 | } else { |
| 3172 | out2_ = out2; |
| 3173 | } |
| 3174 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 3175 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 3176 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3177 | TORCH_INTERNAL_ASSERT(false, |
| 3178 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3179 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3180 | } else { |
| 3181 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3182 | at::AutoDispatchSkipFunctionalize guard; |
| 3183 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::convolution_backward_out::call(grad_output_, input_, weight_, bias_sizes, stride, padding, dilation, transposed, output_padding, groups, output_mask, out0_, out1_, out2_); |
| 3184 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 3185 | } |
| 3186 | } else { |
| 3187 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 3188 | { |
| 3189 | at::AutoDispatchSkipFunctionalize guard; |
| 3190 | tmp_output = at::_ops::convolution_backward::call(grad_output_, input_, weight_, bias_sizes, stride, padding, dilation, transposed, output_padding, groups, output_mask); |
| 3191 | } |
| 3192 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 3193 | at::functionalization::impl::commit_update(out0); |
| 3194 | at::functionalization::impl::sync(out0); |
| 3195 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 3196 | at::functionalization::impl::commit_update(out1); |
| 3197 | at::functionalization::impl::sync(out1); |
| 3198 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 3199 | at::functionalization::impl::commit_update(out2); |
| 3200 | at::functionalization::impl::sync(out2); |
| 3201 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 3202 | } |
| 3203 | } |
| 3204 | |
| 3205 | at::Tensor & _copy_from_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & dst, bool non_blocking, at::Tensor & out) { |
| 3206 | if (false) { |
| 3207 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3208 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3209 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3210 | auto self_meta = to_meta(self); |
| 3211 | auto dst_meta = to_meta(dst); |
| 3212 | auto out_meta = to_meta(out); |
| 3213 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3214 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3215 | at::_ops::_copy_from_out::call(self_meta, dst_meta, non_blocking, out_meta); |
| 3216 | } |
| 3217 | |
| 3218 | at::Tensor self_; |
| 3219 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3220 | at::functionalization::impl::sync(self); |
| 3221 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3222 | } else { |
| 3223 | self_ = self; |
| 3224 | } |
| 3225 | |
| 3226 | at::Tensor dst_; |
| 3227 | if (at::functionalization::impl::isFunctionalTensor(dst)) { |
| 3228 | at::functionalization::impl::sync(dst); |
| 3229 | dst_ = at::functionalization::impl::from_functional_tensor(dst); |
| 3230 | } else { |
| 3231 | dst_ = dst; |
| 3232 | } |
| 3233 | |
| 3234 | at::Tensor out_; |
| 3235 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3236 | at::functionalization::impl::sync(out); |
| 3237 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3238 | } else { |
| 3239 | out_ = out; |
| 3240 | } |
| 3241 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3242 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(dst))) { |
| 3243 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3244 | TORCH_INTERNAL_ASSERT(false, |
| 3245 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3246 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3247 | } else { |
| 3248 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3249 | at::AutoDispatchSkipFunctionalize guard; |
| 3250 | at::Tensor tmp_output = at::_ops::_copy_from_out::call(self_, dst_, non_blocking, out_); |
| 3251 | return out;; |
| 3252 | } |
| 3253 | } else { |
| 3254 | at::Tensor tmp_output; |
| 3255 | { |
| 3256 | at::AutoDispatchSkipFunctionalize guard; |
| 3257 | tmp_output = at::_ops::_copy_from::call(self_, dst_, non_blocking); |
| 3258 | } |
| 3259 | at::functionalization::impl::replace_(out, tmp_output); |
| 3260 | at::functionalization::impl::commit_update(out); |
| 3261 | at::functionalization::impl::sync(out); |
| 3262 | return out; |
| 3263 | } |
| 3264 | } |
| 3265 | |
| 3266 | at::Tensor & cosh_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 3267 | if (false) { |
| 3268 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3269 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3270 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3271 | auto self_meta = to_meta(self); |
| 3272 | auto out_meta = to_meta(out); |
| 3273 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3274 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3275 | at::_ops::cosh_out::call(self_meta, out_meta); |
| 3276 | } |
| 3277 | |
| 3278 | at::Tensor self_; |
| 3279 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3280 | at::functionalization::impl::sync(self); |
| 3281 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3282 | } else { |
| 3283 | self_ = self; |
| 3284 | } |
| 3285 | |
| 3286 | at::Tensor out_; |
| 3287 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3288 | at::functionalization::impl::sync(out); |
| 3289 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3290 | } else { |
| 3291 | out_ = out; |
| 3292 | } |
| 3293 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3294 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 3295 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3296 | TORCH_INTERNAL_ASSERT(false, |
| 3297 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3298 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3299 | } else { |
| 3300 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3301 | at::AutoDispatchSkipFunctionalize guard; |
| 3302 | at::Tensor tmp_output = at::_ops::cosh_out::call(self_, out_); |
| 3303 | return out;; |
| 3304 | } |
| 3305 | } else { |
| 3306 | at::Tensor tmp_output; |
| 3307 | { |
| 3308 | at::AutoDispatchSkipFunctionalize guard; |
| 3309 | tmp_output = at::_ops::cosh::call(self_); |
| 3310 | } |
| 3311 | at::functionalization::impl::replace_(out, tmp_output); |
| 3312 | at::functionalization::impl::commit_update(out); |
| 3313 | at::functionalization::impl::sync(out); |
| 3314 | return out; |
| 3315 | } |
| 3316 | } |
| 3317 | |
| 3318 | at::Tensor & cosh_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 3319 | if (true) { |
| 3320 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3321 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3322 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3323 | auto self_meta = to_meta(self); |
| 3324 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3325 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3326 | at::_ops::cosh_::call(self_meta); |
| 3327 | } |
| 3328 | |
| 3329 | at::Tensor self_; |
| 3330 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3331 | at::functionalization::impl::sync(self); |
| 3332 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3333 | } else { |
| 3334 | self_ = self; |
| 3335 | } |
| 3336 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 3337 | if ((false)) { |
| 3338 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3339 | TORCH_INTERNAL_ASSERT(false, |
| 3340 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3341 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3342 | } else { |
| 3343 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3344 | at::AutoDispatchSkipFunctionalize guard; |
| 3345 | at::Tensor tmp_output = at::_ops::cosh_::call(self_); |
| 3346 | return self;; |
| 3347 | } |
| 3348 | } else { |
| 3349 | at::Tensor tmp_output; |
| 3350 | { |
| 3351 | at::AutoDispatchSkipFunctionalize guard; |
| 3352 | tmp_output = at::_ops::cosh::call(self_); |
| 3353 | } |
| 3354 | at::functionalization::impl::replace_(self, tmp_output); |
| 3355 | at::functionalization::impl::commit_update(self); |
| 3356 | at::functionalization::impl::sync(self); |
| 3357 | return self; |
| 3358 | } |
| 3359 | } |
| 3360 | |
| 3361 | at::Tensor & cudnn_convolution_transpose_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef padding, at::IntArrayRef output_padding, at::IntArrayRef stride, at::IntArrayRef dilation, int64_t groups, bool benchmark, bool deterministic, bool allow_tf32, at::Tensor & out) { |
| 3362 | if (false) { |
| 3363 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3364 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3365 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3366 | auto self_meta = to_meta(self); |
| 3367 | auto weight_meta = to_meta(weight); |
| 3368 | auto out_meta = to_meta(out); |
| 3369 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3370 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3371 | at::_ops::cudnn_convolution_transpose_out::call(self_meta, weight_meta, padding, output_padding, stride, dilation, groups, benchmark, deterministic, allow_tf32, out_meta); |
| 3372 | } |
| 3373 | |
| 3374 | at::Tensor self_; |
| 3375 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3376 | at::functionalization::impl::sync(self); |
| 3377 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3378 | } else { |
| 3379 | self_ = self; |
| 3380 | } |
| 3381 | |
| 3382 | at::Tensor weight_; |
| 3383 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 3384 | at::functionalization::impl::sync(weight); |
| 3385 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 3386 | } else { |
| 3387 | weight_ = weight; |
| 3388 | } |
| 3389 | |
| 3390 | at::Tensor out_; |
| 3391 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3392 | at::functionalization::impl::sync(out); |
| 3393 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3394 | } else { |
| 3395 | out_ = out; |
| 3396 | } |
| 3397 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3398 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 3399 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3400 | TORCH_INTERNAL_ASSERT(false, |
| 3401 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3402 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3403 | } else { |
| 3404 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3405 | at::AutoDispatchSkipFunctionalize guard; |
| 3406 | at::Tensor tmp_output = at::_ops::cudnn_convolution_transpose_out::call(self_, weight_, padding, output_padding, stride, dilation, groups, benchmark, deterministic, allow_tf32, out_); |
| 3407 | return out;; |
| 3408 | } |
| 3409 | } else { |
| 3410 | at::Tensor tmp_output; |
| 3411 | { |
| 3412 | at::AutoDispatchSkipFunctionalize guard; |
| 3413 | tmp_output = at::_ops::cudnn_convolution_transpose::call(self_, weight_, padding, output_padding, stride, dilation, groups, benchmark, deterministic, allow_tf32); |
| 3414 | } |
| 3415 | at::functionalization::impl::replace_(out, tmp_output); |
| 3416 | at::functionalization::impl::commit_update(out); |
| 3417 | at::functionalization::impl::sync(out); |
| 3418 | return out; |
| 3419 | } |
| 3420 | } |
| 3421 | |
| 3422 | at::Tensor & _mps_convolution_transpose_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef padding, at::IntArrayRef output_padding, at::IntArrayRef stride, at::IntArrayRef dilation, int64_t groups, at::Tensor & out) { |
| 3423 | if (false) { |
| 3424 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3425 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3426 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3427 | auto self_meta = to_meta(self); |
| 3428 | auto weight_meta = to_meta(weight); |
| 3429 | auto out_meta = to_meta(out); |
| 3430 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3431 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3432 | at::_ops::_mps_convolution_transpose_out::call(self_meta, weight_meta, padding, output_padding, stride, dilation, groups, out_meta); |
| 3433 | } |
| 3434 | |
| 3435 | at::Tensor self_; |
| 3436 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3437 | at::functionalization::impl::sync(self); |
| 3438 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3439 | } else { |
| 3440 | self_ = self; |
| 3441 | } |
| 3442 | |
| 3443 | at::Tensor weight_; |
| 3444 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 3445 | at::functionalization::impl::sync(weight); |
| 3446 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 3447 | } else { |
| 3448 | weight_ = weight; |
| 3449 | } |
| 3450 | |
| 3451 | at::Tensor out_; |
| 3452 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3453 | at::functionalization::impl::sync(out); |
| 3454 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3455 | } else { |
| 3456 | out_ = out; |
| 3457 | } |
| 3458 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3459 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 3460 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3461 | TORCH_INTERNAL_ASSERT(false, |
| 3462 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3463 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3464 | } else { |
| 3465 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3466 | at::AutoDispatchSkipFunctionalize guard; |
| 3467 | at::Tensor tmp_output = at::_ops::_mps_convolution_transpose_out::call(self_, weight_, padding, output_padding, stride, dilation, groups, out_); |
| 3468 | return out;; |
| 3469 | } |
| 3470 | } else { |
| 3471 | at::Tensor tmp_output; |
| 3472 | { |
| 3473 | at::AutoDispatchSkipFunctionalize guard; |
| 3474 | tmp_output = at::_ops::_mps_convolution_transpose::call(self_, weight_, padding, output_padding, stride, dilation, groups); |
| 3475 | } |
| 3476 | at::functionalization::impl::replace_(out, tmp_output); |
| 3477 | at::functionalization::impl::commit_update(out); |
| 3478 | at::functionalization::impl::sync(out); |
| 3479 | return out; |
| 3480 | } |
| 3481 | } |
| 3482 | |
| 3483 | at::Tensor & cudnn_grid_sampler_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & grid, at::Tensor & out) { |
| 3484 | if (false) { |
| 3485 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3486 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3487 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3488 | auto self_meta = to_meta(self); |
| 3489 | auto grid_meta = to_meta(grid); |
| 3490 | auto out_meta = to_meta(out); |
| 3491 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3492 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3493 | at::_ops::cudnn_grid_sampler_out::call(self_meta, grid_meta, out_meta); |
| 3494 | } |
| 3495 | |
| 3496 | at::Tensor self_; |
| 3497 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3498 | at::functionalization::impl::sync(self); |
| 3499 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3500 | } else { |
| 3501 | self_ = self; |
| 3502 | } |
| 3503 | |
| 3504 | at::Tensor grid_; |
| 3505 | if (at::functionalization::impl::isFunctionalTensor(grid)) { |
| 3506 | at::functionalization::impl::sync(grid); |
| 3507 | grid_ = at::functionalization::impl::from_functional_tensor(grid); |
| 3508 | } else { |
| 3509 | grid_ = grid; |
| 3510 | } |
| 3511 | |
| 3512 | at::Tensor out_; |
| 3513 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3514 | at::functionalization::impl::sync(out); |
| 3515 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3516 | } else { |
| 3517 | out_ = out; |
| 3518 | } |
| 3519 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3520 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(grid))) { |
| 3521 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3522 | TORCH_INTERNAL_ASSERT(false, |
| 3523 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3524 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3525 | } else { |
| 3526 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3527 | at::AutoDispatchSkipFunctionalize guard; |
| 3528 | at::Tensor tmp_output = at::_ops::cudnn_grid_sampler_out::call(self_, grid_, out_); |
| 3529 | return out;; |
| 3530 | } |
| 3531 | } else { |
| 3532 | at::Tensor tmp_output; |
| 3533 | { |
| 3534 | at::AutoDispatchSkipFunctionalize guard; |
| 3535 | tmp_output = at::_ops::cudnn_grid_sampler::call(self_, grid_); |
| 3536 | } |
| 3537 | at::functionalization::impl::replace_(out, tmp_output); |
| 3538 | at::functionalization::impl::commit_update(out); |
| 3539 | at::functionalization::impl::sync(out); |
| 3540 | return out; |
| 3541 | } |
| 3542 | } |
| 3543 | |
| 3544 | at::Tensor & cumprod_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 3545 | if (false) { |
| 3546 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3547 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3548 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3549 | auto self_meta = to_meta(self); |
| 3550 | auto out_meta = to_meta(out); |
| 3551 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3552 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3553 | at::_ops::cumprod_out::call(self_meta, dim, dtype, out_meta); |
| 3554 | } |
| 3555 | |
| 3556 | at::Tensor self_; |
| 3557 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3558 | at::functionalization::impl::sync(self); |
| 3559 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3560 | } else { |
| 3561 | self_ = self; |
| 3562 | } |
| 3563 | |
| 3564 | at::Tensor out_; |
| 3565 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3566 | at::functionalization::impl::sync(out); |
| 3567 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3568 | } else { |
| 3569 | out_ = out; |
| 3570 | } |
| 3571 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3572 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 3573 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3574 | TORCH_INTERNAL_ASSERT(false, |
| 3575 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3576 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3577 | } else { |
| 3578 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3579 | at::AutoDispatchSkipFunctionalize guard; |
| 3580 | at::Tensor tmp_output = at::_ops::cumprod_out::call(self_, dim, dtype, out_); |
| 3581 | return out;; |
| 3582 | } |
| 3583 | } else { |
| 3584 | at::Tensor tmp_output; |
| 3585 | { |
| 3586 | at::AutoDispatchSkipFunctionalize guard; |
| 3587 | tmp_output = at::_ops::cumprod::call(self_, dim, dtype); |
| 3588 | } |
| 3589 | at::functionalization::impl::replace_(out, tmp_output); |
| 3590 | at::functionalization::impl::commit_update(out); |
| 3591 | at::functionalization::impl::sync(out); |
| 3592 | return out; |
| 3593 | } |
| 3594 | } |
| 3595 | |
| 3596 | at::Tensor & cumprod_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, int64_t dim, c10::optional<at::ScalarType> dtype) { |
| 3597 | if (true) { |
| 3598 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3599 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3600 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3601 | auto self_meta = to_meta(self); |
| 3602 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3603 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3604 | at::_ops::cumprod_::call(self_meta, dim, dtype); |
| 3605 | } |
| 3606 | |
| 3607 | at::Tensor self_; |
| 3608 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3609 | at::functionalization::impl::sync(self); |
| 3610 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3611 | } else { |
| 3612 | self_ = self; |
| 3613 | } |
| 3614 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 3615 | if ((false)) { |
| 3616 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3617 | TORCH_INTERNAL_ASSERT(false, |
| 3618 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3619 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3620 | } else { |
| 3621 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3622 | at::AutoDispatchSkipFunctionalize guard; |
| 3623 | at::Tensor tmp_output = at::_ops::cumprod_::call(self_, dim, dtype); |
| 3624 | return self;; |
| 3625 | } |
| 3626 | } else { |
| 3627 | at::Tensor tmp_output; |
| 3628 | { |
| 3629 | at::AutoDispatchSkipFunctionalize guard; |
| 3630 | tmp_output = at::_ops::cumprod::call(self_, dim, dtype); |
| 3631 | } |
| 3632 | at::functionalization::impl::replace_(self, tmp_output); |
| 3633 | at::functionalization::impl::commit_update(self); |
| 3634 | at::functionalization::impl::sync(self); |
| 3635 | return self; |
| 3636 | } |
| 3637 | } |
| 3638 | |
| 3639 | at::Tensor & cumprod_out_dimname_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Dimname dim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 3640 | if (false) { |
| 3641 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3642 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3643 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3644 | auto self_meta = to_meta(self); |
| 3645 | auto out_meta = to_meta(out); |
| 3646 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3647 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3648 | at::_ops::cumprod_dimname_out::call(self_meta, dim, dtype, out_meta); |
| 3649 | } |
| 3650 | |
| 3651 | at::Tensor self_; |
| 3652 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3653 | at::functionalization::impl::sync(self); |
| 3654 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3655 | } else { |
| 3656 | self_ = self; |
| 3657 | } |
| 3658 | |
| 3659 | at::Tensor out_; |
| 3660 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3661 | at::functionalization::impl::sync(out); |
| 3662 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3663 | } else { |
| 3664 | out_ = out; |
| 3665 | } |
| 3666 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3667 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 3668 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3669 | TORCH_INTERNAL_ASSERT(false, |
| 3670 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3671 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3672 | } else { |
| 3673 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3674 | at::AutoDispatchSkipFunctionalize guard; |
| 3675 | at::Tensor tmp_output = at::_ops::cumprod_dimname_out::call(self_, dim, dtype, out_); |
| 3676 | return out;; |
| 3677 | } |
| 3678 | } else { |
| 3679 | at::Tensor tmp_output; |
| 3680 | { |
| 3681 | at::AutoDispatchSkipFunctionalize guard; |
| 3682 | tmp_output = at::_ops::cumprod_dimname::call(self_, dim, dtype); |
| 3683 | } |
| 3684 | at::functionalization::impl::replace_(out, tmp_output); |
| 3685 | at::functionalization::impl::commit_update(out); |
| 3686 | at::functionalization::impl::sync(out); |
| 3687 | return out; |
| 3688 | } |
| 3689 | } |
| 3690 | |
| 3691 | at::Tensor & cumprod__dimname(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, at::Dimname dim, c10::optional<at::ScalarType> dtype) { |
| 3692 | if (true) { |
| 3693 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3694 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3695 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3696 | auto self_meta = to_meta(self); |
| 3697 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3698 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3699 | at::_ops::cumprod__dimname::call(self_meta, dim, dtype); |
| 3700 | } |
| 3701 | |
| 3702 | at::Tensor self_; |
| 3703 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3704 | at::functionalization::impl::sync(self); |
| 3705 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3706 | } else { |
| 3707 | self_ = self; |
| 3708 | } |
| 3709 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 3710 | if ((false)) { |
| 3711 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3712 | TORCH_INTERNAL_ASSERT(false, |
| 3713 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3714 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3715 | } else { |
| 3716 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3717 | at::AutoDispatchSkipFunctionalize guard; |
| 3718 | at::Tensor tmp_output = at::_ops::cumprod__dimname::call(self_, dim, dtype); |
| 3719 | return self;; |
| 3720 | } |
| 3721 | } else { |
| 3722 | at::Tensor tmp_output; |
| 3723 | { |
| 3724 | at::AutoDispatchSkipFunctionalize guard; |
| 3725 | tmp_output = at::_ops::cumprod_dimname::call(self_, dim, dtype); |
| 3726 | } |
| 3727 | at::functionalization::impl::replace_(self, tmp_output); |
| 3728 | at::functionalization::impl::commit_update(self); |
| 3729 | at::functionalization::impl::sync(self); |
| 3730 | return self; |
| 3731 | } |
| 3732 | } |
| 3733 | |
| 3734 | at::Tensor & diag_embed_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t offset, int64_t dim1, int64_t dim2, at::Tensor & out) { |
| 3735 | if (false) { |
| 3736 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3737 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3738 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3739 | auto self_meta = to_meta(self); |
| 3740 | auto out_meta = to_meta(out); |
| 3741 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3742 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3743 | at::_ops::diag_embed_out::call(self_meta, offset, dim1, dim2, out_meta); |
| 3744 | } |
| 3745 | |
| 3746 | at::Tensor self_; |
| 3747 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3748 | at::functionalization::impl::sync(self); |
| 3749 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3750 | } else { |
| 3751 | self_ = self; |
| 3752 | } |
| 3753 | |
| 3754 | at::Tensor out_; |
| 3755 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3756 | at::functionalization::impl::sync(out); |
| 3757 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3758 | } else { |
| 3759 | out_ = out; |
| 3760 | } |
| 3761 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3762 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 3763 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3764 | TORCH_INTERNAL_ASSERT(false, |
| 3765 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3766 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3767 | } else { |
| 3768 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3769 | at::AutoDispatchSkipFunctionalize guard; |
| 3770 | at::Tensor tmp_output = at::_ops::diag_embed_out::call(self_, offset, dim1, dim2, out_); |
| 3771 | return out;; |
| 3772 | } |
| 3773 | } else { |
| 3774 | at::Tensor tmp_output; |
| 3775 | { |
| 3776 | at::AutoDispatchSkipFunctionalize guard; |
| 3777 | tmp_output = at::_ops::diag_embed::call(self_, offset, dim1, dim2); |
| 3778 | } |
| 3779 | at::functionalization::impl::replace_(out, tmp_output); |
| 3780 | at::functionalization::impl::commit_update(out); |
| 3781 | at::functionalization::impl::sync(out); |
| 3782 | return out; |
| 3783 | } |
| 3784 | } |
| 3785 | |
| 3786 | at::Tensor & diagonal_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, c10::SymIntArrayRef input_sizes, int64_t offset, int64_t dim1, int64_t dim2, at::Tensor & out) { |
| 3787 | if (false) { |
| 3788 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3789 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3790 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3791 | auto grad_output_meta = to_meta(grad_output); |
| 3792 | auto out_meta = to_meta(out); |
| 3793 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3794 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3795 | at::_ops::diagonal_backward_out::call(grad_output_meta, input_sizes, offset, dim1, dim2, out_meta); |
| 3796 | } |
| 3797 | |
| 3798 | at::Tensor grad_output_; |
| 3799 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 3800 | at::functionalization::impl::sync(grad_output); |
| 3801 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 3802 | } else { |
| 3803 | grad_output_ = grad_output; |
| 3804 | } |
| 3805 | |
| 3806 | at::Tensor out_; |
| 3807 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3808 | at::functionalization::impl::sync(out); |
| 3809 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3810 | } else { |
| 3811 | out_ = out; |
| 3812 | } |
| 3813 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3814 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output))) { |
| 3815 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3816 | TORCH_INTERNAL_ASSERT(false, |
| 3817 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3818 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3819 | } else { |
| 3820 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3821 | at::AutoDispatchSkipFunctionalize guard; |
| 3822 | at::Tensor tmp_output = at::_ops::diagonal_backward_out::call(grad_output_, input_sizes, offset, dim1, dim2, out_); |
| 3823 | return out;; |
| 3824 | } |
| 3825 | } else { |
| 3826 | at::Tensor tmp_output; |
| 3827 | { |
| 3828 | at::AutoDispatchSkipFunctionalize guard; |
| 3829 | tmp_output = at::_ops::diagonal_backward::call(grad_output_, input_sizes, offset, dim1, dim2); |
| 3830 | } |
| 3831 | at::functionalization::impl::replace_(out, tmp_output); |
| 3832 | at::functionalization::impl::commit_update(out); |
| 3833 | at::functionalization::impl::sync(out); |
| 3834 | return out; |
| 3835 | } |
| 3836 | } |
| 3837 | |
| 3838 | at::Tensor & div_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 3839 | if (false) { |
| 3840 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3841 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3842 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3843 | auto self_meta = to_meta(self); |
| 3844 | auto other_meta = to_meta(other); |
| 3845 | auto out_meta = to_meta(out); |
| 3846 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3847 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3848 | at::_ops::div_out::call(self_meta, other_meta, out_meta); |
| 3849 | } |
| 3850 | |
| 3851 | at::Tensor self_; |
| 3852 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3853 | at::functionalization::impl::sync(self); |
| 3854 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3855 | } else { |
| 3856 | self_ = self; |
| 3857 | } |
| 3858 | |
| 3859 | at::Tensor other_; |
| 3860 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 3861 | at::functionalization::impl::sync(other); |
| 3862 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 3863 | } else { |
| 3864 | other_ = other; |
| 3865 | } |
| 3866 | |
| 3867 | at::Tensor out_; |
| 3868 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3869 | at::functionalization::impl::sync(out); |
| 3870 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3871 | } else { |
| 3872 | out_ = out; |
| 3873 | } |
| 3874 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3875 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 3876 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3877 | TORCH_INTERNAL_ASSERT(false, |
| 3878 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3879 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3880 | } else { |
| 3881 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3882 | at::AutoDispatchSkipFunctionalize guard; |
| 3883 | at::Tensor tmp_output = at::_ops::div_out::call(self_, other_, out_); |
| 3884 | return out;; |
| 3885 | } |
| 3886 | } else { |
| 3887 | at::Tensor tmp_output; |
| 3888 | { |
| 3889 | at::AutoDispatchSkipFunctionalize guard; |
| 3890 | tmp_output = at::_ops::div_Tensor::call(self_, other_); |
| 3891 | } |
| 3892 | at::functionalization::impl::replace_(out, tmp_output); |
| 3893 | at::functionalization::impl::commit_update(out); |
| 3894 | at::functionalization::impl::sync(out); |
| 3895 | return out; |
| 3896 | } |
| 3897 | } |
| 3898 | |
| 3899 | at::Tensor & div__Tensor(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 3900 | if (true) { |
| 3901 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3902 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3903 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3904 | auto self_meta = to_meta(self); |
| 3905 | auto other_meta = to_meta(other); |
| 3906 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3907 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3908 | at::_ops::div__Tensor::call(self_meta, other_meta); |
| 3909 | } |
| 3910 | |
| 3911 | at::Tensor self_; |
| 3912 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3913 | at::functionalization::impl::sync(self); |
| 3914 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3915 | } else { |
| 3916 | self_ = self; |
| 3917 | } |
| 3918 | |
| 3919 | at::Tensor other_; |
| 3920 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 3921 | at::functionalization::impl::sync(other); |
| 3922 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 3923 | } else { |
| 3924 | other_ = other; |
| 3925 | } |
| 3926 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 3927 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 3928 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3929 | TORCH_INTERNAL_ASSERT(false, |
| 3930 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3931 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3932 | } else { |
| 3933 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3934 | at::AutoDispatchSkipFunctionalize guard; |
| 3935 | at::Tensor tmp_output = at::_ops::div__Tensor::call(self_, other_); |
| 3936 | return self;; |
| 3937 | } |
| 3938 | } else { |
| 3939 | at::Tensor tmp_output; |
| 3940 | { |
| 3941 | at::AutoDispatchSkipFunctionalize guard; |
| 3942 | tmp_output = at::_ops::div_Tensor::call(self_, other_); |
| 3943 | } |
| 3944 | at::functionalization::impl::replace_(self, tmp_output); |
| 3945 | at::functionalization::impl::commit_update(self); |
| 3946 | at::functionalization::impl::sync(self); |
| 3947 | return self; |
| 3948 | } |
| 3949 | } |
| 3950 | |
| 3951 | at::Tensor & div_out_out_mode(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, c10::optional<c10::string_view> rounding_mode, at::Tensor & out) { |
| 3952 | if (false) { |
| 3953 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 3954 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 3955 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 3956 | auto self_meta = to_meta(self); |
| 3957 | auto other_meta = to_meta(other); |
| 3958 | auto out_meta = to_meta(out); |
| 3959 | at::AutoDispatchSkipFunctionalize func_guard; |
| 3960 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 3961 | at::_ops::div_out_mode::call(self_meta, other_meta, rounding_mode, out_meta); |
| 3962 | } |
| 3963 | |
| 3964 | at::Tensor self_; |
| 3965 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 3966 | at::functionalization::impl::sync(self); |
| 3967 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 3968 | } else { |
| 3969 | self_ = self; |
| 3970 | } |
| 3971 | |
| 3972 | at::Tensor other_; |
| 3973 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 3974 | at::functionalization::impl::sync(other); |
| 3975 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 3976 | } else { |
| 3977 | other_ = other; |
| 3978 | } |
| 3979 | |
| 3980 | at::Tensor out_; |
| 3981 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 3982 | at::functionalization::impl::sync(out); |
| 3983 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 3984 | } else { |
| 3985 | out_ = out; |
| 3986 | } |
| 3987 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 3988 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 3989 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 3990 | TORCH_INTERNAL_ASSERT(false, |
| 3991 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 3992 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 3993 | } else { |
| 3994 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 3995 | at::AutoDispatchSkipFunctionalize guard; |
| 3996 | at::Tensor tmp_output = at::_ops::div_out_mode::call(self_, other_, rounding_mode, out_); |
| 3997 | return out;; |
| 3998 | } |
| 3999 | } else { |
| 4000 | at::Tensor tmp_output; |
| 4001 | { |
| 4002 | at::AutoDispatchSkipFunctionalize guard; |
| 4003 | tmp_output = at::_ops::div_Tensor_mode::call(self_, other_, rounding_mode); |
| 4004 | } |
| 4005 | at::functionalization::impl::replace_(out, tmp_output); |
| 4006 | at::functionalization::impl::commit_update(out); |
| 4007 | at::functionalization::impl::sync(out); |
| 4008 | return out; |
| 4009 | } |
| 4010 | } |
| 4011 | |
| 4012 | at::Tensor & div__Tensor_mode(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other, c10::optional<c10::string_view> rounding_mode) { |
| 4013 | if (true) { |
| 4014 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4015 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4016 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4017 | auto self_meta = to_meta(self); |
| 4018 | auto other_meta = to_meta(other); |
| 4019 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4020 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4021 | at::_ops::div__Tensor_mode::call(self_meta, other_meta, rounding_mode); |
| 4022 | } |
| 4023 | |
| 4024 | at::Tensor self_; |
| 4025 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4026 | at::functionalization::impl::sync(self); |
| 4027 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4028 | } else { |
| 4029 | self_ = self; |
| 4030 | } |
| 4031 | |
| 4032 | at::Tensor other_; |
| 4033 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 4034 | at::functionalization::impl::sync(other); |
| 4035 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 4036 | } else { |
| 4037 | other_ = other; |
| 4038 | } |
| 4039 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 4040 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 4041 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4042 | TORCH_INTERNAL_ASSERT(false, |
| 4043 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4044 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4045 | } else { |
| 4046 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4047 | at::AutoDispatchSkipFunctionalize guard; |
| 4048 | at::Tensor tmp_output = at::_ops::div__Tensor_mode::call(self_, other_, rounding_mode); |
| 4049 | return self;; |
| 4050 | } |
| 4051 | } else { |
| 4052 | at::Tensor tmp_output; |
| 4053 | { |
| 4054 | at::AutoDispatchSkipFunctionalize guard; |
| 4055 | tmp_output = at::_ops::div_Tensor_mode::call(self_, other_, rounding_mode); |
| 4056 | } |
| 4057 | at::functionalization::impl::replace_(self, tmp_output); |
| 4058 | at::functionalization::impl::commit_update(self); |
| 4059 | at::functionalization::impl::sync(self); |
| 4060 | return self; |
| 4061 | } |
| 4062 | } |
| 4063 | |
| 4064 | at::Tensor & div_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & other, at::Tensor & out) { |
| 4065 | if (false) { |
| 4066 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4067 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4068 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4069 | auto self_meta = to_meta(self); |
| 4070 | auto out_meta = to_meta(out); |
| 4071 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4072 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4073 | at::_ops::div_Scalar_out::call(self_meta, other, out_meta); |
| 4074 | } |
| 4075 | |
| 4076 | at::Tensor self_; |
| 4077 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4078 | at::functionalization::impl::sync(self); |
| 4079 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4080 | } else { |
| 4081 | self_ = self; |
| 4082 | } |
| 4083 | |
| 4084 | at::Tensor out_; |
| 4085 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4086 | at::functionalization::impl::sync(out); |
| 4087 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4088 | } else { |
| 4089 | out_ = out; |
| 4090 | } |
| 4091 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4092 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 4093 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4094 | TORCH_INTERNAL_ASSERT(false, |
| 4095 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4096 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4097 | } else { |
| 4098 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4099 | at::AutoDispatchSkipFunctionalize guard; |
| 4100 | at::Tensor tmp_output = at::_ops::div_Scalar_out::call(self_, other, out_); |
| 4101 | return out;; |
| 4102 | } |
| 4103 | } else { |
| 4104 | at::Tensor tmp_output; |
| 4105 | { |
| 4106 | at::AutoDispatchSkipFunctionalize guard; |
| 4107 | tmp_output = at::_ops::div_Scalar::call(self_, other); |
| 4108 | } |
| 4109 | at::functionalization::impl::replace_(out, tmp_output); |
| 4110 | at::functionalization::impl::commit_update(out); |
| 4111 | at::functionalization::impl::sync(out); |
| 4112 | return out; |
| 4113 | } |
| 4114 | } |
| 4115 | |
| 4116 | at::Tensor & div__Scalar(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & other) { |
| 4117 | if (true) { |
| 4118 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4119 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4120 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4121 | auto self_meta = to_meta(self); |
| 4122 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4123 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4124 | at::_ops::div__Scalar::call(self_meta, other); |
| 4125 | } |
| 4126 | |
| 4127 | at::Tensor self_; |
| 4128 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4129 | at::functionalization::impl::sync(self); |
| 4130 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4131 | } else { |
| 4132 | self_ = self; |
| 4133 | } |
| 4134 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 4135 | if ((false)) { |
| 4136 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4137 | TORCH_INTERNAL_ASSERT(false, |
| 4138 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4139 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4140 | } else { |
| 4141 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4142 | at::AutoDispatchSkipFunctionalize guard; |
| 4143 | at::Tensor tmp_output = at::_ops::div__Scalar::call(self_, other); |
| 4144 | return self;; |
| 4145 | } |
| 4146 | } else { |
| 4147 | at::Tensor tmp_output; |
| 4148 | { |
| 4149 | at::AutoDispatchSkipFunctionalize guard; |
| 4150 | tmp_output = at::_ops::div_Scalar::call(self_, other); |
| 4151 | } |
| 4152 | at::functionalization::impl::replace_(self, tmp_output); |
| 4153 | at::functionalization::impl::commit_update(self); |
| 4154 | at::functionalization::impl::sync(self); |
| 4155 | return self; |
| 4156 | } |
| 4157 | } |
| 4158 | |
| 4159 | at::Tensor & div_out_Scalar_mode_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & other, c10::optional<c10::string_view> rounding_mode, at::Tensor & out) { |
| 4160 | if (false) { |
| 4161 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4162 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4163 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4164 | auto self_meta = to_meta(self); |
| 4165 | auto out_meta = to_meta(out); |
| 4166 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4167 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4168 | at::_ops::div_Scalar_mode_out::call(self_meta, other, rounding_mode, out_meta); |
| 4169 | } |
| 4170 | |
| 4171 | at::Tensor self_; |
| 4172 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4173 | at::functionalization::impl::sync(self); |
| 4174 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4175 | } else { |
| 4176 | self_ = self; |
| 4177 | } |
| 4178 | |
| 4179 | at::Tensor out_; |
| 4180 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4181 | at::functionalization::impl::sync(out); |
| 4182 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4183 | } else { |
| 4184 | out_ = out; |
| 4185 | } |
| 4186 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4187 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 4188 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4189 | TORCH_INTERNAL_ASSERT(false, |
| 4190 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4191 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4192 | } else { |
| 4193 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4194 | at::AutoDispatchSkipFunctionalize guard; |
| 4195 | at::Tensor tmp_output = at::_ops::div_Scalar_mode_out::call(self_, other, rounding_mode, out_); |
| 4196 | return out;; |
| 4197 | } |
| 4198 | } else { |
| 4199 | at::Tensor tmp_output; |
| 4200 | { |
| 4201 | at::AutoDispatchSkipFunctionalize guard; |
| 4202 | tmp_output = at::_ops::div_Scalar_mode::call(self_, other, rounding_mode); |
| 4203 | } |
| 4204 | at::functionalization::impl::replace_(out, tmp_output); |
| 4205 | at::functionalization::impl::commit_update(out); |
| 4206 | at::functionalization::impl::sync(out); |
| 4207 | return out; |
| 4208 | } |
| 4209 | } |
| 4210 | |
| 4211 | at::Tensor & div__Scalar_mode(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & other, c10::optional<c10::string_view> rounding_mode) { |
| 4212 | if (true) { |
| 4213 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4214 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4215 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4216 | auto self_meta = to_meta(self); |
| 4217 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4218 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4219 | at::_ops::div__Scalar_mode::call(self_meta, other, rounding_mode); |
| 4220 | } |
| 4221 | |
| 4222 | at::Tensor self_; |
| 4223 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4224 | at::functionalization::impl::sync(self); |
| 4225 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4226 | } else { |
| 4227 | self_ = self; |
| 4228 | } |
| 4229 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 4230 | if ((false)) { |
| 4231 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4232 | TORCH_INTERNAL_ASSERT(false, |
| 4233 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4234 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4235 | } else { |
| 4236 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4237 | at::AutoDispatchSkipFunctionalize guard; |
| 4238 | at::Tensor tmp_output = at::_ops::div__Scalar_mode::call(self_, other, rounding_mode); |
| 4239 | return self;; |
| 4240 | } |
| 4241 | } else { |
| 4242 | at::Tensor tmp_output; |
| 4243 | { |
| 4244 | at::AutoDispatchSkipFunctionalize guard; |
| 4245 | tmp_output = at::_ops::div_Scalar_mode::call(self_, other, rounding_mode); |
| 4246 | } |
| 4247 | at::functionalization::impl::replace_(self, tmp_output); |
| 4248 | at::functionalization::impl::commit_update(self); |
| 4249 | at::functionalization::impl::sync(self); |
| 4250 | return self; |
| 4251 | } |
| 4252 | } |
| 4253 | |
| 4254 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &> _embedding_bag_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & weight, const at::Tensor & indices, const at::Tensor & offsets, bool scale_grad_by_freq, int64_t mode, bool sparse, const c10::optional<at::Tensor> & per_sample_weights, bool include_last_offset, int64_t padding_idx, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2, at::Tensor & out3) { |
| 4255 | if (false) { |
| 4256 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4257 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4258 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4259 | auto weight_meta = to_meta(weight); |
| 4260 | auto indices_meta = to_meta(indices); |
| 4261 | auto offsets_meta = to_meta(offsets); |
| 4262 | auto per_sample_weights_meta = to_meta(per_sample_weights); |
| 4263 | auto out0_meta = to_meta(out0); |
| 4264 | auto out1_meta = to_meta(out1); |
| 4265 | auto out2_meta = to_meta(out2); |
| 4266 | auto out3_meta = to_meta(out3); |
| 4267 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4268 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4269 | at::_ops::_embedding_bag_out::call(weight_meta, indices_meta, offsets_meta, scale_grad_by_freq, mode, sparse, per_sample_weights_meta, include_last_offset, padding_idx, out0_meta, out1_meta, out2_meta, out3_meta); |
| 4270 | } |
| 4271 | |
| 4272 | at::Tensor weight_; |
| 4273 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 4274 | at::functionalization::impl::sync(weight); |
| 4275 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 4276 | } else { |
| 4277 | weight_ = weight; |
| 4278 | } |
| 4279 | |
| 4280 | at::Tensor indices_; |
| 4281 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 4282 | at::functionalization::impl::sync(indices); |
| 4283 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 4284 | } else { |
| 4285 | indices_ = indices; |
| 4286 | } |
| 4287 | |
| 4288 | at::Tensor offsets_; |
| 4289 | if (at::functionalization::impl::isFunctionalTensor(offsets)) { |
| 4290 | at::functionalization::impl::sync(offsets); |
| 4291 | offsets_ = at::functionalization::impl::from_functional_tensor(offsets); |
| 4292 | } else { |
| 4293 | offsets_ = offsets; |
| 4294 | } |
| 4295 | |
| 4296 | c10::optional<at::Tensor> per_sample_weights_; |
| 4297 | if (at::functionalization::impl::isFunctionalTensor(per_sample_weights)) { |
| 4298 | at::functionalization::impl::sync(per_sample_weights); |
| 4299 | per_sample_weights_ = at::functionalization::impl::from_functional_tensor(per_sample_weights); |
| 4300 | } else { |
| 4301 | per_sample_weights_ = per_sample_weights; |
| 4302 | } |
| 4303 | |
| 4304 | at::Tensor out0_; |
| 4305 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 4306 | at::functionalization::impl::sync(out0); |
| 4307 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 4308 | } else { |
| 4309 | out0_ = out0; |
| 4310 | } |
| 4311 | |
| 4312 | at::Tensor out1_; |
| 4313 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 4314 | at::functionalization::impl::sync(out1); |
| 4315 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 4316 | } else { |
| 4317 | out1_ = out1; |
| 4318 | } |
| 4319 | |
| 4320 | at::Tensor out2_; |
| 4321 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 4322 | at::functionalization::impl::sync(out2); |
| 4323 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 4324 | } else { |
| 4325 | out2_ = out2; |
| 4326 | } |
| 4327 | |
| 4328 | at::Tensor out3_; |
| 4329 | if (at::functionalization::impl::isFunctionalTensor(out3)) { |
| 4330 | at::functionalization::impl::sync(out3); |
| 4331 | out3_ = at::functionalization::impl::from_functional_tensor(out3); |
| 4332 | } else { |
| 4333 | out3_ = out3; |
| 4334 | } |
| 4335 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2) && at::functionalization::impl::isFunctionalTensor(out3))) { |
| 4336 | if ((false || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(indices) || at::functionalization::impl::isFunctionalTensor(offsets) || at::functionalization::impl::isFunctionalTensor(per_sample_weights))) { |
| 4337 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4338 | TORCH_INTERNAL_ASSERT(false, |
| 4339 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4340 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4341 | } else { |
| 4342 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4343 | at::AutoDispatchSkipFunctionalize guard; |
| 4344 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_embedding_bag_out::call(weight_, indices_, offsets_, scale_grad_by_freq, mode, sparse, per_sample_weights_, include_last_offset, padding_idx, out0_, out1_, out2_, out3_); |
| 4345 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2, out3);; |
| 4346 | } |
| 4347 | } else { |
| 4348 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 4349 | { |
| 4350 | at::AutoDispatchSkipFunctionalize guard; |
| 4351 | tmp_output = at::_ops::_embedding_bag::call(weight_, indices_, offsets_, scale_grad_by_freq, mode, sparse, per_sample_weights_, include_last_offset, padding_idx); |
| 4352 | } |
| 4353 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 4354 | at::functionalization::impl::commit_update(out0); |
| 4355 | at::functionalization::impl::sync(out0); |
| 4356 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 4357 | at::functionalization::impl::commit_update(out1); |
| 4358 | at::functionalization::impl::sync(out1); |
| 4359 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 4360 | at::functionalization::impl::commit_update(out2); |
| 4361 | at::functionalization::impl::sync(out2); |
| 4362 | at::functionalization::impl::replace_(out3, std::get<3>(tmp_output)); |
| 4363 | at::functionalization::impl::commit_update(out3); |
| 4364 | at::functionalization::impl::sync(out3); |
| 4365 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2, out3); |
| 4366 | } |
| 4367 | } |
| 4368 | |
| 4369 | at::Tensor & _embedding_bag_per_sample_weights_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad, const at::Tensor & weight, const at::Tensor & indices, const at::Tensor & offsets, const at::Tensor & offset2bag, int64_t mode, int64_t padding_idx, at::Tensor & out) { |
| 4370 | if (false) { |
| 4371 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4372 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4373 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4374 | auto grad_meta = to_meta(grad); |
| 4375 | auto weight_meta = to_meta(weight); |
| 4376 | auto indices_meta = to_meta(indices); |
| 4377 | auto offsets_meta = to_meta(offsets); |
| 4378 | auto offset2bag_meta = to_meta(offset2bag); |
| 4379 | auto out_meta = to_meta(out); |
| 4380 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4381 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4382 | at::_ops::_embedding_bag_per_sample_weights_backward_out::call(grad_meta, weight_meta, indices_meta, offsets_meta, offset2bag_meta, mode, padding_idx, out_meta); |
| 4383 | } |
| 4384 | |
| 4385 | at::Tensor grad_; |
| 4386 | if (at::functionalization::impl::isFunctionalTensor(grad)) { |
| 4387 | at::functionalization::impl::sync(grad); |
| 4388 | grad_ = at::functionalization::impl::from_functional_tensor(grad); |
| 4389 | } else { |
| 4390 | grad_ = grad; |
| 4391 | } |
| 4392 | |
| 4393 | at::Tensor weight_; |
| 4394 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 4395 | at::functionalization::impl::sync(weight); |
| 4396 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 4397 | } else { |
| 4398 | weight_ = weight; |
| 4399 | } |
| 4400 | |
| 4401 | at::Tensor indices_; |
| 4402 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 4403 | at::functionalization::impl::sync(indices); |
| 4404 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 4405 | } else { |
| 4406 | indices_ = indices; |
| 4407 | } |
| 4408 | |
| 4409 | at::Tensor offsets_; |
| 4410 | if (at::functionalization::impl::isFunctionalTensor(offsets)) { |
| 4411 | at::functionalization::impl::sync(offsets); |
| 4412 | offsets_ = at::functionalization::impl::from_functional_tensor(offsets); |
| 4413 | } else { |
| 4414 | offsets_ = offsets; |
| 4415 | } |
| 4416 | |
| 4417 | at::Tensor offset2bag_; |
| 4418 | if (at::functionalization::impl::isFunctionalTensor(offset2bag)) { |
| 4419 | at::functionalization::impl::sync(offset2bag); |
| 4420 | offset2bag_ = at::functionalization::impl::from_functional_tensor(offset2bag); |
| 4421 | } else { |
| 4422 | offset2bag_ = offset2bag; |
| 4423 | } |
| 4424 | |
| 4425 | at::Tensor out_; |
| 4426 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4427 | at::functionalization::impl::sync(out); |
| 4428 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4429 | } else { |
| 4430 | out_ = out; |
| 4431 | } |
| 4432 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4433 | if ((false || at::functionalization::impl::isFunctionalTensor(grad) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(indices) || at::functionalization::impl::isFunctionalTensor(offsets) || at::functionalization::impl::isFunctionalTensor(offset2bag))) { |
| 4434 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4435 | TORCH_INTERNAL_ASSERT(false, |
| 4436 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4437 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4438 | } else { |
| 4439 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4440 | at::AutoDispatchSkipFunctionalize guard; |
| 4441 | at::Tensor tmp_output = at::_ops::_embedding_bag_per_sample_weights_backward_out::call(grad_, weight_, indices_, offsets_, offset2bag_, mode, padding_idx, out_); |
| 4442 | return out;; |
| 4443 | } |
| 4444 | } else { |
| 4445 | at::Tensor tmp_output; |
| 4446 | { |
| 4447 | at::AutoDispatchSkipFunctionalize guard; |
| 4448 | tmp_output = at::_ops::_embedding_bag_per_sample_weights_backward::call(grad_, weight_, indices_, offsets_, offset2bag_, mode, padding_idx); |
| 4449 | } |
| 4450 | at::functionalization::impl::replace_(out, tmp_output); |
| 4451 | at::functionalization::impl::commit_update(out); |
| 4452 | at::functionalization::impl::sync(out); |
| 4453 | return out; |
| 4454 | } |
| 4455 | } |
| 4456 | |
| 4457 | at::Tensor & new_full_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::SymIntArrayRef size, const at::Scalar & fill_value, at::Tensor & out) { |
| 4458 | if (false) { |
| 4459 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4460 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4461 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4462 | auto self_meta = to_meta(self); |
| 4463 | auto out_meta = to_meta(out); |
| 4464 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4465 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4466 | at::_ops::new_full_out::call(self_meta, size, fill_value, out_meta); |
| 4467 | } |
| 4468 | |
| 4469 | at::Tensor self_; |
| 4470 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4471 | at::functionalization::impl::sync(self); |
| 4472 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4473 | } else { |
| 4474 | self_ = self; |
| 4475 | } |
| 4476 | |
| 4477 | at::Tensor out_; |
| 4478 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4479 | at::functionalization::impl::sync(out); |
| 4480 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4481 | } else { |
| 4482 | out_ = out; |
| 4483 | } |
| 4484 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4485 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 4486 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4487 | TORCH_INTERNAL_ASSERT(false, |
| 4488 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4489 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4490 | } else { |
| 4491 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4492 | at::AutoDispatchSkipFunctionalize guard; |
| 4493 | at::Tensor tmp_output = at::_ops::new_full_out::call(self_, size, fill_value, out_); |
| 4494 | return out;; |
| 4495 | } |
| 4496 | } else { |
| 4497 | at::Tensor tmp_output; |
| 4498 | { |
| 4499 | at::AutoDispatchSkipFunctionalize guard; |
| 4500 | tmp_output = at::_ops::new_full::call(self_, size, fill_value, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 4501 | } |
| 4502 | at::functionalization::impl::replace_(out, tmp_output); |
| 4503 | at::functionalization::impl::commit_update(out); |
| 4504 | at::functionalization::impl::sync(out); |
| 4505 | return out; |
| 4506 | } |
| 4507 | } |
| 4508 | |
| 4509 | at::Tensor & empty_quantized_out_out(c10::DispatchKeySet dispatchKeySet, at::IntArrayRef size, const at::Tensor & qtensor, c10::optional<at::MemoryFormat> memory_format, at::Tensor & out) { |
| 4510 | if (false) { |
| 4511 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4512 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4513 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4514 | auto qtensor_meta = to_meta(qtensor); |
| 4515 | auto out_meta = to_meta(out); |
| 4516 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4517 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4518 | at::_ops::empty_quantized_out::call(size, qtensor_meta, memory_format, out_meta); |
| 4519 | } |
| 4520 | |
| 4521 | at::Tensor qtensor_; |
| 4522 | if (at::functionalization::impl::isFunctionalTensor(qtensor)) { |
| 4523 | at::functionalization::impl::sync(qtensor); |
| 4524 | qtensor_ = at::functionalization::impl::from_functional_tensor(qtensor); |
| 4525 | } else { |
| 4526 | qtensor_ = qtensor; |
| 4527 | } |
| 4528 | |
| 4529 | at::Tensor out_; |
| 4530 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4531 | at::functionalization::impl::sync(out); |
| 4532 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4533 | } else { |
| 4534 | out_ = out; |
| 4535 | } |
| 4536 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4537 | if ((false || at::functionalization::impl::isFunctionalTensor(qtensor))) { |
| 4538 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4539 | TORCH_INTERNAL_ASSERT(false, |
| 4540 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4541 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4542 | } else { |
| 4543 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4544 | at::AutoDispatchSkipFunctionalize guard; |
| 4545 | at::Tensor tmp_output = at::_ops::empty_quantized_out::call(size, qtensor_, memory_format, out_); |
| 4546 | return out;; |
| 4547 | } |
| 4548 | } else { |
| 4549 | at::Tensor tmp_output; |
| 4550 | { |
| 4551 | at::AutoDispatchSkipFunctionalize guard; |
| 4552 | tmp_output = at::_ops::empty_quantized::call(size, qtensor_, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt, memory_format); |
| 4553 | } |
| 4554 | at::functionalization::impl::replace_(out, tmp_output); |
| 4555 | at::functionalization::impl::commit_update(out); |
| 4556 | at::functionalization::impl::sync(out); |
| 4557 | return out; |
| 4558 | } |
| 4559 | } |
| 4560 | |
| 4561 | at::Tensor & empty_strided_out_out(c10::DispatchKeySet dispatchKeySet, c10::SymIntArrayRef size, c10::SymIntArrayRef stride, at::Tensor & out) { |
| 4562 | if (false) { |
| 4563 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4564 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4565 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4566 | auto out_meta = to_meta(out); |
| 4567 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4568 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4569 | at::_ops::empty_strided_out::call(size, stride, out_meta); |
| 4570 | } |
| 4571 | |
| 4572 | at::Tensor out_; |
| 4573 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4574 | at::functionalization::impl::sync(out); |
| 4575 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4576 | } else { |
| 4577 | out_ = out; |
| 4578 | } |
| 4579 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4580 | if ((false)) { |
| 4581 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4582 | TORCH_INTERNAL_ASSERT(false, |
| 4583 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4584 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4585 | } else { |
| 4586 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4587 | at::AutoDispatchSkipFunctionalize guard; |
| 4588 | at::Tensor tmp_output = at::_ops::empty_strided_out::call(size, stride, out_); |
| 4589 | return out;; |
| 4590 | } |
| 4591 | } else { |
| 4592 | at::Tensor tmp_output; |
| 4593 | { |
| 4594 | at::AutoDispatchSkipFunctionalize guard; |
| 4595 | tmp_output = at::_ops::empty_strided::call(size, stride, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 4596 | } |
| 4597 | at::functionalization::impl::replace_(out, tmp_output); |
| 4598 | at::functionalization::impl::commit_update(out); |
| 4599 | at::functionalization::impl::sync(out); |
| 4600 | return out; |
| 4601 | } |
| 4602 | } |
| 4603 | |
| 4604 | at::Tensor & exp_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 4605 | if (false) { |
| 4606 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4607 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4608 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4609 | auto self_meta = to_meta(self); |
| 4610 | auto out_meta = to_meta(out); |
| 4611 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4612 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4613 | at::_ops::exp_out::call(self_meta, out_meta); |
| 4614 | } |
| 4615 | |
| 4616 | at::Tensor self_; |
| 4617 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4618 | at::functionalization::impl::sync(self); |
| 4619 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4620 | } else { |
| 4621 | self_ = self; |
| 4622 | } |
| 4623 | |
| 4624 | at::Tensor out_; |
| 4625 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4626 | at::functionalization::impl::sync(out); |
| 4627 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4628 | } else { |
| 4629 | out_ = out; |
| 4630 | } |
| 4631 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4632 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 4633 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4634 | TORCH_INTERNAL_ASSERT(false, |
| 4635 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4636 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4637 | } else { |
| 4638 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4639 | at::AutoDispatchSkipFunctionalize guard; |
| 4640 | at::Tensor tmp_output = at::_ops::exp_out::call(self_, out_); |
| 4641 | return out;; |
| 4642 | } |
| 4643 | } else { |
| 4644 | at::Tensor tmp_output; |
| 4645 | { |
| 4646 | at::AutoDispatchSkipFunctionalize guard; |
| 4647 | tmp_output = at::_ops::exp::call(self_); |
| 4648 | } |
| 4649 | at::functionalization::impl::replace_(out, tmp_output); |
| 4650 | at::functionalization::impl::commit_update(out); |
| 4651 | at::functionalization::impl::sync(out); |
| 4652 | return out; |
| 4653 | } |
| 4654 | } |
| 4655 | |
| 4656 | at::Tensor & exp_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 4657 | if (true) { |
| 4658 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4659 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4660 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4661 | auto self_meta = to_meta(self); |
| 4662 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4663 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4664 | at::_ops::exp_::call(self_meta); |
| 4665 | } |
| 4666 | |
| 4667 | at::Tensor self_; |
| 4668 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4669 | at::functionalization::impl::sync(self); |
| 4670 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4671 | } else { |
| 4672 | self_ = self; |
| 4673 | } |
| 4674 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 4675 | if ((false)) { |
| 4676 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4677 | TORCH_INTERNAL_ASSERT(false, |
| 4678 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4679 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4680 | } else { |
| 4681 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4682 | at::AutoDispatchSkipFunctionalize guard; |
| 4683 | at::Tensor tmp_output = at::_ops::exp_::call(self_); |
| 4684 | return self;; |
| 4685 | } |
| 4686 | } else { |
| 4687 | at::Tensor tmp_output; |
| 4688 | { |
| 4689 | at::AutoDispatchSkipFunctionalize guard; |
| 4690 | tmp_output = at::_ops::exp::call(self_); |
| 4691 | } |
| 4692 | at::functionalization::impl::replace_(self, tmp_output); |
| 4693 | at::functionalization::impl::commit_update(self); |
| 4694 | at::functionalization::impl::sync(self); |
| 4695 | return self; |
| 4696 | } |
| 4697 | } |
| 4698 | |
| 4699 | at::Tensor & expm1_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 4700 | if (false) { |
| 4701 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4702 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4703 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4704 | auto self_meta = to_meta(self); |
| 4705 | auto out_meta = to_meta(out); |
| 4706 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4707 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4708 | at::_ops::expm1_out::call(self_meta, out_meta); |
| 4709 | } |
| 4710 | |
| 4711 | at::Tensor self_; |
| 4712 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4713 | at::functionalization::impl::sync(self); |
| 4714 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4715 | } else { |
| 4716 | self_ = self; |
| 4717 | } |
| 4718 | |
| 4719 | at::Tensor out_; |
| 4720 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4721 | at::functionalization::impl::sync(out); |
| 4722 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4723 | } else { |
| 4724 | out_ = out; |
| 4725 | } |
| 4726 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4727 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 4728 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4729 | TORCH_INTERNAL_ASSERT(false, |
| 4730 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4731 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4732 | } else { |
| 4733 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4734 | at::AutoDispatchSkipFunctionalize guard; |
| 4735 | at::Tensor tmp_output = at::_ops::expm1_out::call(self_, out_); |
| 4736 | return out;; |
| 4737 | } |
| 4738 | } else { |
| 4739 | at::Tensor tmp_output; |
| 4740 | { |
| 4741 | at::AutoDispatchSkipFunctionalize guard; |
| 4742 | tmp_output = at::_ops::expm1::call(self_); |
| 4743 | } |
| 4744 | at::functionalization::impl::replace_(out, tmp_output); |
| 4745 | at::functionalization::impl::commit_update(out); |
| 4746 | at::functionalization::impl::sync(out); |
| 4747 | return out; |
| 4748 | } |
| 4749 | } |
| 4750 | |
| 4751 | at::Tensor & expm1_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 4752 | if (true) { |
| 4753 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4754 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4755 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4756 | auto self_meta = to_meta(self); |
| 4757 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4758 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4759 | at::_ops::expm1_::call(self_meta); |
| 4760 | } |
| 4761 | |
| 4762 | at::Tensor self_; |
| 4763 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4764 | at::functionalization::impl::sync(self); |
| 4765 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4766 | } else { |
| 4767 | self_ = self; |
| 4768 | } |
| 4769 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 4770 | if ((false)) { |
| 4771 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4772 | TORCH_INTERNAL_ASSERT(false, |
| 4773 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4774 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4775 | } else { |
| 4776 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4777 | at::AutoDispatchSkipFunctionalize guard; |
| 4778 | at::Tensor tmp_output = at::_ops::expm1_::call(self_); |
| 4779 | return self;; |
| 4780 | } |
| 4781 | } else { |
| 4782 | at::Tensor tmp_output; |
| 4783 | { |
| 4784 | at::AutoDispatchSkipFunctionalize guard; |
| 4785 | tmp_output = at::_ops::expm1::call(self_); |
| 4786 | } |
| 4787 | at::functionalization::impl::replace_(self, tmp_output); |
| 4788 | at::functionalization::impl::commit_update(self); |
| 4789 | at::functionalization::impl::sync(self); |
| 4790 | return self; |
| 4791 | } |
| 4792 | } |
| 4793 | |
| 4794 | at::Tensor & fill_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & value, at::Tensor & out) { |
| 4795 | if (false) { |
| 4796 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4797 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4798 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4799 | auto self_meta = to_meta(self); |
| 4800 | auto out_meta = to_meta(out); |
| 4801 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4802 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4803 | at::_ops::fill_Scalar_out::call(self_meta, value, out_meta); |
| 4804 | } |
| 4805 | |
| 4806 | at::Tensor self_; |
| 4807 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4808 | at::functionalization::impl::sync(self); |
| 4809 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4810 | } else { |
| 4811 | self_ = self; |
| 4812 | } |
| 4813 | |
| 4814 | at::Tensor out_; |
| 4815 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4816 | at::functionalization::impl::sync(out); |
| 4817 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4818 | } else { |
| 4819 | out_ = out; |
| 4820 | } |
| 4821 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4822 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 4823 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4824 | TORCH_INTERNAL_ASSERT(false, |
| 4825 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4826 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4827 | } else { |
| 4828 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4829 | at::AutoDispatchSkipFunctionalize guard; |
| 4830 | at::Tensor tmp_output = at::_ops::fill_Scalar_out::call(self_, value, out_); |
| 4831 | return out;; |
| 4832 | } |
| 4833 | } else { |
| 4834 | at::Tensor tmp_output; |
| 4835 | { |
| 4836 | at::AutoDispatchSkipFunctionalize guard; |
| 4837 | tmp_output = at::_ops::fill_Scalar::call(self_, value); |
| 4838 | } |
| 4839 | at::functionalization::impl::replace_(out, tmp_output); |
| 4840 | at::functionalization::impl::commit_update(out); |
| 4841 | at::functionalization::impl::sync(out); |
| 4842 | return out; |
| 4843 | } |
| 4844 | } |
| 4845 | |
| 4846 | at::Tensor & fill__Scalar(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & value) { |
| 4847 | if (true) { |
| 4848 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4849 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4850 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4851 | auto self_meta = to_meta(self); |
| 4852 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4853 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4854 | at::_ops::fill__Scalar::call(self_meta, value); |
| 4855 | } |
| 4856 | |
| 4857 | at::Tensor self_; |
| 4858 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4859 | at::functionalization::impl::sync(self); |
| 4860 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4861 | } else { |
| 4862 | self_ = self; |
| 4863 | } |
| 4864 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 4865 | if ((false)) { |
| 4866 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4867 | TORCH_INTERNAL_ASSERT(false, |
| 4868 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4869 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4870 | } else { |
| 4871 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4872 | at::AutoDispatchSkipFunctionalize guard; |
| 4873 | at::Tensor tmp_output = at::_ops::fill__Scalar::call(self_, value); |
| 4874 | return self;; |
| 4875 | } |
| 4876 | } else { |
| 4877 | at::Tensor tmp_output; |
| 4878 | { |
| 4879 | at::AutoDispatchSkipFunctionalize guard; |
| 4880 | tmp_output = at::_ops::fill_Scalar::call(self_, value); |
| 4881 | } |
| 4882 | at::functionalization::impl::replace_(self, tmp_output); |
| 4883 | at::functionalization::impl::commit_update(self); |
| 4884 | at::functionalization::impl::sync(self); |
| 4885 | return self; |
| 4886 | } |
| 4887 | } |
| 4888 | |
| 4889 | at::Tensor & fill_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & value, at::Tensor & out) { |
| 4890 | if (false) { |
| 4891 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4892 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4893 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4894 | auto self_meta = to_meta(self); |
| 4895 | auto value_meta = to_meta(value); |
| 4896 | auto out_meta = to_meta(out); |
| 4897 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4898 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4899 | at::_ops::fill_Tensor_out::call(self_meta, value_meta, out_meta); |
| 4900 | } |
| 4901 | |
| 4902 | at::Tensor self_; |
| 4903 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4904 | at::functionalization::impl::sync(self); |
| 4905 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4906 | } else { |
| 4907 | self_ = self; |
| 4908 | } |
| 4909 | |
| 4910 | at::Tensor value_; |
| 4911 | if (at::functionalization::impl::isFunctionalTensor(value)) { |
| 4912 | at::functionalization::impl::sync(value); |
| 4913 | value_ = at::functionalization::impl::from_functional_tensor(value); |
| 4914 | } else { |
| 4915 | value_ = value; |
| 4916 | } |
| 4917 | |
| 4918 | at::Tensor out_; |
| 4919 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 4920 | at::functionalization::impl::sync(out); |
| 4921 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 4922 | } else { |
| 4923 | out_ = out; |
| 4924 | } |
| 4925 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 4926 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(value))) { |
| 4927 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4928 | TORCH_INTERNAL_ASSERT(false, |
| 4929 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4930 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4931 | } else { |
| 4932 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4933 | at::AutoDispatchSkipFunctionalize guard; |
| 4934 | at::Tensor tmp_output = at::_ops::fill_Tensor_out::call(self_, value_, out_); |
| 4935 | return out;; |
| 4936 | } |
| 4937 | } else { |
| 4938 | at::Tensor tmp_output; |
| 4939 | { |
| 4940 | at::AutoDispatchSkipFunctionalize guard; |
| 4941 | tmp_output = at::_ops::fill_Tensor::call(self_, value_); |
| 4942 | } |
| 4943 | at::functionalization::impl::replace_(out, tmp_output); |
| 4944 | at::functionalization::impl::commit_update(out); |
| 4945 | at::functionalization::impl::sync(out); |
| 4946 | return out; |
| 4947 | } |
| 4948 | } |
| 4949 | |
| 4950 | at::Tensor & fill__Tensor(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & value) { |
| 4951 | if (true) { |
| 4952 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 4953 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 4954 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 4955 | auto self_meta = to_meta(self); |
| 4956 | auto value_meta = to_meta(value); |
| 4957 | at::AutoDispatchSkipFunctionalize func_guard; |
| 4958 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 4959 | at::_ops::fill__Tensor::call(self_meta, value_meta); |
| 4960 | } |
| 4961 | |
| 4962 | at::Tensor self_; |
| 4963 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 4964 | at::functionalization::impl::sync(self); |
| 4965 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 4966 | } else { |
| 4967 | self_ = self; |
| 4968 | } |
| 4969 | |
| 4970 | at::Tensor value_; |
| 4971 | if (at::functionalization::impl::isFunctionalTensor(value)) { |
| 4972 | at::functionalization::impl::sync(value); |
| 4973 | value_ = at::functionalization::impl::from_functional_tensor(value); |
| 4974 | } else { |
| 4975 | value_ = value; |
| 4976 | } |
| 4977 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 4978 | if ((false || at::functionalization::impl::isFunctionalTensor(value))) { |
| 4979 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 4980 | TORCH_INTERNAL_ASSERT(false, |
| 4981 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 4982 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 4983 | } else { |
| 4984 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 4985 | at::AutoDispatchSkipFunctionalize guard; |
| 4986 | at::Tensor tmp_output = at::_ops::fill__Tensor::call(self_, value_); |
| 4987 | return self;; |
| 4988 | } |
| 4989 | } else { |
| 4990 | at::Tensor tmp_output; |
| 4991 | { |
| 4992 | at::AutoDispatchSkipFunctionalize guard; |
| 4993 | tmp_output = at::_ops::fill_Tensor::call(self_, value_); |
| 4994 | } |
| 4995 | at::functionalization::impl::replace_(self, tmp_output); |
| 4996 | at::functionalization::impl::commit_update(self); |
| 4997 | at::functionalization::impl::sync(self); |
| 4998 | return self; |
| 4999 | } |
| 5000 | } |
| 5001 | |
| 5002 | at::Tensor & floor_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 5003 | if (false) { |
| 5004 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5005 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5006 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5007 | auto self_meta = to_meta(self); |
| 5008 | auto out_meta = to_meta(out); |
| 5009 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5010 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5011 | at::_ops::floor_out::call(self_meta, out_meta); |
| 5012 | } |
| 5013 | |
| 5014 | at::Tensor self_; |
| 5015 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5016 | at::functionalization::impl::sync(self); |
| 5017 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5018 | } else { |
| 5019 | self_ = self; |
| 5020 | } |
| 5021 | |
| 5022 | at::Tensor out_; |
| 5023 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5024 | at::functionalization::impl::sync(out); |
| 5025 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5026 | } else { |
| 5027 | out_ = out; |
| 5028 | } |
| 5029 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5030 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 5031 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5032 | TORCH_INTERNAL_ASSERT(false, |
| 5033 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5034 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5035 | } else { |
| 5036 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5037 | at::AutoDispatchSkipFunctionalize guard; |
| 5038 | at::Tensor tmp_output = at::_ops::floor_out::call(self_, out_); |
| 5039 | return out;; |
| 5040 | } |
| 5041 | } else { |
| 5042 | at::Tensor tmp_output; |
| 5043 | { |
| 5044 | at::AutoDispatchSkipFunctionalize guard; |
| 5045 | tmp_output = at::_ops::floor::call(self_); |
| 5046 | } |
| 5047 | at::functionalization::impl::replace_(out, tmp_output); |
| 5048 | at::functionalization::impl::commit_update(out); |
| 5049 | at::functionalization::impl::sync(out); |
| 5050 | return out; |
| 5051 | } |
| 5052 | } |
| 5053 | |
| 5054 | at::Tensor & floor_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 5055 | if (true) { |
| 5056 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5057 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5058 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5059 | auto self_meta = to_meta(self); |
| 5060 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5061 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5062 | at::_ops::floor_::call(self_meta); |
| 5063 | } |
| 5064 | |
| 5065 | at::Tensor self_; |
| 5066 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5067 | at::functionalization::impl::sync(self); |
| 5068 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5069 | } else { |
| 5070 | self_ = self; |
| 5071 | } |
| 5072 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 5073 | if ((false)) { |
| 5074 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5075 | TORCH_INTERNAL_ASSERT(false, |
| 5076 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5077 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5078 | } else { |
| 5079 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5080 | at::AutoDispatchSkipFunctionalize guard; |
| 5081 | at::Tensor tmp_output = at::_ops::floor_::call(self_); |
| 5082 | return self;; |
| 5083 | } |
| 5084 | } else { |
| 5085 | at::Tensor tmp_output; |
| 5086 | { |
| 5087 | at::AutoDispatchSkipFunctionalize guard; |
| 5088 | tmp_output = at::_ops::floor::call(self_); |
| 5089 | } |
| 5090 | at::functionalization::impl::replace_(self, tmp_output); |
| 5091 | at::functionalization::impl::commit_update(self); |
| 5092 | at::functionalization::impl::sync(self); |
| 5093 | return self; |
| 5094 | } |
| 5095 | } |
| 5096 | |
| 5097 | at::Tensor & floor_divide_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 5098 | if (false) { |
| 5099 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5100 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5101 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5102 | auto self_meta = to_meta(self); |
| 5103 | auto other_meta = to_meta(other); |
| 5104 | auto out_meta = to_meta(out); |
| 5105 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5106 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5107 | at::_ops::floor_divide_out::call(self_meta, other_meta, out_meta); |
| 5108 | } |
| 5109 | |
| 5110 | at::Tensor self_; |
| 5111 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5112 | at::functionalization::impl::sync(self); |
| 5113 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5114 | } else { |
| 5115 | self_ = self; |
| 5116 | } |
| 5117 | |
| 5118 | at::Tensor other_; |
| 5119 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 5120 | at::functionalization::impl::sync(other); |
| 5121 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 5122 | } else { |
| 5123 | other_ = other; |
| 5124 | } |
| 5125 | |
| 5126 | at::Tensor out_; |
| 5127 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5128 | at::functionalization::impl::sync(out); |
| 5129 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5130 | } else { |
| 5131 | out_ = out; |
| 5132 | } |
| 5133 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5134 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 5135 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5136 | TORCH_INTERNAL_ASSERT(false, |
| 5137 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5138 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5139 | } else { |
| 5140 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5141 | at::AutoDispatchSkipFunctionalize guard; |
| 5142 | at::Tensor tmp_output = at::_ops::floor_divide_out::call(self_, other_, out_); |
| 5143 | return out;; |
| 5144 | } |
| 5145 | } else { |
| 5146 | at::Tensor tmp_output; |
| 5147 | { |
| 5148 | at::AutoDispatchSkipFunctionalize guard; |
| 5149 | tmp_output = at::_ops::floor_divide::call(self_, other_); |
| 5150 | } |
| 5151 | at::functionalization::impl::replace_(out, tmp_output); |
| 5152 | at::functionalization::impl::commit_update(out); |
| 5153 | at::functionalization::impl::sync(out); |
| 5154 | return out; |
| 5155 | } |
| 5156 | } |
| 5157 | |
| 5158 | at::Tensor & floor_divide__Tensor(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 5159 | if (true) { |
| 5160 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5161 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5162 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5163 | auto self_meta = to_meta(self); |
| 5164 | auto other_meta = to_meta(other); |
| 5165 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5166 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5167 | at::_ops::floor_divide__Tensor::call(self_meta, other_meta); |
| 5168 | } |
| 5169 | |
| 5170 | at::Tensor self_; |
| 5171 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5172 | at::functionalization::impl::sync(self); |
| 5173 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5174 | } else { |
| 5175 | self_ = self; |
| 5176 | } |
| 5177 | |
| 5178 | at::Tensor other_; |
| 5179 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 5180 | at::functionalization::impl::sync(other); |
| 5181 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 5182 | } else { |
| 5183 | other_ = other; |
| 5184 | } |
| 5185 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 5186 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 5187 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5188 | TORCH_INTERNAL_ASSERT(false, |
| 5189 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5190 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5191 | } else { |
| 5192 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5193 | at::AutoDispatchSkipFunctionalize guard; |
| 5194 | at::Tensor tmp_output = at::_ops::floor_divide__Tensor::call(self_, other_); |
| 5195 | return self;; |
| 5196 | } |
| 5197 | } else { |
| 5198 | at::Tensor tmp_output; |
| 5199 | { |
| 5200 | at::AutoDispatchSkipFunctionalize guard; |
| 5201 | tmp_output = at::_ops::floor_divide::call(self_, other_); |
| 5202 | } |
| 5203 | at::functionalization::impl::replace_(self, tmp_output); |
| 5204 | at::functionalization::impl::commit_update(self); |
| 5205 | at::functionalization::impl::sync(self); |
| 5206 | return self; |
| 5207 | } |
| 5208 | } |
| 5209 | |
| 5210 | at::Tensor & full_out_names_out(c10::DispatchKeySet dispatchKeySet, at::IntArrayRef size, const at::Scalar & fill_value, c10::optional<at::DimnameList> names, at::Tensor & out) { |
| 5211 | if (false) { |
| 5212 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5213 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5214 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5215 | auto out_meta = to_meta(out); |
| 5216 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5217 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5218 | at::_ops::full_names_out::call(size, fill_value, names, out_meta); |
| 5219 | } |
| 5220 | |
| 5221 | at::Tensor out_; |
| 5222 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5223 | at::functionalization::impl::sync(out); |
| 5224 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5225 | } else { |
| 5226 | out_ = out; |
| 5227 | } |
| 5228 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5229 | if ((false)) { |
| 5230 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5231 | TORCH_INTERNAL_ASSERT(false, |
| 5232 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5233 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5234 | } else { |
| 5235 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5236 | at::AutoDispatchSkipFunctionalize guard; |
| 5237 | at::Tensor tmp_output = at::_ops::full_names_out::call(size, fill_value, names, out_); |
| 5238 | return out;; |
| 5239 | } |
| 5240 | } else { |
| 5241 | at::Tensor tmp_output; |
| 5242 | { |
| 5243 | at::AutoDispatchSkipFunctionalize guard; |
| 5244 | tmp_output = at::_ops::full_names::call(size, fill_value, names, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 5245 | } |
| 5246 | at::functionalization::impl::replace_(out, tmp_output); |
| 5247 | at::functionalization::impl::commit_update(out); |
| 5248 | at::functionalization::impl::sync(out); |
| 5249 | return out; |
| 5250 | } |
| 5251 | } |
| 5252 | |
| 5253 | at::Tensor & full_out_out(c10::DispatchKeySet dispatchKeySet, c10::SymIntArrayRef size, const at::Scalar & fill_value, at::Tensor & out) { |
| 5254 | if (false) { |
| 5255 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5256 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5257 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5258 | auto out_meta = to_meta(out); |
| 5259 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5260 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5261 | at::_ops::full_out::call(size, fill_value, out_meta); |
| 5262 | } |
| 5263 | |
| 5264 | at::Tensor out_; |
| 5265 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5266 | at::functionalization::impl::sync(out); |
| 5267 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5268 | } else { |
| 5269 | out_ = out; |
| 5270 | } |
| 5271 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5272 | if ((false)) { |
| 5273 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5274 | TORCH_INTERNAL_ASSERT(false, |
| 5275 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5276 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5277 | } else { |
| 5278 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5279 | at::AutoDispatchSkipFunctionalize guard; |
| 5280 | at::Tensor tmp_output = at::_ops::full_out::call(size, fill_value, out_); |
| 5281 | return out;; |
| 5282 | } |
| 5283 | } else { |
| 5284 | at::Tensor tmp_output; |
| 5285 | { |
| 5286 | at::AutoDispatchSkipFunctionalize guard; |
| 5287 | tmp_output = at::_ops::full::call(size, fill_value, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 5288 | } |
| 5289 | at::functionalization::impl::replace_(out, tmp_output); |
| 5290 | at::functionalization::impl::commit_update(out); |
| 5291 | at::functionalization::impl::sync(out); |
| 5292 | return out; |
| 5293 | } |
| 5294 | } |
| 5295 | |
| 5296 | at::Tensor & from_file_out_out(c10::DispatchKeySet dispatchKeySet, c10::string_view filename, c10::optional<bool> shared, c10::optional<int64_t> size, at::Tensor & out) { |
| 5297 | if (false) { |
| 5298 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5299 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5300 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5301 | auto out_meta = to_meta(out); |
| 5302 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5303 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5304 | at::_ops::from_file_out::call(filename, shared, size, out_meta); |
| 5305 | } |
| 5306 | |
| 5307 | at::Tensor out_; |
| 5308 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5309 | at::functionalization::impl::sync(out); |
| 5310 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5311 | } else { |
| 5312 | out_ = out; |
| 5313 | } |
| 5314 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5315 | if ((false)) { |
| 5316 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5317 | TORCH_INTERNAL_ASSERT(false, |
| 5318 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5319 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5320 | } else { |
| 5321 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5322 | at::AutoDispatchSkipFunctionalize guard; |
| 5323 | at::Tensor tmp_output = at::_ops::from_file_out::call(filename, shared, size, out_); |
| 5324 | return out;; |
| 5325 | } |
| 5326 | } else { |
| 5327 | at::Tensor tmp_output; |
| 5328 | { |
| 5329 | at::AutoDispatchSkipFunctionalize guard; |
| 5330 | tmp_output = at::_ops::from_file::call(filename, shared, size, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 5331 | } |
| 5332 | at::functionalization::impl::replace_(out, tmp_output); |
| 5333 | at::functionalization::impl::commit_update(out); |
| 5334 | at::functionalization::impl::sync(out); |
| 5335 | return out; |
| 5336 | } |
| 5337 | } |
| 5338 | |
| 5339 | at::Tensor & grid_sampler_2d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, const at::Tensor & grid, int64_t interpolation_mode, int64_t padding_mode, bool align_corners, at::Tensor & out) { |
| 5340 | if (false) { |
| 5341 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5342 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5343 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5344 | auto input_meta = to_meta(input); |
| 5345 | auto grid_meta = to_meta(grid); |
| 5346 | auto out_meta = to_meta(out); |
| 5347 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5348 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5349 | at::_ops::grid_sampler_2d_out::call(input_meta, grid_meta, interpolation_mode, padding_mode, align_corners, out_meta); |
| 5350 | } |
| 5351 | |
| 5352 | at::Tensor input_; |
| 5353 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 5354 | at::functionalization::impl::sync(input); |
| 5355 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 5356 | } else { |
| 5357 | input_ = input; |
| 5358 | } |
| 5359 | |
| 5360 | at::Tensor grid_; |
| 5361 | if (at::functionalization::impl::isFunctionalTensor(grid)) { |
| 5362 | at::functionalization::impl::sync(grid); |
| 5363 | grid_ = at::functionalization::impl::from_functional_tensor(grid); |
| 5364 | } else { |
| 5365 | grid_ = grid; |
| 5366 | } |
| 5367 | |
| 5368 | at::Tensor out_; |
| 5369 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5370 | at::functionalization::impl::sync(out); |
| 5371 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5372 | } else { |
| 5373 | out_ = out; |
| 5374 | } |
| 5375 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5376 | if ((false || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(grid))) { |
| 5377 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5378 | TORCH_INTERNAL_ASSERT(false, |
| 5379 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5380 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5381 | } else { |
| 5382 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5383 | at::AutoDispatchSkipFunctionalize guard; |
| 5384 | at::Tensor tmp_output = at::_ops::grid_sampler_2d_out::call(input_, grid_, interpolation_mode, padding_mode, align_corners, out_); |
| 5385 | return out;; |
| 5386 | } |
| 5387 | } else { |
| 5388 | at::Tensor tmp_output; |
| 5389 | { |
| 5390 | at::AutoDispatchSkipFunctionalize guard; |
| 5391 | tmp_output = at::_ops::grid_sampler_2d::call(input_, grid_, interpolation_mode, padding_mode, align_corners); |
| 5392 | } |
| 5393 | at::functionalization::impl::replace_(out, tmp_output); |
| 5394 | at::functionalization::impl::commit_update(out); |
| 5395 | at::functionalization::impl::sync(out); |
| 5396 | return out; |
| 5397 | } |
| 5398 | } |
| 5399 | |
| 5400 | at::Tensor & grid_sampler_3d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, const at::Tensor & grid, int64_t interpolation_mode, int64_t padding_mode, bool align_corners, at::Tensor & out) { |
| 5401 | if (false) { |
| 5402 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5403 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5404 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5405 | auto input_meta = to_meta(input); |
| 5406 | auto grid_meta = to_meta(grid); |
| 5407 | auto out_meta = to_meta(out); |
| 5408 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5409 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5410 | at::_ops::grid_sampler_3d_out::call(input_meta, grid_meta, interpolation_mode, padding_mode, align_corners, out_meta); |
| 5411 | } |
| 5412 | |
| 5413 | at::Tensor input_; |
| 5414 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 5415 | at::functionalization::impl::sync(input); |
| 5416 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 5417 | } else { |
| 5418 | input_ = input; |
| 5419 | } |
| 5420 | |
| 5421 | at::Tensor grid_; |
| 5422 | if (at::functionalization::impl::isFunctionalTensor(grid)) { |
| 5423 | at::functionalization::impl::sync(grid); |
| 5424 | grid_ = at::functionalization::impl::from_functional_tensor(grid); |
| 5425 | } else { |
| 5426 | grid_ = grid; |
| 5427 | } |
| 5428 | |
| 5429 | at::Tensor out_; |
| 5430 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5431 | at::functionalization::impl::sync(out); |
| 5432 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5433 | } else { |
| 5434 | out_ = out; |
| 5435 | } |
| 5436 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5437 | if ((false || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(grid))) { |
| 5438 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5439 | TORCH_INTERNAL_ASSERT(false, |
| 5440 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5441 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5442 | } else { |
| 5443 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5444 | at::AutoDispatchSkipFunctionalize guard; |
| 5445 | at::Tensor tmp_output = at::_ops::grid_sampler_3d_out::call(input_, grid_, interpolation_mode, padding_mode, align_corners, out_); |
| 5446 | return out;; |
| 5447 | } |
| 5448 | } else { |
| 5449 | at::Tensor tmp_output; |
| 5450 | { |
| 5451 | at::AutoDispatchSkipFunctionalize guard; |
| 5452 | tmp_output = at::_ops::grid_sampler_3d::call(input_, grid_, interpolation_mode, padding_mode, align_corners); |
| 5453 | } |
| 5454 | at::functionalization::impl::replace_(out, tmp_output); |
| 5455 | at::functionalization::impl::commit_update(out); |
| 5456 | at::functionalization::impl::sync(out); |
| 5457 | return out; |
| 5458 | } |
| 5459 | } |
| 5460 | |
| 5461 | at::Tensor & hamming_window_out_out(c10::DispatchKeySet dispatchKeySet, int64_t window_length, at::Tensor & out) { |
| 5462 | if (false) { |
| 5463 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5464 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5465 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5466 | auto out_meta = to_meta(out); |
| 5467 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5468 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5469 | at::_ops::hamming_window_out::call(window_length, out_meta); |
| 5470 | } |
| 5471 | |
| 5472 | at::Tensor out_; |
| 5473 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5474 | at::functionalization::impl::sync(out); |
| 5475 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5476 | } else { |
| 5477 | out_ = out; |
| 5478 | } |
| 5479 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5480 | if ((false)) { |
| 5481 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5482 | TORCH_INTERNAL_ASSERT(false, |
| 5483 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5484 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5485 | } else { |
| 5486 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5487 | at::AutoDispatchSkipFunctionalize guard; |
| 5488 | at::Tensor tmp_output = at::_ops::hamming_window_out::call(window_length, out_); |
| 5489 | return out;; |
| 5490 | } |
| 5491 | } else { |
| 5492 | at::Tensor tmp_output; |
| 5493 | { |
| 5494 | at::AutoDispatchSkipFunctionalize guard; |
| 5495 | tmp_output = at::_ops::hamming_window::call(window_length, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 5496 | } |
| 5497 | at::functionalization::impl::replace_(out, tmp_output); |
| 5498 | at::functionalization::impl::commit_update(out); |
| 5499 | at::functionalization::impl::sync(out); |
| 5500 | return out; |
| 5501 | } |
| 5502 | } |
| 5503 | |
| 5504 | at::Tensor & hamming_window_out_periodic_out(c10::DispatchKeySet dispatchKeySet, int64_t window_length, bool periodic, at::Tensor & out) { |
| 5505 | if (false) { |
| 5506 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5507 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5508 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5509 | auto out_meta = to_meta(out); |
| 5510 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5511 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5512 | at::_ops::hamming_window_periodic_out::call(window_length, periodic, out_meta); |
| 5513 | } |
| 5514 | |
| 5515 | at::Tensor out_; |
| 5516 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5517 | at::functionalization::impl::sync(out); |
| 5518 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5519 | } else { |
| 5520 | out_ = out; |
| 5521 | } |
| 5522 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5523 | if ((false)) { |
| 5524 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5525 | TORCH_INTERNAL_ASSERT(false, |
| 5526 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5527 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5528 | } else { |
| 5529 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5530 | at::AutoDispatchSkipFunctionalize guard; |
| 5531 | at::Tensor tmp_output = at::_ops::hamming_window_periodic_out::call(window_length, periodic, out_); |
| 5532 | return out;; |
| 5533 | } |
| 5534 | } else { |
| 5535 | at::Tensor tmp_output; |
| 5536 | { |
| 5537 | at::AutoDispatchSkipFunctionalize guard; |
| 5538 | tmp_output = at::_ops::hamming_window_periodic::call(window_length, periodic, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 5539 | } |
| 5540 | at::functionalization::impl::replace_(out, tmp_output); |
| 5541 | at::functionalization::impl::commit_update(out); |
| 5542 | at::functionalization::impl::sync(out); |
| 5543 | return out; |
| 5544 | } |
| 5545 | } |
| 5546 | |
| 5547 | at::Tensor & hamming_window_out_periodic_alpha_out(c10::DispatchKeySet dispatchKeySet, int64_t window_length, bool periodic, double alpha, at::Tensor & out) { |
| 5548 | if (false) { |
| 5549 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5550 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5551 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5552 | auto out_meta = to_meta(out); |
| 5553 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5554 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5555 | at::_ops::hamming_window_periodic_alpha_out::call(window_length, periodic, alpha, out_meta); |
| 5556 | } |
| 5557 | |
| 5558 | at::Tensor out_; |
| 5559 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5560 | at::functionalization::impl::sync(out); |
| 5561 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5562 | } else { |
| 5563 | out_ = out; |
| 5564 | } |
| 5565 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5566 | if ((false)) { |
| 5567 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5568 | TORCH_INTERNAL_ASSERT(false, |
| 5569 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5570 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5571 | } else { |
| 5572 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5573 | at::AutoDispatchSkipFunctionalize guard; |
| 5574 | at::Tensor tmp_output = at::_ops::hamming_window_periodic_alpha_out::call(window_length, periodic, alpha, out_); |
| 5575 | return out;; |
| 5576 | } |
| 5577 | } else { |
| 5578 | at::Tensor tmp_output; |
| 5579 | { |
| 5580 | at::AutoDispatchSkipFunctionalize guard; |
| 5581 | tmp_output = at::_ops::hamming_window_periodic_alpha::call(window_length, periodic, alpha, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 5582 | } |
| 5583 | at::functionalization::impl::replace_(out, tmp_output); |
| 5584 | at::functionalization::impl::commit_update(out); |
| 5585 | at::functionalization::impl::sync(out); |
| 5586 | return out; |
| 5587 | } |
| 5588 | } |
| 5589 | |
| 5590 | at::Tensor & hamming_window_out_periodic_alpha_beta_out(c10::DispatchKeySet dispatchKeySet, int64_t window_length, bool periodic, double alpha, double beta, at::Tensor & out) { |
| 5591 | if (false) { |
| 5592 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5593 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5594 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5595 | auto out_meta = to_meta(out); |
| 5596 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5597 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5598 | at::_ops::hamming_window_periodic_alpha_beta_out::call(window_length, periodic, alpha, beta, out_meta); |
| 5599 | } |
| 5600 | |
| 5601 | at::Tensor out_; |
| 5602 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5603 | at::functionalization::impl::sync(out); |
| 5604 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5605 | } else { |
| 5606 | out_ = out; |
| 5607 | } |
| 5608 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5609 | if ((false)) { |
| 5610 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5611 | TORCH_INTERNAL_ASSERT(false, |
| 5612 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5613 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5614 | } else { |
| 5615 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5616 | at::AutoDispatchSkipFunctionalize guard; |
| 5617 | at::Tensor tmp_output = at::_ops::hamming_window_periodic_alpha_beta_out::call(window_length, periodic, alpha, beta, out_); |
| 5618 | return out;; |
| 5619 | } |
| 5620 | } else { |
| 5621 | at::Tensor tmp_output; |
| 5622 | { |
| 5623 | at::AutoDispatchSkipFunctionalize guard; |
| 5624 | tmp_output = at::_ops::hamming_window_periodic_alpha_beta::call(window_length, periodic, alpha, beta, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 5625 | } |
| 5626 | at::functionalization::impl::replace_(out, tmp_output); |
| 5627 | at::functionalization::impl::commit_update(out); |
| 5628 | at::functionalization::impl::sync(out); |
| 5629 | return out; |
| 5630 | } |
| 5631 | } |
| 5632 | |
| 5633 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> native_group_norm_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, const c10::optional<at::Tensor> & weight, const c10::optional<at::Tensor> & bias, c10::SymInt N, c10::SymInt C, c10::SymInt HxW, int64_t group, double eps, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 5634 | if (false) { |
| 5635 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5636 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5637 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5638 | auto input_meta = to_meta(input); |
| 5639 | auto weight_meta = to_meta(weight); |
| 5640 | auto bias_meta = to_meta(bias); |
| 5641 | auto out0_meta = to_meta(out0); |
| 5642 | auto out1_meta = to_meta(out1); |
| 5643 | auto out2_meta = to_meta(out2); |
| 5644 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5645 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5646 | at::_ops::native_group_norm_out::call(input_meta, weight_meta, bias_meta, N, C, HxW, group, eps, out0_meta, out1_meta, out2_meta); |
| 5647 | } |
| 5648 | |
| 5649 | at::Tensor input_; |
| 5650 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 5651 | at::functionalization::impl::sync(input); |
| 5652 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 5653 | } else { |
| 5654 | input_ = input; |
| 5655 | } |
| 5656 | |
| 5657 | c10::optional<at::Tensor> weight_; |
| 5658 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 5659 | at::functionalization::impl::sync(weight); |
| 5660 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 5661 | } else { |
| 5662 | weight_ = weight; |
| 5663 | } |
| 5664 | |
| 5665 | c10::optional<at::Tensor> bias_; |
| 5666 | if (at::functionalization::impl::isFunctionalTensor(bias)) { |
| 5667 | at::functionalization::impl::sync(bias); |
| 5668 | bias_ = at::functionalization::impl::from_functional_tensor(bias); |
| 5669 | } else { |
| 5670 | bias_ = bias; |
| 5671 | } |
| 5672 | |
| 5673 | at::Tensor out0_; |
| 5674 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 5675 | at::functionalization::impl::sync(out0); |
| 5676 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 5677 | } else { |
| 5678 | out0_ = out0; |
| 5679 | } |
| 5680 | |
| 5681 | at::Tensor out1_; |
| 5682 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 5683 | at::functionalization::impl::sync(out1); |
| 5684 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 5685 | } else { |
| 5686 | out1_ = out1; |
| 5687 | } |
| 5688 | |
| 5689 | at::Tensor out2_; |
| 5690 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 5691 | at::functionalization::impl::sync(out2); |
| 5692 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 5693 | } else { |
| 5694 | out2_ = out2; |
| 5695 | } |
| 5696 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 5697 | if ((false || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(bias))) { |
| 5698 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5699 | TORCH_INTERNAL_ASSERT(false, |
| 5700 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5701 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5702 | } else { |
| 5703 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5704 | at::AutoDispatchSkipFunctionalize guard; |
| 5705 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::native_group_norm_out::call(input_, weight_, bias_, N, C, HxW, group, eps, out0_, out1_, out2_); |
| 5706 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 5707 | } |
| 5708 | } else { |
| 5709 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 5710 | { |
| 5711 | at::AutoDispatchSkipFunctionalize guard; |
| 5712 | tmp_output = at::_ops::native_group_norm::call(input_, weight_, bias_, N, C, HxW, group, eps); |
| 5713 | } |
| 5714 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 5715 | at::functionalization::impl::commit_update(out0); |
| 5716 | at::functionalization::impl::sync(out0); |
| 5717 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 5718 | at::functionalization::impl::commit_update(out1); |
| 5719 | at::functionalization::impl::sync(out1); |
| 5720 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 5721 | at::functionalization::impl::commit_update(out2); |
| 5722 | at::functionalization::impl::sync(out2); |
| 5723 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 5724 | } |
| 5725 | } |
| 5726 | |
| 5727 | at::Tensor & _fft_c2r_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef dim, int64_t normalization, int64_t last_dim_size, at::Tensor & out) { |
| 5728 | if (false) { |
| 5729 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5730 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5731 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5732 | auto self_meta = to_meta(self); |
| 5733 | auto out_meta = to_meta(out); |
| 5734 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5735 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5736 | at::_ops::_fft_c2r_out::call(self_meta, dim, normalization, last_dim_size, out_meta); |
| 5737 | } |
| 5738 | |
| 5739 | at::Tensor self_; |
| 5740 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5741 | at::functionalization::impl::sync(self); |
| 5742 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5743 | } else { |
| 5744 | self_ = self; |
| 5745 | } |
| 5746 | |
| 5747 | at::Tensor out_; |
| 5748 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5749 | at::functionalization::impl::sync(out); |
| 5750 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5751 | } else { |
| 5752 | out_ = out; |
| 5753 | } |
| 5754 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5755 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 5756 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5757 | TORCH_INTERNAL_ASSERT(false, |
| 5758 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5759 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5760 | } else { |
| 5761 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5762 | at::AutoDispatchSkipFunctionalize guard; |
| 5763 | at::Tensor tmp_output = at::_ops::_fft_c2r_out::call(self_, dim, normalization, last_dim_size, out_); |
| 5764 | return out;; |
| 5765 | } |
| 5766 | } else { |
| 5767 | at::Tensor tmp_output; |
| 5768 | { |
| 5769 | at::AutoDispatchSkipFunctionalize guard; |
| 5770 | tmp_output = at::_ops::_fft_c2r::call(self_, dim, normalization, last_dim_size); |
| 5771 | } |
| 5772 | at::functionalization::impl::replace_(out, tmp_output); |
| 5773 | at::functionalization::impl::commit_update(out); |
| 5774 | at::functionalization::impl::sync(out); |
| 5775 | return out; |
| 5776 | } |
| 5777 | } |
| 5778 | |
| 5779 | at::Tensor & isnan_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 5780 | if (false) { |
| 5781 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5782 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5783 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5784 | auto self_meta = to_meta(self); |
| 5785 | auto out_meta = to_meta(out); |
| 5786 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5787 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5788 | at::_ops::isnan_out::call(self_meta, out_meta); |
| 5789 | } |
| 5790 | |
| 5791 | at::Tensor self_; |
| 5792 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5793 | at::functionalization::impl::sync(self); |
| 5794 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5795 | } else { |
| 5796 | self_ = self; |
| 5797 | } |
| 5798 | |
| 5799 | at::Tensor out_; |
| 5800 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5801 | at::functionalization::impl::sync(out); |
| 5802 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5803 | } else { |
| 5804 | out_ = out; |
| 5805 | } |
| 5806 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5807 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 5808 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5809 | TORCH_INTERNAL_ASSERT(false, |
| 5810 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5811 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5812 | } else { |
| 5813 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5814 | at::AutoDispatchSkipFunctionalize guard; |
| 5815 | at::Tensor tmp_output = at::_ops::isnan_out::call(self_, out_); |
| 5816 | return out;; |
| 5817 | } |
| 5818 | } else { |
| 5819 | at::Tensor tmp_output; |
| 5820 | { |
| 5821 | at::AutoDispatchSkipFunctionalize guard; |
| 5822 | tmp_output = at::_ops::isnan::call(self_); |
| 5823 | } |
| 5824 | at::functionalization::impl::replace_(out, tmp_output); |
| 5825 | at::functionalization::impl::commit_update(out); |
| 5826 | at::functionalization::impl::sync(out); |
| 5827 | return out; |
| 5828 | } |
| 5829 | } |
| 5830 | |
| 5831 | at::Tensor & ldexp_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 5832 | if (false) { |
| 5833 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5834 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5835 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5836 | auto self_meta = to_meta(self); |
| 5837 | auto other_meta = to_meta(other); |
| 5838 | auto out_meta = to_meta(out); |
| 5839 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5840 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5841 | at::_ops::ldexp_out::call(self_meta, other_meta, out_meta); |
| 5842 | } |
| 5843 | |
| 5844 | at::Tensor self_; |
| 5845 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5846 | at::functionalization::impl::sync(self); |
| 5847 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5848 | } else { |
| 5849 | self_ = self; |
| 5850 | } |
| 5851 | |
| 5852 | at::Tensor other_; |
| 5853 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 5854 | at::functionalization::impl::sync(other); |
| 5855 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 5856 | } else { |
| 5857 | other_ = other; |
| 5858 | } |
| 5859 | |
| 5860 | at::Tensor out_; |
| 5861 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5862 | at::functionalization::impl::sync(out); |
| 5863 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5864 | } else { |
| 5865 | out_ = out; |
| 5866 | } |
| 5867 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5868 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 5869 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5870 | TORCH_INTERNAL_ASSERT(false, |
| 5871 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5872 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5873 | } else { |
| 5874 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5875 | at::AutoDispatchSkipFunctionalize guard; |
| 5876 | at::Tensor tmp_output = at::_ops::ldexp_out::call(self_, other_, out_); |
| 5877 | return out;; |
| 5878 | } |
| 5879 | } else { |
| 5880 | at::Tensor tmp_output; |
| 5881 | { |
| 5882 | at::AutoDispatchSkipFunctionalize guard; |
| 5883 | tmp_output = at::_ops::ldexp_Tensor::call(self_, other_); |
| 5884 | } |
| 5885 | at::functionalization::impl::replace_(out, tmp_output); |
| 5886 | at::functionalization::impl::commit_update(out); |
| 5887 | at::functionalization::impl::sync(out); |
| 5888 | return out; |
| 5889 | } |
| 5890 | } |
| 5891 | |
| 5892 | at::Tensor & ldexp_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 5893 | if (true) { |
| 5894 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5895 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5896 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5897 | auto self_meta = to_meta(self); |
| 5898 | auto other_meta = to_meta(other); |
| 5899 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5900 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5901 | at::_ops::ldexp_::call(self_meta, other_meta); |
| 5902 | } |
| 5903 | |
| 5904 | at::Tensor self_; |
| 5905 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5906 | at::functionalization::impl::sync(self); |
| 5907 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5908 | } else { |
| 5909 | self_ = self; |
| 5910 | } |
| 5911 | |
| 5912 | at::Tensor other_; |
| 5913 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 5914 | at::functionalization::impl::sync(other); |
| 5915 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 5916 | } else { |
| 5917 | other_ = other; |
| 5918 | } |
| 5919 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 5920 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 5921 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5922 | TORCH_INTERNAL_ASSERT(false, |
| 5923 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5924 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5925 | } else { |
| 5926 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5927 | at::AutoDispatchSkipFunctionalize guard; |
| 5928 | at::Tensor tmp_output = at::_ops::ldexp_::call(self_, other_); |
| 5929 | return self;; |
| 5930 | } |
| 5931 | } else { |
| 5932 | at::Tensor tmp_output; |
| 5933 | { |
| 5934 | at::AutoDispatchSkipFunctionalize guard; |
| 5935 | tmp_output = at::_ops::ldexp_Tensor::call(self_, other_); |
| 5936 | } |
| 5937 | at::functionalization::impl::replace_(self, tmp_output); |
| 5938 | at::functionalization::impl::commit_update(self); |
| 5939 | at::functionalization::impl::sync(self); |
| 5940 | return self; |
| 5941 | } |
| 5942 | } |
| 5943 | |
| 5944 | at::Tensor & log2_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 5945 | if (false) { |
| 5946 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5947 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 5948 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 5949 | auto self_meta = to_meta(self); |
| 5950 | auto out_meta = to_meta(out); |
| 5951 | at::AutoDispatchSkipFunctionalize func_guard; |
| 5952 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 5953 | at::_ops::log2_out::call(self_meta, out_meta); |
| 5954 | } |
| 5955 | |
| 5956 | at::Tensor self_; |
| 5957 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 5958 | at::functionalization::impl::sync(self); |
| 5959 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 5960 | } else { |
| 5961 | self_ = self; |
| 5962 | } |
| 5963 | |
| 5964 | at::Tensor out_; |
| 5965 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 5966 | at::functionalization::impl::sync(out); |
| 5967 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 5968 | } else { |
| 5969 | out_ = out; |
| 5970 | } |
| 5971 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 5972 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 5973 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 5974 | TORCH_INTERNAL_ASSERT(false, |
| 5975 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 5976 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 5977 | } else { |
| 5978 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 5979 | at::AutoDispatchSkipFunctionalize guard; |
| 5980 | at::Tensor tmp_output = at::_ops::log2_out::call(self_, out_); |
| 5981 | return out;; |
| 5982 | } |
| 5983 | } else { |
| 5984 | at::Tensor tmp_output; |
| 5985 | { |
| 5986 | at::AutoDispatchSkipFunctionalize guard; |
| 5987 | tmp_output = at::_ops::log2::call(self_); |
| 5988 | } |
| 5989 | at::functionalization::impl::replace_(out, tmp_output); |
| 5990 | at::functionalization::impl::commit_update(out); |
| 5991 | at::functionalization::impl::sync(out); |
| 5992 | return out; |
| 5993 | } |
| 5994 | } |
| 5995 | |
| 5996 | at::Tensor & log2_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 5997 | if (true) { |
| 5998 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 5999 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6000 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6001 | auto self_meta = to_meta(self); |
| 6002 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6003 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6004 | at::_ops::log2_::call(self_meta); |
| 6005 | } |
| 6006 | |
| 6007 | at::Tensor self_; |
| 6008 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6009 | at::functionalization::impl::sync(self); |
| 6010 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6011 | } else { |
| 6012 | self_ = self; |
| 6013 | } |
| 6014 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 6015 | if ((false)) { |
| 6016 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6017 | TORCH_INTERNAL_ASSERT(false, |
| 6018 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6019 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6020 | } else { |
| 6021 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6022 | at::AutoDispatchSkipFunctionalize guard; |
| 6023 | at::Tensor tmp_output = at::_ops::log2_::call(self_); |
| 6024 | return self;; |
| 6025 | } |
| 6026 | } else { |
| 6027 | at::Tensor tmp_output; |
| 6028 | { |
| 6029 | at::AutoDispatchSkipFunctionalize guard; |
| 6030 | tmp_output = at::_ops::log2::call(self_); |
| 6031 | } |
| 6032 | at::functionalization::impl::replace_(self, tmp_output); |
| 6033 | at::functionalization::impl::commit_update(self); |
| 6034 | at::functionalization::impl::sync(self); |
| 6035 | return self; |
| 6036 | } |
| 6037 | } |
| 6038 | |
| 6039 | at::Tensor & logaddexp2_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 6040 | if (false) { |
| 6041 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6042 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6043 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6044 | auto self_meta = to_meta(self); |
| 6045 | auto other_meta = to_meta(other); |
| 6046 | auto out_meta = to_meta(out); |
| 6047 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6048 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6049 | at::_ops::logaddexp2_out::call(self_meta, other_meta, out_meta); |
| 6050 | } |
| 6051 | |
| 6052 | at::Tensor self_; |
| 6053 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6054 | at::functionalization::impl::sync(self); |
| 6055 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6056 | } else { |
| 6057 | self_ = self; |
| 6058 | } |
| 6059 | |
| 6060 | at::Tensor other_; |
| 6061 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 6062 | at::functionalization::impl::sync(other); |
| 6063 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 6064 | } else { |
| 6065 | other_ = other; |
| 6066 | } |
| 6067 | |
| 6068 | at::Tensor out_; |
| 6069 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6070 | at::functionalization::impl::sync(out); |
| 6071 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6072 | } else { |
| 6073 | out_ = out; |
| 6074 | } |
| 6075 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6076 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 6077 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6078 | TORCH_INTERNAL_ASSERT(false, |
| 6079 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6080 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6081 | } else { |
| 6082 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6083 | at::AutoDispatchSkipFunctionalize guard; |
| 6084 | at::Tensor tmp_output = at::_ops::logaddexp2_out::call(self_, other_, out_); |
| 6085 | return out;; |
| 6086 | } |
| 6087 | } else { |
| 6088 | at::Tensor tmp_output; |
| 6089 | { |
| 6090 | at::AutoDispatchSkipFunctionalize guard; |
| 6091 | tmp_output = at::_ops::logaddexp2::call(self_, other_); |
| 6092 | } |
| 6093 | at::functionalization::impl::replace_(out, tmp_output); |
| 6094 | at::functionalization::impl::commit_update(out); |
| 6095 | at::functionalization::impl::sync(out); |
| 6096 | return out; |
| 6097 | } |
| 6098 | } |
| 6099 | |
| 6100 | at::Tensor & xlogy_out_OutTensor(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 6101 | if (false) { |
| 6102 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6103 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6104 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6105 | auto self_meta = to_meta(self); |
| 6106 | auto other_meta = to_meta(other); |
| 6107 | auto out_meta = to_meta(out); |
| 6108 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6109 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6110 | at::_ops::xlogy_OutTensor::call(self_meta, other_meta, out_meta); |
| 6111 | } |
| 6112 | |
| 6113 | at::Tensor self_; |
| 6114 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6115 | at::functionalization::impl::sync(self); |
| 6116 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6117 | } else { |
| 6118 | self_ = self; |
| 6119 | } |
| 6120 | |
| 6121 | at::Tensor other_; |
| 6122 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 6123 | at::functionalization::impl::sync(other); |
| 6124 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 6125 | } else { |
| 6126 | other_ = other; |
| 6127 | } |
| 6128 | |
| 6129 | at::Tensor out_; |
| 6130 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6131 | at::functionalization::impl::sync(out); |
| 6132 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6133 | } else { |
| 6134 | out_ = out; |
| 6135 | } |
| 6136 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6137 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 6138 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6139 | TORCH_INTERNAL_ASSERT(false, |
| 6140 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6141 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6142 | } else { |
| 6143 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6144 | at::AutoDispatchSkipFunctionalize guard; |
| 6145 | at::Tensor tmp_output = at::_ops::xlogy_OutTensor::call(self_, other_, out_); |
| 6146 | return out;; |
| 6147 | } |
| 6148 | } else { |
| 6149 | at::Tensor tmp_output; |
| 6150 | { |
| 6151 | at::AutoDispatchSkipFunctionalize guard; |
| 6152 | tmp_output = at::_ops::xlogy_Tensor::call(self_, other_); |
| 6153 | } |
| 6154 | at::functionalization::impl::replace_(out, tmp_output); |
| 6155 | at::functionalization::impl::commit_update(out); |
| 6156 | at::functionalization::impl::sync(out); |
| 6157 | return out; |
| 6158 | } |
| 6159 | } |
| 6160 | |
| 6161 | at::Tensor & xlogy__Tensor(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 6162 | if (true) { |
| 6163 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6164 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6165 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6166 | auto self_meta = to_meta(self); |
| 6167 | auto other_meta = to_meta(other); |
| 6168 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6169 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6170 | at::_ops::xlogy__Tensor::call(self_meta, other_meta); |
| 6171 | } |
| 6172 | |
| 6173 | at::Tensor self_; |
| 6174 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6175 | at::functionalization::impl::sync(self); |
| 6176 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6177 | } else { |
| 6178 | self_ = self; |
| 6179 | } |
| 6180 | |
| 6181 | at::Tensor other_; |
| 6182 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 6183 | at::functionalization::impl::sync(other); |
| 6184 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 6185 | } else { |
| 6186 | other_ = other; |
| 6187 | } |
| 6188 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 6189 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 6190 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6191 | TORCH_INTERNAL_ASSERT(false, |
| 6192 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6193 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6194 | } else { |
| 6195 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6196 | at::AutoDispatchSkipFunctionalize guard; |
| 6197 | at::Tensor tmp_output = at::_ops::xlogy__Tensor::call(self_, other_); |
| 6198 | return self;; |
| 6199 | } |
| 6200 | } else { |
| 6201 | at::Tensor tmp_output; |
| 6202 | { |
| 6203 | at::AutoDispatchSkipFunctionalize guard; |
| 6204 | tmp_output = at::_ops::xlogy_Tensor::call(self_, other_); |
| 6205 | } |
| 6206 | at::functionalization::impl::replace_(self, tmp_output); |
| 6207 | at::functionalization::impl::commit_update(self); |
| 6208 | at::functionalization::impl::sync(self); |
| 6209 | return self; |
| 6210 | } |
| 6211 | } |
| 6212 | |
| 6213 | at::Tensor & xlogy_out_OutScalar_Self(c10::DispatchKeySet dispatchKeySet, const at::Scalar & self, const at::Tensor & other, at::Tensor & out) { |
| 6214 | if (false) { |
| 6215 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6216 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6217 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6218 | auto other_meta = to_meta(other); |
| 6219 | auto out_meta = to_meta(out); |
| 6220 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6221 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6222 | at::_ops::xlogy_OutScalar_Self::call(self, other_meta, out_meta); |
| 6223 | } |
| 6224 | |
| 6225 | at::Tensor other_; |
| 6226 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 6227 | at::functionalization::impl::sync(other); |
| 6228 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 6229 | } else { |
| 6230 | other_ = other; |
| 6231 | } |
| 6232 | |
| 6233 | at::Tensor out_; |
| 6234 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6235 | at::functionalization::impl::sync(out); |
| 6236 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6237 | } else { |
| 6238 | out_ = out; |
| 6239 | } |
| 6240 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6241 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 6242 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6243 | TORCH_INTERNAL_ASSERT(false, |
| 6244 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6245 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6246 | } else { |
| 6247 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6248 | at::AutoDispatchSkipFunctionalize guard; |
| 6249 | at::Tensor tmp_output = at::_ops::xlogy_OutScalar_Self::call(self, other_, out_); |
| 6250 | return out;; |
| 6251 | } |
| 6252 | } else { |
| 6253 | at::Tensor tmp_output; |
| 6254 | { |
| 6255 | at::AutoDispatchSkipFunctionalize guard; |
| 6256 | tmp_output = at::_ops::xlogy_Scalar_Self::call(self, other_); |
| 6257 | } |
| 6258 | at::functionalization::impl::replace_(out, tmp_output); |
| 6259 | at::functionalization::impl::commit_update(out); |
| 6260 | at::functionalization::impl::sync(out); |
| 6261 | return out; |
| 6262 | } |
| 6263 | } |
| 6264 | |
| 6265 | at::Tensor & xlogy_out_OutScalar_Other(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & other, at::Tensor & out) { |
| 6266 | if (false) { |
| 6267 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6268 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6269 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6270 | auto self_meta = to_meta(self); |
| 6271 | auto out_meta = to_meta(out); |
| 6272 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6273 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6274 | at::_ops::xlogy_OutScalar_Other::call(self_meta, other, out_meta); |
| 6275 | } |
| 6276 | |
| 6277 | at::Tensor self_; |
| 6278 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6279 | at::functionalization::impl::sync(self); |
| 6280 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6281 | } else { |
| 6282 | self_ = self; |
| 6283 | } |
| 6284 | |
| 6285 | at::Tensor out_; |
| 6286 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6287 | at::functionalization::impl::sync(out); |
| 6288 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6289 | } else { |
| 6290 | out_ = out; |
| 6291 | } |
| 6292 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6293 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 6294 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6295 | TORCH_INTERNAL_ASSERT(false, |
| 6296 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6297 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6298 | } else { |
| 6299 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6300 | at::AutoDispatchSkipFunctionalize guard; |
| 6301 | at::Tensor tmp_output = at::_ops::xlogy_OutScalar_Other::call(self_, other, out_); |
| 6302 | return out;; |
| 6303 | } |
| 6304 | } else { |
| 6305 | at::Tensor tmp_output; |
| 6306 | { |
| 6307 | at::AutoDispatchSkipFunctionalize guard; |
| 6308 | tmp_output = at::_ops::xlogy_Scalar_Other::call(self_, other); |
| 6309 | } |
| 6310 | at::functionalization::impl::replace_(out, tmp_output); |
| 6311 | at::functionalization::impl::commit_update(out); |
| 6312 | at::functionalization::impl::sync(out); |
| 6313 | return out; |
| 6314 | } |
| 6315 | } |
| 6316 | |
| 6317 | at::Tensor & xlogy__Scalar_Other(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & other) { |
| 6318 | if (true) { |
| 6319 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6320 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6321 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6322 | auto self_meta = to_meta(self); |
| 6323 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6324 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6325 | at::_ops::xlogy__Scalar_Other::call(self_meta, other); |
| 6326 | } |
| 6327 | |
| 6328 | at::Tensor self_; |
| 6329 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6330 | at::functionalization::impl::sync(self); |
| 6331 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6332 | } else { |
| 6333 | self_ = self; |
| 6334 | } |
| 6335 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 6336 | if ((false)) { |
| 6337 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6338 | TORCH_INTERNAL_ASSERT(false, |
| 6339 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6340 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6341 | } else { |
| 6342 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6343 | at::AutoDispatchSkipFunctionalize guard; |
| 6344 | at::Tensor tmp_output = at::_ops::xlogy__Scalar_Other::call(self_, other); |
| 6345 | return self;; |
| 6346 | } |
| 6347 | } else { |
| 6348 | at::Tensor tmp_output; |
| 6349 | { |
| 6350 | at::AutoDispatchSkipFunctionalize guard; |
| 6351 | tmp_output = at::_ops::xlogy_Scalar_Other::call(self_, other); |
| 6352 | } |
| 6353 | at::functionalization::impl::replace_(self, tmp_output); |
| 6354 | at::functionalization::impl::commit_update(self); |
| 6355 | at::functionalization::impl::sync(self); |
| 6356 | return self; |
| 6357 | } |
| 6358 | } |
| 6359 | |
| 6360 | at::Tensor & logspace_out_out(c10::DispatchKeySet dispatchKeySet, const at::Scalar & start, const at::Scalar & end, int64_t steps, double base, at::Tensor & out) { |
| 6361 | if (false) { |
| 6362 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6363 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6364 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6365 | auto out_meta = to_meta(out); |
| 6366 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6367 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6368 | at::_ops::logspace_out::call(start, end, steps, base, out_meta); |
| 6369 | } |
| 6370 | |
| 6371 | at::Tensor out_; |
| 6372 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6373 | at::functionalization::impl::sync(out); |
| 6374 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6375 | } else { |
| 6376 | out_ = out; |
| 6377 | } |
| 6378 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6379 | if ((false)) { |
| 6380 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6381 | TORCH_INTERNAL_ASSERT(false, |
| 6382 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6383 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6384 | } else { |
| 6385 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6386 | at::AutoDispatchSkipFunctionalize guard; |
| 6387 | at::Tensor tmp_output = at::_ops::logspace_out::call(start, end, steps, base, out_); |
| 6388 | return out;; |
| 6389 | } |
| 6390 | } else { |
| 6391 | at::Tensor tmp_output; |
| 6392 | { |
| 6393 | at::AutoDispatchSkipFunctionalize guard; |
| 6394 | tmp_output = at::_ops::logspace::call(start, end, steps, base, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 6395 | } |
| 6396 | at::functionalization::impl::replace_(out, tmp_output); |
| 6397 | at::functionalization::impl::commit_update(out); |
| 6398 | at::functionalization::impl::sync(out); |
| 6399 | return out; |
| 6400 | } |
| 6401 | } |
| 6402 | |
| 6403 | at::Tensor & matrix_power_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t n, at::Tensor & out) { |
| 6404 | if (false) { |
| 6405 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6406 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6407 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6408 | auto self_meta = to_meta(self); |
| 6409 | auto out_meta = to_meta(out); |
| 6410 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6411 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6412 | at::_ops::matrix_power_out::call(self_meta, n, out_meta); |
| 6413 | } |
| 6414 | |
| 6415 | at::Tensor self_; |
| 6416 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6417 | at::functionalization::impl::sync(self); |
| 6418 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6419 | } else { |
| 6420 | self_ = self; |
| 6421 | } |
| 6422 | |
| 6423 | at::Tensor out_; |
| 6424 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6425 | at::functionalization::impl::sync(out); |
| 6426 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6427 | } else { |
| 6428 | out_ = out; |
| 6429 | } |
| 6430 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6431 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 6432 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6433 | TORCH_INTERNAL_ASSERT(false, |
| 6434 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6435 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6436 | } else { |
| 6437 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6438 | at::AutoDispatchSkipFunctionalize guard; |
| 6439 | at::Tensor tmp_output = at::_ops::matrix_power_out::call(self_, n, out_); |
| 6440 | return out;; |
| 6441 | } |
| 6442 | } else { |
| 6443 | at::Tensor tmp_output; |
| 6444 | { |
| 6445 | at::AutoDispatchSkipFunctionalize guard; |
| 6446 | tmp_output = at::_ops::matrix_power::call(self_, n); |
| 6447 | } |
| 6448 | at::functionalization::impl::replace_(out, tmp_output); |
| 6449 | at::functionalization::impl::commit_update(out); |
| 6450 | at::functionalization::impl::sync(out); |
| 6451 | return out; |
| 6452 | } |
| 6453 | } |
| 6454 | |
| 6455 | ::std::tuple<at::Tensor &,at::Tensor &> _aminmax_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out0, at::Tensor & out1) { |
| 6456 | if (false) { |
| 6457 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6458 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6459 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6460 | auto self_meta = to_meta(self); |
| 6461 | auto out0_meta = to_meta(out0); |
| 6462 | auto out1_meta = to_meta(out1); |
| 6463 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6464 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6465 | at::_ops::_aminmax_out::call(self_meta, out0_meta, out1_meta); |
| 6466 | } |
| 6467 | |
| 6468 | at::Tensor self_; |
| 6469 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6470 | at::functionalization::impl::sync(self); |
| 6471 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6472 | } else { |
| 6473 | self_ = self; |
| 6474 | } |
| 6475 | |
| 6476 | at::Tensor out0_; |
| 6477 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 6478 | at::functionalization::impl::sync(out0); |
| 6479 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 6480 | } else { |
| 6481 | out0_ = out0; |
| 6482 | } |
| 6483 | |
| 6484 | at::Tensor out1_; |
| 6485 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 6486 | at::functionalization::impl::sync(out1); |
| 6487 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 6488 | } else { |
| 6489 | out1_ = out1; |
| 6490 | } |
| 6491 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1))) { |
| 6492 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 6493 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6494 | TORCH_INTERNAL_ASSERT(false, |
| 6495 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6496 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6497 | } else { |
| 6498 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6499 | at::AutoDispatchSkipFunctionalize guard; |
| 6500 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::_aminmax_out::call(self_, out0_, out1_); |
| 6501 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1);; |
| 6502 | } |
| 6503 | } else { |
| 6504 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 6505 | { |
| 6506 | at::AutoDispatchSkipFunctionalize guard; |
| 6507 | tmp_output = at::_ops::_aminmax::call(self_); |
| 6508 | } |
| 6509 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 6510 | at::functionalization::impl::commit_update(out0); |
| 6511 | at::functionalization::impl::sync(out0); |
| 6512 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 6513 | at::functionalization::impl::commit_update(out1); |
| 6514 | at::functionalization::impl::sync(out1); |
| 6515 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1); |
| 6516 | } |
| 6517 | } |
| 6518 | |
| 6519 | ::std::tuple<at::Tensor &,at::Tensor &> _aminmax_out_dim_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dim, bool keepdim, at::Tensor & out0, at::Tensor & out1) { |
| 6520 | if (false) { |
| 6521 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6522 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6523 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6524 | auto self_meta = to_meta(self); |
| 6525 | auto out0_meta = to_meta(out0); |
| 6526 | auto out1_meta = to_meta(out1); |
| 6527 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6528 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6529 | at::_ops::_aminmax_dim_out::call(self_meta, dim, keepdim, out0_meta, out1_meta); |
| 6530 | } |
| 6531 | |
| 6532 | at::Tensor self_; |
| 6533 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6534 | at::functionalization::impl::sync(self); |
| 6535 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6536 | } else { |
| 6537 | self_ = self; |
| 6538 | } |
| 6539 | |
| 6540 | at::Tensor out0_; |
| 6541 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 6542 | at::functionalization::impl::sync(out0); |
| 6543 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 6544 | } else { |
| 6545 | out0_ = out0; |
| 6546 | } |
| 6547 | |
| 6548 | at::Tensor out1_; |
| 6549 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 6550 | at::functionalization::impl::sync(out1); |
| 6551 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 6552 | } else { |
| 6553 | out1_ = out1; |
| 6554 | } |
| 6555 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1))) { |
| 6556 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 6557 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6558 | TORCH_INTERNAL_ASSERT(false, |
| 6559 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6560 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6561 | } else { |
| 6562 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6563 | at::AutoDispatchSkipFunctionalize guard; |
| 6564 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::_aminmax_dim_out::call(self_, dim, keepdim, out0_, out1_); |
| 6565 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1);; |
| 6566 | } |
| 6567 | } else { |
| 6568 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 6569 | { |
| 6570 | at::AutoDispatchSkipFunctionalize guard; |
| 6571 | tmp_output = at::_ops::_aminmax_dim::call(self_, dim, keepdim); |
| 6572 | } |
| 6573 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 6574 | at::functionalization::impl::commit_update(out0); |
| 6575 | at::functionalization::impl::sync(out0); |
| 6576 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 6577 | at::functionalization::impl::commit_update(out1); |
| 6578 | at::functionalization::impl::sync(out1); |
| 6579 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1); |
| 6580 | } |
| 6581 | } |
| 6582 | |
| 6583 | ::std::tuple<at::Tensor &,at::Tensor &> aminmax_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<int64_t> dim, bool keepdim, at::Tensor & min, at::Tensor & max) { |
| 6584 | if (false) { |
| 6585 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6586 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6587 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6588 | auto self_meta = to_meta(self); |
| 6589 | auto min_meta = to_meta(min); |
| 6590 | auto max_meta = to_meta(max); |
| 6591 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6592 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6593 | at::_ops::aminmax_out::call(self_meta, dim, keepdim, min_meta, max_meta); |
| 6594 | } |
| 6595 | |
| 6596 | at::Tensor self_; |
| 6597 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6598 | at::functionalization::impl::sync(self); |
| 6599 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6600 | } else { |
| 6601 | self_ = self; |
| 6602 | } |
| 6603 | |
| 6604 | at::Tensor min_; |
| 6605 | if (at::functionalization::impl::isFunctionalTensor(min)) { |
| 6606 | at::functionalization::impl::sync(min); |
| 6607 | min_ = at::functionalization::impl::from_functional_tensor(min); |
| 6608 | } else { |
| 6609 | min_ = min; |
| 6610 | } |
| 6611 | |
| 6612 | at::Tensor max_; |
| 6613 | if (at::functionalization::impl::isFunctionalTensor(max)) { |
| 6614 | at::functionalization::impl::sync(max); |
| 6615 | max_ = at::functionalization::impl::from_functional_tensor(max); |
| 6616 | } else { |
| 6617 | max_ = max; |
| 6618 | } |
| 6619 | if (!(true && at::functionalization::impl::isFunctionalTensor(min) && at::functionalization::impl::isFunctionalTensor(max))) { |
| 6620 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 6621 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6622 | TORCH_INTERNAL_ASSERT(false, |
| 6623 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6624 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6625 | } else { |
| 6626 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6627 | at::AutoDispatchSkipFunctionalize guard; |
| 6628 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::aminmax_out::call(self_, dim, keepdim, min_, max_); |
| 6629 | return ::std::tuple<at::Tensor &,at::Tensor &>(min, max);; |
| 6630 | } |
| 6631 | } else { |
| 6632 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 6633 | { |
| 6634 | at::AutoDispatchSkipFunctionalize guard; |
| 6635 | tmp_output = at::_ops::aminmax::call(self_, dim, keepdim); |
| 6636 | } |
| 6637 | at::functionalization::impl::replace_(min, std::get<0>(tmp_output)); |
| 6638 | at::functionalization::impl::commit_update(min); |
| 6639 | at::functionalization::impl::sync(min); |
| 6640 | at::functionalization::impl::replace_(max, std::get<1>(tmp_output)); |
| 6641 | at::functionalization::impl::commit_update(max); |
| 6642 | at::functionalization::impl::sync(max); |
| 6643 | return ::std::tuple<at::Tensor &,at::Tensor &>(min, max); |
| 6644 | } |
| 6645 | } |
| 6646 | |
| 6647 | at::Tensor & _compute_linear_combination_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, const at::Tensor & coefficients, at::Tensor & out) { |
| 6648 | if (false) { |
| 6649 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6650 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6651 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6652 | auto input_meta = to_meta(input); |
| 6653 | auto coefficients_meta = to_meta(coefficients); |
| 6654 | auto out_meta = to_meta(out); |
| 6655 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6656 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6657 | at::_ops::_compute_linear_combination_out::call(input_meta, coefficients_meta, out_meta); |
| 6658 | } |
| 6659 | |
| 6660 | at::Tensor input_; |
| 6661 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 6662 | at::functionalization::impl::sync(input); |
| 6663 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 6664 | } else { |
| 6665 | input_ = input; |
| 6666 | } |
| 6667 | |
| 6668 | at::Tensor coefficients_; |
| 6669 | if (at::functionalization::impl::isFunctionalTensor(coefficients)) { |
| 6670 | at::functionalization::impl::sync(coefficients); |
| 6671 | coefficients_ = at::functionalization::impl::from_functional_tensor(coefficients); |
| 6672 | } else { |
| 6673 | coefficients_ = coefficients; |
| 6674 | } |
| 6675 | |
| 6676 | at::Tensor out_; |
| 6677 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6678 | at::functionalization::impl::sync(out); |
| 6679 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6680 | } else { |
| 6681 | out_ = out; |
| 6682 | } |
| 6683 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6684 | if ((false || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(coefficients))) { |
| 6685 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6686 | TORCH_INTERNAL_ASSERT(false, |
| 6687 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6688 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6689 | } else { |
| 6690 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6691 | at::AutoDispatchSkipFunctionalize guard; |
| 6692 | at::Tensor tmp_output = at::_ops::_compute_linear_combination_out::call(input_, coefficients_, out_); |
| 6693 | return out;; |
| 6694 | } |
| 6695 | } else { |
| 6696 | at::Tensor tmp_output; |
| 6697 | { |
| 6698 | at::AutoDispatchSkipFunctionalize guard; |
| 6699 | tmp_output = at::_ops::_compute_linear_combination::call(input_, coefficients_); |
| 6700 | } |
| 6701 | at::functionalization::impl::replace_(out, tmp_output); |
| 6702 | at::functionalization::impl::commit_update(out); |
| 6703 | at::functionalization::impl::sync(out); |
| 6704 | return out; |
| 6705 | } |
| 6706 | } |
| 6707 | |
| 6708 | at::Tensor & _mps_max_pool2d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef kernel_size, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, bool ceil_mode, at::Tensor & out) { |
| 6709 | if (false) { |
| 6710 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6711 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6712 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6713 | auto self_meta = to_meta(self); |
| 6714 | auto out_meta = to_meta(out); |
| 6715 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6716 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6717 | at::_ops::_mps_max_pool2d_out::call(self_meta, kernel_size, stride, padding, dilation, ceil_mode, out_meta); |
| 6718 | } |
| 6719 | |
| 6720 | at::Tensor self_; |
| 6721 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6722 | at::functionalization::impl::sync(self); |
| 6723 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6724 | } else { |
| 6725 | self_ = self; |
| 6726 | } |
| 6727 | |
| 6728 | at::Tensor out_; |
| 6729 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6730 | at::functionalization::impl::sync(out); |
| 6731 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6732 | } else { |
| 6733 | out_ = out; |
| 6734 | } |
| 6735 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6736 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 6737 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6738 | TORCH_INTERNAL_ASSERT(false, |
| 6739 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6740 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6741 | } else { |
| 6742 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6743 | at::AutoDispatchSkipFunctionalize guard; |
| 6744 | at::Tensor tmp_output = at::_ops::_mps_max_pool2d_out::call(self_, kernel_size, stride, padding, dilation, ceil_mode, out_); |
| 6745 | return out;; |
| 6746 | } |
| 6747 | } else { |
| 6748 | at::Tensor tmp_output; |
| 6749 | { |
| 6750 | at::AutoDispatchSkipFunctionalize guard; |
| 6751 | tmp_output = at::_ops::_mps_max_pool2d::call(self_, kernel_size, stride, padding, dilation, ceil_mode); |
| 6752 | } |
| 6753 | at::functionalization::impl::replace_(out, tmp_output); |
| 6754 | at::functionalization::impl::commit_update(out); |
| 6755 | at::functionalization::impl::sync(out); |
| 6756 | return out; |
| 6757 | } |
| 6758 | } |
| 6759 | |
| 6760 | at::Tensor & mkldnn_max_pool3d_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & output, const at::Tensor & input, at::IntArrayRef kernel_size, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, bool ceil_mode, at::Tensor & out) { |
| 6761 | if (false) { |
| 6762 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6763 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6764 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6765 | auto grad_output_meta = to_meta(grad_output); |
| 6766 | auto output_meta = to_meta(output); |
| 6767 | auto input_meta = to_meta(input); |
| 6768 | auto out_meta = to_meta(out); |
| 6769 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6770 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6771 | at::_ops::mkldnn_max_pool3d_backward_out::call(grad_output_meta, output_meta, input_meta, kernel_size, stride, padding, dilation, ceil_mode, out_meta); |
| 6772 | } |
| 6773 | |
| 6774 | at::Tensor grad_output_; |
| 6775 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 6776 | at::functionalization::impl::sync(grad_output); |
| 6777 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 6778 | } else { |
| 6779 | grad_output_ = grad_output; |
| 6780 | } |
| 6781 | |
| 6782 | at::Tensor output_; |
| 6783 | if (at::functionalization::impl::isFunctionalTensor(output)) { |
| 6784 | at::functionalization::impl::sync(output); |
| 6785 | output_ = at::functionalization::impl::from_functional_tensor(output); |
| 6786 | } else { |
| 6787 | output_ = output; |
| 6788 | } |
| 6789 | |
| 6790 | at::Tensor input_; |
| 6791 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 6792 | at::functionalization::impl::sync(input); |
| 6793 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 6794 | } else { |
| 6795 | input_ = input; |
| 6796 | } |
| 6797 | |
| 6798 | at::Tensor out_; |
| 6799 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 6800 | at::functionalization::impl::sync(out); |
| 6801 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 6802 | } else { |
| 6803 | out_ = out; |
| 6804 | } |
| 6805 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 6806 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(output) || at::functionalization::impl::isFunctionalTensor(input))) { |
| 6807 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6808 | TORCH_INTERNAL_ASSERT(false, |
| 6809 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6810 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6811 | } else { |
| 6812 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6813 | at::AutoDispatchSkipFunctionalize guard; |
| 6814 | at::Tensor tmp_output = at::_ops::mkldnn_max_pool3d_backward_out::call(grad_output_, output_, input_, kernel_size, stride, padding, dilation, ceil_mode, out_); |
| 6815 | return out;; |
| 6816 | } |
| 6817 | } else { |
| 6818 | at::Tensor tmp_output; |
| 6819 | { |
| 6820 | at::AutoDispatchSkipFunctionalize guard; |
| 6821 | tmp_output = at::_ops::mkldnn_max_pool3d_backward::call(grad_output_, output_, input_, kernel_size, stride, padding, dilation, ceil_mode); |
| 6822 | } |
| 6823 | at::functionalization::impl::replace_(out, tmp_output); |
| 6824 | at::functionalization::impl::commit_update(out); |
| 6825 | at::functionalization::impl::sync(out); |
| 6826 | return out; |
| 6827 | } |
| 6828 | } |
| 6829 | |
| 6830 | ::std::tuple<at::Tensor &,at::Tensor &> min_out_dim_min(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dim, bool keepdim, at::Tensor & min, at::Tensor & min_indices) { |
| 6831 | if (false) { |
| 6832 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6833 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6834 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6835 | auto self_meta = to_meta(self); |
| 6836 | auto min_meta = to_meta(min); |
| 6837 | auto min_indices_meta = to_meta(min_indices); |
| 6838 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6839 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6840 | at::_ops::min_dim_min::call(self_meta, dim, keepdim, min_meta, min_indices_meta); |
| 6841 | } |
| 6842 | |
| 6843 | at::Tensor self_; |
| 6844 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6845 | at::functionalization::impl::sync(self); |
| 6846 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6847 | } else { |
| 6848 | self_ = self; |
| 6849 | } |
| 6850 | |
| 6851 | at::Tensor min_; |
| 6852 | if (at::functionalization::impl::isFunctionalTensor(min)) { |
| 6853 | at::functionalization::impl::sync(min); |
| 6854 | min_ = at::functionalization::impl::from_functional_tensor(min); |
| 6855 | } else { |
| 6856 | min_ = min; |
| 6857 | } |
| 6858 | |
| 6859 | at::Tensor min_indices_; |
| 6860 | if (at::functionalization::impl::isFunctionalTensor(min_indices)) { |
| 6861 | at::functionalization::impl::sync(min_indices); |
| 6862 | min_indices_ = at::functionalization::impl::from_functional_tensor(min_indices); |
| 6863 | } else { |
| 6864 | min_indices_ = min_indices; |
| 6865 | } |
| 6866 | if (!(true && at::functionalization::impl::isFunctionalTensor(min) && at::functionalization::impl::isFunctionalTensor(min_indices))) { |
| 6867 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 6868 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6869 | TORCH_INTERNAL_ASSERT(false, |
| 6870 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6871 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6872 | } else { |
| 6873 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6874 | at::AutoDispatchSkipFunctionalize guard; |
| 6875 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::min_dim_min::call(self_, dim, keepdim, min_, min_indices_); |
| 6876 | return ::std::tuple<at::Tensor &,at::Tensor &>(min, min_indices);; |
| 6877 | } |
| 6878 | } else { |
| 6879 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 6880 | { |
| 6881 | at::AutoDispatchSkipFunctionalize guard; |
| 6882 | tmp_output = at::_ops::min_dim::call(self_, dim, keepdim); |
| 6883 | } |
| 6884 | at::functionalization::impl::replace_(min, std::get<0>(tmp_output)); |
| 6885 | at::functionalization::impl::commit_update(min); |
| 6886 | at::functionalization::impl::sync(min); |
| 6887 | at::functionalization::impl::replace_(min_indices, std::get<1>(tmp_output)); |
| 6888 | at::functionalization::impl::commit_update(min_indices); |
| 6889 | at::functionalization::impl::sync(min_indices); |
| 6890 | return ::std::tuple<at::Tensor &,at::Tensor &>(min, min_indices); |
| 6891 | } |
| 6892 | } |
| 6893 | |
| 6894 | ::std::tuple<at::Tensor &,at::Tensor &> min_out_names_dim_min(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Dimname dim, bool keepdim, at::Tensor & min, at::Tensor & min_indices) { |
| 6895 | if (false) { |
| 6896 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6897 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6898 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6899 | auto self_meta = to_meta(self); |
| 6900 | auto min_meta = to_meta(min); |
| 6901 | auto min_indices_meta = to_meta(min_indices); |
| 6902 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6903 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6904 | at::_ops::min_names_dim_min::call(self_meta, dim, keepdim, min_meta, min_indices_meta); |
| 6905 | } |
| 6906 | |
| 6907 | at::Tensor self_; |
| 6908 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6909 | at::functionalization::impl::sync(self); |
| 6910 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6911 | } else { |
| 6912 | self_ = self; |
| 6913 | } |
| 6914 | |
| 6915 | at::Tensor min_; |
| 6916 | if (at::functionalization::impl::isFunctionalTensor(min)) { |
| 6917 | at::functionalization::impl::sync(min); |
| 6918 | min_ = at::functionalization::impl::from_functional_tensor(min); |
| 6919 | } else { |
| 6920 | min_ = min; |
| 6921 | } |
| 6922 | |
| 6923 | at::Tensor min_indices_; |
| 6924 | if (at::functionalization::impl::isFunctionalTensor(min_indices)) { |
| 6925 | at::functionalization::impl::sync(min_indices); |
| 6926 | min_indices_ = at::functionalization::impl::from_functional_tensor(min_indices); |
| 6927 | } else { |
| 6928 | min_indices_ = min_indices; |
| 6929 | } |
| 6930 | if (!(true && at::functionalization::impl::isFunctionalTensor(min) && at::functionalization::impl::isFunctionalTensor(min_indices))) { |
| 6931 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 6932 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 6933 | TORCH_INTERNAL_ASSERT(false, |
| 6934 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 6935 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 6936 | } else { |
| 6937 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 6938 | at::AutoDispatchSkipFunctionalize guard; |
| 6939 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::min_names_dim_min::call(self_, dim, keepdim, min_, min_indices_); |
| 6940 | return ::std::tuple<at::Tensor &,at::Tensor &>(min, min_indices);; |
| 6941 | } |
| 6942 | } else { |
| 6943 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 6944 | { |
| 6945 | at::AutoDispatchSkipFunctionalize guard; |
| 6946 | tmp_output = at::_ops::min_names_dim::call(self_, dim, keepdim); |
| 6947 | } |
| 6948 | at::functionalization::impl::replace_(min, std::get<0>(tmp_output)); |
| 6949 | at::functionalization::impl::commit_update(min); |
| 6950 | at::functionalization::impl::sync(min); |
| 6951 | at::functionalization::impl::replace_(min_indices, std::get<1>(tmp_output)); |
| 6952 | at::functionalization::impl::commit_update(min_indices); |
| 6953 | at::functionalization::impl::sync(min_indices); |
| 6954 | return ::std::tuple<at::Tensor &,at::Tensor &>(min, min_indices); |
| 6955 | } |
| 6956 | } |
| 6957 | |
| 6958 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> mps_convolution_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & grad_output, const at::Tensor & weight, at::IntArrayRef padding, at::IntArrayRef stride, at::IntArrayRef dilation, int64_t groups, ::std::array<bool,3> output_mask, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 6959 | if (false) { |
| 6960 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 6961 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 6962 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 6963 | auto self_meta = to_meta(self); |
| 6964 | auto grad_output_meta = to_meta(grad_output); |
| 6965 | auto weight_meta = to_meta(weight); |
| 6966 | auto out0_meta = to_meta(out0); |
| 6967 | auto out1_meta = to_meta(out1); |
| 6968 | auto out2_meta = to_meta(out2); |
| 6969 | at::AutoDispatchSkipFunctionalize func_guard; |
| 6970 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 6971 | at::_ops::mps_convolution_backward_out::call(self_meta, grad_output_meta, weight_meta, padding, stride, dilation, groups, output_mask, out0_meta, out1_meta, out2_meta); |
| 6972 | } |
| 6973 | |
| 6974 | at::Tensor self_; |
| 6975 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 6976 | at::functionalization::impl::sync(self); |
| 6977 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 6978 | } else { |
| 6979 | self_ = self; |
| 6980 | } |
| 6981 | |
| 6982 | at::Tensor grad_output_; |
| 6983 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 6984 | at::functionalization::impl::sync(grad_output); |
| 6985 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 6986 | } else { |
| 6987 | grad_output_ = grad_output; |
| 6988 | } |
| 6989 | |
| 6990 | at::Tensor weight_; |
| 6991 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 6992 | at::functionalization::impl::sync(weight); |
| 6993 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 6994 | } else { |
| 6995 | weight_ = weight; |
| 6996 | } |
| 6997 | |
| 6998 | at::Tensor out0_; |
| 6999 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 7000 | at::functionalization::impl::sync(out0); |
| 7001 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 7002 | } else { |
| 7003 | out0_ = out0; |
| 7004 | } |
| 7005 | |
| 7006 | at::Tensor out1_; |
| 7007 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 7008 | at::functionalization::impl::sync(out1); |
| 7009 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 7010 | } else { |
| 7011 | out1_ = out1; |
| 7012 | } |
| 7013 | |
| 7014 | at::Tensor out2_; |
| 7015 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 7016 | at::functionalization::impl::sync(out2); |
| 7017 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 7018 | } else { |
| 7019 | out2_ = out2; |
| 7020 | } |
| 7021 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 7022 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 7023 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7024 | TORCH_INTERNAL_ASSERT(false, |
| 7025 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7026 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7027 | } else { |
| 7028 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7029 | at::AutoDispatchSkipFunctionalize guard; |
| 7030 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::mps_convolution_backward_out::call(self_, grad_output_, weight_, padding, stride, dilation, groups, output_mask, out0_, out1_, out2_); |
| 7031 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 7032 | } |
| 7033 | } else { |
| 7034 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 7035 | { |
| 7036 | at::AutoDispatchSkipFunctionalize guard; |
| 7037 | tmp_output = at::_ops::mps_convolution_backward::call(self_, grad_output_, weight_, padding, stride, dilation, groups, output_mask); |
| 7038 | } |
| 7039 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 7040 | at::functionalization::impl::commit_update(out0); |
| 7041 | at::functionalization::impl::sync(out0); |
| 7042 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 7043 | at::functionalization::impl::commit_update(out1); |
| 7044 | at::functionalization::impl::sync(out1); |
| 7045 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 7046 | at::functionalization::impl::commit_update(out2); |
| 7047 | at::functionalization::impl::sync(out2); |
| 7048 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 7049 | } |
| 7050 | } |
| 7051 | |
| 7052 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &> miopen_rnn_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, at::TensorList weight, int64_t weight_stride0, const at::Tensor & hx, const c10::optional<at::Tensor> & cx, int64_t mode, int64_t hidden_size, int64_t num_layers, bool batch_first, double dropout, bool train, bool bidirectional, at::IntArrayRef batch_sizes, const c10::optional<at::Tensor> & dropout_state, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2, at::Tensor & out3, at::Tensor & out4) { |
| 7053 | if (false) { |
| 7054 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7055 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7056 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7057 | auto input_meta = to_meta(input); |
| 7058 | auto weight_meta = to_meta(weight); |
| 7059 | auto hx_meta = to_meta(hx); |
| 7060 | auto cx_meta = to_meta(cx); |
| 7061 | auto dropout_state_meta = to_meta(dropout_state); |
| 7062 | auto out0_meta = to_meta(out0); |
| 7063 | auto out1_meta = to_meta(out1); |
| 7064 | auto out2_meta = to_meta(out2); |
| 7065 | auto out3_meta = to_meta(out3); |
| 7066 | auto out4_meta = to_meta(out4); |
| 7067 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7068 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7069 | at::_ops::miopen_rnn_out::call(input_meta, weight_meta, weight_stride0, hx_meta, cx_meta, mode, hidden_size, num_layers, batch_first, dropout, train, bidirectional, batch_sizes, dropout_state_meta, out0_meta, out1_meta, out2_meta, out3_meta, out4_meta); |
| 7070 | } |
| 7071 | |
| 7072 | at::Tensor input_; |
| 7073 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 7074 | at::functionalization::impl::sync(input); |
| 7075 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 7076 | } else { |
| 7077 | input_ = input; |
| 7078 | } |
| 7079 | |
| 7080 | ::std::vector<at::Tensor> weight_; |
| 7081 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 7082 | at::functionalization::impl::sync(weight); |
| 7083 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 7084 | } else { |
| 7085 | weight_ = weight.vec(); |
| 7086 | } |
| 7087 | |
| 7088 | at::Tensor hx_; |
| 7089 | if (at::functionalization::impl::isFunctionalTensor(hx)) { |
| 7090 | at::functionalization::impl::sync(hx); |
| 7091 | hx_ = at::functionalization::impl::from_functional_tensor(hx); |
| 7092 | } else { |
| 7093 | hx_ = hx; |
| 7094 | } |
| 7095 | |
| 7096 | c10::optional<at::Tensor> cx_; |
| 7097 | if (at::functionalization::impl::isFunctionalTensor(cx)) { |
| 7098 | at::functionalization::impl::sync(cx); |
| 7099 | cx_ = at::functionalization::impl::from_functional_tensor(cx); |
| 7100 | } else { |
| 7101 | cx_ = cx; |
| 7102 | } |
| 7103 | |
| 7104 | c10::optional<at::Tensor> dropout_state_; |
| 7105 | if (at::functionalization::impl::isFunctionalTensor(dropout_state)) { |
| 7106 | at::functionalization::impl::sync(dropout_state); |
| 7107 | dropout_state_ = at::functionalization::impl::from_functional_tensor(dropout_state); |
| 7108 | } else { |
| 7109 | dropout_state_ = dropout_state; |
| 7110 | } |
| 7111 | |
| 7112 | at::Tensor out0_; |
| 7113 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 7114 | at::functionalization::impl::sync(out0); |
| 7115 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 7116 | } else { |
| 7117 | out0_ = out0; |
| 7118 | } |
| 7119 | |
| 7120 | at::Tensor out1_; |
| 7121 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 7122 | at::functionalization::impl::sync(out1); |
| 7123 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 7124 | } else { |
| 7125 | out1_ = out1; |
| 7126 | } |
| 7127 | |
| 7128 | at::Tensor out2_; |
| 7129 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 7130 | at::functionalization::impl::sync(out2); |
| 7131 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 7132 | } else { |
| 7133 | out2_ = out2; |
| 7134 | } |
| 7135 | |
| 7136 | at::Tensor out3_; |
| 7137 | if (at::functionalization::impl::isFunctionalTensor(out3)) { |
| 7138 | at::functionalization::impl::sync(out3); |
| 7139 | out3_ = at::functionalization::impl::from_functional_tensor(out3); |
| 7140 | } else { |
| 7141 | out3_ = out3; |
| 7142 | } |
| 7143 | |
| 7144 | at::Tensor out4_; |
| 7145 | if (at::functionalization::impl::isFunctionalTensor(out4)) { |
| 7146 | at::functionalization::impl::sync(out4); |
| 7147 | out4_ = at::functionalization::impl::from_functional_tensor(out4); |
| 7148 | } else { |
| 7149 | out4_ = out4; |
| 7150 | } |
| 7151 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2) && at::functionalization::impl::isFunctionalTensor(out3) && at::functionalization::impl::isFunctionalTensor(out4))) { |
| 7152 | if ((false || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(hx) || at::functionalization::impl::isFunctionalTensor(cx) || at::functionalization::impl::isFunctionalTensor(dropout_state))) { |
| 7153 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7154 | TORCH_INTERNAL_ASSERT(false, |
| 7155 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7156 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7157 | } else { |
| 7158 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7159 | at::AutoDispatchSkipFunctionalize guard; |
| 7160 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::miopen_rnn_out::call(input_, weight_, weight_stride0, hx_, cx_, mode, hidden_size, num_layers, batch_first, dropout, train, bidirectional, batch_sizes, dropout_state_, out0_, out1_, out2_, out3_, out4_); |
| 7161 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2, out3, out4);; |
| 7162 | } |
| 7163 | } else { |
| 7164 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 7165 | { |
| 7166 | at::AutoDispatchSkipFunctionalize guard; |
| 7167 | tmp_output = at::_ops::miopen_rnn::call(input_, weight_, weight_stride0, hx_, cx_, mode, hidden_size, num_layers, batch_first, dropout, train, bidirectional, batch_sizes, dropout_state_); |
| 7168 | } |
| 7169 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 7170 | at::functionalization::impl::commit_update(out0); |
| 7171 | at::functionalization::impl::sync(out0); |
| 7172 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 7173 | at::functionalization::impl::commit_update(out1); |
| 7174 | at::functionalization::impl::sync(out1); |
| 7175 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 7176 | at::functionalization::impl::commit_update(out2); |
| 7177 | at::functionalization::impl::sync(out2); |
| 7178 | at::functionalization::impl::replace_(out3, std::get<3>(tmp_output)); |
| 7179 | at::functionalization::impl::commit_update(out3); |
| 7180 | at::functionalization::impl::sync(out3); |
| 7181 | at::functionalization::impl::replace_(out4, std::get<4>(tmp_output)); |
| 7182 | at::functionalization::impl::commit_update(out4); |
| 7183 | at::functionalization::impl::sync(out4); |
| 7184 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2, out3, out4); |
| 7185 | } |
| 7186 | } |
| 7187 | |
| 7188 | at::Tensor & mv_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & vec, at::Tensor & out) { |
| 7189 | if (false) { |
| 7190 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7191 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7192 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7193 | auto self_meta = to_meta(self); |
| 7194 | auto vec_meta = to_meta(vec); |
| 7195 | auto out_meta = to_meta(out); |
| 7196 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7197 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7198 | at::_ops::mv_out::call(self_meta, vec_meta, out_meta); |
| 7199 | } |
| 7200 | |
| 7201 | at::Tensor self_; |
| 7202 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 7203 | at::functionalization::impl::sync(self); |
| 7204 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 7205 | } else { |
| 7206 | self_ = self; |
| 7207 | } |
| 7208 | |
| 7209 | at::Tensor vec_; |
| 7210 | if (at::functionalization::impl::isFunctionalTensor(vec)) { |
| 7211 | at::functionalization::impl::sync(vec); |
| 7212 | vec_ = at::functionalization::impl::from_functional_tensor(vec); |
| 7213 | } else { |
| 7214 | vec_ = vec; |
| 7215 | } |
| 7216 | |
| 7217 | at::Tensor out_; |
| 7218 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 7219 | at::functionalization::impl::sync(out); |
| 7220 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 7221 | } else { |
| 7222 | out_ = out; |
| 7223 | } |
| 7224 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 7225 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(vec))) { |
| 7226 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7227 | TORCH_INTERNAL_ASSERT(false, |
| 7228 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7229 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7230 | } else { |
| 7231 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7232 | at::AutoDispatchSkipFunctionalize guard; |
| 7233 | at::Tensor tmp_output = at::_ops::mv_out::call(self_, vec_, out_); |
| 7234 | return out;; |
| 7235 | } |
| 7236 | } else { |
| 7237 | at::Tensor tmp_output; |
| 7238 | { |
| 7239 | at::AutoDispatchSkipFunctionalize guard; |
| 7240 | tmp_output = at::_ops::mv::call(self_, vec_); |
| 7241 | } |
| 7242 | at::functionalization::impl::replace_(out, tmp_output); |
| 7243 | at::functionalization::impl::commit_update(out); |
| 7244 | at::functionalization::impl::sync(out); |
| 7245 | return out; |
| 7246 | } |
| 7247 | } |
| 7248 | |
| 7249 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> _native_batch_norm_legit_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, const c10::optional<at::Tensor> & weight, const c10::optional<at::Tensor> & bias, at::Tensor & running_mean, at::Tensor & running_var, bool training, double momentum, double eps, at::Tensor & out, at::Tensor & save_mean, at::Tensor & save_invstd) { |
| 7250 | if (false) { |
| 7251 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7252 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7253 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7254 | auto input_meta = to_meta(input); |
| 7255 | auto weight_meta = to_meta(weight); |
| 7256 | auto bias_meta = to_meta(bias); |
| 7257 | auto running_mean_meta = to_meta(running_mean); |
| 7258 | auto running_var_meta = to_meta(running_var); |
| 7259 | auto out_meta = to_meta(out); |
| 7260 | auto save_mean_meta = to_meta(save_mean); |
| 7261 | auto save_invstd_meta = to_meta(save_invstd); |
| 7262 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7263 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7264 | at::_ops::_native_batch_norm_legit_out::call(input_meta, weight_meta, bias_meta, running_mean_meta, running_var_meta, training, momentum, eps, out_meta, save_mean_meta, save_invstd_meta); |
| 7265 | } |
| 7266 | |
| 7267 | at::Tensor input_; |
| 7268 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 7269 | at::functionalization::impl::sync(input); |
| 7270 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 7271 | } else { |
| 7272 | input_ = input; |
| 7273 | } |
| 7274 | |
| 7275 | c10::optional<at::Tensor> weight_; |
| 7276 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 7277 | at::functionalization::impl::sync(weight); |
| 7278 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 7279 | } else { |
| 7280 | weight_ = weight; |
| 7281 | } |
| 7282 | |
| 7283 | c10::optional<at::Tensor> bias_; |
| 7284 | if (at::functionalization::impl::isFunctionalTensor(bias)) { |
| 7285 | at::functionalization::impl::sync(bias); |
| 7286 | bias_ = at::functionalization::impl::from_functional_tensor(bias); |
| 7287 | } else { |
| 7288 | bias_ = bias; |
| 7289 | } |
| 7290 | |
| 7291 | at::Tensor running_mean_; |
| 7292 | if (at::functionalization::impl::isFunctionalTensor(running_mean)) { |
| 7293 | at::functionalization::impl::sync(running_mean); |
| 7294 | running_mean_ = at::functionalization::impl::from_functional_tensor(running_mean); |
| 7295 | } else { |
| 7296 | running_mean_ = running_mean; |
| 7297 | } |
| 7298 | |
| 7299 | at::Tensor running_var_; |
| 7300 | if (at::functionalization::impl::isFunctionalTensor(running_var)) { |
| 7301 | at::functionalization::impl::sync(running_var); |
| 7302 | running_var_ = at::functionalization::impl::from_functional_tensor(running_var); |
| 7303 | } else { |
| 7304 | running_var_ = running_var; |
| 7305 | } |
| 7306 | |
| 7307 | at::Tensor out_; |
| 7308 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 7309 | at::functionalization::impl::sync(out); |
| 7310 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 7311 | } else { |
| 7312 | out_ = out; |
| 7313 | } |
| 7314 | |
| 7315 | at::Tensor save_mean_; |
| 7316 | if (at::functionalization::impl::isFunctionalTensor(save_mean)) { |
| 7317 | at::functionalization::impl::sync(save_mean); |
| 7318 | save_mean_ = at::functionalization::impl::from_functional_tensor(save_mean); |
| 7319 | } else { |
| 7320 | save_mean_ = save_mean; |
| 7321 | } |
| 7322 | |
| 7323 | at::Tensor save_invstd_; |
| 7324 | if (at::functionalization::impl::isFunctionalTensor(save_invstd)) { |
| 7325 | at::functionalization::impl::sync(save_invstd); |
| 7326 | save_invstd_ = at::functionalization::impl::from_functional_tensor(save_invstd); |
| 7327 | } else { |
| 7328 | save_invstd_ = save_invstd; |
| 7329 | } |
| 7330 | if (!(true && at::functionalization::impl::isFunctionalTensor(running_mean) && at::functionalization::impl::isFunctionalTensor(running_var) && at::functionalization::impl::isFunctionalTensor(out) && at::functionalization::impl::isFunctionalTensor(save_mean) && at::functionalization::impl::isFunctionalTensor(save_invstd))) { |
| 7331 | if ((false || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(bias))) { |
| 7332 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7333 | TORCH_INTERNAL_ASSERT(false, |
| 7334 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7335 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7336 | } else { |
| 7337 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7338 | at::AutoDispatchSkipFunctionalize guard; |
| 7339 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_native_batch_norm_legit_out::call(input_, weight_, bias_, running_mean_, running_var_, training, momentum, eps, out_, save_mean_, save_invstd_); |
| 7340 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out, save_mean, save_invstd);; |
| 7341 | } |
| 7342 | } else { |
| 7343 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 7344 | { |
| 7345 | at::AutoDispatchSkipFunctionalize guard; |
| 7346 | tmp_output = at::_ops::_native_batch_norm_legit_functional::call(input_, weight_, bias_, running_mean_, running_var_, training, momentum, eps); |
| 7347 | } |
| 7348 | at::functionalization::impl::replace_(running_mean, std::get<0>(tmp_output)); |
| 7349 | at::functionalization::impl::commit_update(running_mean); |
| 7350 | at::functionalization::impl::sync(running_mean); |
| 7351 | at::functionalization::impl::replace_(running_var, std::get<1>(tmp_output)); |
| 7352 | at::functionalization::impl::commit_update(running_var); |
| 7353 | at::functionalization::impl::sync(running_var); |
| 7354 | at::functionalization::impl::replace_(out, std::get<2>(tmp_output)); |
| 7355 | at::functionalization::impl::commit_update(out); |
| 7356 | at::functionalization::impl::sync(out); |
| 7357 | at::functionalization::impl::replace_(save_mean, std::get<3>(tmp_output)); |
| 7358 | at::functionalization::impl::commit_update(save_mean); |
| 7359 | at::functionalization::impl::sync(save_mean); |
| 7360 | at::functionalization::impl::replace_(save_invstd, std::get<4>(tmp_output)); |
| 7361 | at::functionalization::impl::commit_update(save_invstd); |
| 7362 | at::functionalization::impl::sync(save_invstd); |
| 7363 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out, save_mean, save_invstd); |
| 7364 | } |
| 7365 | } |
| 7366 | |
| 7367 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> _native_batch_norm_legit(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, const c10::optional<at::Tensor> & weight, const c10::optional<at::Tensor> & bias, at::Tensor & running_mean, at::Tensor & running_var, bool training, double momentum, double eps) { |
| 7368 | if (false) { |
| 7369 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7370 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7371 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7372 | auto input_meta = to_meta(input); |
| 7373 | auto weight_meta = to_meta(weight); |
| 7374 | auto bias_meta = to_meta(bias); |
| 7375 | auto running_mean_meta = to_meta(running_mean); |
| 7376 | auto running_var_meta = to_meta(running_var); |
| 7377 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7378 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7379 | at::_ops::_native_batch_norm_legit::call(input_meta, weight_meta, bias_meta, running_mean_meta, running_var_meta, training, momentum, eps); |
| 7380 | } |
| 7381 | |
| 7382 | at::Tensor input_; |
| 7383 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 7384 | at::functionalization::impl::sync(input); |
| 7385 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 7386 | } else { |
| 7387 | input_ = input; |
| 7388 | } |
| 7389 | |
| 7390 | c10::optional<at::Tensor> weight_; |
| 7391 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 7392 | at::functionalization::impl::sync(weight); |
| 7393 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 7394 | } else { |
| 7395 | weight_ = weight; |
| 7396 | } |
| 7397 | |
| 7398 | c10::optional<at::Tensor> bias_; |
| 7399 | if (at::functionalization::impl::isFunctionalTensor(bias)) { |
| 7400 | at::functionalization::impl::sync(bias); |
| 7401 | bias_ = at::functionalization::impl::from_functional_tensor(bias); |
| 7402 | } else { |
| 7403 | bias_ = bias; |
| 7404 | } |
| 7405 | |
| 7406 | at::Tensor running_mean_; |
| 7407 | if (at::functionalization::impl::isFunctionalTensor(running_mean)) { |
| 7408 | at::functionalization::impl::sync(running_mean); |
| 7409 | running_mean_ = at::functionalization::impl::from_functional_tensor(running_mean); |
| 7410 | } else { |
| 7411 | running_mean_ = running_mean; |
| 7412 | } |
| 7413 | |
| 7414 | at::Tensor running_var_; |
| 7415 | if (at::functionalization::impl::isFunctionalTensor(running_var)) { |
| 7416 | at::functionalization::impl::sync(running_var); |
| 7417 | running_var_ = at::functionalization::impl::from_functional_tensor(running_var); |
| 7418 | } else { |
| 7419 | running_var_ = running_var; |
| 7420 | } |
| 7421 | if (!(true && at::functionalization::impl::isFunctionalTensor(running_mean) && at::functionalization::impl::isFunctionalTensor(running_var))) { |
| 7422 | if ((false || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(bias))) { |
| 7423 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7424 | TORCH_INTERNAL_ASSERT(false, |
| 7425 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7426 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7427 | } else { |
| 7428 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7429 | at::AutoDispatchSkipFunctionalize guard; |
| 7430 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_native_batch_norm_legit::call(input_, weight_, bias_, running_mean_, running_var_, training, momentum, eps); |
| 7431 | return ::std::tuple<at::Tensor,at::Tensor,at::Tensor>(std::get<0>(tmp_output), std::get<1>(tmp_output), std::get<2>(tmp_output));; |
| 7432 | } |
| 7433 | } else { |
| 7434 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 7435 | { |
| 7436 | at::AutoDispatchSkipFunctionalize guard; |
| 7437 | tmp_output = at::_ops::_native_batch_norm_legit_functional::call(input_, weight_, bias_, running_mean_, running_var_, training, momentum, eps); |
| 7438 | } |
| 7439 | auto output_0 = at::functionalization::impl::to_functional_tensor(std::get<0>(tmp_output)); |
| 7440 | auto output_1 = at::functionalization::impl::to_functional_tensor(std::get<1>(tmp_output)); |
| 7441 | auto output_2 = at::functionalization::impl::to_functional_tensor(std::get<2>(tmp_output)); |
| 7442 | at::functionalization::impl::replace_(running_mean, std::get<3>(tmp_output)); |
| 7443 | at::functionalization::impl::commit_update(running_mean); |
| 7444 | at::functionalization::impl::sync(running_mean); |
| 7445 | at::functionalization::impl::replace_(running_var, std::get<4>(tmp_output)); |
| 7446 | at::functionalization::impl::commit_update(running_var); |
| 7447 | at::functionalization::impl::sync(running_var); |
| 7448 | return ::std::tuple<at::Tensor,at::Tensor,at::Tensor>(output_0, output_1, output_2); |
| 7449 | } |
| 7450 | } |
| 7451 | |
| 7452 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> _native_batch_norm_legit_out_no_stats_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, const c10::optional<at::Tensor> & weight, const c10::optional<at::Tensor> & bias, bool training, double momentum, double eps, at::Tensor & out, at::Tensor & save_mean, at::Tensor & save_invstd) { |
| 7453 | if (false) { |
| 7454 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7455 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7456 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7457 | auto input_meta = to_meta(input); |
| 7458 | auto weight_meta = to_meta(weight); |
| 7459 | auto bias_meta = to_meta(bias); |
| 7460 | auto out_meta = to_meta(out); |
| 7461 | auto save_mean_meta = to_meta(save_mean); |
| 7462 | auto save_invstd_meta = to_meta(save_invstd); |
| 7463 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7464 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7465 | at::_ops::_native_batch_norm_legit_no_stats_out::call(input_meta, weight_meta, bias_meta, training, momentum, eps, out_meta, save_mean_meta, save_invstd_meta); |
| 7466 | } |
| 7467 | |
| 7468 | at::Tensor input_; |
| 7469 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 7470 | at::functionalization::impl::sync(input); |
| 7471 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 7472 | } else { |
| 7473 | input_ = input; |
| 7474 | } |
| 7475 | |
| 7476 | c10::optional<at::Tensor> weight_; |
| 7477 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 7478 | at::functionalization::impl::sync(weight); |
| 7479 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 7480 | } else { |
| 7481 | weight_ = weight; |
| 7482 | } |
| 7483 | |
| 7484 | c10::optional<at::Tensor> bias_; |
| 7485 | if (at::functionalization::impl::isFunctionalTensor(bias)) { |
| 7486 | at::functionalization::impl::sync(bias); |
| 7487 | bias_ = at::functionalization::impl::from_functional_tensor(bias); |
| 7488 | } else { |
| 7489 | bias_ = bias; |
| 7490 | } |
| 7491 | |
| 7492 | at::Tensor out_; |
| 7493 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 7494 | at::functionalization::impl::sync(out); |
| 7495 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 7496 | } else { |
| 7497 | out_ = out; |
| 7498 | } |
| 7499 | |
| 7500 | at::Tensor save_mean_; |
| 7501 | if (at::functionalization::impl::isFunctionalTensor(save_mean)) { |
| 7502 | at::functionalization::impl::sync(save_mean); |
| 7503 | save_mean_ = at::functionalization::impl::from_functional_tensor(save_mean); |
| 7504 | } else { |
| 7505 | save_mean_ = save_mean; |
| 7506 | } |
| 7507 | |
| 7508 | at::Tensor save_invstd_; |
| 7509 | if (at::functionalization::impl::isFunctionalTensor(save_invstd)) { |
| 7510 | at::functionalization::impl::sync(save_invstd); |
| 7511 | save_invstd_ = at::functionalization::impl::from_functional_tensor(save_invstd); |
| 7512 | } else { |
| 7513 | save_invstd_ = save_invstd; |
| 7514 | } |
| 7515 | if (!(true && at::functionalization::impl::isFunctionalTensor(out) && at::functionalization::impl::isFunctionalTensor(save_mean) && at::functionalization::impl::isFunctionalTensor(save_invstd))) { |
| 7516 | if ((false || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(bias))) { |
| 7517 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7518 | TORCH_INTERNAL_ASSERT(false, |
| 7519 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7520 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7521 | } else { |
| 7522 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7523 | at::AutoDispatchSkipFunctionalize guard; |
| 7524 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_native_batch_norm_legit_no_stats_out::call(input_, weight_, bias_, training, momentum, eps, out_, save_mean_, save_invstd_); |
| 7525 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out, save_mean, save_invstd);; |
| 7526 | } |
| 7527 | } else { |
| 7528 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 7529 | { |
| 7530 | at::AutoDispatchSkipFunctionalize guard; |
| 7531 | tmp_output = at::_ops::_native_batch_norm_legit_no_stats::call(input_, weight_, bias_, training, momentum, eps); |
| 7532 | } |
| 7533 | at::functionalization::impl::replace_(out, std::get<0>(tmp_output)); |
| 7534 | at::functionalization::impl::commit_update(out); |
| 7535 | at::functionalization::impl::sync(out); |
| 7536 | at::functionalization::impl::replace_(save_mean, std::get<1>(tmp_output)); |
| 7537 | at::functionalization::impl::commit_update(save_mean); |
| 7538 | at::functionalization::impl::sync(save_mean); |
| 7539 | at::functionalization::impl::replace_(save_invstd, std::get<2>(tmp_output)); |
| 7540 | at::functionalization::impl::commit_update(save_invstd); |
| 7541 | at::functionalization::impl::sync(save_invstd); |
| 7542 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out, save_mean, save_invstd); |
| 7543 | } |
| 7544 | } |
| 7545 | |
| 7546 | ::std::tuple<at::Tensor &,at::Tensor &> batch_norm_stats_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input, double eps, at::Tensor & out0, at::Tensor & out1) { |
| 7547 | if (false) { |
| 7548 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7549 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7550 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7551 | auto input_meta = to_meta(input); |
| 7552 | auto out0_meta = to_meta(out0); |
| 7553 | auto out1_meta = to_meta(out1); |
| 7554 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7555 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7556 | at::_ops::batch_norm_stats_out::call(input_meta, eps, out0_meta, out1_meta); |
| 7557 | } |
| 7558 | |
| 7559 | at::Tensor input_; |
| 7560 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 7561 | at::functionalization::impl::sync(input); |
| 7562 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 7563 | } else { |
| 7564 | input_ = input; |
| 7565 | } |
| 7566 | |
| 7567 | at::Tensor out0_; |
| 7568 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 7569 | at::functionalization::impl::sync(out0); |
| 7570 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 7571 | } else { |
| 7572 | out0_ = out0; |
| 7573 | } |
| 7574 | |
| 7575 | at::Tensor out1_; |
| 7576 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 7577 | at::functionalization::impl::sync(out1); |
| 7578 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 7579 | } else { |
| 7580 | out1_ = out1; |
| 7581 | } |
| 7582 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1))) { |
| 7583 | if ((false || at::functionalization::impl::isFunctionalTensor(input))) { |
| 7584 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7585 | TORCH_INTERNAL_ASSERT(false, |
| 7586 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7587 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7588 | } else { |
| 7589 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7590 | at::AutoDispatchSkipFunctionalize guard; |
| 7591 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::batch_norm_stats_out::call(input_, eps, out0_, out1_); |
| 7592 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1);; |
| 7593 | } |
| 7594 | } else { |
| 7595 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 7596 | { |
| 7597 | at::AutoDispatchSkipFunctionalize guard; |
| 7598 | tmp_output = at::_ops::batch_norm_stats::call(input_, eps); |
| 7599 | } |
| 7600 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 7601 | at::functionalization::impl::commit_update(out0); |
| 7602 | at::functionalization::impl::sync(out0); |
| 7603 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 7604 | at::functionalization::impl::commit_update(out1); |
| 7605 | at::functionalization::impl::sync(out1); |
| 7606 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1); |
| 7607 | } |
| 7608 | } |
| 7609 | |
| 7610 | at::Tensor & batch_norm_backward_elemt_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_out, const at::Tensor & input, const at::Tensor & mean, const at::Tensor & invstd, const c10::optional<at::Tensor> & weight, const at::Tensor & mean_dy, const at::Tensor & mean_dy_xmu, const at::Tensor & count, at::Tensor & out) { |
| 7611 | if (false) { |
| 7612 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7613 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7614 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7615 | auto grad_out_meta = to_meta(grad_out); |
| 7616 | auto input_meta = to_meta(input); |
| 7617 | auto mean_meta = to_meta(mean); |
| 7618 | auto invstd_meta = to_meta(invstd); |
| 7619 | auto weight_meta = to_meta(weight); |
| 7620 | auto mean_dy_meta = to_meta(mean_dy); |
| 7621 | auto mean_dy_xmu_meta = to_meta(mean_dy_xmu); |
| 7622 | auto count_meta = to_meta(count); |
| 7623 | auto out_meta = to_meta(out); |
| 7624 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7625 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7626 | at::_ops::batch_norm_backward_elemt_out::call(grad_out_meta, input_meta, mean_meta, invstd_meta, weight_meta, mean_dy_meta, mean_dy_xmu_meta, count_meta, out_meta); |
| 7627 | } |
| 7628 | |
| 7629 | at::Tensor grad_out_; |
| 7630 | if (at::functionalization::impl::isFunctionalTensor(grad_out)) { |
| 7631 | at::functionalization::impl::sync(grad_out); |
| 7632 | grad_out_ = at::functionalization::impl::from_functional_tensor(grad_out); |
| 7633 | } else { |
| 7634 | grad_out_ = grad_out; |
| 7635 | } |
| 7636 | |
| 7637 | at::Tensor input_; |
| 7638 | if (at::functionalization::impl::isFunctionalTensor(input)) { |
| 7639 | at::functionalization::impl::sync(input); |
| 7640 | input_ = at::functionalization::impl::from_functional_tensor(input); |
| 7641 | } else { |
| 7642 | input_ = input; |
| 7643 | } |
| 7644 | |
| 7645 | at::Tensor mean_; |
| 7646 | if (at::functionalization::impl::isFunctionalTensor(mean)) { |
| 7647 | at::functionalization::impl::sync(mean); |
| 7648 | mean_ = at::functionalization::impl::from_functional_tensor(mean); |
| 7649 | } else { |
| 7650 | mean_ = mean; |
| 7651 | } |
| 7652 | |
| 7653 | at::Tensor invstd_; |
| 7654 | if (at::functionalization::impl::isFunctionalTensor(invstd)) { |
| 7655 | at::functionalization::impl::sync(invstd); |
| 7656 | invstd_ = at::functionalization::impl::from_functional_tensor(invstd); |
| 7657 | } else { |
| 7658 | invstd_ = invstd; |
| 7659 | } |
| 7660 | |
| 7661 | c10::optional<at::Tensor> weight_; |
| 7662 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 7663 | at::functionalization::impl::sync(weight); |
| 7664 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 7665 | } else { |
| 7666 | weight_ = weight; |
| 7667 | } |
| 7668 | |
| 7669 | at::Tensor mean_dy_; |
| 7670 | if (at::functionalization::impl::isFunctionalTensor(mean_dy)) { |
| 7671 | at::functionalization::impl::sync(mean_dy); |
| 7672 | mean_dy_ = at::functionalization::impl::from_functional_tensor(mean_dy); |
| 7673 | } else { |
| 7674 | mean_dy_ = mean_dy; |
| 7675 | } |
| 7676 | |
| 7677 | at::Tensor mean_dy_xmu_; |
| 7678 | if (at::functionalization::impl::isFunctionalTensor(mean_dy_xmu)) { |
| 7679 | at::functionalization::impl::sync(mean_dy_xmu); |
| 7680 | mean_dy_xmu_ = at::functionalization::impl::from_functional_tensor(mean_dy_xmu); |
| 7681 | } else { |
| 7682 | mean_dy_xmu_ = mean_dy_xmu; |
| 7683 | } |
| 7684 | |
| 7685 | at::Tensor count_; |
| 7686 | if (at::functionalization::impl::isFunctionalTensor(count)) { |
| 7687 | at::functionalization::impl::sync(count); |
| 7688 | count_ = at::functionalization::impl::from_functional_tensor(count); |
| 7689 | } else { |
| 7690 | count_ = count; |
| 7691 | } |
| 7692 | |
| 7693 | at::Tensor out_; |
| 7694 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 7695 | at::functionalization::impl::sync(out); |
| 7696 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 7697 | } else { |
| 7698 | out_ = out; |
| 7699 | } |
| 7700 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 7701 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_out) || at::functionalization::impl::isFunctionalTensor(input) || at::functionalization::impl::isFunctionalTensor(mean) || at::functionalization::impl::isFunctionalTensor(invstd) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(mean_dy) || at::functionalization::impl::isFunctionalTensor(mean_dy_xmu) || at::functionalization::impl::isFunctionalTensor(count))) { |
| 7702 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7703 | TORCH_INTERNAL_ASSERT(false, |
| 7704 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7705 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7706 | } else { |
| 7707 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7708 | at::AutoDispatchSkipFunctionalize guard; |
| 7709 | at::Tensor tmp_output = at::_ops::batch_norm_backward_elemt_out::call(grad_out_, input_, mean_, invstd_, weight_, mean_dy_, mean_dy_xmu_, count_, out_); |
| 7710 | return out;; |
| 7711 | } |
| 7712 | } else { |
| 7713 | at::Tensor tmp_output; |
| 7714 | { |
| 7715 | at::AutoDispatchSkipFunctionalize guard; |
| 7716 | tmp_output = at::_ops::batch_norm_backward_elemt::call(grad_out_, input_, mean_, invstd_, weight_, mean_dy_, mean_dy_xmu_, count_); |
| 7717 | } |
| 7718 | at::functionalization::impl::replace_(out, tmp_output); |
| 7719 | at::functionalization::impl::commit_update(out); |
| 7720 | at::functionalization::impl::sync(out); |
| 7721 | return out; |
| 7722 | } |
| 7723 | } |
| 7724 | |
| 7725 | at::Tensor & _euclidean_dist_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & x1, const at::Tensor & x2, at::Tensor & out) { |
| 7726 | if (false) { |
| 7727 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7728 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7729 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7730 | auto x1_meta = to_meta(x1); |
| 7731 | auto x2_meta = to_meta(x2); |
| 7732 | auto out_meta = to_meta(out); |
| 7733 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7734 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7735 | at::_ops::_euclidean_dist_out::call(x1_meta, x2_meta, out_meta); |
| 7736 | } |
| 7737 | |
| 7738 | at::Tensor x1_; |
| 7739 | if (at::functionalization::impl::isFunctionalTensor(x1)) { |
| 7740 | at::functionalization::impl::sync(x1); |
| 7741 | x1_ = at::functionalization::impl::from_functional_tensor(x1); |
| 7742 | } else { |
| 7743 | x1_ = x1; |
| 7744 | } |
| 7745 | |
| 7746 | at::Tensor x2_; |
| 7747 | if (at::functionalization::impl::isFunctionalTensor(x2)) { |
| 7748 | at::functionalization::impl::sync(x2); |
| 7749 | x2_ = at::functionalization::impl::from_functional_tensor(x2); |
| 7750 | } else { |
| 7751 | x2_ = x2; |
| 7752 | } |
| 7753 | |
| 7754 | at::Tensor out_; |
| 7755 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 7756 | at::functionalization::impl::sync(out); |
| 7757 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 7758 | } else { |
| 7759 | out_ = out; |
| 7760 | } |
| 7761 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 7762 | if ((false || at::functionalization::impl::isFunctionalTensor(x1) || at::functionalization::impl::isFunctionalTensor(x2))) { |
| 7763 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7764 | TORCH_INTERNAL_ASSERT(false, |
| 7765 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7766 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7767 | } else { |
| 7768 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7769 | at::AutoDispatchSkipFunctionalize guard; |
| 7770 | at::Tensor tmp_output = at::_ops::_euclidean_dist_out::call(x1_, x2_, out_); |
| 7771 | return out;; |
| 7772 | } |
| 7773 | } else { |
| 7774 | at::Tensor tmp_output; |
| 7775 | { |
| 7776 | at::AutoDispatchSkipFunctionalize guard; |
| 7777 | tmp_output = at::_ops::_euclidean_dist::call(x1_, x2_); |
| 7778 | } |
| 7779 | at::functionalization::impl::replace_(out, tmp_output); |
| 7780 | at::functionalization::impl::commit_update(out); |
| 7781 | at::functionalization::impl::sync(out); |
| 7782 | return out; |
| 7783 | } |
| 7784 | } |
| 7785 | |
| 7786 | at::Tensor & _cdist_forward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & x1, const at::Tensor & x2, double p, c10::optional<int64_t> compute_mode, at::Tensor & out) { |
| 7787 | if (false) { |
| 7788 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7789 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7790 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7791 | auto x1_meta = to_meta(x1); |
| 7792 | auto x2_meta = to_meta(x2); |
| 7793 | auto out_meta = to_meta(out); |
| 7794 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7795 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7796 | at::_ops::_cdist_forward_out::call(x1_meta, x2_meta, p, compute_mode, out_meta); |
| 7797 | } |
| 7798 | |
| 7799 | at::Tensor x1_; |
| 7800 | if (at::functionalization::impl::isFunctionalTensor(x1)) { |
| 7801 | at::functionalization::impl::sync(x1); |
| 7802 | x1_ = at::functionalization::impl::from_functional_tensor(x1); |
| 7803 | } else { |
| 7804 | x1_ = x1; |
| 7805 | } |
| 7806 | |
| 7807 | at::Tensor x2_; |
| 7808 | if (at::functionalization::impl::isFunctionalTensor(x2)) { |
| 7809 | at::functionalization::impl::sync(x2); |
| 7810 | x2_ = at::functionalization::impl::from_functional_tensor(x2); |
| 7811 | } else { |
| 7812 | x2_ = x2; |
| 7813 | } |
| 7814 | |
| 7815 | at::Tensor out_; |
| 7816 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 7817 | at::functionalization::impl::sync(out); |
| 7818 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 7819 | } else { |
| 7820 | out_ = out; |
| 7821 | } |
| 7822 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 7823 | if ((false || at::functionalization::impl::isFunctionalTensor(x1) || at::functionalization::impl::isFunctionalTensor(x2))) { |
| 7824 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7825 | TORCH_INTERNAL_ASSERT(false, |
| 7826 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7827 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7828 | } else { |
| 7829 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7830 | at::AutoDispatchSkipFunctionalize guard; |
| 7831 | at::Tensor tmp_output = at::_ops::_cdist_forward_out::call(x1_, x2_, p, compute_mode, out_); |
| 7832 | return out;; |
| 7833 | } |
| 7834 | } else { |
| 7835 | at::Tensor tmp_output; |
| 7836 | { |
| 7837 | at::AutoDispatchSkipFunctionalize guard; |
| 7838 | tmp_output = at::_ops::_cdist_forward::call(x1_, x2_, p, compute_mode); |
| 7839 | } |
| 7840 | at::functionalization::impl::replace_(out, tmp_output); |
| 7841 | at::functionalization::impl::commit_update(out); |
| 7842 | at::functionalization::impl::sync(out); |
| 7843 | return out; |
| 7844 | } |
| 7845 | } |
| 7846 | |
| 7847 | at::Tensor & _cdist_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad, const at::Tensor & x1, const at::Tensor & x2, double p, const at::Tensor & cdist, at::Tensor & out) { |
| 7848 | if (false) { |
| 7849 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7850 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7851 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7852 | auto grad_meta = to_meta(grad); |
| 7853 | auto x1_meta = to_meta(x1); |
| 7854 | auto x2_meta = to_meta(x2); |
| 7855 | auto cdist_meta = to_meta(cdist); |
| 7856 | auto out_meta = to_meta(out); |
| 7857 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7858 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7859 | at::_ops::_cdist_backward_out::call(grad_meta, x1_meta, x2_meta, p, cdist_meta, out_meta); |
| 7860 | } |
| 7861 | |
| 7862 | at::Tensor grad_; |
| 7863 | if (at::functionalization::impl::isFunctionalTensor(grad)) { |
| 7864 | at::functionalization::impl::sync(grad); |
| 7865 | grad_ = at::functionalization::impl::from_functional_tensor(grad); |
| 7866 | } else { |
| 7867 | grad_ = grad; |
| 7868 | } |
| 7869 | |
| 7870 | at::Tensor x1_; |
| 7871 | if (at::functionalization::impl::isFunctionalTensor(x1)) { |
| 7872 | at::functionalization::impl::sync(x1); |
| 7873 | x1_ = at::functionalization::impl::from_functional_tensor(x1); |
| 7874 | } else { |
| 7875 | x1_ = x1; |
| 7876 | } |
| 7877 | |
| 7878 | at::Tensor x2_; |
| 7879 | if (at::functionalization::impl::isFunctionalTensor(x2)) { |
| 7880 | at::functionalization::impl::sync(x2); |
| 7881 | x2_ = at::functionalization::impl::from_functional_tensor(x2); |
| 7882 | } else { |
| 7883 | x2_ = x2; |
| 7884 | } |
| 7885 | |
| 7886 | at::Tensor cdist_; |
| 7887 | if (at::functionalization::impl::isFunctionalTensor(cdist)) { |
| 7888 | at::functionalization::impl::sync(cdist); |
| 7889 | cdist_ = at::functionalization::impl::from_functional_tensor(cdist); |
| 7890 | } else { |
| 7891 | cdist_ = cdist; |
| 7892 | } |
| 7893 | |
| 7894 | at::Tensor out_; |
| 7895 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 7896 | at::functionalization::impl::sync(out); |
| 7897 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 7898 | } else { |
| 7899 | out_ = out; |
| 7900 | } |
| 7901 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 7902 | if ((false || at::functionalization::impl::isFunctionalTensor(grad) || at::functionalization::impl::isFunctionalTensor(x1) || at::functionalization::impl::isFunctionalTensor(x2) || at::functionalization::impl::isFunctionalTensor(cdist))) { |
| 7903 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7904 | TORCH_INTERNAL_ASSERT(false, |
| 7905 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7906 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7907 | } else { |
| 7908 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7909 | at::AutoDispatchSkipFunctionalize guard; |
| 7910 | at::Tensor tmp_output = at::_ops::_cdist_backward_out::call(grad_, x1_, x2_, p, cdist_, out_); |
| 7911 | return out;; |
| 7912 | } |
| 7913 | } else { |
| 7914 | at::Tensor tmp_output; |
| 7915 | { |
| 7916 | at::AutoDispatchSkipFunctionalize guard; |
| 7917 | tmp_output = at::_ops::_cdist_backward::call(grad_, x1_, x2_, p, cdist_); |
| 7918 | } |
| 7919 | at::functionalization::impl::replace_(out, tmp_output); |
| 7920 | at::functionalization::impl::commit_update(out); |
| 7921 | at::functionalization::impl::sync(out); |
| 7922 | return out; |
| 7923 | } |
| 7924 | } |
| 7925 | |
| 7926 | at::Tensor & pixel_unshuffle_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t downscale_factor, at::Tensor & out) { |
| 7927 | if (false) { |
| 7928 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7929 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7930 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7931 | auto self_meta = to_meta(self); |
| 7932 | auto out_meta = to_meta(out); |
| 7933 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7934 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7935 | at::_ops::pixel_unshuffle_out::call(self_meta, downscale_factor, out_meta); |
| 7936 | } |
| 7937 | |
| 7938 | at::Tensor self_; |
| 7939 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 7940 | at::functionalization::impl::sync(self); |
| 7941 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 7942 | } else { |
| 7943 | self_ = self; |
| 7944 | } |
| 7945 | |
| 7946 | at::Tensor out_; |
| 7947 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 7948 | at::functionalization::impl::sync(out); |
| 7949 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 7950 | } else { |
| 7951 | out_ = out; |
| 7952 | } |
| 7953 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 7954 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 7955 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 7956 | TORCH_INTERNAL_ASSERT(false, |
| 7957 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 7958 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 7959 | } else { |
| 7960 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 7961 | at::AutoDispatchSkipFunctionalize guard; |
| 7962 | at::Tensor tmp_output = at::_ops::pixel_unshuffle_out::call(self_, downscale_factor, out_); |
| 7963 | return out;; |
| 7964 | } |
| 7965 | } else { |
| 7966 | at::Tensor tmp_output; |
| 7967 | { |
| 7968 | at::AutoDispatchSkipFunctionalize guard; |
| 7969 | tmp_output = at::_ops::pixel_unshuffle::call(self_, downscale_factor); |
| 7970 | } |
| 7971 | at::functionalization::impl::replace_(out, tmp_output); |
| 7972 | at::functionalization::impl::commit_update(out); |
| 7973 | at::functionalization::impl::sync(out); |
| 7974 | return out; |
| 7975 | } |
| 7976 | } |
| 7977 | |
| 7978 | at::Tensor & rad2deg_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 7979 | if (false) { |
| 7980 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 7981 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 7982 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 7983 | auto self_meta = to_meta(self); |
| 7984 | auto out_meta = to_meta(out); |
| 7985 | at::AutoDispatchSkipFunctionalize func_guard; |
| 7986 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 7987 | at::_ops::rad2deg_out::call(self_meta, out_meta); |
| 7988 | } |
| 7989 | |
| 7990 | at::Tensor self_; |
| 7991 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 7992 | at::functionalization::impl::sync(self); |
| 7993 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 7994 | } else { |
| 7995 | self_ = self; |
| 7996 | } |
| 7997 | |
| 7998 | at::Tensor out_; |
| 7999 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8000 | at::functionalization::impl::sync(out); |
| 8001 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8002 | } else { |
| 8003 | out_ = out; |
| 8004 | } |
| 8005 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8006 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8007 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8008 | TORCH_INTERNAL_ASSERT(false, |
| 8009 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8010 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8011 | } else { |
| 8012 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8013 | at::AutoDispatchSkipFunctionalize guard; |
| 8014 | at::Tensor tmp_output = at::_ops::rad2deg_out::call(self_, out_); |
| 8015 | return out;; |
| 8016 | } |
| 8017 | } else { |
| 8018 | at::Tensor tmp_output; |
| 8019 | { |
| 8020 | at::AutoDispatchSkipFunctionalize guard; |
| 8021 | tmp_output = at::_ops::rad2deg::call(self_); |
| 8022 | } |
| 8023 | at::functionalization::impl::replace_(out, tmp_output); |
| 8024 | at::functionalization::impl::commit_update(out); |
| 8025 | at::functionalization::impl::sync(out); |
| 8026 | return out; |
| 8027 | } |
| 8028 | } |
| 8029 | |
| 8030 | at::Tensor & rad2deg_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 8031 | if (true) { |
| 8032 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8033 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8034 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8035 | auto self_meta = to_meta(self); |
| 8036 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8037 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8038 | at::_ops::rad2deg_::call(self_meta); |
| 8039 | } |
| 8040 | |
| 8041 | at::Tensor self_; |
| 8042 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8043 | at::functionalization::impl::sync(self); |
| 8044 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8045 | } else { |
| 8046 | self_ = self; |
| 8047 | } |
| 8048 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 8049 | if ((false)) { |
| 8050 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8051 | TORCH_INTERNAL_ASSERT(false, |
| 8052 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8053 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8054 | } else { |
| 8055 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8056 | at::AutoDispatchSkipFunctionalize guard; |
| 8057 | at::Tensor tmp_output = at::_ops::rad2deg_::call(self_); |
| 8058 | return self;; |
| 8059 | } |
| 8060 | } else { |
| 8061 | at::Tensor tmp_output; |
| 8062 | { |
| 8063 | at::AutoDispatchSkipFunctionalize guard; |
| 8064 | tmp_output = at::_ops::rad2deg::call(self_); |
| 8065 | } |
| 8066 | at::functionalization::impl::replace_(self, tmp_output); |
| 8067 | at::functionalization::impl::commit_update(self); |
| 8068 | at::functionalization::impl::sync(self); |
| 8069 | return self; |
| 8070 | } |
| 8071 | } |
| 8072 | |
| 8073 | at::Tensor & scalar_tensor_out_out(c10::DispatchKeySet dispatchKeySet, const at::Scalar & s, at::Tensor & out) { |
| 8074 | if (false) { |
| 8075 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8076 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8077 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8078 | auto out_meta = to_meta(out); |
| 8079 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8080 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8081 | at::_ops::scalar_tensor_out::call(s, out_meta); |
| 8082 | } |
| 8083 | |
| 8084 | at::Tensor out_; |
| 8085 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8086 | at::functionalization::impl::sync(out); |
| 8087 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8088 | } else { |
| 8089 | out_ = out; |
| 8090 | } |
| 8091 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8092 | if ((false)) { |
| 8093 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8094 | TORCH_INTERNAL_ASSERT(false, |
| 8095 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8096 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8097 | } else { |
| 8098 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8099 | at::AutoDispatchSkipFunctionalize guard; |
| 8100 | at::Tensor tmp_output = at::_ops::scalar_tensor_out::call(s, out_); |
| 8101 | return out;; |
| 8102 | } |
| 8103 | } else { |
| 8104 | at::Tensor tmp_output; |
| 8105 | { |
| 8106 | at::AutoDispatchSkipFunctionalize guard; |
| 8107 | tmp_output = at::_ops::scalar_tensor::call(s, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 8108 | } |
| 8109 | at::functionalization::impl::replace_(out, tmp_output); |
| 8110 | at::functionalization::impl::commit_update(out); |
| 8111 | at::functionalization::impl::sync(out); |
| 8112 | return out; |
| 8113 | } |
| 8114 | } |
| 8115 | |
| 8116 | at::Tensor & rand_out_names_out(c10::DispatchKeySet dispatchKeySet, c10::SymIntArrayRef size, c10::optional<at::DimnameList> names, at::Tensor & out) { |
| 8117 | if (false) { |
| 8118 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8119 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8120 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8121 | auto out_meta = to_meta(out); |
| 8122 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8123 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8124 | at::_ops::rand_names_out::call(size, names, out_meta); |
| 8125 | } |
| 8126 | |
| 8127 | at::Tensor out_; |
| 8128 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8129 | at::functionalization::impl::sync(out); |
| 8130 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8131 | } else { |
| 8132 | out_ = out; |
| 8133 | } |
| 8134 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8135 | if ((false)) { |
| 8136 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8137 | TORCH_INTERNAL_ASSERT(false, |
| 8138 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8139 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8140 | } else { |
| 8141 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8142 | at::AutoDispatchSkipFunctionalize guard; |
| 8143 | at::Tensor tmp_output = at::_ops::rand_names_out::call(size, names, out_); |
| 8144 | return out;; |
| 8145 | } |
| 8146 | } else { |
| 8147 | at::Tensor tmp_output; |
| 8148 | { |
| 8149 | at::AutoDispatchSkipFunctionalize guard; |
| 8150 | tmp_output = at::_ops::rand_names::call(size, names, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 8151 | } |
| 8152 | at::functionalization::impl::replace_(out, tmp_output); |
| 8153 | at::functionalization::impl::commit_update(out); |
| 8154 | at::functionalization::impl::sync(out); |
| 8155 | return out; |
| 8156 | } |
| 8157 | } |
| 8158 | |
| 8159 | at::Tensor & rand_out_generator_with_names_out(c10::DispatchKeySet dispatchKeySet, c10::SymIntArrayRef size, c10::optional<at::Generator> generator, c10::optional<at::DimnameList> names, at::Tensor & out) { |
| 8160 | if (false) { |
| 8161 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8162 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8163 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8164 | auto out_meta = to_meta(out); |
| 8165 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8166 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8167 | at::_ops::rand_generator_with_names_out::call(size, generator, names, out_meta); |
| 8168 | } |
| 8169 | |
| 8170 | at::Tensor out_; |
| 8171 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8172 | at::functionalization::impl::sync(out); |
| 8173 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8174 | } else { |
| 8175 | out_ = out; |
| 8176 | } |
| 8177 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8178 | if ((false)) { |
| 8179 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8180 | TORCH_INTERNAL_ASSERT(false, |
| 8181 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8182 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8183 | } else { |
| 8184 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8185 | at::AutoDispatchSkipFunctionalize guard; |
| 8186 | at::Tensor tmp_output = at::_ops::rand_generator_with_names_out::call(size, generator, names, out_); |
| 8187 | return out;; |
| 8188 | } |
| 8189 | } else { |
| 8190 | at::Tensor tmp_output; |
| 8191 | { |
| 8192 | at::AutoDispatchSkipFunctionalize guard; |
| 8193 | tmp_output = at::_ops::rand_generator_with_names::call(size, generator, names, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 8194 | } |
| 8195 | at::functionalization::impl::replace_(out, tmp_output); |
| 8196 | at::functionalization::impl::commit_update(out); |
| 8197 | at::functionalization::impl::sync(out); |
| 8198 | return out; |
| 8199 | } |
| 8200 | } |
| 8201 | |
| 8202 | at::Tensor & rand_out_out(c10::DispatchKeySet dispatchKeySet, c10::SymIntArrayRef size, at::Tensor & out) { |
| 8203 | if (false) { |
| 8204 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8205 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8206 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8207 | auto out_meta = to_meta(out); |
| 8208 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8209 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8210 | at::_ops::rand_out::call(size, out_meta); |
| 8211 | } |
| 8212 | |
| 8213 | at::Tensor out_; |
| 8214 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8215 | at::functionalization::impl::sync(out); |
| 8216 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8217 | } else { |
| 8218 | out_ = out; |
| 8219 | } |
| 8220 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8221 | if ((false)) { |
| 8222 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8223 | TORCH_INTERNAL_ASSERT(false, |
| 8224 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8225 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8226 | } else { |
| 8227 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8228 | at::AutoDispatchSkipFunctionalize guard; |
| 8229 | at::Tensor tmp_output = at::_ops::rand_out::call(size, out_); |
| 8230 | return out;; |
| 8231 | } |
| 8232 | } else { |
| 8233 | at::Tensor tmp_output; |
| 8234 | { |
| 8235 | at::AutoDispatchSkipFunctionalize guard; |
| 8236 | tmp_output = at::_ops::rand::call(size, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 8237 | } |
| 8238 | at::functionalization::impl::replace_(out, tmp_output); |
| 8239 | at::functionalization::impl::commit_update(out); |
| 8240 | at::functionalization::impl::sync(out); |
| 8241 | return out; |
| 8242 | } |
| 8243 | } |
| 8244 | |
| 8245 | at::Tensor & rand_out_generator_out(c10::DispatchKeySet dispatchKeySet, c10::SymIntArrayRef size, c10::optional<at::Generator> generator, at::Tensor & out) { |
| 8246 | if (false) { |
| 8247 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8248 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8249 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8250 | auto out_meta = to_meta(out); |
| 8251 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8252 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8253 | at::_ops::rand_generator_out::call(size, generator, out_meta); |
| 8254 | } |
| 8255 | |
| 8256 | at::Tensor out_; |
| 8257 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8258 | at::functionalization::impl::sync(out); |
| 8259 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8260 | } else { |
| 8261 | out_ = out; |
| 8262 | } |
| 8263 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8264 | if ((false)) { |
| 8265 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8266 | TORCH_INTERNAL_ASSERT(false, |
| 8267 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8268 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8269 | } else { |
| 8270 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8271 | at::AutoDispatchSkipFunctionalize guard; |
| 8272 | at::Tensor tmp_output = at::_ops::rand_generator_out::call(size, generator, out_); |
| 8273 | return out;; |
| 8274 | } |
| 8275 | } else { |
| 8276 | at::Tensor tmp_output; |
| 8277 | { |
| 8278 | at::AutoDispatchSkipFunctionalize guard; |
| 8279 | tmp_output = at::_ops::rand_generator::call(size, generator, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 8280 | } |
| 8281 | at::functionalization::impl::replace_(out, tmp_output); |
| 8282 | at::functionalization::impl::commit_update(out); |
| 8283 | at::functionalization::impl::sync(out); |
| 8284 | return out; |
| 8285 | } |
| 8286 | } |
| 8287 | |
| 8288 | at::Tensor & rand_like_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<at::MemoryFormat> memory_format, at::Tensor & out) { |
| 8289 | if (false) { |
| 8290 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8291 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8292 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8293 | auto self_meta = to_meta(self); |
| 8294 | auto out_meta = to_meta(out); |
| 8295 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8296 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8297 | at::_ops::rand_like_out::call(self_meta, memory_format, out_meta); |
| 8298 | } |
| 8299 | |
| 8300 | at::Tensor self_; |
| 8301 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8302 | at::functionalization::impl::sync(self); |
| 8303 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8304 | } else { |
| 8305 | self_ = self; |
| 8306 | } |
| 8307 | |
| 8308 | at::Tensor out_; |
| 8309 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8310 | at::functionalization::impl::sync(out); |
| 8311 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8312 | } else { |
| 8313 | out_ = out; |
| 8314 | } |
| 8315 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8316 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8317 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8318 | TORCH_INTERNAL_ASSERT(false, |
| 8319 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8320 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8321 | } else { |
| 8322 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8323 | at::AutoDispatchSkipFunctionalize guard; |
| 8324 | at::Tensor tmp_output = at::_ops::rand_like_out::call(self_, memory_format, out_); |
| 8325 | return out;; |
| 8326 | } |
| 8327 | } else { |
| 8328 | at::Tensor tmp_output; |
| 8329 | { |
| 8330 | at::AutoDispatchSkipFunctionalize guard; |
| 8331 | tmp_output = at::_ops::rand_like::call(self_, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt, memory_format); |
| 8332 | } |
| 8333 | at::functionalization::impl::replace_(out, tmp_output); |
| 8334 | at::functionalization::impl::commit_update(out); |
| 8335 | at::functionalization::impl::sync(out); |
| 8336 | return out; |
| 8337 | } |
| 8338 | } |
| 8339 | |
| 8340 | at::Tensor & relu_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 8341 | if (false) { |
| 8342 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8343 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8344 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8345 | auto self_meta = to_meta(self); |
| 8346 | auto out_meta = to_meta(out); |
| 8347 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8348 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8349 | at::_ops::relu_out::call(self_meta, out_meta); |
| 8350 | } |
| 8351 | |
| 8352 | at::Tensor self_; |
| 8353 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8354 | at::functionalization::impl::sync(self); |
| 8355 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8356 | } else { |
| 8357 | self_ = self; |
| 8358 | } |
| 8359 | |
| 8360 | at::Tensor out_; |
| 8361 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8362 | at::functionalization::impl::sync(out); |
| 8363 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8364 | } else { |
| 8365 | out_ = out; |
| 8366 | } |
| 8367 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8368 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8369 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8370 | TORCH_INTERNAL_ASSERT(false, |
| 8371 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8372 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8373 | } else { |
| 8374 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8375 | at::AutoDispatchSkipFunctionalize guard; |
| 8376 | at::Tensor tmp_output = at::_ops::relu_out::call(self_, out_); |
| 8377 | return out;; |
| 8378 | } |
| 8379 | } else { |
| 8380 | at::Tensor tmp_output; |
| 8381 | { |
| 8382 | at::AutoDispatchSkipFunctionalize guard; |
| 8383 | tmp_output = at::_ops::relu::call(self_); |
| 8384 | } |
| 8385 | at::functionalization::impl::replace_(out, tmp_output); |
| 8386 | at::functionalization::impl::commit_update(out); |
| 8387 | at::functionalization::impl::sync(out); |
| 8388 | return out; |
| 8389 | } |
| 8390 | } |
| 8391 | |
| 8392 | at::Tensor & relu_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 8393 | if (true) { |
| 8394 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8395 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8396 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8397 | auto self_meta = to_meta(self); |
| 8398 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8399 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8400 | at::_ops::relu_::call(self_meta); |
| 8401 | } |
| 8402 | |
| 8403 | at::Tensor self_; |
| 8404 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8405 | at::functionalization::impl::sync(self); |
| 8406 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8407 | } else { |
| 8408 | self_ = self; |
| 8409 | } |
| 8410 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 8411 | if ((false)) { |
| 8412 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8413 | TORCH_INTERNAL_ASSERT(false, |
| 8414 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8415 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8416 | } else { |
| 8417 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8418 | at::AutoDispatchSkipFunctionalize guard; |
| 8419 | at::Tensor tmp_output = at::_ops::relu_::call(self_); |
| 8420 | return self;; |
| 8421 | } |
| 8422 | } else { |
| 8423 | at::Tensor tmp_output; |
| 8424 | { |
| 8425 | at::AutoDispatchSkipFunctionalize guard; |
| 8426 | tmp_output = at::_ops::relu::call(self_); |
| 8427 | } |
| 8428 | at::functionalization::impl::replace_(self, tmp_output); |
| 8429 | at::functionalization::impl::commit_update(self); |
| 8430 | at::functionalization::impl::sync(self); |
| 8431 | return self; |
| 8432 | } |
| 8433 | } |
| 8434 | |
| 8435 | at::Tensor & logit_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<double> eps, at::Tensor & out) { |
| 8436 | if (false) { |
| 8437 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8438 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8439 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8440 | auto self_meta = to_meta(self); |
| 8441 | auto out_meta = to_meta(out); |
| 8442 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8443 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8444 | at::_ops::logit_out::call(self_meta, eps, out_meta); |
| 8445 | } |
| 8446 | |
| 8447 | at::Tensor self_; |
| 8448 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8449 | at::functionalization::impl::sync(self); |
| 8450 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8451 | } else { |
| 8452 | self_ = self; |
| 8453 | } |
| 8454 | |
| 8455 | at::Tensor out_; |
| 8456 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8457 | at::functionalization::impl::sync(out); |
| 8458 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8459 | } else { |
| 8460 | out_ = out; |
| 8461 | } |
| 8462 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8463 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8464 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8465 | TORCH_INTERNAL_ASSERT(false, |
| 8466 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8467 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8468 | } else { |
| 8469 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8470 | at::AutoDispatchSkipFunctionalize guard; |
| 8471 | at::Tensor tmp_output = at::_ops::logit_out::call(self_, eps, out_); |
| 8472 | return out;; |
| 8473 | } |
| 8474 | } else { |
| 8475 | at::Tensor tmp_output; |
| 8476 | { |
| 8477 | at::AutoDispatchSkipFunctionalize guard; |
| 8478 | tmp_output = at::_ops::logit::call(self_, eps); |
| 8479 | } |
| 8480 | at::functionalization::impl::replace_(out, tmp_output); |
| 8481 | at::functionalization::impl::commit_update(out); |
| 8482 | at::functionalization::impl::sync(out); |
| 8483 | return out; |
| 8484 | } |
| 8485 | } |
| 8486 | |
| 8487 | at::Tensor & logit_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, c10::optional<double> eps) { |
| 8488 | if (true) { |
| 8489 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8490 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8491 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8492 | auto self_meta = to_meta(self); |
| 8493 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8494 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8495 | at::_ops::logit_::call(self_meta, eps); |
| 8496 | } |
| 8497 | |
| 8498 | at::Tensor self_; |
| 8499 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8500 | at::functionalization::impl::sync(self); |
| 8501 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8502 | } else { |
| 8503 | self_ = self; |
| 8504 | } |
| 8505 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 8506 | if ((false)) { |
| 8507 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8508 | TORCH_INTERNAL_ASSERT(false, |
| 8509 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8510 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8511 | } else { |
| 8512 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8513 | at::AutoDispatchSkipFunctionalize guard; |
| 8514 | at::Tensor tmp_output = at::_ops::logit_::call(self_, eps); |
| 8515 | return self;; |
| 8516 | } |
| 8517 | } else { |
| 8518 | at::Tensor tmp_output; |
| 8519 | { |
| 8520 | at::AutoDispatchSkipFunctionalize guard; |
| 8521 | tmp_output = at::_ops::logit::call(self_, eps); |
| 8522 | } |
| 8523 | at::functionalization::impl::replace_(self, tmp_output); |
| 8524 | at::functionalization::impl::commit_update(self); |
| 8525 | at::functionalization::impl::sync(self); |
| 8526 | return self; |
| 8527 | } |
| 8528 | } |
| 8529 | |
| 8530 | at::Tensor & select_scatter_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & src, int64_t dim, c10::SymInt index, at::Tensor & out) { |
| 8531 | if (false) { |
| 8532 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8533 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8534 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8535 | auto self_meta = to_meta(self); |
| 8536 | auto src_meta = to_meta(src); |
| 8537 | auto out_meta = to_meta(out); |
| 8538 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8539 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8540 | at::_ops::select_scatter_out::call(self_meta, src_meta, dim, index, out_meta); |
| 8541 | } |
| 8542 | |
| 8543 | at::Tensor self_; |
| 8544 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8545 | at::functionalization::impl::sync(self); |
| 8546 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8547 | } else { |
| 8548 | self_ = self; |
| 8549 | } |
| 8550 | |
| 8551 | at::Tensor src_; |
| 8552 | if (at::functionalization::impl::isFunctionalTensor(src)) { |
| 8553 | at::functionalization::impl::sync(src); |
| 8554 | src_ = at::functionalization::impl::from_functional_tensor(src); |
| 8555 | } else { |
| 8556 | src_ = src; |
| 8557 | } |
| 8558 | |
| 8559 | at::Tensor out_; |
| 8560 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8561 | at::functionalization::impl::sync(out); |
| 8562 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8563 | } else { |
| 8564 | out_ = out; |
| 8565 | } |
| 8566 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8567 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(src))) { |
| 8568 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8569 | TORCH_INTERNAL_ASSERT(false, |
| 8570 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8571 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8572 | } else { |
| 8573 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8574 | at::AutoDispatchSkipFunctionalize guard; |
| 8575 | at::Tensor tmp_output = at::_ops::select_scatter_out::call(self_, src_, dim, index, out_); |
| 8576 | return out;; |
| 8577 | } |
| 8578 | } else { |
| 8579 | at::Tensor tmp_output; |
| 8580 | { |
| 8581 | at::AutoDispatchSkipFunctionalize guard; |
| 8582 | tmp_output = at::_ops::select_scatter::call(self_, src_, dim, index); |
| 8583 | } |
| 8584 | at::functionalization::impl::replace_(out, tmp_output); |
| 8585 | at::functionalization::impl::commit_update(out); |
| 8586 | at::functionalization::impl::sync(out); |
| 8587 | return out; |
| 8588 | } |
| 8589 | } |
| 8590 | |
| 8591 | at::Tensor & softmax_out_int_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 8592 | if (false) { |
| 8593 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8594 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8595 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8596 | auto self_meta = to_meta(self); |
| 8597 | auto out_meta = to_meta(out); |
| 8598 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8599 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8600 | at::_ops::softmax_int_out::call(self_meta, dim, dtype, out_meta); |
| 8601 | } |
| 8602 | |
| 8603 | at::Tensor self_; |
| 8604 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8605 | at::functionalization::impl::sync(self); |
| 8606 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8607 | } else { |
| 8608 | self_ = self; |
| 8609 | } |
| 8610 | |
| 8611 | at::Tensor out_; |
| 8612 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8613 | at::functionalization::impl::sync(out); |
| 8614 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8615 | } else { |
| 8616 | out_ = out; |
| 8617 | } |
| 8618 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8619 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8620 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8621 | TORCH_INTERNAL_ASSERT(false, |
| 8622 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8623 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8624 | } else { |
| 8625 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8626 | at::AutoDispatchSkipFunctionalize guard; |
| 8627 | at::Tensor tmp_output = at::_ops::softmax_int_out::call(self_, dim, dtype, out_); |
| 8628 | return out;; |
| 8629 | } |
| 8630 | } else { |
| 8631 | at::Tensor tmp_output; |
| 8632 | { |
| 8633 | at::AutoDispatchSkipFunctionalize guard; |
| 8634 | tmp_output = at::_ops::softmax_int::call(self_, dim, dtype); |
| 8635 | } |
| 8636 | at::functionalization::impl::replace_(out, tmp_output); |
| 8637 | at::functionalization::impl::commit_update(out); |
| 8638 | at::functionalization::impl::sync(out); |
| 8639 | return out; |
| 8640 | } |
| 8641 | } |
| 8642 | |
| 8643 | at::Tensor & stack_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList tensors, int64_t dim, at::Tensor & out) { |
| 8644 | if (false) { |
| 8645 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8646 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8647 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8648 | auto tensors_meta = to_meta(tensors); |
| 8649 | auto out_meta = to_meta(out); |
| 8650 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8651 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8652 | at::_ops::stack_out::call(tensors_meta, dim, out_meta); |
| 8653 | } |
| 8654 | |
| 8655 | ::std::vector<at::Tensor> tensors_; |
| 8656 | if (at::functionalization::impl::isFunctionalTensor(tensors)) { |
| 8657 | at::functionalization::impl::sync(tensors); |
| 8658 | tensors_ = at::functionalization::impl::from_functional_tensor(tensors); |
| 8659 | } else { |
| 8660 | tensors_ = tensors.vec(); |
| 8661 | } |
| 8662 | |
| 8663 | at::Tensor out_; |
| 8664 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8665 | at::functionalization::impl::sync(out); |
| 8666 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8667 | } else { |
| 8668 | out_ = out; |
| 8669 | } |
| 8670 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8671 | if ((false || at::functionalization::impl::isFunctionalTensor(tensors))) { |
| 8672 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8673 | TORCH_INTERNAL_ASSERT(false, |
| 8674 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8675 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8676 | } else { |
| 8677 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8678 | at::AutoDispatchSkipFunctionalize guard; |
| 8679 | at::Tensor tmp_output = at::_ops::stack_out::call(tensors_, dim, out_); |
| 8680 | return out;; |
| 8681 | } |
| 8682 | } else { |
| 8683 | at::Tensor tmp_output; |
| 8684 | { |
| 8685 | at::AutoDispatchSkipFunctionalize guard; |
| 8686 | tmp_output = at::_ops::stack::call(tensors_, dim); |
| 8687 | } |
| 8688 | at::functionalization::impl::replace_(out, tmp_output); |
| 8689 | at::functionalization::impl::commit_update(out); |
| 8690 | at::functionalization::impl::sync(out); |
| 8691 | return out; |
| 8692 | } |
| 8693 | } |
| 8694 | |
| 8695 | at::Tensor & vstack_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList tensors, at::Tensor & out) { |
| 8696 | if (false) { |
| 8697 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8698 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8699 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8700 | auto tensors_meta = to_meta(tensors); |
| 8701 | auto out_meta = to_meta(out); |
| 8702 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8703 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8704 | at::_ops::vstack_out::call(tensors_meta, out_meta); |
| 8705 | } |
| 8706 | |
| 8707 | ::std::vector<at::Tensor> tensors_; |
| 8708 | if (at::functionalization::impl::isFunctionalTensor(tensors)) { |
| 8709 | at::functionalization::impl::sync(tensors); |
| 8710 | tensors_ = at::functionalization::impl::from_functional_tensor(tensors); |
| 8711 | } else { |
| 8712 | tensors_ = tensors.vec(); |
| 8713 | } |
| 8714 | |
| 8715 | at::Tensor out_; |
| 8716 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8717 | at::functionalization::impl::sync(out); |
| 8718 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8719 | } else { |
| 8720 | out_ = out; |
| 8721 | } |
| 8722 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8723 | if ((false || at::functionalization::impl::isFunctionalTensor(tensors))) { |
| 8724 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8725 | TORCH_INTERNAL_ASSERT(false, |
| 8726 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8727 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8728 | } else { |
| 8729 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8730 | at::AutoDispatchSkipFunctionalize guard; |
| 8731 | at::Tensor tmp_output = at::_ops::vstack_out::call(tensors_, out_); |
| 8732 | return out;; |
| 8733 | } |
| 8734 | } else { |
| 8735 | at::Tensor tmp_output; |
| 8736 | { |
| 8737 | at::AutoDispatchSkipFunctionalize guard; |
| 8738 | tmp_output = at::_ops::vstack::call(tensors_); |
| 8739 | } |
| 8740 | at::functionalization::impl::replace_(out, tmp_output); |
| 8741 | at::functionalization::impl::commit_update(out); |
| 8742 | at::functionalization::impl::sync(out); |
| 8743 | return out; |
| 8744 | } |
| 8745 | } |
| 8746 | |
| 8747 | at::Tensor & nansum_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::OptionalIntArrayRef dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 8748 | if (false) { |
| 8749 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8750 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8751 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8752 | auto self_meta = to_meta(self); |
| 8753 | auto out_meta = to_meta(out); |
| 8754 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8755 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8756 | at::_ops::nansum_out::call(self_meta, dim, keepdim, dtype, out_meta); |
| 8757 | } |
| 8758 | |
| 8759 | at::Tensor self_; |
| 8760 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8761 | at::functionalization::impl::sync(self); |
| 8762 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8763 | } else { |
| 8764 | self_ = self; |
| 8765 | } |
| 8766 | |
| 8767 | at::Tensor out_; |
| 8768 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8769 | at::functionalization::impl::sync(out); |
| 8770 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8771 | } else { |
| 8772 | out_ = out; |
| 8773 | } |
| 8774 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8775 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8776 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8777 | TORCH_INTERNAL_ASSERT(false, |
| 8778 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8779 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8780 | } else { |
| 8781 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8782 | at::AutoDispatchSkipFunctionalize guard; |
| 8783 | at::Tensor tmp_output = at::_ops::nansum_out::call(self_, dim, keepdim, dtype, out_); |
| 8784 | return out;; |
| 8785 | } |
| 8786 | } else { |
| 8787 | at::Tensor tmp_output; |
| 8788 | { |
| 8789 | at::AutoDispatchSkipFunctionalize guard; |
| 8790 | tmp_output = at::_ops::nansum::call(self_, dim, keepdim, dtype); |
| 8791 | } |
| 8792 | at::functionalization::impl::replace_(out, tmp_output); |
| 8793 | at::functionalization::impl::commit_update(out); |
| 8794 | at::functionalization::impl::sync(out); |
| 8795 | return out; |
| 8796 | } |
| 8797 | } |
| 8798 | |
| 8799 | at::Tensor & sqrt_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 8800 | if (false) { |
| 8801 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8802 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8803 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8804 | auto self_meta = to_meta(self); |
| 8805 | auto out_meta = to_meta(out); |
| 8806 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8807 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8808 | at::_ops::sqrt_out::call(self_meta, out_meta); |
| 8809 | } |
| 8810 | |
| 8811 | at::Tensor self_; |
| 8812 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8813 | at::functionalization::impl::sync(self); |
| 8814 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8815 | } else { |
| 8816 | self_ = self; |
| 8817 | } |
| 8818 | |
| 8819 | at::Tensor out_; |
| 8820 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8821 | at::functionalization::impl::sync(out); |
| 8822 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8823 | } else { |
| 8824 | out_ = out; |
| 8825 | } |
| 8826 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8827 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8828 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8829 | TORCH_INTERNAL_ASSERT(false, |
| 8830 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8831 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8832 | } else { |
| 8833 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8834 | at::AutoDispatchSkipFunctionalize guard; |
| 8835 | at::Tensor tmp_output = at::_ops::sqrt_out::call(self_, out_); |
| 8836 | return out;; |
| 8837 | } |
| 8838 | } else { |
| 8839 | at::Tensor tmp_output; |
| 8840 | { |
| 8841 | at::AutoDispatchSkipFunctionalize guard; |
| 8842 | tmp_output = at::_ops::sqrt::call(self_); |
| 8843 | } |
| 8844 | at::functionalization::impl::replace_(out, tmp_output); |
| 8845 | at::functionalization::impl::commit_update(out); |
| 8846 | at::functionalization::impl::sync(out); |
| 8847 | return out; |
| 8848 | } |
| 8849 | } |
| 8850 | |
| 8851 | at::Tensor & sqrt_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 8852 | if (true) { |
| 8853 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8854 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8855 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8856 | auto self_meta = to_meta(self); |
| 8857 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8858 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8859 | at::_ops::sqrt_::call(self_meta); |
| 8860 | } |
| 8861 | |
| 8862 | at::Tensor self_; |
| 8863 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8864 | at::functionalization::impl::sync(self); |
| 8865 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8866 | } else { |
| 8867 | self_ = self; |
| 8868 | } |
| 8869 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 8870 | if ((false)) { |
| 8871 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8872 | TORCH_INTERNAL_ASSERT(false, |
| 8873 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8874 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8875 | } else { |
| 8876 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8877 | at::AutoDispatchSkipFunctionalize guard; |
| 8878 | at::Tensor tmp_output = at::_ops::sqrt_::call(self_); |
| 8879 | return self;; |
| 8880 | } |
| 8881 | } else { |
| 8882 | at::Tensor tmp_output; |
| 8883 | { |
| 8884 | at::AutoDispatchSkipFunctionalize guard; |
| 8885 | tmp_output = at::_ops::sqrt::call(self_); |
| 8886 | } |
| 8887 | at::functionalization::impl::replace_(self, tmp_output); |
| 8888 | at::functionalization::impl::commit_update(self); |
| 8889 | at::functionalization::impl::sync(self); |
| 8890 | return self; |
| 8891 | } |
| 8892 | } |
| 8893 | |
| 8894 | at::Tensor & prod_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 8895 | if (false) { |
| 8896 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8897 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8898 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8899 | auto self_meta = to_meta(self); |
| 8900 | auto out_meta = to_meta(out); |
| 8901 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8902 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8903 | at::_ops::prod_out::call(self_meta, dtype, out_meta); |
| 8904 | } |
| 8905 | |
| 8906 | at::Tensor self_; |
| 8907 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8908 | at::functionalization::impl::sync(self); |
| 8909 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8910 | } else { |
| 8911 | self_ = self; |
| 8912 | } |
| 8913 | |
| 8914 | at::Tensor out_; |
| 8915 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8916 | at::functionalization::impl::sync(out); |
| 8917 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8918 | } else { |
| 8919 | out_ = out; |
| 8920 | } |
| 8921 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8922 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8923 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8924 | TORCH_INTERNAL_ASSERT(false, |
| 8925 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8926 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8927 | } else { |
| 8928 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8929 | at::AutoDispatchSkipFunctionalize guard; |
| 8930 | at::Tensor tmp_output = at::_ops::prod_out::call(self_, dtype, out_); |
| 8931 | return out;; |
| 8932 | } |
| 8933 | } else { |
| 8934 | at::Tensor tmp_output; |
| 8935 | { |
| 8936 | at::AutoDispatchSkipFunctionalize guard; |
| 8937 | tmp_output = at::_ops::prod::call(self_, dtype); |
| 8938 | } |
| 8939 | at::functionalization::impl::replace_(out, tmp_output); |
| 8940 | at::functionalization::impl::commit_update(out); |
| 8941 | at::functionalization::impl::sync(out); |
| 8942 | return out; |
| 8943 | } |
| 8944 | } |
| 8945 | |
| 8946 | at::Tensor & prod_out_int_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 8947 | if (false) { |
| 8948 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 8949 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 8950 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 8951 | auto self_meta = to_meta(self); |
| 8952 | auto out_meta = to_meta(out); |
| 8953 | at::AutoDispatchSkipFunctionalize func_guard; |
| 8954 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 8955 | at::_ops::prod_int_out::call(self_meta, dim, keepdim, dtype, out_meta); |
| 8956 | } |
| 8957 | |
| 8958 | at::Tensor self_; |
| 8959 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 8960 | at::functionalization::impl::sync(self); |
| 8961 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 8962 | } else { |
| 8963 | self_ = self; |
| 8964 | } |
| 8965 | |
| 8966 | at::Tensor out_; |
| 8967 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 8968 | at::functionalization::impl::sync(out); |
| 8969 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 8970 | } else { |
| 8971 | out_ = out; |
| 8972 | } |
| 8973 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 8974 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 8975 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 8976 | TORCH_INTERNAL_ASSERT(false, |
| 8977 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 8978 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 8979 | } else { |
| 8980 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 8981 | at::AutoDispatchSkipFunctionalize guard; |
| 8982 | at::Tensor tmp_output = at::_ops::prod_int_out::call(self_, dim, keepdim, dtype, out_); |
| 8983 | return out;; |
| 8984 | } |
| 8985 | } else { |
| 8986 | at::Tensor tmp_output; |
| 8987 | { |
| 8988 | at::AutoDispatchSkipFunctionalize guard; |
| 8989 | tmp_output = at::_ops::prod_dim_int::call(self_, dim, keepdim, dtype); |
| 8990 | } |
| 8991 | at::functionalization::impl::replace_(out, tmp_output); |
| 8992 | at::functionalization::impl::commit_update(out); |
| 8993 | at::functionalization::impl::sync(out); |
| 8994 | return out; |
| 8995 | } |
| 8996 | } |
| 8997 | |
| 8998 | at::Tensor & prod_out_Dimname_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Dimname dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 8999 | if (false) { |
| 9000 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9001 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9002 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9003 | auto self_meta = to_meta(self); |
| 9004 | auto out_meta = to_meta(out); |
| 9005 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9006 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9007 | at::_ops::prod_Dimname_out::call(self_meta, dim, keepdim, dtype, out_meta); |
| 9008 | } |
| 9009 | |
| 9010 | at::Tensor self_; |
| 9011 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9012 | at::functionalization::impl::sync(self); |
| 9013 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9014 | } else { |
| 9015 | self_ = self; |
| 9016 | } |
| 9017 | |
| 9018 | at::Tensor out_; |
| 9019 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9020 | at::functionalization::impl::sync(out); |
| 9021 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9022 | } else { |
| 9023 | out_ = out; |
| 9024 | } |
| 9025 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9026 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9027 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9028 | TORCH_INTERNAL_ASSERT(false, |
| 9029 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9030 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9031 | } else { |
| 9032 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9033 | at::AutoDispatchSkipFunctionalize guard; |
| 9034 | at::Tensor tmp_output = at::_ops::prod_Dimname_out::call(self_, dim, keepdim, dtype, out_); |
| 9035 | return out;; |
| 9036 | } |
| 9037 | } else { |
| 9038 | at::Tensor tmp_output; |
| 9039 | { |
| 9040 | at::AutoDispatchSkipFunctionalize guard; |
| 9041 | tmp_output = at::_ops::prod_dim_Dimname::call(self_, dim, keepdim, dtype); |
| 9042 | } |
| 9043 | at::functionalization::impl::replace_(out, tmp_output); |
| 9044 | at::functionalization::impl::commit_update(out); |
| 9045 | at::functionalization::impl::sync(out); |
| 9046 | return out; |
| 9047 | } |
| 9048 | } |
| 9049 | |
| 9050 | at::Tensor & threshold_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, const at::Scalar & threshold, at::Tensor & grad_input) { |
| 9051 | if (false) { |
| 9052 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9053 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9054 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9055 | auto grad_output_meta = to_meta(grad_output); |
| 9056 | auto self_meta = to_meta(self); |
| 9057 | auto grad_input_meta = to_meta(grad_input); |
| 9058 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9059 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9060 | at::_ops::threshold_backward_grad_input::call(grad_output_meta, self_meta, threshold, grad_input_meta); |
| 9061 | } |
| 9062 | |
| 9063 | at::Tensor grad_output_; |
| 9064 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 9065 | at::functionalization::impl::sync(grad_output); |
| 9066 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 9067 | } else { |
| 9068 | grad_output_ = grad_output; |
| 9069 | } |
| 9070 | |
| 9071 | at::Tensor self_; |
| 9072 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9073 | at::functionalization::impl::sync(self); |
| 9074 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9075 | } else { |
| 9076 | self_ = self; |
| 9077 | } |
| 9078 | |
| 9079 | at::Tensor grad_input_; |
| 9080 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 9081 | at::functionalization::impl::sync(grad_input); |
| 9082 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 9083 | } else { |
| 9084 | grad_input_ = grad_input; |
| 9085 | } |
| 9086 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 9087 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9088 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9089 | TORCH_INTERNAL_ASSERT(false, |
| 9090 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9091 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9092 | } else { |
| 9093 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9094 | at::AutoDispatchSkipFunctionalize guard; |
| 9095 | at::Tensor tmp_output = at::_ops::threshold_backward_grad_input::call(grad_output_, self_, threshold, grad_input_); |
| 9096 | return grad_input;; |
| 9097 | } |
| 9098 | } else { |
| 9099 | at::Tensor tmp_output; |
| 9100 | { |
| 9101 | at::AutoDispatchSkipFunctionalize guard; |
| 9102 | tmp_output = at::_ops::threshold_backward::call(grad_output_, self_, threshold); |
| 9103 | } |
| 9104 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 9105 | at::functionalization::impl::commit_update(grad_input); |
| 9106 | at::functionalization::impl::sync(grad_input); |
| 9107 | return grad_input; |
| 9108 | } |
| 9109 | } |
| 9110 | |
| 9111 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> _transform_bias_rescale_qkv_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & qkv, const at::Tensor & qkv_bias, int64_t num_heads, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 9112 | if (false) { |
| 9113 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9114 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9115 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9116 | auto qkv_meta = to_meta(qkv); |
| 9117 | auto qkv_bias_meta = to_meta(qkv_bias); |
| 9118 | auto out0_meta = to_meta(out0); |
| 9119 | auto out1_meta = to_meta(out1); |
| 9120 | auto out2_meta = to_meta(out2); |
| 9121 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9122 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9123 | at::_ops::_transform_bias_rescale_qkv_out::call(qkv_meta, qkv_bias_meta, num_heads, out0_meta, out1_meta, out2_meta); |
| 9124 | } |
| 9125 | |
| 9126 | at::Tensor qkv_; |
| 9127 | if (at::functionalization::impl::isFunctionalTensor(qkv)) { |
| 9128 | at::functionalization::impl::sync(qkv); |
| 9129 | qkv_ = at::functionalization::impl::from_functional_tensor(qkv); |
| 9130 | } else { |
| 9131 | qkv_ = qkv; |
| 9132 | } |
| 9133 | |
| 9134 | at::Tensor qkv_bias_; |
| 9135 | if (at::functionalization::impl::isFunctionalTensor(qkv_bias)) { |
| 9136 | at::functionalization::impl::sync(qkv_bias); |
| 9137 | qkv_bias_ = at::functionalization::impl::from_functional_tensor(qkv_bias); |
| 9138 | } else { |
| 9139 | qkv_bias_ = qkv_bias; |
| 9140 | } |
| 9141 | |
| 9142 | at::Tensor out0_; |
| 9143 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 9144 | at::functionalization::impl::sync(out0); |
| 9145 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 9146 | } else { |
| 9147 | out0_ = out0; |
| 9148 | } |
| 9149 | |
| 9150 | at::Tensor out1_; |
| 9151 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 9152 | at::functionalization::impl::sync(out1); |
| 9153 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 9154 | } else { |
| 9155 | out1_ = out1; |
| 9156 | } |
| 9157 | |
| 9158 | at::Tensor out2_; |
| 9159 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 9160 | at::functionalization::impl::sync(out2); |
| 9161 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 9162 | } else { |
| 9163 | out2_ = out2; |
| 9164 | } |
| 9165 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 9166 | if ((false || at::functionalization::impl::isFunctionalTensor(qkv) || at::functionalization::impl::isFunctionalTensor(qkv_bias))) { |
| 9167 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9168 | TORCH_INTERNAL_ASSERT(false, |
| 9169 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9170 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9171 | } else { |
| 9172 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9173 | at::AutoDispatchSkipFunctionalize guard; |
| 9174 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_transform_bias_rescale_qkv_out::call(qkv_, qkv_bias_, num_heads, out0_, out1_, out2_); |
| 9175 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 9176 | } |
| 9177 | } else { |
| 9178 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 9179 | { |
| 9180 | at::AutoDispatchSkipFunctionalize guard; |
| 9181 | tmp_output = at::_ops::_transform_bias_rescale_qkv::call(qkv_, qkv_bias_, num_heads); |
| 9182 | } |
| 9183 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 9184 | at::functionalization::impl::commit_update(out0); |
| 9185 | at::functionalization::impl::sync(out0); |
| 9186 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 9187 | at::functionalization::impl::commit_update(out1); |
| 9188 | at::functionalization::impl::sync(out1); |
| 9189 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 9190 | at::functionalization::impl::commit_update(out2); |
| 9191 | at::functionalization::impl::sync(out2); |
| 9192 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 9193 | } |
| 9194 | } |
| 9195 | |
| 9196 | at::Tensor & _nested_from_padded_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & padded, const at::Tensor & cpu_nested_shape_example, bool fuse_transform_0213, at::Tensor & out) { |
| 9197 | if (false) { |
| 9198 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9199 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9200 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9201 | auto padded_meta = to_meta(padded); |
| 9202 | auto cpu_nested_shape_example_meta = to_meta(cpu_nested_shape_example); |
| 9203 | auto out_meta = to_meta(out); |
| 9204 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9205 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9206 | at::_ops::_nested_from_padded_out::call(padded_meta, cpu_nested_shape_example_meta, fuse_transform_0213, out_meta); |
| 9207 | } |
| 9208 | |
| 9209 | at::Tensor padded_; |
| 9210 | if (at::functionalization::impl::isFunctionalTensor(padded)) { |
| 9211 | at::functionalization::impl::sync(padded); |
| 9212 | padded_ = at::functionalization::impl::from_functional_tensor(padded); |
| 9213 | } else { |
| 9214 | padded_ = padded; |
| 9215 | } |
| 9216 | |
| 9217 | at::Tensor cpu_nested_shape_example_; |
| 9218 | if (at::functionalization::impl::isFunctionalTensor(cpu_nested_shape_example)) { |
| 9219 | at::functionalization::impl::sync(cpu_nested_shape_example); |
| 9220 | cpu_nested_shape_example_ = at::functionalization::impl::from_functional_tensor(cpu_nested_shape_example); |
| 9221 | } else { |
| 9222 | cpu_nested_shape_example_ = cpu_nested_shape_example; |
| 9223 | } |
| 9224 | |
| 9225 | at::Tensor out_; |
| 9226 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9227 | at::functionalization::impl::sync(out); |
| 9228 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9229 | } else { |
| 9230 | out_ = out; |
| 9231 | } |
| 9232 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9233 | if ((false || at::functionalization::impl::isFunctionalTensor(padded) || at::functionalization::impl::isFunctionalTensor(cpu_nested_shape_example))) { |
| 9234 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9235 | TORCH_INTERNAL_ASSERT(false, |
| 9236 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9237 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9238 | } else { |
| 9239 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9240 | at::AutoDispatchSkipFunctionalize guard; |
| 9241 | at::Tensor tmp_output = at::_ops::_nested_from_padded_out::call(padded_, cpu_nested_shape_example_, fuse_transform_0213, out_); |
| 9242 | return out;; |
| 9243 | } |
| 9244 | } else { |
| 9245 | at::Tensor tmp_output; |
| 9246 | { |
| 9247 | at::AutoDispatchSkipFunctionalize guard; |
| 9248 | tmp_output = at::_ops::_nested_from_padded::call(padded_, cpu_nested_shape_example_, fuse_transform_0213); |
| 9249 | } |
| 9250 | at::functionalization::impl::replace_(out, tmp_output); |
| 9251 | at::functionalization::impl::commit_update(out); |
| 9252 | at::functionalization::impl::sync(out); |
| 9253 | return out; |
| 9254 | } |
| 9255 | } |
| 9256 | |
| 9257 | at::Tensor & _nested_tensor_size_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 9258 | if (false) { |
| 9259 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9260 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9261 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9262 | auto self_meta = to_meta(self); |
| 9263 | auto out_meta = to_meta(out); |
| 9264 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9265 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9266 | at::_ops::_nested_tensor_size_out::call(self_meta, out_meta); |
| 9267 | } |
| 9268 | |
| 9269 | at::Tensor self_; |
| 9270 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9271 | at::functionalization::impl::sync(self); |
| 9272 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9273 | } else { |
| 9274 | self_ = self; |
| 9275 | } |
| 9276 | |
| 9277 | at::Tensor out_; |
| 9278 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9279 | at::functionalization::impl::sync(out); |
| 9280 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9281 | } else { |
| 9282 | out_ = out; |
| 9283 | } |
| 9284 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9285 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9286 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9287 | TORCH_INTERNAL_ASSERT(false, |
| 9288 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9289 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9290 | } else { |
| 9291 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9292 | at::AutoDispatchSkipFunctionalize guard; |
| 9293 | at::Tensor tmp_output = at::_ops::_nested_tensor_size_out::call(self_, out_); |
| 9294 | return out;; |
| 9295 | } |
| 9296 | } else { |
| 9297 | at::Tensor tmp_output; |
| 9298 | { |
| 9299 | at::AutoDispatchSkipFunctionalize guard; |
| 9300 | tmp_output = at::_ops::_nested_tensor_size::call(self_); |
| 9301 | } |
| 9302 | at::functionalization::impl::replace_(out, tmp_output); |
| 9303 | at::functionalization::impl::commit_update(out); |
| 9304 | at::functionalization::impl::sync(out); |
| 9305 | return out; |
| 9306 | } |
| 9307 | } |
| 9308 | |
| 9309 | at::Tensor & _nested_view_from_buffer_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & nested_size, const at::Tensor & nested_strides, at::IntArrayRef offsets, at::Tensor & out) { |
| 9310 | if (false) { |
| 9311 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9312 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9313 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9314 | auto self_meta = to_meta(self); |
| 9315 | auto nested_size_meta = to_meta(nested_size); |
| 9316 | auto nested_strides_meta = to_meta(nested_strides); |
| 9317 | auto out_meta = to_meta(out); |
| 9318 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9319 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9320 | at::_ops::_nested_view_from_buffer_copy_out::call(self_meta, nested_size_meta, nested_strides_meta, offsets, out_meta); |
| 9321 | } |
| 9322 | |
| 9323 | at::Tensor self_; |
| 9324 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9325 | at::functionalization::impl::sync(self); |
| 9326 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9327 | } else { |
| 9328 | self_ = self; |
| 9329 | } |
| 9330 | |
| 9331 | at::Tensor nested_size_; |
| 9332 | if (at::functionalization::impl::isFunctionalTensor(nested_size)) { |
| 9333 | at::functionalization::impl::sync(nested_size); |
| 9334 | nested_size_ = at::functionalization::impl::from_functional_tensor(nested_size); |
| 9335 | } else { |
| 9336 | nested_size_ = nested_size; |
| 9337 | } |
| 9338 | |
| 9339 | at::Tensor nested_strides_; |
| 9340 | if (at::functionalization::impl::isFunctionalTensor(nested_strides)) { |
| 9341 | at::functionalization::impl::sync(nested_strides); |
| 9342 | nested_strides_ = at::functionalization::impl::from_functional_tensor(nested_strides); |
| 9343 | } else { |
| 9344 | nested_strides_ = nested_strides; |
| 9345 | } |
| 9346 | |
| 9347 | at::Tensor out_; |
| 9348 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9349 | at::functionalization::impl::sync(out); |
| 9350 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9351 | } else { |
| 9352 | out_ = out; |
| 9353 | } |
| 9354 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9355 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(nested_size) || at::functionalization::impl::isFunctionalTensor(nested_strides))) { |
| 9356 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9357 | TORCH_INTERNAL_ASSERT(false, |
| 9358 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9359 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9360 | } else { |
| 9361 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9362 | at::AutoDispatchSkipFunctionalize guard; |
| 9363 | at::Tensor tmp_output = at::_ops::_nested_view_from_buffer_copy_out::call(self_, nested_size_, nested_strides_, offsets, out_); |
| 9364 | return out;; |
| 9365 | } |
| 9366 | } else { |
| 9367 | at::Tensor tmp_output; |
| 9368 | { |
| 9369 | at::AutoDispatchSkipFunctionalize guard; |
| 9370 | tmp_output = at::_ops::_nested_view_from_buffer_copy::call(self_, nested_size_, nested_strides_, offsets); |
| 9371 | } |
| 9372 | at::functionalization::impl::replace_(out, tmp_output); |
| 9373 | at::functionalization::impl::commit_update(out); |
| 9374 | at::functionalization::impl::sync(out); |
| 9375 | return out; |
| 9376 | } |
| 9377 | } |
| 9378 | |
| 9379 | at::Tensor & trunc_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 9380 | if (false) { |
| 9381 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9382 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9383 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9384 | auto self_meta = to_meta(self); |
| 9385 | auto out_meta = to_meta(out); |
| 9386 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9387 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9388 | at::_ops::trunc_out::call(self_meta, out_meta); |
| 9389 | } |
| 9390 | |
| 9391 | at::Tensor self_; |
| 9392 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9393 | at::functionalization::impl::sync(self); |
| 9394 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9395 | } else { |
| 9396 | self_ = self; |
| 9397 | } |
| 9398 | |
| 9399 | at::Tensor out_; |
| 9400 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9401 | at::functionalization::impl::sync(out); |
| 9402 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9403 | } else { |
| 9404 | out_ = out; |
| 9405 | } |
| 9406 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9407 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9408 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9409 | TORCH_INTERNAL_ASSERT(false, |
| 9410 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9411 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9412 | } else { |
| 9413 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9414 | at::AutoDispatchSkipFunctionalize guard; |
| 9415 | at::Tensor tmp_output = at::_ops::trunc_out::call(self_, out_); |
| 9416 | return out;; |
| 9417 | } |
| 9418 | } else { |
| 9419 | at::Tensor tmp_output; |
| 9420 | { |
| 9421 | at::AutoDispatchSkipFunctionalize guard; |
| 9422 | tmp_output = at::_ops::trunc::call(self_); |
| 9423 | } |
| 9424 | at::functionalization::impl::replace_(out, tmp_output); |
| 9425 | at::functionalization::impl::commit_update(out); |
| 9426 | at::functionalization::impl::sync(out); |
| 9427 | return out; |
| 9428 | } |
| 9429 | } |
| 9430 | |
| 9431 | at::Tensor & trunc_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 9432 | if (true) { |
| 9433 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9434 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9435 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9436 | auto self_meta = to_meta(self); |
| 9437 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9438 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9439 | at::_ops::trunc_::call(self_meta); |
| 9440 | } |
| 9441 | |
| 9442 | at::Tensor self_; |
| 9443 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9444 | at::functionalization::impl::sync(self); |
| 9445 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9446 | } else { |
| 9447 | self_ = self; |
| 9448 | } |
| 9449 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 9450 | if ((false)) { |
| 9451 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9452 | TORCH_INTERNAL_ASSERT(false, |
| 9453 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9454 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9455 | } else { |
| 9456 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9457 | at::AutoDispatchSkipFunctionalize guard; |
| 9458 | at::Tensor tmp_output = at::_ops::trunc_::call(self_); |
| 9459 | return self;; |
| 9460 | } |
| 9461 | } else { |
| 9462 | at::Tensor tmp_output; |
| 9463 | { |
| 9464 | at::AutoDispatchSkipFunctionalize guard; |
| 9465 | tmp_output = at::_ops::trunc::call(self_); |
| 9466 | } |
| 9467 | at::functionalization::impl::replace_(self, tmp_output); |
| 9468 | at::functionalization::impl::commit_update(self); |
| 9469 | at::functionalization::impl::sync(self); |
| 9470 | return self; |
| 9471 | } |
| 9472 | } |
| 9473 | |
| 9474 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> unique_dim_consecutive_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dim, bool return_inverse, bool return_counts, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 9475 | if (false) { |
| 9476 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9477 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9478 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9479 | auto self_meta = to_meta(self); |
| 9480 | auto out0_meta = to_meta(out0); |
| 9481 | auto out1_meta = to_meta(out1); |
| 9482 | auto out2_meta = to_meta(out2); |
| 9483 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9484 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9485 | at::_ops::unique_dim_consecutive_out::call(self_meta, dim, return_inverse, return_counts, out0_meta, out1_meta, out2_meta); |
| 9486 | } |
| 9487 | |
| 9488 | at::Tensor self_; |
| 9489 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9490 | at::functionalization::impl::sync(self); |
| 9491 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9492 | } else { |
| 9493 | self_ = self; |
| 9494 | } |
| 9495 | |
| 9496 | at::Tensor out0_; |
| 9497 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 9498 | at::functionalization::impl::sync(out0); |
| 9499 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 9500 | } else { |
| 9501 | out0_ = out0; |
| 9502 | } |
| 9503 | |
| 9504 | at::Tensor out1_; |
| 9505 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 9506 | at::functionalization::impl::sync(out1); |
| 9507 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 9508 | } else { |
| 9509 | out1_ = out1; |
| 9510 | } |
| 9511 | |
| 9512 | at::Tensor out2_; |
| 9513 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 9514 | at::functionalization::impl::sync(out2); |
| 9515 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 9516 | } else { |
| 9517 | out2_ = out2; |
| 9518 | } |
| 9519 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 9520 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9521 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9522 | TORCH_INTERNAL_ASSERT(false, |
| 9523 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9524 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9525 | } else { |
| 9526 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9527 | at::AutoDispatchSkipFunctionalize guard; |
| 9528 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::unique_dim_consecutive_out::call(self_, dim, return_inverse, return_counts, out0_, out1_, out2_); |
| 9529 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 9530 | } |
| 9531 | } else { |
| 9532 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 9533 | { |
| 9534 | at::AutoDispatchSkipFunctionalize guard; |
| 9535 | tmp_output = at::_ops::unique_dim_consecutive::call(self_, dim, return_inverse, return_counts); |
| 9536 | } |
| 9537 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 9538 | at::functionalization::impl::commit_update(out0); |
| 9539 | at::functionalization::impl::sync(out0); |
| 9540 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 9541 | at::functionalization::impl::commit_update(out1); |
| 9542 | at::functionalization::impl::sync(out1); |
| 9543 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 9544 | at::functionalization::impl::commit_update(out2); |
| 9545 | at::functionalization::impl::sync(out2); |
| 9546 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 9547 | } |
| 9548 | } |
| 9549 | |
| 9550 | at::Tensor & where_out_self_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & condition, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 9551 | if (false) { |
| 9552 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9553 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9554 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9555 | auto condition_meta = to_meta(condition); |
| 9556 | auto self_meta = to_meta(self); |
| 9557 | auto other_meta = to_meta(other); |
| 9558 | auto out_meta = to_meta(out); |
| 9559 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9560 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9561 | at::_ops::where_self_out::call(condition_meta, self_meta, other_meta, out_meta); |
| 9562 | } |
| 9563 | |
| 9564 | at::Tensor condition_; |
| 9565 | if (at::functionalization::impl::isFunctionalTensor(condition)) { |
| 9566 | at::functionalization::impl::sync(condition); |
| 9567 | condition_ = at::functionalization::impl::from_functional_tensor(condition); |
| 9568 | } else { |
| 9569 | condition_ = condition; |
| 9570 | } |
| 9571 | |
| 9572 | at::Tensor self_; |
| 9573 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9574 | at::functionalization::impl::sync(self); |
| 9575 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9576 | } else { |
| 9577 | self_ = self; |
| 9578 | } |
| 9579 | |
| 9580 | at::Tensor other_; |
| 9581 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 9582 | at::functionalization::impl::sync(other); |
| 9583 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 9584 | } else { |
| 9585 | other_ = other; |
| 9586 | } |
| 9587 | |
| 9588 | at::Tensor out_; |
| 9589 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9590 | at::functionalization::impl::sync(out); |
| 9591 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9592 | } else { |
| 9593 | out_ = out; |
| 9594 | } |
| 9595 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9596 | if ((false || at::functionalization::impl::isFunctionalTensor(condition) || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 9597 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9598 | TORCH_INTERNAL_ASSERT(false, |
| 9599 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9600 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9601 | } else { |
| 9602 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9603 | at::AutoDispatchSkipFunctionalize guard; |
| 9604 | at::Tensor tmp_output = at::_ops::where_self_out::call(condition_, self_, other_, out_); |
| 9605 | return out;; |
| 9606 | } |
| 9607 | } else { |
| 9608 | at::Tensor tmp_output; |
| 9609 | { |
| 9610 | at::AutoDispatchSkipFunctionalize guard; |
| 9611 | tmp_output = at::_ops::where_self::call(condition_, self_, other_); |
| 9612 | } |
| 9613 | at::functionalization::impl::replace_(out, tmp_output); |
| 9614 | at::functionalization::impl::commit_update(out); |
| 9615 | at::functionalization::impl::sync(out); |
| 9616 | return out; |
| 9617 | } |
| 9618 | } |
| 9619 | |
| 9620 | ::std::tuple<at::Tensor &,at::Tensor &> _weight_norm_interface_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_w, const at::Tensor & saved_v, const at::Tensor & saved_g, const at::Tensor & saved_norms, int64_t dim, at::Tensor & out0, at::Tensor & out1) { |
| 9621 | if (false) { |
| 9622 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9623 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9624 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9625 | auto grad_w_meta = to_meta(grad_w); |
| 9626 | auto saved_v_meta = to_meta(saved_v); |
| 9627 | auto saved_g_meta = to_meta(saved_g); |
| 9628 | auto saved_norms_meta = to_meta(saved_norms); |
| 9629 | auto out0_meta = to_meta(out0); |
| 9630 | auto out1_meta = to_meta(out1); |
| 9631 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9632 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9633 | at::_ops::_weight_norm_interface_backward_out::call(grad_w_meta, saved_v_meta, saved_g_meta, saved_norms_meta, dim, out0_meta, out1_meta); |
| 9634 | } |
| 9635 | |
| 9636 | at::Tensor grad_w_; |
| 9637 | if (at::functionalization::impl::isFunctionalTensor(grad_w)) { |
| 9638 | at::functionalization::impl::sync(grad_w); |
| 9639 | grad_w_ = at::functionalization::impl::from_functional_tensor(grad_w); |
| 9640 | } else { |
| 9641 | grad_w_ = grad_w; |
| 9642 | } |
| 9643 | |
| 9644 | at::Tensor saved_v_; |
| 9645 | if (at::functionalization::impl::isFunctionalTensor(saved_v)) { |
| 9646 | at::functionalization::impl::sync(saved_v); |
| 9647 | saved_v_ = at::functionalization::impl::from_functional_tensor(saved_v); |
| 9648 | } else { |
| 9649 | saved_v_ = saved_v; |
| 9650 | } |
| 9651 | |
| 9652 | at::Tensor saved_g_; |
| 9653 | if (at::functionalization::impl::isFunctionalTensor(saved_g)) { |
| 9654 | at::functionalization::impl::sync(saved_g); |
| 9655 | saved_g_ = at::functionalization::impl::from_functional_tensor(saved_g); |
| 9656 | } else { |
| 9657 | saved_g_ = saved_g; |
| 9658 | } |
| 9659 | |
| 9660 | at::Tensor saved_norms_; |
| 9661 | if (at::functionalization::impl::isFunctionalTensor(saved_norms)) { |
| 9662 | at::functionalization::impl::sync(saved_norms); |
| 9663 | saved_norms_ = at::functionalization::impl::from_functional_tensor(saved_norms); |
| 9664 | } else { |
| 9665 | saved_norms_ = saved_norms; |
| 9666 | } |
| 9667 | |
| 9668 | at::Tensor out0_; |
| 9669 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 9670 | at::functionalization::impl::sync(out0); |
| 9671 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 9672 | } else { |
| 9673 | out0_ = out0; |
| 9674 | } |
| 9675 | |
| 9676 | at::Tensor out1_; |
| 9677 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 9678 | at::functionalization::impl::sync(out1); |
| 9679 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 9680 | } else { |
| 9681 | out1_ = out1; |
| 9682 | } |
| 9683 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1))) { |
| 9684 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_w) || at::functionalization::impl::isFunctionalTensor(saved_v) || at::functionalization::impl::isFunctionalTensor(saved_g) || at::functionalization::impl::isFunctionalTensor(saved_norms))) { |
| 9685 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9686 | TORCH_INTERNAL_ASSERT(false, |
| 9687 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9688 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9689 | } else { |
| 9690 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9691 | at::AutoDispatchSkipFunctionalize guard; |
| 9692 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::_weight_norm_interface_backward_out::call(grad_w_, saved_v_, saved_g_, saved_norms_, dim, out0_, out1_); |
| 9693 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1);; |
| 9694 | } |
| 9695 | } else { |
| 9696 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 9697 | { |
| 9698 | at::AutoDispatchSkipFunctionalize guard; |
| 9699 | tmp_output = at::_ops::_weight_norm_interface_backward::call(grad_w_, saved_v_, saved_g_, saved_norms_, dim); |
| 9700 | } |
| 9701 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 9702 | at::functionalization::impl::commit_update(out0); |
| 9703 | at::functionalization::impl::sync(out0); |
| 9704 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 9705 | at::functionalization::impl::commit_update(out1); |
| 9706 | at::functionalization::impl::sync(out1); |
| 9707 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1); |
| 9708 | } |
| 9709 | } |
| 9710 | |
| 9711 | at::Tensor & _sample_dirichlet_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<at::Generator> generator, at::Tensor & out) { |
| 9712 | if (false) { |
| 9713 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9714 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9715 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9716 | auto self_meta = to_meta(self); |
| 9717 | auto out_meta = to_meta(out); |
| 9718 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9719 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9720 | at::_ops::_sample_dirichlet_out::call(self_meta, generator, out_meta); |
| 9721 | } |
| 9722 | |
| 9723 | at::Tensor self_; |
| 9724 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9725 | at::functionalization::impl::sync(self); |
| 9726 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9727 | } else { |
| 9728 | self_ = self; |
| 9729 | } |
| 9730 | |
| 9731 | at::Tensor out_; |
| 9732 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9733 | at::functionalization::impl::sync(out); |
| 9734 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9735 | } else { |
| 9736 | out_ = out; |
| 9737 | } |
| 9738 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9739 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9740 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9741 | TORCH_INTERNAL_ASSERT(false, |
| 9742 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9743 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9744 | } else { |
| 9745 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9746 | at::AutoDispatchSkipFunctionalize guard; |
| 9747 | at::Tensor tmp_output = at::_ops::_sample_dirichlet_out::call(self_, generator, out_); |
| 9748 | return out;; |
| 9749 | } |
| 9750 | } else { |
| 9751 | at::Tensor tmp_output; |
| 9752 | { |
| 9753 | at::AutoDispatchSkipFunctionalize guard; |
| 9754 | tmp_output = at::_ops::_sample_dirichlet::call(self_, generator); |
| 9755 | } |
| 9756 | at::functionalization::impl::replace_(out, tmp_output); |
| 9757 | at::functionalization::impl::commit_update(out); |
| 9758 | at::functionalization::impl::sync(out); |
| 9759 | return out; |
| 9760 | } |
| 9761 | } |
| 9762 | |
| 9763 | at::Tensor & binomial_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & count, const at::Tensor & prob, c10::optional<at::Generator> generator, at::Tensor & out) { |
| 9764 | if (false) { |
| 9765 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9766 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9767 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9768 | auto count_meta = to_meta(count); |
| 9769 | auto prob_meta = to_meta(prob); |
| 9770 | auto out_meta = to_meta(out); |
| 9771 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9772 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9773 | at::_ops::binomial_out::call(count_meta, prob_meta, generator, out_meta); |
| 9774 | } |
| 9775 | |
| 9776 | at::Tensor count_; |
| 9777 | if (at::functionalization::impl::isFunctionalTensor(count)) { |
| 9778 | at::functionalization::impl::sync(count); |
| 9779 | count_ = at::functionalization::impl::from_functional_tensor(count); |
| 9780 | } else { |
| 9781 | count_ = count; |
| 9782 | } |
| 9783 | |
| 9784 | at::Tensor prob_; |
| 9785 | if (at::functionalization::impl::isFunctionalTensor(prob)) { |
| 9786 | at::functionalization::impl::sync(prob); |
| 9787 | prob_ = at::functionalization::impl::from_functional_tensor(prob); |
| 9788 | } else { |
| 9789 | prob_ = prob; |
| 9790 | } |
| 9791 | |
| 9792 | at::Tensor out_; |
| 9793 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9794 | at::functionalization::impl::sync(out); |
| 9795 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9796 | } else { |
| 9797 | out_ = out; |
| 9798 | } |
| 9799 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9800 | if ((false || at::functionalization::impl::isFunctionalTensor(count) || at::functionalization::impl::isFunctionalTensor(prob))) { |
| 9801 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9802 | TORCH_INTERNAL_ASSERT(false, |
| 9803 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9804 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9805 | } else { |
| 9806 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9807 | at::AutoDispatchSkipFunctionalize guard; |
| 9808 | at::Tensor tmp_output = at::_ops::binomial_out::call(count_, prob_, generator, out_); |
| 9809 | return out;; |
| 9810 | } |
| 9811 | } else { |
| 9812 | at::Tensor tmp_output; |
| 9813 | { |
| 9814 | at::AutoDispatchSkipFunctionalize guard; |
| 9815 | tmp_output = at::_ops::binomial::call(count_, prob_, generator); |
| 9816 | } |
| 9817 | at::functionalization::impl::replace_(out, tmp_output); |
| 9818 | at::functionalization::impl::commit_update(out); |
| 9819 | at::functionalization::impl::sync(out); |
| 9820 | return out; |
| 9821 | } |
| 9822 | } |
| 9823 | |
| 9824 | at::Tensor & native_norm_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & p, at::Tensor & out) { |
| 9825 | if (false) { |
| 9826 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9827 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9828 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9829 | auto self_meta = to_meta(self); |
| 9830 | auto out_meta = to_meta(out); |
| 9831 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9832 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9833 | at::_ops::native_norm_out::call(self_meta, p, out_meta); |
| 9834 | } |
| 9835 | |
| 9836 | at::Tensor self_; |
| 9837 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9838 | at::functionalization::impl::sync(self); |
| 9839 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9840 | } else { |
| 9841 | self_ = self; |
| 9842 | } |
| 9843 | |
| 9844 | at::Tensor out_; |
| 9845 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9846 | at::functionalization::impl::sync(out); |
| 9847 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9848 | } else { |
| 9849 | out_ = out; |
| 9850 | } |
| 9851 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9852 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9853 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9854 | TORCH_INTERNAL_ASSERT(false, |
| 9855 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9856 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9857 | } else { |
| 9858 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9859 | at::AutoDispatchSkipFunctionalize guard; |
| 9860 | at::Tensor tmp_output = at::_ops::native_norm_out::call(self_, p, out_); |
| 9861 | return out;; |
| 9862 | } |
| 9863 | } else { |
| 9864 | at::Tensor tmp_output; |
| 9865 | { |
| 9866 | at::AutoDispatchSkipFunctionalize guard; |
| 9867 | tmp_output = at::_ops::native_norm::call(self_, p); |
| 9868 | } |
| 9869 | at::functionalization::impl::replace_(out, tmp_output); |
| 9870 | at::functionalization::impl::commit_update(out); |
| 9871 | at::functionalization::impl::sync(out); |
| 9872 | return out; |
| 9873 | } |
| 9874 | } |
| 9875 | |
| 9876 | at::Tensor & native_norm_out_ScalarOpt_dim_dtype_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const c10::optional<at::Scalar> & p, at::IntArrayRef dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 9877 | if (false) { |
| 9878 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9879 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9880 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9881 | auto self_meta = to_meta(self); |
| 9882 | auto out_meta = to_meta(out); |
| 9883 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9884 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9885 | at::_ops::native_norm_ScalarOpt_dim_dtype_out::call(self_meta, p, dim, keepdim, dtype, out_meta); |
| 9886 | } |
| 9887 | |
| 9888 | at::Tensor self_; |
| 9889 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9890 | at::functionalization::impl::sync(self); |
| 9891 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9892 | } else { |
| 9893 | self_ = self; |
| 9894 | } |
| 9895 | |
| 9896 | at::Tensor out_; |
| 9897 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9898 | at::functionalization::impl::sync(out); |
| 9899 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9900 | } else { |
| 9901 | out_ = out; |
| 9902 | } |
| 9903 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9904 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9905 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9906 | TORCH_INTERNAL_ASSERT(false, |
| 9907 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9908 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9909 | } else { |
| 9910 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9911 | at::AutoDispatchSkipFunctionalize guard; |
| 9912 | at::Tensor tmp_output = at::_ops::native_norm_ScalarOpt_dim_dtype_out::call(self_, p, dim, keepdim, dtype, out_); |
| 9913 | return out;; |
| 9914 | } |
| 9915 | } else { |
| 9916 | at::Tensor tmp_output; |
| 9917 | { |
| 9918 | at::AutoDispatchSkipFunctionalize guard; |
| 9919 | tmp_output = at::_ops::native_norm_ScalarOpt_dim_dtype::call(self_, p, dim, keepdim, dtype); |
| 9920 | } |
| 9921 | at::functionalization::impl::replace_(out, tmp_output); |
| 9922 | at::functionalization::impl::commit_update(out); |
| 9923 | at::functionalization::impl::sync(out); |
| 9924 | return out; |
| 9925 | } |
| 9926 | } |
| 9927 | |
| 9928 | at::Tensor & _sparse_sum_out_dim_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef dim, at::Tensor & out) { |
| 9929 | if (false) { |
| 9930 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9931 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9932 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9933 | auto self_meta = to_meta(self); |
| 9934 | auto out_meta = to_meta(out); |
| 9935 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9936 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9937 | at::_ops::_sparse_sum_dim_out::call(self_meta, dim, out_meta); |
| 9938 | } |
| 9939 | |
| 9940 | at::Tensor self_; |
| 9941 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 9942 | at::functionalization::impl::sync(self); |
| 9943 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 9944 | } else { |
| 9945 | self_ = self; |
| 9946 | } |
| 9947 | |
| 9948 | at::Tensor out_; |
| 9949 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 9950 | at::functionalization::impl::sync(out); |
| 9951 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 9952 | } else { |
| 9953 | out_ = out; |
| 9954 | } |
| 9955 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 9956 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 9957 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 9958 | TORCH_INTERNAL_ASSERT(false, |
| 9959 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 9960 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 9961 | } else { |
| 9962 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 9963 | at::AutoDispatchSkipFunctionalize guard; |
| 9964 | at::Tensor tmp_output = at::_ops::_sparse_sum_dim_out::call(self_, dim, out_); |
| 9965 | return out;; |
| 9966 | } |
| 9967 | } else { |
| 9968 | at::Tensor tmp_output; |
| 9969 | { |
| 9970 | at::AutoDispatchSkipFunctionalize guard; |
| 9971 | tmp_output = at::_ops::_sparse_sum_dim::call(self_, dim); |
| 9972 | } |
| 9973 | at::functionalization::impl::replace_(out, tmp_output); |
| 9974 | at::functionalization::impl::commit_update(out); |
| 9975 | at::functionalization::impl::sync(out); |
| 9976 | return out; |
| 9977 | } |
| 9978 | } |
| 9979 | |
| 9980 | at::Tensor & _sparse_sum_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad, const at::Tensor & self, at::IntArrayRef dim, at::Tensor & out) { |
| 9981 | if (false) { |
| 9982 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 9983 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 9984 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 9985 | auto grad_meta = to_meta(grad); |
| 9986 | auto self_meta = to_meta(self); |
| 9987 | auto out_meta = to_meta(out); |
| 9988 | at::AutoDispatchSkipFunctionalize func_guard; |
| 9989 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 9990 | at::_ops::_sparse_sum_backward_out::call(grad_meta, self_meta, dim, out_meta); |
| 9991 | } |
| 9992 | |
| 9993 | at::Tensor grad_; |
| 9994 | if (at::functionalization::impl::isFunctionalTensor(grad)) { |
| 9995 | at::functionalization::impl::sync(grad); |
| 9996 | grad_ = at::functionalization::impl::from_functional_tensor(grad); |
| 9997 | } else { |
| 9998 | grad_ = grad; |
| 9999 | } |
| 10000 | |
| 10001 | at::Tensor self_; |
| 10002 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10003 | at::functionalization::impl::sync(self); |
| 10004 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10005 | } else { |
| 10006 | self_ = self; |
| 10007 | } |
| 10008 | |
| 10009 | at::Tensor out_; |
| 10010 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10011 | at::functionalization::impl::sync(out); |
| 10012 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10013 | } else { |
| 10014 | out_ = out; |
| 10015 | } |
| 10016 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10017 | if ((false || at::functionalization::impl::isFunctionalTensor(grad) || at::functionalization::impl::isFunctionalTensor(self))) { |
| 10018 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10019 | TORCH_INTERNAL_ASSERT(false, |
| 10020 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10021 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10022 | } else { |
| 10023 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10024 | at::AutoDispatchSkipFunctionalize guard; |
| 10025 | at::Tensor tmp_output = at::_ops::_sparse_sum_backward_out::call(grad_, self_, dim, out_); |
| 10026 | return out;; |
| 10027 | } |
| 10028 | } else { |
| 10029 | at::Tensor tmp_output; |
| 10030 | { |
| 10031 | at::AutoDispatchSkipFunctionalize guard; |
| 10032 | tmp_output = at::_ops::_sparse_sum_backward::call(grad_, self_, dim); |
| 10033 | } |
| 10034 | at::functionalization::impl::replace_(out, tmp_output); |
| 10035 | at::functionalization::impl::commit_update(out); |
| 10036 | at::functionalization::impl::sync(out); |
| 10037 | return out; |
| 10038 | } |
| 10039 | } |
| 10040 | |
| 10041 | at::Tensor & _sparse_softmax_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dim, bool half_to_float, at::Tensor & out) { |
| 10042 | if (false) { |
| 10043 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10044 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10045 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10046 | auto self_meta = to_meta(self); |
| 10047 | auto out_meta = to_meta(out); |
| 10048 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10049 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10050 | at::_ops::_sparse_softmax_out::call(self_meta, dim, half_to_float, out_meta); |
| 10051 | } |
| 10052 | |
| 10053 | at::Tensor self_; |
| 10054 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10055 | at::functionalization::impl::sync(self); |
| 10056 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10057 | } else { |
| 10058 | self_ = self; |
| 10059 | } |
| 10060 | |
| 10061 | at::Tensor out_; |
| 10062 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10063 | at::functionalization::impl::sync(out); |
| 10064 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10065 | } else { |
| 10066 | out_ = out; |
| 10067 | } |
| 10068 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10069 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 10070 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10071 | TORCH_INTERNAL_ASSERT(false, |
| 10072 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10073 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10074 | } else { |
| 10075 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10076 | at::AutoDispatchSkipFunctionalize guard; |
| 10077 | at::Tensor tmp_output = at::_ops::_sparse_softmax_out::call(self_, dim, half_to_float, out_); |
| 10078 | return out;; |
| 10079 | } |
| 10080 | } else { |
| 10081 | at::Tensor tmp_output; |
| 10082 | { |
| 10083 | at::AutoDispatchSkipFunctionalize guard; |
| 10084 | tmp_output = at::_ops::_sparse_softmax::call(self_, dim, half_to_float); |
| 10085 | } |
| 10086 | at::functionalization::impl::replace_(out, tmp_output); |
| 10087 | at::functionalization::impl::commit_update(out); |
| 10088 | at::functionalization::impl::sync(out); |
| 10089 | return out; |
| 10090 | } |
| 10091 | } |
| 10092 | |
| 10093 | at::Tensor & clone_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<at::MemoryFormat> memory_format, at::Tensor & out) { |
| 10094 | if (false) { |
| 10095 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10096 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10097 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10098 | auto self_meta = to_meta(self); |
| 10099 | auto out_meta = to_meta(out); |
| 10100 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10101 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10102 | at::_ops::clone_out::call(self_meta, memory_format, out_meta); |
| 10103 | } |
| 10104 | |
| 10105 | at::Tensor self_; |
| 10106 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10107 | at::functionalization::impl::sync(self); |
| 10108 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10109 | } else { |
| 10110 | self_ = self; |
| 10111 | } |
| 10112 | |
| 10113 | at::Tensor out_; |
| 10114 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10115 | at::functionalization::impl::sync(out); |
| 10116 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10117 | } else { |
| 10118 | out_ = out; |
| 10119 | } |
| 10120 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10121 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 10122 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10123 | TORCH_INTERNAL_ASSERT(false, |
| 10124 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10125 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10126 | } else { |
| 10127 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10128 | at::AutoDispatchSkipFunctionalize guard; |
| 10129 | at::Tensor tmp_output = at::_ops::clone_out::call(self_, memory_format, out_); |
| 10130 | return out;; |
| 10131 | } |
| 10132 | } else { |
| 10133 | at::Tensor tmp_output; |
| 10134 | { |
| 10135 | at::AutoDispatchSkipFunctionalize guard; |
| 10136 | tmp_output = at::_ops::clone::call(self_, memory_format); |
| 10137 | } |
| 10138 | at::functionalization::impl::replace_(out, tmp_output); |
| 10139 | at::functionalization::impl::commit_update(out); |
| 10140 | at::functionalization::impl::sync(out); |
| 10141 | return out; |
| 10142 | } |
| 10143 | } |
| 10144 | |
| 10145 | const at::Tensor & resize_as_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & the_template, c10::optional<at::MemoryFormat> memory_format, const at::Tensor & out) { |
| 10146 | if (false) { |
| 10147 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10148 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10149 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10150 | auto self_meta = to_meta(self); |
| 10151 | auto the_template_meta = to_meta(the_template); |
| 10152 | auto out_meta = to_meta(out); |
| 10153 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10154 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10155 | at::_ops::resize_as_out::call(self_meta, the_template_meta, memory_format, out_meta); |
| 10156 | } |
| 10157 | |
| 10158 | at::Tensor self_; |
| 10159 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10160 | at::functionalization::impl::sync(self); |
| 10161 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10162 | } else { |
| 10163 | self_ = self; |
| 10164 | } |
| 10165 | |
| 10166 | at::Tensor the_template_; |
| 10167 | if (at::functionalization::impl::isFunctionalTensor(the_template)) { |
| 10168 | at::functionalization::impl::sync(the_template); |
| 10169 | the_template_ = at::functionalization::impl::from_functional_tensor(the_template); |
| 10170 | } else { |
| 10171 | the_template_ = the_template; |
| 10172 | } |
| 10173 | |
| 10174 | at::Tensor out_; |
| 10175 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10176 | at::functionalization::impl::sync(out); |
| 10177 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10178 | } else { |
| 10179 | out_ = out; |
| 10180 | } |
| 10181 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10182 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(the_template))) { |
| 10183 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10184 | TORCH_INTERNAL_ASSERT(false, |
| 10185 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10186 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10187 | } else { |
| 10188 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10189 | at::AutoDispatchSkipFunctionalize guard; |
| 10190 | at::Tensor tmp_output = at::_ops::resize_as_out::call(self_, the_template_, memory_format, out_); |
| 10191 | return out;; |
| 10192 | } |
| 10193 | } else { |
| 10194 | at::Tensor tmp_output; |
| 10195 | { |
| 10196 | at::AutoDispatchSkipFunctionalize guard; |
| 10197 | tmp_output = at::_ops::resize_as::call(self_, the_template_, memory_format); |
| 10198 | } |
| 10199 | at::functionalization::impl::replace_(out, tmp_output); |
| 10200 | at::functionalization::impl::commit_update(out); |
| 10201 | at::functionalization::impl::sync(out); |
| 10202 | return out; |
| 10203 | } |
| 10204 | } |
| 10205 | |
| 10206 | const at::Tensor & resize_as_(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & the_template, c10::optional<at::MemoryFormat> memory_format) { |
| 10207 | if (true) { |
| 10208 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10209 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10210 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10211 | auto self_meta = to_meta(self); |
| 10212 | auto the_template_meta = to_meta(the_template); |
| 10213 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10214 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10215 | at::_ops::resize_as_::call(self_meta, the_template_meta, memory_format); |
| 10216 | } |
| 10217 | |
| 10218 | at::Tensor self_; |
| 10219 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10220 | at::functionalization::impl::sync(self); |
| 10221 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10222 | } else { |
| 10223 | self_ = self; |
| 10224 | } |
| 10225 | |
| 10226 | at::Tensor the_template_; |
| 10227 | if (at::functionalization::impl::isFunctionalTensor(the_template)) { |
| 10228 | at::functionalization::impl::sync(the_template); |
| 10229 | the_template_ = at::functionalization::impl::from_functional_tensor(the_template); |
| 10230 | } else { |
| 10231 | the_template_ = the_template; |
| 10232 | } |
| 10233 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 10234 | if ((false || at::functionalization::impl::isFunctionalTensor(the_template))) { |
| 10235 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10236 | TORCH_INTERNAL_ASSERT(false, |
| 10237 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10238 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10239 | } else { |
| 10240 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10241 | at::AutoDispatchSkipFunctionalize guard; |
| 10242 | at::Tensor tmp_output = at::_ops::resize_as_::call(self_, the_template_, memory_format); |
| 10243 | return self;; |
| 10244 | } |
| 10245 | } else { |
| 10246 | at::Tensor tmp_output; |
| 10247 | { |
| 10248 | at::AutoDispatchSkipFunctionalize guard; |
| 10249 | tmp_output = at::_ops::resize_as::call(self_, the_template_, memory_format); |
| 10250 | } |
| 10251 | at::functionalization::impl::replace_(self, tmp_output); |
| 10252 | at::functionalization::impl::commit_update(self); |
| 10253 | at::functionalization::impl::sync(self); |
| 10254 | return self; |
| 10255 | } |
| 10256 | } |
| 10257 | |
| 10258 | at::Tensor & zero_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 10259 | if (false) { |
| 10260 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10261 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10262 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10263 | auto self_meta = to_meta(self); |
| 10264 | auto out_meta = to_meta(out); |
| 10265 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10266 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10267 | at::_ops::zero_out::call(self_meta, out_meta); |
| 10268 | } |
| 10269 | |
| 10270 | at::Tensor self_; |
| 10271 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10272 | at::functionalization::impl::sync(self); |
| 10273 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10274 | } else { |
| 10275 | self_ = self; |
| 10276 | } |
| 10277 | |
| 10278 | at::Tensor out_; |
| 10279 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10280 | at::functionalization::impl::sync(out); |
| 10281 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10282 | } else { |
| 10283 | out_ = out; |
| 10284 | } |
| 10285 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10286 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 10287 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10288 | TORCH_INTERNAL_ASSERT(false, |
| 10289 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10290 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10291 | } else { |
| 10292 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10293 | at::AutoDispatchSkipFunctionalize guard; |
| 10294 | at::Tensor tmp_output = at::_ops::zero_out::call(self_, out_); |
| 10295 | return out;; |
| 10296 | } |
| 10297 | } else { |
| 10298 | at::Tensor tmp_output; |
| 10299 | { |
| 10300 | at::AutoDispatchSkipFunctionalize guard; |
| 10301 | tmp_output = at::_ops::zero::call(self_); |
| 10302 | } |
| 10303 | at::functionalization::impl::replace_(out, tmp_output); |
| 10304 | at::functionalization::impl::commit_update(out); |
| 10305 | at::functionalization::impl::sync(out); |
| 10306 | return out; |
| 10307 | } |
| 10308 | } |
| 10309 | |
| 10310 | at::Tensor & zero_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 10311 | if (true) { |
| 10312 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10313 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10314 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10315 | auto self_meta = to_meta(self); |
| 10316 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10317 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10318 | at::_ops::zero_::call(self_meta); |
| 10319 | } |
| 10320 | |
| 10321 | at::Tensor self_; |
| 10322 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10323 | at::functionalization::impl::sync(self); |
| 10324 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10325 | } else { |
| 10326 | self_ = self; |
| 10327 | } |
| 10328 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 10329 | if ((false)) { |
| 10330 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10331 | TORCH_INTERNAL_ASSERT(false, |
| 10332 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10333 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10334 | } else { |
| 10335 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10336 | at::AutoDispatchSkipFunctionalize guard; |
| 10337 | at::Tensor tmp_output = at::_ops::zero_::call(self_); |
| 10338 | return self;; |
| 10339 | } |
| 10340 | } else { |
| 10341 | at::Tensor tmp_output; |
| 10342 | { |
| 10343 | at::AutoDispatchSkipFunctionalize guard; |
| 10344 | tmp_output = at::_ops::zero::call(self_); |
| 10345 | } |
| 10346 | at::functionalization::impl::replace_(self, tmp_output); |
| 10347 | at::functionalization::impl::commit_update(self); |
| 10348 | at::functionalization::impl::sync(self); |
| 10349 | return self; |
| 10350 | } |
| 10351 | } |
| 10352 | |
| 10353 | at::Tensor & heaviside_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & values, at::Tensor & out) { |
| 10354 | if (false) { |
| 10355 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10356 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10357 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10358 | auto self_meta = to_meta(self); |
| 10359 | auto values_meta = to_meta(values); |
| 10360 | auto out_meta = to_meta(out); |
| 10361 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10362 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10363 | at::_ops::heaviside_out::call(self_meta, values_meta, out_meta); |
| 10364 | } |
| 10365 | |
| 10366 | at::Tensor self_; |
| 10367 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10368 | at::functionalization::impl::sync(self); |
| 10369 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10370 | } else { |
| 10371 | self_ = self; |
| 10372 | } |
| 10373 | |
| 10374 | at::Tensor values_; |
| 10375 | if (at::functionalization::impl::isFunctionalTensor(values)) { |
| 10376 | at::functionalization::impl::sync(values); |
| 10377 | values_ = at::functionalization::impl::from_functional_tensor(values); |
| 10378 | } else { |
| 10379 | values_ = values; |
| 10380 | } |
| 10381 | |
| 10382 | at::Tensor out_; |
| 10383 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10384 | at::functionalization::impl::sync(out); |
| 10385 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10386 | } else { |
| 10387 | out_ = out; |
| 10388 | } |
| 10389 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10390 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(values))) { |
| 10391 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10392 | TORCH_INTERNAL_ASSERT(false, |
| 10393 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10394 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10395 | } else { |
| 10396 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10397 | at::AutoDispatchSkipFunctionalize guard; |
| 10398 | at::Tensor tmp_output = at::_ops::heaviside_out::call(self_, values_, out_); |
| 10399 | return out;; |
| 10400 | } |
| 10401 | } else { |
| 10402 | at::Tensor tmp_output; |
| 10403 | { |
| 10404 | at::AutoDispatchSkipFunctionalize guard; |
| 10405 | tmp_output = at::_ops::heaviside::call(self_, values_); |
| 10406 | } |
| 10407 | at::functionalization::impl::replace_(out, tmp_output); |
| 10408 | at::functionalization::impl::commit_update(out); |
| 10409 | at::functionalization::impl::sync(out); |
| 10410 | return out; |
| 10411 | } |
| 10412 | } |
| 10413 | |
| 10414 | at::Tensor & heaviside_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & values) { |
| 10415 | if (true) { |
| 10416 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10417 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10418 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10419 | auto self_meta = to_meta(self); |
| 10420 | auto values_meta = to_meta(values); |
| 10421 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10422 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10423 | at::_ops::heaviside_::call(self_meta, values_meta); |
| 10424 | } |
| 10425 | |
| 10426 | at::Tensor self_; |
| 10427 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10428 | at::functionalization::impl::sync(self); |
| 10429 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10430 | } else { |
| 10431 | self_ = self; |
| 10432 | } |
| 10433 | |
| 10434 | at::Tensor values_; |
| 10435 | if (at::functionalization::impl::isFunctionalTensor(values)) { |
| 10436 | at::functionalization::impl::sync(values); |
| 10437 | values_ = at::functionalization::impl::from_functional_tensor(values); |
| 10438 | } else { |
| 10439 | values_ = values; |
| 10440 | } |
| 10441 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 10442 | if ((false || at::functionalization::impl::isFunctionalTensor(values))) { |
| 10443 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10444 | TORCH_INTERNAL_ASSERT(false, |
| 10445 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10446 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10447 | } else { |
| 10448 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10449 | at::AutoDispatchSkipFunctionalize guard; |
| 10450 | at::Tensor tmp_output = at::_ops::heaviside_::call(self_, values_); |
| 10451 | return self;; |
| 10452 | } |
| 10453 | } else { |
| 10454 | at::Tensor tmp_output; |
| 10455 | { |
| 10456 | at::AutoDispatchSkipFunctionalize guard; |
| 10457 | tmp_output = at::_ops::heaviside::call(self_, values_); |
| 10458 | } |
| 10459 | at::functionalization::impl::replace_(self, tmp_output); |
| 10460 | at::functionalization::impl::commit_update(self); |
| 10461 | at::functionalization::impl::sync(self); |
| 10462 | return self; |
| 10463 | } |
| 10464 | } |
| 10465 | |
| 10466 | at::Tensor & addmm_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & mat1, const at::Tensor & mat2, const at::Scalar & beta, const at::Scalar & alpha, at::Tensor & out) { |
| 10467 | if (false) { |
| 10468 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10469 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10470 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10471 | auto self_meta = to_meta(self); |
| 10472 | auto mat1_meta = to_meta(mat1); |
| 10473 | auto mat2_meta = to_meta(mat2); |
| 10474 | auto out_meta = to_meta(out); |
| 10475 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10476 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10477 | at::_ops::addmm_out::call(self_meta, mat1_meta, mat2_meta, beta, alpha, out_meta); |
| 10478 | } |
| 10479 | |
| 10480 | at::Tensor self_; |
| 10481 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10482 | at::functionalization::impl::sync(self); |
| 10483 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10484 | } else { |
| 10485 | self_ = self; |
| 10486 | } |
| 10487 | |
| 10488 | at::Tensor mat1_; |
| 10489 | if (at::functionalization::impl::isFunctionalTensor(mat1)) { |
| 10490 | at::functionalization::impl::sync(mat1); |
| 10491 | mat1_ = at::functionalization::impl::from_functional_tensor(mat1); |
| 10492 | } else { |
| 10493 | mat1_ = mat1; |
| 10494 | } |
| 10495 | |
| 10496 | at::Tensor mat2_; |
| 10497 | if (at::functionalization::impl::isFunctionalTensor(mat2)) { |
| 10498 | at::functionalization::impl::sync(mat2); |
| 10499 | mat2_ = at::functionalization::impl::from_functional_tensor(mat2); |
| 10500 | } else { |
| 10501 | mat2_ = mat2; |
| 10502 | } |
| 10503 | |
| 10504 | at::Tensor out_; |
| 10505 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10506 | at::functionalization::impl::sync(out); |
| 10507 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10508 | } else { |
| 10509 | out_ = out; |
| 10510 | } |
| 10511 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10512 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(mat1) || at::functionalization::impl::isFunctionalTensor(mat2))) { |
| 10513 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10514 | TORCH_INTERNAL_ASSERT(false, |
| 10515 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10516 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10517 | } else { |
| 10518 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10519 | at::AutoDispatchSkipFunctionalize guard; |
| 10520 | at::Tensor tmp_output = at::_ops::addmm_out::call(self_, mat1_, mat2_, beta, alpha, out_); |
| 10521 | return out;; |
| 10522 | } |
| 10523 | } else { |
| 10524 | at::Tensor tmp_output; |
| 10525 | { |
| 10526 | at::AutoDispatchSkipFunctionalize guard; |
| 10527 | tmp_output = at::_ops::addmm::call(self_, mat1_, mat2_, beta, alpha); |
| 10528 | } |
| 10529 | at::functionalization::impl::replace_(out, tmp_output); |
| 10530 | at::functionalization::impl::commit_update(out); |
| 10531 | at::functionalization::impl::sync(out); |
| 10532 | return out; |
| 10533 | } |
| 10534 | } |
| 10535 | |
| 10536 | at::Tensor & addmm_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & mat1, const at::Tensor & mat2, const at::Scalar & beta, const at::Scalar & alpha) { |
| 10537 | if (true) { |
| 10538 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10539 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10540 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10541 | auto self_meta = to_meta(self); |
| 10542 | auto mat1_meta = to_meta(mat1); |
| 10543 | auto mat2_meta = to_meta(mat2); |
| 10544 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10545 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10546 | at::_ops::addmm_::call(self_meta, mat1_meta, mat2_meta, beta, alpha); |
| 10547 | } |
| 10548 | |
| 10549 | at::Tensor self_; |
| 10550 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10551 | at::functionalization::impl::sync(self); |
| 10552 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10553 | } else { |
| 10554 | self_ = self; |
| 10555 | } |
| 10556 | |
| 10557 | at::Tensor mat1_; |
| 10558 | if (at::functionalization::impl::isFunctionalTensor(mat1)) { |
| 10559 | at::functionalization::impl::sync(mat1); |
| 10560 | mat1_ = at::functionalization::impl::from_functional_tensor(mat1); |
| 10561 | } else { |
| 10562 | mat1_ = mat1; |
| 10563 | } |
| 10564 | |
| 10565 | at::Tensor mat2_; |
| 10566 | if (at::functionalization::impl::isFunctionalTensor(mat2)) { |
| 10567 | at::functionalization::impl::sync(mat2); |
| 10568 | mat2_ = at::functionalization::impl::from_functional_tensor(mat2); |
| 10569 | } else { |
| 10570 | mat2_ = mat2; |
| 10571 | } |
| 10572 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 10573 | if ((false || at::functionalization::impl::isFunctionalTensor(mat1) || at::functionalization::impl::isFunctionalTensor(mat2))) { |
| 10574 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10575 | TORCH_INTERNAL_ASSERT(false, |
| 10576 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10577 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10578 | } else { |
| 10579 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10580 | at::AutoDispatchSkipFunctionalize guard; |
| 10581 | at::Tensor tmp_output = at::_ops::addmm_::call(self_, mat1_, mat2_, beta, alpha); |
| 10582 | return self;; |
| 10583 | } |
| 10584 | } else { |
| 10585 | at::Tensor tmp_output; |
| 10586 | { |
| 10587 | at::AutoDispatchSkipFunctionalize guard; |
| 10588 | tmp_output = at::_ops::addmm::call(self_, mat1_, mat2_, beta, alpha); |
| 10589 | } |
| 10590 | at::functionalization::impl::replace_(self, tmp_output); |
| 10591 | at::functionalization::impl::commit_update(self); |
| 10592 | at::functionalization::impl::sync(self); |
| 10593 | return self; |
| 10594 | } |
| 10595 | } |
| 10596 | |
| 10597 | at::Tensor & _sparse_coo_tensor_with_dims_out_out(c10::DispatchKeySet dispatchKeySet, int64_t sparse_dim, int64_t dense_dim, at::IntArrayRef size, at::Tensor & out) { |
| 10598 | if (false) { |
| 10599 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10600 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10601 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10602 | auto out_meta = to_meta(out); |
| 10603 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10604 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10605 | at::_ops::_sparse_coo_tensor_with_dims_out::call(sparse_dim, dense_dim, size, out_meta); |
| 10606 | } |
| 10607 | |
| 10608 | at::Tensor out_; |
| 10609 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10610 | at::functionalization::impl::sync(out); |
| 10611 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10612 | } else { |
| 10613 | out_ = out; |
| 10614 | } |
| 10615 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10616 | if ((false)) { |
| 10617 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10618 | TORCH_INTERNAL_ASSERT(false, |
| 10619 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10620 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10621 | } else { |
| 10622 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10623 | at::AutoDispatchSkipFunctionalize guard; |
| 10624 | at::Tensor tmp_output = at::_ops::_sparse_coo_tensor_with_dims_out::call(sparse_dim, dense_dim, size, out_); |
| 10625 | return out;; |
| 10626 | } |
| 10627 | } else { |
| 10628 | at::Tensor tmp_output; |
| 10629 | { |
| 10630 | at::AutoDispatchSkipFunctionalize guard; |
| 10631 | tmp_output = at::_ops::_sparse_coo_tensor_with_dims::call(sparse_dim, dense_dim, size, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 10632 | } |
| 10633 | at::functionalization::impl::replace_(out, tmp_output); |
| 10634 | at::functionalization::impl::commit_update(out); |
| 10635 | at::functionalization::impl::sync(out); |
| 10636 | return out; |
| 10637 | } |
| 10638 | } |
| 10639 | |
| 10640 | at::Tensor & _sparse_coo_tensor_with_dims_and_tensors_out_out(c10::DispatchKeySet dispatchKeySet, int64_t sparse_dim, int64_t dense_dim, c10::SymIntArrayRef size, const at::Tensor & indices, const at::Tensor & values, at::Tensor & out) { |
| 10641 | if (false) { |
| 10642 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10643 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10644 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10645 | auto indices_meta = to_meta(indices); |
| 10646 | auto values_meta = to_meta(values); |
| 10647 | auto out_meta = to_meta(out); |
| 10648 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10649 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10650 | at::_ops::_sparse_coo_tensor_with_dims_and_tensors_out::call(sparse_dim, dense_dim, size, indices_meta, values_meta, out_meta); |
| 10651 | } |
| 10652 | |
| 10653 | at::Tensor indices_; |
| 10654 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 10655 | at::functionalization::impl::sync(indices); |
| 10656 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 10657 | } else { |
| 10658 | indices_ = indices; |
| 10659 | } |
| 10660 | |
| 10661 | at::Tensor values_; |
| 10662 | if (at::functionalization::impl::isFunctionalTensor(values)) { |
| 10663 | at::functionalization::impl::sync(values); |
| 10664 | values_ = at::functionalization::impl::from_functional_tensor(values); |
| 10665 | } else { |
| 10666 | values_ = values; |
| 10667 | } |
| 10668 | |
| 10669 | at::Tensor out_; |
| 10670 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10671 | at::functionalization::impl::sync(out); |
| 10672 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10673 | } else { |
| 10674 | out_ = out; |
| 10675 | } |
| 10676 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10677 | if ((false || at::functionalization::impl::isFunctionalTensor(indices) || at::functionalization::impl::isFunctionalTensor(values))) { |
| 10678 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10679 | TORCH_INTERNAL_ASSERT(false, |
| 10680 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10681 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10682 | } else { |
| 10683 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10684 | at::AutoDispatchSkipFunctionalize guard; |
| 10685 | at::Tensor tmp_output = at::_ops::_sparse_coo_tensor_with_dims_and_tensors_out::call(sparse_dim, dense_dim, size, indices_, values_, out_); |
| 10686 | return out;; |
| 10687 | } |
| 10688 | } else { |
| 10689 | at::Tensor tmp_output; |
| 10690 | { |
| 10691 | at::AutoDispatchSkipFunctionalize guard; |
| 10692 | tmp_output = at::_ops::_sparse_coo_tensor_with_dims_and_tensors::call(sparse_dim, dense_dim, size, indices_, values_, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 10693 | } |
| 10694 | at::functionalization::impl::replace_(out, tmp_output); |
| 10695 | at::functionalization::impl::commit_update(out); |
| 10696 | at::functionalization::impl::sync(out); |
| 10697 | return out; |
| 10698 | } |
| 10699 | } |
| 10700 | |
| 10701 | const at::Tensor & sparse_resize_and_clear_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef size, int64_t sparse_dim, int64_t dense_dim, const at::Tensor & out) { |
| 10702 | if (false) { |
| 10703 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10704 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10705 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10706 | auto self_meta = to_meta(self); |
| 10707 | auto out_meta = to_meta(out); |
| 10708 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10709 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10710 | at::_ops::sparse_resize_and_clear_out::call(self_meta, size, sparse_dim, dense_dim, out_meta); |
| 10711 | } |
| 10712 | |
| 10713 | at::Tensor self_; |
| 10714 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10715 | at::functionalization::impl::sync(self); |
| 10716 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10717 | } else { |
| 10718 | self_ = self; |
| 10719 | } |
| 10720 | |
| 10721 | at::Tensor out_; |
| 10722 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10723 | at::functionalization::impl::sync(out); |
| 10724 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10725 | } else { |
| 10726 | out_ = out; |
| 10727 | } |
| 10728 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10729 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 10730 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10731 | TORCH_INTERNAL_ASSERT(false, |
| 10732 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10733 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10734 | } else { |
| 10735 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10736 | at::AutoDispatchSkipFunctionalize guard; |
| 10737 | at::Tensor tmp_output = at::_ops::sparse_resize_and_clear_out::call(self_, size, sparse_dim, dense_dim, out_); |
| 10738 | return out;; |
| 10739 | } |
| 10740 | } else { |
| 10741 | at::Tensor tmp_output; |
| 10742 | { |
| 10743 | at::AutoDispatchSkipFunctionalize guard; |
| 10744 | tmp_output = at::_ops::sparse_resize_and_clear::call(self_, size, sparse_dim, dense_dim); |
| 10745 | } |
| 10746 | at::functionalization::impl::replace_(out, tmp_output); |
| 10747 | at::functionalization::impl::commit_update(out); |
| 10748 | at::functionalization::impl::sync(out); |
| 10749 | return out; |
| 10750 | } |
| 10751 | } |
| 10752 | |
| 10753 | const at::Tensor & sparse_resize_and_clear_(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef size, int64_t sparse_dim, int64_t dense_dim) { |
| 10754 | if (true) { |
| 10755 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10756 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10757 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10758 | auto self_meta = to_meta(self); |
| 10759 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10760 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10761 | at::_ops::sparse_resize_and_clear_::call(self_meta, size, sparse_dim, dense_dim); |
| 10762 | } |
| 10763 | |
| 10764 | at::Tensor self_; |
| 10765 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10766 | at::functionalization::impl::sync(self); |
| 10767 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10768 | } else { |
| 10769 | self_ = self; |
| 10770 | } |
| 10771 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 10772 | if ((false)) { |
| 10773 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10774 | TORCH_INTERNAL_ASSERT(false, |
| 10775 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10776 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10777 | } else { |
| 10778 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10779 | at::AutoDispatchSkipFunctionalize guard; |
| 10780 | at::Tensor tmp_output = at::_ops::sparse_resize_and_clear_::call(self_, size, sparse_dim, dense_dim); |
| 10781 | return self;; |
| 10782 | } |
| 10783 | } else { |
| 10784 | at::Tensor tmp_output; |
| 10785 | { |
| 10786 | at::AutoDispatchSkipFunctionalize guard; |
| 10787 | tmp_output = at::_ops::sparse_resize_and_clear::call(self_, size, sparse_dim, dense_dim); |
| 10788 | } |
| 10789 | at::functionalization::impl::replace_(self, tmp_output); |
| 10790 | at::functionalization::impl::commit_update(self); |
| 10791 | at::functionalization::impl::sync(self); |
| 10792 | return self; |
| 10793 | } |
| 10794 | } |
| 10795 | |
| 10796 | at::Tensor & hspmm_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & mat1, const at::Tensor & mat2, at::Tensor & out) { |
| 10797 | if (false) { |
| 10798 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10799 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10800 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10801 | auto mat1_meta = to_meta(mat1); |
| 10802 | auto mat2_meta = to_meta(mat2); |
| 10803 | auto out_meta = to_meta(out); |
| 10804 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10805 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10806 | at::_ops::hspmm_out::call(mat1_meta, mat2_meta, out_meta); |
| 10807 | } |
| 10808 | |
| 10809 | at::Tensor mat1_; |
| 10810 | if (at::functionalization::impl::isFunctionalTensor(mat1)) { |
| 10811 | at::functionalization::impl::sync(mat1); |
| 10812 | mat1_ = at::functionalization::impl::from_functional_tensor(mat1); |
| 10813 | } else { |
| 10814 | mat1_ = mat1; |
| 10815 | } |
| 10816 | |
| 10817 | at::Tensor mat2_; |
| 10818 | if (at::functionalization::impl::isFunctionalTensor(mat2)) { |
| 10819 | at::functionalization::impl::sync(mat2); |
| 10820 | mat2_ = at::functionalization::impl::from_functional_tensor(mat2); |
| 10821 | } else { |
| 10822 | mat2_ = mat2; |
| 10823 | } |
| 10824 | |
| 10825 | at::Tensor out_; |
| 10826 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10827 | at::functionalization::impl::sync(out); |
| 10828 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10829 | } else { |
| 10830 | out_ = out; |
| 10831 | } |
| 10832 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10833 | if ((false || at::functionalization::impl::isFunctionalTensor(mat1) || at::functionalization::impl::isFunctionalTensor(mat2))) { |
| 10834 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10835 | TORCH_INTERNAL_ASSERT(false, |
| 10836 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10837 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10838 | } else { |
| 10839 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10840 | at::AutoDispatchSkipFunctionalize guard; |
| 10841 | at::Tensor tmp_output = at::_ops::hspmm_out::call(mat1_, mat2_, out_); |
| 10842 | return out;; |
| 10843 | } |
| 10844 | } else { |
| 10845 | at::Tensor tmp_output; |
| 10846 | { |
| 10847 | at::AutoDispatchSkipFunctionalize guard; |
| 10848 | tmp_output = at::_ops::hspmm::call(mat1_, mat2_); |
| 10849 | } |
| 10850 | at::functionalization::impl::replace_(out, tmp_output); |
| 10851 | at::functionalization::impl::commit_update(out); |
| 10852 | at::functionalization::impl::sync(out); |
| 10853 | return out; |
| 10854 | } |
| 10855 | } |
| 10856 | |
| 10857 | at::Tensor & to_sparse_out_sparse_dim_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t sparse_dim, at::Tensor & out) { |
| 10858 | if (false) { |
| 10859 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10860 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10861 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10862 | auto self_meta = to_meta(self); |
| 10863 | auto out_meta = to_meta(out); |
| 10864 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10865 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10866 | at::_ops::to_sparse_sparse_dim_out::call(self_meta, sparse_dim, out_meta); |
| 10867 | } |
| 10868 | |
| 10869 | at::Tensor self_; |
| 10870 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10871 | at::functionalization::impl::sync(self); |
| 10872 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10873 | } else { |
| 10874 | self_ = self; |
| 10875 | } |
| 10876 | |
| 10877 | at::Tensor out_; |
| 10878 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10879 | at::functionalization::impl::sync(out); |
| 10880 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10881 | } else { |
| 10882 | out_ = out; |
| 10883 | } |
| 10884 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10885 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 10886 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10887 | TORCH_INTERNAL_ASSERT(false, |
| 10888 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10889 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10890 | } else { |
| 10891 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10892 | at::AutoDispatchSkipFunctionalize guard; |
| 10893 | at::Tensor tmp_output = at::_ops::to_sparse_sparse_dim_out::call(self_, sparse_dim, out_); |
| 10894 | return out;; |
| 10895 | } |
| 10896 | } else { |
| 10897 | at::Tensor tmp_output; |
| 10898 | { |
| 10899 | at::AutoDispatchSkipFunctionalize guard; |
| 10900 | tmp_output = at::_ops::to_sparse_sparse_dim::call(self_, sparse_dim); |
| 10901 | } |
| 10902 | at::functionalization::impl::replace_(out, tmp_output); |
| 10903 | at::functionalization::impl::commit_update(out); |
| 10904 | at::functionalization::impl::sync(out); |
| 10905 | return out; |
| 10906 | } |
| 10907 | } |
| 10908 | |
| 10909 | at::Tensor & to_sparse_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<at::Layout> layout, at::OptionalIntArrayRef blocksize, c10::optional<int64_t> dense_dim, at::Tensor & out) { |
| 10910 | if (false) { |
| 10911 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10912 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10913 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10914 | auto self_meta = to_meta(self); |
| 10915 | auto out_meta = to_meta(out); |
| 10916 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10917 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10918 | at::_ops::to_sparse_out::call(self_meta, layout, blocksize, dense_dim, out_meta); |
| 10919 | } |
| 10920 | |
| 10921 | at::Tensor self_; |
| 10922 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10923 | at::functionalization::impl::sync(self); |
| 10924 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10925 | } else { |
| 10926 | self_ = self; |
| 10927 | } |
| 10928 | |
| 10929 | at::Tensor out_; |
| 10930 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10931 | at::functionalization::impl::sync(out); |
| 10932 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10933 | } else { |
| 10934 | out_ = out; |
| 10935 | } |
| 10936 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10937 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 10938 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10939 | TORCH_INTERNAL_ASSERT(false, |
| 10940 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10941 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10942 | } else { |
| 10943 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10944 | at::AutoDispatchSkipFunctionalize guard; |
| 10945 | at::Tensor tmp_output = at::_ops::to_sparse_out::call(self_, layout, blocksize, dense_dim, out_); |
| 10946 | return out;; |
| 10947 | } |
| 10948 | } else { |
| 10949 | at::Tensor tmp_output; |
| 10950 | { |
| 10951 | at::AutoDispatchSkipFunctionalize guard; |
| 10952 | tmp_output = at::_ops::to_sparse::call(self_, layout, blocksize, dense_dim); |
| 10953 | } |
| 10954 | at::functionalization::impl::replace_(out, tmp_output); |
| 10955 | at::functionalization::impl::commit_update(out); |
| 10956 | at::functionalization::impl::sync(out); |
| 10957 | return out; |
| 10958 | } |
| 10959 | } |
| 10960 | |
| 10961 | at::Tensor & to_sparse_bsr_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef blocksize, c10::optional<int64_t> dense_dim, at::Tensor & out) { |
| 10962 | if (false) { |
| 10963 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 10964 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 10965 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 10966 | auto self_meta = to_meta(self); |
| 10967 | auto out_meta = to_meta(out); |
| 10968 | at::AutoDispatchSkipFunctionalize func_guard; |
| 10969 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 10970 | at::_ops::to_sparse_bsr_out::call(self_meta, blocksize, dense_dim, out_meta); |
| 10971 | } |
| 10972 | |
| 10973 | at::Tensor self_; |
| 10974 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 10975 | at::functionalization::impl::sync(self); |
| 10976 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 10977 | } else { |
| 10978 | self_ = self; |
| 10979 | } |
| 10980 | |
| 10981 | at::Tensor out_; |
| 10982 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 10983 | at::functionalization::impl::sync(out); |
| 10984 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 10985 | } else { |
| 10986 | out_ = out; |
| 10987 | } |
| 10988 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 10989 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 10990 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 10991 | TORCH_INTERNAL_ASSERT(false, |
| 10992 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 10993 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 10994 | } else { |
| 10995 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 10996 | at::AutoDispatchSkipFunctionalize guard; |
| 10997 | at::Tensor tmp_output = at::_ops::to_sparse_bsr_out::call(self_, blocksize, dense_dim, out_); |
| 10998 | return out;; |
| 10999 | } |
| 11000 | } else { |
| 11001 | at::Tensor tmp_output; |
| 11002 | { |
| 11003 | at::AutoDispatchSkipFunctionalize guard; |
| 11004 | tmp_output = at::_ops::to_sparse_bsr::call(self_, blocksize, dense_dim); |
| 11005 | } |
| 11006 | at::functionalization::impl::replace_(out, tmp_output); |
| 11007 | at::functionalization::impl::commit_update(out); |
| 11008 | at::functionalization::impl::sync(out); |
| 11009 | return out; |
| 11010 | } |
| 11011 | } |
| 11012 | |
| 11013 | at::Tensor & to_mkldnn_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 11014 | if (false) { |
| 11015 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11016 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11017 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11018 | auto self_meta = to_meta(self); |
| 11019 | auto out_meta = to_meta(out); |
| 11020 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11021 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11022 | at::_ops::to_mkldnn_out::call(self_meta, dtype, out_meta); |
| 11023 | } |
| 11024 | |
| 11025 | at::Tensor self_; |
| 11026 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11027 | at::functionalization::impl::sync(self); |
| 11028 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11029 | } else { |
| 11030 | self_ = self; |
| 11031 | } |
| 11032 | |
| 11033 | at::Tensor out_; |
| 11034 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11035 | at::functionalization::impl::sync(out); |
| 11036 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11037 | } else { |
| 11038 | out_ = out; |
| 11039 | } |
| 11040 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11041 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 11042 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11043 | TORCH_INTERNAL_ASSERT(false, |
| 11044 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11045 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11046 | } else { |
| 11047 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11048 | at::AutoDispatchSkipFunctionalize guard; |
| 11049 | at::Tensor tmp_output = at::_ops::to_mkldnn_out::call(self_, dtype, out_); |
| 11050 | return out;; |
| 11051 | } |
| 11052 | } else { |
| 11053 | at::Tensor tmp_output; |
| 11054 | { |
| 11055 | at::AutoDispatchSkipFunctionalize guard; |
| 11056 | tmp_output = at::_ops::to_mkldnn::call(self_, dtype); |
| 11057 | } |
| 11058 | at::functionalization::impl::replace_(out, tmp_output); |
| 11059 | at::functionalization::impl::commit_update(out); |
| 11060 | at::functionalization::impl::sync(out); |
| 11061 | return out; |
| 11062 | } |
| 11063 | } |
| 11064 | |
| 11065 | at::Tensor & mkldnn_reorder_conv3d_weight_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef padding, at::IntArrayRef stride, at::IntArrayRef dilation, int64_t groups, at::Tensor & out) { |
| 11066 | if (false) { |
| 11067 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11068 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11069 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11070 | auto self_meta = to_meta(self); |
| 11071 | auto out_meta = to_meta(out); |
| 11072 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11073 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11074 | at::_ops::mkldnn_reorder_conv3d_weight_out::call(self_meta, padding, stride, dilation, groups, out_meta); |
| 11075 | } |
| 11076 | |
| 11077 | at::Tensor self_; |
| 11078 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11079 | at::functionalization::impl::sync(self); |
| 11080 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11081 | } else { |
| 11082 | self_ = self; |
| 11083 | } |
| 11084 | |
| 11085 | at::Tensor out_; |
| 11086 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11087 | at::functionalization::impl::sync(out); |
| 11088 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11089 | } else { |
| 11090 | out_ = out; |
| 11091 | } |
| 11092 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11093 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 11094 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11095 | TORCH_INTERNAL_ASSERT(false, |
| 11096 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11097 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11098 | } else { |
| 11099 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11100 | at::AutoDispatchSkipFunctionalize guard; |
| 11101 | at::Tensor tmp_output = at::_ops::mkldnn_reorder_conv3d_weight_out::call(self_, padding, stride, dilation, groups, out_); |
| 11102 | return out;; |
| 11103 | } |
| 11104 | } else { |
| 11105 | at::Tensor tmp_output; |
| 11106 | { |
| 11107 | at::AutoDispatchSkipFunctionalize guard; |
| 11108 | tmp_output = at::_ops::mkldnn_reorder_conv3d_weight::call(self_, padding, stride, dilation, groups); |
| 11109 | } |
| 11110 | at::functionalization::impl::replace_(out, tmp_output); |
| 11111 | at::functionalization::impl::commit_update(out); |
| 11112 | at::functionalization::impl::sync(out); |
| 11113 | return out; |
| 11114 | } |
| 11115 | } |
| 11116 | |
| 11117 | at::Tensor & q_per_channel_scales_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 11118 | if (false) { |
| 11119 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11120 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11121 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11122 | auto self_meta = to_meta(self); |
| 11123 | auto out_meta = to_meta(out); |
| 11124 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11125 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11126 | at::_ops::q_per_channel_scales_out::call(self_meta, out_meta); |
| 11127 | } |
| 11128 | |
| 11129 | at::Tensor self_; |
| 11130 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11131 | at::functionalization::impl::sync(self); |
| 11132 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11133 | } else { |
| 11134 | self_ = self; |
| 11135 | } |
| 11136 | |
| 11137 | at::Tensor out_; |
| 11138 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11139 | at::functionalization::impl::sync(out); |
| 11140 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11141 | } else { |
| 11142 | out_ = out; |
| 11143 | } |
| 11144 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11145 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 11146 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11147 | TORCH_INTERNAL_ASSERT(false, |
| 11148 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11149 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11150 | } else { |
| 11151 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11152 | at::AutoDispatchSkipFunctionalize guard; |
| 11153 | at::Tensor tmp_output = at::_ops::q_per_channel_scales_out::call(self_, out_); |
| 11154 | return out;; |
| 11155 | } |
| 11156 | } else { |
| 11157 | at::Tensor tmp_output; |
| 11158 | { |
| 11159 | at::AutoDispatchSkipFunctionalize guard; |
| 11160 | tmp_output = at::_ops::q_per_channel_scales::call(self_); |
| 11161 | } |
| 11162 | at::functionalization::impl::replace_(out, tmp_output); |
| 11163 | at::functionalization::impl::commit_update(out); |
| 11164 | at::functionalization::impl::sync(out); |
| 11165 | return out; |
| 11166 | } |
| 11167 | } |
| 11168 | |
| 11169 | at::Tensor & int_repr_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 11170 | if (false) { |
| 11171 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11172 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11173 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11174 | auto self_meta = to_meta(self); |
| 11175 | auto out_meta = to_meta(out); |
| 11176 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11177 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11178 | at::_ops::int_repr_out::call(self_meta, out_meta); |
| 11179 | } |
| 11180 | |
| 11181 | at::Tensor self_; |
| 11182 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11183 | at::functionalization::impl::sync(self); |
| 11184 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11185 | } else { |
| 11186 | self_ = self; |
| 11187 | } |
| 11188 | |
| 11189 | at::Tensor out_; |
| 11190 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11191 | at::functionalization::impl::sync(out); |
| 11192 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11193 | } else { |
| 11194 | out_ = out; |
| 11195 | } |
| 11196 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11197 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 11198 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11199 | TORCH_INTERNAL_ASSERT(false, |
| 11200 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11201 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11202 | } else { |
| 11203 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11204 | at::AutoDispatchSkipFunctionalize guard; |
| 11205 | at::Tensor tmp_output = at::_ops::int_repr_out::call(self_, out_); |
| 11206 | return out;; |
| 11207 | } |
| 11208 | } else { |
| 11209 | at::Tensor tmp_output; |
| 11210 | { |
| 11211 | at::AutoDispatchSkipFunctionalize guard; |
| 11212 | tmp_output = at::_ops::int_repr::call(self_); |
| 11213 | } |
| 11214 | at::functionalization::impl::replace_(out, tmp_output); |
| 11215 | at::functionalization::impl::commit_update(out); |
| 11216 | at::functionalization::impl::sync(out); |
| 11217 | return out; |
| 11218 | } |
| 11219 | } |
| 11220 | |
| 11221 | at::Tensor & _make_per_tensor_quantized_tensor_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, double scale, int64_t zero_point, at::Tensor & out) { |
| 11222 | if (false) { |
| 11223 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11224 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11225 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11226 | auto self_meta = to_meta(self); |
| 11227 | auto out_meta = to_meta(out); |
| 11228 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11229 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11230 | at::_ops::_make_per_tensor_quantized_tensor_out::call(self_meta, scale, zero_point, out_meta); |
| 11231 | } |
| 11232 | |
| 11233 | at::Tensor self_; |
| 11234 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11235 | at::functionalization::impl::sync(self); |
| 11236 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11237 | } else { |
| 11238 | self_ = self; |
| 11239 | } |
| 11240 | |
| 11241 | at::Tensor out_; |
| 11242 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11243 | at::functionalization::impl::sync(out); |
| 11244 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11245 | } else { |
| 11246 | out_ = out; |
| 11247 | } |
| 11248 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11249 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 11250 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11251 | TORCH_INTERNAL_ASSERT(false, |
| 11252 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11253 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11254 | } else { |
| 11255 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11256 | at::AutoDispatchSkipFunctionalize guard; |
| 11257 | at::Tensor tmp_output = at::_ops::_make_per_tensor_quantized_tensor_out::call(self_, scale, zero_point, out_); |
| 11258 | return out;; |
| 11259 | } |
| 11260 | } else { |
| 11261 | at::Tensor tmp_output; |
| 11262 | { |
| 11263 | at::AutoDispatchSkipFunctionalize guard; |
| 11264 | tmp_output = at::_ops::_make_per_tensor_quantized_tensor::call(self_, scale, zero_point); |
| 11265 | } |
| 11266 | at::functionalization::impl::replace_(out, tmp_output); |
| 11267 | at::functionalization::impl::commit_update(out); |
| 11268 | at::functionalization::impl::sync(out); |
| 11269 | return out; |
| 11270 | } |
| 11271 | } |
| 11272 | |
| 11273 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> _thnn_fused_lstm_cell_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & input_gates, const at::Tensor & hidden_gates, const at::Tensor & cx, const c10::optional<at::Tensor> & input_bias, const c10::optional<at::Tensor> & hidden_bias, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 11274 | if (false) { |
| 11275 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11276 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11277 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11278 | auto input_gates_meta = to_meta(input_gates); |
| 11279 | auto hidden_gates_meta = to_meta(hidden_gates); |
| 11280 | auto cx_meta = to_meta(cx); |
| 11281 | auto input_bias_meta = to_meta(input_bias); |
| 11282 | auto hidden_bias_meta = to_meta(hidden_bias); |
| 11283 | auto out0_meta = to_meta(out0); |
| 11284 | auto out1_meta = to_meta(out1); |
| 11285 | auto out2_meta = to_meta(out2); |
| 11286 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11287 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11288 | at::_ops::_thnn_fused_lstm_cell_out::call(input_gates_meta, hidden_gates_meta, cx_meta, input_bias_meta, hidden_bias_meta, out0_meta, out1_meta, out2_meta); |
| 11289 | } |
| 11290 | |
| 11291 | at::Tensor input_gates_; |
| 11292 | if (at::functionalization::impl::isFunctionalTensor(input_gates)) { |
| 11293 | at::functionalization::impl::sync(input_gates); |
| 11294 | input_gates_ = at::functionalization::impl::from_functional_tensor(input_gates); |
| 11295 | } else { |
| 11296 | input_gates_ = input_gates; |
| 11297 | } |
| 11298 | |
| 11299 | at::Tensor hidden_gates_; |
| 11300 | if (at::functionalization::impl::isFunctionalTensor(hidden_gates)) { |
| 11301 | at::functionalization::impl::sync(hidden_gates); |
| 11302 | hidden_gates_ = at::functionalization::impl::from_functional_tensor(hidden_gates); |
| 11303 | } else { |
| 11304 | hidden_gates_ = hidden_gates; |
| 11305 | } |
| 11306 | |
| 11307 | at::Tensor cx_; |
| 11308 | if (at::functionalization::impl::isFunctionalTensor(cx)) { |
| 11309 | at::functionalization::impl::sync(cx); |
| 11310 | cx_ = at::functionalization::impl::from_functional_tensor(cx); |
| 11311 | } else { |
| 11312 | cx_ = cx; |
| 11313 | } |
| 11314 | |
| 11315 | c10::optional<at::Tensor> input_bias_; |
| 11316 | if (at::functionalization::impl::isFunctionalTensor(input_bias)) { |
| 11317 | at::functionalization::impl::sync(input_bias); |
| 11318 | input_bias_ = at::functionalization::impl::from_functional_tensor(input_bias); |
| 11319 | } else { |
| 11320 | input_bias_ = input_bias; |
| 11321 | } |
| 11322 | |
| 11323 | c10::optional<at::Tensor> hidden_bias_; |
| 11324 | if (at::functionalization::impl::isFunctionalTensor(hidden_bias)) { |
| 11325 | at::functionalization::impl::sync(hidden_bias); |
| 11326 | hidden_bias_ = at::functionalization::impl::from_functional_tensor(hidden_bias); |
| 11327 | } else { |
| 11328 | hidden_bias_ = hidden_bias; |
| 11329 | } |
| 11330 | |
| 11331 | at::Tensor out0_; |
| 11332 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 11333 | at::functionalization::impl::sync(out0); |
| 11334 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 11335 | } else { |
| 11336 | out0_ = out0; |
| 11337 | } |
| 11338 | |
| 11339 | at::Tensor out1_; |
| 11340 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 11341 | at::functionalization::impl::sync(out1); |
| 11342 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 11343 | } else { |
| 11344 | out1_ = out1; |
| 11345 | } |
| 11346 | |
| 11347 | at::Tensor out2_; |
| 11348 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 11349 | at::functionalization::impl::sync(out2); |
| 11350 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 11351 | } else { |
| 11352 | out2_ = out2; |
| 11353 | } |
| 11354 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 11355 | if ((false || at::functionalization::impl::isFunctionalTensor(input_gates) || at::functionalization::impl::isFunctionalTensor(hidden_gates) || at::functionalization::impl::isFunctionalTensor(cx) || at::functionalization::impl::isFunctionalTensor(input_bias) || at::functionalization::impl::isFunctionalTensor(hidden_bias))) { |
| 11356 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11357 | TORCH_INTERNAL_ASSERT(false, |
| 11358 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11359 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11360 | } else { |
| 11361 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11362 | at::AutoDispatchSkipFunctionalize guard; |
| 11363 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_thnn_fused_lstm_cell_out::call(input_gates_, hidden_gates_, cx_, input_bias_, hidden_bias_, out0_, out1_, out2_); |
| 11364 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 11365 | } |
| 11366 | } else { |
| 11367 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 11368 | { |
| 11369 | at::AutoDispatchSkipFunctionalize guard; |
| 11370 | tmp_output = at::_ops::_thnn_fused_lstm_cell::call(input_gates_, hidden_gates_, cx_, input_bias_, hidden_bias_); |
| 11371 | } |
| 11372 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 11373 | at::functionalization::impl::commit_update(out0); |
| 11374 | at::functionalization::impl::sync(out0); |
| 11375 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 11376 | at::functionalization::impl::commit_update(out1); |
| 11377 | at::functionalization::impl::sync(out1); |
| 11378 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 11379 | at::functionalization::impl::commit_update(out2); |
| 11380 | at::functionalization::impl::sync(out2); |
| 11381 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 11382 | } |
| 11383 | } |
| 11384 | |
| 11385 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> _thnn_fused_lstm_cell_backward_impl_out_out(c10::DispatchKeySet dispatchKeySet, const c10::optional<at::Tensor> & grad_hy, const c10::optional<at::Tensor> & grad_cy, const at::Tensor & cx, const at::Tensor & cy, const at::Tensor & workspace, bool has_bias, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 11386 | if (false) { |
| 11387 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11388 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11389 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11390 | auto grad_hy_meta = to_meta(grad_hy); |
| 11391 | auto grad_cy_meta = to_meta(grad_cy); |
| 11392 | auto cx_meta = to_meta(cx); |
| 11393 | auto cy_meta = to_meta(cy); |
| 11394 | auto workspace_meta = to_meta(workspace); |
| 11395 | auto out0_meta = to_meta(out0); |
| 11396 | auto out1_meta = to_meta(out1); |
| 11397 | auto out2_meta = to_meta(out2); |
| 11398 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11399 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11400 | at::_ops::_thnn_fused_lstm_cell_backward_impl_out::call(grad_hy_meta, grad_cy_meta, cx_meta, cy_meta, workspace_meta, has_bias, out0_meta, out1_meta, out2_meta); |
| 11401 | } |
| 11402 | |
| 11403 | c10::optional<at::Tensor> grad_hy_; |
| 11404 | if (at::functionalization::impl::isFunctionalTensor(grad_hy)) { |
| 11405 | at::functionalization::impl::sync(grad_hy); |
| 11406 | grad_hy_ = at::functionalization::impl::from_functional_tensor(grad_hy); |
| 11407 | } else { |
| 11408 | grad_hy_ = grad_hy; |
| 11409 | } |
| 11410 | |
| 11411 | c10::optional<at::Tensor> grad_cy_; |
| 11412 | if (at::functionalization::impl::isFunctionalTensor(grad_cy)) { |
| 11413 | at::functionalization::impl::sync(grad_cy); |
| 11414 | grad_cy_ = at::functionalization::impl::from_functional_tensor(grad_cy); |
| 11415 | } else { |
| 11416 | grad_cy_ = grad_cy; |
| 11417 | } |
| 11418 | |
| 11419 | at::Tensor cx_; |
| 11420 | if (at::functionalization::impl::isFunctionalTensor(cx)) { |
| 11421 | at::functionalization::impl::sync(cx); |
| 11422 | cx_ = at::functionalization::impl::from_functional_tensor(cx); |
| 11423 | } else { |
| 11424 | cx_ = cx; |
| 11425 | } |
| 11426 | |
| 11427 | at::Tensor cy_; |
| 11428 | if (at::functionalization::impl::isFunctionalTensor(cy)) { |
| 11429 | at::functionalization::impl::sync(cy); |
| 11430 | cy_ = at::functionalization::impl::from_functional_tensor(cy); |
| 11431 | } else { |
| 11432 | cy_ = cy; |
| 11433 | } |
| 11434 | |
| 11435 | at::Tensor workspace_; |
| 11436 | if (at::functionalization::impl::isFunctionalTensor(workspace)) { |
| 11437 | at::functionalization::impl::sync(workspace); |
| 11438 | workspace_ = at::functionalization::impl::from_functional_tensor(workspace); |
| 11439 | } else { |
| 11440 | workspace_ = workspace; |
| 11441 | } |
| 11442 | |
| 11443 | at::Tensor out0_; |
| 11444 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 11445 | at::functionalization::impl::sync(out0); |
| 11446 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 11447 | } else { |
| 11448 | out0_ = out0; |
| 11449 | } |
| 11450 | |
| 11451 | at::Tensor out1_; |
| 11452 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 11453 | at::functionalization::impl::sync(out1); |
| 11454 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 11455 | } else { |
| 11456 | out1_ = out1; |
| 11457 | } |
| 11458 | |
| 11459 | at::Tensor out2_; |
| 11460 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 11461 | at::functionalization::impl::sync(out2); |
| 11462 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 11463 | } else { |
| 11464 | out2_ = out2; |
| 11465 | } |
| 11466 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 11467 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_hy) || at::functionalization::impl::isFunctionalTensor(grad_cy) || at::functionalization::impl::isFunctionalTensor(cx) || at::functionalization::impl::isFunctionalTensor(cy) || at::functionalization::impl::isFunctionalTensor(workspace))) { |
| 11468 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11469 | TORCH_INTERNAL_ASSERT(false, |
| 11470 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11471 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11472 | } else { |
| 11473 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11474 | at::AutoDispatchSkipFunctionalize guard; |
| 11475 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_thnn_fused_lstm_cell_backward_impl_out::call(grad_hy_, grad_cy_, cx_, cy_, workspace_, has_bias, out0_, out1_, out2_); |
| 11476 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 11477 | } |
| 11478 | } else { |
| 11479 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 11480 | { |
| 11481 | at::AutoDispatchSkipFunctionalize guard; |
| 11482 | tmp_output = at::_ops::_thnn_fused_lstm_cell_backward_impl::call(grad_hy_, grad_cy_, cx_, cy_, workspace_, has_bias); |
| 11483 | } |
| 11484 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 11485 | at::functionalization::impl::commit_update(out0); |
| 11486 | at::functionalization::impl::sync(out0); |
| 11487 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 11488 | at::functionalization::impl::commit_update(out1); |
| 11489 | at::functionalization::impl::sync(out1); |
| 11490 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 11491 | at::functionalization::impl::commit_update(out2); |
| 11492 | at::functionalization::impl::sync(out2); |
| 11493 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 11494 | } |
| 11495 | } |
| 11496 | |
| 11497 | at::Tensor & masked_fill_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & mask, const at::Scalar & value, at::Tensor & out) { |
| 11498 | if (false) { |
| 11499 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11500 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11501 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11502 | auto self_meta = to_meta(self); |
| 11503 | auto mask_meta = to_meta(mask); |
| 11504 | auto out_meta = to_meta(out); |
| 11505 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11506 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11507 | at::_ops::masked_fill_Scalar_out::call(self_meta, mask_meta, value, out_meta); |
| 11508 | } |
| 11509 | |
| 11510 | at::Tensor self_; |
| 11511 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11512 | at::functionalization::impl::sync(self); |
| 11513 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11514 | } else { |
| 11515 | self_ = self; |
| 11516 | } |
| 11517 | |
| 11518 | at::Tensor mask_; |
| 11519 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 11520 | at::functionalization::impl::sync(mask); |
| 11521 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 11522 | } else { |
| 11523 | mask_ = mask; |
| 11524 | } |
| 11525 | |
| 11526 | at::Tensor out_; |
| 11527 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11528 | at::functionalization::impl::sync(out); |
| 11529 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11530 | } else { |
| 11531 | out_ = out; |
| 11532 | } |
| 11533 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11534 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(mask))) { |
| 11535 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11536 | TORCH_INTERNAL_ASSERT(false, |
| 11537 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11538 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11539 | } else { |
| 11540 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11541 | at::AutoDispatchSkipFunctionalize guard; |
| 11542 | at::Tensor tmp_output = at::_ops::masked_fill_Scalar_out::call(self_, mask_, value, out_); |
| 11543 | return out;; |
| 11544 | } |
| 11545 | } else { |
| 11546 | at::Tensor tmp_output; |
| 11547 | { |
| 11548 | at::AutoDispatchSkipFunctionalize guard; |
| 11549 | tmp_output = at::_ops::masked_fill_Scalar::call(self_, mask_, value); |
| 11550 | } |
| 11551 | at::functionalization::impl::replace_(out, tmp_output); |
| 11552 | at::functionalization::impl::commit_update(out); |
| 11553 | at::functionalization::impl::sync(out); |
| 11554 | return out; |
| 11555 | } |
| 11556 | } |
| 11557 | |
| 11558 | at::Tensor & masked_fill__Scalar(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & mask, const at::Scalar & value) { |
| 11559 | if (true) { |
| 11560 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11561 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11562 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11563 | auto self_meta = to_meta(self); |
| 11564 | auto mask_meta = to_meta(mask); |
| 11565 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11566 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11567 | at::_ops::masked_fill__Scalar::call(self_meta, mask_meta, value); |
| 11568 | } |
| 11569 | |
| 11570 | at::Tensor self_; |
| 11571 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11572 | at::functionalization::impl::sync(self); |
| 11573 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11574 | } else { |
| 11575 | self_ = self; |
| 11576 | } |
| 11577 | |
| 11578 | at::Tensor mask_; |
| 11579 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 11580 | at::functionalization::impl::sync(mask); |
| 11581 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 11582 | } else { |
| 11583 | mask_ = mask; |
| 11584 | } |
| 11585 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 11586 | if ((false || at::functionalization::impl::isFunctionalTensor(mask))) { |
| 11587 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11588 | TORCH_INTERNAL_ASSERT(false, |
| 11589 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11590 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11591 | } else { |
| 11592 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11593 | at::AutoDispatchSkipFunctionalize guard; |
| 11594 | at::Tensor tmp_output = at::_ops::masked_fill__Scalar::call(self_, mask_, value); |
| 11595 | return self;; |
| 11596 | } |
| 11597 | } else { |
| 11598 | at::Tensor tmp_output; |
| 11599 | { |
| 11600 | at::AutoDispatchSkipFunctionalize guard; |
| 11601 | tmp_output = at::_ops::masked_fill_Scalar::call(self_, mask_, value); |
| 11602 | } |
| 11603 | at::functionalization::impl::replace_(self, tmp_output); |
| 11604 | at::functionalization::impl::commit_update(self); |
| 11605 | at::functionalization::impl::sync(self); |
| 11606 | return self; |
| 11607 | } |
| 11608 | } |
| 11609 | |
| 11610 | at::Tensor & masked_fill_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & mask, const at::Tensor & value, at::Tensor & out) { |
| 11611 | if (false) { |
| 11612 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11613 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11614 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11615 | auto self_meta = to_meta(self); |
| 11616 | auto mask_meta = to_meta(mask); |
| 11617 | auto value_meta = to_meta(value); |
| 11618 | auto out_meta = to_meta(out); |
| 11619 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11620 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11621 | at::_ops::masked_fill_Tensor_out::call(self_meta, mask_meta, value_meta, out_meta); |
| 11622 | } |
| 11623 | |
| 11624 | at::Tensor self_; |
| 11625 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11626 | at::functionalization::impl::sync(self); |
| 11627 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11628 | } else { |
| 11629 | self_ = self; |
| 11630 | } |
| 11631 | |
| 11632 | at::Tensor mask_; |
| 11633 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 11634 | at::functionalization::impl::sync(mask); |
| 11635 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 11636 | } else { |
| 11637 | mask_ = mask; |
| 11638 | } |
| 11639 | |
| 11640 | at::Tensor value_; |
| 11641 | if (at::functionalization::impl::isFunctionalTensor(value)) { |
| 11642 | at::functionalization::impl::sync(value); |
| 11643 | value_ = at::functionalization::impl::from_functional_tensor(value); |
| 11644 | } else { |
| 11645 | value_ = value; |
| 11646 | } |
| 11647 | |
| 11648 | at::Tensor out_; |
| 11649 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11650 | at::functionalization::impl::sync(out); |
| 11651 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11652 | } else { |
| 11653 | out_ = out; |
| 11654 | } |
| 11655 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11656 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(mask) || at::functionalization::impl::isFunctionalTensor(value))) { |
| 11657 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11658 | TORCH_INTERNAL_ASSERT(false, |
| 11659 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11660 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11661 | } else { |
| 11662 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11663 | at::AutoDispatchSkipFunctionalize guard; |
| 11664 | at::Tensor tmp_output = at::_ops::masked_fill_Tensor_out::call(self_, mask_, value_, out_); |
| 11665 | return out;; |
| 11666 | } |
| 11667 | } else { |
| 11668 | at::Tensor tmp_output; |
| 11669 | { |
| 11670 | at::AutoDispatchSkipFunctionalize guard; |
| 11671 | tmp_output = at::_ops::masked_fill_Tensor::call(self_, mask_, value_); |
| 11672 | } |
| 11673 | at::functionalization::impl::replace_(out, tmp_output); |
| 11674 | at::functionalization::impl::commit_update(out); |
| 11675 | at::functionalization::impl::sync(out); |
| 11676 | return out; |
| 11677 | } |
| 11678 | } |
| 11679 | |
| 11680 | at::Tensor & masked_fill__Tensor(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & mask, const at::Tensor & value) { |
| 11681 | if (true) { |
| 11682 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11683 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11684 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11685 | auto self_meta = to_meta(self); |
| 11686 | auto mask_meta = to_meta(mask); |
| 11687 | auto value_meta = to_meta(value); |
| 11688 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11689 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11690 | at::_ops::masked_fill__Tensor::call(self_meta, mask_meta, value_meta); |
| 11691 | } |
| 11692 | |
| 11693 | at::Tensor self_; |
| 11694 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11695 | at::functionalization::impl::sync(self); |
| 11696 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11697 | } else { |
| 11698 | self_ = self; |
| 11699 | } |
| 11700 | |
| 11701 | at::Tensor mask_; |
| 11702 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 11703 | at::functionalization::impl::sync(mask); |
| 11704 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 11705 | } else { |
| 11706 | mask_ = mask; |
| 11707 | } |
| 11708 | |
| 11709 | at::Tensor value_; |
| 11710 | if (at::functionalization::impl::isFunctionalTensor(value)) { |
| 11711 | at::functionalization::impl::sync(value); |
| 11712 | value_ = at::functionalization::impl::from_functional_tensor(value); |
| 11713 | } else { |
| 11714 | value_ = value; |
| 11715 | } |
| 11716 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 11717 | if ((false || at::functionalization::impl::isFunctionalTensor(mask) || at::functionalization::impl::isFunctionalTensor(value))) { |
| 11718 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11719 | TORCH_INTERNAL_ASSERT(false, |
| 11720 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11721 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11722 | } else { |
| 11723 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11724 | at::AutoDispatchSkipFunctionalize guard; |
| 11725 | at::Tensor tmp_output = at::_ops::masked_fill__Tensor::call(self_, mask_, value_); |
| 11726 | return self;; |
| 11727 | } |
| 11728 | } else { |
| 11729 | at::Tensor tmp_output; |
| 11730 | { |
| 11731 | at::AutoDispatchSkipFunctionalize guard; |
| 11732 | tmp_output = at::_ops::masked_fill_Tensor::call(self_, mask_, value_); |
| 11733 | } |
| 11734 | at::functionalization::impl::replace_(self, tmp_output); |
| 11735 | at::functionalization::impl::commit_update(self); |
| 11736 | at::functionalization::impl::sync(self); |
| 11737 | return self; |
| 11738 | } |
| 11739 | } |
| 11740 | |
| 11741 | at::Tensor & _masked_softmax_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & mask, c10::optional<int64_t> dim, c10::optional<int64_t> mask_type, at::Tensor & out) { |
| 11742 | if (false) { |
| 11743 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11744 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11745 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11746 | auto self_meta = to_meta(self); |
| 11747 | auto mask_meta = to_meta(mask); |
| 11748 | auto out_meta = to_meta(out); |
| 11749 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11750 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11751 | at::_ops::_masked_softmax_out::call(self_meta, mask_meta, dim, mask_type, out_meta); |
| 11752 | } |
| 11753 | |
| 11754 | at::Tensor self_; |
| 11755 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11756 | at::functionalization::impl::sync(self); |
| 11757 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11758 | } else { |
| 11759 | self_ = self; |
| 11760 | } |
| 11761 | |
| 11762 | at::Tensor mask_; |
| 11763 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 11764 | at::functionalization::impl::sync(mask); |
| 11765 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 11766 | } else { |
| 11767 | mask_ = mask; |
| 11768 | } |
| 11769 | |
| 11770 | at::Tensor out_; |
| 11771 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11772 | at::functionalization::impl::sync(out); |
| 11773 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11774 | } else { |
| 11775 | out_ = out; |
| 11776 | } |
| 11777 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11778 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(mask))) { |
| 11779 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11780 | TORCH_INTERNAL_ASSERT(false, |
| 11781 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11782 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11783 | } else { |
| 11784 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11785 | at::AutoDispatchSkipFunctionalize guard; |
| 11786 | at::Tensor tmp_output = at::_ops::_masked_softmax_out::call(self_, mask_, dim, mask_type, out_); |
| 11787 | return out;; |
| 11788 | } |
| 11789 | } else { |
| 11790 | at::Tensor tmp_output; |
| 11791 | { |
| 11792 | at::AutoDispatchSkipFunctionalize guard; |
| 11793 | tmp_output = at::_ops::_masked_softmax::call(self_, mask_, dim, mask_type); |
| 11794 | } |
| 11795 | at::functionalization::impl::replace_(out, tmp_output); |
| 11796 | at::functionalization::impl::commit_update(out); |
| 11797 | at::functionalization::impl::sync(out); |
| 11798 | return out; |
| 11799 | } |
| 11800 | } |
| 11801 | |
| 11802 | at::Tensor & bitwise_right_shift_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 11803 | if (false) { |
| 11804 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11805 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11806 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11807 | auto self_meta = to_meta(self); |
| 11808 | auto other_meta = to_meta(other); |
| 11809 | auto out_meta = to_meta(out); |
| 11810 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11811 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11812 | at::_ops::bitwise_right_shift_Tensor_out::call(self_meta, other_meta, out_meta); |
| 11813 | } |
| 11814 | |
| 11815 | at::Tensor self_; |
| 11816 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11817 | at::functionalization::impl::sync(self); |
| 11818 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11819 | } else { |
| 11820 | self_ = self; |
| 11821 | } |
| 11822 | |
| 11823 | at::Tensor other_; |
| 11824 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 11825 | at::functionalization::impl::sync(other); |
| 11826 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 11827 | } else { |
| 11828 | other_ = other; |
| 11829 | } |
| 11830 | |
| 11831 | at::Tensor out_; |
| 11832 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11833 | at::functionalization::impl::sync(out); |
| 11834 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11835 | } else { |
| 11836 | out_ = out; |
| 11837 | } |
| 11838 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11839 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 11840 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11841 | TORCH_INTERNAL_ASSERT(false, |
| 11842 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11843 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11844 | } else { |
| 11845 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11846 | at::AutoDispatchSkipFunctionalize guard; |
| 11847 | at::Tensor tmp_output = at::_ops::bitwise_right_shift_Tensor_out::call(self_, other_, out_); |
| 11848 | return out;; |
| 11849 | } |
| 11850 | } else { |
| 11851 | at::Tensor tmp_output; |
| 11852 | { |
| 11853 | at::AutoDispatchSkipFunctionalize guard; |
| 11854 | tmp_output = at::_ops::bitwise_right_shift_Tensor::call(self_, other_); |
| 11855 | } |
| 11856 | at::functionalization::impl::replace_(out, tmp_output); |
| 11857 | at::functionalization::impl::commit_update(out); |
| 11858 | at::functionalization::impl::sync(out); |
| 11859 | return out; |
| 11860 | } |
| 11861 | } |
| 11862 | |
| 11863 | at::Tensor & bitwise_right_shift__Tensor(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 11864 | if (true) { |
| 11865 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11866 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11867 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11868 | auto self_meta = to_meta(self); |
| 11869 | auto other_meta = to_meta(other); |
| 11870 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11871 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11872 | at::_ops::bitwise_right_shift__Tensor::call(self_meta, other_meta); |
| 11873 | } |
| 11874 | |
| 11875 | at::Tensor self_; |
| 11876 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11877 | at::functionalization::impl::sync(self); |
| 11878 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11879 | } else { |
| 11880 | self_ = self; |
| 11881 | } |
| 11882 | |
| 11883 | at::Tensor other_; |
| 11884 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 11885 | at::functionalization::impl::sync(other); |
| 11886 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 11887 | } else { |
| 11888 | other_ = other; |
| 11889 | } |
| 11890 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 11891 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 11892 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11893 | TORCH_INTERNAL_ASSERT(false, |
| 11894 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11895 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11896 | } else { |
| 11897 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11898 | at::AutoDispatchSkipFunctionalize guard; |
| 11899 | at::Tensor tmp_output = at::_ops::bitwise_right_shift__Tensor::call(self_, other_); |
| 11900 | return self;; |
| 11901 | } |
| 11902 | } else { |
| 11903 | at::Tensor tmp_output; |
| 11904 | { |
| 11905 | at::AutoDispatchSkipFunctionalize guard; |
| 11906 | tmp_output = at::_ops::bitwise_right_shift_Tensor::call(self_, other_); |
| 11907 | } |
| 11908 | at::functionalization::impl::replace_(self, tmp_output); |
| 11909 | at::functionalization::impl::commit_update(self); |
| 11910 | at::functionalization::impl::sync(self); |
| 11911 | return self; |
| 11912 | } |
| 11913 | } |
| 11914 | |
| 11915 | at::Tensor & bitwise_right_shift_out_Tensor_Scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & other, at::Tensor & out) { |
| 11916 | if (false) { |
| 11917 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11918 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11919 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11920 | auto self_meta = to_meta(self); |
| 11921 | auto out_meta = to_meta(out); |
| 11922 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11923 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11924 | at::_ops::bitwise_right_shift_Tensor_Scalar_out::call(self_meta, other, out_meta); |
| 11925 | } |
| 11926 | |
| 11927 | at::Tensor self_; |
| 11928 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11929 | at::functionalization::impl::sync(self); |
| 11930 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11931 | } else { |
| 11932 | self_ = self; |
| 11933 | } |
| 11934 | |
| 11935 | at::Tensor out_; |
| 11936 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 11937 | at::functionalization::impl::sync(out); |
| 11938 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 11939 | } else { |
| 11940 | out_ = out; |
| 11941 | } |
| 11942 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 11943 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 11944 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11945 | TORCH_INTERNAL_ASSERT(false, |
| 11946 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11947 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11948 | } else { |
| 11949 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11950 | at::AutoDispatchSkipFunctionalize guard; |
| 11951 | at::Tensor tmp_output = at::_ops::bitwise_right_shift_Tensor_Scalar_out::call(self_, other, out_); |
| 11952 | return out;; |
| 11953 | } |
| 11954 | } else { |
| 11955 | at::Tensor tmp_output; |
| 11956 | { |
| 11957 | at::AutoDispatchSkipFunctionalize guard; |
| 11958 | tmp_output = at::_ops::bitwise_right_shift_Tensor_Scalar::call(self_, other); |
| 11959 | } |
| 11960 | at::functionalization::impl::replace_(out, tmp_output); |
| 11961 | at::functionalization::impl::commit_update(out); |
| 11962 | at::functionalization::impl::sync(out); |
| 11963 | return out; |
| 11964 | } |
| 11965 | } |
| 11966 | |
| 11967 | at::Tensor & bitwise_right_shift__Tensor_Scalar(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & other) { |
| 11968 | if (true) { |
| 11969 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 11970 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 11971 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 11972 | auto self_meta = to_meta(self); |
| 11973 | at::AutoDispatchSkipFunctionalize func_guard; |
| 11974 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 11975 | at::_ops::bitwise_right_shift__Tensor_Scalar::call(self_meta, other); |
| 11976 | } |
| 11977 | |
| 11978 | at::Tensor self_; |
| 11979 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 11980 | at::functionalization::impl::sync(self); |
| 11981 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 11982 | } else { |
| 11983 | self_ = self; |
| 11984 | } |
| 11985 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 11986 | if ((false)) { |
| 11987 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 11988 | TORCH_INTERNAL_ASSERT(false, |
| 11989 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 11990 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 11991 | } else { |
| 11992 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 11993 | at::AutoDispatchSkipFunctionalize guard; |
| 11994 | at::Tensor tmp_output = at::_ops::bitwise_right_shift__Tensor_Scalar::call(self_, other); |
| 11995 | return self;; |
| 11996 | } |
| 11997 | } else { |
| 11998 | at::Tensor tmp_output; |
| 11999 | { |
| 12000 | at::AutoDispatchSkipFunctionalize guard; |
| 12001 | tmp_output = at::_ops::bitwise_right_shift_Tensor_Scalar::call(self_, other); |
| 12002 | } |
| 12003 | at::functionalization::impl::replace_(self, tmp_output); |
| 12004 | at::functionalization::impl::commit_update(self); |
| 12005 | at::functionalization::impl::sync(self); |
| 12006 | return self; |
| 12007 | } |
| 12008 | } |
| 12009 | |
| 12010 | at::Tensor & bitwise_right_shift_out_Scalar_Tensor_out(c10::DispatchKeySet dispatchKeySet, const at::Scalar & self, const at::Tensor & other, at::Tensor & out) { |
| 12011 | if (false) { |
| 12012 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12013 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12014 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12015 | auto other_meta = to_meta(other); |
| 12016 | auto out_meta = to_meta(out); |
| 12017 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12018 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12019 | at::_ops::bitwise_right_shift_Scalar_Tensor_out::call(self, other_meta, out_meta); |
| 12020 | } |
| 12021 | |
| 12022 | at::Tensor other_; |
| 12023 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 12024 | at::functionalization::impl::sync(other); |
| 12025 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 12026 | } else { |
| 12027 | other_ = other; |
| 12028 | } |
| 12029 | |
| 12030 | at::Tensor out_; |
| 12031 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12032 | at::functionalization::impl::sync(out); |
| 12033 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12034 | } else { |
| 12035 | out_ = out; |
| 12036 | } |
| 12037 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12038 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 12039 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12040 | TORCH_INTERNAL_ASSERT(false, |
| 12041 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12042 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12043 | } else { |
| 12044 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12045 | at::AutoDispatchSkipFunctionalize guard; |
| 12046 | at::Tensor tmp_output = at::_ops::bitwise_right_shift_Scalar_Tensor_out::call(self, other_, out_); |
| 12047 | return out;; |
| 12048 | } |
| 12049 | } else { |
| 12050 | at::Tensor tmp_output; |
| 12051 | { |
| 12052 | at::AutoDispatchSkipFunctionalize guard; |
| 12053 | tmp_output = at::_ops::bitwise_right_shift_Scalar_Tensor::call(self, other_); |
| 12054 | } |
| 12055 | at::functionalization::impl::replace_(out, tmp_output); |
| 12056 | at::functionalization::impl::commit_update(out); |
| 12057 | at::functionalization::impl::sync(out); |
| 12058 | return out; |
| 12059 | } |
| 12060 | } |
| 12061 | |
| 12062 | at::Tensor & cauchy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, double median, double sigma, c10::optional<at::Generator> generator, at::Tensor & out) { |
| 12063 | if (false) { |
| 12064 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12065 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12066 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12067 | auto self_meta = to_meta(self); |
| 12068 | auto out_meta = to_meta(out); |
| 12069 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12070 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12071 | at::_ops::cauchy_out::call(self_meta, median, sigma, generator, out_meta); |
| 12072 | } |
| 12073 | |
| 12074 | at::Tensor self_; |
| 12075 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12076 | at::functionalization::impl::sync(self); |
| 12077 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12078 | } else { |
| 12079 | self_ = self; |
| 12080 | } |
| 12081 | |
| 12082 | at::Tensor out_; |
| 12083 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12084 | at::functionalization::impl::sync(out); |
| 12085 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12086 | } else { |
| 12087 | out_ = out; |
| 12088 | } |
| 12089 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12090 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 12091 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12092 | TORCH_INTERNAL_ASSERT(false, |
| 12093 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12094 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12095 | } else { |
| 12096 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12097 | at::AutoDispatchSkipFunctionalize guard; |
| 12098 | at::Tensor tmp_output = at::_ops::cauchy_out::call(self_, median, sigma, generator, out_); |
| 12099 | return out;; |
| 12100 | } |
| 12101 | } else { |
| 12102 | at::Tensor tmp_output; |
| 12103 | { |
| 12104 | at::AutoDispatchSkipFunctionalize guard; |
| 12105 | tmp_output = at::_ops::cauchy::call(self_, median, sigma, generator); |
| 12106 | } |
| 12107 | at::functionalization::impl::replace_(out, tmp_output); |
| 12108 | at::functionalization::impl::commit_update(out); |
| 12109 | at::functionalization::impl::sync(out); |
| 12110 | return out; |
| 12111 | } |
| 12112 | } |
| 12113 | |
| 12114 | at::Tensor & cauchy_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, double median, double sigma, c10::optional<at::Generator> generator) { |
| 12115 | if (true) { |
| 12116 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12117 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12118 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12119 | auto self_meta = to_meta(self); |
| 12120 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12121 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12122 | at::_ops::cauchy_::call(self_meta, median, sigma, generator); |
| 12123 | } |
| 12124 | |
| 12125 | at::Tensor self_; |
| 12126 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12127 | at::functionalization::impl::sync(self); |
| 12128 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12129 | } else { |
| 12130 | self_ = self; |
| 12131 | } |
| 12132 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 12133 | if ((false)) { |
| 12134 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12135 | TORCH_INTERNAL_ASSERT(false, |
| 12136 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12137 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12138 | } else { |
| 12139 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12140 | at::AutoDispatchSkipFunctionalize guard; |
| 12141 | at::Tensor tmp_output = at::_ops::cauchy_::call(self_, median, sigma, generator); |
| 12142 | return self;; |
| 12143 | } |
| 12144 | } else { |
| 12145 | at::Tensor tmp_output; |
| 12146 | { |
| 12147 | at::AutoDispatchSkipFunctionalize guard; |
| 12148 | tmp_output = at::_ops::cauchy::call(self_, median, sigma, generator); |
| 12149 | } |
| 12150 | at::functionalization::impl::replace_(self, tmp_output); |
| 12151 | at::functionalization::impl::commit_update(self); |
| 12152 | at::functionalization::impl::sync(self); |
| 12153 | return self; |
| 12154 | } |
| 12155 | } |
| 12156 | |
| 12157 | at::Tensor & log_normal_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, double mean, double std, c10::optional<at::Generator> generator, at::Tensor & out) { |
| 12158 | if (false) { |
| 12159 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12160 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12161 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12162 | auto self_meta = to_meta(self); |
| 12163 | auto out_meta = to_meta(out); |
| 12164 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12165 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12166 | at::_ops::log_normal_out::call(self_meta, mean, std, generator, out_meta); |
| 12167 | } |
| 12168 | |
| 12169 | at::Tensor self_; |
| 12170 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12171 | at::functionalization::impl::sync(self); |
| 12172 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12173 | } else { |
| 12174 | self_ = self; |
| 12175 | } |
| 12176 | |
| 12177 | at::Tensor out_; |
| 12178 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12179 | at::functionalization::impl::sync(out); |
| 12180 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12181 | } else { |
| 12182 | out_ = out; |
| 12183 | } |
| 12184 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12185 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 12186 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12187 | TORCH_INTERNAL_ASSERT(false, |
| 12188 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12189 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12190 | } else { |
| 12191 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12192 | at::AutoDispatchSkipFunctionalize guard; |
| 12193 | at::Tensor tmp_output = at::_ops::log_normal_out::call(self_, mean, std, generator, out_); |
| 12194 | return out;; |
| 12195 | } |
| 12196 | } else { |
| 12197 | at::Tensor tmp_output; |
| 12198 | { |
| 12199 | at::AutoDispatchSkipFunctionalize guard; |
| 12200 | tmp_output = at::_ops::log_normal::call(self_, mean, std, generator); |
| 12201 | } |
| 12202 | at::functionalization::impl::replace_(out, tmp_output); |
| 12203 | at::functionalization::impl::commit_update(out); |
| 12204 | at::functionalization::impl::sync(out); |
| 12205 | return out; |
| 12206 | } |
| 12207 | } |
| 12208 | |
| 12209 | at::Tensor & log_normal_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, double mean, double std, c10::optional<at::Generator> generator) { |
| 12210 | if (true) { |
| 12211 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12212 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12213 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12214 | auto self_meta = to_meta(self); |
| 12215 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12216 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12217 | at::_ops::log_normal_::call(self_meta, mean, std, generator); |
| 12218 | } |
| 12219 | |
| 12220 | at::Tensor self_; |
| 12221 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12222 | at::functionalization::impl::sync(self); |
| 12223 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12224 | } else { |
| 12225 | self_ = self; |
| 12226 | } |
| 12227 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 12228 | if ((false)) { |
| 12229 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12230 | TORCH_INTERNAL_ASSERT(false, |
| 12231 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12232 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12233 | } else { |
| 12234 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12235 | at::AutoDispatchSkipFunctionalize guard; |
| 12236 | at::Tensor tmp_output = at::_ops::log_normal_::call(self_, mean, std, generator); |
| 12237 | return self;; |
| 12238 | } |
| 12239 | } else { |
| 12240 | at::Tensor tmp_output; |
| 12241 | { |
| 12242 | at::AutoDispatchSkipFunctionalize guard; |
| 12243 | tmp_output = at::_ops::log_normal::call(self_, mean, std, generator); |
| 12244 | } |
| 12245 | at::functionalization::impl::replace_(self, tmp_output); |
| 12246 | at::functionalization::impl::commit_update(self); |
| 12247 | at::functionalization::impl::sync(self); |
| 12248 | return self; |
| 12249 | } |
| 12250 | } |
| 12251 | |
| 12252 | at::Tensor & diag_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t diagonal, at::Tensor & out) { |
| 12253 | if (false) { |
| 12254 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12255 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12256 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12257 | auto self_meta = to_meta(self); |
| 12258 | auto out_meta = to_meta(out); |
| 12259 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12260 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12261 | at::_ops::diag_out::call(self_meta, diagonal, out_meta); |
| 12262 | } |
| 12263 | |
| 12264 | at::Tensor self_; |
| 12265 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12266 | at::functionalization::impl::sync(self); |
| 12267 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12268 | } else { |
| 12269 | self_ = self; |
| 12270 | } |
| 12271 | |
| 12272 | at::Tensor out_; |
| 12273 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12274 | at::functionalization::impl::sync(out); |
| 12275 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12276 | } else { |
| 12277 | out_ = out; |
| 12278 | } |
| 12279 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12280 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 12281 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12282 | TORCH_INTERNAL_ASSERT(false, |
| 12283 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12284 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12285 | } else { |
| 12286 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12287 | at::AutoDispatchSkipFunctionalize guard; |
| 12288 | at::Tensor tmp_output = at::_ops::diag_out::call(self_, diagonal, out_); |
| 12289 | return out;; |
| 12290 | } |
| 12291 | } else { |
| 12292 | at::Tensor tmp_output; |
| 12293 | { |
| 12294 | at::AutoDispatchSkipFunctionalize guard; |
| 12295 | tmp_output = at::_ops::diag::call(self_, diagonal); |
| 12296 | } |
| 12297 | at::functionalization::impl::replace_(out, tmp_output); |
| 12298 | at::functionalization::impl::commit_update(out); |
| 12299 | at::functionalization::impl::sync(out); |
| 12300 | return out; |
| 12301 | } |
| 12302 | } |
| 12303 | |
| 12304 | at::Tensor & ne_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & other, at::Tensor & out) { |
| 12305 | if (false) { |
| 12306 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12307 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12308 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12309 | auto self_meta = to_meta(self); |
| 12310 | auto out_meta = to_meta(out); |
| 12311 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12312 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12313 | at::_ops::ne_Scalar_out::call(self_meta, other, out_meta); |
| 12314 | } |
| 12315 | |
| 12316 | at::Tensor self_; |
| 12317 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12318 | at::functionalization::impl::sync(self); |
| 12319 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12320 | } else { |
| 12321 | self_ = self; |
| 12322 | } |
| 12323 | |
| 12324 | at::Tensor out_; |
| 12325 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12326 | at::functionalization::impl::sync(out); |
| 12327 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12328 | } else { |
| 12329 | out_ = out; |
| 12330 | } |
| 12331 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12332 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 12333 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12334 | TORCH_INTERNAL_ASSERT(false, |
| 12335 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12336 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12337 | } else { |
| 12338 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12339 | at::AutoDispatchSkipFunctionalize guard; |
| 12340 | at::Tensor tmp_output = at::_ops::ne_Scalar_out::call(self_, other, out_); |
| 12341 | return out;; |
| 12342 | } |
| 12343 | } else { |
| 12344 | at::Tensor tmp_output; |
| 12345 | { |
| 12346 | at::AutoDispatchSkipFunctionalize guard; |
| 12347 | tmp_output = at::_ops::ne_Scalar::call(self_, other); |
| 12348 | } |
| 12349 | at::functionalization::impl::replace_(out, tmp_output); |
| 12350 | at::functionalization::impl::commit_update(out); |
| 12351 | at::functionalization::impl::sync(out); |
| 12352 | return out; |
| 12353 | } |
| 12354 | } |
| 12355 | |
| 12356 | at::Tensor & ne__Scalar(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & other) { |
| 12357 | if (true) { |
| 12358 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12359 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12360 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12361 | auto self_meta = to_meta(self); |
| 12362 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12363 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12364 | at::_ops::ne__Scalar::call(self_meta, other); |
| 12365 | } |
| 12366 | |
| 12367 | at::Tensor self_; |
| 12368 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12369 | at::functionalization::impl::sync(self); |
| 12370 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12371 | } else { |
| 12372 | self_ = self; |
| 12373 | } |
| 12374 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 12375 | if ((false)) { |
| 12376 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12377 | TORCH_INTERNAL_ASSERT(false, |
| 12378 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12379 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12380 | } else { |
| 12381 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12382 | at::AutoDispatchSkipFunctionalize guard; |
| 12383 | at::Tensor tmp_output = at::_ops::ne__Scalar::call(self_, other); |
| 12384 | return self;; |
| 12385 | } |
| 12386 | } else { |
| 12387 | at::Tensor tmp_output; |
| 12388 | { |
| 12389 | at::AutoDispatchSkipFunctionalize guard; |
| 12390 | tmp_output = at::_ops::ne_Scalar::call(self_, other); |
| 12391 | } |
| 12392 | at::functionalization::impl::replace_(self, tmp_output); |
| 12393 | at::functionalization::impl::commit_update(self); |
| 12394 | at::functionalization::impl::sync(self); |
| 12395 | return self; |
| 12396 | } |
| 12397 | } |
| 12398 | |
| 12399 | at::Tensor & ne_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 12400 | if (false) { |
| 12401 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12402 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12403 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12404 | auto self_meta = to_meta(self); |
| 12405 | auto other_meta = to_meta(other); |
| 12406 | auto out_meta = to_meta(out); |
| 12407 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12408 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12409 | at::_ops::ne_Tensor_out::call(self_meta, other_meta, out_meta); |
| 12410 | } |
| 12411 | |
| 12412 | at::Tensor self_; |
| 12413 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12414 | at::functionalization::impl::sync(self); |
| 12415 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12416 | } else { |
| 12417 | self_ = self; |
| 12418 | } |
| 12419 | |
| 12420 | at::Tensor other_; |
| 12421 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 12422 | at::functionalization::impl::sync(other); |
| 12423 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 12424 | } else { |
| 12425 | other_ = other; |
| 12426 | } |
| 12427 | |
| 12428 | at::Tensor out_; |
| 12429 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12430 | at::functionalization::impl::sync(out); |
| 12431 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12432 | } else { |
| 12433 | out_ = out; |
| 12434 | } |
| 12435 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12436 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 12437 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12438 | TORCH_INTERNAL_ASSERT(false, |
| 12439 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12440 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12441 | } else { |
| 12442 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12443 | at::AutoDispatchSkipFunctionalize guard; |
| 12444 | at::Tensor tmp_output = at::_ops::ne_Tensor_out::call(self_, other_, out_); |
| 12445 | return out;; |
| 12446 | } |
| 12447 | } else { |
| 12448 | at::Tensor tmp_output; |
| 12449 | { |
| 12450 | at::AutoDispatchSkipFunctionalize guard; |
| 12451 | tmp_output = at::_ops::ne_Tensor::call(self_, other_); |
| 12452 | } |
| 12453 | at::functionalization::impl::replace_(out, tmp_output); |
| 12454 | at::functionalization::impl::commit_update(out); |
| 12455 | at::functionalization::impl::sync(out); |
| 12456 | return out; |
| 12457 | } |
| 12458 | } |
| 12459 | |
| 12460 | at::Tensor & ne__Tensor(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 12461 | if (true) { |
| 12462 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12463 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12464 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12465 | auto self_meta = to_meta(self); |
| 12466 | auto other_meta = to_meta(other); |
| 12467 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12468 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12469 | at::_ops::ne__Tensor::call(self_meta, other_meta); |
| 12470 | } |
| 12471 | |
| 12472 | at::Tensor self_; |
| 12473 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12474 | at::functionalization::impl::sync(self); |
| 12475 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12476 | } else { |
| 12477 | self_ = self; |
| 12478 | } |
| 12479 | |
| 12480 | at::Tensor other_; |
| 12481 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 12482 | at::functionalization::impl::sync(other); |
| 12483 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 12484 | } else { |
| 12485 | other_ = other; |
| 12486 | } |
| 12487 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 12488 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 12489 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12490 | TORCH_INTERNAL_ASSERT(false, |
| 12491 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12492 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12493 | } else { |
| 12494 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12495 | at::AutoDispatchSkipFunctionalize guard; |
| 12496 | at::Tensor tmp_output = at::_ops::ne__Tensor::call(self_, other_); |
| 12497 | return self;; |
| 12498 | } |
| 12499 | } else { |
| 12500 | at::Tensor tmp_output; |
| 12501 | { |
| 12502 | at::AutoDispatchSkipFunctionalize guard; |
| 12503 | tmp_output = at::_ops::ne_Tensor::call(self_, other_); |
| 12504 | } |
| 12505 | at::functionalization::impl::replace_(self, tmp_output); |
| 12506 | at::functionalization::impl::commit_update(self); |
| 12507 | at::functionalization::impl::sync(self); |
| 12508 | return self; |
| 12509 | } |
| 12510 | } |
| 12511 | |
| 12512 | at::Tensor & not_equal_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & other, at::Tensor & out) { |
| 12513 | if (false) { |
| 12514 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12515 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12516 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12517 | auto self_meta = to_meta(self); |
| 12518 | auto out_meta = to_meta(out); |
| 12519 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12520 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12521 | at::_ops::not_equal_Scalar_out::call(self_meta, other, out_meta); |
| 12522 | } |
| 12523 | |
| 12524 | at::Tensor self_; |
| 12525 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12526 | at::functionalization::impl::sync(self); |
| 12527 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12528 | } else { |
| 12529 | self_ = self; |
| 12530 | } |
| 12531 | |
| 12532 | at::Tensor out_; |
| 12533 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12534 | at::functionalization::impl::sync(out); |
| 12535 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12536 | } else { |
| 12537 | out_ = out; |
| 12538 | } |
| 12539 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12540 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 12541 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12542 | TORCH_INTERNAL_ASSERT(false, |
| 12543 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12544 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12545 | } else { |
| 12546 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12547 | at::AutoDispatchSkipFunctionalize guard; |
| 12548 | at::Tensor tmp_output = at::_ops::not_equal_Scalar_out::call(self_, other, out_); |
| 12549 | return out;; |
| 12550 | } |
| 12551 | } else { |
| 12552 | at::Tensor tmp_output; |
| 12553 | { |
| 12554 | at::AutoDispatchSkipFunctionalize guard; |
| 12555 | tmp_output = at::_ops::not_equal_Scalar::call(self_, other); |
| 12556 | } |
| 12557 | at::functionalization::impl::replace_(out, tmp_output); |
| 12558 | at::functionalization::impl::commit_update(out); |
| 12559 | at::functionalization::impl::sync(out); |
| 12560 | return out; |
| 12561 | } |
| 12562 | } |
| 12563 | |
| 12564 | at::Tensor & not_equal__Scalar(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & other) { |
| 12565 | if (true) { |
| 12566 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12567 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12568 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12569 | auto self_meta = to_meta(self); |
| 12570 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12571 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12572 | at::_ops::not_equal__Scalar::call(self_meta, other); |
| 12573 | } |
| 12574 | |
| 12575 | at::Tensor self_; |
| 12576 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12577 | at::functionalization::impl::sync(self); |
| 12578 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12579 | } else { |
| 12580 | self_ = self; |
| 12581 | } |
| 12582 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 12583 | if ((false)) { |
| 12584 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12585 | TORCH_INTERNAL_ASSERT(false, |
| 12586 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12587 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12588 | } else { |
| 12589 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12590 | at::AutoDispatchSkipFunctionalize guard; |
| 12591 | at::Tensor tmp_output = at::_ops::not_equal__Scalar::call(self_, other); |
| 12592 | return self;; |
| 12593 | } |
| 12594 | } else { |
| 12595 | at::Tensor tmp_output; |
| 12596 | { |
| 12597 | at::AutoDispatchSkipFunctionalize guard; |
| 12598 | tmp_output = at::_ops::not_equal_Scalar::call(self_, other); |
| 12599 | } |
| 12600 | at::functionalization::impl::replace_(self, tmp_output); |
| 12601 | at::functionalization::impl::commit_update(self); |
| 12602 | at::functionalization::impl::sync(self); |
| 12603 | return self; |
| 12604 | } |
| 12605 | } |
| 12606 | |
| 12607 | at::Tensor & not_equal_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 12608 | if (false) { |
| 12609 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12610 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12611 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12612 | auto self_meta = to_meta(self); |
| 12613 | auto other_meta = to_meta(other); |
| 12614 | auto out_meta = to_meta(out); |
| 12615 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12616 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12617 | at::_ops::not_equal_Tensor_out::call(self_meta, other_meta, out_meta); |
| 12618 | } |
| 12619 | |
| 12620 | at::Tensor self_; |
| 12621 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12622 | at::functionalization::impl::sync(self); |
| 12623 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12624 | } else { |
| 12625 | self_ = self; |
| 12626 | } |
| 12627 | |
| 12628 | at::Tensor other_; |
| 12629 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 12630 | at::functionalization::impl::sync(other); |
| 12631 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 12632 | } else { |
| 12633 | other_ = other; |
| 12634 | } |
| 12635 | |
| 12636 | at::Tensor out_; |
| 12637 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12638 | at::functionalization::impl::sync(out); |
| 12639 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12640 | } else { |
| 12641 | out_ = out; |
| 12642 | } |
| 12643 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12644 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 12645 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12646 | TORCH_INTERNAL_ASSERT(false, |
| 12647 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12648 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12649 | } else { |
| 12650 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12651 | at::AutoDispatchSkipFunctionalize guard; |
| 12652 | at::Tensor tmp_output = at::_ops::not_equal_Tensor_out::call(self_, other_, out_); |
| 12653 | return out;; |
| 12654 | } |
| 12655 | } else { |
| 12656 | at::Tensor tmp_output; |
| 12657 | { |
| 12658 | at::AutoDispatchSkipFunctionalize guard; |
| 12659 | tmp_output = at::_ops::not_equal_Tensor::call(self_, other_); |
| 12660 | } |
| 12661 | at::functionalization::impl::replace_(out, tmp_output); |
| 12662 | at::functionalization::impl::commit_update(out); |
| 12663 | at::functionalization::impl::sync(out); |
| 12664 | return out; |
| 12665 | } |
| 12666 | } |
| 12667 | |
| 12668 | at::Tensor & not_equal__Tensor(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 12669 | if (true) { |
| 12670 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12671 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12672 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12673 | auto self_meta = to_meta(self); |
| 12674 | auto other_meta = to_meta(other); |
| 12675 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12676 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12677 | at::_ops::not_equal__Tensor::call(self_meta, other_meta); |
| 12678 | } |
| 12679 | |
| 12680 | at::Tensor self_; |
| 12681 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12682 | at::functionalization::impl::sync(self); |
| 12683 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12684 | } else { |
| 12685 | self_ = self; |
| 12686 | } |
| 12687 | |
| 12688 | at::Tensor other_; |
| 12689 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 12690 | at::functionalization::impl::sync(other); |
| 12691 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 12692 | } else { |
| 12693 | other_ = other; |
| 12694 | } |
| 12695 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 12696 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 12697 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12698 | TORCH_INTERNAL_ASSERT(false, |
| 12699 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12700 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12701 | } else { |
| 12702 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12703 | at::AutoDispatchSkipFunctionalize guard; |
| 12704 | at::Tensor tmp_output = at::_ops::not_equal__Tensor::call(self_, other_); |
| 12705 | return self;; |
| 12706 | } |
| 12707 | } else { |
| 12708 | at::Tensor tmp_output; |
| 12709 | { |
| 12710 | at::AutoDispatchSkipFunctionalize guard; |
| 12711 | tmp_output = at::_ops::not_equal_Tensor::call(self_, other_); |
| 12712 | } |
| 12713 | at::functionalization::impl::replace_(self, tmp_output); |
| 12714 | at::functionalization::impl::commit_update(self); |
| 12715 | at::functionalization::impl::sync(self); |
| 12716 | return self; |
| 12717 | } |
| 12718 | } |
| 12719 | |
| 12720 | at::Tensor & addcmul_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & tensor1, const at::Tensor & tensor2, const at::Scalar & value, at::Tensor & out) { |
| 12721 | if (false) { |
| 12722 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12723 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12724 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12725 | auto self_meta = to_meta(self); |
| 12726 | auto tensor1_meta = to_meta(tensor1); |
| 12727 | auto tensor2_meta = to_meta(tensor2); |
| 12728 | auto out_meta = to_meta(out); |
| 12729 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12730 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12731 | at::_ops::addcmul_out::call(self_meta, tensor1_meta, tensor2_meta, value, out_meta); |
| 12732 | } |
| 12733 | |
| 12734 | at::Tensor self_; |
| 12735 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12736 | at::functionalization::impl::sync(self); |
| 12737 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12738 | } else { |
| 12739 | self_ = self; |
| 12740 | } |
| 12741 | |
| 12742 | at::Tensor tensor1_; |
| 12743 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 12744 | at::functionalization::impl::sync(tensor1); |
| 12745 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 12746 | } else { |
| 12747 | tensor1_ = tensor1; |
| 12748 | } |
| 12749 | |
| 12750 | at::Tensor tensor2_; |
| 12751 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 12752 | at::functionalization::impl::sync(tensor2); |
| 12753 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 12754 | } else { |
| 12755 | tensor2_ = tensor2; |
| 12756 | } |
| 12757 | |
| 12758 | at::Tensor out_; |
| 12759 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12760 | at::functionalization::impl::sync(out); |
| 12761 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12762 | } else { |
| 12763 | out_ = out; |
| 12764 | } |
| 12765 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12766 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 12767 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12768 | TORCH_INTERNAL_ASSERT(false, |
| 12769 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12770 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12771 | } else { |
| 12772 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12773 | at::AutoDispatchSkipFunctionalize guard; |
| 12774 | at::Tensor tmp_output = at::_ops::addcmul_out::call(self_, tensor1_, tensor2_, value, out_); |
| 12775 | return out;; |
| 12776 | } |
| 12777 | } else { |
| 12778 | at::Tensor tmp_output; |
| 12779 | { |
| 12780 | at::AutoDispatchSkipFunctionalize guard; |
| 12781 | tmp_output = at::_ops::addcmul::call(self_, tensor1_, tensor2_, value); |
| 12782 | } |
| 12783 | at::functionalization::impl::replace_(out, tmp_output); |
| 12784 | at::functionalization::impl::commit_update(out); |
| 12785 | at::functionalization::impl::sync(out); |
| 12786 | return out; |
| 12787 | } |
| 12788 | } |
| 12789 | |
| 12790 | at::Tensor & addcmul_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & tensor1, const at::Tensor & tensor2, const at::Scalar & value) { |
| 12791 | if (true) { |
| 12792 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12793 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12794 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12795 | auto self_meta = to_meta(self); |
| 12796 | auto tensor1_meta = to_meta(tensor1); |
| 12797 | auto tensor2_meta = to_meta(tensor2); |
| 12798 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12799 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12800 | at::_ops::addcmul_::call(self_meta, tensor1_meta, tensor2_meta, value); |
| 12801 | } |
| 12802 | |
| 12803 | at::Tensor self_; |
| 12804 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12805 | at::functionalization::impl::sync(self); |
| 12806 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12807 | } else { |
| 12808 | self_ = self; |
| 12809 | } |
| 12810 | |
| 12811 | at::Tensor tensor1_; |
| 12812 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 12813 | at::functionalization::impl::sync(tensor1); |
| 12814 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 12815 | } else { |
| 12816 | tensor1_ = tensor1; |
| 12817 | } |
| 12818 | |
| 12819 | at::Tensor tensor2_; |
| 12820 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 12821 | at::functionalization::impl::sync(tensor2); |
| 12822 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 12823 | } else { |
| 12824 | tensor2_ = tensor2; |
| 12825 | } |
| 12826 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 12827 | if ((false || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 12828 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12829 | TORCH_INTERNAL_ASSERT(false, |
| 12830 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12831 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12832 | } else { |
| 12833 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12834 | at::AutoDispatchSkipFunctionalize guard; |
| 12835 | at::Tensor tmp_output = at::_ops::addcmul_::call(self_, tensor1_, tensor2_, value); |
| 12836 | return self;; |
| 12837 | } |
| 12838 | } else { |
| 12839 | at::Tensor tmp_output; |
| 12840 | { |
| 12841 | at::AutoDispatchSkipFunctionalize guard; |
| 12842 | tmp_output = at::_ops::addcmul::call(self_, tensor1_, tensor2_, value); |
| 12843 | } |
| 12844 | at::functionalization::impl::replace_(self, tmp_output); |
| 12845 | at::functionalization::impl::commit_update(self); |
| 12846 | at::functionalization::impl::sync(self); |
| 12847 | return self; |
| 12848 | } |
| 12849 | } |
| 12850 | |
| 12851 | at::Tensor & ormqr_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & input2, const at::Tensor & input3, bool left, bool transpose, at::Tensor & out) { |
| 12852 | if (false) { |
| 12853 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12854 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12855 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12856 | auto self_meta = to_meta(self); |
| 12857 | auto input2_meta = to_meta(input2); |
| 12858 | auto input3_meta = to_meta(input3); |
| 12859 | auto out_meta = to_meta(out); |
| 12860 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12861 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12862 | at::_ops::ormqr_out::call(self_meta, input2_meta, input3_meta, left, transpose, out_meta); |
| 12863 | } |
| 12864 | |
| 12865 | at::Tensor self_; |
| 12866 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 12867 | at::functionalization::impl::sync(self); |
| 12868 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 12869 | } else { |
| 12870 | self_ = self; |
| 12871 | } |
| 12872 | |
| 12873 | at::Tensor input2_; |
| 12874 | if (at::functionalization::impl::isFunctionalTensor(input2)) { |
| 12875 | at::functionalization::impl::sync(input2); |
| 12876 | input2_ = at::functionalization::impl::from_functional_tensor(input2); |
| 12877 | } else { |
| 12878 | input2_ = input2; |
| 12879 | } |
| 12880 | |
| 12881 | at::Tensor input3_; |
| 12882 | if (at::functionalization::impl::isFunctionalTensor(input3)) { |
| 12883 | at::functionalization::impl::sync(input3); |
| 12884 | input3_ = at::functionalization::impl::from_functional_tensor(input3); |
| 12885 | } else { |
| 12886 | input3_ = input3; |
| 12887 | } |
| 12888 | |
| 12889 | at::Tensor out_; |
| 12890 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 12891 | at::functionalization::impl::sync(out); |
| 12892 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 12893 | } else { |
| 12894 | out_ = out; |
| 12895 | } |
| 12896 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 12897 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(input2) || at::functionalization::impl::isFunctionalTensor(input3))) { |
| 12898 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12899 | TORCH_INTERNAL_ASSERT(false, |
| 12900 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12901 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12902 | } else { |
| 12903 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12904 | at::AutoDispatchSkipFunctionalize guard; |
| 12905 | at::Tensor tmp_output = at::_ops::ormqr_out::call(self_, input2_, input3_, left, transpose, out_); |
| 12906 | return out;; |
| 12907 | } |
| 12908 | } else { |
| 12909 | at::Tensor tmp_output; |
| 12910 | { |
| 12911 | at::AutoDispatchSkipFunctionalize guard; |
| 12912 | tmp_output = at::_ops::ormqr::call(self_, input2_, input3_, left, transpose); |
| 12913 | } |
| 12914 | at::functionalization::impl::replace_(out, tmp_output); |
| 12915 | at::functionalization::impl::commit_update(out); |
| 12916 | at::functionalization::impl::sync(out); |
| 12917 | return out; |
| 12918 | } |
| 12919 | } |
| 12920 | |
| 12921 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> lu_unpack_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & LU_data, const at::Tensor & LU_pivots, bool unpack_data, bool unpack_pivots, at::Tensor & P, at::Tensor & L, at::Tensor & U) { |
| 12922 | if (false) { |
| 12923 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 12924 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 12925 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 12926 | auto LU_data_meta = to_meta(LU_data); |
| 12927 | auto LU_pivots_meta = to_meta(LU_pivots); |
| 12928 | auto P_meta = to_meta(P); |
| 12929 | auto L_meta = to_meta(L); |
| 12930 | auto U_meta = to_meta(U); |
| 12931 | at::AutoDispatchSkipFunctionalize func_guard; |
| 12932 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 12933 | at::_ops::lu_unpack_out::call(LU_data_meta, LU_pivots_meta, unpack_data, unpack_pivots, P_meta, L_meta, U_meta); |
| 12934 | } |
| 12935 | |
| 12936 | at::Tensor LU_data_; |
| 12937 | if (at::functionalization::impl::isFunctionalTensor(LU_data)) { |
| 12938 | at::functionalization::impl::sync(LU_data); |
| 12939 | LU_data_ = at::functionalization::impl::from_functional_tensor(LU_data); |
| 12940 | } else { |
| 12941 | LU_data_ = LU_data; |
| 12942 | } |
| 12943 | |
| 12944 | at::Tensor LU_pivots_; |
| 12945 | if (at::functionalization::impl::isFunctionalTensor(LU_pivots)) { |
| 12946 | at::functionalization::impl::sync(LU_pivots); |
| 12947 | LU_pivots_ = at::functionalization::impl::from_functional_tensor(LU_pivots); |
| 12948 | } else { |
| 12949 | LU_pivots_ = LU_pivots; |
| 12950 | } |
| 12951 | |
| 12952 | at::Tensor P_; |
| 12953 | if (at::functionalization::impl::isFunctionalTensor(P)) { |
| 12954 | at::functionalization::impl::sync(P); |
| 12955 | P_ = at::functionalization::impl::from_functional_tensor(P); |
| 12956 | } else { |
| 12957 | P_ = P; |
| 12958 | } |
| 12959 | |
| 12960 | at::Tensor L_; |
| 12961 | if (at::functionalization::impl::isFunctionalTensor(L)) { |
| 12962 | at::functionalization::impl::sync(L); |
| 12963 | L_ = at::functionalization::impl::from_functional_tensor(L); |
| 12964 | } else { |
| 12965 | L_ = L; |
| 12966 | } |
| 12967 | |
| 12968 | at::Tensor U_; |
| 12969 | if (at::functionalization::impl::isFunctionalTensor(U)) { |
| 12970 | at::functionalization::impl::sync(U); |
| 12971 | U_ = at::functionalization::impl::from_functional_tensor(U); |
| 12972 | } else { |
| 12973 | U_ = U; |
| 12974 | } |
| 12975 | if (!(true && at::functionalization::impl::isFunctionalTensor(P) && at::functionalization::impl::isFunctionalTensor(L) && at::functionalization::impl::isFunctionalTensor(U))) { |
| 12976 | if ((false || at::functionalization::impl::isFunctionalTensor(LU_data) || at::functionalization::impl::isFunctionalTensor(LU_pivots))) { |
| 12977 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 12978 | TORCH_INTERNAL_ASSERT(false, |
| 12979 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 12980 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 12981 | } else { |
| 12982 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 12983 | at::AutoDispatchSkipFunctionalize guard; |
| 12984 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::lu_unpack_out::call(LU_data_, LU_pivots_, unpack_data, unpack_pivots, P_, L_, U_); |
| 12985 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(P, L, U);; |
| 12986 | } |
| 12987 | } else { |
| 12988 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 12989 | { |
| 12990 | at::AutoDispatchSkipFunctionalize guard; |
| 12991 | tmp_output = at::_ops::lu_unpack::call(LU_data_, LU_pivots_, unpack_data, unpack_pivots); |
| 12992 | } |
| 12993 | at::functionalization::impl::replace_(P, std::get<0>(tmp_output)); |
| 12994 | at::functionalization::impl::commit_update(P); |
| 12995 | at::functionalization::impl::sync(P); |
| 12996 | at::functionalization::impl::replace_(L, std::get<1>(tmp_output)); |
| 12997 | at::functionalization::impl::commit_update(L); |
| 12998 | at::functionalization::impl::sync(L); |
| 12999 | at::functionalization::impl::replace_(U, std::get<2>(tmp_output)); |
| 13000 | at::functionalization::impl::commit_update(U); |
| 13001 | at::functionalization::impl::sync(U); |
| 13002 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(P, L, U); |
| 13003 | } |
| 13004 | } |
| 13005 | |
| 13006 | at::Tensor & dist_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, const at::Scalar & p, at::Tensor & out) { |
| 13007 | if (false) { |
| 13008 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13009 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13010 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13011 | auto self_meta = to_meta(self); |
| 13012 | auto other_meta = to_meta(other); |
| 13013 | auto out_meta = to_meta(out); |
| 13014 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13015 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13016 | at::_ops::dist_out::call(self_meta, other_meta, p, out_meta); |
| 13017 | } |
| 13018 | |
| 13019 | at::Tensor self_; |
| 13020 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13021 | at::functionalization::impl::sync(self); |
| 13022 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13023 | } else { |
| 13024 | self_ = self; |
| 13025 | } |
| 13026 | |
| 13027 | at::Tensor other_; |
| 13028 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13029 | at::functionalization::impl::sync(other); |
| 13030 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13031 | } else { |
| 13032 | other_ = other; |
| 13033 | } |
| 13034 | |
| 13035 | at::Tensor out_; |
| 13036 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13037 | at::functionalization::impl::sync(out); |
| 13038 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13039 | } else { |
| 13040 | out_ = out; |
| 13041 | } |
| 13042 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13043 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13044 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13045 | TORCH_INTERNAL_ASSERT(false, |
| 13046 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13047 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13048 | } else { |
| 13049 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13050 | at::AutoDispatchSkipFunctionalize guard; |
| 13051 | at::Tensor tmp_output = at::_ops::dist_out::call(self_, other_, p, out_); |
| 13052 | return out;; |
| 13053 | } |
| 13054 | } else { |
| 13055 | at::Tensor tmp_output; |
| 13056 | { |
| 13057 | at::AutoDispatchSkipFunctionalize guard; |
| 13058 | tmp_output = at::_ops::dist::call(self_, other_, p); |
| 13059 | } |
| 13060 | at::functionalization::impl::replace_(out, tmp_output); |
| 13061 | at::functionalization::impl::commit_update(out); |
| 13062 | at::functionalization::impl::sync(out); |
| 13063 | return out; |
| 13064 | } |
| 13065 | } |
| 13066 | |
| 13067 | at::Tensor & arctan2_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 13068 | if (false) { |
| 13069 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13070 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13071 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13072 | auto self_meta = to_meta(self); |
| 13073 | auto other_meta = to_meta(other); |
| 13074 | auto out_meta = to_meta(out); |
| 13075 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13076 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13077 | at::_ops::arctan2_out::call(self_meta, other_meta, out_meta); |
| 13078 | } |
| 13079 | |
| 13080 | at::Tensor self_; |
| 13081 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13082 | at::functionalization::impl::sync(self); |
| 13083 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13084 | } else { |
| 13085 | self_ = self; |
| 13086 | } |
| 13087 | |
| 13088 | at::Tensor other_; |
| 13089 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13090 | at::functionalization::impl::sync(other); |
| 13091 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13092 | } else { |
| 13093 | other_ = other; |
| 13094 | } |
| 13095 | |
| 13096 | at::Tensor out_; |
| 13097 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13098 | at::functionalization::impl::sync(out); |
| 13099 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13100 | } else { |
| 13101 | out_ = out; |
| 13102 | } |
| 13103 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13104 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13105 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13106 | TORCH_INTERNAL_ASSERT(false, |
| 13107 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13108 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13109 | } else { |
| 13110 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13111 | at::AutoDispatchSkipFunctionalize guard; |
| 13112 | at::Tensor tmp_output = at::_ops::arctan2_out::call(self_, other_, out_); |
| 13113 | return out;; |
| 13114 | } |
| 13115 | } else { |
| 13116 | at::Tensor tmp_output; |
| 13117 | { |
| 13118 | at::AutoDispatchSkipFunctionalize guard; |
| 13119 | tmp_output = at::_ops::arctan2::call(self_, other_); |
| 13120 | } |
| 13121 | at::functionalization::impl::replace_(out, tmp_output); |
| 13122 | at::functionalization::impl::commit_update(out); |
| 13123 | at::functionalization::impl::sync(out); |
| 13124 | return out; |
| 13125 | } |
| 13126 | } |
| 13127 | |
| 13128 | at::Tensor & arctan2_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 13129 | if (true) { |
| 13130 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13131 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13132 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13133 | auto self_meta = to_meta(self); |
| 13134 | auto other_meta = to_meta(other); |
| 13135 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13136 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13137 | at::_ops::arctan2_::call(self_meta, other_meta); |
| 13138 | } |
| 13139 | |
| 13140 | at::Tensor self_; |
| 13141 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13142 | at::functionalization::impl::sync(self); |
| 13143 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13144 | } else { |
| 13145 | self_ = self; |
| 13146 | } |
| 13147 | |
| 13148 | at::Tensor other_; |
| 13149 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13150 | at::functionalization::impl::sync(other); |
| 13151 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13152 | } else { |
| 13153 | other_ = other; |
| 13154 | } |
| 13155 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 13156 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13157 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13158 | TORCH_INTERNAL_ASSERT(false, |
| 13159 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13160 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13161 | } else { |
| 13162 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13163 | at::AutoDispatchSkipFunctionalize guard; |
| 13164 | at::Tensor tmp_output = at::_ops::arctan2_::call(self_, other_); |
| 13165 | return self;; |
| 13166 | } |
| 13167 | } else { |
| 13168 | at::Tensor tmp_output; |
| 13169 | { |
| 13170 | at::AutoDispatchSkipFunctionalize guard; |
| 13171 | tmp_output = at::_ops::arctan2::call(self_, other_); |
| 13172 | } |
| 13173 | at::functionalization::impl::replace_(self, tmp_output); |
| 13174 | at::functionalization::impl::commit_update(self); |
| 13175 | at::functionalization::impl::sync(self); |
| 13176 | return self; |
| 13177 | } |
| 13178 | } |
| 13179 | |
| 13180 | at::Tensor & histc_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t bins, const at::Scalar & min, const at::Scalar & max, at::Tensor & out) { |
| 13181 | if (false) { |
| 13182 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13183 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13184 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13185 | auto self_meta = to_meta(self); |
| 13186 | auto out_meta = to_meta(out); |
| 13187 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13188 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13189 | at::_ops::histc_out::call(self_meta, bins, min, max, out_meta); |
| 13190 | } |
| 13191 | |
| 13192 | at::Tensor self_; |
| 13193 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13194 | at::functionalization::impl::sync(self); |
| 13195 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13196 | } else { |
| 13197 | self_ = self; |
| 13198 | } |
| 13199 | |
| 13200 | at::Tensor out_; |
| 13201 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13202 | at::functionalization::impl::sync(out); |
| 13203 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13204 | } else { |
| 13205 | out_ = out; |
| 13206 | } |
| 13207 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13208 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 13209 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13210 | TORCH_INTERNAL_ASSERT(false, |
| 13211 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13212 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13213 | } else { |
| 13214 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13215 | at::AutoDispatchSkipFunctionalize guard; |
| 13216 | at::Tensor tmp_output = at::_ops::histc_out::call(self_, bins, min, max, out_); |
| 13217 | return out;; |
| 13218 | } |
| 13219 | } else { |
| 13220 | at::Tensor tmp_output; |
| 13221 | { |
| 13222 | at::AutoDispatchSkipFunctionalize guard; |
| 13223 | tmp_output = at::_ops::histc::call(self_, bins, min, max); |
| 13224 | } |
| 13225 | at::functionalization::impl::replace_(out, tmp_output); |
| 13226 | at::functionalization::impl::commit_update(out); |
| 13227 | at::functionalization::impl::sync(out); |
| 13228 | return out; |
| 13229 | } |
| 13230 | } |
| 13231 | |
| 13232 | at::Tensor & _histogramdd_from_bin_cts_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef bins, c10::optional<at::ArrayRef<double>> range, const c10::optional<at::Tensor> & weight, bool density, at::Tensor & out) { |
| 13233 | if (false) { |
| 13234 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13235 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13236 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13237 | auto self_meta = to_meta(self); |
| 13238 | auto weight_meta = to_meta(weight); |
| 13239 | auto out_meta = to_meta(out); |
| 13240 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13241 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13242 | at::_ops::_histogramdd_from_bin_cts_out::call(self_meta, bins, range, weight_meta, density, out_meta); |
| 13243 | } |
| 13244 | |
| 13245 | at::Tensor self_; |
| 13246 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13247 | at::functionalization::impl::sync(self); |
| 13248 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13249 | } else { |
| 13250 | self_ = self; |
| 13251 | } |
| 13252 | |
| 13253 | c10::optional<at::Tensor> weight_; |
| 13254 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 13255 | at::functionalization::impl::sync(weight); |
| 13256 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 13257 | } else { |
| 13258 | weight_ = weight; |
| 13259 | } |
| 13260 | |
| 13261 | at::Tensor out_; |
| 13262 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13263 | at::functionalization::impl::sync(out); |
| 13264 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13265 | } else { |
| 13266 | out_ = out; |
| 13267 | } |
| 13268 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13269 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 13270 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13271 | TORCH_INTERNAL_ASSERT(false, |
| 13272 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13273 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13274 | } else { |
| 13275 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13276 | at::AutoDispatchSkipFunctionalize guard; |
| 13277 | at::Tensor tmp_output = at::_ops::_histogramdd_from_bin_cts_out::call(self_, bins, range, weight_, density, out_); |
| 13278 | return out;; |
| 13279 | } |
| 13280 | } else { |
| 13281 | at::Tensor tmp_output; |
| 13282 | { |
| 13283 | at::AutoDispatchSkipFunctionalize guard; |
| 13284 | tmp_output = at::_ops::_histogramdd_from_bin_cts::call(self_, bins, range, weight_, density); |
| 13285 | } |
| 13286 | at::functionalization::impl::replace_(out, tmp_output); |
| 13287 | at::functionalization::impl::commit_update(out); |
| 13288 | at::functionalization::impl::sync(out); |
| 13289 | return out; |
| 13290 | } |
| 13291 | } |
| 13292 | |
| 13293 | at::Tensor & hypot_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 13294 | if (false) { |
| 13295 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13296 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13297 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13298 | auto self_meta = to_meta(self); |
| 13299 | auto other_meta = to_meta(other); |
| 13300 | auto out_meta = to_meta(out); |
| 13301 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13302 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13303 | at::_ops::hypot_out::call(self_meta, other_meta, out_meta); |
| 13304 | } |
| 13305 | |
| 13306 | at::Tensor self_; |
| 13307 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13308 | at::functionalization::impl::sync(self); |
| 13309 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13310 | } else { |
| 13311 | self_ = self; |
| 13312 | } |
| 13313 | |
| 13314 | at::Tensor other_; |
| 13315 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13316 | at::functionalization::impl::sync(other); |
| 13317 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13318 | } else { |
| 13319 | other_ = other; |
| 13320 | } |
| 13321 | |
| 13322 | at::Tensor out_; |
| 13323 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13324 | at::functionalization::impl::sync(out); |
| 13325 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13326 | } else { |
| 13327 | out_ = out; |
| 13328 | } |
| 13329 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13330 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13331 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13332 | TORCH_INTERNAL_ASSERT(false, |
| 13333 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13334 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13335 | } else { |
| 13336 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13337 | at::AutoDispatchSkipFunctionalize guard; |
| 13338 | at::Tensor tmp_output = at::_ops::hypot_out::call(self_, other_, out_); |
| 13339 | return out;; |
| 13340 | } |
| 13341 | } else { |
| 13342 | at::Tensor tmp_output; |
| 13343 | { |
| 13344 | at::AutoDispatchSkipFunctionalize guard; |
| 13345 | tmp_output = at::_ops::hypot::call(self_, other_); |
| 13346 | } |
| 13347 | at::functionalization::impl::replace_(out, tmp_output); |
| 13348 | at::functionalization::impl::commit_update(out); |
| 13349 | at::functionalization::impl::sync(out); |
| 13350 | return out; |
| 13351 | } |
| 13352 | } |
| 13353 | |
| 13354 | at::Tensor & hypot_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Tensor & other) { |
| 13355 | if (true) { |
| 13356 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13357 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13358 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13359 | auto self_meta = to_meta(self); |
| 13360 | auto other_meta = to_meta(other); |
| 13361 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13362 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13363 | at::_ops::hypot_::call(self_meta, other_meta); |
| 13364 | } |
| 13365 | |
| 13366 | at::Tensor self_; |
| 13367 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13368 | at::functionalization::impl::sync(self); |
| 13369 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13370 | } else { |
| 13371 | self_ = self; |
| 13372 | } |
| 13373 | |
| 13374 | at::Tensor other_; |
| 13375 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13376 | at::functionalization::impl::sync(other); |
| 13377 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13378 | } else { |
| 13379 | other_ = other; |
| 13380 | } |
| 13381 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 13382 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13383 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13384 | TORCH_INTERNAL_ASSERT(false, |
| 13385 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13386 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13387 | } else { |
| 13388 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13389 | at::AutoDispatchSkipFunctionalize guard; |
| 13390 | at::Tensor tmp_output = at::_ops::hypot_::call(self_, other_); |
| 13391 | return self;; |
| 13392 | } |
| 13393 | } else { |
| 13394 | at::Tensor tmp_output; |
| 13395 | { |
| 13396 | at::AutoDispatchSkipFunctionalize guard; |
| 13397 | tmp_output = at::_ops::hypot::call(self_, other_); |
| 13398 | } |
| 13399 | at::functionalization::impl::replace_(self, tmp_output); |
| 13400 | at::functionalization::impl::commit_update(self); |
| 13401 | at::functionalization::impl::sync(self); |
| 13402 | return self; |
| 13403 | } |
| 13404 | } |
| 13405 | |
| 13406 | at::Tensor & min_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 13407 | if (false) { |
| 13408 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13409 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13410 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13411 | auto self_meta = to_meta(self); |
| 13412 | auto other_meta = to_meta(other); |
| 13413 | auto out_meta = to_meta(out); |
| 13414 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13415 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13416 | at::_ops::min_out::call(self_meta, other_meta, out_meta); |
| 13417 | } |
| 13418 | |
| 13419 | at::Tensor self_; |
| 13420 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13421 | at::functionalization::impl::sync(self); |
| 13422 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13423 | } else { |
| 13424 | self_ = self; |
| 13425 | } |
| 13426 | |
| 13427 | at::Tensor other_; |
| 13428 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13429 | at::functionalization::impl::sync(other); |
| 13430 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13431 | } else { |
| 13432 | other_ = other; |
| 13433 | } |
| 13434 | |
| 13435 | at::Tensor out_; |
| 13436 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13437 | at::functionalization::impl::sync(out); |
| 13438 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13439 | } else { |
| 13440 | out_ = out; |
| 13441 | } |
| 13442 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13443 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13444 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13445 | TORCH_INTERNAL_ASSERT(false, |
| 13446 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13447 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13448 | } else { |
| 13449 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13450 | at::AutoDispatchSkipFunctionalize guard; |
| 13451 | at::Tensor tmp_output = at::_ops::min_out::call(self_, other_, out_); |
| 13452 | return out;; |
| 13453 | } |
| 13454 | } else { |
| 13455 | at::Tensor tmp_output; |
| 13456 | { |
| 13457 | at::AutoDispatchSkipFunctionalize guard; |
| 13458 | tmp_output = at::_ops::min_other::call(self_, other_); |
| 13459 | } |
| 13460 | at::functionalization::impl::replace_(out, tmp_output); |
| 13461 | at::functionalization::impl::commit_update(out); |
| 13462 | at::functionalization::impl::sync(out); |
| 13463 | return out; |
| 13464 | } |
| 13465 | } |
| 13466 | |
| 13467 | at::Tensor & msort_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 13468 | if (false) { |
| 13469 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13470 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13471 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13472 | auto self_meta = to_meta(self); |
| 13473 | auto out_meta = to_meta(out); |
| 13474 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13475 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13476 | at::_ops::msort_out::call(self_meta, out_meta); |
| 13477 | } |
| 13478 | |
| 13479 | at::Tensor self_; |
| 13480 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13481 | at::functionalization::impl::sync(self); |
| 13482 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13483 | } else { |
| 13484 | self_ = self; |
| 13485 | } |
| 13486 | |
| 13487 | at::Tensor out_; |
| 13488 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13489 | at::functionalization::impl::sync(out); |
| 13490 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13491 | } else { |
| 13492 | out_ = out; |
| 13493 | } |
| 13494 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13495 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 13496 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13497 | TORCH_INTERNAL_ASSERT(false, |
| 13498 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13499 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13500 | } else { |
| 13501 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13502 | at::AutoDispatchSkipFunctionalize guard; |
| 13503 | at::Tensor tmp_output = at::_ops::msort_out::call(self_, out_); |
| 13504 | return out;; |
| 13505 | } |
| 13506 | } else { |
| 13507 | at::Tensor tmp_output; |
| 13508 | { |
| 13509 | at::AutoDispatchSkipFunctionalize guard; |
| 13510 | tmp_output = at::_ops::msort::call(self_); |
| 13511 | } |
| 13512 | at::functionalization::impl::replace_(out, tmp_output); |
| 13513 | at::functionalization::impl::commit_update(out); |
| 13514 | at::functionalization::impl::sync(out); |
| 13515 | return out; |
| 13516 | } |
| 13517 | } |
| 13518 | |
| 13519 | void _foreach_add_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, const at::Scalar & scalar, at::TensorList out) { |
| 13520 | if (false) { |
| 13521 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13522 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13523 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13524 | auto self_meta = to_meta(self); |
| 13525 | auto out_meta = to_meta(out); |
| 13526 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13527 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13528 | at::_ops::_foreach_add_Scalar_out::call(self_meta, scalar, out_meta); |
| 13529 | } |
| 13530 | |
| 13531 | ::std::vector<at::Tensor> self_; |
| 13532 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13533 | at::functionalization::impl::sync(self); |
| 13534 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13535 | } else { |
| 13536 | self_ = self.vec(); |
| 13537 | } |
| 13538 | |
| 13539 | ::std::vector<at::Tensor> out_; |
| 13540 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13541 | at::functionalization::impl::sync(out); |
| 13542 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13543 | } else { |
| 13544 | out_ = out.vec(); |
| 13545 | } |
| 13546 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13547 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 13548 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13549 | TORCH_INTERNAL_ASSERT(false, |
| 13550 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13551 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13552 | } else { |
| 13553 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13554 | at::AutoDispatchSkipFunctionalize guard; |
| 13555 | at::_ops::_foreach_add_Scalar_out::call(self_, scalar, out_); |
| 13556 | ; |
| 13557 | } |
| 13558 | } else { |
| 13559 | ::std::vector<at::Tensor> tmp_output; |
| 13560 | { |
| 13561 | at::AutoDispatchSkipFunctionalize guard; |
| 13562 | tmp_output = at::_ops::_foreach_add_Scalar::call(self_, scalar); |
| 13563 | } |
| 13564 | at::functionalization::impl::replace_(out, tmp_output); |
| 13565 | at::functionalization::impl::commit_update(out); |
| 13566 | at::functionalization::impl::sync(out); |
| 13567 | |
| 13568 | } |
| 13569 | } |
| 13570 | |
| 13571 | void _foreach_add__Scalar(c10::DispatchKeySet dispatchKeySet, at::TensorList self, const at::Scalar & scalar) { |
| 13572 | if (true) { |
| 13573 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13574 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13575 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13576 | auto self_meta = to_meta(self); |
| 13577 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13578 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13579 | at::_ops::_foreach_add__Scalar::call(self_meta, scalar); |
| 13580 | } |
| 13581 | |
| 13582 | ::std::vector<at::Tensor> self_; |
| 13583 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13584 | at::functionalization::impl::sync(self); |
| 13585 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13586 | } else { |
| 13587 | self_ = self.vec(); |
| 13588 | } |
| 13589 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 13590 | if ((false)) { |
| 13591 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13592 | TORCH_INTERNAL_ASSERT(false, |
| 13593 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13594 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13595 | } else { |
| 13596 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13597 | at::AutoDispatchSkipFunctionalize guard; |
| 13598 | at::_ops::_foreach_add__Scalar::call(self_, scalar); |
| 13599 | ; |
| 13600 | } |
| 13601 | } else { |
| 13602 | ::std::vector<at::Tensor> tmp_output; |
| 13603 | { |
| 13604 | at::AutoDispatchSkipFunctionalize guard; |
| 13605 | tmp_output = at::_ops::_foreach_add_Scalar::call(self_, scalar); |
| 13606 | } |
| 13607 | at::functionalization::impl::replace_(self, tmp_output); |
| 13608 | at::functionalization::impl::commit_update(self); |
| 13609 | at::functionalization::impl::sync(self); |
| 13610 | |
| 13611 | } |
| 13612 | } |
| 13613 | |
| 13614 | void _foreach_clamp_min_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, const at::Scalar & scalar, at::TensorList out) { |
| 13615 | if (false) { |
| 13616 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13617 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13618 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13619 | auto self_meta = to_meta(self); |
| 13620 | auto out_meta = to_meta(out); |
| 13621 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13622 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13623 | at::_ops::_foreach_clamp_min_Scalar_out::call(self_meta, scalar, out_meta); |
| 13624 | } |
| 13625 | |
| 13626 | ::std::vector<at::Tensor> self_; |
| 13627 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13628 | at::functionalization::impl::sync(self); |
| 13629 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13630 | } else { |
| 13631 | self_ = self.vec(); |
| 13632 | } |
| 13633 | |
| 13634 | ::std::vector<at::Tensor> out_; |
| 13635 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13636 | at::functionalization::impl::sync(out); |
| 13637 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13638 | } else { |
| 13639 | out_ = out.vec(); |
| 13640 | } |
| 13641 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13642 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 13643 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13644 | TORCH_INTERNAL_ASSERT(false, |
| 13645 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13646 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13647 | } else { |
| 13648 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13649 | at::AutoDispatchSkipFunctionalize guard; |
| 13650 | at::_ops::_foreach_clamp_min_Scalar_out::call(self_, scalar, out_); |
| 13651 | ; |
| 13652 | } |
| 13653 | } else { |
| 13654 | ::std::vector<at::Tensor> tmp_output; |
| 13655 | { |
| 13656 | at::AutoDispatchSkipFunctionalize guard; |
| 13657 | tmp_output = at::_ops::_foreach_clamp_min_Scalar::call(self_, scalar); |
| 13658 | } |
| 13659 | at::functionalization::impl::replace_(out, tmp_output); |
| 13660 | at::functionalization::impl::commit_update(out); |
| 13661 | at::functionalization::impl::sync(out); |
| 13662 | |
| 13663 | } |
| 13664 | } |
| 13665 | |
| 13666 | void _foreach_clamp_min__Scalar(c10::DispatchKeySet dispatchKeySet, at::TensorList self, const at::Scalar & scalar) { |
| 13667 | if (true) { |
| 13668 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13669 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13670 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13671 | auto self_meta = to_meta(self); |
| 13672 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13673 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13674 | at::_ops::_foreach_clamp_min__Scalar::call(self_meta, scalar); |
| 13675 | } |
| 13676 | |
| 13677 | ::std::vector<at::Tensor> self_; |
| 13678 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13679 | at::functionalization::impl::sync(self); |
| 13680 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13681 | } else { |
| 13682 | self_ = self.vec(); |
| 13683 | } |
| 13684 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 13685 | if ((false)) { |
| 13686 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13687 | TORCH_INTERNAL_ASSERT(false, |
| 13688 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13689 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13690 | } else { |
| 13691 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13692 | at::AutoDispatchSkipFunctionalize guard; |
| 13693 | at::_ops::_foreach_clamp_min__Scalar::call(self_, scalar); |
| 13694 | ; |
| 13695 | } |
| 13696 | } else { |
| 13697 | ::std::vector<at::Tensor> tmp_output; |
| 13698 | { |
| 13699 | at::AutoDispatchSkipFunctionalize guard; |
| 13700 | tmp_output = at::_ops::_foreach_clamp_min_Scalar::call(self_, scalar); |
| 13701 | } |
| 13702 | at::functionalization::impl::replace_(self, tmp_output); |
| 13703 | at::functionalization::impl::commit_update(self); |
| 13704 | at::functionalization::impl::sync(self); |
| 13705 | |
| 13706 | } |
| 13707 | } |
| 13708 | |
| 13709 | void _foreach_add_out_List_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList other, const at::Scalar & alpha, at::TensorList out) { |
| 13710 | if (false) { |
| 13711 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13712 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13713 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13714 | auto self_meta = to_meta(self); |
| 13715 | auto other_meta = to_meta(other); |
| 13716 | auto out_meta = to_meta(out); |
| 13717 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13718 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13719 | at::_ops::_foreach_add_List_out::call(self_meta, other_meta, alpha, out_meta); |
| 13720 | } |
| 13721 | |
| 13722 | ::std::vector<at::Tensor> self_; |
| 13723 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13724 | at::functionalization::impl::sync(self); |
| 13725 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13726 | } else { |
| 13727 | self_ = self.vec(); |
| 13728 | } |
| 13729 | |
| 13730 | ::std::vector<at::Tensor> other_; |
| 13731 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13732 | at::functionalization::impl::sync(other); |
| 13733 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13734 | } else { |
| 13735 | other_ = other.vec(); |
| 13736 | } |
| 13737 | |
| 13738 | ::std::vector<at::Tensor> out_; |
| 13739 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13740 | at::functionalization::impl::sync(out); |
| 13741 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13742 | } else { |
| 13743 | out_ = out.vec(); |
| 13744 | } |
| 13745 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13746 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13747 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13748 | TORCH_INTERNAL_ASSERT(false, |
| 13749 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13750 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13751 | } else { |
| 13752 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13753 | at::AutoDispatchSkipFunctionalize guard; |
| 13754 | at::_ops::_foreach_add_List_out::call(self_, other_, alpha, out_); |
| 13755 | ; |
| 13756 | } |
| 13757 | } else { |
| 13758 | ::std::vector<at::Tensor> tmp_output; |
| 13759 | { |
| 13760 | at::AutoDispatchSkipFunctionalize guard; |
| 13761 | tmp_output = at::_ops::_foreach_add_List::call(self_, other_, alpha); |
| 13762 | } |
| 13763 | at::functionalization::impl::replace_(out, tmp_output); |
| 13764 | at::functionalization::impl::commit_update(out); |
| 13765 | at::functionalization::impl::sync(out); |
| 13766 | |
| 13767 | } |
| 13768 | } |
| 13769 | |
| 13770 | void _foreach_add__List(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList other, const at::Scalar & alpha) { |
| 13771 | if (true) { |
| 13772 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13773 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13774 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13775 | auto self_meta = to_meta(self); |
| 13776 | auto other_meta = to_meta(other); |
| 13777 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13778 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13779 | at::_ops::_foreach_add__List::call(self_meta, other_meta, alpha); |
| 13780 | } |
| 13781 | |
| 13782 | ::std::vector<at::Tensor> self_; |
| 13783 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13784 | at::functionalization::impl::sync(self); |
| 13785 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13786 | } else { |
| 13787 | self_ = self.vec(); |
| 13788 | } |
| 13789 | |
| 13790 | ::std::vector<at::Tensor> other_; |
| 13791 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13792 | at::functionalization::impl::sync(other); |
| 13793 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13794 | } else { |
| 13795 | other_ = other.vec(); |
| 13796 | } |
| 13797 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 13798 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13799 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13800 | TORCH_INTERNAL_ASSERT(false, |
| 13801 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13802 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13803 | } else { |
| 13804 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13805 | at::AutoDispatchSkipFunctionalize guard; |
| 13806 | at::_ops::_foreach_add__List::call(self_, other_, alpha); |
| 13807 | ; |
| 13808 | } |
| 13809 | } else { |
| 13810 | ::std::vector<at::Tensor> tmp_output; |
| 13811 | { |
| 13812 | at::AutoDispatchSkipFunctionalize guard; |
| 13813 | tmp_output = at::_ops::_foreach_add_List::call(self_, other_, alpha); |
| 13814 | } |
| 13815 | at::functionalization::impl::replace_(self, tmp_output); |
| 13816 | at::functionalization::impl::commit_update(self); |
| 13817 | at::functionalization::impl::sync(self); |
| 13818 | |
| 13819 | } |
| 13820 | } |
| 13821 | |
| 13822 | void _foreach_clamp_min_out_List_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList other, at::TensorList out) { |
| 13823 | if (false) { |
| 13824 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13825 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13826 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13827 | auto self_meta = to_meta(self); |
| 13828 | auto other_meta = to_meta(other); |
| 13829 | auto out_meta = to_meta(out); |
| 13830 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13831 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13832 | at::_ops::_foreach_clamp_min_List_out::call(self_meta, other_meta, out_meta); |
| 13833 | } |
| 13834 | |
| 13835 | ::std::vector<at::Tensor> self_; |
| 13836 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13837 | at::functionalization::impl::sync(self); |
| 13838 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13839 | } else { |
| 13840 | self_ = self.vec(); |
| 13841 | } |
| 13842 | |
| 13843 | ::std::vector<at::Tensor> other_; |
| 13844 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13845 | at::functionalization::impl::sync(other); |
| 13846 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13847 | } else { |
| 13848 | other_ = other.vec(); |
| 13849 | } |
| 13850 | |
| 13851 | ::std::vector<at::Tensor> out_; |
| 13852 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13853 | at::functionalization::impl::sync(out); |
| 13854 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13855 | } else { |
| 13856 | out_ = out.vec(); |
| 13857 | } |
| 13858 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13859 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13860 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13861 | TORCH_INTERNAL_ASSERT(false, |
| 13862 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13863 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13864 | } else { |
| 13865 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13866 | at::AutoDispatchSkipFunctionalize guard; |
| 13867 | at::_ops::_foreach_clamp_min_List_out::call(self_, other_, out_); |
| 13868 | ; |
| 13869 | } |
| 13870 | } else { |
| 13871 | ::std::vector<at::Tensor> tmp_output; |
| 13872 | { |
| 13873 | at::AutoDispatchSkipFunctionalize guard; |
| 13874 | tmp_output = at::_ops::_foreach_clamp_min_List::call(self_, other_); |
| 13875 | } |
| 13876 | at::functionalization::impl::replace_(out, tmp_output); |
| 13877 | at::functionalization::impl::commit_update(out); |
| 13878 | at::functionalization::impl::sync(out); |
| 13879 | |
| 13880 | } |
| 13881 | } |
| 13882 | |
| 13883 | void _foreach_clamp_min__List(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList other) { |
| 13884 | if (true) { |
| 13885 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13886 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13887 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13888 | auto self_meta = to_meta(self); |
| 13889 | auto other_meta = to_meta(other); |
| 13890 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13891 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13892 | at::_ops::_foreach_clamp_min__List::call(self_meta, other_meta); |
| 13893 | } |
| 13894 | |
| 13895 | ::std::vector<at::Tensor> self_; |
| 13896 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13897 | at::functionalization::impl::sync(self); |
| 13898 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13899 | } else { |
| 13900 | self_ = self.vec(); |
| 13901 | } |
| 13902 | |
| 13903 | ::std::vector<at::Tensor> other_; |
| 13904 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 13905 | at::functionalization::impl::sync(other); |
| 13906 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 13907 | } else { |
| 13908 | other_ = other.vec(); |
| 13909 | } |
| 13910 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 13911 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 13912 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13913 | TORCH_INTERNAL_ASSERT(false, |
| 13914 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13915 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13916 | } else { |
| 13917 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13918 | at::AutoDispatchSkipFunctionalize guard; |
| 13919 | at::_ops::_foreach_clamp_min__List::call(self_, other_); |
| 13920 | ; |
| 13921 | } |
| 13922 | } else { |
| 13923 | ::std::vector<at::Tensor> tmp_output; |
| 13924 | { |
| 13925 | at::AutoDispatchSkipFunctionalize guard; |
| 13926 | tmp_output = at::_ops::_foreach_clamp_min_List::call(self_, other_); |
| 13927 | } |
| 13928 | at::functionalization::impl::replace_(self, tmp_output); |
| 13929 | at::functionalization::impl::commit_update(self); |
| 13930 | at::functionalization::impl::sync(self); |
| 13931 | |
| 13932 | } |
| 13933 | } |
| 13934 | |
| 13935 | void _foreach_add_out_ScalarList_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::ArrayRef<at::Scalar> scalars, at::TensorList out) { |
| 13936 | if (false) { |
| 13937 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13938 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13939 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13940 | auto self_meta = to_meta(self); |
| 13941 | auto out_meta = to_meta(out); |
| 13942 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13943 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13944 | at::_ops::_foreach_add_ScalarList_out::call(self_meta, scalars, out_meta); |
| 13945 | } |
| 13946 | |
| 13947 | ::std::vector<at::Tensor> self_; |
| 13948 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 13949 | at::functionalization::impl::sync(self); |
| 13950 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 13951 | } else { |
| 13952 | self_ = self.vec(); |
| 13953 | } |
| 13954 | |
| 13955 | ::std::vector<at::Tensor> out_; |
| 13956 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 13957 | at::functionalization::impl::sync(out); |
| 13958 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 13959 | } else { |
| 13960 | out_ = out.vec(); |
| 13961 | } |
| 13962 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 13963 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 13964 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 13965 | TORCH_INTERNAL_ASSERT(false, |
| 13966 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 13967 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 13968 | } else { |
| 13969 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 13970 | at::AutoDispatchSkipFunctionalize guard; |
| 13971 | at::_ops::_foreach_add_ScalarList_out::call(self_, scalars, out_); |
| 13972 | ; |
| 13973 | } |
| 13974 | } else { |
| 13975 | ::std::vector<at::Tensor> tmp_output; |
| 13976 | { |
| 13977 | at::AutoDispatchSkipFunctionalize guard; |
| 13978 | tmp_output = at::_ops::_foreach_add_ScalarList::call(self_, scalars); |
| 13979 | } |
| 13980 | at::functionalization::impl::replace_(out, tmp_output); |
| 13981 | at::functionalization::impl::commit_update(out); |
| 13982 | at::functionalization::impl::sync(out); |
| 13983 | |
| 13984 | } |
| 13985 | } |
| 13986 | |
| 13987 | void _foreach_add__ScalarList(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::ArrayRef<at::Scalar> scalars) { |
| 13988 | if (true) { |
| 13989 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 13990 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 13991 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 13992 | auto self_meta = to_meta(self); |
| 13993 | at::AutoDispatchSkipFunctionalize func_guard; |
| 13994 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 13995 | at::_ops::_foreach_add__ScalarList::call(self_meta, scalars); |
| 13996 | } |
| 13997 | |
| 13998 | ::std::vector<at::Tensor> self_; |
| 13999 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14000 | at::functionalization::impl::sync(self); |
| 14001 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14002 | } else { |
| 14003 | self_ = self.vec(); |
| 14004 | } |
| 14005 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14006 | if ((false)) { |
| 14007 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14008 | TORCH_INTERNAL_ASSERT(false, |
| 14009 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14010 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14011 | } else { |
| 14012 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14013 | at::AutoDispatchSkipFunctionalize guard; |
| 14014 | at::_ops::_foreach_add__ScalarList::call(self_, scalars); |
| 14015 | ; |
| 14016 | } |
| 14017 | } else { |
| 14018 | ::std::vector<at::Tensor> tmp_output; |
| 14019 | { |
| 14020 | at::AutoDispatchSkipFunctionalize guard; |
| 14021 | tmp_output = at::_ops::_foreach_add_ScalarList::call(self_, scalars); |
| 14022 | } |
| 14023 | at::functionalization::impl::replace_(self, tmp_output); |
| 14024 | at::functionalization::impl::commit_update(self); |
| 14025 | at::functionalization::impl::sync(self); |
| 14026 | |
| 14027 | } |
| 14028 | } |
| 14029 | |
| 14030 | void _foreach_clamp_min_out_ScalarList_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::ArrayRef<at::Scalar> scalars, at::TensorList out) { |
| 14031 | if (false) { |
| 14032 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14033 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14034 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14035 | auto self_meta = to_meta(self); |
| 14036 | auto out_meta = to_meta(out); |
| 14037 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14038 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14039 | at::_ops::_foreach_clamp_min_ScalarList_out::call(self_meta, scalars, out_meta); |
| 14040 | } |
| 14041 | |
| 14042 | ::std::vector<at::Tensor> self_; |
| 14043 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14044 | at::functionalization::impl::sync(self); |
| 14045 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14046 | } else { |
| 14047 | self_ = self.vec(); |
| 14048 | } |
| 14049 | |
| 14050 | ::std::vector<at::Tensor> out_; |
| 14051 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14052 | at::functionalization::impl::sync(out); |
| 14053 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14054 | } else { |
| 14055 | out_ = out.vec(); |
| 14056 | } |
| 14057 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14058 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 14059 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14060 | TORCH_INTERNAL_ASSERT(false, |
| 14061 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14062 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14063 | } else { |
| 14064 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14065 | at::AutoDispatchSkipFunctionalize guard; |
| 14066 | at::_ops::_foreach_clamp_min_ScalarList_out::call(self_, scalars, out_); |
| 14067 | ; |
| 14068 | } |
| 14069 | } else { |
| 14070 | ::std::vector<at::Tensor> tmp_output; |
| 14071 | { |
| 14072 | at::AutoDispatchSkipFunctionalize guard; |
| 14073 | tmp_output = at::_ops::_foreach_clamp_min_ScalarList::call(self_, scalars); |
| 14074 | } |
| 14075 | at::functionalization::impl::replace_(out, tmp_output); |
| 14076 | at::functionalization::impl::commit_update(out); |
| 14077 | at::functionalization::impl::sync(out); |
| 14078 | |
| 14079 | } |
| 14080 | } |
| 14081 | |
| 14082 | void _foreach_clamp_min__ScalarList(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::ArrayRef<at::Scalar> scalars) { |
| 14083 | if (true) { |
| 14084 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14085 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14086 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14087 | auto self_meta = to_meta(self); |
| 14088 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14089 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14090 | at::_ops::_foreach_clamp_min__ScalarList::call(self_meta, scalars); |
| 14091 | } |
| 14092 | |
| 14093 | ::std::vector<at::Tensor> self_; |
| 14094 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14095 | at::functionalization::impl::sync(self); |
| 14096 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14097 | } else { |
| 14098 | self_ = self.vec(); |
| 14099 | } |
| 14100 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14101 | if ((false)) { |
| 14102 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14103 | TORCH_INTERNAL_ASSERT(false, |
| 14104 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14105 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14106 | } else { |
| 14107 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14108 | at::AutoDispatchSkipFunctionalize guard; |
| 14109 | at::_ops::_foreach_clamp_min__ScalarList::call(self_, scalars); |
| 14110 | ; |
| 14111 | } |
| 14112 | } else { |
| 14113 | ::std::vector<at::Tensor> tmp_output; |
| 14114 | { |
| 14115 | at::AutoDispatchSkipFunctionalize guard; |
| 14116 | tmp_output = at::_ops::_foreach_clamp_min_ScalarList::call(self_, scalars); |
| 14117 | } |
| 14118 | at::functionalization::impl::replace_(self, tmp_output); |
| 14119 | at::functionalization::impl::commit_update(self); |
| 14120 | at::functionalization::impl::sync(self); |
| 14121 | |
| 14122 | } |
| 14123 | } |
| 14124 | |
| 14125 | void _foreach_zero_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList out) { |
| 14126 | if (false) { |
| 14127 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14128 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14129 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14130 | auto self_meta = to_meta(self); |
| 14131 | auto out_meta = to_meta(out); |
| 14132 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14133 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14134 | at::_ops::_foreach_zero_out::call(self_meta, out_meta); |
| 14135 | } |
| 14136 | |
| 14137 | ::std::vector<at::Tensor> self_; |
| 14138 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14139 | at::functionalization::impl::sync(self); |
| 14140 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14141 | } else { |
| 14142 | self_ = self.vec(); |
| 14143 | } |
| 14144 | |
| 14145 | ::std::vector<at::Tensor> out_; |
| 14146 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14147 | at::functionalization::impl::sync(out); |
| 14148 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14149 | } else { |
| 14150 | out_ = out.vec(); |
| 14151 | } |
| 14152 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14153 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 14154 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14155 | TORCH_INTERNAL_ASSERT(false, |
| 14156 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14157 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14158 | } else { |
| 14159 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14160 | at::AutoDispatchSkipFunctionalize guard; |
| 14161 | at::_ops::_foreach_zero_out::call(self_, out_); |
| 14162 | ; |
| 14163 | } |
| 14164 | } else { |
| 14165 | ::std::vector<at::Tensor> tmp_output; |
| 14166 | { |
| 14167 | at::AutoDispatchSkipFunctionalize guard; |
| 14168 | tmp_output = at::_ops::_foreach_zero::call(self_); |
| 14169 | } |
| 14170 | at::functionalization::impl::replace_(out, tmp_output); |
| 14171 | at::functionalization::impl::commit_update(out); |
| 14172 | at::functionalization::impl::sync(out); |
| 14173 | |
| 14174 | } |
| 14175 | } |
| 14176 | |
| 14177 | void _foreach_zero_(c10::DispatchKeySet dispatchKeySet, at::TensorList self) { |
| 14178 | if (true) { |
| 14179 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14180 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14181 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14182 | auto self_meta = to_meta(self); |
| 14183 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14184 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14185 | at::_ops::_foreach_zero_::call(self_meta); |
| 14186 | } |
| 14187 | |
| 14188 | ::std::vector<at::Tensor> self_; |
| 14189 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14190 | at::functionalization::impl::sync(self); |
| 14191 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14192 | } else { |
| 14193 | self_ = self.vec(); |
| 14194 | } |
| 14195 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14196 | if ((false)) { |
| 14197 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14198 | TORCH_INTERNAL_ASSERT(false, |
| 14199 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14200 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14201 | } else { |
| 14202 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14203 | at::AutoDispatchSkipFunctionalize guard; |
| 14204 | at::_ops::_foreach_zero_::call(self_); |
| 14205 | ; |
| 14206 | } |
| 14207 | } else { |
| 14208 | ::std::vector<at::Tensor> tmp_output; |
| 14209 | { |
| 14210 | at::AutoDispatchSkipFunctionalize guard; |
| 14211 | tmp_output = at::_ops::_foreach_zero::call(self_); |
| 14212 | } |
| 14213 | at::functionalization::impl::replace_(self, tmp_output); |
| 14214 | at::functionalization::impl::commit_update(self); |
| 14215 | at::functionalization::impl::sync(self); |
| 14216 | |
| 14217 | } |
| 14218 | } |
| 14219 | |
| 14220 | void _foreach_asin_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList out) { |
| 14221 | if (false) { |
| 14222 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14223 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14224 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14225 | auto self_meta = to_meta(self); |
| 14226 | auto out_meta = to_meta(out); |
| 14227 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14228 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14229 | at::_ops::_foreach_asin_out::call(self_meta, out_meta); |
| 14230 | } |
| 14231 | |
| 14232 | ::std::vector<at::Tensor> self_; |
| 14233 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14234 | at::functionalization::impl::sync(self); |
| 14235 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14236 | } else { |
| 14237 | self_ = self.vec(); |
| 14238 | } |
| 14239 | |
| 14240 | ::std::vector<at::Tensor> out_; |
| 14241 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14242 | at::functionalization::impl::sync(out); |
| 14243 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14244 | } else { |
| 14245 | out_ = out.vec(); |
| 14246 | } |
| 14247 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14248 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 14249 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14250 | TORCH_INTERNAL_ASSERT(false, |
| 14251 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14252 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14253 | } else { |
| 14254 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14255 | at::AutoDispatchSkipFunctionalize guard; |
| 14256 | at::_ops::_foreach_asin_out::call(self_, out_); |
| 14257 | ; |
| 14258 | } |
| 14259 | } else { |
| 14260 | ::std::vector<at::Tensor> tmp_output; |
| 14261 | { |
| 14262 | at::AutoDispatchSkipFunctionalize guard; |
| 14263 | tmp_output = at::_ops::_foreach_asin::call(self_); |
| 14264 | } |
| 14265 | at::functionalization::impl::replace_(out, tmp_output); |
| 14266 | at::functionalization::impl::commit_update(out); |
| 14267 | at::functionalization::impl::sync(out); |
| 14268 | |
| 14269 | } |
| 14270 | } |
| 14271 | |
| 14272 | void _foreach_asin_(c10::DispatchKeySet dispatchKeySet, at::TensorList self) { |
| 14273 | if (true) { |
| 14274 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14275 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14276 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14277 | auto self_meta = to_meta(self); |
| 14278 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14279 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14280 | at::_ops::_foreach_asin_::call(self_meta); |
| 14281 | } |
| 14282 | |
| 14283 | ::std::vector<at::Tensor> self_; |
| 14284 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14285 | at::functionalization::impl::sync(self); |
| 14286 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14287 | } else { |
| 14288 | self_ = self.vec(); |
| 14289 | } |
| 14290 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14291 | if ((false)) { |
| 14292 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14293 | TORCH_INTERNAL_ASSERT(false, |
| 14294 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14295 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14296 | } else { |
| 14297 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14298 | at::AutoDispatchSkipFunctionalize guard; |
| 14299 | at::_ops::_foreach_asin_::call(self_); |
| 14300 | ; |
| 14301 | } |
| 14302 | } else { |
| 14303 | ::std::vector<at::Tensor> tmp_output; |
| 14304 | { |
| 14305 | at::AutoDispatchSkipFunctionalize guard; |
| 14306 | tmp_output = at::_ops::_foreach_asin::call(self_); |
| 14307 | } |
| 14308 | at::functionalization::impl::replace_(self, tmp_output); |
| 14309 | at::functionalization::impl::commit_update(self); |
| 14310 | at::functionalization::impl::sync(self); |
| 14311 | |
| 14312 | } |
| 14313 | } |
| 14314 | |
| 14315 | void _foreach_ceil_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList out) { |
| 14316 | if (false) { |
| 14317 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14318 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14319 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14320 | auto self_meta = to_meta(self); |
| 14321 | auto out_meta = to_meta(out); |
| 14322 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14323 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14324 | at::_ops::_foreach_ceil_out::call(self_meta, out_meta); |
| 14325 | } |
| 14326 | |
| 14327 | ::std::vector<at::Tensor> self_; |
| 14328 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14329 | at::functionalization::impl::sync(self); |
| 14330 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14331 | } else { |
| 14332 | self_ = self.vec(); |
| 14333 | } |
| 14334 | |
| 14335 | ::std::vector<at::Tensor> out_; |
| 14336 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14337 | at::functionalization::impl::sync(out); |
| 14338 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14339 | } else { |
| 14340 | out_ = out.vec(); |
| 14341 | } |
| 14342 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14343 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 14344 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14345 | TORCH_INTERNAL_ASSERT(false, |
| 14346 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14347 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14348 | } else { |
| 14349 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14350 | at::AutoDispatchSkipFunctionalize guard; |
| 14351 | at::_ops::_foreach_ceil_out::call(self_, out_); |
| 14352 | ; |
| 14353 | } |
| 14354 | } else { |
| 14355 | ::std::vector<at::Tensor> tmp_output; |
| 14356 | { |
| 14357 | at::AutoDispatchSkipFunctionalize guard; |
| 14358 | tmp_output = at::_ops::_foreach_ceil::call(self_); |
| 14359 | } |
| 14360 | at::functionalization::impl::replace_(out, tmp_output); |
| 14361 | at::functionalization::impl::commit_update(out); |
| 14362 | at::functionalization::impl::sync(out); |
| 14363 | |
| 14364 | } |
| 14365 | } |
| 14366 | |
| 14367 | void _foreach_ceil_(c10::DispatchKeySet dispatchKeySet, at::TensorList self) { |
| 14368 | if (true) { |
| 14369 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14370 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14371 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14372 | auto self_meta = to_meta(self); |
| 14373 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14374 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14375 | at::_ops::_foreach_ceil_::call(self_meta); |
| 14376 | } |
| 14377 | |
| 14378 | ::std::vector<at::Tensor> self_; |
| 14379 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14380 | at::functionalization::impl::sync(self); |
| 14381 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14382 | } else { |
| 14383 | self_ = self.vec(); |
| 14384 | } |
| 14385 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14386 | if ((false)) { |
| 14387 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14388 | TORCH_INTERNAL_ASSERT(false, |
| 14389 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14390 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14391 | } else { |
| 14392 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14393 | at::AutoDispatchSkipFunctionalize guard; |
| 14394 | at::_ops::_foreach_ceil_::call(self_); |
| 14395 | ; |
| 14396 | } |
| 14397 | } else { |
| 14398 | ::std::vector<at::Tensor> tmp_output; |
| 14399 | { |
| 14400 | at::AutoDispatchSkipFunctionalize guard; |
| 14401 | tmp_output = at::_ops::_foreach_ceil::call(self_); |
| 14402 | } |
| 14403 | at::functionalization::impl::replace_(self, tmp_output); |
| 14404 | at::functionalization::impl::commit_update(self); |
| 14405 | at::functionalization::impl::sync(self); |
| 14406 | |
| 14407 | } |
| 14408 | } |
| 14409 | |
| 14410 | void _foreach_cosh_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList out) { |
| 14411 | if (false) { |
| 14412 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14413 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14414 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14415 | auto self_meta = to_meta(self); |
| 14416 | auto out_meta = to_meta(out); |
| 14417 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14418 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14419 | at::_ops::_foreach_cosh_out::call(self_meta, out_meta); |
| 14420 | } |
| 14421 | |
| 14422 | ::std::vector<at::Tensor> self_; |
| 14423 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14424 | at::functionalization::impl::sync(self); |
| 14425 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14426 | } else { |
| 14427 | self_ = self.vec(); |
| 14428 | } |
| 14429 | |
| 14430 | ::std::vector<at::Tensor> out_; |
| 14431 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14432 | at::functionalization::impl::sync(out); |
| 14433 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14434 | } else { |
| 14435 | out_ = out.vec(); |
| 14436 | } |
| 14437 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14438 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 14439 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14440 | TORCH_INTERNAL_ASSERT(false, |
| 14441 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14442 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14443 | } else { |
| 14444 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14445 | at::AutoDispatchSkipFunctionalize guard; |
| 14446 | at::_ops::_foreach_cosh_out::call(self_, out_); |
| 14447 | ; |
| 14448 | } |
| 14449 | } else { |
| 14450 | ::std::vector<at::Tensor> tmp_output; |
| 14451 | { |
| 14452 | at::AutoDispatchSkipFunctionalize guard; |
| 14453 | tmp_output = at::_ops::_foreach_cosh::call(self_); |
| 14454 | } |
| 14455 | at::functionalization::impl::replace_(out, tmp_output); |
| 14456 | at::functionalization::impl::commit_update(out); |
| 14457 | at::functionalization::impl::sync(out); |
| 14458 | |
| 14459 | } |
| 14460 | } |
| 14461 | |
| 14462 | void _foreach_cosh_(c10::DispatchKeySet dispatchKeySet, at::TensorList self) { |
| 14463 | if (true) { |
| 14464 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14465 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14466 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14467 | auto self_meta = to_meta(self); |
| 14468 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14469 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14470 | at::_ops::_foreach_cosh_::call(self_meta); |
| 14471 | } |
| 14472 | |
| 14473 | ::std::vector<at::Tensor> self_; |
| 14474 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14475 | at::functionalization::impl::sync(self); |
| 14476 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14477 | } else { |
| 14478 | self_ = self.vec(); |
| 14479 | } |
| 14480 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14481 | if ((false)) { |
| 14482 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14483 | TORCH_INTERNAL_ASSERT(false, |
| 14484 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14485 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14486 | } else { |
| 14487 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14488 | at::AutoDispatchSkipFunctionalize guard; |
| 14489 | at::_ops::_foreach_cosh_::call(self_); |
| 14490 | ; |
| 14491 | } |
| 14492 | } else { |
| 14493 | ::std::vector<at::Tensor> tmp_output; |
| 14494 | { |
| 14495 | at::AutoDispatchSkipFunctionalize guard; |
| 14496 | tmp_output = at::_ops::_foreach_cosh::call(self_); |
| 14497 | } |
| 14498 | at::functionalization::impl::replace_(self, tmp_output); |
| 14499 | at::functionalization::impl::commit_update(self); |
| 14500 | at::functionalization::impl::sync(self); |
| 14501 | |
| 14502 | } |
| 14503 | } |
| 14504 | |
| 14505 | void _foreach_log1p_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList out) { |
| 14506 | if (false) { |
| 14507 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14508 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14509 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14510 | auto self_meta = to_meta(self); |
| 14511 | auto out_meta = to_meta(out); |
| 14512 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14513 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14514 | at::_ops::_foreach_log1p_out::call(self_meta, out_meta); |
| 14515 | } |
| 14516 | |
| 14517 | ::std::vector<at::Tensor> self_; |
| 14518 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14519 | at::functionalization::impl::sync(self); |
| 14520 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14521 | } else { |
| 14522 | self_ = self.vec(); |
| 14523 | } |
| 14524 | |
| 14525 | ::std::vector<at::Tensor> out_; |
| 14526 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14527 | at::functionalization::impl::sync(out); |
| 14528 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14529 | } else { |
| 14530 | out_ = out.vec(); |
| 14531 | } |
| 14532 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14533 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 14534 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14535 | TORCH_INTERNAL_ASSERT(false, |
| 14536 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14537 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14538 | } else { |
| 14539 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14540 | at::AutoDispatchSkipFunctionalize guard; |
| 14541 | at::_ops::_foreach_log1p_out::call(self_, out_); |
| 14542 | ; |
| 14543 | } |
| 14544 | } else { |
| 14545 | ::std::vector<at::Tensor> tmp_output; |
| 14546 | { |
| 14547 | at::AutoDispatchSkipFunctionalize guard; |
| 14548 | tmp_output = at::_ops::_foreach_log1p::call(self_); |
| 14549 | } |
| 14550 | at::functionalization::impl::replace_(out, tmp_output); |
| 14551 | at::functionalization::impl::commit_update(out); |
| 14552 | at::functionalization::impl::sync(out); |
| 14553 | |
| 14554 | } |
| 14555 | } |
| 14556 | |
| 14557 | void _foreach_log1p_(c10::DispatchKeySet dispatchKeySet, at::TensorList self) { |
| 14558 | if (true) { |
| 14559 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14560 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14561 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14562 | auto self_meta = to_meta(self); |
| 14563 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14564 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14565 | at::_ops::_foreach_log1p_::call(self_meta); |
| 14566 | } |
| 14567 | |
| 14568 | ::std::vector<at::Tensor> self_; |
| 14569 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14570 | at::functionalization::impl::sync(self); |
| 14571 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14572 | } else { |
| 14573 | self_ = self.vec(); |
| 14574 | } |
| 14575 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14576 | if ((false)) { |
| 14577 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14578 | TORCH_INTERNAL_ASSERT(false, |
| 14579 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14580 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14581 | } else { |
| 14582 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14583 | at::AutoDispatchSkipFunctionalize guard; |
| 14584 | at::_ops::_foreach_log1p_::call(self_); |
| 14585 | ; |
| 14586 | } |
| 14587 | } else { |
| 14588 | ::std::vector<at::Tensor> tmp_output; |
| 14589 | { |
| 14590 | at::AutoDispatchSkipFunctionalize guard; |
| 14591 | tmp_output = at::_ops::_foreach_log1p::call(self_); |
| 14592 | } |
| 14593 | at::functionalization::impl::replace_(self, tmp_output); |
| 14594 | at::functionalization::impl::commit_update(self); |
| 14595 | at::functionalization::impl::sync(self); |
| 14596 | |
| 14597 | } |
| 14598 | } |
| 14599 | |
| 14600 | void _foreach_log2_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList out) { |
| 14601 | if (false) { |
| 14602 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14603 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14604 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14605 | auto self_meta = to_meta(self); |
| 14606 | auto out_meta = to_meta(out); |
| 14607 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14608 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14609 | at::_ops::_foreach_log2_out::call(self_meta, out_meta); |
| 14610 | } |
| 14611 | |
| 14612 | ::std::vector<at::Tensor> self_; |
| 14613 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14614 | at::functionalization::impl::sync(self); |
| 14615 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14616 | } else { |
| 14617 | self_ = self.vec(); |
| 14618 | } |
| 14619 | |
| 14620 | ::std::vector<at::Tensor> out_; |
| 14621 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14622 | at::functionalization::impl::sync(out); |
| 14623 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14624 | } else { |
| 14625 | out_ = out.vec(); |
| 14626 | } |
| 14627 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14628 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 14629 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14630 | TORCH_INTERNAL_ASSERT(false, |
| 14631 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14632 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14633 | } else { |
| 14634 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14635 | at::AutoDispatchSkipFunctionalize guard; |
| 14636 | at::_ops::_foreach_log2_out::call(self_, out_); |
| 14637 | ; |
| 14638 | } |
| 14639 | } else { |
| 14640 | ::std::vector<at::Tensor> tmp_output; |
| 14641 | { |
| 14642 | at::AutoDispatchSkipFunctionalize guard; |
| 14643 | tmp_output = at::_ops::_foreach_log2::call(self_); |
| 14644 | } |
| 14645 | at::functionalization::impl::replace_(out, tmp_output); |
| 14646 | at::functionalization::impl::commit_update(out); |
| 14647 | at::functionalization::impl::sync(out); |
| 14648 | |
| 14649 | } |
| 14650 | } |
| 14651 | |
| 14652 | void _foreach_log2_(c10::DispatchKeySet dispatchKeySet, at::TensorList self) { |
| 14653 | if (true) { |
| 14654 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14655 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14656 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14657 | auto self_meta = to_meta(self); |
| 14658 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14659 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14660 | at::_ops::_foreach_log2_::call(self_meta); |
| 14661 | } |
| 14662 | |
| 14663 | ::std::vector<at::Tensor> self_; |
| 14664 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14665 | at::functionalization::impl::sync(self); |
| 14666 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14667 | } else { |
| 14668 | self_ = self.vec(); |
| 14669 | } |
| 14670 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14671 | if ((false)) { |
| 14672 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14673 | TORCH_INTERNAL_ASSERT(false, |
| 14674 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14675 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14676 | } else { |
| 14677 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14678 | at::AutoDispatchSkipFunctionalize guard; |
| 14679 | at::_ops::_foreach_log2_::call(self_); |
| 14680 | ; |
| 14681 | } |
| 14682 | } else { |
| 14683 | ::std::vector<at::Tensor> tmp_output; |
| 14684 | { |
| 14685 | at::AutoDispatchSkipFunctionalize guard; |
| 14686 | tmp_output = at::_ops::_foreach_log2::call(self_); |
| 14687 | } |
| 14688 | at::functionalization::impl::replace_(self, tmp_output); |
| 14689 | at::functionalization::impl::commit_update(self); |
| 14690 | at::functionalization::impl::sync(self); |
| 14691 | |
| 14692 | } |
| 14693 | } |
| 14694 | |
| 14695 | void _foreach_round_out_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList out) { |
| 14696 | if (false) { |
| 14697 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14698 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14699 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14700 | auto self_meta = to_meta(self); |
| 14701 | auto out_meta = to_meta(out); |
| 14702 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14703 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14704 | at::_ops::_foreach_round_out::call(self_meta, out_meta); |
| 14705 | } |
| 14706 | |
| 14707 | ::std::vector<at::Tensor> self_; |
| 14708 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14709 | at::functionalization::impl::sync(self); |
| 14710 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14711 | } else { |
| 14712 | self_ = self.vec(); |
| 14713 | } |
| 14714 | |
| 14715 | ::std::vector<at::Tensor> out_; |
| 14716 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14717 | at::functionalization::impl::sync(out); |
| 14718 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14719 | } else { |
| 14720 | out_ = out.vec(); |
| 14721 | } |
| 14722 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14723 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 14724 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14725 | TORCH_INTERNAL_ASSERT(false, |
| 14726 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14727 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14728 | } else { |
| 14729 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14730 | at::AutoDispatchSkipFunctionalize guard; |
| 14731 | at::_ops::_foreach_round_out::call(self_, out_); |
| 14732 | ; |
| 14733 | } |
| 14734 | } else { |
| 14735 | ::std::vector<at::Tensor> tmp_output; |
| 14736 | { |
| 14737 | at::AutoDispatchSkipFunctionalize guard; |
| 14738 | tmp_output = at::_ops::_foreach_round::call(self_); |
| 14739 | } |
| 14740 | at::functionalization::impl::replace_(out, tmp_output); |
| 14741 | at::functionalization::impl::commit_update(out); |
| 14742 | at::functionalization::impl::sync(out); |
| 14743 | |
| 14744 | } |
| 14745 | } |
| 14746 | |
| 14747 | void _foreach_round_(c10::DispatchKeySet dispatchKeySet, at::TensorList self) { |
| 14748 | if (true) { |
| 14749 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14750 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14751 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14752 | auto self_meta = to_meta(self); |
| 14753 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14754 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14755 | at::_ops::_foreach_round_::call(self_meta); |
| 14756 | } |
| 14757 | |
| 14758 | ::std::vector<at::Tensor> self_; |
| 14759 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14760 | at::functionalization::impl::sync(self); |
| 14761 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14762 | } else { |
| 14763 | self_ = self.vec(); |
| 14764 | } |
| 14765 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14766 | if ((false)) { |
| 14767 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14768 | TORCH_INTERNAL_ASSERT(false, |
| 14769 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14770 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14771 | } else { |
| 14772 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14773 | at::AutoDispatchSkipFunctionalize guard; |
| 14774 | at::_ops::_foreach_round_::call(self_); |
| 14775 | ; |
| 14776 | } |
| 14777 | } else { |
| 14778 | ::std::vector<at::Tensor> tmp_output; |
| 14779 | { |
| 14780 | at::AutoDispatchSkipFunctionalize guard; |
| 14781 | tmp_output = at::_ops::_foreach_round::call(self_); |
| 14782 | } |
| 14783 | at::functionalization::impl::replace_(self, tmp_output); |
| 14784 | at::functionalization::impl::commit_update(self); |
| 14785 | at::functionalization::impl::sync(self); |
| 14786 | |
| 14787 | } |
| 14788 | } |
| 14789 | |
| 14790 | void _foreach_addcdiv_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, const at::Scalar & value, at::TensorList out) { |
| 14791 | if (false) { |
| 14792 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14793 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14794 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14795 | auto self_meta = to_meta(self); |
| 14796 | auto tensor1_meta = to_meta(tensor1); |
| 14797 | auto tensor2_meta = to_meta(tensor2); |
| 14798 | auto out_meta = to_meta(out); |
| 14799 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14800 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14801 | at::_ops::_foreach_addcdiv_Scalar_out::call(self_meta, tensor1_meta, tensor2_meta, value, out_meta); |
| 14802 | } |
| 14803 | |
| 14804 | ::std::vector<at::Tensor> self_; |
| 14805 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14806 | at::functionalization::impl::sync(self); |
| 14807 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14808 | } else { |
| 14809 | self_ = self.vec(); |
| 14810 | } |
| 14811 | |
| 14812 | ::std::vector<at::Tensor> tensor1_; |
| 14813 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 14814 | at::functionalization::impl::sync(tensor1); |
| 14815 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 14816 | } else { |
| 14817 | tensor1_ = tensor1.vec(); |
| 14818 | } |
| 14819 | |
| 14820 | ::std::vector<at::Tensor> tensor2_; |
| 14821 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 14822 | at::functionalization::impl::sync(tensor2); |
| 14823 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 14824 | } else { |
| 14825 | tensor2_ = tensor2.vec(); |
| 14826 | } |
| 14827 | |
| 14828 | ::std::vector<at::Tensor> out_; |
| 14829 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14830 | at::functionalization::impl::sync(out); |
| 14831 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14832 | } else { |
| 14833 | out_ = out.vec(); |
| 14834 | } |
| 14835 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14836 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 14837 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14838 | TORCH_INTERNAL_ASSERT(false, |
| 14839 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14840 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14841 | } else { |
| 14842 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14843 | at::AutoDispatchSkipFunctionalize guard; |
| 14844 | at::_ops::_foreach_addcdiv_Scalar_out::call(self_, tensor1_, tensor2_, value, out_); |
| 14845 | ; |
| 14846 | } |
| 14847 | } else { |
| 14848 | ::std::vector<at::Tensor> tmp_output; |
| 14849 | { |
| 14850 | at::AutoDispatchSkipFunctionalize guard; |
| 14851 | tmp_output = at::_ops::_foreach_addcdiv_Scalar::call(self_, tensor1_, tensor2_, value); |
| 14852 | } |
| 14853 | at::functionalization::impl::replace_(out, tmp_output); |
| 14854 | at::functionalization::impl::commit_update(out); |
| 14855 | at::functionalization::impl::sync(out); |
| 14856 | |
| 14857 | } |
| 14858 | } |
| 14859 | |
| 14860 | void _foreach_addcdiv__Scalar(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, const at::Scalar & value) { |
| 14861 | if (true) { |
| 14862 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14863 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14864 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14865 | auto self_meta = to_meta(self); |
| 14866 | auto tensor1_meta = to_meta(tensor1); |
| 14867 | auto tensor2_meta = to_meta(tensor2); |
| 14868 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14869 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14870 | at::_ops::_foreach_addcdiv__Scalar::call(self_meta, tensor1_meta, tensor2_meta, value); |
| 14871 | } |
| 14872 | |
| 14873 | ::std::vector<at::Tensor> self_; |
| 14874 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14875 | at::functionalization::impl::sync(self); |
| 14876 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14877 | } else { |
| 14878 | self_ = self.vec(); |
| 14879 | } |
| 14880 | |
| 14881 | ::std::vector<at::Tensor> tensor1_; |
| 14882 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 14883 | at::functionalization::impl::sync(tensor1); |
| 14884 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 14885 | } else { |
| 14886 | tensor1_ = tensor1.vec(); |
| 14887 | } |
| 14888 | |
| 14889 | ::std::vector<at::Tensor> tensor2_; |
| 14890 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 14891 | at::functionalization::impl::sync(tensor2); |
| 14892 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 14893 | } else { |
| 14894 | tensor2_ = tensor2.vec(); |
| 14895 | } |
| 14896 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 14897 | if ((false || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 14898 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14899 | TORCH_INTERNAL_ASSERT(false, |
| 14900 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14901 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14902 | } else { |
| 14903 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14904 | at::AutoDispatchSkipFunctionalize guard; |
| 14905 | at::_ops::_foreach_addcdiv__Scalar::call(self_, tensor1_, tensor2_, value); |
| 14906 | ; |
| 14907 | } |
| 14908 | } else { |
| 14909 | ::std::vector<at::Tensor> tmp_output; |
| 14910 | { |
| 14911 | at::AutoDispatchSkipFunctionalize guard; |
| 14912 | tmp_output = at::_ops::_foreach_addcdiv_Scalar::call(self_, tensor1_, tensor2_, value); |
| 14913 | } |
| 14914 | at::functionalization::impl::replace_(self, tmp_output); |
| 14915 | at::functionalization::impl::commit_update(self); |
| 14916 | at::functionalization::impl::sync(self); |
| 14917 | |
| 14918 | } |
| 14919 | } |
| 14920 | |
| 14921 | void _foreach_addcmul_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, const at::Scalar & value, at::TensorList out) { |
| 14922 | if (false) { |
| 14923 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14924 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14925 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14926 | auto self_meta = to_meta(self); |
| 14927 | auto tensor1_meta = to_meta(tensor1); |
| 14928 | auto tensor2_meta = to_meta(tensor2); |
| 14929 | auto out_meta = to_meta(out); |
| 14930 | at::AutoDispatchSkipFunctionalize func_guard; |
| 14931 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 14932 | at::_ops::_foreach_addcmul_Scalar_out::call(self_meta, tensor1_meta, tensor2_meta, value, out_meta); |
| 14933 | } |
| 14934 | |
| 14935 | ::std::vector<at::Tensor> self_; |
| 14936 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 14937 | at::functionalization::impl::sync(self); |
| 14938 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 14939 | } else { |
| 14940 | self_ = self.vec(); |
| 14941 | } |
| 14942 | |
| 14943 | ::std::vector<at::Tensor> tensor1_; |
| 14944 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 14945 | at::functionalization::impl::sync(tensor1); |
| 14946 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 14947 | } else { |
| 14948 | tensor1_ = tensor1.vec(); |
| 14949 | } |
| 14950 | |
| 14951 | ::std::vector<at::Tensor> tensor2_; |
| 14952 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 14953 | at::functionalization::impl::sync(tensor2); |
| 14954 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 14955 | } else { |
| 14956 | tensor2_ = tensor2.vec(); |
| 14957 | } |
| 14958 | |
| 14959 | ::std::vector<at::Tensor> out_; |
| 14960 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 14961 | at::functionalization::impl::sync(out); |
| 14962 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 14963 | } else { |
| 14964 | out_ = out.vec(); |
| 14965 | } |
| 14966 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 14967 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 14968 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 14969 | TORCH_INTERNAL_ASSERT(false, |
| 14970 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 14971 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 14972 | } else { |
| 14973 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 14974 | at::AutoDispatchSkipFunctionalize guard; |
| 14975 | at::_ops::_foreach_addcmul_Scalar_out::call(self_, tensor1_, tensor2_, value, out_); |
| 14976 | ; |
| 14977 | } |
| 14978 | } else { |
| 14979 | ::std::vector<at::Tensor> tmp_output; |
| 14980 | { |
| 14981 | at::AutoDispatchSkipFunctionalize guard; |
| 14982 | tmp_output = at::_ops::_foreach_addcmul_Scalar::call(self_, tensor1_, tensor2_, value); |
| 14983 | } |
| 14984 | at::functionalization::impl::replace_(out, tmp_output); |
| 14985 | at::functionalization::impl::commit_update(out); |
| 14986 | at::functionalization::impl::sync(out); |
| 14987 | |
| 14988 | } |
| 14989 | } |
| 14990 | |
| 14991 | void _foreach_addcmul__Scalar(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, const at::Scalar & value) { |
| 14992 | if (true) { |
| 14993 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 14994 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 14995 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 14996 | auto self_meta = to_meta(self); |
| 14997 | auto tensor1_meta = to_meta(tensor1); |
| 14998 | auto tensor2_meta = to_meta(tensor2); |
| 14999 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15000 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15001 | at::_ops::_foreach_addcmul__Scalar::call(self_meta, tensor1_meta, tensor2_meta, value); |
| 15002 | } |
| 15003 | |
| 15004 | ::std::vector<at::Tensor> self_; |
| 15005 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15006 | at::functionalization::impl::sync(self); |
| 15007 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15008 | } else { |
| 15009 | self_ = self.vec(); |
| 15010 | } |
| 15011 | |
| 15012 | ::std::vector<at::Tensor> tensor1_; |
| 15013 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15014 | at::functionalization::impl::sync(tensor1); |
| 15015 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15016 | } else { |
| 15017 | tensor1_ = tensor1.vec(); |
| 15018 | } |
| 15019 | |
| 15020 | ::std::vector<at::Tensor> tensor2_; |
| 15021 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15022 | at::functionalization::impl::sync(tensor2); |
| 15023 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15024 | } else { |
| 15025 | tensor2_ = tensor2.vec(); |
| 15026 | } |
| 15027 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 15028 | if ((false || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 15029 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15030 | TORCH_INTERNAL_ASSERT(false, |
| 15031 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15032 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15033 | } else { |
| 15034 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15035 | at::AutoDispatchSkipFunctionalize guard; |
| 15036 | at::_ops::_foreach_addcmul__Scalar::call(self_, tensor1_, tensor2_, value); |
| 15037 | ; |
| 15038 | } |
| 15039 | } else { |
| 15040 | ::std::vector<at::Tensor> tmp_output; |
| 15041 | { |
| 15042 | at::AutoDispatchSkipFunctionalize guard; |
| 15043 | tmp_output = at::_ops::_foreach_addcmul_Scalar::call(self_, tensor1_, tensor2_, value); |
| 15044 | } |
| 15045 | at::functionalization::impl::replace_(self, tmp_output); |
| 15046 | at::functionalization::impl::commit_update(self); |
| 15047 | at::functionalization::impl::sync(self); |
| 15048 | |
| 15049 | } |
| 15050 | } |
| 15051 | |
| 15052 | void _foreach_addcdiv_out_ScalarList_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, at::ArrayRef<at::Scalar> scalars, at::TensorList out) { |
| 15053 | if (false) { |
| 15054 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15055 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15056 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15057 | auto self_meta = to_meta(self); |
| 15058 | auto tensor1_meta = to_meta(tensor1); |
| 15059 | auto tensor2_meta = to_meta(tensor2); |
| 15060 | auto out_meta = to_meta(out); |
| 15061 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15062 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15063 | at::_ops::_foreach_addcdiv_ScalarList_out::call(self_meta, tensor1_meta, tensor2_meta, scalars, out_meta); |
| 15064 | } |
| 15065 | |
| 15066 | ::std::vector<at::Tensor> self_; |
| 15067 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15068 | at::functionalization::impl::sync(self); |
| 15069 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15070 | } else { |
| 15071 | self_ = self.vec(); |
| 15072 | } |
| 15073 | |
| 15074 | ::std::vector<at::Tensor> tensor1_; |
| 15075 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15076 | at::functionalization::impl::sync(tensor1); |
| 15077 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15078 | } else { |
| 15079 | tensor1_ = tensor1.vec(); |
| 15080 | } |
| 15081 | |
| 15082 | ::std::vector<at::Tensor> tensor2_; |
| 15083 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15084 | at::functionalization::impl::sync(tensor2); |
| 15085 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15086 | } else { |
| 15087 | tensor2_ = tensor2.vec(); |
| 15088 | } |
| 15089 | |
| 15090 | ::std::vector<at::Tensor> out_; |
| 15091 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15092 | at::functionalization::impl::sync(out); |
| 15093 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15094 | } else { |
| 15095 | out_ = out.vec(); |
| 15096 | } |
| 15097 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15098 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 15099 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15100 | TORCH_INTERNAL_ASSERT(false, |
| 15101 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15102 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15103 | } else { |
| 15104 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15105 | at::AutoDispatchSkipFunctionalize guard; |
| 15106 | at::_ops::_foreach_addcdiv_ScalarList_out::call(self_, tensor1_, tensor2_, scalars, out_); |
| 15107 | ; |
| 15108 | } |
| 15109 | } else { |
| 15110 | ::std::vector<at::Tensor> tmp_output; |
| 15111 | { |
| 15112 | at::AutoDispatchSkipFunctionalize guard; |
| 15113 | tmp_output = at::_ops::_foreach_addcdiv_ScalarList::call(self_, tensor1_, tensor2_, scalars); |
| 15114 | } |
| 15115 | at::functionalization::impl::replace_(out, tmp_output); |
| 15116 | at::functionalization::impl::commit_update(out); |
| 15117 | at::functionalization::impl::sync(out); |
| 15118 | |
| 15119 | } |
| 15120 | } |
| 15121 | |
| 15122 | void _foreach_addcdiv__ScalarList(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, at::ArrayRef<at::Scalar> scalars) { |
| 15123 | if (true) { |
| 15124 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15125 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15126 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15127 | auto self_meta = to_meta(self); |
| 15128 | auto tensor1_meta = to_meta(tensor1); |
| 15129 | auto tensor2_meta = to_meta(tensor2); |
| 15130 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15131 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15132 | at::_ops::_foreach_addcdiv__ScalarList::call(self_meta, tensor1_meta, tensor2_meta, scalars); |
| 15133 | } |
| 15134 | |
| 15135 | ::std::vector<at::Tensor> self_; |
| 15136 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15137 | at::functionalization::impl::sync(self); |
| 15138 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15139 | } else { |
| 15140 | self_ = self.vec(); |
| 15141 | } |
| 15142 | |
| 15143 | ::std::vector<at::Tensor> tensor1_; |
| 15144 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15145 | at::functionalization::impl::sync(tensor1); |
| 15146 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15147 | } else { |
| 15148 | tensor1_ = tensor1.vec(); |
| 15149 | } |
| 15150 | |
| 15151 | ::std::vector<at::Tensor> tensor2_; |
| 15152 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15153 | at::functionalization::impl::sync(tensor2); |
| 15154 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15155 | } else { |
| 15156 | tensor2_ = tensor2.vec(); |
| 15157 | } |
| 15158 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 15159 | if ((false || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 15160 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15161 | TORCH_INTERNAL_ASSERT(false, |
| 15162 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15163 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15164 | } else { |
| 15165 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15166 | at::AutoDispatchSkipFunctionalize guard; |
| 15167 | at::_ops::_foreach_addcdiv__ScalarList::call(self_, tensor1_, tensor2_, scalars); |
| 15168 | ; |
| 15169 | } |
| 15170 | } else { |
| 15171 | ::std::vector<at::Tensor> tmp_output; |
| 15172 | { |
| 15173 | at::AutoDispatchSkipFunctionalize guard; |
| 15174 | tmp_output = at::_ops::_foreach_addcdiv_ScalarList::call(self_, tensor1_, tensor2_, scalars); |
| 15175 | } |
| 15176 | at::functionalization::impl::replace_(self, tmp_output); |
| 15177 | at::functionalization::impl::commit_update(self); |
| 15178 | at::functionalization::impl::sync(self); |
| 15179 | |
| 15180 | } |
| 15181 | } |
| 15182 | |
| 15183 | void _foreach_addcdiv_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, const at::Tensor & scalars, at::TensorList out) { |
| 15184 | if (false) { |
| 15185 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15186 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15187 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15188 | auto self_meta = to_meta(self); |
| 15189 | auto tensor1_meta = to_meta(tensor1); |
| 15190 | auto tensor2_meta = to_meta(tensor2); |
| 15191 | auto scalars_meta = to_meta(scalars); |
| 15192 | auto out_meta = to_meta(out); |
| 15193 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15194 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15195 | at::_ops::_foreach_addcdiv_Tensor_out::call(self_meta, tensor1_meta, tensor2_meta, scalars_meta, out_meta); |
| 15196 | } |
| 15197 | |
| 15198 | ::std::vector<at::Tensor> self_; |
| 15199 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15200 | at::functionalization::impl::sync(self); |
| 15201 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15202 | } else { |
| 15203 | self_ = self.vec(); |
| 15204 | } |
| 15205 | |
| 15206 | ::std::vector<at::Tensor> tensor1_; |
| 15207 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15208 | at::functionalization::impl::sync(tensor1); |
| 15209 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15210 | } else { |
| 15211 | tensor1_ = tensor1.vec(); |
| 15212 | } |
| 15213 | |
| 15214 | ::std::vector<at::Tensor> tensor2_; |
| 15215 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15216 | at::functionalization::impl::sync(tensor2); |
| 15217 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15218 | } else { |
| 15219 | tensor2_ = tensor2.vec(); |
| 15220 | } |
| 15221 | |
| 15222 | at::Tensor scalars_; |
| 15223 | if (at::functionalization::impl::isFunctionalTensor(scalars)) { |
| 15224 | at::functionalization::impl::sync(scalars); |
| 15225 | scalars_ = at::functionalization::impl::from_functional_tensor(scalars); |
| 15226 | } else { |
| 15227 | scalars_ = scalars; |
| 15228 | } |
| 15229 | |
| 15230 | ::std::vector<at::Tensor> out_; |
| 15231 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15232 | at::functionalization::impl::sync(out); |
| 15233 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15234 | } else { |
| 15235 | out_ = out.vec(); |
| 15236 | } |
| 15237 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15238 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2) || at::functionalization::impl::isFunctionalTensor(scalars))) { |
| 15239 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15240 | TORCH_INTERNAL_ASSERT(false, |
| 15241 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15242 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15243 | } else { |
| 15244 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15245 | at::AutoDispatchSkipFunctionalize guard; |
| 15246 | at::_ops::_foreach_addcdiv_Tensor_out::call(self_, tensor1_, tensor2_, scalars_, out_); |
| 15247 | ; |
| 15248 | } |
| 15249 | } else { |
| 15250 | ::std::vector<at::Tensor> tmp_output; |
| 15251 | { |
| 15252 | at::AutoDispatchSkipFunctionalize guard; |
| 15253 | tmp_output = at::_ops::_foreach_addcdiv_Tensor::call(self_, tensor1_, tensor2_, scalars_); |
| 15254 | } |
| 15255 | at::functionalization::impl::replace_(out, tmp_output); |
| 15256 | at::functionalization::impl::commit_update(out); |
| 15257 | at::functionalization::impl::sync(out); |
| 15258 | |
| 15259 | } |
| 15260 | } |
| 15261 | |
| 15262 | void _foreach_addcdiv__Tensor(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, const at::Tensor & scalars) { |
| 15263 | if (true) { |
| 15264 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15265 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15266 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15267 | auto self_meta = to_meta(self); |
| 15268 | auto tensor1_meta = to_meta(tensor1); |
| 15269 | auto tensor2_meta = to_meta(tensor2); |
| 15270 | auto scalars_meta = to_meta(scalars); |
| 15271 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15272 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15273 | at::_ops::_foreach_addcdiv__Tensor::call(self_meta, tensor1_meta, tensor2_meta, scalars_meta); |
| 15274 | } |
| 15275 | |
| 15276 | ::std::vector<at::Tensor> self_; |
| 15277 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15278 | at::functionalization::impl::sync(self); |
| 15279 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15280 | } else { |
| 15281 | self_ = self.vec(); |
| 15282 | } |
| 15283 | |
| 15284 | ::std::vector<at::Tensor> tensor1_; |
| 15285 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15286 | at::functionalization::impl::sync(tensor1); |
| 15287 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15288 | } else { |
| 15289 | tensor1_ = tensor1.vec(); |
| 15290 | } |
| 15291 | |
| 15292 | ::std::vector<at::Tensor> tensor2_; |
| 15293 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15294 | at::functionalization::impl::sync(tensor2); |
| 15295 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15296 | } else { |
| 15297 | tensor2_ = tensor2.vec(); |
| 15298 | } |
| 15299 | |
| 15300 | at::Tensor scalars_; |
| 15301 | if (at::functionalization::impl::isFunctionalTensor(scalars)) { |
| 15302 | at::functionalization::impl::sync(scalars); |
| 15303 | scalars_ = at::functionalization::impl::from_functional_tensor(scalars); |
| 15304 | } else { |
| 15305 | scalars_ = scalars; |
| 15306 | } |
| 15307 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 15308 | if ((false || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2) || at::functionalization::impl::isFunctionalTensor(scalars))) { |
| 15309 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15310 | TORCH_INTERNAL_ASSERT(false, |
| 15311 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15312 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15313 | } else { |
| 15314 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15315 | at::AutoDispatchSkipFunctionalize guard; |
| 15316 | at::_ops::_foreach_addcdiv__Tensor::call(self_, tensor1_, tensor2_, scalars_); |
| 15317 | ; |
| 15318 | } |
| 15319 | } else { |
| 15320 | ::std::vector<at::Tensor> tmp_output; |
| 15321 | { |
| 15322 | at::AutoDispatchSkipFunctionalize guard; |
| 15323 | tmp_output = at::_ops::_foreach_addcdiv_Tensor::call(self_, tensor1_, tensor2_, scalars_); |
| 15324 | } |
| 15325 | at::functionalization::impl::replace_(self, tmp_output); |
| 15326 | at::functionalization::impl::commit_update(self); |
| 15327 | at::functionalization::impl::sync(self); |
| 15328 | |
| 15329 | } |
| 15330 | } |
| 15331 | |
| 15332 | void _foreach_addcmul_out_ScalarList_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, at::ArrayRef<at::Scalar> scalars, at::TensorList out) { |
| 15333 | if (false) { |
| 15334 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15335 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15336 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15337 | auto self_meta = to_meta(self); |
| 15338 | auto tensor1_meta = to_meta(tensor1); |
| 15339 | auto tensor2_meta = to_meta(tensor2); |
| 15340 | auto out_meta = to_meta(out); |
| 15341 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15342 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15343 | at::_ops::_foreach_addcmul_ScalarList_out::call(self_meta, tensor1_meta, tensor2_meta, scalars, out_meta); |
| 15344 | } |
| 15345 | |
| 15346 | ::std::vector<at::Tensor> self_; |
| 15347 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15348 | at::functionalization::impl::sync(self); |
| 15349 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15350 | } else { |
| 15351 | self_ = self.vec(); |
| 15352 | } |
| 15353 | |
| 15354 | ::std::vector<at::Tensor> tensor1_; |
| 15355 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15356 | at::functionalization::impl::sync(tensor1); |
| 15357 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15358 | } else { |
| 15359 | tensor1_ = tensor1.vec(); |
| 15360 | } |
| 15361 | |
| 15362 | ::std::vector<at::Tensor> tensor2_; |
| 15363 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15364 | at::functionalization::impl::sync(tensor2); |
| 15365 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15366 | } else { |
| 15367 | tensor2_ = tensor2.vec(); |
| 15368 | } |
| 15369 | |
| 15370 | ::std::vector<at::Tensor> out_; |
| 15371 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15372 | at::functionalization::impl::sync(out); |
| 15373 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15374 | } else { |
| 15375 | out_ = out.vec(); |
| 15376 | } |
| 15377 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15378 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 15379 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15380 | TORCH_INTERNAL_ASSERT(false, |
| 15381 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15382 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15383 | } else { |
| 15384 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15385 | at::AutoDispatchSkipFunctionalize guard; |
| 15386 | at::_ops::_foreach_addcmul_ScalarList_out::call(self_, tensor1_, tensor2_, scalars, out_); |
| 15387 | ; |
| 15388 | } |
| 15389 | } else { |
| 15390 | ::std::vector<at::Tensor> tmp_output; |
| 15391 | { |
| 15392 | at::AutoDispatchSkipFunctionalize guard; |
| 15393 | tmp_output = at::_ops::_foreach_addcmul_ScalarList::call(self_, tensor1_, tensor2_, scalars); |
| 15394 | } |
| 15395 | at::functionalization::impl::replace_(out, tmp_output); |
| 15396 | at::functionalization::impl::commit_update(out); |
| 15397 | at::functionalization::impl::sync(out); |
| 15398 | |
| 15399 | } |
| 15400 | } |
| 15401 | |
| 15402 | void _foreach_addcmul__ScalarList(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, at::ArrayRef<at::Scalar> scalars) { |
| 15403 | if (true) { |
| 15404 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15405 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15406 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15407 | auto self_meta = to_meta(self); |
| 15408 | auto tensor1_meta = to_meta(tensor1); |
| 15409 | auto tensor2_meta = to_meta(tensor2); |
| 15410 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15411 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15412 | at::_ops::_foreach_addcmul__ScalarList::call(self_meta, tensor1_meta, tensor2_meta, scalars); |
| 15413 | } |
| 15414 | |
| 15415 | ::std::vector<at::Tensor> self_; |
| 15416 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15417 | at::functionalization::impl::sync(self); |
| 15418 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15419 | } else { |
| 15420 | self_ = self.vec(); |
| 15421 | } |
| 15422 | |
| 15423 | ::std::vector<at::Tensor> tensor1_; |
| 15424 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15425 | at::functionalization::impl::sync(tensor1); |
| 15426 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15427 | } else { |
| 15428 | tensor1_ = tensor1.vec(); |
| 15429 | } |
| 15430 | |
| 15431 | ::std::vector<at::Tensor> tensor2_; |
| 15432 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15433 | at::functionalization::impl::sync(tensor2); |
| 15434 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15435 | } else { |
| 15436 | tensor2_ = tensor2.vec(); |
| 15437 | } |
| 15438 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 15439 | if ((false || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2))) { |
| 15440 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15441 | TORCH_INTERNAL_ASSERT(false, |
| 15442 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15443 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15444 | } else { |
| 15445 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15446 | at::AutoDispatchSkipFunctionalize guard; |
| 15447 | at::_ops::_foreach_addcmul__ScalarList::call(self_, tensor1_, tensor2_, scalars); |
| 15448 | ; |
| 15449 | } |
| 15450 | } else { |
| 15451 | ::std::vector<at::Tensor> tmp_output; |
| 15452 | { |
| 15453 | at::AutoDispatchSkipFunctionalize guard; |
| 15454 | tmp_output = at::_ops::_foreach_addcmul_ScalarList::call(self_, tensor1_, tensor2_, scalars); |
| 15455 | } |
| 15456 | at::functionalization::impl::replace_(self, tmp_output); |
| 15457 | at::functionalization::impl::commit_update(self); |
| 15458 | at::functionalization::impl::sync(self); |
| 15459 | |
| 15460 | } |
| 15461 | } |
| 15462 | |
| 15463 | void _foreach_addcmul_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, const at::Tensor & scalars, at::TensorList out) { |
| 15464 | if (false) { |
| 15465 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15466 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15467 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15468 | auto self_meta = to_meta(self); |
| 15469 | auto tensor1_meta = to_meta(tensor1); |
| 15470 | auto tensor2_meta = to_meta(tensor2); |
| 15471 | auto scalars_meta = to_meta(scalars); |
| 15472 | auto out_meta = to_meta(out); |
| 15473 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15474 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15475 | at::_ops::_foreach_addcmul_Tensor_out::call(self_meta, tensor1_meta, tensor2_meta, scalars_meta, out_meta); |
| 15476 | } |
| 15477 | |
| 15478 | ::std::vector<at::Tensor> self_; |
| 15479 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15480 | at::functionalization::impl::sync(self); |
| 15481 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15482 | } else { |
| 15483 | self_ = self.vec(); |
| 15484 | } |
| 15485 | |
| 15486 | ::std::vector<at::Tensor> tensor1_; |
| 15487 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15488 | at::functionalization::impl::sync(tensor1); |
| 15489 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15490 | } else { |
| 15491 | tensor1_ = tensor1.vec(); |
| 15492 | } |
| 15493 | |
| 15494 | ::std::vector<at::Tensor> tensor2_; |
| 15495 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15496 | at::functionalization::impl::sync(tensor2); |
| 15497 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15498 | } else { |
| 15499 | tensor2_ = tensor2.vec(); |
| 15500 | } |
| 15501 | |
| 15502 | at::Tensor scalars_; |
| 15503 | if (at::functionalization::impl::isFunctionalTensor(scalars)) { |
| 15504 | at::functionalization::impl::sync(scalars); |
| 15505 | scalars_ = at::functionalization::impl::from_functional_tensor(scalars); |
| 15506 | } else { |
| 15507 | scalars_ = scalars; |
| 15508 | } |
| 15509 | |
| 15510 | ::std::vector<at::Tensor> out_; |
| 15511 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15512 | at::functionalization::impl::sync(out); |
| 15513 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15514 | } else { |
| 15515 | out_ = out.vec(); |
| 15516 | } |
| 15517 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15518 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2) || at::functionalization::impl::isFunctionalTensor(scalars))) { |
| 15519 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15520 | TORCH_INTERNAL_ASSERT(false, |
| 15521 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15522 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15523 | } else { |
| 15524 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15525 | at::AutoDispatchSkipFunctionalize guard; |
| 15526 | at::_ops::_foreach_addcmul_Tensor_out::call(self_, tensor1_, tensor2_, scalars_, out_); |
| 15527 | ; |
| 15528 | } |
| 15529 | } else { |
| 15530 | ::std::vector<at::Tensor> tmp_output; |
| 15531 | { |
| 15532 | at::AutoDispatchSkipFunctionalize guard; |
| 15533 | tmp_output = at::_ops::_foreach_addcmul_Tensor::call(self_, tensor1_, tensor2_, scalars_); |
| 15534 | } |
| 15535 | at::functionalization::impl::replace_(out, tmp_output); |
| 15536 | at::functionalization::impl::commit_update(out); |
| 15537 | at::functionalization::impl::sync(out); |
| 15538 | |
| 15539 | } |
| 15540 | } |
| 15541 | |
| 15542 | void _foreach_addcmul__Tensor(c10::DispatchKeySet dispatchKeySet, at::TensorList self, at::TensorList tensor1, at::TensorList tensor2, const at::Tensor & scalars) { |
| 15543 | if (true) { |
| 15544 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15545 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15546 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15547 | auto self_meta = to_meta(self); |
| 15548 | auto tensor1_meta = to_meta(tensor1); |
| 15549 | auto tensor2_meta = to_meta(tensor2); |
| 15550 | auto scalars_meta = to_meta(scalars); |
| 15551 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15552 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15553 | at::_ops::_foreach_addcmul__Tensor::call(self_meta, tensor1_meta, tensor2_meta, scalars_meta); |
| 15554 | } |
| 15555 | |
| 15556 | ::std::vector<at::Tensor> self_; |
| 15557 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15558 | at::functionalization::impl::sync(self); |
| 15559 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15560 | } else { |
| 15561 | self_ = self.vec(); |
| 15562 | } |
| 15563 | |
| 15564 | ::std::vector<at::Tensor> tensor1_; |
| 15565 | if (at::functionalization::impl::isFunctionalTensor(tensor1)) { |
| 15566 | at::functionalization::impl::sync(tensor1); |
| 15567 | tensor1_ = at::functionalization::impl::from_functional_tensor(tensor1); |
| 15568 | } else { |
| 15569 | tensor1_ = tensor1.vec(); |
| 15570 | } |
| 15571 | |
| 15572 | ::std::vector<at::Tensor> tensor2_; |
| 15573 | if (at::functionalization::impl::isFunctionalTensor(tensor2)) { |
| 15574 | at::functionalization::impl::sync(tensor2); |
| 15575 | tensor2_ = at::functionalization::impl::from_functional_tensor(tensor2); |
| 15576 | } else { |
| 15577 | tensor2_ = tensor2.vec(); |
| 15578 | } |
| 15579 | |
| 15580 | at::Tensor scalars_; |
| 15581 | if (at::functionalization::impl::isFunctionalTensor(scalars)) { |
| 15582 | at::functionalization::impl::sync(scalars); |
| 15583 | scalars_ = at::functionalization::impl::from_functional_tensor(scalars); |
| 15584 | } else { |
| 15585 | scalars_ = scalars; |
| 15586 | } |
| 15587 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 15588 | if ((false || at::functionalization::impl::isFunctionalTensor(tensor1) || at::functionalization::impl::isFunctionalTensor(tensor2) || at::functionalization::impl::isFunctionalTensor(scalars))) { |
| 15589 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15590 | TORCH_INTERNAL_ASSERT(false, |
| 15591 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15592 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15593 | } else { |
| 15594 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15595 | at::AutoDispatchSkipFunctionalize guard; |
| 15596 | at::_ops::_foreach_addcmul__Tensor::call(self_, tensor1_, tensor2_, scalars_); |
| 15597 | ; |
| 15598 | } |
| 15599 | } else { |
| 15600 | ::std::vector<at::Tensor> tmp_output; |
| 15601 | { |
| 15602 | at::AutoDispatchSkipFunctionalize guard; |
| 15603 | tmp_output = at::_ops::_foreach_addcmul_Tensor::call(self_, tensor1_, tensor2_, scalars_); |
| 15604 | } |
| 15605 | at::functionalization::impl::replace_(self, tmp_output); |
| 15606 | at::functionalization::impl::commit_update(self); |
| 15607 | at::functionalization::impl::sync(self); |
| 15608 | |
| 15609 | } |
| 15610 | } |
| 15611 | |
| 15612 | at::Tensor & bucketize_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & boundaries, bool out_int32, bool right, at::Tensor & out) { |
| 15613 | if (false) { |
| 15614 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15615 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15616 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15617 | auto self_meta = to_meta(self); |
| 15618 | auto boundaries_meta = to_meta(boundaries); |
| 15619 | auto out_meta = to_meta(out); |
| 15620 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15621 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15622 | at::_ops::bucketize_Tensor_out::call(self_meta, boundaries_meta, out_int32, right, out_meta); |
| 15623 | } |
| 15624 | |
| 15625 | at::Tensor self_; |
| 15626 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15627 | at::functionalization::impl::sync(self); |
| 15628 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15629 | } else { |
| 15630 | self_ = self; |
| 15631 | } |
| 15632 | |
| 15633 | at::Tensor boundaries_; |
| 15634 | if (at::functionalization::impl::isFunctionalTensor(boundaries)) { |
| 15635 | at::functionalization::impl::sync(boundaries); |
| 15636 | boundaries_ = at::functionalization::impl::from_functional_tensor(boundaries); |
| 15637 | } else { |
| 15638 | boundaries_ = boundaries; |
| 15639 | } |
| 15640 | |
| 15641 | at::Tensor out_; |
| 15642 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15643 | at::functionalization::impl::sync(out); |
| 15644 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15645 | } else { |
| 15646 | out_ = out; |
| 15647 | } |
| 15648 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15649 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(boundaries))) { |
| 15650 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15651 | TORCH_INTERNAL_ASSERT(false, |
| 15652 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15653 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15654 | } else { |
| 15655 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15656 | at::AutoDispatchSkipFunctionalize guard; |
| 15657 | at::Tensor tmp_output = at::_ops::bucketize_Tensor_out::call(self_, boundaries_, out_int32, right, out_); |
| 15658 | return out;; |
| 15659 | } |
| 15660 | } else { |
| 15661 | at::Tensor tmp_output; |
| 15662 | { |
| 15663 | at::AutoDispatchSkipFunctionalize guard; |
| 15664 | tmp_output = at::_ops::bucketize_Tensor::call(self_, boundaries_, out_int32, right); |
| 15665 | } |
| 15666 | at::functionalization::impl::replace_(out, tmp_output); |
| 15667 | at::functionalization::impl::commit_update(out); |
| 15668 | at::functionalization::impl::sync(out); |
| 15669 | return out; |
| 15670 | } |
| 15671 | } |
| 15672 | |
| 15673 | at::Tensor & bucketize_out_Scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Scalar & self, const at::Tensor & boundaries, bool out_int32, bool right, at::Tensor & out) { |
| 15674 | if (false) { |
| 15675 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15676 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15677 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15678 | auto boundaries_meta = to_meta(boundaries); |
| 15679 | auto out_meta = to_meta(out); |
| 15680 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15681 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15682 | at::_ops::bucketize_Scalar_out::call(self, boundaries_meta, out_int32, right, out_meta); |
| 15683 | } |
| 15684 | |
| 15685 | at::Tensor boundaries_; |
| 15686 | if (at::functionalization::impl::isFunctionalTensor(boundaries)) { |
| 15687 | at::functionalization::impl::sync(boundaries); |
| 15688 | boundaries_ = at::functionalization::impl::from_functional_tensor(boundaries); |
| 15689 | } else { |
| 15690 | boundaries_ = boundaries; |
| 15691 | } |
| 15692 | |
| 15693 | at::Tensor out_; |
| 15694 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15695 | at::functionalization::impl::sync(out); |
| 15696 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15697 | } else { |
| 15698 | out_ = out; |
| 15699 | } |
| 15700 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15701 | if ((false || at::functionalization::impl::isFunctionalTensor(boundaries))) { |
| 15702 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15703 | TORCH_INTERNAL_ASSERT(false, |
| 15704 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15705 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15706 | } else { |
| 15707 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15708 | at::AutoDispatchSkipFunctionalize guard; |
| 15709 | at::Tensor tmp_output = at::_ops::bucketize_Scalar_out::call(self, boundaries_, out_int32, right, out_); |
| 15710 | return out;; |
| 15711 | } |
| 15712 | } else { |
| 15713 | at::Tensor tmp_output; |
| 15714 | { |
| 15715 | at::AutoDispatchSkipFunctionalize guard; |
| 15716 | tmp_output = at::_ops::bucketize_Scalar::call(self, boundaries_, out_int32, right); |
| 15717 | } |
| 15718 | at::functionalization::impl::replace_(out, tmp_output); |
| 15719 | at::functionalization::impl::commit_update(out); |
| 15720 | at::functionalization::impl::sync(out); |
| 15721 | return out; |
| 15722 | } |
| 15723 | } |
| 15724 | |
| 15725 | at::Tensor & mse_loss_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & target, int64_t reduction, at::Tensor & out) { |
| 15726 | if (false) { |
| 15727 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15728 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15729 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15730 | auto self_meta = to_meta(self); |
| 15731 | auto target_meta = to_meta(target); |
| 15732 | auto out_meta = to_meta(out); |
| 15733 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15734 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15735 | at::_ops::mse_loss_out::call(self_meta, target_meta, reduction, out_meta); |
| 15736 | } |
| 15737 | |
| 15738 | at::Tensor self_; |
| 15739 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15740 | at::functionalization::impl::sync(self); |
| 15741 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15742 | } else { |
| 15743 | self_ = self; |
| 15744 | } |
| 15745 | |
| 15746 | at::Tensor target_; |
| 15747 | if (at::functionalization::impl::isFunctionalTensor(target)) { |
| 15748 | at::functionalization::impl::sync(target); |
| 15749 | target_ = at::functionalization::impl::from_functional_tensor(target); |
| 15750 | } else { |
| 15751 | target_ = target; |
| 15752 | } |
| 15753 | |
| 15754 | at::Tensor out_; |
| 15755 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15756 | at::functionalization::impl::sync(out); |
| 15757 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15758 | } else { |
| 15759 | out_ = out; |
| 15760 | } |
| 15761 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15762 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(target))) { |
| 15763 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15764 | TORCH_INTERNAL_ASSERT(false, |
| 15765 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15766 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15767 | } else { |
| 15768 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15769 | at::AutoDispatchSkipFunctionalize guard; |
| 15770 | at::Tensor tmp_output = at::_ops::mse_loss_out::call(self_, target_, reduction, out_); |
| 15771 | return out;; |
| 15772 | } |
| 15773 | } else { |
| 15774 | at::Tensor tmp_output; |
| 15775 | { |
| 15776 | at::AutoDispatchSkipFunctionalize guard; |
| 15777 | tmp_output = at::_ops::mse_loss::call(self_, target_, reduction); |
| 15778 | } |
| 15779 | at::functionalization::impl::replace_(out, tmp_output); |
| 15780 | at::functionalization::impl::commit_update(out); |
| 15781 | at::functionalization::impl::sync(out); |
| 15782 | return out; |
| 15783 | } |
| 15784 | } |
| 15785 | |
| 15786 | at::Tensor & multi_margin_loss_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & target, const at::Scalar & p, const at::Scalar & margin, const c10::optional<at::Tensor> & weight, int64_t reduction, at::Tensor & out) { |
| 15787 | if (false) { |
| 15788 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15789 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15790 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15791 | auto self_meta = to_meta(self); |
| 15792 | auto target_meta = to_meta(target); |
| 15793 | auto weight_meta = to_meta(weight); |
| 15794 | auto out_meta = to_meta(out); |
| 15795 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15796 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15797 | at::_ops::multi_margin_loss_out::call(self_meta, target_meta, p, margin, weight_meta, reduction, out_meta); |
| 15798 | } |
| 15799 | |
| 15800 | at::Tensor self_; |
| 15801 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15802 | at::functionalization::impl::sync(self); |
| 15803 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15804 | } else { |
| 15805 | self_ = self; |
| 15806 | } |
| 15807 | |
| 15808 | at::Tensor target_; |
| 15809 | if (at::functionalization::impl::isFunctionalTensor(target)) { |
| 15810 | at::functionalization::impl::sync(target); |
| 15811 | target_ = at::functionalization::impl::from_functional_tensor(target); |
| 15812 | } else { |
| 15813 | target_ = target; |
| 15814 | } |
| 15815 | |
| 15816 | c10::optional<at::Tensor> weight_; |
| 15817 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 15818 | at::functionalization::impl::sync(weight); |
| 15819 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 15820 | } else { |
| 15821 | weight_ = weight; |
| 15822 | } |
| 15823 | |
| 15824 | at::Tensor out_; |
| 15825 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15826 | at::functionalization::impl::sync(out); |
| 15827 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15828 | } else { |
| 15829 | out_ = out; |
| 15830 | } |
| 15831 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15832 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(target) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 15833 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15834 | TORCH_INTERNAL_ASSERT(false, |
| 15835 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15836 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15837 | } else { |
| 15838 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15839 | at::AutoDispatchSkipFunctionalize guard; |
| 15840 | at::Tensor tmp_output = at::_ops::multi_margin_loss_out::call(self_, target_, p, margin, weight_, reduction, out_); |
| 15841 | return out;; |
| 15842 | } |
| 15843 | } else { |
| 15844 | at::Tensor tmp_output; |
| 15845 | { |
| 15846 | at::AutoDispatchSkipFunctionalize guard; |
| 15847 | tmp_output = at::_ops::multi_margin_loss::call(self_, target_, p, margin, weight_, reduction); |
| 15848 | } |
| 15849 | at::functionalization::impl::replace_(out, tmp_output); |
| 15850 | at::functionalization::impl::commit_update(out); |
| 15851 | at::functionalization::impl::sync(out); |
| 15852 | return out; |
| 15853 | } |
| 15854 | } |
| 15855 | |
| 15856 | at::Tensor & multilabel_margin_loss_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & target, int64_t reduction, at::Tensor & out) { |
| 15857 | if (false) { |
| 15858 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15859 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15860 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15861 | auto self_meta = to_meta(self); |
| 15862 | auto target_meta = to_meta(target); |
| 15863 | auto out_meta = to_meta(out); |
| 15864 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15865 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15866 | at::_ops::multilabel_margin_loss_out::call(self_meta, target_meta, reduction, out_meta); |
| 15867 | } |
| 15868 | |
| 15869 | at::Tensor self_; |
| 15870 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15871 | at::functionalization::impl::sync(self); |
| 15872 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15873 | } else { |
| 15874 | self_ = self; |
| 15875 | } |
| 15876 | |
| 15877 | at::Tensor target_; |
| 15878 | if (at::functionalization::impl::isFunctionalTensor(target)) { |
| 15879 | at::functionalization::impl::sync(target); |
| 15880 | target_ = at::functionalization::impl::from_functional_tensor(target); |
| 15881 | } else { |
| 15882 | target_ = target; |
| 15883 | } |
| 15884 | |
| 15885 | at::Tensor out_; |
| 15886 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 15887 | at::functionalization::impl::sync(out); |
| 15888 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 15889 | } else { |
| 15890 | out_ = out; |
| 15891 | } |
| 15892 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 15893 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(target))) { |
| 15894 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15895 | TORCH_INTERNAL_ASSERT(false, |
| 15896 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15897 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15898 | } else { |
| 15899 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15900 | at::AutoDispatchSkipFunctionalize guard; |
| 15901 | at::Tensor tmp_output = at::_ops::multilabel_margin_loss_out::call(self_, target_, reduction, out_); |
| 15902 | return out;; |
| 15903 | } |
| 15904 | } else { |
| 15905 | at::Tensor tmp_output; |
| 15906 | { |
| 15907 | at::AutoDispatchSkipFunctionalize guard; |
| 15908 | tmp_output = at::_ops::multilabel_margin_loss::call(self_, target_, reduction); |
| 15909 | } |
| 15910 | at::functionalization::impl::replace_(out, tmp_output); |
| 15911 | at::functionalization::impl::commit_update(out); |
| 15912 | at::functionalization::impl::sync(out); |
| 15913 | return out; |
| 15914 | } |
| 15915 | } |
| 15916 | |
| 15917 | ::std::tuple<at::Tensor &,at::Tensor &> multilabel_margin_loss_forward_out_output(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & target, int64_t reduction, at::Tensor & output, at::Tensor & is_target) { |
| 15918 | if (false) { |
| 15919 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15920 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15921 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15922 | auto self_meta = to_meta(self); |
| 15923 | auto target_meta = to_meta(target); |
| 15924 | auto output_meta = to_meta(output); |
| 15925 | auto is_target_meta = to_meta(is_target); |
| 15926 | at::AutoDispatchSkipFunctionalize func_guard; |
| 15927 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 15928 | at::_ops::multilabel_margin_loss_forward_output::call(self_meta, target_meta, reduction, output_meta, is_target_meta); |
| 15929 | } |
| 15930 | |
| 15931 | at::Tensor self_; |
| 15932 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 15933 | at::functionalization::impl::sync(self); |
| 15934 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 15935 | } else { |
| 15936 | self_ = self; |
| 15937 | } |
| 15938 | |
| 15939 | at::Tensor target_; |
| 15940 | if (at::functionalization::impl::isFunctionalTensor(target)) { |
| 15941 | at::functionalization::impl::sync(target); |
| 15942 | target_ = at::functionalization::impl::from_functional_tensor(target); |
| 15943 | } else { |
| 15944 | target_ = target; |
| 15945 | } |
| 15946 | |
| 15947 | at::Tensor output_; |
| 15948 | if (at::functionalization::impl::isFunctionalTensor(output)) { |
| 15949 | at::functionalization::impl::sync(output); |
| 15950 | output_ = at::functionalization::impl::from_functional_tensor(output); |
| 15951 | } else { |
| 15952 | output_ = output; |
| 15953 | } |
| 15954 | |
| 15955 | at::Tensor is_target_; |
| 15956 | if (at::functionalization::impl::isFunctionalTensor(is_target)) { |
| 15957 | at::functionalization::impl::sync(is_target); |
| 15958 | is_target_ = at::functionalization::impl::from_functional_tensor(is_target); |
| 15959 | } else { |
| 15960 | is_target_ = is_target; |
| 15961 | } |
| 15962 | if (!(true && at::functionalization::impl::isFunctionalTensor(output) && at::functionalization::impl::isFunctionalTensor(is_target))) { |
| 15963 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(target))) { |
| 15964 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 15965 | TORCH_INTERNAL_ASSERT(false, |
| 15966 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 15967 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 15968 | } else { |
| 15969 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 15970 | at::AutoDispatchSkipFunctionalize guard; |
| 15971 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::multilabel_margin_loss_forward_output::call(self_, target_, reduction, output_, is_target_); |
| 15972 | return ::std::tuple<at::Tensor &,at::Tensor &>(output, is_target);; |
| 15973 | } |
| 15974 | } else { |
| 15975 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 15976 | { |
| 15977 | at::AutoDispatchSkipFunctionalize guard; |
| 15978 | tmp_output = at::_ops::multilabel_margin_loss_forward::call(self_, target_, reduction); |
| 15979 | } |
| 15980 | at::functionalization::impl::replace_(output, std::get<0>(tmp_output)); |
| 15981 | at::functionalization::impl::commit_update(output); |
| 15982 | at::functionalization::impl::sync(output); |
| 15983 | at::functionalization::impl::replace_(is_target, std::get<1>(tmp_output)); |
| 15984 | at::functionalization::impl::commit_update(is_target); |
| 15985 | at::functionalization::impl::sync(is_target); |
| 15986 | return ::std::tuple<at::Tensor &,at::Tensor &>(output, is_target); |
| 15987 | } |
| 15988 | } |
| 15989 | |
| 15990 | ::std::tuple<at::Tensor &,at::Tensor &> nll_loss2d_forward_out_output(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & target, const c10::optional<at::Tensor> & weight, int64_t reduction, c10::SymInt ignore_index, at::Tensor & output, at::Tensor & total_weight) { |
| 15991 | if (false) { |
| 15992 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 15993 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 15994 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 15995 | auto self_meta = to_meta(self); |
| 15996 | auto target_meta = to_meta(target); |
| 15997 | auto weight_meta = to_meta(weight); |
| 15998 | auto output_meta = to_meta(output); |
| 15999 | auto total_weight_meta = to_meta(total_weight); |
| 16000 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16001 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16002 | at::_ops::nll_loss2d_forward_output::call(self_meta, target_meta, weight_meta, reduction, ignore_index, output_meta, total_weight_meta); |
| 16003 | } |
| 16004 | |
| 16005 | at::Tensor self_; |
| 16006 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16007 | at::functionalization::impl::sync(self); |
| 16008 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16009 | } else { |
| 16010 | self_ = self; |
| 16011 | } |
| 16012 | |
| 16013 | at::Tensor target_; |
| 16014 | if (at::functionalization::impl::isFunctionalTensor(target)) { |
| 16015 | at::functionalization::impl::sync(target); |
| 16016 | target_ = at::functionalization::impl::from_functional_tensor(target); |
| 16017 | } else { |
| 16018 | target_ = target; |
| 16019 | } |
| 16020 | |
| 16021 | c10::optional<at::Tensor> weight_; |
| 16022 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 16023 | at::functionalization::impl::sync(weight); |
| 16024 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 16025 | } else { |
| 16026 | weight_ = weight; |
| 16027 | } |
| 16028 | |
| 16029 | at::Tensor output_; |
| 16030 | if (at::functionalization::impl::isFunctionalTensor(output)) { |
| 16031 | at::functionalization::impl::sync(output); |
| 16032 | output_ = at::functionalization::impl::from_functional_tensor(output); |
| 16033 | } else { |
| 16034 | output_ = output; |
| 16035 | } |
| 16036 | |
| 16037 | at::Tensor total_weight_; |
| 16038 | if (at::functionalization::impl::isFunctionalTensor(total_weight)) { |
| 16039 | at::functionalization::impl::sync(total_weight); |
| 16040 | total_weight_ = at::functionalization::impl::from_functional_tensor(total_weight); |
| 16041 | } else { |
| 16042 | total_weight_ = total_weight; |
| 16043 | } |
| 16044 | if (!(true && at::functionalization::impl::isFunctionalTensor(output) && at::functionalization::impl::isFunctionalTensor(total_weight))) { |
| 16045 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(target) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 16046 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16047 | TORCH_INTERNAL_ASSERT(false, |
| 16048 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16049 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16050 | } else { |
| 16051 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16052 | at::AutoDispatchSkipFunctionalize guard; |
| 16053 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::nll_loss2d_forward_output::call(self_, target_, weight_, reduction, ignore_index, output_, total_weight_); |
| 16054 | return ::std::tuple<at::Tensor &,at::Tensor &>(output, total_weight);; |
| 16055 | } |
| 16056 | } else { |
| 16057 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 16058 | { |
| 16059 | at::AutoDispatchSkipFunctionalize guard; |
| 16060 | tmp_output = at::_ops::nll_loss2d_forward::call(self_, target_, weight_, reduction, ignore_index); |
| 16061 | } |
| 16062 | at::functionalization::impl::replace_(output, std::get<0>(tmp_output)); |
| 16063 | at::functionalization::impl::commit_update(output); |
| 16064 | at::functionalization::impl::sync(output); |
| 16065 | at::functionalization::impl::replace_(total_weight, std::get<1>(tmp_output)); |
| 16066 | at::functionalization::impl::commit_update(total_weight); |
| 16067 | at::functionalization::impl::sync(total_weight); |
| 16068 | return ::std::tuple<at::Tensor &,at::Tensor &>(output, total_weight); |
| 16069 | } |
| 16070 | } |
| 16071 | |
| 16072 | at::Tensor & soft_margin_loss_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & target, int64_t reduction, at::Tensor & out) { |
| 16073 | if (false) { |
| 16074 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16075 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16076 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16077 | auto self_meta = to_meta(self); |
| 16078 | auto target_meta = to_meta(target); |
| 16079 | auto out_meta = to_meta(out); |
| 16080 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16081 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16082 | at::_ops::soft_margin_loss_out::call(self_meta, target_meta, reduction, out_meta); |
| 16083 | } |
| 16084 | |
| 16085 | at::Tensor self_; |
| 16086 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16087 | at::functionalization::impl::sync(self); |
| 16088 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16089 | } else { |
| 16090 | self_ = self; |
| 16091 | } |
| 16092 | |
| 16093 | at::Tensor target_; |
| 16094 | if (at::functionalization::impl::isFunctionalTensor(target)) { |
| 16095 | at::functionalization::impl::sync(target); |
| 16096 | target_ = at::functionalization::impl::from_functional_tensor(target); |
| 16097 | } else { |
| 16098 | target_ = target; |
| 16099 | } |
| 16100 | |
| 16101 | at::Tensor out_; |
| 16102 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 16103 | at::functionalization::impl::sync(out); |
| 16104 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 16105 | } else { |
| 16106 | out_ = out; |
| 16107 | } |
| 16108 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 16109 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(target))) { |
| 16110 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16111 | TORCH_INTERNAL_ASSERT(false, |
| 16112 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16113 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16114 | } else { |
| 16115 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16116 | at::AutoDispatchSkipFunctionalize guard; |
| 16117 | at::Tensor tmp_output = at::_ops::soft_margin_loss_out::call(self_, target_, reduction, out_); |
| 16118 | return out;; |
| 16119 | } |
| 16120 | } else { |
| 16121 | at::Tensor tmp_output; |
| 16122 | { |
| 16123 | at::AutoDispatchSkipFunctionalize guard; |
| 16124 | tmp_output = at::_ops::soft_margin_loss::call(self_, target_, reduction); |
| 16125 | } |
| 16126 | at::functionalization::impl::replace_(out, tmp_output); |
| 16127 | at::functionalization::impl::commit_update(out); |
| 16128 | at::functionalization::impl::sync(out); |
| 16129 | return out; |
| 16130 | } |
| 16131 | } |
| 16132 | |
| 16133 | at::Tensor & glu_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, int64_t dim, at::Tensor & grad_input) { |
| 16134 | if (false) { |
| 16135 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16136 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16137 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16138 | auto grad_output_meta = to_meta(grad_output); |
| 16139 | auto self_meta = to_meta(self); |
| 16140 | auto grad_input_meta = to_meta(grad_input); |
| 16141 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16142 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16143 | at::_ops::glu_backward_grad_input::call(grad_output_meta, self_meta, dim, grad_input_meta); |
| 16144 | } |
| 16145 | |
| 16146 | at::Tensor grad_output_; |
| 16147 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 16148 | at::functionalization::impl::sync(grad_output); |
| 16149 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 16150 | } else { |
| 16151 | grad_output_ = grad_output; |
| 16152 | } |
| 16153 | |
| 16154 | at::Tensor self_; |
| 16155 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16156 | at::functionalization::impl::sync(self); |
| 16157 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16158 | } else { |
| 16159 | self_ = self; |
| 16160 | } |
| 16161 | |
| 16162 | at::Tensor grad_input_; |
| 16163 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 16164 | at::functionalization::impl::sync(grad_input); |
| 16165 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 16166 | } else { |
| 16167 | grad_input_ = grad_input; |
| 16168 | } |
| 16169 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 16170 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self))) { |
| 16171 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16172 | TORCH_INTERNAL_ASSERT(false, |
| 16173 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16174 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16175 | } else { |
| 16176 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16177 | at::AutoDispatchSkipFunctionalize guard; |
| 16178 | at::Tensor tmp_output = at::_ops::glu_backward_grad_input::call(grad_output_, self_, dim, grad_input_); |
| 16179 | return grad_input;; |
| 16180 | } |
| 16181 | } else { |
| 16182 | at::Tensor tmp_output; |
| 16183 | { |
| 16184 | at::AutoDispatchSkipFunctionalize guard; |
| 16185 | tmp_output = at::_ops::glu_backward::call(grad_output_, self_, dim); |
| 16186 | } |
| 16187 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 16188 | at::functionalization::impl::commit_update(grad_input); |
| 16189 | at::functionalization::impl::sync(grad_input); |
| 16190 | return grad_input; |
| 16191 | } |
| 16192 | } |
| 16193 | |
| 16194 | at::Tensor & glu_backward_jvp_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_x, const at::Tensor & grad_glu, const at::Tensor & x, const at::Tensor & dgrad_glu, const at::Tensor & dx, int64_t dim, at::Tensor & out) { |
| 16195 | if (false) { |
| 16196 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16197 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16198 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16199 | auto grad_x_meta = to_meta(grad_x); |
| 16200 | auto grad_glu_meta = to_meta(grad_glu); |
| 16201 | auto x_meta = to_meta(x); |
| 16202 | auto dgrad_glu_meta = to_meta(dgrad_glu); |
| 16203 | auto dx_meta = to_meta(dx); |
| 16204 | auto out_meta = to_meta(out); |
| 16205 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16206 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16207 | at::_ops::glu_backward_jvp_out::call(grad_x_meta, grad_glu_meta, x_meta, dgrad_glu_meta, dx_meta, dim, out_meta); |
| 16208 | } |
| 16209 | |
| 16210 | at::Tensor grad_x_; |
| 16211 | if (at::functionalization::impl::isFunctionalTensor(grad_x)) { |
| 16212 | at::functionalization::impl::sync(grad_x); |
| 16213 | grad_x_ = at::functionalization::impl::from_functional_tensor(grad_x); |
| 16214 | } else { |
| 16215 | grad_x_ = grad_x; |
| 16216 | } |
| 16217 | |
| 16218 | at::Tensor grad_glu_; |
| 16219 | if (at::functionalization::impl::isFunctionalTensor(grad_glu)) { |
| 16220 | at::functionalization::impl::sync(grad_glu); |
| 16221 | grad_glu_ = at::functionalization::impl::from_functional_tensor(grad_glu); |
| 16222 | } else { |
| 16223 | grad_glu_ = grad_glu; |
| 16224 | } |
| 16225 | |
| 16226 | at::Tensor x_; |
| 16227 | if (at::functionalization::impl::isFunctionalTensor(x)) { |
| 16228 | at::functionalization::impl::sync(x); |
| 16229 | x_ = at::functionalization::impl::from_functional_tensor(x); |
| 16230 | } else { |
| 16231 | x_ = x; |
| 16232 | } |
| 16233 | |
| 16234 | at::Tensor dgrad_glu_; |
| 16235 | if (at::functionalization::impl::isFunctionalTensor(dgrad_glu)) { |
| 16236 | at::functionalization::impl::sync(dgrad_glu); |
| 16237 | dgrad_glu_ = at::functionalization::impl::from_functional_tensor(dgrad_glu); |
| 16238 | } else { |
| 16239 | dgrad_glu_ = dgrad_glu; |
| 16240 | } |
| 16241 | |
| 16242 | at::Tensor dx_; |
| 16243 | if (at::functionalization::impl::isFunctionalTensor(dx)) { |
| 16244 | at::functionalization::impl::sync(dx); |
| 16245 | dx_ = at::functionalization::impl::from_functional_tensor(dx); |
| 16246 | } else { |
| 16247 | dx_ = dx; |
| 16248 | } |
| 16249 | |
| 16250 | at::Tensor out_; |
| 16251 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 16252 | at::functionalization::impl::sync(out); |
| 16253 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 16254 | } else { |
| 16255 | out_ = out; |
| 16256 | } |
| 16257 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 16258 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_x) || at::functionalization::impl::isFunctionalTensor(grad_glu) || at::functionalization::impl::isFunctionalTensor(x) || at::functionalization::impl::isFunctionalTensor(dgrad_glu) || at::functionalization::impl::isFunctionalTensor(dx))) { |
| 16259 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16260 | TORCH_INTERNAL_ASSERT(false, |
| 16261 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16262 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16263 | } else { |
| 16264 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16265 | at::AutoDispatchSkipFunctionalize guard; |
| 16266 | at::Tensor tmp_output = at::_ops::glu_backward_jvp_out::call(grad_x_, grad_glu_, x_, dgrad_glu_, dx_, dim, out_); |
| 16267 | return out;; |
| 16268 | } |
| 16269 | } else { |
| 16270 | at::Tensor tmp_output; |
| 16271 | { |
| 16272 | at::AutoDispatchSkipFunctionalize guard; |
| 16273 | tmp_output = at::_ops::glu_backward_jvp::call(grad_x_, grad_glu_, x_, dgrad_glu_, dx_, dim); |
| 16274 | } |
| 16275 | at::functionalization::impl::replace_(out, tmp_output); |
| 16276 | at::functionalization::impl::commit_update(out); |
| 16277 | at::functionalization::impl::sync(out); |
| 16278 | return out; |
| 16279 | } |
| 16280 | } |
| 16281 | |
| 16282 | at::Tensor & hardtanh_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & min_val, const at::Scalar & max_val, at::Tensor & out) { |
| 16283 | if (false) { |
| 16284 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16285 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16286 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16287 | auto self_meta = to_meta(self); |
| 16288 | auto out_meta = to_meta(out); |
| 16289 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16290 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16291 | at::_ops::hardtanh_out::call(self_meta, min_val, max_val, out_meta); |
| 16292 | } |
| 16293 | |
| 16294 | at::Tensor self_; |
| 16295 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16296 | at::functionalization::impl::sync(self); |
| 16297 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16298 | } else { |
| 16299 | self_ = self; |
| 16300 | } |
| 16301 | |
| 16302 | at::Tensor out_; |
| 16303 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 16304 | at::functionalization::impl::sync(out); |
| 16305 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 16306 | } else { |
| 16307 | out_ = out; |
| 16308 | } |
| 16309 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 16310 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 16311 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16312 | TORCH_INTERNAL_ASSERT(false, |
| 16313 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16314 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16315 | } else { |
| 16316 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16317 | at::AutoDispatchSkipFunctionalize guard; |
| 16318 | at::Tensor tmp_output = at::_ops::hardtanh_out::call(self_, min_val, max_val, out_); |
| 16319 | return out;; |
| 16320 | } |
| 16321 | } else { |
| 16322 | at::Tensor tmp_output; |
| 16323 | { |
| 16324 | at::AutoDispatchSkipFunctionalize guard; |
| 16325 | tmp_output = at::_ops::hardtanh::call(self_, min_val, max_val); |
| 16326 | } |
| 16327 | at::functionalization::impl::replace_(out, tmp_output); |
| 16328 | at::functionalization::impl::commit_update(out); |
| 16329 | at::functionalization::impl::sync(out); |
| 16330 | return out; |
| 16331 | } |
| 16332 | } |
| 16333 | |
| 16334 | at::Tensor & hardtanh_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & min_val, const at::Scalar & max_val) { |
| 16335 | if (true) { |
| 16336 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16337 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16338 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16339 | auto self_meta = to_meta(self); |
| 16340 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16341 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16342 | at::_ops::hardtanh_::call(self_meta, min_val, max_val); |
| 16343 | } |
| 16344 | |
| 16345 | at::Tensor self_; |
| 16346 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16347 | at::functionalization::impl::sync(self); |
| 16348 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16349 | } else { |
| 16350 | self_ = self; |
| 16351 | } |
| 16352 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 16353 | if ((false)) { |
| 16354 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16355 | TORCH_INTERNAL_ASSERT(false, |
| 16356 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16357 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16358 | } else { |
| 16359 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16360 | at::AutoDispatchSkipFunctionalize guard; |
| 16361 | at::Tensor tmp_output = at::_ops::hardtanh_::call(self_, min_val, max_val); |
| 16362 | return self;; |
| 16363 | } |
| 16364 | } else { |
| 16365 | at::Tensor tmp_output; |
| 16366 | { |
| 16367 | at::AutoDispatchSkipFunctionalize guard; |
| 16368 | tmp_output = at::_ops::hardtanh::call(self_, min_val, max_val); |
| 16369 | } |
| 16370 | at::functionalization::impl::replace_(self, tmp_output); |
| 16371 | at::functionalization::impl::commit_update(self); |
| 16372 | at::functionalization::impl::sync(self); |
| 16373 | return self; |
| 16374 | } |
| 16375 | } |
| 16376 | |
| 16377 | at::Tensor & hardtanh_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, const at::Scalar & min_val, const at::Scalar & max_val, at::Tensor & grad_input) { |
| 16378 | if (false) { |
| 16379 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16380 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16381 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16382 | auto grad_output_meta = to_meta(grad_output); |
| 16383 | auto self_meta = to_meta(self); |
| 16384 | auto grad_input_meta = to_meta(grad_input); |
| 16385 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16386 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16387 | at::_ops::hardtanh_backward_grad_input::call(grad_output_meta, self_meta, min_val, max_val, grad_input_meta); |
| 16388 | } |
| 16389 | |
| 16390 | at::Tensor grad_output_; |
| 16391 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 16392 | at::functionalization::impl::sync(grad_output); |
| 16393 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 16394 | } else { |
| 16395 | grad_output_ = grad_output; |
| 16396 | } |
| 16397 | |
| 16398 | at::Tensor self_; |
| 16399 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16400 | at::functionalization::impl::sync(self); |
| 16401 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16402 | } else { |
| 16403 | self_ = self; |
| 16404 | } |
| 16405 | |
| 16406 | at::Tensor grad_input_; |
| 16407 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 16408 | at::functionalization::impl::sync(grad_input); |
| 16409 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 16410 | } else { |
| 16411 | grad_input_ = grad_input; |
| 16412 | } |
| 16413 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 16414 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self))) { |
| 16415 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16416 | TORCH_INTERNAL_ASSERT(false, |
| 16417 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16418 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16419 | } else { |
| 16420 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16421 | at::AutoDispatchSkipFunctionalize guard; |
| 16422 | at::Tensor tmp_output = at::_ops::hardtanh_backward_grad_input::call(grad_output_, self_, min_val, max_val, grad_input_); |
| 16423 | return grad_input;; |
| 16424 | } |
| 16425 | } else { |
| 16426 | at::Tensor tmp_output; |
| 16427 | { |
| 16428 | at::AutoDispatchSkipFunctionalize guard; |
| 16429 | tmp_output = at::_ops::hardtanh_backward::call(grad_output_, self_, min_val, max_val); |
| 16430 | } |
| 16431 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 16432 | at::functionalization::impl::commit_update(grad_input); |
| 16433 | at::functionalization::impl::sync(grad_input); |
| 16434 | return grad_input; |
| 16435 | } |
| 16436 | } |
| 16437 | |
| 16438 | at::Tensor & leaky_relu_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & negative_slope, at::Tensor & out) { |
| 16439 | if (false) { |
| 16440 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16441 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16442 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16443 | auto self_meta = to_meta(self); |
| 16444 | auto out_meta = to_meta(out); |
| 16445 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16446 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16447 | at::_ops::leaky_relu_out::call(self_meta, negative_slope, out_meta); |
| 16448 | } |
| 16449 | |
| 16450 | at::Tensor self_; |
| 16451 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16452 | at::functionalization::impl::sync(self); |
| 16453 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16454 | } else { |
| 16455 | self_ = self; |
| 16456 | } |
| 16457 | |
| 16458 | at::Tensor out_; |
| 16459 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 16460 | at::functionalization::impl::sync(out); |
| 16461 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 16462 | } else { |
| 16463 | out_ = out; |
| 16464 | } |
| 16465 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 16466 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 16467 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16468 | TORCH_INTERNAL_ASSERT(false, |
| 16469 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16470 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16471 | } else { |
| 16472 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16473 | at::AutoDispatchSkipFunctionalize guard; |
| 16474 | at::Tensor tmp_output = at::_ops::leaky_relu_out::call(self_, negative_slope, out_); |
| 16475 | return out;; |
| 16476 | } |
| 16477 | } else { |
| 16478 | at::Tensor tmp_output; |
| 16479 | { |
| 16480 | at::AutoDispatchSkipFunctionalize guard; |
| 16481 | tmp_output = at::_ops::leaky_relu::call(self_, negative_slope); |
| 16482 | } |
| 16483 | at::functionalization::impl::replace_(out, tmp_output); |
| 16484 | at::functionalization::impl::commit_update(out); |
| 16485 | at::functionalization::impl::sync(out); |
| 16486 | return out; |
| 16487 | } |
| 16488 | } |
| 16489 | |
| 16490 | at::Tensor & leaky_relu_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self, const at::Scalar & negative_slope) { |
| 16491 | if (true) { |
| 16492 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16493 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16494 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16495 | auto self_meta = to_meta(self); |
| 16496 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16497 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16498 | at::_ops::leaky_relu_::call(self_meta, negative_slope); |
| 16499 | } |
| 16500 | |
| 16501 | at::Tensor self_; |
| 16502 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16503 | at::functionalization::impl::sync(self); |
| 16504 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16505 | } else { |
| 16506 | self_ = self; |
| 16507 | } |
| 16508 | if (!(true && at::functionalization::impl::isFunctionalTensor(self))) { |
| 16509 | if ((false)) { |
| 16510 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16511 | TORCH_INTERNAL_ASSERT(false, |
| 16512 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16513 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16514 | } else { |
| 16515 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16516 | at::AutoDispatchSkipFunctionalize guard; |
| 16517 | at::Tensor tmp_output = at::_ops::leaky_relu_::call(self_, negative_slope); |
| 16518 | return self;; |
| 16519 | } |
| 16520 | } else { |
| 16521 | at::Tensor tmp_output; |
| 16522 | { |
| 16523 | at::AutoDispatchSkipFunctionalize guard; |
| 16524 | tmp_output = at::_ops::leaky_relu::call(self_, negative_slope); |
| 16525 | } |
| 16526 | at::functionalization::impl::replace_(self, tmp_output); |
| 16527 | at::functionalization::impl::commit_update(self); |
| 16528 | at::functionalization::impl::sync(self); |
| 16529 | return self; |
| 16530 | } |
| 16531 | } |
| 16532 | |
| 16533 | at::Tensor & log_sigmoid_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, const at::Tensor & buffer, at::Tensor & grad_input) { |
| 16534 | if (false) { |
| 16535 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16536 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16537 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16538 | auto grad_output_meta = to_meta(grad_output); |
| 16539 | auto self_meta = to_meta(self); |
| 16540 | auto buffer_meta = to_meta(buffer); |
| 16541 | auto grad_input_meta = to_meta(grad_input); |
| 16542 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16543 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16544 | at::_ops::log_sigmoid_backward_grad_input::call(grad_output_meta, self_meta, buffer_meta, grad_input_meta); |
| 16545 | } |
| 16546 | |
| 16547 | at::Tensor grad_output_; |
| 16548 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 16549 | at::functionalization::impl::sync(grad_output); |
| 16550 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 16551 | } else { |
| 16552 | grad_output_ = grad_output; |
| 16553 | } |
| 16554 | |
| 16555 | at::Tensor self_; |
| 16556 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16557 | at::functionalization::impl::sync(self); |
| 16558 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16559 | } else { |
| 16560 | self_ = self; |
| 16561 | } |
| 16562 | |
| 16563 | at::Tensor buffer_; |
| 16564 | if (at::functionalization::impl::isFunctionalTensor(buffer)) { |
| 16565 | at::functionalization::impl::sync(buffer); |
| 16566 | buffer_ = at::functionalization::impl::from_functional_tensor(buffer); |
| 16567 | } else { |
| 16568 | buffer_ = buffer; |
| 16569 | } |
| 16570 | |
| 16571 | at::Tensor grad_input_; |
| 16572 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 16573 | at::functionalization::impl::sync(grad_input); |
| 16574 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 16575 | } else { |
| 16576 | grad_input_ = grad_input; |
| 16577 | } |
| 16578 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 16579 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(buffer))) { |
| 16580 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16581 | TORCH_INTERNAL_ASSERT(false, |
| 16582 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16583 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16584 | } else { |
| 16585 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16586 | at::AutoDispatchSkipFunctionalize guard; |
| 16587 | at::Tensor tmp_output = at::_ops::log_sigmoid_backward_grad_input::call(grad_output_, self_, buffer_, grad_input_); |
| 16588 | return grad_input;; |
| 16589 | } |
| 16590 | } else { |
| 16591 | at::Tensor tmp_output; |
| 16592 | { |
| 16593 | at::AutoDispatchSkipFunctionalize guard; |
| 16594 | tmp_output = at::_ops::log_sigmoid_backward::call(grad_output_, self_, buffer_); |
| 16595 | } |
| 16596 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 16597 | at::functionalization::impl::commit_update(grad_input); |
| 16598 | at::functionalization::impl::sync(grad_input); |
| 16599 | return grad_input; |
| 16600 | } |
| 16601 | } |
| 16602 | |
| 16603 | at::Tensor & softplus_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & beta, const at::Scalar & threshold, at::Tensor & out) { |
| 16604 | if (false) { |
| 16605 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16606 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16607 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16608 | auto self_meta = to_meta(self); |
| 16609 | auto out_meta = to_meta(out); |
| 16610 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16611 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16612 | at::_ops::softplus_out::call(self_meta, beta, threshold, out_meta); |
| 16613 | } |
| 16614 | |
| 16615 | at::Tensor self_; |
| 16616 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16617 | at::functionalization::impl::sync(self); |
| 16618 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16619 | } else { |
| 16620 | self_ = self; |
| 16621 | } |
| 16622 | |
| 16623 | at::Tensor out_; |
| 16624 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 16625 | at::functionalization::impl::sync(out); |
| 16626 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 16627 | } else { |
| 16628 | out_ = out; |
| 16629 | } |
| 16630 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 16631 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 16632 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16633 | TORCH_INTERNAL_ASSERT(false, |
| 16634 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16635 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16636 | } else { |
| 16637 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16638 | at::AutoDispatchSkipFunctionalize guard; |
| 16639 | at::Tensor tmp_output = at::_ops::softplus_out::call(self_, beta, threshold, out_); |
| 16640 | return out;; |
| 16641 | } |
| 16642 | } else { |
| 16643 | at::Tensor tmp_output; |
| 16644 | { |
| 16645 | at::AutoDispatchSkipFunctionalize guard; |
| 16646 | tmp_output = at::_ops::softplus::call(self_, beta, threshold); |
| 16647 | } |
| 16648 | at::functionalization::impl::replace_(out, tmp_output); |
| 16649 | at::functionalization::impl::commit_update(out); |
| 16650 | at::functionalization::impl::sync(out); |
| 16651 | return out; |
| 16652 | } |
| 16653 | } |
| 16654 | |
| 16655 | at::Tensor & adaptive_avg_pool2d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::SymIntArrayRef output_size, at::Tensor & out) { |
| 16656 | if (false) { |
| 16657 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16658 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16659 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16660 | auto self_meta = to_meta(self); |
| 16661 | auto out_meta = to_meta(out); |
| 16662 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16663 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16664 | at::_ops::adaptive_avg_pool2d_out::call(self_meta, output_size, out_meta); |
| 16665 | } |
| 16666 | |
| 16667 | at::Tensor self_; |
| 16668 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16669 | at::functionalization::impl::sync(self); |
| 16670 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16671 | } else { |
| 16672 | self_ = self; |
| 16673 | } |
| 16674 | |
| 16675 | at::Tensor out_; |
| 16676 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 16677 | at::functionalization::impl::sync(out); |
| 16678 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 16679 | } else { |
| 16680 | out_ = out; |
| 16681 | } |
| 16682 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 16683 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 16684 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16685 | TORCH_INTERNAL_ASSERT(false, |
| 16686 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16687 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16688 | } else { |
| 16689 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16690 | at::AutoDispatchSkipFunctionalize guard; |
| 16691 | at::Tensor tmp_output = at::_ops::adaptive_avg_pool2d_out::call(self_, output_size, out_); |
| 16692 | return out;; |
| 16693 | } |
| 16694 | } else { |
| 16695 | at::Tensor tmp_output; |
| 16696 | { |
| 16697 | at::AutoDispatchSkipFunctionalize guard; |
| 16698 | tmp_output = at::_ops::adaptive_avg_pool2d::call(self_, output_size); |
| 16699 | } |
| 16700 | at::functionalization::impl::replace_(out, tmp_output); |
| 16701 | at::functionalization::impl::commit_update(out); |
| 16702 | at::functionalization::impl::sync(out); |
| 16703 | return out; |
| 16704 | } |
| 16705 | } |
| 16706 | |
| 16707 | at::Tensor & _adaptive_avg_pool3d_backward_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, at::Tensor & out) { |
| 16708 | if (false) { |
| 16709 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16710 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16711 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16712 | auto grad_output_meta = to_meta(grad_output); |
| 16713 | auto self_meta = to_meta(self); |
| 16714 | auto out_meta = to_meta(out); |
| 16715 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16716 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16717 | at::_ops::_adaptive_avg_pool3d_backward_out::call(grad_output_meta, self_meta, out_meta); |
| 16718 | } |
| 16719 | |
| 16720 | at::Tensor grad_output_; |
| 16721 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 16722 | at::functionalization::impl::sync(grad_output); |
| 16723 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 16724 | } else { |
| 16725 | grad_output_ = grad_output; |
| 16726 | } |
| 16727 | |
| 16728 | at::Tensor self_; |
| 16729 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16730 | at::functionalization::impl::sync(self); |
| 16731 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16732 | } else { |
| 16733 | self_ = self; |
| 16734 | } |
| 16735 | |
| 16736 | at::Tensor out_; |
| 16737 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 16738 | at::functionalization::impl::sync(out); |
| 16739 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 16740 | } else { |
| 16741 | out_ = out; |
| 16742 | } |
| 16743 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 16744 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self))) { |
| 16745 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16746 | TORCH_INTERNAL_ASSERT(false, |
| 16747 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16748 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16749 | } else { |
| 16750 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16751 | at::AutoDispatchSkipFunctionalize guard; |
| 16752 | at::Tensor tmp_output = at::_ops::_adaptive_avg_pool3d_backward_out::call(grad_output_, self_, out_); |
| 16753 | return out;; |
| 16754 | } |
| 16755 | } else { |
| 16756 | at::Tensor tmp_output; |
| 16757 | { |
| 16758 | at::AutoDispatchSkipFunctionalize guard; |
| 16759 | tmp_output = at::_ops::_adaptive_avg_pool3d_backward::call(grad_output_, self_); |
| 16760 | } |
| 16761 | at::functionalization::impl::replace_(out, tmp_output); |
| 16762 | at::functionalization::impl::commit_update(out); |
| 16763 | at::functionalization::impl::sync(out); |
| 16764 | return out; |
| 16765 | } |
| 16766 | } |
| 16767 | |
| 16768 | at::Tensor & adaptive_max_pool2d_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, const at::Tensor & indices, at::Tensor & grad_input) { |
| 16769 | if (false) { |
| 16770 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16771 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16772 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16773 | auto grad_output_meta = to_meta(grad_output); |
| 16774 | auto self_meta = to_meta(self); |
| 16775 | auto indices_meta = to_meta(indices); |
| 16776 | auto grad_input_meta = to_meta(grad_input); |
| 16777 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16778 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16779 | at::_ops::adaptive_max_pool2d_backward_grad_input::call(grad_output_meta, self_meta, indices_meta, grad_input_meta); |
| 16780 | } |
| 16781 | |
| 16782 | at::Tensor grad_output_; |
| 16783 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 16784 | at::functionalization::impl::sync(grad_output); |
| 16785 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 16786 | } else { |
| 16787 | grad_output_ = grad_output; |
| 16788 | } |
| 16789 | |
| 16790 | at::Tensor self_; |
| 16791 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16792 | at::functionalization::impl::sync(self); |
| 16793 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16794 | } else { |
| 16795 | self_ = self; |
| 16796 | } |
| 16797 | |
| 16798 | at::Tensor indices_; |
| 16799 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 16800 | at::functionalization::impl::sync(indices); |
| 16801 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 16802 | } else { |
| 16803 | indices_ = indices; |
| 16804 | } |
| 16805 | |
| 16806 | at::Tensor grad_input_; |
| 16807 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 16808 | at::functionalization::impl::sync(grad_input); |
| 16809 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 16810 | } else { |
| 16811 | grad_input_ = grad_input; |
| 16812 | } |
| 16813 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 16814 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(indices))) { |
| 16815 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16816 | TORCH_INTERNAL_ASSERT(false, |
| 16817 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16818 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16819 | } else { |
| 16820 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16821 | at::AutoDispatchSkipFunctionalize guard; |
| 16822 | at::Tensor tmp_output = at::_ops::adaptive_max_pool2d_backward_grad_input::call(grad_output_, self_, indices_, grad_input_); |
| 16823 | return grad_input;; |
| 16824 | } |
| 16825 | } else { |
| 16826 | at::Tensor tmp_output; |
| 16827 | { |
| 16828 | at::AutoDispatchSkipFunctionalize guard; |
| 16829 | tmp_output = at::_ops::adaptive_max_pool2d_backward::call(grad_output_, self_, indices_); |
| 16830 | } |
| 16831 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 16832 | at::functionalization::impl::commit_update(grad_input); |
| 16833 | at::functionalization::impl::sync(grad_input); |
| 16834 | return grad_input; |
| 16835 | } |
| 16836 | } |
| 16837 | |
| 16838 | at::Tensor & fractional_max_pool2d_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, at::IntArrayRef kernel_size, at::IntArrayRef output_size, const at::Tensor & indices, at::Tensor & grad_input) { |
| 16839 | if (false) { |
| 16840 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16841 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16842 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16843 | auto grad_output_meta = to_meta(grad_output); |
| 16844 | auto self_meta = to_meta(self); |
| 16845 | auto indices_meta = to_meta(indices); |
| 16846 | auto grad_input_meta = to_meta(grad_input); |
| 16847 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16848 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16849 | at::_ops::fractional_max_pool2d_backward_grad_input::call(grad_output_meta, self_meta, kernel_size, output_size, indices_meta, grad_input_meta); |
| 16850 | } |
| 16851 | |
| 16852 | at::Tensor grad_output_; |
| 16853 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 16854 | at::functionalization::impl::sync(grad_output); |
| 16855 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 16856 | } else { |
| 16857 | grad_output_ = grad_output; |
| 16858 | } |
| 16859 | |
| 16860 | at::Tensor self_; |
| 16861 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16862 | at::functionalization::impl::sync(self); |
| 16863 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16864 | } else { |
| 16865 | self_ = self; |
| 16866 | } |
| 16867 | |
| 16868 | at::Tensor indices_; |
| 16869 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 16870 | at::functionalization::impl::sync(indices); |
| 16871 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 16872 | } else { |
| 16873 | indices_ = indices; |
| 16874 | } |
| 16875 | |
| 16876 | at::Tensor grad_input_; |
| 16877 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 16878 | at::functionalization::impl::sync(grad_input); |
| 16879 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 16880 | } else { |
| 16881 | grad_input_ = grad_input; |
| 16882 | } |
| 16883 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 16884 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(indices))) { |
| 16885 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16886 | TORCH_INTERNAL_ASSERT(false, |
| 16887 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16888 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16889 | } else { |
| 16890 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16891 | at::AutoDispatchSkipFunctionalize guard; |
| 16892 | at::Tensor tmp_output = at::_ops::fractional_max_pool2d_backward_grad_input::call(grad_output_, self_, kernel_size, output_size, indices_, grad_input_); |
| 16893 | return grad_input;; |
| 16894 | } |
| 16895 | } else { |
| 16896 | at::Tensor tmp_output; |
| 16897 | { |
| 16898 | at::AutoDispatchSkipFunctionalize guard; |
| 16899 | tmp_output = at::_ops::fractional_max_pool2d_backward::call(grad_output_, self_, kernel_size, output_size, indices_); |
| 16900 | } |
| 16901 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 16902 | at::functionalization::impl::commit_update(grad_input); |
| 16903 | at::functionalization::impl::sync(grad_input); |
| 16904 | return grad_input; |
| 16905 | } |
| 16906 | } |
| 16907 | |
| 16908 | at::Tensor & fractional_max_pool3d_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, at::IntArrayRef kernel_size, at::IntArrayRef output_size, const at::Tensor & indices, at::Tensor & grad_input) { |
| 16909 | if (false) { |
| 16910 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16911 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16912 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16913 | auto grad_output_meta = to_meta(grad_output); |
| 16914 | auto self_meta = to_meta(self); |
| 16915 | auto indices_meta = to_meta(indices); |
| 16916 | auto grad_input_meta = to_meta(grad_input); |
| 16917 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16918 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16919 | at::_ops::fractional_max_pool3d_backward_grad_input::call(grad_output_meta, self_meta, kernel_size, output_size, indices_meta, grad_input_meta); |
| 16920 | } |
| 16921 | |
| 16922 | at::Tensor grad_output_; |
| 16923 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 16924 | at::functionalization::impl::sync(grad_output); |
| 16925 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 16926 | } else { |
| 16927 | grad_output_ = grad_output; |
| 16928 | } |
| 16929 | |
| 16930 | at::Tensor self_; |
| 16931 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16932 | at::functionalization::impl::sync(self); |
| 16933 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16934 | } else { |
| 16935 | self_ = self; |
| 16936 | } |
| 16937 | |
| 16938 | at::Tensor indices_; |
| 16939 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 16940 | at::functionalization::impl::sync(indices); |
| 16941 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 16942 | } else { |
| 16943 | indices_ = indices; |
| 16944 | } |
| 16945 | |
| 16946 | at::Tensor grad_input_; |
| 16947 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 16948 | at::functionalization::impl::sync(grad_input); |
| 16949 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 16950 | } else { |
| 16951 | grad_input_ = grad_input; |
| 16952 | } |
| 16953 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 16954 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(indices))) { |
| 16955 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 16956 | TORCH_INTERNAL_ASSERT(false, |
| 16957 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 16958 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 16959 | } else { |
| 16960 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 16961 | at::AutoDispatchSkipFunctionalize guard; |
| 16962 | at::Tensor tmp_output = at::_ops::fractional_max_pool3d_backward_grad_input::call(grad_output_, self_, kernel_size, output_size, indices_, grad_input_); |
| 16963 | return grad_input;; |
| 16964 | } |
| 16965 | } else { |
| 16966 | at::Tensor tmp_output; |
| 16967 | { |
| 16968 | at::AutoDispatchSkipFunctionalize guard; |
| 16969 | tmp_output = at::_ops::fractional_max_pool3d_backward::call(grad_output_, self_, kernel_size, output_size, indices_); |
| 16970 | } |
| 16971 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 16972 | at::functionalization::impl::commit_update(grad_input); |
| 16973 | at::functionalization::impl::sync(grad_input); |
| 16974 | return grad_input; |
| 16975 | } |
| 16976 | } |
| 16977 | |
| 16978 | ::std::tuple<at::Tensor &,at::Tensor &> max_pool3d_with_indices_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef kernel_size, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, bool ceil_mode, at::Tensor & out, at::Tensor & indices) { |
| 16979 | if (false) { |
| 16980 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 16981 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 16982 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 16983 | auto self_meta = to_meta(self); |
| 16984 | auto out_meta = to_meta(out); |
| 16985 | auto indices_meta = to_meta(indices); |
| 16986 | at::AutoDispatchSkipFunctionalize func_guard; |
| 16987 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 16988 | at::_ops::max_pool3d_with_indices_out::call(self_meta, kernel_size, stride, padding, dilation, ceil_mode, out_meta, indices_meta); |
| 16989 | } |
| 16990 | |
| 16991 | at::Tensor self_; |
| 16992 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 16993 | at::functionalization::impl::sync(self); |
| 16994 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 16995 | } else { |
| 16996 | self_ = self; |
| 16997 | } |
| 16998 | |
| 16999 | at::Tensor out_; |
| 17000 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 17001 | at::functionalization::impl::sync(out); |
| 17002 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 17003 | } else { |
| 17004 | out_ = out; |
| 17005 | } |
| 17006 | |
| 17007 | at::Tensor indices_; |
| 17008 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 17009 | at::functionalization::impl::sync(indices); |
| 17010 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 17011 | } else { |
| 17012 | indices_ = indices; |
| 17013 | } |
| 17014 | if (!(true && at::functionalization::impl::isFunctionalTensor(out) && at::functionalization::impl::isFunctionalTensor(indices))) { |
| 17015 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 17016 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17017 | TORCH_INTERNAL_ASSERT(false, |
| 17018 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17019 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17020 | } else { |
| 17021 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17022 | at::AutoDispatchSkipFunctionalize guard; |
| 17023 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::max_pool3d_with_indices_out::call(self_, kernel_size, stride, padding, dilation, ceil_mode, out_, indices_); |
| 17024 | return ::std::tuple<at::Tensor &,at::Tensor &>(out, indices);; |
| 17025 | } |
| 17026 | } else { |
| 17027 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 17028 | { |
| 17029 | at::AutoDispatchSkipFunctionalize guard; |
| 17030 | tmp_output = at::_ops::max_pool3d_with_indices::call(self_, kernel_size, stride, padding, dilation, ceil_mode); |
| 17031 | } |
| 17032 | at::functionalization::impl::replace_(out, std::get<0>(tmp_output)); |
| 17033 | at::functionalization::impl::commit_update(out); |
| 17034 | at::functionalization::impl::sync(out); |
| 17035 | at::functionalization::impl::replace_(indices, std::get<1>(tmp_output)); |
| 17036 | at::functionalization::impl::commit_update(indices); |
| 17037 | at::functionalization::impl::sync(indices); |
| 17038 | return ::std::tuple<at::Tensor &,at::Tensor &>(out, indices); |
| 17039 | } |
| 17040 | } |
| 17041 | |
| 17042 | at::Tensor & max_pool3d_with_indices_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, at::IntArrayRef kernel_size, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, bool ceil_mode, const at::Tensor & indices, at::Tensor & grad_input) { |
| 17043 | if (false) { |
| 17044 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17045 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17046 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17047 | auto grad_output_meta = to_meta(grad_output); |
| 17048 | auto self_meta = to_meta(self); |
| 17049 | auto indices_meta = to_meta(indices); |
| 17050 | auto grad_input_meta = to_meta(grad_input); |
| 17051 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17052 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17053 | at::_ops::max_pool3d_with_indices_backward_grad_input::call(grad_output_meta, self_meta, kernel_size, stride, padding, dilation, ceil_mode, indices_meta, grad_input_meta); |
| 17054 | } |
| 17055 | |
| 17056 | at::Tensor grad_output_; |
| 17057 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 17058 | at::functionalization::impl::sync(grad_output); |
| 17059 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 17060 | } else { |
| 17061 | grad_output_ = grad_output; |
| 17062 | } |
| 17063 | |
| 17064 | at::Tensor self_; |
| 17065 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17066 | at::functionalization::impl::sync(self); |
| 17067 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17068 | } else { |
| 17069 | self_ = self; |
| 17070 | } |
| 17071 | |
| 17072 | at::Tensor indices_; |
| 17073 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 17074 | at::functionalization::impl::sync(indices); |
| 17075 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 17076 | } else { |
| 17077 | indices_ = indices; |
| 17078 | } |
| 17079 | |
| 17080 | at::Tensor grad_input_; |
| 17081 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 17082 | at::functionalization::impl::sync(grad_input); |
| 17083 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 17084 | } else { |
| 17085 | grad_input_ = grad_input; |
| 17086 | } |
| 17087 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 17088 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(indices))) { |
| 17089 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17090 | TORCH_INTERNAL_ASSERT(false, |
| 17091 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17092 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17093 | } else { |
| 17094 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17095 | at::AutoDispatchSkipFunctionalize guard; |
| 17096 | at::Tensor tmp_output = at::_ops::max_pool3d_with_indices_backward_grad_input::call(grad_output_, self_, kernel_size, stride, padding, dilation, ceil_mode, indices_, grad_input_); |
| 17097 | return grad_input;; |
| 17098 | } |
| 17099 | } else { |
| 17100 | at::Tensor tmp_output; |
| 17101 | { |
| 17102 | at::AutoDispatchSkipFunctionalize guard; |
| 17103 | tmp_output = at::_ops::max_pool3d_with_indices_backward::call(grad_output_, self_, kernel_size, stride, padding, dilation, ceil_mode, indices_); |
| 17104 | } |
| 17105 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 17106 | at::functionalization::impl::commit_update(grad_input); |
| 17107 | at::functionalization::impl::sync(grad_input); |
| 17108 | return grad_input; |
| 17109 | } |
| 17110 | } |
| 17111 | |
| 17112 | at::Tensor & max_unpool2d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & indices, at::IntArrayRef output_size, at::Tensor & out) { |
| 17113 | if (false) { |
| 17114 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17115 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17116 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17117 | auto self_meta = to_meta(self); |
| 17118 | auto indices_meta = to_meta(indices); |
| 17119 | auto out_meta = to_meta(out); |
| 17120 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17121 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17122 | at::_ops::max_unpool2d_out::call(self_meta, indices_meta, output_size, out_meta); |
| 17123 | } |
| 17124 | |
| 17125 | at::Tensor self_; |
| 17126 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17127 | at::functionalization::impl::sync(self); |
| 17128 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17129 | } else { |
| 17130 | self_ = self; |
| 17131 | } |
| 17132 | |
| 17133 | at::Tensor indices_; |
| 17134 | if (at::functionalization::impl::isFunctionalTensor(indices)) { |
| 17135 | at::functionalization::impl::sync(indices); |
| 17136 | indices_ = at::functionalization::impl::from_functional_tensor(indices); |
| 17137 | } else { |
| 17138 | indices_ = indices; |
| 17139 | } |
| 17140 | |
| 17141 | at::Tensor out_; |
| 17142 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 17143 | at::functionalization::impl::sync(out); |
| 17144 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 17145 | } else { |
| 17146 | out_ = out; |
| 17147 | } |
| 17148 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 17149 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(indices))) { |
| 17150 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17151 | TORCH_INTERNAL_ASSERT(false, |
| 17152 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17153 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17154 | } else { |
| 17155 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17156 | at::AutoDispatchSkipFunctionalize guard; |
| 17157 | at::Tensor tmp_output = at::_ops::max_unpool2d_out::call(self_, indices_, output_size, out_); |
| 17158 | return out;; |
| 17159 | } |
| 17160 | } else { |
| 17161 | at::Tensor tmp_output; |
| 17162 | { |
| 17163 | at::AutoDispatchSkipFunctionalize guard; |
| 17164 | tmp_output = at::_ops::max_unpool2d::call(self_, indices_, output_size); |
| 17165 | } |
| 17166 | at::functionalization::impl::replace_(out, tmp_output); |
| 17167 | at::functionalization::impl::commit_update(out); |
| 17168 | at::functionalization::impl::sync(out); |
| 17169 | return out; |
| 17170 | } |
| 17171 | } |
| 17172 | |
| 17173 | at::Tensor & reflection_pad2d_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, c10::SymIntArrayRef padding, at::Tensor & grad_input) { |
| 17174 | if (false) { |
| 17175 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17176 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17177 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17178 | auto grad_output_meta = to_meta(grad_output); |
| 17179 | auto self_meta = to_meta(self); |
| 17180 | auto grad_input_meta = to_meta(grad_input); |
| 17181 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17182 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17183 | at::_ops::reflection_pad2d_backward_grad_input::call(grad_output_meta, self_meta, padding, grad_input_meta); |
| 17184 | } |
| 17185 | |
| 17186 | at::Tensor grad_output_; |
| 17187 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 17188 | at::functionalization::impl::sync(grad_output); |
| 17189 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 17190 | } else { |
| 17191 | grad_output_ = grad_output; |
| 17192 | } |
| 17193 | |
| 17194 | at::Tensor self_; |
| 17195 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17196 | at::functionalization::impl::sync(self); |
| 17197 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17198 | } else { |
| 17199 | self_ = self; |
| 17200 | } |
| 17201 | |
| 17202 | at::Tensor grad_input_; |
| 17203 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 17204 | at::functionalization::impl::sync(grad_input); |
| 17205 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 17206 | } else { |
| 17207 | grad_input_ = grad_input; |
| 17208 | } |
| 17209 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 17210 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self))) { |
| 17211 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17212 | TORCH_INTERNAL_ASSERT(false, |
| 17213 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17214 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17215 | } else { |
| 17216 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17217 | at::AutoDispatchSkipFunctionalize guard; |
| 17218 | at::Tensor tmp_output = at::_ops::reflection_pad2d_backward_grad_input::call(grad_output_, self_, padding, grad_input_); |
| 17219 | return grad_input;; |
| 17220 | } |
| 17221 | } else { |
| 17222 | at::Tensor tmp_output; |
| 17223 | { |
| 17224 | at::AutoDispatchSkipFunctionalize guard; |
| 17225 | tmp_output = at::_ops::reflection_pad2d_backward::call(grad_output_, self_, padding); |
| 17226 | } |
| 17227 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 17228 | at::functionalization::impl::commit_update(grad_input); |
| 17229 | at::functionalization::impl::sync(grad_input); |
| 17230 | return grad_input; |
| 17231 | } |
| 17232 | } |
| 17233 | |
| 17234 | at::Tensor & upsample_bilinear2d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::SymIntArrayRef output_size, bool align_corners, c10::optional<double> scales_h, c10::optional<double> scales_w, at::Tensor & out) { |
| 17235 | if (false) { |
| 17236 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17237 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17238 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17239 | auto self_meta = to_meta(self); |
| 17240 | auto out_meta = to_meta(out); |
| 17241 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17242 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17243 | at::_ops::upsample_bilinear2d_out::call(self_meta, output_size, align_corners, scales_h, scales_w, out_meta); |
| 17244 | } |
| 17245 | |
| 17246 | at::Tensor self_; |
| 17247 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17248 | at::functionalization::impl::sync(self); |
| 17249 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17250 | } else { |
| 17251 | self_ = self; |
| 17252 | } |
| 17253 | |
| 17254 | at::Tensor out_; |
| 17255 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 17256 | at::functionalization::impl::sync(out); |
| 17257 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 17258 | } else { |
| 17259 | out_ = out; |
| 17260 | } |
| 17261 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 17262 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 17263 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17264 | TORCH_INTERNAL_ASSERT(false, |
| 17265 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17266 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17267 | } else { |
| 17268 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17269 | at::AutoDispatchSkipFunctionalize guard; |
| 17270 | at::Tensor tmp_output = at::_ops::upsample_bilinear2d_out::call(self_, output_size, align_corners, scales_h, scales_w, out_); |
| 17271 | return out;; |
| 17272 | } |
| 17273 | } else { |
| 17274 | at::Tensor tmp_output; |
| 17275 | { |
| 17276 | at::AutoDispatchSkipFunctionalize guard; |
| 17277 | tmp_output = at::_ops::upsample_bilinear2d::call(self_, output_size, align_corners, scales_h, scales_w); |
| 17278 | } |
| 17279 | at::functionalization::impl::replace_(out, tmp_output); |
| 17280 | at::functionalization::impl::commit_update(out); |
| 17281 | at::functionalization::impl::sync(out); |
| 17282 | return out; |
| 17283 | } |
| 17284 | } |
| 17285 | |
| 17286 | at::Tensor & upsample_bilinear2d_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, c10::SymIntArrayRef output_size, c10::SymIntArrayRef input_size, bool align_corners, c10::optional<double> scales_h, c10::optional<double> scales_w, at::Tensor & grad_input) { |
| 17287 | if (false) { |
| 17288 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17289 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17290 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17291 | auto grad_output_meta = to_meta(grad_output); |
| 17292 | auto grad_input_meta = to_meta(grad_input); |
| 17293 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17294 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17295 | at::_ops::upsample_bilinear2d_backward_grad_input::call(grad_output_meta, output_size, input_size, align_corners, scales_h, scales_w, grad_input_meta); |
| 17296 | } |
| 17297 | |
| 17298 | at::Tensor grad_output_; |
| 17299 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 17300 | at::functionalization::impl::sync(grad_output); |
| 17301 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 17302 | } else { |
| 17303 | grad_output_ = grad_output; |
| 17304 | } |
| 17305 | |
| 17306 | at::Tensor grad_input_; |
| 17307 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 17308 | at::functionalization::impl::sync(grad_input); |
| 17309 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 17310 | } else { |
| 17311 | grad_input_ = grad_input; |
| 17312 | } |
| 17313 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 17314 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output))) { |
| 17315 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17316 | TORCH_INTERNAL_ASSERT(false, |
| 17317 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17318 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17319 | } else { |
| 17320 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17321 | at::AutoDispatchSkipFunctionalize guard; |
| 17322 | at::Tensor tmp_output = at::_ops::upsample_bilinear2d_backward_grad_input::call(grad_output_, output_size, input_size, align_corners, scales_h, scales_w, grad_input_); |
| 17323 | return grad_input;; |
| 17324 | } |
| 17325 | } else { |
| 17326 | at::Tensor tmp_output; |
| 17327 | { |
| 17328 | at::AutoDispatchSkipFunctionalize guard; |
| 17329 | tmp_output = at::_ops::upsample_bilinear2d_backward::call(grad_output_, output_size, input_size, align_corners, scales_h, scales_w); |
| 17330 | } |
| 17331 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 17332 | at::functionalization::impl::commit_update(grad_input); |
| 17333 | at::functionalization::impl::sync(grad_input); |
| 17334 | return grad_input; |
| 17335 | } |
| 17336 | } |
| 17337 | |
| 17338 | at::Tensor & _upsample_bilinear2d_aa_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::SymIntArrayRef output_size, bool align_corners, c10::optional<double> scales_h, c10::optional<double> scales_w, at::Tensor & out) { |
| 17339 | if (false) { |
| 17340 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17341 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17342 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17343 | auto self_meta = to_meta(self); |
| 17344 | auto out_meta = to_meta(out); |
| 17345 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17346 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17347 | at::_ops::_upsample_bilinear2d_aa_out::call(self_meta, output_size, align_corners, scales_h, scales_w, out_meta); |
| 17348 | } |
| 17349 | |
| 17350 | at::Tensor self_; |
| 17351 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17352 | at::functionalization::impl::sync(self); |
| 17353 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17354 | } else { |
| 17355 | self_ = self; |
| 17356 | } |
| 17357 | |
| 17358 | at::Tensor out_; |
| 17359 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 17360 | at::functionalization::impl::sync(out); |
| 17361 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 17362 | } else { |
| 17363 | out_ = out; |
| 17364 | } |
| 17365 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 17366 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 17367 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17368 | TORCH_INTERNAL_ASSERT(false, |
| 17369 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17370 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17371 | } else { |
| 17372 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17373 | at::AutoDispatchSkipFunctionalize guard; |
| 17374 | at::Tensor tmp_output = at::_ops::_upsample_bilinear2d_aa_out::call(self_, output_size, align_corners, scales_h, scales_w, out_); |
| 17375 | return out;; |
| 17376 | } |
| 17377 | } else { |
| 17378 | at::Tensor tmp_output; |
| 17379 | { |
| 17380 | at::AutoDispatchSkipFunctionalize guard; |
| 17381 | tmp_output = at::_ops::_upsample_bilinear2d_aa::call(self_, output_size, align_corners, scales_h, scales_w); |
| 17382 | } |
| 17383 | at::functionalization::impl::replace_(out, tmp_output); |
| 17384 | at::functionalization::impl::commit_update(out); |
| 17385 | at::functionalization::impl::sync(out); |
| 17386 | return out; |
| 17387 | } |
| 17388 | } |
| 17389 | |
| 17390 | at::Tensor & upsample_trilinear3d_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, c10::SymIntArrayRef output_size, c10::SymIntArrayRef input_size, bool align_corners, c10::optional<double> scales_d, c10::optional<double> scales_h, c10::optional<double> scales_w, at::Tensor & grad_input) { |
| 17391 | if (false) { |
| 17392 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17393 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17394 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17395 | auto grad_output_meta = to_meta(grad_output); |
| 17396 | auto grad_input_meta = to_meta(grad_input); |
| 17397 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17398 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17399 | at::_ops::upsample_trilinear3d_backward_grad_input::call(grad_output_meta, output_size, input_size, align_corners, scales_d, scales_h, scales_w, grad_input_meta); |
| 17400 | } |
| 17401 | |
| 17402 | at::Tensor grad_output_; |
| 17403 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 17404 | at::functionalization::impl::sync(grad_output); |
| 17405 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 17406 | } else { |
| 17407 | grad_output_ = grad_output; |
| 17408 | } |
| 17409 | |
| 17410 | at::Tensor grad_input_; |
| 17411 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 17412 | at::functionalization::impl::sync(grad_input); |
| 17413 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 17414 | } else { |
| 17415 | grad_input_ = grad_input; |
| 17416 | } |
| 17417 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 17418 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output))) { |
| 17419 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17420 | TORCH_INTERNAL_ASSERT(false, |
| 17421 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17422 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17423 | } else { |
| 17424 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17425 | at::AutoDispatchSkipFunctionalize guard; |
| 17426 | at::Tensor tmp_output = at::_ops::upsample_trilinear3d_backward_grad_input::call(grad_output_, output_size, input_size, align_corners, scales_d, scales_h, scales_w, grad_input_); |
| 17427 | return grad_input;; |
| 17428 | } |
| 17429 | } else { |
| 17430 | at::Tensor tmp_output; |
| 17431 | { |
| 17432 | at::AutoDispatchSkipFunctionalize guard; |
| 17433 | tmp_output = at::_ops::upsample_trilinear3d_backward::call(grad_output_, output_size, input_size, align_corners, scales_d, scales_h, scales_w); |
| 17434 | } |
| 17435 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 17436 | at::functionalization::impl::commit_update(grad_input); |
| 17437 | at::functionalization::impl::sync(grad_input); |
| 17438 | return grad_input; |
| 17439 | } |
| 17440 | } |
| 17441 | |
| 17442 | at::Tensor & _upsample_nearest_exact3d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::SymIntArrayRef output_size, c10::optional<double> scales_d, c10::optional<double> scales_h, c10::optional<double> scales_w, at::Tensor & out) { |
| 17443 | if (false) { |
| 17444 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17445 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17446 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17447 | auto self_meta = to_meta(self); |
| 17448 | auto out_meta = to_meta(out); |
| 17449 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17450 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17451 | at::_ops::_upsample_nearest_exact3d_out::call(self_meta, output_size, scales_d, scales_h, scales_w, out_meta); |
| 17452 | } |
| 17453 | |
| 17454 | at::Tensor self_; |
| 17455 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17456 | at::functionalization::impl::sync(self); |
| 17457 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17458 | } else { |
| 17459 | self_ = self; |
| 17460 | } |
| 17461 | |
| 17462 | at::Tensor out_; |
| 17463 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 17464 | at::functionalization::impl::sync(out); |
| 17465 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 17466 | } else { |
| 17467 | out_ = out; |
| 17468 | } |
| 17469 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 17470 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 17471 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17472 | TORCH_INTERNAL_ASSERT(false, |
| 17473 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17474 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17475 | } else { |
| 17476 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17477 | at::AutoDispatchSkipFunctionalize guard; |
| 17478 | at::Tensor tmp_output = at::_ops::_upsample_nearest_exact3d_out::call(self_, output_size, scales_d, scales_h, scales_w, out_); |
| 17479 | return out;; |
| 17480 | } |
| 17481 | } else { |
| 17482 | at::Tensor tmp_output; |
| 17483 | { |
| 17484 | at::AutoDispatchSkipFunctionalize guard; |
| 17485 | tmp_output = at::_ops::_upsample_nearest_exact3d::call(self_, output_size, scales_d, scales_h, scales_w); |
| 17486 | } |
| 17487 | at::functionalization::impl::replace_(out, tmp_output); |
| 17488 | at::functionalization::impl::commit_update(out); |
| 17489 | at::functionalization::impl::sync(out); |
| 17490 | return out; |
| 17491 | } |
| 17492 | } |
| 17493 | |
| 17494 | at::Tensor & upsample_nearest3d_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, c10::SymIntArrayRef output_size, c10::SymIntArrayRef input_size, c10::optional<double> scales_d, c10::optional<double> scales_h, c10::optional<double> scales_w, at::Tensor & grad_input) { |
| 17495 | if (false) { |
| 17496 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17497 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17498 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17499 | auto grad_output_meta = to_meta(grad_output); |
| 17500 | auto grad_input_meta = to_meta(grad_input); |
| 17501 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17502 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17503 | at::_ops::upsample_nearest3d_backward_grad_input::call(grad_output_meta, output_size, input_size, scales_d, scales_h, scales_w, grad_input_meta); |
| 17504 | } |
| 17505 | |
| 17506 | at::Tensor grad_output_; |
| 17507 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 17508 | at::functionalization::impl::sync(grad_output); |
| 17509 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 17510 | } else { |
| 17511 | grad_output_ = grad_output; |
| 17512 | } |
| 17513 | |
| 17514 | at::Tensor grad_input_; |
| 17515 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 17516 | at::functionalization::impl::sync(grad_input); |
| 17517 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 17518 | } else { |
| 17519 | grad_input_ = grad_input; |
| 17520 | } |
| 17521 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 17522 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output))) { |
| 17523 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17524 | TORCH_INTERNAL_ASSERT(false, |
| 17525 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17526 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17527 | } else { |
| 17528 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17529 | at::AutoDispatchSkipFunctionalize guard; |
| 17530 | at::Tensor tmp_output = at::_ops::upsample_nearest3d_backward_grad_input::call(grad_output_, output_size, input_size, scales_d, scales_h, scales_w, grad_input_); |
| 17531 | return grad_input;; |
| 17532 | } |
| 17533 | } else { |
| 17534 | at::Tensor tmp_output; |
| 17535 | { |
| 17536 | at::AutoDispatchSkipFunctionalize guard; |
| 17537 | tmp_output = at::_ops::upsample_nearest3d_backward::call(grad_output_, output_size, input_size, scales_d, scales_h, scales_w); |
| 17538 | } |
| 17539 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 17540 | at::functionalization::impl::commit_update(grad_input); |
| 17541 | at::functionalization::impl::sync(grad_input); |
| 17542 | return grad_input; |
| 17543 | } |
| 17544 | } |
| 17545 | |
| 17546 | at::Tensor & logit_backward_out_grad_input(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, c10::optional<double> eps, at::Tensor & grad_input) { |
| 17547 | if (false) { |
| 17548 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17549 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17550 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17551 | auto grad_output_meta = to_meta(grad_output); |
| 17552 | auto self_meta = to_meta(self); |
| 17553 | auto grad_input_meta = to_meta(grad_input); |
| 17554 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17555 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17556 | at::_ops::logit_backward_grad_input::call(grad_output_meta, self_meta, eps, grad_input_meta); |
| 17557 | } |
| 17558 | |
| 17559 | at::Tensor grad_output_; |
| 17560 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 17561 | at::functionalization::impl::sync(grad_output); |
| 17562 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 17563 | } else { |
| 17564 | grad_output_ = grad_output; |
| 17565 | } |
| 17566 | |
| 17567 | at::Tensor self_; |
| 17568 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17569 | at::functionalization::impl::sync(self); |
| 17570 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17571 | } else { |
| 17572 | self_ = self; |
| 17573 | } |
| 17574 | |
| 17575 | at::Tensor grad_input_; |
| 17576 | if (at::functionalization::impl::isFunctionalTensor(grad_input)) { |
| 17577 | at::functionalization::impl::sync(grad_input); |
| 17578 | grad_input_ = at::functionalization::impl::from_functional_tensor(grad_input); |
| 17579 | } else { |
| 17580 | grad_input_ = grad_input; |
| 17581 | } |
| 17582 | if (!(true && at::functionalization::impl::isFunctionalTensor(grad_input))) { |
| 17583 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self))) { |
| 17584 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17585 | TORCH_INTERNAL_ASSERT(false, |
| 17586 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17587 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17588 | } else { |
| 17589 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17590 | at::AutoDispatchSkipFunctionalize guard; |
| 17591 | at::Tensor tmp_output = at::_ops::logit_backward_grad_input::call(grad_output_, self_, eps, grad_input_); |
| 17592 | return grad_input;; |
| 17593 | } |
| 17594 | } else { |
| 17595 | at::Tensor tmp_output; |
| 17596 | { |
| 17597 | at::AutoDispatchSkipFunctionalize guard; |
| 17598 | tmp_output = at::_ops::logit_backward::call(grad_output_, self_, eps); |
| 17599 | } |
| 17600 | at::functionalization::impl::replace_(grad_input, tmp_output); |
| 17601 | at::functionalization::impl::commit_update(grad_input); |
| 17602 | at::functionalization::impl::sync(grad_input); |
| 17603 | return grad_input; |
| 17604 | } |
| 17605 | } |
| 17606 | |
| 17607 | at::Tensor & thnn_conv2d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::Tensor & out) { |
| 17608 | if (false) { |
| 17609 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17610 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17611 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17612 | auto self_meta = to_meta(self); |
| 17613 | auto weight_meta = to_meta(weight); |
| 17614 | auto bias_meta = to_meta(bias); |
| 17615 | auto out_meta = to_meta(out); |
| 17616 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17617 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17618 | at::_ops::thnn_conv2d_out::call(self_meta, weight_meta, kernel_size, bias_meta, stride, padding, out_meta); |
| 17619 | } |
| 17620 | |
| 17621 | at::Tensor self_; |
| 17622 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17623 | at::functionalization::impl::sync(self); |
| 17624 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17625 | } else { |
| 17626 | self_ = self; |
| 17627 | } |
| 17628 | |
| 17629 | at::Tensor weight_; |
| 17630 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 17631 | at::functionalization::impl::sync(weight); |
| 17632 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 17633 | } else { |
| 17634 | weight_ = weight; |
| 17635 | } |
| 17636 | |
| 17637 | c10::optional<at::Tensor> bias_; |
| 17638 | if (at::functionalization::impl::isFunctionalTensor(bias)) { |
| 17639 | at::functionalization::impl::sync(bias); |
| 17640 | bias_ = at::functionalization::impl::from_functional_tensor(bias); |
| 17641 | } else { |
| 17642 | bias_ = bias; |
| 17643 | } |
| 17644 | |
| 17645 | at::Tensor out_; |
| 17646 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 17647 | at::functionalization::impl::sync(out); |
| 17648 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 17649 | } else { |
| 17650 | out_ = out; |
| 17651 | } |
| 17652 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 17653 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(bias))) { |
| 17654 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17655 | TORCH_INTERNAL_ASSERT(false, |
| 17656 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17657 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17658 | } else { |
| 17659 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17660 | at::AutoDispatchSkipFunctionalize guard; |
| 17661 | at::Tensor tmp_output = at::_ops::thnn_conv2d_out::call(self_, weight_, kernel_size, bias_, stride, padding, out_); |
| 17662 | return out;; |
| 17663 | } |
| 17664 | } else { |
| 17665 | at::Tensor tmp_output; |
| 17666 | { |
| 17667 | at::AutoDispatchSkipFunctionalize guard; |
| 17668 | tmp_output = at::_ops::thnn_conv2d::call(self_, weight_, kernel_size, bias_, stride, padding); |
| 17669 | } |
| 17670 | at::functionalization::impl::replace_(out, tmp_output); |
| 17671 | at::functionalization::impl::commit_update(out); |
| 17672 | at::functionalization::impl::sync(out); |
| 17673 | return out; |
| 17674 | } |
| 17675 | } |
| 17676 | |
| 17677 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> _slow_conv2d_backward_out_output_mask_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & grad_output, const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, at::IntArrayRef stride, at::IntArrayRef padding, ::std::array<bool,3> output_mask, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 17678 | if (false) { |
| 17679 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17680 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17681 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17682 | auto grad_output_meta = to_meta(grad_output); |
| 17683 | auto self_meta = to_meta(self); |
| 17684 | auto weight_meta = to_meta(weight); |
| 17685 | auto out0_meta = to_meta(out0); |
| 17686 | auto out1_meta = to_meta(out1); |
| 17687 | auto out2_meta = to_meta(out2); |
| 17688 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17689 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17690 | at::_ops::_slow_conv2d_backward_output_mask_out::call(grad_output_meta, self_meta, weight_meta, kernel_size, stride, padding, output_mask, out0_meta, out1_meta, out2_meta); |
| 17691 | } |
| 17692 | |
| 17693 | at::Tensor grad_output_; |
| 17694 | if (at::functionalization::impl::isFunctionalTensor(grad_output)) { |
| 17695 | at::functionalization::impl::sync(grad_output); |
| 17696 | grad_output_ = at::functionalization::impl::from_functional_tensor(grad_output); |
| 17697 | } else { |
| 17698 | grad_output_ = grad_output; |
| 17699 | } |
| 17700 | |
| 17701 | at::Tensor self_; |
| 17702 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17703 | at::functionalization::impl::sync(self); |
| 17704 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17705 | } else { |
| 17706 | self_ = self; |
| 17707 | } |
| 17708 | |
| 17709 | at::Tensor weight_; |
| 17710 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 17711 | at::functionalization::impl::sync(weight); |
| 17712 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 17713 | } else { |
| 17714 | weight_ = weight; |
| 17715 | } |
| 17716 | |
| 17717 | at::Tensor out0_; |
| 17718 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 17719 | at::functionalization::impl::sync(out0); |
| 17720 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 17721 | } else { |
| 17722 | out0_ = out0; |
| 17723 | } |
| 17724 | |
| 17725 | at::Tensor out1_; |
| 17726 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 17727 | at::functionalization::impl::sync(out1); |
| 17728 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 17729 | } else { |
| 17730 | out1_ = out1; |
| 17731 | } |
| 17732 | |
| 17733 | at::Tensor out2_; |
| 17734 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 17735 | at::functionalization::impl::sync(out2); |
| 17736 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 17737 | } else { |
| 17738 | out2_ = out2; |
| 17739 | } |
| 17740 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 17741 | if ((false || at::functionalization::impl::isFunctionalTensor(grad_output) || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(weight))) { |
| 17742 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17743 | TORCH_INTERNAL_ASSERT(false, |
| 17744 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17745 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17746 | } else { |
| 17747 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17748 | at::AutoDispatchSkipFunctionalize guard; |
| 17749 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_slow_conv2d_backward_output_mask_out::call(grad_output_, self_, weight_, kernel_size, stride, padding, output_mask, out0_, out1_, out2_); |
| 17750 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 17751 | } |
| 17752 | } else { |
| 17753 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 17754 | { |
| 17755 | at::AutoDispatchSkipFunctionalize guard; |
| 17756 | tmp_output = at::_ops::_slow_conv2d_backward_output_mask::call(grad_output_, self_, weight_, kernel_size, stride, padding, output_mask); |
| 17757 | } |
| 17758 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 17759 | at::functionalization::impl::commit_update(out0); |
| 17760 | at::functionalization::impl::sync(out0); |
| 17761 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 17762 | at::functionalization::impl::commit_update(out1); |
| 17763 | at::functionalization::impl::sync(out1); |
| 17764 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 17765 | at::functionalization::impl::commit_update(out2); |
| 17766 | at::functionalization::impl::sync(out2); |
| 17767 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 17768 | } |
| 17769 | } |
| 17770 | |
| 17771 | at::Tensor & slow_conv3d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, c10::SymIntArrayRef padding, at::Tensor & out) { |
| 17772 | if (false) { |
| 17773 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17774 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17775 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17776 | auto self_meta = to_meta(self); |
| 17777 | auto weight_meta = to_meta(weight); |
| 17778 | auto bias_meta = to_meta(bias); |
| 17779 | auto out_meta = to_meta(out); |
| 17780 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17781 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17782 | at::_ops::slow_conv3d_out::call(self_meta, weight_meta, kernel_size, bias_meta, stride, padding, out_meta); |
| 17783 | } |
| 17784 | |
| 17785 | at::Tensor self_; |
| 17786 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17787 | at::functionalization::impl::sync(self); |
| 17788 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17789 | } else { |
| 17790 | self_ = self; |
| 17791 | } |
| 17792 | |
| 17793 | at::Tensor weight_; |
| 17794 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 17795 | at::functionalization::impl::sync(weight); |
| 17796 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 17797 | } else { |
| 17798 | weight_ = weight; |
| 17799 | } |
| 17800 | |
| 17801 | c10::optional<at::Tensor> bias_; |
| 17802 | if (at::functionalization::impl::isFunctionalTensor(bias)) { |
| 17803 | at::functionalization::impl::sync(bias); |
| 17804 | bias_ = at::functionalization::impl::from_functional_tensor(bias); |
| 17805 | } else { |
| 17806 | bias_ = bias; |
| 17807 | } |
| 17808 | |
| 17809 | at::Tensor out_; |
| 17810 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 17811 | at::functionalization::impl::sync(out); |
| 17812 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 17813 | } else { |
| 17814 | out_ = out; |
| 17815 | } |
| 17816 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 17817 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(bias))) { |
| 17818 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17819 | TORCH_INTERNAL_ASSERT(false, |
| 17820 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17821 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17822 | } else { |
| 17823 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17824 | at::AutoDispatchSkipFunctionalize guard; |
| 17825 | at::Tensor tmp_output = at::_ops::slow_conv3d_out::call(self_, weight_, kernel_size, bias_, stride, padding, out_); |
| 17826 | return out;; |
| 17827 | } |
| 17828 | } else { |
| 17829 | at::Tensor tmp_output; |
| 17830 | { |
| 17831 | at::AutoDispatchSkipFunctionalize guard; |
| 17832 | tmp_output = at::_ops::slow_conv3d::call(self_, weight_, kernel_size, bias_, stride, padding); |
| 17833 | } |
| 17834 | at::functionalization::impl::replace_(out, tmp_output); |
| 17835 | at::functionalization::impl::commit_update(out); |
| 17836 | at::functionalization::impl::sync(out); |
| 17837 | return out; |
| 17838 | } |
| 17839 | } |
| 17840 | |
| 17841 | at::Tensor & slow_conv3d_forward_out_output(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, c10::SymIntArrayRef padding, at::Tensor & output) { |
| 17842 | if (false) { |
| 17843 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17844 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17845 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17846 | auto self_meta = to_meta(self); |
| 17847 | auto weight_meta = to_meta(weight); |
| 17848 | auto bias_meta = to_meta(bias); |
| 17849 | auto output_meta = to_meta(output); |
| 17850 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17851 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17852 | at::_ops::slow_conv3d_forward_output::call(self_meta, weight_meta, kernel_size, bias_meta, stride, padding, output_meta); |
| 17853 | } |
| 17854 | |
| 17855 | at::Tensor self_; |
| 17856 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17857 | at::functionalization::impl::sync(self); |
| 17858 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17859 | } else { |
| 17860 | self_ = self; |
| 17861 | } |
| 17862 | |
| 17863 | at::Tensor weight_; |
| 17864 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 17865 | at::functionalization::impl::sync(weight); |
| 17866 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 17867 | } else { |
| 17868 | weight_ = weight; |
| 17869 | } |
| 17870 | |
| 17871 | c10::optional<at::Tensor> bias_; |
| 17872 | if (at::functionalization::impl::isFunctionalTensor(bias)) { |
| 17873 | at::functionalization::impl::sync(bias); |
| 17874 | bias_ = at::functionalization::impl::from_functional_tensor(bias); |
| 17875 | } else { |
| 17876 | bias_ = bias; |
| 17877 | } |
| 17878 | |
| 17879 | at::Tensor output_; |
| 17880 | if (at::functionalization::impl::isFunctionalTensor(output)) { |
| 17881 | at::functionalization::impl::sync(output); |
| 17882 | output_ = at::functionalization::impl::from_functional_tensor(output); |
| 17883 | } else { |
| 17884 | output_ = output; |
| 17885 | } |
| 17886 | if (!(true && at::functionalization::impl::isFunctionalTensor(output))) { |
| 17887 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(bias))) { |
| 17888 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17889 | TORCH_INTERNAL_ASSERT(false, |
| 17890 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17891 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17892 | } else { |
| 17893 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17894 | at::AutoDispatchSkipFunctionalize guard; |
| 17895 | at::Tensor tmp_output = at::_ops::slow_conv3d_forward_output::call(self_, weight_, kernel_size, bias_, stride, padding, output_); |
| 17896 | return output;; |
| 17897 | } |
| 17898 | } else { |
| 17899 | at::Tensor tmp_output; |
| 17900 | { |
| 17901 | at::AutoDispatchSkipFunctionalize guard; |
| 17902 | tmp_output = at::_ops::slow_conv3d_forward::call(self_, weight_, kernel_size, bias_, stride, padding); |
| 17903 | } |
| 17904 | at::functionalization::impl::replace_(output, tmp_output); |
| 17905 | at::functionalization::impl::commit_update(output); |
| 17906 | at::functionalization::impl::sync(output); |
| 17907 | return output; |
| 17908 | } |
| 17909 | } |
| 17910 | |
| 17911 | at::Tensor & slow_conv_dilated3d_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, c10::SymIntArrayRef padding, at::IntArrayRef dilation, at::Tensor & out) { |
| 17912 | if (false) { |
| 17913 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17914 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17915 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17916 | auto self_meta = to_meta(self); |
| 17917 | auto weight_meta = to_meta(weight); |
| 17918 | auto bias_meta = to_meta(bias); |
| 17919 | auto out_meta = to_meta(out); |
| 17920 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17921 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17922 | at::_ops::slow_conv_dilated3d_out::call(self_meta, weight_meta, kernel_size, bias_meta, stride, padding, dilation, out_meta); |
| 17923 | } |
| 17924 | |
| 17925 | at::Tensor self_; |
| 17926 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17927 | at::functionalization::impl::sync(self); |
| 17928 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17929 | } else { |
| 17930 | self_ = self; |
| 17931 | } |
| 17932 | |
| 17933 | at::Tensor weight_; |
| 17934 | if (at::functionalization::impl::isFunctionalTensor(weight)) { |
| 17935 | at::functionalization::impl::sync(weight); |
| 17936 | weight_ = at::functionalization::impl::from_functional_tensor(weight); |
| 17937 | } else { |
| 17938 | weight_ = weight; |
| 17939 | } |
| 17940 | |
| 17941 | c10::optional<at::Tensor> bias_; |
| 17942 | if (at::functionalization::impl::isFunctionalTensor(bias)) { |
| 17943 | at::functionalization::impl::sync(bias); |
| 17944 | bias_ = at::functionalization::impl::from_functional_tensor(bias); |
| 17945 | } else { |
| 17946 | bias_ = bias; |
| 17947 | } |
| 17948 | |
| 17949 | at::Tensor out_; |
| 17950 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 17951 | at::functionalization::impl::sync(out); |
| 17952 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 17953 | } else { |
| 17954 | out_ = out; |
| 17955 | } |
| 17956 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 17957 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(weight) || at::functionalization::impl::isFunctionalTensor(bias))) { |
| 17958 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 17959 | TORCH_INTERNAL_ASSERT(false, |
| 17960 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 17961 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 17962 | } else { |
| 17963 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 17964 | at::AutoDispatchSkipFunctionalize guard; |
| 17965 | at::Tensor tmp_output = at::_ops::slow_conv_dilated3d_out::call(self_, weight_, kernel_size, bias_, stride, padding, dilation, out_); |
| 17966 | return out;; |
| 17967 | } |
| 17968 | } else { |
| 17969 | at::Tensor tmp_output; |
| 17970 | { |
| 17971 | at::AutoDispatchSkipFunctionalize guard; |
| 17972 | tmp_output = at::_ops::slow_conv_dilated3d::call(self_, weight_, kernel_size, bias_, stride, padding, dilation); |
| 17973 | } |
| 17974 | at::functionalization::impl::replace_(out, tmp_output); |
| 17975 | at::functionalization::impl::commit_update(out); |
| 17976 | at::functionalization::impl::sync(out); |
| 17977 | return out; |
| 17978 | } |
| 17979 | } |
| 17980 | |
| 17981 | at::Tensor & special_log_ndtr_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 17982 | if (false) { |
| 17983 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 17984 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 17985 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 17986 | auto self_meta = to_meta(self); |
| 17987 | auto out_meta = to_meta(out); |
| 17988 | at::AutoDispatchSkipFunctionalize func_guard; |
| 17989 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 17990 | at::_ops::special_log_ndtr_out::call(self_meta, out_meta); |
| 17991 | } |
| 17992 | |
| 17993 | at::Tensor self_; |
| 17994 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 17995 | at::functionalization::impl::sync(self); |
| 17996 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 17997 | } else { |
| 17998 | self_ = self; |
| 17999 | } |
| 18000 | |
| 18001 | at::Tensor out_; |
| 18002 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18003 | at::functionalization::impl::sync(out); |
| 18004 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18005 | } else { |
| 18006 | out_ = out; |
| 18007 | } |
| 18008 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18009 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18010 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18011 | TORCH_INTERNAL_ASSERT(false, |
| 18012 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18013 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18014 | } else { |
| 18015 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18016 | at::AutoDispatchSkipFunctionalize guard; |
| 18017 | at::Tensor tmp_output = at::_ops::special_log_ndtr_out::call(self_, out_); |
| 18018 | return out;; |
| 18019 | } |
| 18020 | } else { |
| 18021 | at::Tensor tmp_output; |
| 18022 | { |
| 18023 | at::AutoDispatchSkipFunctionalize guard; |
| 18024 | tmp_output = at::_ops::special_log_ndtr::call(self_); |
| 18025 | } |
| 18026 | at::functionalization::impl::replace_(out, tmp_output); |
| 18027 | at::functionalization::impl::commit_update(out); |
| 18028 | at::functionalization::impl::sync(out); |
| 18029 | return out; |
| 18030 | } |
| 18031 | } |
| 18032 | |
| 18033 | at::Tensor & special_exp2_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 18034 | if (false) { |
| 18035 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18036 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18037 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18038 | auto self_meta = to_meta(self); |
| 18039 | auto out_meta = to_meta(out); |
| 18040 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18041 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18042 | at::_ops::special_exp2_out::call(self_meta, out_meta); |
| 18043 | } |
| 18044 | |
| 18045 | at::Tensor self_; |
| 18046 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18047 | at::functionalization::impl::sync(self); |
| 18048 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18049 | } else { |
| 18050 | self_ = self; |
| 18051 | } |
| 18052 | |
| 18053 | at::Tensor out_; |
| 18054 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18055 | at::functionalization::impl::sync(out); |
| 18056 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18057 | } else { |
| 18058 | out_ = out; |
| 18059 | } |
| 18060 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18061 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18062 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18063 | TORCH_INTERNAL_ASSERT(false, |
| 18064 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18065 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18066 | } else { |
| 18067 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18068 | at::AutoDispatchSkipFunctionalize guard; |
| 18069 | at::Tensor tmp_output = at::_ops::special_exp2_out::call(self_, out_); |
| 18070 | return out;; |
| 18071 | } |
| 18072 | } else { |
| 18073 | at::Tensor tmp_output; |
| 18074 | { |
| 18075 | at::AutoDispatchSkipFunctionalize guard; |
| 18076 | tmp_output = at::_ops::special_exp2::call(self_); |
| 18077 | } |
| 18078 | at::functionalization::impl::replace_(out, tmp_output); |
| 18079 | at::functionalization::impl::commit_update(out); |
| 18080 | at::functionalization::impl::sync(out); |
| 18081 | return out; |
| 18082 | } |
| 18083 | } |
| 18084 | |
| 18085 | at::Tensor & special_digamma_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 18086 | if (false) { |
| 18087 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18088 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18089 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18090 | auto self_meta = to_meta(self); |
| 18091 | auto out_meta = to_meta(out); |
| 18092 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18093 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18094 | at::_ops::special_digamma_out::call(self_meta, out_meta); |
| 18095 | } |
| 18096 | |
| 18097 | at::Tensor self_; |
| 18098 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18099 | at::functionalization::impl::sync(self); |
| 18100 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18101 | } else { |
| 18102 | self_ = self; |
| 18103 | } |
| 18104 | |
| 18105 | at::Tensor out_; |
| 18106 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18107 | at::functionalization::impl::sync(out); |
| 18108 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18109 | } else { |
| 18110 | out_ = out; |
| 18111 | } |
| 18112 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18113 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18114 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18115 | TORCH_INTERNAL_ASSERT(false, |
| 18116 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18117 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18118 | } else { |
| 18119 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18120 | at::AutoDispatchSkipFunctionalize guard; |
| 18121 | at::Tensor tmp_output = at::_ops::special_digamma_out::call(self_, out_); |
| 18122 | return out;; |
| 18123 | } |
| 18124 | } else { |
| 18125 | at::Tensor tmp_output; |
| 18126 | { |
| 18127 | at::AutoDispatchSkipFunctionalize guard; |
| 18128 | tmp_output = at::_ops::special_digamma::call(self_); |
| 18129 | } |
| 18130 | at::functionalization::impl::replace_(out, tmp_output); |
| 18131 | at::functionalization::impl::commit_update(out); |
| 18132 | at::functionalization::impl::sync(out); |
| 18133 | return out; |
| 18134 | } |
| 18135 | } |
| 18136 | |
| 18137 | at::Tensor & special_gammaln_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 18138 | if (false) { |
| 18139 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18140 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18141 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18142 | auto self_meta = to_meta(self); |
| 18143 | auto out_meta = to_meta(out); |
| 18144 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18145 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18146 | at::_ops::special_gammaln_out::call(self_meta, out_meta); |
| 18147 | } |
| 18148 | |
| 18149 | at::Tensor self_; |
| 18150 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18151 | at::functionalization::impl::sync(self); |
| 18152 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18153 | } else { |
| 18154 | self_ = self; |
| 18155 | } |
| 18156 | |
| 18157 | at::Tensor out_; |
| 18158 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18159 | at::functionalization::impl::sync(out); |
| 18160 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18161 | } else { |
| 18162 | out_ = out; |
| 18163 | } |
| 18164 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18165 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18166 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18167 | TORCH_INTERNAL_ASSERT(false, |
| 18168 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18169 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18170 | } else { |
| 18171 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18172 | at::AutoDispatchSkipFunctionalize guard; |
| 18173 | at::Tensor tmp_output = at::_ops::special_gammaln_out::call(self_, out_); |
| 18174 | return out;; |
| 18175 | } |
| 18176 | } else { |
| 18177 | at::Tensor tmp_output; |
| 18178 | { |
| 18179 | at::AutoDispatchSkipFunctionalize guard; |
| 18180 | tmp_output = at::_ops::special_gammaln::call(self_); |
| 18181 | } |
| 18182 | at::functionalization::impl::replace_(out, tmp_output); |
| 18183 | at::functionalization::impl::commit_update(out); |
| 18184 | at::functionalization::impl::sync(out); |
| 18185 | return out; |
| 18186 | } |
| 18187 | } |
| 18188 | |
| 18189 | at::Tensor & special_erfcx_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 18190 | if (false) { |
| 18191 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18192 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18193 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18194 | auto self_meta = to_meta(self); |
| 18195 | auto out_meta = to_meta(out); |
| 18196 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18197 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18198 | at::_ops::special_erfcx_out::call(self_meta, out_meta); |
| 18199 | } |
| 18200 | |
| 18201 | at::Tensor self_; |
| 18202 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18203 | at::functionalization::impl::sync(self); |
| 18204 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18205 | } else { |
| 18206 | self_ = self; |
| 18207 | } |
| 18208 | |
| 18209 | at::Tensor out_; |
| 18210 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18211 | at::functionalization::impl::sync(out); |
| 18212 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18213 | } else { |
| 18214 | out_ = out; |
| 18215 | } |
| 18216 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18217 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18218 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18219 | TORCH_INTERNAL_ASSERT(false, |
| 18220 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18221 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18222 | } else { |
| 18223 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18224 | at::AutoDispatchSkipFunctionalize guard; |
| 18225 | at::Tensor tmp_output = at::_ops::special_erfcx_out::call(self_, out_); |
| 18226 | return out;; |
| 18227 | } |
| 18228 | } else { |
| 18229 | at::Tensor tmp_output; |
| 18230 | { |
| 18231 | at::AutoDispatchSkipFunctionalize guard; |
| 18232 | tmp_output = at::_ops::special_erfcx::call(self_); |
| 18233 | } |
| 18234 | at::functionalization::impl::replace_(out, tmp_output); |
| 18235 | at::functionalization::impl::commit_update(out); |
| 18236 | at::functionalization::impl::sync(out); |
| 18237 | return out; |
| 18238 | } |
| 18239 | } |
| 18240 | |
| 18241 | at::Tensor & special_xlog1py_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 18242 | if (false) { |
| 18243 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18244 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18245 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18246 | auto self_meta = to_meta(self); |
| 18247 | auto other_meta = to_meta(other); |
| 18248 | auto out_meta = to_meta(out); |
| 18249 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18250 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18251 | at::_ops::special_xlog1py_out::call(self_meta, other_meta, out_meta); |
| 18252 | } |
| 18253 | |
| 18254 | at::Tensor self_; |
| 18255 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18256 | at::functionalization::impl::sync(self); |
| 18257 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18258 | } else { |
| 18259 | self_ = self; |
| 18260 | } |
| 18261 | |
| 18262 | at::Tensor other_; |
| 18263 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 18264 | at::functionalization::impl::sync(other); |
| 18265 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 18266 | } else { |
| 18267 | other_ = other; |
| 18268 | } |
| 18269 | |
| 18270 | at::Tensor out_; |
| 18271 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18272 | at::functionalization::impl::sync(out); |
| 18273 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18274 | } else { |
| 18275 | out_ = out; |
| 18276 | } |
| 18277 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18278 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 18279 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18280 | TORCH_INTERNAL_ASSERT(false, |
| 18281 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18282 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18283 | } else { |
| 18284 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18285 | at::AutoDispatchSkipFunctionalize guard; |
| 18286 | at::Tensor tmp_output = at::_ops::special_xlog1py_out::call(self_, other_, out_); |
| 18287 | return out;; |
| 18288 | } |
| 18289 | } else { |
| 18290 | at::Tensor tmp_output; |
| 18291 | { |
| 18292 | at::AutoDispatchSkipFunctionalize guard; |
| 18293 | tmp_output = at::_ops::special_xlog1py::call(self_, other_); |
| 18294 | } |
| 18295 | at::functionalization::impl::replace_(out, tmp_output); |
| 18296 | at::functionalization::impl::commit_update(out); |
| 18297 | at::functionalization::impl::sync(out); |
| 18298 | return out; |
| 18299 | } |
| 18300 | } |
| 18301 | |
| 18302 | at::Tensor & special_xlog1py_out_self_scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Scalar & self, const at::Tensor & other, at::Tensor & out) { |
| 18303 | if (false) { |
| 18304 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18305 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18306 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18307 | auto other_meta = to_meta(other); |
| 18308 | auto out_meta = to_meta(out); |
| 18309 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18310 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18311 | at::_ops::special_xlog1py_self_scalar_out::call(self, other_meta, out_meta); |
| 18312 | } |
| 18313 | |
| 18314 | at::Tensor other_; |
| 18315 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 18316 | at::functionalization::impl::sync(other); |
| 18317 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 18318 | } else { |
| 18319 | other_ = other; |
| 18320 | } |
| 18321 | |
| 18322 | at::Tensor out_; |
| 18323 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18324 | at::functionalization::impl::sync(out); |
| 18325 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18326 | } else { |
| 18327 | out_ = out; |
| 18328 | } |
| 18329 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18330 | if ((false || at::functionalization::impl::isFunctionalTensor(other))) { |
| 18331 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18332 | TORCH_INTERNAL_ASSERT(false, |
| 18333 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18334 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18335 | } else { |
| 18336 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18337 | at::AutoDispatchSkipFunctionalize guard; |
| 18338 | at::Tensor tmp_output = at::_ops::special_xlog1py_self_scalar_out::call(self, other_, out_); |
| 18339 | return out;; |
| 18340 | } |
| 18341 | } else { |
| 18342 | at::Tensor tmp_output; |
| 18343 | { |
| 18344 | at::AutoDispatchSkipFunctionalize guard; |
| 18345 | tmp_output = at::_ops::special_xlog1py_self_scalar::call(self, other_); |
| 18346 | } |
| 18347 | at::functionalization::impl::replace_(out, tmp_output); |
| 18348 | at::functionalization::impl::commit_update(out); |
| 18349 | at::functionalization::impl::sync(out); |
| 18350 | return out; |
| 18351 | } |
| 18352 | } |
| 18353 | |
| 18354 | at::Tensor & special_xlog1py_out_other_scalar_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & other, at::Tensor & out) { |
| 18355 | if (false) { |
| 18356 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18357 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18358 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18359 | auto self_meta = to_meta(self); |
| 18360 | auto out_meta = to_meta(out); |
| 18361 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18362 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18363 | at::_ops::special_xlog1py_other_scalar_out::call(self_meta, other, out_meta); |
| 18364 | } |
| 18365 | |
| 18366 | at::Tensor self_; |
| 18367 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18368 | at::functionalization::impl::sync(self); |
| 18369 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18370 | } else { |
| 18371 | self_ = self; |
| 18372 | } |
| 18373 | |
| 18374 | at::Tensor out_; |
| 18375 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18376 | at::functionalization::impl::sync(out); |
| 18377 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18378 | } else { |
| 18379 | out_ = out; |
| 18380 | } |
| 18381 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18382 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18383 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18384 | TORCH_INTERNAL_ASSERT(false, |
| 18385 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18386 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18387 | } else { |
| 18388 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18389 | at::AutoDispatchSkipFunctionalize guard; |
| 18390 | at::Tensor tmp_output = at::_ops::special_xlog1py_other_scalar_out::call(self_, other, out_); |
| 18391 | return out;; |
| 18392 | } |
| 18393 | } else { |
| 18394 | at::Tensor tmp_output; |
| 18395 | { |
| 18396 | at::AutoDispatchSkipFunctionalize guard; |
| 18397 | tmp_output = at::_ops::special_xlog1py_other_scalar::call(self_, other); |
| 18398 | } |
| 18399 | at::functionalization::impl::replace_(out, tmp_output); |
| 18400 | at::functionalization::impl::commit_update(out); |
| 18401 | at::functionalization::impl::sync(out); |
| 18402 | return out; |
| 18403 | } |
| 18404 | } |
| 18405 | |
| 18406 | at::Tensor & special_i1_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 18407 | if (false) { |
| 18408 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18409 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18410 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18411 | auto self_meta = to_meta(self); |
| 18412 | auto out_meta = to_meta(out); |
| 18413 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18414 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18415 | at::_ops::special_i1_out::call(self_meta, out_meta); |
| 18416 | } |
| 18417 | |
| 18418 | at::Tensor self_; |
| 18419 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18420 | at::functionalization::impl::sync(self); |
| 18421 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18422 | } else { |
| 18423 | self_ = self; |
| 18424 | } |
| 18425 | |
| 18426 | at::Tensor out_; |
| 18427 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18428 | at::functionalization::impl::sync(out); |
| 18429 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18430 | } else { |
| 18431 | out_ = out; |
| 18432 | } |
| 18433 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18434 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18435 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18436 | TORCH_INTERNAL_ASSERT(false, |
| 18437 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18438 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18439 | } else { |
| 18440 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18441 | at::AutoDispatchSkipFunctionalize guard; |
| 18442 | at::Tensor tmp_output = at::_ops::special_i1_out::call(self_, out_); |
| 18443 | return out;; |
| 18444 | } |
| 18445 | } else { |
| 18446 | at::Tensor tmp_output; |
| 18447 | { |
| 18448 | at::AutoDispatchSkipFunctionalize guard; |
| 18449 | tmp_output = at::_ops::special_i1::call(self_); |
| 18450 | } |
| 18451 | at::functionalization::impl::replace_(out, tmp_output); |
| 18452 | at::functionalization::impl::commit_update(out); |
| 18453 | at::functionalization::impl::sync(out); |
| 18454 | return out; |
| 18455 | } |
| 18456 | } |
| 18457 | |
| 18458 | at::Tensor & special_i1e_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 18459 | if (false) { |
| 18460 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18461 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18462 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18463 | auto self_meta = to_meta(self); |
| 18464 | auto out_meta = to_meta(out); |
| 18465 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18466 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18467 | at::_ops::special_i1e_out::call(self_meta, out_meta); |
| 18468 | } |
| 18469 | |
| 18470 | at::Tensor self_; |
| 18471 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18472 | at::functionalization::impl::sync(self); |
| 18473 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18474 | } else { |
| 18475 | self_ = self; |
| 18476 | } |
| 18477 | |
| 18478 | at::Tensor out_; |
| 18479 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18480 | at::functionalization::impl::sync(out); |
| 18481 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18482 | } else { |
| 18483 | out_ = out; |
| 18484 | } |
| 18485 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18486 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18487 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18488 | TORCH_INTERNAL_ASSERT(false, |
| 18489 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18490 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18491 | } else { |
| 18492 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18493 | at::AutoDispatchSkipFunctionalize guard; |
| 18494 | at::Tensor tmp_output = at::_ops::special_i1e_out::call(self_, out_); |
| 18495 | return out;; |
| 18496 | } |
| 18497 | } else { |
| 18498 | at::Tensor tmp_output; |
| 18499 | { |
| 18500 | at::AutoDispatchSkipFunctionalize guard; |
| 18501 | tmp_output = at::_ops::special_i1e::call(self_); |
| 18502 | } |
| 18503 | at::functionalization::impl::replace_(out, tmp_output); |
| 18504 | at::functionalization::impl::commit_update(out); |
| 18505 | at::functionalization::impl::sync(out); |
| 18506 | return out; |
| 18507 | } |
| 18508 | } |
| 18509 | |
| 18510 | at::Tensor & fft_fft_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm, at::Tensor & out) { |
| 18511 | if (false) { |
| 18512 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18513 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18514 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18515 | auto self_meta = to_meta(self); |
| 18516 | auto out_meta = to_meta(out); |
| 18517 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18518 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18519 | at::_ops::fft_fft_out::call(self_meta, n, dim, norm, out_meta); |
| 18520 | } |
| 18521 | |
| 18522 | at::Tensor self_; |
| 18523 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18524 | at::functionalization::impl::sync(self); |
| 18525 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18526 | } else { |
| 18527 | self_ = self; |
| 18528 | } |
| 18529 | |
| 18530 | at::Tensor out_; |
| 18531 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18532 | at::functionalization::impl::sync(out); |
| 18533 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18534 | } else { |
| 18535 | out_ = out; |
| 18536 | } |
| 18537 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18538 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18539 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18540 | TORCH_INTERNAL_ASSERT(false, |
| 18541 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18542 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18543 | } else { |
| 18544 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18545 | at::AutoDispatchSkipFunctionalize guard; |
| 18546 | at::Tensor tmp_output = at::_ops::fft_fft_out::call(self_, n, dim, norm, out_); |
| 18547 | return out;; |
| 18548 | } |
| 18549 | } else { |
| 18550 | at::Tensor tmp_output; |
| 18551 | { |
| 18552 | at::AutoDispatchSkipFunctionalize guard; |
| 18553 | tmp_output = at::_ops::fft_fft::call(self_, n, dim, norm); |
| 18554 | } |
| 18555 | at::functionalization::impl::replace_(out, tmp_output); |
| 18556 | at::functionalization::impl::commit_update(out); |
| 18557 | at::functionalization::impl::sync(out); |
| 18558 | return out; |
| 18559 | } |
| 18560 | } |
| 18561 | |
| 18562 | at::Tensor & fft_rfft_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm, at::Tensor & out) { |
| 18563 | if (false) { |
| 18564 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18565 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18566 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18567 | auto self_meta = to_meta(self); |
| 18568 | auto out_meta = to_meta(out); |
| 18569 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18570 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18571 | at::_ops::fft_rfft_out::call(self_meta, n, dim, norm, out_meta); |
| 18572 | } |
| 18573 | |
| 18574 | at::Tensor self_; |
| 18575 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18576 | at::functionalization::impl::sync(self); |
| 18577 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18578 | } else { |
| 18579 | self_ = self; |
| 18580 | } |
| 18581 | |
| 18582 | at::Tensor out_; |
| 18583 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18584 | at::functionalization::impl::sync(out); |
| 18585 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18586 | } else { |
| 18587 | out_ = out; |
| 18588 | } |
| 18589 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18590 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18591 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18592 | TORCH_INTERNAL_ASSERT(false, |
| 18593 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18594 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18595 | } else { |
| 18596 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18597 | at::AutoDispatchSkipFunctionalize guard; |
| 18598 | at::Tensor tmp_output = at::_ops::fft_rfft_out::call(self_, n, dim, norm, out_); |
| 18599 | return out;; |
| 18600 | } |
| 18601 | } else { |
| 18602 | at::Tensor tmp_output; |
| 18603 | { |
| 18604 | at::AutoDispatchSkipFunctionalize guard; |
| 18605 | tmp_output = at::_ops::fft_rfft::call(self_, n, dim, norm); |
| 18606 | } |
| 18607 | at::functionalization::impl::replace_(out, tmp_output); |
| 18608 | at::functionalization::impl::commit_update(out); |
| 18609 | at::functionalization::impl::sync(out); |
| 18610 | return out; |
| 18611 | } |
| 18612 | } |
| 18613 | |
| 18614 | at::Tensor & fft_hfft_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm, at::Tensor & out) { |
| 18615 | if (false) { |
| 18616 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18617 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18618 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18619 | auto self_meta = to_meta(self); |
| 18620 | auto out_meta = to_meta(out); |
| 18621 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18622 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18623 | at::_ops::fft_hfft_out::call(self_meta, n, dim, norm, out_meta); |
| 18624 | } |
| 18625 | |
| 18626 | at::Tensor self_; |
| 18627 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18628 | at::functionalization::impl::sync(self); |
| 18629 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18630 | } else { |
| 18631 | self_ = self; |
| 18632 | } |
| 18633 | |
| 18634 | at::Tensor out_; |
| 18635 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18636 | at::functionalization::impl::sync(out); |
| 18637 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18638 | } else { |
| 18639 | out_ = out; |
| 18640 | } |
| 18641 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18642 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18643 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18644 | TORCH_INTERNAL_ASSERT(false, |
| 18645 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18646 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18647 | } else { |
| 18648 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18649 | at::AutoDispatchSkipFunctionalize guard; |
| 18650 | at::Tensor tmp_output = at::_ops::fft_hfft_out::call(self_, n, dim, norm, out_); |
| 18651 | return out;; |
| 18652 | } |
| 18653 | } else { |
| 18654 | at::Tensor tmp_output; |
| 18655 | { |
| 18656 | at::AutoDispatchSkipFunctionalize guard; |
| 18657 | tmp_output = at::_ops::fft_hfft::call(self_, n, dim, norm); |
| 18658 | } |
| 18659 | at::functionalization::impl::replace_(out, tmp_output); |
| 18660 | at::functionalization::impl::commit_update(out); |
| 18661 | at::functionalization::impl::sync(out); |
| 18662 | return out; |
| 18663 | } |
| 18664 | } |
| 18665 | |
| 18666 | const at::Tensor & fft_hfft2_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::OptionalIntArrayRef s, at::IntArrayRef dim, c10::optional<c10::string_view> norm, const at::Tensor & out) { |
| 18667 | if (false) { |
| 18668 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18669 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18670 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18671 | auto self_meta = to_meta(self); |
| 18672 | auto out_meta = to_meta(out); |
| 18673 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18674 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18675 | at::_ops::fft_hfft2_out::call(self_meta, s, dim, norm, out_meta); |
| 18676 | } |
| 18677 | |
| 18678 | at::Tensor self_; |
| 18679 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18680 | at::functionalization::impl::sync(self); |
| 18681 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18682 | } else { |
| 18683 | self_ = self; |
| 18684 | } |
| 18685 | |
| 18686 | at::Tensor out_; |
| 18687 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18688 | at::functionalization::impl::sync(out); |
| 18689 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18690 | } else { |
| 18691 | out_ = out; |
| 18692 | } |
| 18693 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18694 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18695 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18696 | TORCH_INTERNAL_ASSERT(false, |
| 18697 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18698 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18699 | } else { |
| 18700 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18701 | at::AutoDispatchSkipFunctionalize guard; |
| 18702 | at::Tensor tmp_output = at::_ops::fft_hfft2_out::call(self_, s, dim, norm, out_); |
| 18703 | return out;; |
| 18704 | } |
| 18705 | } else { |
| 18706 | at::Tensor tmp_output; |
| 18707 | { |
| 18708 | at::AutoDispatchSkipFunctionalize guard; |
| 18709 | tmp_output = at::_ops::fft_hfft2::call(self_, s, dim, norm); |
| 18710 | } |
| 18711 | at::functionalization::impl::replace_(out, tmp_output); |
| 18712 | at::functionalization::impl::commit_update(out); |
| 18713 | at::functionalization::impl::sync(out); |
| 18714 | return out; |
| 18715 | } |
| 18716 | } |
| 18717 | |
| 18718 | at::Tensor & fft_ifftn_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm, at::Tensor & out) { |
| 18719 | if (false) { |
| 18720 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18721 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18722 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18723 | auto self_meta = to_meta(self); |
| 18724 | auto out_meta = to_meta(out); |
| 18725 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18726 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18727 | at::_ops::fft_ifftn_out::call(self_meta, s, dim, norm, out_meta); |
| 18728 | } |
| 18729 | |
| 18730 | at::Tensor self_; |
| 18731 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18732 | at::functionalization::impl::sync(self); |
| 18733 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18734 | } else { |
| 18735 | self_ = self; |
| 18736 | } |
| 18737 | |
| 18738 | at::Tensor out_; |
| 18739 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18740 | at::functionalization::impl::sync(out); |
| 18741 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18742 | } else { |
| 18743 | out_ = out; |
| 18744 | } |
| 18745 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18746 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18747 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18748 | TORCH_INTERNAL_ASSERT(false, |
| 18749 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18750 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18751 | } else { |
| 18752 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18753 | at::AutoDispatchSkipFunctionalize guard; |
| 18754 | at::Tensor tmp_output = at::_ops::fft_ifftn_out::call(self_, s, dim, norm, out_); |
| 18755 | return out;; |
| 18756 | } |
| 18757 | } else { |
| 18758 | at::Tensor tmp_output; |
| 18759 | { |
| 18760 | at::AutoDispatchSkipFunctionalize guard; |
| 18761 | tmp_output = at::_ops::fft_ifftn::call(self_, s, dim, norm); |
| 18762 | } |
| 18763 | at::functionalization::impl::replace_(out, tmp_output); |
| 18764 | at::functionalization::impl::commit_update(out); |
| 18765 | at::functionalization::impl::sync(out); |
| 18766 | return out; |
| 18767 | } |
| 18768 | } |
| 18769 | |
| 18770 | at::Tensor & fft_rfftn_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm, at::Tensor & out) { |
| 18771 | if (false) { |
| 18772 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18773 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18774 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18775 | auto self_meta = to_meta(self); |
| 18776 | auto out_meta = to_meta(out); |
| 18777 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18778 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18779 | at::_ops::fft_rfftn_out::call(self_meta, s, dim, norm, out_meta); |
| 18780 | } |
| 18781 | |
| 18782 | at::Tensor self_; |
| 18783 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18784 | at::functionalization::impl::sync(self); |
| 18785 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18786 | } else { |
| 18787 | self_ = self; |
| 18788 | } |
| 18789 | |
| 18790 | at::Tensor out_; |
| 18791 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18792 | at::functionalization::impl::sync(out); |
| 18793 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18794 | } else { |
| 18795 | out_ = out; |
| 18796 | } |
| 18797 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18798 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18799 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18800 | TORCH_INTERNAL_ASSERT(false, |
| 18801 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18802 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18803 | } else { |
| 18804 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18805 | at::AutoDispatchSkipFunctionalize guard; |
| 18806 | at::Tensor tmp_output = at::_ops::fft_rfftn_out::call(self_, s, dim, norm, out_); |
| 18807 | return out;; |
| 18808 | } |
| 18809 | } else { |
| 18810 | at::Tensor tmp_output; |
| 18811 | { |
| 18812 | at::AutoDispatchSkipFunctionalize guard; |
| 18813 | tmp_output = at::_ops::fft_rfftn::call(self_, s, dim, norm); |
| 18814 | } |
| 18815 | at::functionalization::impl::replace_(out, tmp_output); |
| 18816 | at::functionalization::impl::commit_update(out); |
| 18817 | at::functionalization::impl::sync(out); |
| 18818 | return out; |
| 18819 | } |
| 18820 | } |
| 18821 | |
| 18822 | const at::Tensor & fft_hfftn_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm, const at::Tensor & out) { |
| 18823 | if (false) { |
| 18824 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18825 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18826 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18827 | auto self_meta = to_meta(self); |
| 18828 | auto out_meta = to_meta(out); |
| 18829 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18830 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18831 | at::_ops::fft_hfftn_out::call(self_meta, s, dim, norm, out_meta); |
| 18832 | } |
| 18833 | |
| 18834 | at::Tensor self_; |
| 18835 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18836 | at::functionalization::impl::sync(self); |
| 18837 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18838 | } else { |
| 18839 | self_ = self; |
| 18840 | } |
| 18841 | |
| 18842 | at::Tensor out_; |
| 18843 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18844 | at::functionalization::impl::sync(out); |
| 18845 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18846 | } else { |
| 18847 | out_ = out; |
| 18848 | } |
| 18849 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18850 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 18851 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18852 | TORCH_INTERNAL_ASSERT(false, |
| 18853 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18854 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18855 | } else { |
| 18856 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18857 | at::AutoDispatchSkipFunctionalize guard; |
| 18858 | at::Tensor tmp_output = at::_ops::fft_hfftn_out::call(self_, s, dim, norm, out_); |
| 18859 | return out;; |
| 18860 | } |
| 18861 | } else { |
| 18862 | at::Tensor tmp_output; |
| 18863 | { |
| 18864 | at::AutoDispatchSkipFunctionalize guard; |
| 18865 | tmp_output = at::_ops::fft_hfftn::call(self_, s, dim, norm); |
| 18866 | } |
| 18867 | at::functionalization::impl::replace_(out, tmp_output); |
| 18868 | at::functionalization::impl::commit_update(out); |
| 18869 | at::functionalization::impl::sync(out); |
| 18870 | return out; |
| 18871 | } |
| 18872 | } |
| 18873 | |
| 18874 | at::Tensor & fft_fftfreq_out_out(c10::DispatchKeySet dispatchKeySet, int64_t n, double d, at::Tensor & out) { |
| 18875 | if (false) { |
| 18876 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18877 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18878 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18879 | auto out_meta = to_meta(out); |
| 18880 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18881 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18882 | at::_ops::fft_fftfreq_out::call(n, d, out_meta); |
| 18883 | } |
| 18884 | |
| 18885 | at::Tensor out_; |
| 18886 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18887 | at::functionalization::impl::sync(out); |
| 18888 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18889 | } else { |
| 18890 | out_ = out; |
| 18891 | } |
| 18892 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18893 | if ((false)) { |
| 18894 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18895 | TORCH_INTERNAL_ASSERT(false, |
| 18896 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18897 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18898 | } else { |
| 18899 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18900 | at::AutoDispatchSkipFunctionalize guard; |
| 18901 | at::Tensor tmp_output = at::_ops::fft_fftfreq_out::call(n, d, out_); |
| 18902 | return out;; |
| 18903 | } |
| 18904 | } else { |
| 18905 | at::Tensor tmp_output; |
| 18906 | { |
| 18907 | at::AutoDispatchSkipFunctionalize guard; |
| 18908 | tmp_output = at::_ops::fft_fftfreq::call(n, d, out_.scalar_type(), out_.layout(), out_.device(), c10::nullopt); |
| 18909 | } |
| 18910 | at::functionalization::impl::replace_(out, tmp_output); |
| 18911 | at::functionalization::impl::commit_update(out); |
| 18912 | at::functionalization::impl::sync(out); |
| 18913 | return out; |
| 18914 | } |
| 18915 | } |
| 18916 | |
| 18917 | at::Tensor & linalg_det_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & A, at::Tensor & out) { |
| 18918 | if (false) { |
| 18919 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18920 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18921 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18922 | auto A_meta = to_meta(A); |
| 18923 | auto out_meta = to_meta(out); |
| 18924 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18925 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18926 | at::_ops::linalg_det_out::call(A_meta, out_meta); |
| 18927 | } |
| 18928 | |
| 18929 | at::Tensor A_; |
| 18930 | if (at::functionalization::impl::isFunctionalTensor(A)) { |
| 18931 | at::functionalization::impl::sync(A); |
| 18932 | A_ = at::functionalization::impl::from_functional_tensor(A); |
| 18933 | } else { |
| 18934 | A_ = A; |
| 18935 | } |
| 18936 | |
| 18937 | at::Tensor out_; |
| 18938 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 18939 | at::functionalization::impl::sync(out); |
| 18940 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 18941 | } else { |
| 18942 | out_ = out; |
| 18943 | } |
| 18944 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 18945 | if ((false || at::functionalization::impl::isFunctionalTensor(A))) { |
| 18946 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 18947 | TORCH_INTERNAL_ASSERT(false, |
| 18948 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 18949 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 18950 | } else { |
| 18951 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 18952 | at::AutoDispatchSkipFunctionalize guard; |
| 18953 | at::Tensor tmp_output = at::_ops::linalg_det_out::call(A_, out_); |
| 18954 | return out;; |
| 18955 | } |
| 18956 | } else { |
| 18957 | at::Tensor tmp_output; |
| 18958 | { |
| 18959 | at::AutoDispatchSkipFunctionalize guard; |
| 18960 | tmp_output = at::_ops::linalg_det::call(A_); |
| 18961 | } |
| 18962 | at::functionalization::impl::replace_(out, tmp_output); |
| 18963 | at::functionalization::impl::commit_update(out); |
| 18964 | at::functionalization::impl::sync(out); |
| 18965 | return out; |
| 18966 | } |
| 18967 | } |
| 18968 | |
| 18969 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> linalg_ldl_factor_ex_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, bool hermitian, bool check_errors, at::Tensor & LD, at::Tensor & pivots, at::Tensor & info) { |
| 18970 | if (false) { |
| 18971 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 18972 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 18973 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 18974 | auto self_meta = to_meta(self); |
| 18975 | auto LD_meta = to_meta(LD); |
| 18976 | auto pivots_meta = to_meta(pivots); |
| 18977 | auto info_meta = to_meta(info); |
| 18978 | at::AutoDispatchSkipFunctionalize func_guard; |
| 18979 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 18980 | at::_ops::linalg_ldl_factor_ex_out::call(self_meta, hermitian, check_errors, LD_meta, pivots_meta, info_meta); |
| 18981 | } |
| 18982 | |
| 18983 | at::Tensor self_; |
| 18984 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 18985 | at::functionalization::impl::sync(self); |
| 18986 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 18987 | } else { |
| 18988 | self_ = self; |
| 18989 | } |
| 18990 | |
| 18991 | at::Tensor LD_; |
| 18992 | if (at::functionalization::impl::isFunctionalTensor(LD)) { |
| 18993 | at::functionalization::impl::sync(LD); |
| 18994 | LD_ = at::functionalization::impl::from_functional_tensor(LD); |
| 18995 | } else { |
| 18996 | LD_ = LD; |
| 18997 | } |
| 18998 | |
| 18999 | at::Tensor pivots_; |
| 19000 | if (at::functionalization::impl::isFunctionalTensor(pivots)) { |
| 19001 | at::functionalization::impl::sync(pivots); |
| 19002 | pivots_ = at::functionalization::impl::from_functional_tensor(pivots); |
| 19003 | } else { |
| 19004 | pivots_ = pivots; |
| 19005 | } |
| 19006 | |
| 19007 | at::Tensor info_; |
| 19008 | if (at::functionalization::impl::isFunctionalTensor(info)) { |
| 19009 | at::functionalization::impl::sync(info); |
| 19010 | info_ = at::functionalization::impl::from_functional_tensor(info); |
| 19011 | } else { |
| 19012 | info_ = info; |
| 19013 | } |
| 19014 | if (!(true && at::functionalization::impl::isFunctionalTensor(LD) && at::functionalization::impl::isFunctionalTensor(pivots) && at::functionalization::impl::isFunctionalTensor(info))) { |
| 19015 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19016 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19017 | TORCH_INTERNAL_ASSERT(false, |
| 19018 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19019 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19020 | } else { |
| 19021 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19022 | at::AutoDispatchSkipFunctionalize guard; |
| 19023 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::linalg_ldl_factor_ex_out::call(self_, hermitian, check_errors, LD_, pivots_, info_); |
| 19024 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(LD, pivots, info);; |
| 19025 | } |
| 19026 | } else { |
| 19027 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 19028 | { |
| 19029 | at::AutoDispatchSkipFunctionalize guard; |
| 19030 | tmp_output = at::_ops::linalg_ldl_factor_ex::call(self_, hermitian, check_errors); |
| 19031 | } |
| 19032 | at::functionalization::impl::replace_(LD, std::get<0>(tmp_output)); |
| 19033 | at::functionalization::impl::commit_update(LD); |
| 19034 | at::functionalization::impl::sync(LD); |
| 19035 | at::functionalization::impl::replace_(pivots, std::get<1>(tmp_output)); |
| 19036 | at::functionalization::impl::commit_update(pivots); |
| 19037 | at::functionalization::impl::sync(pivots); |
| 19038 | at::functionalization::impl::replace_(info, std::get<2>(tmp_output)); |
| 19039 | at::functionalization::impl::commit_update(info); |
| 19040 | at::functionalization::impl::sync(info); |
| 19041 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(LD, pivots, info); |
| 19042 | } |
| 19043 | } |
| 19044 | |
| 19045 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &> linalg_lstsq_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & b, c10::optional<double> rcond, c10::optional<c10::string_view> driver, at::Tensor & solution, at::Tensor & residuals, at::Tensor & rank, at::Tensor & singular_values) { |
| 19046 | if (false) { |
| 19047 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19048 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19049 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19050 | auto self_meta = to_meta(self); |
| 19051 | auto b_meta = to_meta(b); |
| 19052 | auto solution_meta = to_meta(solution); |
| 19053 | auto residuals_meta = to_meta(residuals); |
| 19054 | auto rank_meta = to_meta(rank); |
| 19055 | auto singular_values_meta = to_meta(singular_values); |
| 19056 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19057 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19058 | at::_ops::linalg_lstsq_out::call(self_meta, b_meta, rcond, driver, solution_meta, residuals_meta, rank_meta, singular_values_meta); |
| 19059 | } |
| 19060 | |
| 19061 | at::Tensor self_; |
| 19062 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19063 | at::functionalization::impl::sync(self); |
| 19064 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19065 | } else { |
| 19066 | self_ = self; |
| 19067 | } |
| 19068 | |
| 19069 | at::Tensor b_; |
| 19070 | if (at::functionalization::impl::isFunctionalTensor(b)) { |
| 19071 | at::functionalization::impl::sync(b); |
| 19072 | b_ = at::functionalization::impl::from_functional_tensor(b); |
| 19073 | } else { |
| 19074 | b_ = b; |
| 19075 | } |
| 19076 | |
| 19077 | at::Tensor solution_; |
| 19078 | if (at::functionalization::impl::isFunctionalTensor(solution)) { |
| 19079 | at::functionalization::impl::sync(solution); |
| 19080 | solution_ = at::functionalization::impl::from_functional_tensor(solution); |
| 19081 | } else { |
| 19082 | solution_ = solution; |
| 19083 | } |
| 19084 | |
| 19085 | at::Tensor residuals_; |
| 19086 | if (at::functionalization::impl::isFunctionalTensor(residuals)) { |
| 19087 | at::functionalization::impl::sync(residuals); |
| 19088 | residuals_ = at::functionalization::impl::from_functional_tensor(residuals); |
| 19089 | } else { |
| 19090 | residuals_ = residuals; |
| 19091 | } |
| 19092 | |
| 19093 | at::Tensor rank_; |
| 19094 | if (at::functionalization::impl::isFunctionalTensor(rank)) { |
| 19095 | at::functionalization::impl::sync(rank); |
| 19096 | rank_ = at::functionalization::impl::from_functional_tensor(rank); |
| 19097 | } else { |
| 19098 | rank_ = rank; |
| 19099 | } |
| 19100 | |
| 19101 | at::Tensor singular_values_; |
| 19102 | if (at::functionalization::impl::isFunctionalTensor(singular_values)) { |
| 19103 | at::functionalization::impl::sync(singular_values); |
| 19104 | singular_values_ = at::functionalization::impl::from_functional_tensor(singular_values); |
| 19105 | } else { |
| 19106 | singular_values_ = singular_values; |
| 19107 | } |
| 19108 | if (!(true && at::functionalization::impl::isFunctionalTensor(solution) && at::functionalization::impl::isFunctionalTensor(residuals) && at::functionalization::impl::isFunctionalTensor(rank) && at::functionalization::impl::isFunctionalTensor(singular_values))) { |
| 19109 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(b))) { |
| 19110 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19111 | TORCH_INTERNAL_ASSERT(false, |
| 19112 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19113 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19114 | } else { |
| 19115 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19116 | at::AutoDispatchSkipFunctionalize guard; |
| 19117 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::linalg_lstsq_out::call(self_, b_, rcond, driver, solution_, residuals_, rank_, singular_values_); |
| 19118 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &>(solution, residuals, rank, singular_values);; |
| 19119 | } |
| 19120 | } else { |
| 19121 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 19122 | { |
| 19123 | at::AutoDispatchSkipFunctionalize guard; |
| 19124 | tmp_output = at::_ops::linalg_lstsq::call(self_, b_, rcond, driver); |
| 19125 | } |
| 19126 | at::functionalization::impl::replace_(solution, std::get<0>(tmp_output)); |
| 19127 | at::functionalization::impl::commit_update(solution); |
| 19128 | at::functionalization::impl::sync(solution); |
| 19129 | at::functionalization::impl::replace_(residuals, std::get<1>(tmp_output)); |
| 19130 | at::functionalization::impl::commit_update(residuals); |
| 19131 | at::functionalization::impl::sync(residuals); |
| 19132 | at::functionalization::impl::replace_(rank, std::get<2>(tmp_output)); |
| 19133 | at::functionalization::impl::commit_update(rank); |
| 19134 | at::functionalization::impl::sync(rank); |
| 19135 | at::functionalization::impl::replace_(singular_values, std::get<3>(tmp_output)); |
| 19136 | at::functionalization::impl::commit_update(singular_values); |
| 19137 | at::functionalization::impl::sync(singular_values); |
| 19138 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &>(solution, residuals, rank, singular_values); |
| 19139 | } |
| 19140 | } |
| 19141 | |
| 19142 | at::Tensor & linalg_matrix_exp_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 19143 | if (false) { |
| 19144 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19145 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19146 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19147 | auto self_meta = to_meta(self); |
| 19148 | auto out_meta = to_meta(out); |
| 19149 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19150 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19151 | at::_ops::linalg_matrix_exp_out::call(self_meta, out_meta); |
| 19152 | } |
| 19153 | |
| 19154 | at::Tensor self_; |
| 19155 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19156 | at::functionalization::impl::sync(self); |
| 19157 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19158 | } else { |
| 19159 | self_ = self; |
| 19160 | } |
| 19161 | |
| 19162 | at::Tensor out_; |
| 19163 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19164 | at::functionalization::impl::sync(out); |
| 19165 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19166 | } else { |
| 19167 | out_ = out; |
| 19168 | } |
| 19169 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19170 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19171 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19172 | TORCH_INTERNAL_ASSERT(false, |
| 19173 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19174 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19175 | } else { |
| 19176 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19177 | at::AutoDispatchSkipFunctionalize guard; |
| 19178 | at::Tensor tmp_output = at::_ops::linalg_matrix_exp_out::call(self_, out_); |
| 19179 | return out;; |
| 19180 | } |
| 19181 | } else { |
| 19182 | at::Tensor tmp_output; |
| 19183 | { |
| 19184 | at::AutoDispatchSkipFunctionalize guard; |
| 19185 | tmp_output = at::_ops::linalg_matrix_exp::call(self_); |
| 19186 | } |
| 19187 | at::functionalization::impl::replace_(out, tmp_output); |
| 19188 | at::functionalization::impl::commit_update(out); |
| 19189 | at::functionalization::impl::sync(out); |
| 19190 | return out; |
| 19191 | } |
| 19192 | } |
| 19193 | |
| 19194 | ::std::tuple<at::Tensor &,at::Tensor &> linalg_slogdet_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & A, at::Tensor & sign, at::Tensor & logabsdet) { |
| 19195 | if (false) { |
| 19196 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19197 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19198 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19199 | auto A_meta = to_meta(A); |
| 19200 | auto sign_meta = to_meta(sign); |
| 19201 | auto logabsdet_meta = to_meta(logabsdet); |
| 19202 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19203 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19204 | at::_ops::linalg_slogdet_out::call(A_meta, sign_meta, logabsdet_meta); |
| 19205 | } |
| 19206 | |
| 19207 | at::Tensor A_; |
| 19208 | if (at::functionalization::impl::isFunctionalTensor(A)) { |
| 19209 | at::functionalization::impl::sync(A); |
| 19210 | A_ = at::functionalization::impl::from_functional_tensor(A); |
| 19211 | } else { |
| 19212 | A_ = A; |
| 19213 | } |
| 19214 | |
| 19215 | at::Tensor sign_; |
| 19216 | if (at::functionalization::impl::isFunctionalTensor(sign)) { |
| 19217 | at::functionalization::impl::sync(sign); |
| 19218 | sign_ = at::functionalization::impl::from_functional_tensor(sign); |
| 19219 | } else { |
| 19220 | sign_ = sign; |
| 19221 | } |
| 19222 | |
| 19223 | at::Tensor logabsdet_; |
| 19224 | if (at::functionalization::impl::isFunctionalTensor(logabsdet)) { |
| 19225 | at::functionalization::impl::sync(logabsdet); |
| 19226 | logabsdet_ = at::functionalization::impl::from_functional_tensor(logabsdet); |
| 19227 | } else { |
| 19228 | logabsdet_ = logabsdet; |
| 19229 | } |
| 19230 | if (!(true && at::functionalization::impl::isFunctionalTensor(sign) && at::functionalization::impl::isFunctionalTensor(logabsdet))) { |
| 19231 | if ((false || at::functionalization::impl::isFunctionalTensor(A))) { |
| 19232 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19233 | TORCH_INTERNAL_ASSERT(false, |
| 19234 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19235 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19236 | } else { |
| 19237 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19238 | at::AutoDispatchSkipFunctionalize guard; |
| 19239 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::linalg_slogdet_out::call(A_, sign_, logabsdet_); |
| 19240 | return ::std::tuple<at::Tensor &,at::Tensor &>(sign, logabsdet);; |
| 19241 | } |
| 19242 | } else { |
| 19243 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 19244 | { |
| 19245 | at::AutoDispatchSkipFunctionalize guard; |
| 19246 | tmp_output = at::_ops::linalg_slogdet::call(A_); |
| 19247 | } |
| 19248 | at::functionalization::impl::replace_(sign, std::get<0>(tmp_output)); |
| 19249 | at::functionalization::impl::commit_update(sign); |
| 19250 | at::functionalization::impl::sync(sign); |
| 19251 | at::functionalization::impl::replace_(logabsdet, std::get<1>(tmp_output)); |
| 19252 | at::functionalization::impl::commit_update(logabsdet); |
| 19253 | at::functionalization::impl::sync(logabsdet); |
| 19254 | return ::std::tuple<at::Tensor &,at::Tensor &>(sign, logabsdet); |
| 19255 | } |
| 19256 | } |
| 19257 | |
| 19258 | ::std::tuple<at::Tensor &,at::Tensor &> _linalg_eigh_out_eigenvalues(c10::DispatchKeySet dispatchKeySet, const at::Tensor & A, c10::string_view UPLO, bool compute_v, at::Tensor & eigenvalues, at::Tensor & eigenvectors) { |
| 19259 | if (false) { |
| 19260 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19261 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19262 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19263 | auto A_meta = to_meta(A); |
| 19264 | auto eigenvalues_meta = to_meta(eigenvalues); |
| 19265 | auto eigenvectors_meta = to_meta(eigenvectors); |
| 19266 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19267 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19268 | at::_ops::_linalg_eigh_eigenvalues::call(A_meta, UPLO, compute_v, eigenvalues_meta, eigenvectors_meta); |
| 19269 | } |
| 19270 | |
| 19271 | at::Tensor A_; |
| 19272 | if (at::functionalization::impl::isFunctionalTensor(A)) { |
| 19273 | at::functionalization::impl::sync(A); |
| 19274 | A_ = at::functionalization::impl::from_functional_tensor(A); |
| 19275 | } else { |
| 19276 | A_ = A; |
| 19277 | } |
| 19278 | |
| 19279 | at::Tensor eigenvalues_; |
| 19280 | if (at::functionalization::impl::isFunctionalTensor(eigenvalues)) { |
| 19281 | at::functionalization::impl::sync(eigenvalues); |
| 19282 | eigenvalues_ = at::functionalization::impl::from_functional_tensor(eigenvalues); |
| 19283 | } else { |
| 19284 | eigenvalues_ = eigenvalues; |
| 19285 | } |
| 19286 | |
| 19287 | at::Tensor eigenvectors_; |
| 19288 | if (at::functionalization::impl::isFunctionalTensor(eigenvectors)) { |
| 19289 | at::functionalization::impl::sync(eigenvectors); |
| 19290 | eigenvectors_ = at::functionalization::impl::from_functional_tensor(eigenvectors); |
| 19291 | } else { |
| 19292 | eigenvectors_ = eigenvectors; |
| 19293 | } |
| 19294 | if (!(true && at::functionalization::impl::isFunctionalTensor(eigenvalues) && at::functionalization::impl::isFunctionalTensor(eigenvectors))) { |
| 19295 | if ((false || at::functionalization::impl::isFunctionalTensor(A))) { |
| 19296 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19297 | TORCH_INTERNAL_ASSERT(false, |
| 19298 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19299 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19300 | } else { |
| 19301 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19302 | at::AutoDispatchSkipFunctionalize guard; |
| 19303 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::_linalg_eigh_eigenvalues::call(A_, UPLO, compute_v, eigenvalues_, eigenvectors_); |
| 19304 | return ::std::tuple<at::Tensor &,at::Tensor &>(eigenvalues, eigenvectors);; |
| 19305 | } |
| 19306 | } else { |
| 19307 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 19308 | { |
| 19309 | at::AutoDispatchSkipFunctionalize guard; |
| 19310 | tmp_output = at::_ops::_linalg_eigh::call(A_, UPLO, compute_v); |
| 19311 | } |
| 19312 | at::functionalization::impl::replace_(eigenvalues, std::get<0>(tmp_output)); |
| 19313 | at::functionalization::impl::commit_update(eigenvalues); |
| 19314 | at::functionalization::impl::sync(eigenvalues); |
| 19315 | at::functionalization::impl::replace_(eigenvectors, std::get<1>(tmp_output)); |
| 19316 | at::functionalization::impl::commit_update(eigenvectors); |
| 19317 | at::functionalization::impl::sync(eigenvectors); |
| 19318 | return ::std::tuple<at::Tensor &,at::Tensor &>(eigenvalues, eigenvectors); |
| 19319 | } |
| 19320 | } |
| 19321 | |
| 19322 | at::Tensor & inner_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Tensor & other, at::Tensor & out) { |
| 19323 | if (false) { |
| 19324 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19325 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19326 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19327 | auto self_meta = to_meta(self); |
| 19328 | auto other_meta = to_meta(other); |
| 19329 | auto out_meta = to_meta(out); |
| 19330 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19331 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19332 | at::_ops::inner_out::call(self_meta, other_meta, out_meta); |
| 19333 | } |
| 19334 | |
| 19335 | at::Tensor self_; |
| 19336 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19337 | at::functionalization::impl::sync(self); |
| 19338 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19339 | } else { |
| 19340 | self_ = self; |
| 19341 | } |
| 19342 | |
| 19343 | at::Tensor other_; |
| 19344 | if (at::functionalization::impl::isFunctionalTensor(other)) { |
| 19345 | at::functionalization::impl::sync(other); |
| 19346 | other_ = at::functionalization::impl::from_functional_tensor(other); |
| 19347 | } else { |
| 19348 | other_ = other; |
| 19349 | } |
| 19350 | |
| 19351 | at::Tensor out_; |
| 19352 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19353 | at::functionalization::impl::sync(out); |
| 19354 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19355 | } else { |
| 19356 | out_ = out; |
| 19357 | } |
| 19358 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19359 | if ((false || at::functionalization::impl::isFunctionalTensor(self) || at::functionalization::impl::isFunctionalTensor(other))) { |
| 19360 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19361 | TORCH_INTERNAL_ASSERT(false, |
| 19362 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19363 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19364 | } else { |
| 19365 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19366 | at::AutoDispatchSkipFunctionalize guard; |
| 19367 | at::Tensor tmp_output = at::_ops::inner_out::call(self_, other_, out_); |
| 19368 | return out;; |
| 19369 | } |
| 19370 | } else { |
| 19371 | at::Tensor tmp_output; |
| 19372 | { |
| 19373 | at::AutoDispatchSkipFunctionalize guard; |
| 19374 | tmp_output = at::_ops::inner::call(self_, other_); |
| 19375 | } |
| 19376 | at::functionalization::impl::replace_(out, tmp_output); |
| 19377 | at::functionalization::impl::commit_update(out); |
| 19378 | at::functionalization::impl::sync(out); |
| 19379 | return out; |
| 19380 | } |
| 19381 | } |
| 19382 | |
| 19383 | at::Tensor & linalg_matrix_norm_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, const at::Scalar & ord, at::IntArrayRef dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 19384 | if (false) { |
| 19385 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19386 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19387 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19388 | auto self_meta = to_meta(self); |
| 19389 | auto out_meta = to_meta(out); |
| 19390 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19391 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19392 | at::_ops::linalg_matrix_norm_out::call(self_meta, ord, dim, keepdim, dtype, out_meta); |
| 19393 | } |
| 19394 | |
| 19395 | at::Tensor self_; |
| 19396 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19397 | at::functionalization::impl::sync(self); |
| 19398 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19399 | } else { |
| 19400 | self_ = self; |
| 19401 | } |
| 19402 | |
| 19403 | at::Tensor out_; |
| 19404 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19405 | at::functionalization::impl::sync(out); |
| 19406 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19407 | } else { |
| 19408 | out_ = out; |
| 19409 | } |
| 19410 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19411 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19412 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19413 | TORCH_INTERNAL_ASSERT(false, |
| 19414 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19415 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19416 | } else { |
| 19417 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19418 | at::AutoDispatchSkipFunctionalize guard; |
| 19419 | at::Tensor tmp_output = at::_ops::linalg_matrix_norm_out::call(self_, ord, dim, keepdim, dtype, out_); |
| 19420 | return out;; |
| 19421 | } |
| 19422 | } else { |
| 19423 | at::Tensor tmp_output; |
| 19424 | { |
| 19425 | at::AutoDispatchSkipFunctionalize guard; |
| 19426 | tmp_output = at::_ops::linalg_matrix_norm::call(self_, ord, dim, keepdim, dtype); |
| 19427 | } |
| 19428 | at::functionalization::impl::replace_(out, tmp_output); |
| 19429 | at::functionalization::impl::commit_update(out); |
| 19430 | at::functionalization::impl::sync(out); |
| 19431 | return out; |
| 19432 | } |
| 19433 | } |
| 19434 | |
| 19435 | at::Tensor & linalg_matrix_norm_out_str_ord_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::string_view ord, at::IntArrayRef dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out) { |
| 19436 | if (false) { |
| 19437 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19438 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19439 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19440 | auto self_meta = to_meta(self); |
| 19441 | auto out_meta = to_meta(out); |
| 19442 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19443 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19444 | at::_ops::linalg_matrix_norm_str_ord_out::call(self_meta, ord, dim, keepdim, dtype, out_meta); |
| 19445 | } |
| 19446 | |
| 19447 | at::Tensor self_; |
| 19448 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19449 | at::functionalization::impl::sync(self); |
| 19450 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19451 | } else { |
| 19452 | self_ = self; |
| 19453 | } |
| 19454 | |
| 19455 | at::Tensor out_; |
| 19456 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19457 | at::functionalization::impl::sync(out); |
| 19458 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19459 | } else { |
| 19460 | out_ = out; |
| 19461 | } |
| 19462 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19463 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19464 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19465 | TORCH_INTERNAL_ASSERT(false, |
| 19466 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19467 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19468 | } else { |
| 19469 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19470 | at::AutoDispatchSkipFunctionalize guard; |
| 19471 | at::Tensor tmp_output = at::_ops::linalg_matrix_norm_str_ord_out::call(self_, ord, dim, keepdim, dtype, out_); |
| 19472 | return out;; |
| 19473 | } |
| 19474 | } else { |
| 19475 | at::Tensor tmp_output; |
| 19476 | { |
| 19477 | at::AutoDispatchSkipFunctionalize guard; |
| 19478 | tmp_output = at::_ops::linalg_matrix_norm_str_ord::call(self_, ord, dim, keepdim, dtype); |
| 19479 | } |
| 19480 | at::functionalization::impl::replace_(out, tmp_output); |
| 19481 | at::functionalization::impl::commit_update(out); |
| 19482 | at::functionalization::impl::sync(out); |
| 19483 | return out; |
| 19484 | } |
| 19485 | } |
| 19486 | |
| 19487 | at::Tensor & linalg_tensorinv_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t ind, at::Tensor & out) { |
| 19488 | if (false) { |
| 19489 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19490 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19491 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19492 | auto self_meta = to_meta(self); |
| 19493 | auto out_meta = to_meta(out); |
| 19494 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19495 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19496 | at::_ops::linalg_tensorinv_out::call(self_meta, ind, out_meta); |
| 19497 | } |
| 19498 | |
| 19499 | at::Tensor self_; |
| 19500 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19501 | at::functionalization::impl::sync(self); |
| 19502 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19503 | } else { |
| 19504 | self_ = self; |
| 19505 | } |
| 19506 | |
| 19507 | at::Tensor out_; |
| 19508 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19509 | at::functionalization::impl::sync(out); |
| 19510 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19511 | } else { |
| 19512 | out_ = out; |
| 19513 | } |
| 19514 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19515 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19516 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19517 | TORCH_INTERNAL_ASSERT(false, |
| 19518 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19519 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19520 | } else { |
| 19521 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19522 | at::AutoDispatchSkipFunctionalize guard; |
| 19523 | at::Tensor tmp_output = at::_ops::linalg_tensorinv_out::call(self_, ind, out_); |
| 19524 | return out;; |
| 19525 | } |
| 19526 | } else { |
| 19527 | at::Tensor tmp_output; |
| 19528 | { |
| 19529 | at::AutoDispatchSkipFunctionalize guard; |
| 19530 | tmp_output = at::_ops::linalg_tensorinv::call(self_, ind); |
| 19531 | } |
| 19532 | at::functionalization::impl::replace_(out, tmp_output); |
| 19533 | at::functionalization::impl::commit_update(out); |
| 19534 | at::functionalization::impl::sync(out); |
| 19535 | return out; |
| 19536 | } |
| 19537 | } |
| 19538 | |
| 19539 | at::Tensor & linalg_matrix_power_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t n, at::Tensor & out) { |
| 19540 | if (false) { |
| 19541 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19542 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19543 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19544 | auto self_meta = to_meta(self); |
| 19545 | auto out_meta = to_meta(out); |
| 19546 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19547 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19548 | at::_ops::linalg_matrix_power_out::call(self_meta, n, out_meta); |
| 19549 | } |
| 19550 | |
| 19551 | at::Tensor self_; |
| 19552 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19553 | at::functionalization::impl::sync(self); |
| 19554 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19555 | } else { |
| 19556 | self_ = self; |
| 19557 | } |
| 19558 | |
| 19559 | at::Tensor out_; |
| 19560 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19561 | at::functionalization::impl::sync(out); |
| 19562 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19563 | } else { |
| 19564 | out_ = out; |
| 19565 | } |
| 19566 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19567 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19568 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19569 | TORCH_INTERNAL_ASSERT(false, |
| 19570 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19571 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19572 | } else { |
| 19573 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19574 | at::AutoDispatchSkipFunctionalize guard; |
| 19575 | at::Tensor tmp_output = at::_ops::linalg_matrix_power_out::call(self_, n, out_); |
| 19576 | return out;; |
| 19577 | } |
| 19578 | } else { |
| 19579 | at::Tensor tmp_output; |
| 19580 | { |
| 19581 | at::AutoDispatchSkipFunctionalize guard; |
| 19582 | tmp_output = at::_ops::linalg_matrix_power::call(self_, n); |
| 19583 | } |
| 19584 | at::functionalization::impl::replace_(out, tmp_output); |
| 19585 | at::functionalization::impl::commit_update(out); |
| 19586 | at::functionalization::impl::sync(out); |
| 19587 | return out; |
| 19588 | } |
| 19589 | } |
| 19590 | |
| 19591 | at::Tensor & _make_dual_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & primal, const at::Tensor & tangent, int64_t level, at::Tensor & out) { |
| 19592 | if (false) { |
| 19593 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19594 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19595 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19596 | auto primal_meta = to_meta(primal); |
| 19597 | auto tangent_meta = to_meta(tangent); |
| 19598 | auto out_meta = to_meta(out); |
| 19599 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19600 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19601 | at::_ops::_make_dual_copy_out::call(primal_meta, tangent_meta, level, out_meta); |
| 19602 | } |
| 19603 | |
| 19604 | at::Tensor primal_; |
| 19605 | if (at::functionalization::impl::isFunctionalTensor(primal)) { |
| 19606 | at::functionalization::impl::sync(primal); |
| 19607 | primal_ = at::functionalization::impl::from_functional_tensor(primal); |
| 19608 | } else { |
| 19609 | primal_ = primal; |
| 19610 | } |
| 19611 | |
| 19612 | at::Tensor tangent_; |
| 19613 | if (at::functionalization::impl::isFunctionalTensor(tangent)) { |
| 19614 | at::functionalization::impl::sync(tangent); |
| 19615 | tangent_ = at::functionalization::impl::from_functional_tensor(tangent); |
| 19616 | } else { |
| 19617 | tangent_ = tangent; |
| 19618 | } |
| 19619 | |
| 19620 | at::Tensor out_; |
| 19621 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19622 | at::functionalization::impl::sync(out); |
| 19623 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19624 | } else { |
| 19625 | out_ = out; |
| 19626 | } |
| 19627 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19628 | if ((false || at::functionalization::impl::isFunctionalTensor(primal) || at::functionalization::impl::isFunctionalTensor(tangent))) { |
| 19629 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19630 | TORCH_INTERNAL_ASSERT(false, |
| 19631 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19632 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19633 | } else { |
| 19634 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19635 | at::AutoDispatchSkipFunctionalize guard; |
| 19636 | at::Tensor tmp_output = at::_ops::_make_dual_copy_out::call(primal_, tangent_, level, out_); |
| 19637 | return out;; |
| 19638 | } |
| 19639 | } else { |
| 19640 | at::Tensor tmp_output; |
| 19641 | { |
| 19642 | at::AutoDispatchSkipFunctionalize guard; |
| 19643 | tmp_output = at::_ops::_make_dual_copy::call(primal_, tangent_, level); |
| 19644 | } |
| 19645 | at::functionalization::impl::replace_(out, tmp_output); |
| 19646 | at::functionalization::impl::commit_update(out); |
| 19647 | at::functionalization::impl::sync(out); |
| 19648 | return out; |
| 19649 | } |
| 19650 | } |
| 19651 | |
| 19652 | at::Tensor & view_as_real_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 19653 | if (false) { |
| 19654 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19655 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19656 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19657 | auto self_meta = to_meta(self); |
| 19658 | auto out_meta = to_meta(out); |
| 19659 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19660 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19661 | at::_ops::view_as_real_copy_out::call(self_meta, out_meta); |
| 19662 | } |
| 19663 | |
| 19664 | at::Tensor self_; |
| 19665 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19666 | at::functionalization::impl::sync(self); |
| 19667 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19668 | } else { |
| 19669 | self_ = self; |
| 19670 | } |
| 19671 | |
| 19672 | at::Tensor out_; |
| 19673 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19674 | at::functionalization::impl::sync(out); |
| 19675 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19676 | } else { |
| 19677 | out_ = out; |
| 19678 | } |
| 19679 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19680 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19681 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19682 | TORCH_INTERNAL_ASSERT(false, |
| 19683 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19684 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19685 | } else { |
| 19686 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19687 | at::AutoDispatchSkipFunctionalize guard; |
| 19688 | at::Tensor tmp_output = at::_ops::view_as_real_copy_out::call(self_, out_); |
| 19689 | return out;; |
| 19690 | } |
| 19691 | } else { |
| 19692 | at::Tensor tmp_output; |
| 19693 | { |
| 19694 | at::AutoDispatchSkipFunctionalize guard; |
| 19695 | tmp_output = at::_ops::view_as_real_copy::call(self_); |
| 19696 | } |
| 19697 | at::functionalization::impl::replace_(out, tmp_output); |
| 19698 | at::functionalization::impl::commit_update(out); |
| 19699 | at::functionalization::impl::sync(out); |
| 19700 | return out; |
| 19701 | } |
| 19702 | } |
| 19703 | |
| 19704 | at::Tensor & view_as_complex_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 19705 | if (false) { |
| 19706 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19707 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19708 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19709 | auto self_meta = to_meta(self); |
| 19710 | auto out_meta = to_meta(out); |
| 19711 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19712 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19713 | at::_ops::view_as_complex_copy_out::call(self_meta, out_meta); |
| 19714 | } |
| 19715 | |
| 19716 | at::Tensor self_; |
| 19717 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19718 | at::functionalization::impl::sync(self); |
| 19719 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19720 | } else { |
| 19721 | self_ = self; |
| 19722 | } |
| 19723 | |
| 19724 | at::Tensor out_; |
| 19725 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19726 | at::functionalization::impl::sync(out); |
| 19727 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19728 | } else { |
| 19729 | out_ = out; |
| 19730 | } |
| 19731 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19732 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19733 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19734 | TORCH_INTERNAL_ASSERT(false, |
| 19735 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19736 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19737 | } else { |
| 19738 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19739 | at::AutoDispatchSkipFunctionalize guard; |
| 19740 | at::Tensor tmp_output = at::_ops::view_as_complex_copy_out::call(self_, out_); |
| 19741 | return out;; |
| 19742 | } |
| 19743 | } else { |
| 19744 | at::Tensor tmp_output; |
| 19745 | { |
| 19746 | at::AutoDispatchSkipFunctionalize guard; |
| 19747 | tmp_output = at::_ops::view_as_complex_copy::call(self_); |
| 19748 | } |
| 19749 | at::functionalization::impl::replace_(out, tmp_output); |
| 19750 | at::functionalization::impl::commit_update(out); |
| 19751 | at::functionalization::impl::sync(out); |
| 19752 | return out; |
| 19753 | } |
| 19754 | } |
| 19755 | |
| 19756 | at::Tensor & _conj_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 19757 | if (false) { |
| 19758 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19759 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19760 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19761 | auto self_meta = to_meta(self); |
| 19762 | auto out_meta = to_meta(out); |
| 19763 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19764 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19765 | at::_ops::_conj_copy_out::call(self_meta, out_meta); |
| 19766 | } |
| 19767 | |
| 19768 | at::Tensor self_; |
| 19769 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19770 | at::functionalization::impl::sync(self); |
| 19771 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19772 | } else { |
| 19773 | self_ = self; |
| 19774 | } |
| 19775 | |
| 19776 | at::Tensor out_; |
| 19777 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19778 | at::functionalization::impl::sync(out); |
| 19779 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19780 | } else { |
| 19781 | out_ = out; |
| 19782 | } |
| 19783 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19784 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19785 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19786 | TORCH_INTERNAL_ASSERT(false, |
| 19787 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19788 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19789 | } else { |
| 19790 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19791 | at::AutoDispatchSkipFunctionalize guard; |
| 19792 | at::Tensor tmp_output = at::_ops::_conj_copy_out::call(self_, out_); |
| 19793 | return out;; |
| 19794 | } |
| 19795 | } else { |
| 19796 | at::Tensor tmp_output; |
| 19797 | { |
| 19798 | at::AutoDispatchSkipFunctionalize guard; |
| 19799 | tmp_output = at::_ops::_conj_copy::call(self_); |
| 19800 | } |
| 19801 | at::functionalization::impl::replace_(out, tmp_output); |
| 19802 | at::functionalization::impl::commit_update(out); |
| 19803 | at::functionalization::impl::sync(out); |
| 19804 | return out; |
| 19805 | } |
| 19806 | } |
| 19807 | |
| 19808 | at::Tensor & _neg_view_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 19809 | if (false) { |
| 19810 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19811 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19812 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19813 | auto self_meta = to_meta(self); |
| 19814 | auto out_meta = to_meta(out); |
| 19815 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19816 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19817 | at::_ops::_neg_view_copy_out::call(self_meta, out_meta); |
| 19818 | } |
| 19819 | |
| 19820 | at::Tensor self_; |
| 19821 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19822 | at::functionalization::impl::sync(self); |
| 19823 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19824 | } else { |
| 19825 | self_ = self; |
| 19826 | } |
| 19827 | |
| 19828 | at::Tensor out_; |
| 19829 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19830 | at::functionalization::impl::sync(out); |
| 19831 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19832 | } else { |
| 19833 | out_ = out; |
| 19834 | } |
| 19835 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19836 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19837 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19838 | TORCH_INTERNAL_ASSERT(false, |
| 19839 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19840 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19841 | } else { |
| 19842 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19843 | at::AutoDispatchSkipFunctionalize guard; |
| 19844 | at::Tensor tmp_output = at::_ops::_neg_view_copy_out::call(self_, out_); |
| 19845 | return out;; |
| 19846 | } |
| 19847 | } else { |
| 19848 | at::Tensor tmp_output; |
| 19849 | { |
| 19850 | at::AutoDispatchSkipFunctionalize guard; |
| 19851 | tmp_output = at::_ops::_neg_view_copy::call(self_); |
| 19852 | } |
| 19853 | at::functionalization::impl::replace_(out, tmp_output); |
| 19854 | at::functionalization::impl::commit_update(out); |
| 19855 | at::functionalization::impl::sync(out); |
| 19856 | return out; |
| 19857 | } |
| 19858 | } |
| 19859 | |
| 19860 | at::Tensor & permute_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::IntArrayRef dims, at::Tensor & out) { |
| 19861 | if (false) { |
| 19862 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19863 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19864 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19865 | auto self_meta = to_meta(self); |
| 19866 | auto out_meta = to_meta(out); |
| 19867 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19868 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19869 | at::_ops::permute_copy_out::call(self_meta, dims, out_meta); |
| 19870 | } |
| 19871 | |
| 19872 | at::Tensor self_; |
| 19873 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19874 | at::functionalization::impl::sync(self); |
| 19875 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19876 | } else { |
| 19877 | self_ = self; |
| 19878 | } |
| 19879 | |
| 19880 | at::Tensor out_; |
| 19881 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19882 | at::functionalization::impl::sync(out); |
| 19883 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19884 | } else { |
| 19885 | out_ = out; |
| 19886 | } |
| 19887 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19888 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19889 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19890 | TORCH_INTERNAL_ASSERT(false, |
| 19891 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19892 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19893 | } else { |
| 19894 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19895 | at::AutoDispatchSkipFunctionalize guard; |
| 19896 | at::Tensor tmp_output = at::_ops::permute_copy_out::call(self_, dims, out_); |
| 19897 | return out;; |
| 19898 | } |
| 19899 | } else { |
| 19900 | at::Tensor tmp_output; |
| 19901 | { |
| 19902 | at::AutoDispatchSkipFunctionalize guard; |
| 19903 | tmp_output = at::_ops::permute_copy::call(self_, dims); |
| 19904 | } |
| 19905 | at::functionalization::impl::replace_(out, tmp_output); |
| 19906 | at::functionalization::impl::commit_update(out); |
| 19907 | at::functionalization::impl::sync(out); |
| 19908 | return out; |
| 19909 | } |
| 19910 | } |
| 19911 | |
| 19912 | void split_copy_out_Tensor_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, c10::SymInt split_size, int64_t dim, at::TensorList out) { |
| 19913 | if (false) { |
| 19914 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19915 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19916 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19917 | auto self_meta = to_meta(self); |
| 19918 | auto out_meta = to_meta(out); |
| 19919 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19920 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19921 | at::_ops::split_copy_Tensor_out::call(self_meta, split_size, dim, out_meta); |
| 19922 | } |
| 19923 | |
| 19924 | at::Tensor self_; |
| 19925 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19926 | at::functionalization::impl::sync(self); |
| 19927 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19928 | } else { |
| 19929 | self_ = self; |
| 19930 | } |
| 19931 | |
| 19932 | ::std::vector<at::Tensor> out_; |
| 19933 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19934 | at::functionalization::impl::sync(out); |
| 19935 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19936 | } else { |
| 19937 | out_ = out.vec(); |
| 19938 | } |
| 19939 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19940 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19941 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19942 | TORCH_INTERNAL_ASSERT(false, |
| 19943 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19944 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19945 | } else { |
| 19946 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19947 | at::AutoDispatchSkipFunctionalize guard; |
| 19948 | at::_ops::split_copy_Tensor_out::call(self_, split_size, dim, out_); |
| 19949 | ; |
| 19950 | } |
| 19951 | } else { |
| 19952 | ::std::vector<at::Tensor> tmp_output; |
| 19953 | { |
| 19954 | at::AutoDispatchSkipFunctionalize guard; |
| 19955 | tmp_output = at::_ops::split_copy_Tensor::call(self_, split_size, dim); |
| 19956 | } |
| 19957 | at::functionalization::impl::replace_(out, tmp_output); |
| 19958 | at::functionalization::impl::commit_update(out); |
| 19959 | at::functionalization::impl::sync(out); |
| 19960 | |
| 19961 | } |
| 19962 | } |
| 19963 | |
| 19964 | at::Tensor & t_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 19965 | if (false) { |
| 19966 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 19967 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 19968 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 19969 | auto self_meta = to_meta(self); |
| 19970 | auto out_meta = to_meta(out); |
| 19971 | at::AutoDispatchSkipFunctionalize func_guard; |
| 19972 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 19973 | at::_ops::t_copy_out::call(self_meta, out_meta); |
| 19974 | } |
| 19975 | |
| 19976 | at::Tensor self_; |
| 19977 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 19978 | at::functionalization::impl::sync(self); |
| 19979 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 19980 | } else { |
| 19981 | self_ = self; |
| 19982 | } |
| 19983 | |
| 19984 | at::Tensor out_; |
| 19985 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 19986 | at::functionalization::impl::sync(out); |
| 19987 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 19988 | } else { |
| 19989 | out_ = out; |
| 19990 | } |
| 19991 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 19992 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 19993 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 19994 | TORCH_INTERNAL_ASSERT(false, |
| 19995 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 19996 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 19997 | } else { |
| 19998 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 19999 | at::AutoDispatchSkipFunctionalize guard; |
| 20000 | at::Tensor tmp_output = at::_ops::t_copy_out::call(self_, out_); |
| 20001 | return out;; |
| 20002 | } |
| 20003 | } else { |
| 20004 | at::Tensor tmp_output; |
| 20005 | { |
| 20006 | at::AutoDispatchSkipFunctionalize guard; |
| 20007 | tmp_output = at::_ops::t_copy::call(self_); |
| 20008 | } |
| 20009 | at::functionalization::impl::replace_(out, tmp_output); |
| 20010 | at::functionalization::impl::commit_update(out); |
| 20011 | at::functionalization::impl::sync(out); |
| 20012 | return out; |
| 20013 | } |
| 20014 | } |
| 20015 | |
| 20016 | at::Tensor & ccol_indices_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 20017 | if (false) { |
| 20018 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20019 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20020 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20021 | auto self_meta = to_meta(self); |
| 20022 | auto out_meta = to_meta(out); |
| 20023 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20024 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20025 | at::_ops::ccol_indices_copy_out::call(self_meta, out_meta); |
| 20026 | } |
| 20027 | |
| 20028 | at::Tensor self_; |
| 20029 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 20030 | at::functionalization::impl::sync(self); |
| 20031 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 20032 | } else { |
| 20033 | self_ = self; |
| 20034 | } |
| 20035 | |
| 20036 | at::Tensor out_; |
| 20037 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 20038 | at::functionalization::impl::sync(out); |
| 20039 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 20040 | } else { |
| 20041 | out_ = out; |
| 20042 | } |
| 20043 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 20044 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 20045 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20046 | TORCH_INTERNAL_ASSERT(false, |
| 20047 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20048 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20049 | } else { |
| 20050 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20051 | at::AutoDispatchSkipFunctionalize guard; |
| 20052 | at::Tensor tmp_output = at::_ops::ccol_indices_copy_out::call(self_, out_); |
| 20053 | return out;; |
| 20054 | } |
| 20055 | } else { |
| 20056 | at::Tensor tmp_output; |
| 20057 | { |
| 20058 | at::AutoDispatchSkipFunctionalize guard; |
| 20059 | tmp_output = at::_ops::ccol_indices_copy::call(self_); |
| 20060 | } |
| 20061 | at::functionalization::impl::replace_(out, tmp_output); |
| 20062 | at::functionalization::impl::commit_update(out); |
| 20063 | at::functionalization::impl::sync(out); |
| 20064 | return out; |
| 20065 | } |
| 20066 | } |
| 20067 | |
| 20068 | at::Tensor & unfold_copy_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dimension, int64_t size, int64_t step, at::Tensor & out) { |
| 20069 | if (false) { |
| 20070 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20071 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20072 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20073 | auto self_meta = to_meta(self); |
| 20074 | auto out_meta = to_meta(out); |
| 20075 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20076 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20077 | at::_ops::unfold_copy_out::call(self_meta, dimension, size, step, out_meta); |
| 20078 | } |
| 20079 | |
| 20080 | at::Tensor self_; |
| 20081 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 20082 | at::functionalization::impl::sync(self); |
| 20083 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 20084 | } else { |
| 20085 | self_ = self; |
| 20086 | } |
| 20087 | |
| 20088 | at::Tensor out_; |
| 20089 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 20090 | at::functionalization::impl::sync(out); |
| 20091 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 20092 | } else { |
| 20093 | out_ = out; |
| 20094 | } |
| 20095 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 20096 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 20097 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20098 | TORCH_INTERNAL_ASSERT(false, |
| 20099 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20100 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20101 | } else { |
| 20102 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20103 | at::AutoDispatchSkipFunctionalize guard; |
| 20104 | at::Tensor tmp_output = at::_ops::unfold_copy_out::call(self_, dimension, size, step, out_); |
| 20105 | return out;; |
| 20106 | } |
| 20107 | } else { |
| 20108 | at::Tensor tmp_output; |
| 20109 | { |
| 20110 | at::AutoDispatchSkipFunctionalize guard; |
| 20111 | tmp_output = at::_ops::unfold_copy::call(self_, dimension, size, step); |
| 20112 | } |
| 20113 | at::functionalization::impl::replace_(out, tmp_output); |
| 20114 | at::functionalization::impl::commit_update(out); |
| 20115 | at::functionalization::impl::sync(out); |
| 20116 | return out; |
| 20117 | } |
| 20118 | } |
| 20119 | |
| 20120 | at::Tensor & _transformer_encoder_layer_fwd_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & src, int64_t embed_dim, int64_t num_heads, const at::Tensor & qkv_weight, const at::Tensor & qkv_bias, const at::Tensor & proj_weight, const at::Tensor & proj_bias, bool use_gelu, bool norm_first, double eps, const at::Tensor & norm_weight_1, const at::Tensor & norm_bias_1, const at::Tensor & norm_weight_2, const at::Tensor & norm_bias_2, const at::Tensor & ffn_weight_1, const at::Tensor & ffn_bias_1, const at::Tensor & ffn_weight_2, const at::Tensor & ffn_bias_2, const c10::optional<at::Tensor> & mask, c10::optional<int64_t> mask_type, at::Tensor & out) { |
| 20121 | if (false) { |
| 20122 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20123 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20124 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20125 | auto src_meta = to_meta(src); |
| 20126 | auto qkv_weight_meta = to_meta(qkv_weight); |
| 20127 | auto qkv_bias_meta = to_meta(qkv_bias); |
| 20128 | auto proj_weight_meta = to_meta(proj_weight); |
| 20129 | auto proj_bias_meta = to_meta(proj_bias); |
| 20130 | auto norm_weight_1_meta = to_meta(norm_weight_1); |
| 20131 | auto norm_bias_1_meta = to_meta(norm_bias_1); |
| 20132 | auto norm_weight_2_meta = to_meta(norm_weight_2); |
| 20133 | auto norm_bias_2_meta = to_meta(norm_bias_2); |
| 20134 | auto ffn_weight_1_meta = to_meta(ffn_weight_1); |
| 20135 | auto ffn_bias_1_meta = to_meta(ffn_bias_1); |
| 20136 | auto ffn_weight_2_meta = to_meta(ffn_weight_2); |
| 20137 | auto ffn_bias_2_meta = to_meta(ffn_bias_2); |
| 20138 | auto mask_meta = to_meta(mask); |
| 20139 | auto out_meta = to_meta(out); |
| 20140 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20141 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20142 | at::_ops::_transformer_encoder_layer_fwd_out::call(src_meta, embed_dim, num_heads, qkv_weight_meta, qkv_bias_meta, proj_weight_meta, proj_bias_meta, use_gelu, norm_first, eps, norm_weight_1_meta, norm_bias_1_meta, norm_weight_2_meta, norm_bias_2_meta, ffn_weight_1_meta, ffn_bias_1_meta, ffn_weight_2_meta, ffn_bias_2_meta, mask_meta, mask_type, out_meta); |
| 20143 | } |
| 20144 | |
| 20145 | at::Tensor src_; |
| 20146 | if (at::functionalization::impl::isFunctionalTensor(src)) { |
| 20147 | at::functionalization::impl::sync(src); |
| 20148 | src_ = at::functionalization::impl::from_functional_tensor(src); |
| 20149 | } else { |
| 20150 | src_ = src; |
| 20151 | } |
| 20152 | |
| 20153 | at::Tensor qkv_weight_; |
| 20154 | if (at::functionalization::impl::isFunctionalTensor(qkv_weight)) { |
| 20155 | at::functionalization::impl::sync(qkv_weight); |
| 20156 | qkv_weight_ = at::functionalization::impl::from_functional_tensor(qkv_weight); |
| 20157 | } else { |
| 20158 | qkv_weight_ = qkv_weight; |
| 20159 | } |
| 20160 | |
| 20161 | at::Tensor qkv_bias_; |
| 20162 | if (at::functionalization::impl::isFunctionalTensor(qkv_bias)) { |
| 20163 | at::functionalization::impl::sync(qkv_bias); |
| 20164 | qkv_bias_ = at::functionalization::impl::from_functional_tensor(qkv_bias); |
| 20165 | } else { |
| 20166 | qkv_bias_ = qkv_bias; |
| 20167 | } |
| 20168 | |
| 20169 | at::Tensor proj_weight_; |
| 20170 | if (at::functionalization::impl::isFunctionalTensor(proj_weight)) { |
| 20171 | at::functionalization::impl::sync(proj_weight); |
| 20172 | proj_weight_ = at::functionalization::impl::from_functional_tensor(proj_weight); |
| 20173 | } else { |
| 20174 | proj_weight_ = proj_weight; |
| 20175 | } |
| 20176 | |
| 20177 | at::Tensor proj_bias_; |
| 20178 | if (at::functionalization::impl::isFunctionalTensor(proj_bias)) { |
| 20179 | at::functionalization::impl::sync(proj_bias); |
| 20180 | proj_bias_ = at::functionalization::impl::from_functional_tensor(proj_bias); |
| 20181 | } else { |
| 20182 | proj_bias_ = proj_bias; |
| 20183 | } |
| 20184 | |
| 20185 | at::Tensor norm_weight_1_; |
| 20186 | if (at::functionalization::impl::isFunctionalTensor(norm_weight_1)) { |
| 20187 | at::functionalization::impl::sync(norm_weight_1); |
| 20188 | norm_weight_1_ = at::functionalization::impl::from_functional_tensor(norm_weight_1); |
| 20189 | } else { |
| 20190 | norm_weight_1_ = norm_weight_1; |
| 20191 | } |
| 20192 | |
| 20193 | at::Tensor norm_bias_1_; |
| 20194 | if (at::functionalization::impl::isFunctionalTensor(norm_bias_1)) { |
| 20195 | at::functionalization::impl::sync(norm_bias_1); |
| 20196 | norm_bias_1_ = at::functionalization::impl::from_functional_tensor(norm_bias_1); |
| 20197 | } else { |
| 20198 | norm_bias_1_ = norm_bias_1; |
| 20199 | } |
| 20200 | |
| 20201 | at::Tensor norm_weight_2_; |
| 20202 | if (at::functionalization::impl::isFunctionalTensor(norm_weight_2)) { |
| 20203 | at::functionalization::impl::sync(norm_weight_2); |
| 20204 | norm_weight_2_ = at::functionalization::impl::from_functional_tensor(norm_weight_2); |
| 20205 | } else { |
| 20206 | norm_weight_2_ = norm_weight_2; |
| 20207 | } |
| 20208 | |
| 20209 | at::Tensor norm_bias_2_; |
| 20210 | if (at::functionalization::impl::isFunctionalTensor(norm_bias_2)) { |
| 20211 | at::functionalization::impl::sync(norm_bias_2); |
| 20212 | norm_bias_2_ = at::functionalization::impl::from_functional_tensor(norm_bias_2); |
| 20213 | } else { |
| 20214 | norm_bias_2_ = norm_bias_2; |
| 20215 | } |
| 20216 | |
| 20217 | at::Tensor ffn_weight_1_; |
| 20218 | if (at::functionalization::impl::isFunctionalTensor(ffn_weight_1)) { |
| 20219 | at::functionalization::impl::sync(ffn_weight_1); |
| 20220 | ffn_weight_1_ = at::functionalization::impl::from_functional_tensor(ffn_weight_1); |
| 20221 | } else { |
| 20222 | ffn_weight_1_ = ffn_weight_1; |
| 20223 | } |
| 20224 | |
| 20225 | at::Tensor ffn_bias_1_; |
| 20226 | if (at::functionalization::impl::isFunctionalTensor(ffn_bias_1)) { |
| 20227 | at::functionalization::impl::sync(ffn_bias_1); |
| 20228 | ffn_bias_1_ = at::functionalization::impl::from_functional_tensor(ffn_bias_1); |
| 20229 | } else { |
| 20230 | ffn_bias_1_ = ffn_bias_1; |
| 20231 | } |
| 20232 | |
| 20233 | at::Tensor ffn_weight_2_; |
| 20234 | if (at::functionalization::impl::isFunctionalTensor(ffn_weight_2)) { |
| 20235 | at::functionalization::impl::sync(ffn_weight_2); |
| 20236 | ffn_weight_2_ = at::functionalization::impl::from_functional_tensor(ffn_weight_2); |
| 20237 | } else { |
| 20238 | ffn_weight_2_ = ffn_weight_2; |
| 20239 | } |
| 20240 | |
| 20241 | at::Tensor ffn_bias_2_; |
| 20242 | if (at::functionalization::impl::isFunctionalTensor(ffn_bias_2)) { |
| 20243 | at::functionalization::impl::sync(ffn_bias_2); |
| 20244 | ffn_bias_2_ = at::functionalization::impl::from_functional_tensor(ffn_bias_2); |
| 20245 | } else { |
| 20246 | ffn_bias_2_ = ffn_bias_2; |
| 20247 | } |
| 20248 | |
| 20249 | c10::optional<at::Tensor> mask_; |
| 20250 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 20251 | at::functionalization::impl::sync(mask); |
| 20252 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 20253 | } else { |
| 20254 | mask_ = mask; |
| 20255 | } |
| 20256 | |
| 20257 | at::Tensor out_; |
| 20258 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 20259 | at::functionalization::impl::sync(out); |
| 20260 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 20261 | } else { |
| 20262 | out_ = out; |
| 20263 | } |
| 20264 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 20265 | if ((false || at::functionalization::impl::isFunctionalTensor(src) || at::functionalization::impl::isFunctionalTensor(qkv_weight) || at::functionalization::impl::isFunctionalTensor(qkv_bias) || at::functionalization::impl::isFunctionalTensor(proj_weight) || at::functionalization::impl::isFunctionalTensor(proj_bias) || at::functionalization::impl::isFunctionalTensor(norm_weight_1) || at::functionalization::impl::isFunctionalTensor(norm_bias_1) || at::functionalization::impl::isFunctionalTensor(norm_weight_2) || at::functionalization::impl::isFunctionalTensor(norm_bias_2) || at::functionalization::impl::isFunctionalTensor(ffn_weight_1) || at::functionalization::impl::isFunctionalTensor(ffn_bias_1) || at::functionalization::impl::isFunctionalTensor(ffn_weight_2) || at::functionalization::impl::isFunctionalTensor(ffn_bias_2) || at::functionalization::impl::isFunctionalTensor(mask))) { |
| 20266 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20267 | TORCH_INTERNAL_ASSERT(false, |
| 20268 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20269 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20270 | } else { |
| 20271 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20272 | at::AutoDispatchSkipFunctionalize guard; |
| 20273 | at::Tensor tmp_output = at::_ops::_transformer_encoder_layer_fwd_out::call(src_, embed_dim, num_heads, qkv_weight_, qkv_bias_, proj_weight_, proj_bias_, use_gelu, norm_first, eps, norm_weight_1_, norm_bias_1_, norm_weight_2_, norm_bias_2_, ffn_weight_1_, ffn_bias_1_, ffn_weight_2_, ffn_bias_2_, mask_, mask_type, out_); |
| 20274 | return out;; |
| 20275 | } |
| 20276 | } else { |
| 20277 | at::Tensor tmp_output; |
| 20278 | { |
| 20279 | at::AutoDispatchSkipFunctionalize guard; |
| 20280 | tmp_output = at::_ops::_transformer_encoder_layer_fwd::call(src_, embed_dim, num_heads, qkv_weight_, qkv_bias_, proj_weight_, proj_bias_, use_gelu, norm_first, eps, norm_weight_1_, norm_bias_1_, norm_weight_2_, norm_bias_2_, ffn_weight_1_, ffn_bias_1_, ffn_weight_2_, ffn_bias_2_, mask_, mask_type); |
| 20281 | } |
| 20282 | at::functionalization::impl::replace_(out, tmp_output); |
| 20283 | at::functionalization::impl::commit_update(out); |
| 20284 | at::functionalization::impl::sync(out); |
| 20285 | return out; |
| 20286 | } |
| 20287 | } |
| 20288 | |
| 20289 | ::std::tuple<at::Tensor &,at::Tensor &> _native_multi_head_attention_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, int64_t embed_dim, int64_t num_head, const at::Tensor & qkv_weight, const at::Tensor & qkv_bias, const at::Tensor & proj_weight, const at::Tensor & proj_bias, const c10::optional<at::Tensor> & mask, bool need_weights, bool average_attn_weights, c10::optional<int64_t> mask_type, at::Tensor & out0, at::Tensor & out1) { |
| 20290 | if (false) { |
| 20291 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20292 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20293 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20294 | auto query_meta = to_meta(query); |
| 20295 | auto key_meta = to_meta(key); |
| 20296 | auto value_meta = to_meta(value); |
| 20297 | auto qkv_weight_meta = to_meta(qkv_weight); |
| 20298 | auto qkv_bias_meta = to_meta(qkv_bias); |
| 20299 | auto proj_weight_meta = to_meta(proj_weight); |
| 20300 | auto proj_bias_meta = to_meta(proj_bias); |
| 20301 | auto mask_meta = to_meta(mask); |
| 20302 | auto out0_meta = to_meta(out0); |
| 20303 | auto out1_meta = to_meta(out1); |
| 20304 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20305 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20306 | at::_ops::_native_multi_head_attention_out::call(query_meta, key_meta, value_meta, embed_dim, num_head, qkv_weight_meta, qkv_bias_meta, proj_weight_meta, proj_bias_meta, mask_meta, need_weights, average_attn_weights, mask_type, out0_meta, out1_meta); |
| 20307 | } |
| 20308 | |
| 20309 | at::Tensor query_; |
| 20310 | if (at::functionalization::impl::isFunctionalTensor(query)) { |
| 20311 | at::functionalization::impl::sync(query); |
| 20312 | query_ = at::functionalization::impl::from_functional_tensor(query); |
| 20313 | } else { |
| 20314 | query_ = query; |
| 20315 | } |
| 20316 | |
| 20317 | at::Tensor key_; |
| 20318 | if (at::functionalization::impl::isFunctionalTensor(key)) { |
| 20319 | at::functionalization::impl::sync(key); |
| 20320 | key_ = at::functionalization::impl::from_functional_tensor(key); |
| 20321 | } else { |
| 20322 | key_ = key; |
| 20323 | } |
| 20324 | |
| 20325 | at::Tensor value_; |
| 20326 | if (at::functionalization::impl::isFunctionalTensor(value)) { |
| 20327 | at::functionalization::impl::sync(value); |
| 20328 | value_ = at::functionalization::impl::from_functional_tensor(value); |
| 20329 | } else { |
| 20330 | value_ = value; |
| 20331 | } |
| 20332 | |
| 20333 | at::Tensor qkv_weight_; |
| 20334 | if (at::functionalization::impl::isFunctionalTensor(qkv_weight)) { |
| 20335 | at::functionalization::impl::sync(qkv_weight); |
| 20336 | qkv_weight_ = at::functionalization::impl::from_functional_tensor(qkv_weight); |
| 20337 | } else { |
| 20338 | qkv_weight_ = qkv_weight; |
| 20339 | } |
| 20340 | |
| 20341 | at::Tensor qkv_bias_; |
| 20342 | if (at::functionalization::impl::isFunctionalTensor(qkv_bias)) { |
| 20343 | at::functionalization::impl::sync(qkv_bias); |
| 20344 | qkv_bias_ = at::functionalization::impl::from_functional_tensor(qkv_bias); |
| 20345 | } else { |
| 20346 | qkv_bias_ = qkv_bias; |
| 20347 | } |
| 20348 | |
| 20349 | at::Tensor proj_weight_; |
| 20350 | if (at::functionalization::impl::isFunctionalTensor(proj_weight)) { |
| 20351 | at::functionalization::impl::sync(proj_weight); |
| 20352 | proj_weight_ = at::functionalization::impl::from_functional_tensor(proj_weight); |
| 20353 | } else { |
| 20354 | proj_weight_ = proj_weight; |
| 20355 | } |
| 20356 | |
| 20357 | at::Tensor proj_bias_; |
| 20358 | if (at::functionalization::impl::isFunctionalTensor(proj_bias)) { |
| 20359 | at::functionalization::impl::sync(proj_bias); |
| 20360 | proj_bias_ = at::functionalization::impl::from_functional_tensor(proj_bias); |
| 20361 | } else { |
| 20362 | proj_bias_ = proj_bias; |
| 20363 | } |
| 20364 | |
| 20365 | c10::optional<at::Tensor> mask_; |
| 20366 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 20367 | at::functionalization::impl::sync(mask); |
| 20368 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 20369 | } else { |
| 20370 | mask_ = mask; |
| 20371 | } |
| 20372 | |
| 20373 | at::Tensor out0_; |
| 20374 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 20375 | at::functionalization::impl::sync(out0); |
| 20376 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 20377 | } else { |
| 20378 | out0_ = out0; |
| 20379 | } |
| 20380 | |
| 20381 | at::Tensor out1_; |
| 20382 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 20383 | at::functionalization::impl::sync(out1); |
| 20384 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 20385 | } else { |
| 20386 | out1_ = out1; |
| 20387 | } |
| 20388 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1))) { |
| 20389 | if ((false || at::functionalization::impl::isFunctionalTensor(query) || at::functionalization::impl::isFunctionalTensor(key) || at::functionalization::impl::isFunctionalTensor(value) || at::functionalization::impl::isFunctionalTensor(qkv_weight) || at::functionalization::impl::isFunctionalTensor(qkv_bias) || at::functionalization::impl::isFunctionalTensor(proj_weight) || at::functionalization::impl::isFunctionalTensor(proj_bias) || at::functionalization::impl::isFunctionalTensor(mask))) { |
| 20390 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20391 | TORCH_INTERNAL_ASSERT(false, |
| 20392 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20393 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20394 | } else { |
| 20395 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20396 | at::AutoDispatchSkipFunctionalize guard; |
| 20397 | ::std::tuple<at::Tensor,at::Tensor> tmp_output = at::_ops::_native_multi_head_attention_out::call(query_, key_, value_, embed_dim, num_head, qkv_weight_, qkv_bias_, proj_weight_, proj_bias_, mask_, need_weights, average_attn_weights, mask_type, out0_, out1_); |
| 20398 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1);; |
| 20399 | } |
| 20400 | } else { |
| 20401 | ::std::tuple<at::Tensor,at::Tensor> tmp_output; |
| 20402 | { |
| 20403 | at::AutoDispatchSkipFunctionalize guard; |
| 20404 | tmp_output = at::_ops::_native_multi_head_attention::call(query_, key_, value_, embed_dim, num_head, qkv_weight_, qkv_bias_, proj_weight_, proj_bias_, mask_, need_weights, average_attn_weights, mask_type); |
| 20405 | } |
| 20406 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 20407 | at::functionalization::impl::commit_update(out0); |
| 20408 | at::functionalization::impl::sync(out0); |
| 20409 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 20410 | at::functionalization::impl::commit_update(out1); |
| 20411 | at::functionalization::impl::sync(out1); |
| 20412 | return ::std::tuple<at::Tensor &,at::Tensor &>(out0, out1); |
| 20413 | } |
| 20414 | } |
| 20415 | |
| 20416 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &> _transformer_decoder_only_layer_fwd_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & src, int64_t embed_dim, int64_t num_heads, const at::Tensor & qkv_weight, const at::Tensor & qkv_bias, const at::Tensor & proj_weight, const at::Tensor & proj_bias, bool use_gelu, bool norm_first, double eps, const at::Tensor & norm_weight_1, const at::Tensor & norm_bias_1, const at::Tensor & norm_weight_2, const at::Tensor & norm_bias_2, const at::Tensor & ffn_weight_1, const at::Tensor & ffn_bias_1, const at::Tensor & ffn_weight_2, const at::Tensor & ffn_bias_2, const c10::optional<at::Tensor> & mask, const c10::optional<at::Tensor> & incr_key, const c10::optional<at::Tensor> & incr_value, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2) { |
| 20417 | if (false) { |
| 20418 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20419 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20420 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20421 | auto src_meta = to_meta(src); |
| 20422 | auto qkv_weight_meta = to_meta(qkv_weight); |
| 20423 | auto qkv_bias_meta = to_meta(qkv_bias); |
| 20424 | auto proj_weight_meta = to_meta(proj_weight); |
| 20425 | auto proj_bias_meta = to_meta(proj_bias); |
| 20426 | auto norm_weight_1_meta = to_meta(norm_weight_1); |
| 20427 | auto norm_bias_1_meta = to_meta(norm_bias_1); |
| 20428 | auto norm_weight_2_meta = to_meta(norm_weight_2); |
| 20429 | auto norm_bias_2_meta = to_meta(norm_bias_2); |
| 20430 | auto ffn_weight_1_meta = to_meta(ffn_weight_1); |
| 20431 | auto ffn_bias_1_meta = to_meta(ffn_bias_1); |
| 20432 | auto ffn_weight_2_meta = to_meta(ffn_weight_2); |
| 20433 | auto ffn_bias_2_meta = to_meta(ffn_bias_2); |
| 20434 | auto mask_meta = to_meta(mask); |
| 20435 | auto incr_key_meta = to_meta(incr_key); |
| 20436 | auto incr_value_meta = to_meta(incr_value); |
| 20437 | auto out0_meta = to_meta(out0); |
| 20438 | auto out1_meta = to_meta(out1); |
| 20439 | auto out2_meta = to_meta(out2); |
| 20440 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20441 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20442 | at::_ops::_transformer_decoder_only_layer_fwd_out::call(src_meta, embed_dim, num_heads, qkv_weight_meta, qkv_bias_meta, proj_weight_meta, proj_bias_meta, use_gelu, norm_first, eps, norm_weight_1_meta, norm_bias_1_meta, norm_weight_2_meta, norm_bias_2_meta, ffn_weight_1_meta, ffn_bias_1_meta, ffn_weight_2_meta, ffn_bias_2_meta, mask_meta, incr_key_meta, incr_value_meta, out0_meta, out1_meta, out2_meta); |
| 20443 | } |
| 20444 | |
| 20445 | at::Tensor src_; |
| 20446 | if (at::functionalization::impl::isFunctionalTensor(src)) { |
| 20447 | at::functionalization::impl::sync(src); |
| 20448 | src_ = at::functionalization::impl::from_functional_tensor(src); |
| 20449 | } else { |
| 20450 | src_ = src; |
| 20451 | } |
| 20452 | |
| 20453 | at::Tensor qkv_weight_; |
| 20454 | if (at::functionalization::impl::isFunctionalTensor(qkv_weight)) { |
| 20455 | at::functionalization::impl::sync(qkv_weight); |
| 20456 | qkv_weight_ = at::functionalization::impl::from_functional_tensor(qkv_weight); |
| 20457 | } else { |
| 20458 | qkv_weight_ = qkv_weight; |
| 20459 | } |
| 20460 | |
| 20461 | at::Tensor qkv_bias_; |
| 20462 | if (at::functionalization::impl::isFunctionalTensor(qkv_bias)) { |
| 20463 | at::functionalization::impl::sync(qkv_bias); |
| 20464 | qkv_bias_ = at::functionalization::impl::from_functional_tensor(qkv_bias); |
| 20465 | } else { |
| 20466 | qkv_bias_ = qkv_bias; |
| 20467 | } |
| 20468 | |
| 20469 | at::Tensor proj_weight_; |
| 20470 | if (at::functionalization::impl::isFunctionalTensor(proj_weight)) { |
| 20471 | at::functionalization::impl::sync(proj_weight); |
| 20472 | proj_weight_ = at::functionalization::impl::from_functional_tensor(proj_weight); |
| 20473 | } else { |
| 20474 | proj_weight_ = proj_weight; |
| 20475 | } |
| 20476 | |
| 20477 | at::Tensor proj_bias_; |
| 20478 | if (at::functionalization::impl::isFunctionalTensor(proj_bias)) { |
| 20479 | at::functionalization::impl::sync(proj_bias); |
| 20480 | proj_bias_ = at::functionalization::impl::from_functional_tensor(proj_bias); |
| 20481 | } else { |
| 20482 | proj_bias_ = proj_bias; |
| 20483 | } |
| 20484 | |
| 20485 | at::Tensor norm_weight_1_; |
| 20486 | if (at::functionalization::impl::isFunctionalTensor(norm_weight_1)) { |
| 20487 | at::functionalization::impl::sync(norm_weight_1); |
| 20488 | norm_weight_1_ = at::functionalization::impl::from_functional_tensor(norm_weight_1); |
| 20489 | } else { |
| 20490 | norm_weight_1_ = norm_weight_1; |
| 20491 | } |
| 20492 | |
| 20493 | at::Tensor norm_bias_1_; |
| 20494 | if (at::functionalization::impl::isFunctionalTensor(norm_bias_1)) { |
| 20495 | at::functionalization::impl::sync(norm_bias_1); |
| 20496 | norm_bias_1_ = at::functionalization::impl::from_functional_tensor(norm_bias_1); |
| 20497 | } else { |
| 20498 | norm_bias_1_ = norm_bias_1; |
| 20499 | } |
| 20500 | |
| 20501 | at::Tensor norm_weight_2_; |
| 20502 | if (at::functionalization::impl::isFunctionalTensor(norm_weight_2)) { |
| 20503 | at::functionalization::impl::sync(norm_weight_2); |
| 20504 | norm_weight_2_ = at::functionalization::impl::from_functional_tensor(norm_weight_2); |
| 20505 | } else { |
| 20506 | norm_weight_2_ = norm_weight_2; |
| 20507 | } |
| 20508 | |
| 20509 | at::Tensor norm_bias_2_; |
| 20510 | if (at::functionalization::impl::isFunctionalTensor(norm_bias_2)) { |
| 20511 | at::functionalization::impl::sync(norm_bias_2); |
| 20512 | norm_bias_2_ = at::functionalization::impl::from_functional_tensor(norm_bias_2); |
| 20513 | } else { |
| 20514 | norm_bias_2_ = norm_bias_2; |
| 20515 | } |
| 20516 | |
| 20517 | at::Tensor ffn_weight_1_; |
| 20518 | if (at::functionalization::impl::isFunctionalTensor(ffn_weight_1)) { |
| 20519 | at::functionalization::impl::sync(ffn_weight_1); |
| 20520 | ffn_weight_1_ = at::functionalization::impl::from_functional_tensor(ffn_weight_1); |
| 20521 | } else { |
| 20522 | ffn_weight_1_ = ffn_weight_1; |
| 20523 | } |
| 20524 | |
| 20525 | at::Tensor ffn_bias_1_; |
| 20526 | if (at::functionalization::impl::isFunctionalTensor(ffn_bias_1)) { |
| 20527 | at::functionalization::impl::sync(ffn_bias_1); |
| 20528 | ffn_bias_1_ = at::functionalization::impl::from_functional_tensor(ffn_bias_1); |
| 20529 | } else { |
| 20530 | ffn_bias_1_ = ffn_bias_1; |
| 20531 | } |
| 20532 | |
| 20533 | at::Tensor ffn_weight_2_; |
| 20534 | if (at::functionalization::impl::isFunctionalTensor(ffn_weight_2)) { |
| 20535 | at::functionalization::impl::sync(ffn_weight_2); |
| 20536 | ffn_weight_2_ = at::functionalization::impl::from_functional_tensor(ffn_weight_2); |
| 20537 | } else { |
| 20538 | ffn_weight_2_ = ffn_weight_2; |
| 20539 | } |
| 20540 | |
| 20541 | at::Tensor ffn_bias_2_; |
| 20542 | if (at::functionalization::impl::isFunctionalTensor(ffn_bias_2)) { |
| 20543 | at::functionalization::impl::sync(ffn_bias_2); |
| 20544 | ffn_bias_2_ = at::functionalization::impl::from_functional_tensor(ffn_bias_2); |
| 20545 | } else { |
| 20546 | ffn_bias_2_ = ffn_bias_2; |
| 20547 | } |
| 20548 | |
| 20549 | c10::optional<at::Tensor> mask_; |
| 20550 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 20551 | at::functionalization::impl::sync(mask); |
| 20552 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 20553 | } else { |
| 20554 | mask_ = mask; |
| 20555 | } |
| 20556 | |
| 20557 | c10::optional<at::Tensor> incr_key_; |
| 20558 | if (at::functionalization::impl::isFunctionalTensor(incr_key)) { |
| 20559 | at::functionalization::impl::sync(incr_key); |
| 20560 | incr_key_ = at::functionalization::impl::from_functional_tensor(incr_key); |
| 20561 | } else { |
| 20562 | incr_key_ = incr_key; |
| 20563 | } |
| 20564 | |
| 20565 | c10::optional<at::Tensor> incr_value_; |
| 20566 | if (at::functionalization::impl::isFunctionalTensor(incr_value)) { |
| 20567 | at::functionalization::impl::sync(incr_value); |
| 20568 | incr_value_ = at::functionalization::impl::from_functional_tensor(incr_value); |
| 20569 | } else { |
| 20570 | incr_value_ = incr_value; |
| 20571 | } |
| 20572 | |
| 20573 | at::Tensor out0_; |
| 20574 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 20575 | at::functionalization::impl::sync(out0); |
| 20576 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 20577 | } else { |
| 20578 | out0_ = out0; |
| 20579 | } |
| 20580 | |
| 20581 | at::Tensor out1_; |
| 20582 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 20583 | at::functionalization::impl::sync(out1); |
| 20584 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 20585 | } else { |
| 20586 | out1_ = out1; |
| 20587 | } |
| 20588 | |
| 20589 | at::Tensor out2_; |
| 20590 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 20591 | at::functionalization::impl::sync(out2); |
| 20592 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 20593 | } else { |
| 20594 | out2_ = out2; |
| 20595 | } |
| 20596 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2))) { |
| 20597 | if ((false || at::functionalization::impl::isFunctionalTensor(src) || at::functionalization::impl::isFunctionalTensor(qkv_weight) || at::functionalization::impl::isFunctionalTensor(qkv_bias) || at::functionalization::impl::isFunctionalTensor(proj_weight) || at::functionalization::impl::isFunctionalTensor(proj_bias) || at::functionalization::impl::isFunctionalTensor(norm_weight_1) || at::functionalization::impl::isFunctionalTensor(norm_bias_1) || at::functionalization::impl::isFunctionalTensor(norm_weight_2) || at::functionalization::impl::isFunctionalTensor(norm_bias_2) || at::functionalization::impl::isFunctionalTensor(ffn_weight_1) || at::functionalization::impl::isFunctionalTensor(ffn_bias_1) || at::functionalization::impl::isFunctionalTensor(ffn_weight_2) || at::functionalization::impl::isFunctionalTensor(ffn_bias_2) || at::functionalization::impl::isFunctionalTensor(mask) || at::functionalization::impl::isFunctionalTensor(incr_key) || at::functionalization::impl::isFunctionalTensor(incr_value))) { |
| 20598 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20599 | TORCH_INTERNAL_ASSERT(false, |
| 20600 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20601 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20602 | } else { |
| 20603 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20604 | at::AutoDispatchSkipFunctionalize guard; |
| 20605 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_transformer_decoder_only_layer_fwd_out::call(src_, embed_dim, num_heads, qkv_weight_, qkv_bias_, proj_weight_, proj_bias_, use_gelu, norm_first, eps, norm_weight_1_, norm_bias_1_, norm_weight_2_, norm_bias_2_, ffn_weight_1_, ffn_bias_1_, ffn_weight_2_, ffn_bias_2_, mask_, incr_key_, incr_value_, out0_, out1_, out2_); |
| 20606 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2);; |
| 20607 | } |
| 20608 | } else { |
| 20609 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 20610 | { |
| 20611 | at::AutoDispatchSkipFunctionalize guard; |
| 20612 | tmp_output = at::_ops::_transformer_decoder_only_layer_fwd::call(src_, embed_dim, num_heads, qkv_weight_, qkv_bias_, proj_weight_, proj_bias_, use_gelu, norm_first, eps, norm_weight_1_, norm_bias_1_, norm_weight_2_, norm_bias_2_, ffn_weight_1_, ffn_bias_1_, ffn_weight_2_, ffn_bias_2_, mask_, incr_key_, incr_value_); |
| 20613 | } |
| 20614 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 20615 | at::functionalization::impl::commit_update(out0); |
| 20616 | at::functionalization::impl::sync(out0); |
| 20617 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 20618 | at::functionalization::impl::commit_update(out1); |
| 20619 | at::functionalization::impl::sync(out1); |
| 20620 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 20621 | at::functionalization::impl::commit_update(out2); |
| 20622 | at::functionalization::impl::sync(out2); |
| 20623 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2); |
| 20624 | } |
| 20625 | } |
| 20626 | |
| 20627 | ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &> _native_decoder_only_multi_head_attention_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, int64_t embed_dim, int64_t num_head, const at::Tensor & qkv_weight, const at::Tensor & qkv_bias, const at::Tensor & proj_weight, const at::Tensor & proj_bias, const c10::optional<at::Tensor> & mask, const c10::optional<at::Tensor> & incr_key, const c10::optional<at::Tensor> & incr_value, bool need_weights, bool average_attn_weights, at::Tensor & out0, at::Tensor & out1, at::Tensor & out2, at::Tensor & out3) { |
| 20628 | if (false) { |
| 20629 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20630 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20631 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20632 | auto query_meta = to_meta(query); |
| 20633 | auto key_meta = to_meta(key); |
| 20634 | auto value_meta = to_meta(value); |
| 20635 | auto qkv_weight_meta = to_meta(qkv_weight); |
| 20636 | auto qkv_bias_meta = to_meta(qkv_bias); |
| 20637 | auto proj_weight_meta = to_meta(proj_weight); |
| 20638 | auto proj_bias_meta = to_meta(proj_bias); |
| 20639 | auto mask_meta = to_meta(mask); |
| 20640 | auto incr_key_meta = to_meta(incr_key); |
| 20641 | auto incr_value_meta = to_meta(incr_value); |
| 20642 | auto out0_meta = to_meta(out0); |
| 20643 | auto out1_meta = to_meta(out1); |
| 20644 | auto out2_meta = to_meta(out2); |
| 20645 | auto out3_meta = to_meta(out3); |
| 20646 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20647 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20648 | at::_ops::_native_decoder_only_multi_head_attention_out::call(query_meta, key_meta, value_meta, embed_dim, num_head, qkv_weight_meta, qkv_bias_meta, proj_weight_meta, proj_bias_meta, mask_meta, incr_key_meta, incr_value_meta, need_weights, average_attn_weights, out0_meta, out1_meta, out2_meta, out3_meta); |
| 20649 | } |
| 20650 | |
| 20651 | at::Tensor query_; |
| 20652 | if (at::functionalization::impl::isFunctionalTensor(query)) { |
| 20653 | at::functionalization::impl::sync(query); |
| 20654 | query_ = at::functionalization::impl::from_functional_tensor(query); |
| 20655 | } else { |
| 20656 | query_ = query; |
| 20657 | } |
| 20658 | |
| 20659 | at::Tensor key_; |
| 20660 | if (at::functionalization::impl::isFunctionalTensor(key)) { |
| 20661 | at::functionalization::impl::sync(key); |
| 20662 | key_ = at::functionalization::impl::from_functional_tensor(key); |
| 20663 | } else { |
| 20664 | key_ = key; |
| 20665 | } |
| 20666 | |
| 20667 | at::Tensor value_; |
| 20668 | if (at::functionalization::impl::isFunctionalTensor(value)) { |
| 20669 | at::functionalization::impl::sync(value); |
| 20670 | value_ = at::functionalization::impl::from_functional_tensor(value); |
| 20671 | } else { |
| 20672 | value_ = value; |
| 20673 | } |
| 20674 | |
| 20675 | at::Tensor qkv_weight_; |
| 20676 | if (at::functionalization::impl::isFunctionalTensor(qkv_weight)) { |
| 20677 | at::functionalization::impl::sync(qkv_weight); |
| 20678 | qkv_weight_ = at::functionalization::impl::from_functional_tensor(qkv_weight); |
| 20679 | } else { |
| 20680 | qkv_weight_ = qkv_weight; |
| 20681 | } |
| 20682 | |
| 20683 | at::Tensor qkv_bias_; |
| 20684 | if (at::functionalization::impl::isFunctionalTensor(qkv_bias)) { |
| 20685 | at::functionalization::impl::sync(qkv_bias); |
| 20686 | qkv_bias_ = at::functionalization::impl::from_functional_tensor(qkv_bias); |
| 20687 | } else { |
| 20688 | qkv_bias_ = qkv_bias; |
| 20689 | } |
| 20690 | |
| 20691 | at::Tensor proj_weight_; |
| 20692 | if (at::functionalization::impl::isFunctionalTensor(proj_weight)) { |
| 20693 | at::functionalization::impl::sync(proj_weight); |
| 20694 | proj_weight_ = at::functionalization::impl::from_functional_tensor(proj_weight); |
| 20695 | } else { |
| 20696 | proj_weight_ = proj_weight; |
| 20697 | } |
| 20698 | |
| 20699 | at::Tensor proj_bias_; |
| 20700 | if (at::functionalization::impl::isFunctionalTensor(proj_bias)) { |
| 20701 | at::functionalization::impl::sync(proj_bias); |
| 20702 | proj_bias_ = at::functionalization::impl::from_functional_tensor(proj_bias); |
| 20703 | } else { |
| 20704 | proj_bias_ = proj_bias; |
| 20705 | } |
| 20706 | |
| 20707 | c10::optional<at::Tensor> mask_; |
| 20708 | if (at::functionalization::impl::isFunctionalTensor(mask)) { |
| 20709 | at::functionalization::impl::sync(mask); |
| 20710 | mask_ = at::functionalization::impl::from_functional_tensor(mask); |
| 20711 | } else { |
| 20712 | mask_ = mask; |
| 20713 | } |
| 20714 | |
| 20715 | c10::optional<at::Tensor> incr_key_; |
| 20716 | if (at::functionalization::impl::isFunctionalTensor(incr_key)) { |
| 20717 | at::functionalization::impl::sync(incr_key); |
| 20718 | incr_key_ = at::functionalization::impl::from_functional_tensor(incr_key); |
| 20719 | } else { |
| 20720 | incr_key_ = incr_key; |
| 20721 | } |
| 20722 | |
| 20723 | c10::optional<at::Tensor> incr_value_; |
| 20724 | if (at::functionalization::impl::isFunctionalTensor(incr_value)) { |
| 20725 | at::functionalization::impl::sync(incr_value); |
| 20726 | incr_value_ = at::functionalization::impl::from_functional_tensor(incr_value); |
| 20727 | } else { |
| 20728 | incr_value_ = incr_value; |
| 20729 | } |
| 20730 | |
| 20731 | at::Tensor out0_; |
| 20732 | if (at::functionalization::impl::isFunctionalTensor(out0)) { |
| 20733 | at::functionalization::impl::sync(out0); |
| 20734 | out0_ = at::functionalization::impl::from_functional_tensor(out0); |
| 20735 | } else { |
| 20736 | out0_ = out0; |
| 20737 | } |
| 20738 | |
| 20739 | at::Tensor out1_; |
| 20740 | if (at::functionalization::impl::isFunctionalTensor(out1)) { |
| 20741 | at::functionalization::impl::sync(out1); |
| 20742 | out1_ = at::functionalization::impl::from_functional_tensor(out1); |
| 20743 | } else { |
| 20744 | out1_ = out1; |
| 20745 | } |
| 20746 | |
| 20747 | at::Tensor out2_; |
| 20748 | if (at::functionalization::impl::isFunctionalTensor(out2)) { |
| 20749 | at::functionalization::impl::sync(out2); |
| 20750 | out2_ = at::functionalization::impl::from_functional_tensor(out2); |
| 20751 | } else { |
| 20752 | out2_ = out2; |
| 20753 | } |
| 20754 | |
| 20755 | at::Tensor out3_; |
| 20756 | if (at::functionalization::impl::isFunctionalTensor(out3)) { |
| 20757 | at::functionalization::impl::sync(out3); |
| 20758 | out3_ = at::functionalization::impl::from_functional_tensor(out3); |
| 20759 | } else { |
| 20760 | out3_ = out3; |
| 20761 | } |
| 20762 | if (!(true && at::functionalization::impl::isFunctionalTensor(out0) && at::functionalization::impl::isFunctionalTensor(out1) && at::functionalization::impl::isFunctionalTensor(out2) && at::functionalization::impl::isFunctionalTensor(out3))) { |
| 20763 | if ((false || at::functionalization::impl::isFunctionalTensor(query) || at::functionalization::impl::isFunctionalTensor(key) || at::functionalization::impl::isFunctionalTensor(value) || at::functionalization::impl::isFunctionalTensor(qkv_weight) || at::functionalization::impl::isFunctionalTensor(qkv_bias) || at::functionalization::impl::isFunctionalTensor(proj_weight) || at::functionalization::impl::isFunctionalTensor(proj_bias) || at::functionalization::impl::isFunctionalTensor(mask) || at::functionalization::impl::isFunctionalTensor(incr_key) || at::functionalization::impl::isFunctionalTensor(incr_value))) { |
| 20764 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20765 | TORCH_INTERNAL_ASSERT(false, |
| 20766 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20767 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20768 | } else { |
| 20769 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20770 | at::AutoDispatchSkipFunctionalize guard; |
| 20771 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output = at::_ops::_native_decoder_only_multi_head_attention_out::call(query_, key_, value_, embed_dim, num_head, qkv_weight_, qkv_bias_, proj_weight_, proj_bias_, mask_, incr_key_, incr_value_, need_weights, average_attn_weights, out0_, out1_, out2_, out3_); |
| 20772 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2, out3);; |
| 20773 | } |
| 20774 | } else { |
| 20775 | ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor> tmp_output; |
| 20776 | { |
| 20777 | at::AutoDispatchSkipFunctionalize guard; |
| 20778 | tmp_output = at::_ops::_native_decoder_only_multi_head_attention::call(query_, key_, value_, embed_dim, num_head, qkv_weight_, qkv_bias_, proj_weight_, proj_bias_, mask_, incr_key_, incr_value_, need_weights, average_attn_weights); |
| 20779 | } |
| 20780 | at::functionalization::impl::replace_(out0, std::get<0>(tmp_output)); |
| 20781 | at::functionalization::impl::commit_update(out0); |
| 20782 | at::functionalization::impl::sync(out0); |
| 20783 | at::functionalization::impl::replace_(out1, std::get<1>(tmp_output)); |
| 20784 | at::functionalization::impl::commit_update(out1); |
| 20785 | at::functionalization::impl::sync(out1); |
| 20786 | at::functionalization::impl::replace_(out2, std::get<2>(tmp_output)); |
| 20787 | at::functionalization::impl::commit_update(out2); |
| 20788 | at::functionalization::impl::sync(out2); |
| 20789 | at::functionalization::impl::replace_(out3, std::get<3>(tmp_output)); |
| 20790 | at::functionalization::impl::commit_update(out3); |
| 20791 | at::functionalization::impl::sync(out3); |
| 20792 | return ::std::tuple<at::Tensor &,at::Tensor &,at::Tensor &,at::Tensor &>(out0, out1, out2, out3); |
| 20793 | } |
| 20794 | } |
| 20795 | |
| 20796 | at::Tensor & special_modified_bessel_i1_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 20797 | if (false) { |
| 20798 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20799 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20800 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20801 | auto self_meta = to_meta(self); |
| 20802 | auto out_meta = to_meta(out); |
| 20803 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20804 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20805 | at::_ops::special_modified_bessel_i1_out::call(self_meta, out_meta); |
| 20806 | } |
| 20807 | |
| 20808 | at::Tensor self_; |
| 20809 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 20810 | at::functionalization::impl::sync(self); |
| 20811 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 20812 | } else { |
| 20813 | self_ = self; |
| 20814 | } |
| 20815 | |
| 20816 | at::Tensor out_; |
| 20817 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 20818 | at::functionalization::impl::sync(out); |
| 20819 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 20820 | } else { |
| 20821 | out_ = out; |
| 20822 | } |
| 20823 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 20824 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 20825 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20826 | TORCH_INTERNAL_ASSERT(false, |
| 20827 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20828 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20829 | } else { |
| 20830 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20831 | at::AutoDispatchSkipFunctionalize guard; |
| 20832 | at::Tensor tmp_output = at::_ops::special_modified_bessel_i1_out::call(self_, out_); |
| 20833 | return out;; |
| 20834 | } |
| 20835 | } else { |
| 20836 | at::Tensor tmp_output; |
| 20837 | { |
| 20838 | at::AutoDispatchSkipFunctionalize guard; |
| 20839 | tmp_output = at::_ops::special_modified_bessel_i1::call(self_); |
| 20840 | } |
| 20841 | at::functionalization::impl::replace_(out, tmp_output); |
| 20842 | at::functionalization::impl::commit_update(out); |
| 20843 | at::functionalization::impl::sync(out); |
| 20844 | return out; |
| 20845 | } |
| 20846 | } |
| 20847 | |
| 20848 | at::Tensor & special_modified_bessel_k0_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, at::Tensor & out) { |
| 20849 | if (false) { |
| 20850 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20851 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20852 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20853 | auto self_meta = to_meta(self); |
| 20854 | auto out_meta = to_meta(out); |
| 20855 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20856 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20857 | at::_ops::special_modified_bessel_k0_out::call(self_meta, out_meta); |
| 20858 | } |
| 20859 | |
| 20860 | at::Tensor self_; |
| 20861 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 20862 | at::functionalization::impl::sync(self); |
| 20863 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 20864 | } else { |
| 20865 | self_ = self; |
| 20866 | } |
| 20867 | |
| 20868 | at::Tensor out_; |
| 20869 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 20870 | at::functionalization::impl::sync(out); |
| 20871 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 20872 | } else { |
| 20873 | out_ = out; |
| 20874 | } |
| 20875 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 20876 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 20877 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20878 | TORCH_INTERNAL_ASSERT(false, |
| 20879 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20880 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20881 | } else { |
| 20882 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20883 | at::AutoDispatchSkipFunctionalize guard; |
| 20884 | at::Tensor tmp_output = at::_ops::special_modified_bessel_k0_out::call(self_, out_); |
| 20885 | return out;; |
| 20886 | } |
| 20887 | } else { |
| 20888 | at::Tensor tmp_output; |
| 20889 | { |
| 20890 | at::AutoDispatchSkipFunctionalize guard; |
| 20891 | tmp_output = at::_ops::special_modified_bessel_k0::call(self_); |
| 20892 | } |
| 20893 | at::functionalization::impl::replace_(out, tmp_output); |
| 20894 | at::functionalization::impl::commit_update(out); |
| 20895 | at::functionalization::impl::sync(out); |
| 20896 | return out; |
| 20897 | } |
| 20898 | } |
| 20899 | |
| 20900 | at::Tensor & _foobar_out_out(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, bool arg1, bool arg2, bool arg3, at::Tensor & out) { |
| 20901 | if (false) { |
| 20902 | // Before converting the mutable op to its functional variant, run meta tensors through the original op. |
| 20903 | // This will help us catch shape errors that apply to inplace ops that wouldn't apply to their functional variants. |
| 20904 | // (We can only do this for inplace ops today though, because they technicaly all support meta tensors). |
| 20905 | auto self_meta = to_meta(self); |
| 20906 | auto out_meta = to_meta(out); |
| 20907 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20908 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20909 | at::_ops::_foobar_out::call(self_meta, arg1, arg2, arg3, out_meta); |
| 20910 | } |
| 20911 | |
| 20912 | at::Tensor self_; |
| 20913 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 20914 | at::functionalization::impl::sync(self); |
| 20915 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 20916 | } else { |
| 20917 | self_ = self; |
| 20918 | } |
| 20919 | |
| 20920 | at::Tensor out_; |
| 20921 | if (at::functionalization::impl::isFunctionalTensor(out)) { |
| 20922 | at::functionalization::impl::sync(out); |
| 20923 | out_ = at::functionalization::impl::from_functional_tensor(out); |
| 20924 | } else { |
| 20925 | out_ = out; |
| 20926 | } |
| 20927 | if (!(true && at::functionalization::impl::isFunctionalTensor(out))) { |
| 20928 | if ((false || at::functionalization::impl::isFunctionalTensor(self))) { |
| 20929 | // case 1: trying to mutate a non functional tensor with a functional tensor is an error |
| 20930 | TORCH_INTERNAL_ASSERT(false, |
| 20931 | "mutating a non-functional tensor with a functional tensor is not allowed.", |
| 20932 | " Please ensure that all of your inputs are wrapped inside of a functionalize() call."); |
| 20933 | } else { |
| 20934 | // case 2: arguments are not functional tensors, so we no-op and redispatch. |
| 20935 | at::AutoDispatchSkipFunctionalize guard; |
| 20936 | at::Tensor tmp_output = at::_ops::_foobar_out::call(self_, arg1, arg2, arg3, out_); |
| 20937 | return out;; |
| 20938 | } |
| 20939 | } else { |
| 20940 | at::Tensor tmp_output; |
| 20941 | { |
| 20942 | at::AutoDispatchSkipFunctionalize guard; |
| 20943 | tmp_output = at::_ops::_foobar::call(self_, arg1, arg2, arg3); |
| 20944 | } |
| 20945 | at::functionalization::impl::replace_(out, tmp_output); |
| 20946 | at::functionalization::impl::commit_update(out); |
| 20947 | at::functionalization::impl::sync(out); |
| 20948 | return out; |
| 20949 | } |
| 20950 | } |
| 20951 | |
| 20952 | at::Tensor view_as_real(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self) { |
| 20953 | |
| 20954 | at::Tensor self_; |
| 20955 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 20956 | |
| 20957 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 20958 | } else { |
| 20959 | self_ = self; |
| 20960 | } |
| 20961 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 20962 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 20963 | at::AutoDispatchSkipFunctionalize guard; |
| 20964 | return at::_ops::view_as_real::call(self_); |
| 20965 | } |
| 20966 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 20967 | auto compute_reference_meta = |
| 20968 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 20969 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 20970 | at::Tensor reference_tensor_output; |
| 20971 | if (compute_reference_meta) { |
| 20972 | auto self_meta = to_meta(self); |
| 20973 | at::AutoDispatchSkipFunctionalize func_guard; |
| 20974 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 20975 | reference_tensor_output = at::_ops::view_as_real::call(self_meta); |
| 20976 | } |
| 20977 | at::Tensor tmp_output; |
| 20978 | { |
| 20979 | at::AutoDispatchSkipFunctionalize guard; |
| 20980 | if (reapply_views) { |
| 20981 | tmp_output = at::_ops::view_as_real::call(self_); |
| 20982 | } else { |
| 20983 | tmp_output = at::_ops::view_as_real_copy::call(self_); |
| 20984 | } |
| 20985 | } |
| 20986 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 20987 | [reapply_views = reapply_views](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 20988 | if (reapply_views) { |
| 20989 | return at::_ops::view_as_real::call(base); |
| 20990 | } else { |
| 20991 | return at::_ops::view_as_real_copy::call(base); |
| 20992 | } |
| 20993 | }, |
| 20994 | [reapply_views = reapply_views](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 20995 | return at::functionalization::FunctionalInverses::view_as_real_copy_inverse(base, mutated_view, reapply_views); |
| 20996 | } |
| 20997 | ); |
| 20998 | auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, view_meta); |
| 20999 | // See Note [Propagating strides in the functionalization pass] |
| 21000 | if (compute_reference_meta) { |
| 21001 | at::functionalization::impl::set_sizes_strides_offset(out, reference_tensor_output); |
| 21002 | } |
| 21003 | return out; |
| 21004 | } |
| 21005 | |
| 21006 | at::Tensor slice_Tensor(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dim, c10::optional<c10::SymInt> start, c10::optional<c10::SymInt> end, c10::SymInt step) { |
| 21007 | |
| 21008 | at::Tensor self_; |
| 21009 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 21010 | |
| 21011 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 21012 | } else { |
| 21013 | self_ = self; |
| 21014 | } |
| 21015 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 21016 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 21017 | at::AutoDispatchSkipFunctionalize guard; |
| 21018 | return at::_ops::slice_Tensor::call(self_, dim, start, end, step); |
| 21019 | } |
| 21020 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 21021 | auto compute_reference_meta = |
| 21022 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 21023 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 21024 | at::Tensor reference_tensor_output; |
| 21025 | if (compute_reference_meta) { |
| 21026 | auto self_meta = to_meta(self); |
| 21027 | at::AutoDispatchSkipFunctionalize func_guard; |
| 21028 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 21029 | reference_tensor_output = at::_ops::slice_Tensor::call(self_meta, dim, start, end, step); |
| 21030 | } |
| 21031 | at::Tensor tmp_output; |
| 21032 | { |
| 21033 | at::AutoDispatchSkipFunctionalize guard; |
| 21034 | if (reapply_views) { |
| 21035 | tmp_output = at::_ops::slice_Tensor::call(self_, dim, start, end, step); |
| 21036 | } else { |
| 21037 | tmp_output = at::_ops::slice_copy_Tensor::call(self_, dim, start, end, step); |
| 21038 | } |
| 21039 | } |
| 21040 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 21041 | [reapply_views = reapply_views, dim = dim, start = start, end = end, step = step](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 21042 | if (reapply_views) { |
| 21043 | return at::_ops::slice_Tensor::call(base, dim, start, end, step); |
| 21044 | } else { |
| 21045 | return at::_ops::slice_copy_Tensor::call(base, dim, start, end, step); |
| 21046 | } |
| 21047 | }, |
| 21048 | [reapply_views = reapply_views, dim = dim, start = start, end = end, step = step](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 21049 | return at::functionalization::FunctionalInverses::slice_copy_Tensor_inverse(base, mutated_view, reapply_views, dim, start, end, step); |
| 21050 | } |
| 21051 | ); |
| 21052 | auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, view_meta); |
| 21053 | // See Note [Propagating strides in the functionalization pass] |
| 21054 | if (compute_reference_meta) { |
| 21055 | at::functionalization::impl::set_sizes_strides_offset(out, reference_tensor_output); |
| 21056 | } |
| 21057 | return out; |
| 21058 | } |
| 21059 | |
| 21060 | at::Tensor t(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self) { |
| 21061 | |
| 21062 | at::Tensor self_; |
| 21063 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 21064 | |
| 21065 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 21066 | } else { |
| 21067 | self_ = self; |
| 21068 | } |
| 21069 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 21070 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 21071 | at::AutoDispatchSkipFunctionalize guard; |
| 21072 | return at::_ops::t::call(self_); |
| 21073 | } |
| 21074 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 21075 | auto compute_reference_meta = |
| 21076 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 21077 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 21078 | at::Tensor reference_tensor_output; |
| 21079 | if (compute_reference_meta) { |
| 21080 | auto self_meta = to_meta(self); |
| 21081 | at::AutoDispatchSkipFunctionalize func_guard; |
| 21082 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 21083 | reference_tensor_output = at::_ops::t::call(self_meta); |
| 21084 | } |
| 21085 | at::Tensor tmp_output; |
| 21086 | { |
| 21087 | at::AutoDispatchSkipFunctionalize guard; |
| 21088 | if (reapply_views) { |
| 21089 | tmp_output = at::_ops::t::call(self_); |
| 21090 | } else { |
| 21091 | tmp_output = at::_ops::t_copy::call(self_); |
| 21092 | } |
| 21093 | } |
| 21094 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 21095 | [reapply_views = reapply_views](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 21096 | if (reapply_views) { |
| 21097 | return at::_ops::t::call(base); |
| 21098 | } else { |
| 21099 | return at::_ops::t_copy::call(base); |
| 21100 | } |
| 21101 | }, |
| 21102 | [reapply_views = reapply_views](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 21103 | return at::functionalization::FunctionalInverses::t_copy_inverse(base, mutated_view, reapply_views); |
| 21104 | } |
| 21105 | ); |
| 21106 | auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, view_meta); |
| 21107 | // See Note [Propagating strides in the functionalization pass] |
| 21108 | if (compute_reference_meta) { |
| 21109 | at::functionalization::impl::set_sizes_strides_offset(out, reference_tensor_output); |
| 21110 | } |
| 21111 | return out; |
| 21112 | } |
| 21113 | |
| 21114 | at::Tensor & t_(c10::DispatchKeySet dispatchKeySet, at::Tensor & self) { |
| 21115 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 21116 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 21117 | |
| 21118 | at::Tensor self_; |
| 21119 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 21120 | |
| 21121 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 21122 | } else { |
| 21123 | self_ = self; |
| 21124 | } |
| 21125 | at::AutoDispatchSkipFunctionalize guard; |
| 21126 | return at::_ops::t_::call(self_); |
| 21127 | } |
| 21128 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 21129 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 21130 | [reapply_views = reapply_views](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 21131 | if (reapply_views) { |
| 21132 | return at::_ops::t::call(base); |
| 21133 | } else { |
| 21134 | return at::_ops::t_copy::call(base); |
| 21135 | } |
| 21136 | }, |
| 21137 | [reapply_views = reapply_views](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 21138 | return at::functionalization::FunctionalInverses::t_copy_inverse(base, mutated_view, reapply_views); |
| 21139 | } |
| 21140 | ); |
| 21141 | auto compute_reference_meta = |
| 21142 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 21143 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 21144 | at::Tensor reference_tensor_output; |
| 21145 | if (compute_reference_meta) { |
| 21146 | auto self_meta = to_meta(self); |
| 21147 | at::AutoDispatchSkipFunctionalize func_guard; |
| 21148 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 21149 | reference_tensor_output = at::_ops::t_::call(self_meta); |
| 21150 | } |
| 21151 | // This function adds the above view meta to the current tensor and replays them off the base, |
| 21152 | // mutating the size/stride info of the current FunctionalTensorWrapper. |
| 21153 | // Because of this, we need to make sure to run the reference shape function above, |
| 21154 | // BEFORE doing this (otherwise we'll end up runnin the reference function using the wrong sizes/strides) |
| 21155 | at::functionalization::impl::mutate_view_meta(self, view_meta); |
| 21156 | // See Note [Propagating strides in the functionalization pass] |
| 21157 | // XLA/LTC don't implement the logic to propagate strides correctly, so we need to rely |
| 21158 | // on a reference implementation here (instead of relying on the output from the forward lambda |
| 21159 | // having the correct stride info) |
| 21160 | if (compute_reference_meta) { |
| 21161 | at::functionalization::impl::set_sizes_strides_offset(self, reference_tensor_output); |
| 21162 | } |
| 21163 | return self; |
| 21164 | } |
| 21165 | |
| 21166 | at::Tensor _indices(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self) { |
| 21167 | |
| 21168 | at::Tensor self_; |
| 21169 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 21170 | |
| 21171 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 21172 | } else { |
| 21173 | self_ = self; |
| 21174 | } |
| 21175 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 21176 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 21177 | at::AutoDispatchSkipFunctionalize guard; |
| 21178 | return at::_ops::_indices::call(self_); |
| 21179 | } |
| 21180 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 21181 | auto compute_reference_meta = |
| 21182 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 21183 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 21184 | at::Tensor reference_tensor_output; |
| 21185 | if (compute_reference_meta) { |
| 21186 | auto self_meta = to_meta(self); |
| 21187 | at::AutoDispatchSkipFunctionalize func_guard; |
| 21188 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 21189 | reference_tensor_output = at::_ops::_indices::call(self_meta); |
| 21190 | } |
| 21191 | at::Tensor tmp_output; |
| 21192 | { |
| 21193 | at::AutoDispatchSkipFunctionalize guard; |
| 21194 | if (reapply_views) { |
| 21195 | tmp_output = at::_ops::_indices::call(self_); |
| 21196 | } else { |
| 21197 | tmp_output = at::_ops::_indices_copy::call(self_); |
| 21198 | } |
| 21199 | } |
| 21200 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 21201 | [reapply_views = reapply_views](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 21202 | if (reapply_views) { |
| 21203 | return at::_ops::_indices::call(base); |
| 21204 | } else { |
| 21205 | return at::_ops::_indices_copy::call(base); |
| 21206 | } |
| 21207 | }, |
| 21208 | [reapply_views = reapply_views](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 21209 | return at::functionalization::FunctionalInverses::_indices_copy_inverse(base, mutated_view, reapply_views); |
| 21210 | } |
| 21211 | ); |
| 21212 | auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, view_meta); |
| 21213 | // See Note [Propagating strides in the functionalization pass] |
| 21214 | if (compute_reference_meta) { |
| 21215 | at::functionalization::impl::set_sizes_strides_offset(out, reference_tensor_output); |
| 21216 | } |
| 21217 | return out; |
| 21218 | } |
| 21219 | |
| 21220 | at::Tensor indices(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self) { |
| 21221 | |
| 21222 | at::Tensor self_; |
| 21223 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 21224 | |
| 21225 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 21226 | } else { |
| 21227 | self_ = self; |
| 21228 | } |
| 21229 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 21230 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 21231 | at::AutoDispatchSkipFunctionalize guard; |
| 21232 | return at::_ops::indices::call(self_); |
| 21233 | } |
| 21234 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 21235 | auto compute_reference_meta = |
| 21236 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 21237 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 21238 | at::Tensor reference_tensor_output; |
| 21239 | if (compute_reference_meta) { |
| 21240 | auto self_meta = to_meta(self); |
| 21241 | at::AutoDispatchSkipFunctionalize func_guard; |
| 21242 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 21243 | reference_tensor_output = at::_ops::indices::call(self_meta); |
| 21244 | } |
| 21245 | at::Tensor tmp_output; |
| 21246 | { |
| 21247 | at::AutoDispatchSkipFunctionalize guard; |
| 21248 | if (reapply_views) { |
| 21249 | tmp_output = at::_ops::indices::call(self_); |
| 21250 | } else { |
| 21251 | tmp_output = at::_ops::indices_copy::call(self_); |
| 21252 | } |
| 21253 | } |
| 21254 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 21255 | [reapply_views = reapply_views](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 21256 | if (reapply_views) { |
| 21257 | return at::_ops::indices::call(base); |
| 21258 | } else { |
| 21259 | return at::_ops::indices_copy::call(base); |
| 21260 | } |
| 21261 | }, |
| 21262 | [reapply_views = reapply_views](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 21263 | return at::functionalization::FunctionalInverses::indices_copy_inverse(base, mutated_view, reapply_views); |
| 21264 | } |
| 21265 | ); |
| 21266 | auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, view_meta); |
| 21267 | // See Note [Propagating strides in the functionalization pass] |
| 21268 | if (compute_reference_meta) { |
| 21269 | at::functionalization::impl::set_sizes_strides_offset(out, reference_tensor_output); |
| 21270 | } |
| 21271 | return out; |
| 21272 | } |
| 21273 | |
| 21274 | at::Tensor crow_indices(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self) { |
| 21275 | |
| 21276 | at::Tensor self_; |
| 21277 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 21278 | |
| 21279 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 21280 | } else { |
| 21281 | self_ = self; |
| 21282 | } |
| 21283 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 21284 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 21285 | at::AutoDispatchSkipFunctionalize guard; |
| 21286 | return at::_ops::crow_indices::call(self_); |
| 21287 | } |
| 21288 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 21289 | auto compute_reference_meta = |
| 21290 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 21291 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 21292 | at::Tensor reference_tensor_output; |
| 21293 | if (compute_reference_meta) { |
| 21294 | auto self_meta = to_meta(self); |
| 21295 | at::AutoDispatchSkipFunctionalize func_guard; |
| 21296 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 21297 | reference_tensor_output = at::_ops::crow_indices::call(self_meta); |
| 21298 | } |
| 21299 | at::Tensor tmp_output; |
| 21300 | { |
| 21301 | at::AutoDispatchSkipFunctionalize guard; |
| 21302 | if (reapply_views) { |
| 21303 | tmp_output = at::_ops::crow_indices::call(self_); |
| 21304 | } else { |
| 21305 | tmp_output = at::_ops::crow_indices_copy::call(self_); |
| 21306 | } |
| 21307 | } |
| 21308 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 21309 | [reapply_views = reapply_views](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 21310 | if (reapply_views) { |
| 21311 | return at::_ops::crow_indices::call(base); |
| 21312 | } else { |
| 21313 | return at::_ops::crow_indices_copy::call(base); |
| 21314 | } |
| 21315 | }, |
| 21316 | [reapply_views = reapply_views](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 21317 | return at::functionalization::FunctionalInverses::crow_indices_copy_inverse(base, mutated_view, reapply_views); |
| 21318 | } |
| 21319 | ); |
| 21320 | auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, view_meta); |
| 21321 | // See Note [Propagating strides in the functionalization pass] |
| 21322 | if (compute_reference_meta) { |
| 21323 | at::functionalization::impl::set_sizes_strides_offset(out, reference_tensor_output); |
| 21324 | } |
| 21325 | return out; |
| 21326 | } |
| 21327 | |
| 21328 | at::Tensor col_indices(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self) { |
| 21329 | |
| 21330 | at::Tensor self_; |
| 21331 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 21332 | |
| 21333 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 21334 | } else { |
| 21335 | self_ = self; |
| 21336 | } |
| 21337 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 21338 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 21339 | at::AutoDispatchSkipFunctionalize guard; |
| 21340 | return at::_ops::col_indices::call(self_); |
| 21341 | } |
| 21342 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 21343 | auto compute_reference_meta = |
| 21344 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 21345 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 21346 | at::Tensor reference_tensor_output; |
| 21347 | if (compute_reference_meta) { |
| 21348 | auto self_meta = to_meta(self); |
| 21349 | at::AutoDispatchSkipFunctionalize func_guard; |
| 21350 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 21351 | reference_tensor_output = at::_ops::col_indices::call(self_meta); |
| 21352 | } |
| 21353 | at::Tensor tmp_output; |
| 21354 | { |
| 21355 | at::AutoDispatchSkipFunctionalize guard; |
| 21356 | if (reapply_views) { |
| 21357 | tmp_output = at::_ops::col_indices::call(self_); |
| 21358 | } else { |
| 21359 | tmp_output = at::_ops::col_indices_copy::call(self_); |
| 21360 | } |
| 21361 | } |
| 21362 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 21363 | [reapply_views = reapply_views](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 21364 | if (reapply_views) { |
| 21365 | return at::_ops::col_indices::call(base); |
| 21366 | } else { |
| 21367 | return at::_ops::col_indices_copy::call(base); |
| 21368 | } |
| 21369 | }, |
| 21370 | [reapply_views = reapply_views](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 21371 | return at::functionalization::FunctionalInverses::col_indices_copy_inverse(base, mutated_view, reapply_views); |
| 21372 | } |
| 21373 | ); |
| 21374 | auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, view_meta); |
| 21375 | // See Note [Propagating strides in the functionalization pass] |
| 21376 | if (compute_reference_meta) { |
| 21377 | at::functionalization::impl::set_sizes_strides_offset(out, reference_tensor_output); |
| 21378 | } |
| 21379 | return out; |
| 21380 | } |
| 21381 | |
| 21382 | at::Tensor unfold(c10::DispatchKeySet dispatchKeySet, const at::Tensor & self, int64_t dimension, int64_t size, int64_t step) { |
| 21383 | |
| 21384 | at::Tensor self_; |
| 21385 | if (at::functionalization::impl::isFunctionalTensor(self)) { |
| 21386 | |
| 21387 | self_ = at::functionalization::impl::from_functional_tensor(self); |
| 21388 | } else { |
| 21389 | self_ = self; |
| 21390 | } |
| 21391 | if (!at::functionalization::impl::isFunctionalTensor(self)) { |
| 21392 | // functionalization is re-entrant, but will no-op if it wasn't passed a FunctionalTensorWrapper. |
| 21393 | at::AutoDispatchSkipFunctionalize guard; |
| 21394 | return at::_ops::unfold::call(self_, dimension, size, step); |
| 21395 | } |
| 21396 | auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS(); |
| 21397 | auto compute_reference_meta = |
| 21398 | self.key_set().has_backend(c10::BackendComponent::XLABit) || |
| 21399 | self.key_set().has_backend(c10::BackendComponent::LazyBit); |
| 21400 | at::Tensor reference_tensor_output; |
| 21401 | if (compute_reference_meta) { |
| 21402 | auto self_meta = to_meta(self); |
| 21403 | at::AutoDispatchSkipFunctionalize func_guard; |
| 21404 | c10::impl::ExcludeDispatchKeyGuard guard(exclude_keys_for_meta_dispatch); |
| 21405 | reference_tensor_output = at::_ops::unfold::call(self_meta, dimension, size, step); |
| 21406 | } |
| 21407 | at::Tensor tmp_output; |
| 21408 | { |
| 21409 | at::AutoDispatchSkipFunctionalize guard; |
| 21410 | if (reapply_views) { |
| 21411 | tmp_output = at::_ops::unfold::call(self_, dimension, size, step); |
| 21412 | } else { |
| 21413 | tmp_output = at::_ops::unfold_copy::call(self_, dimension, size, step); |
| 21414 | } |
| 21415 | } |
| 21416 | at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta( |
| 21417 | [reapply_views = reapply_views, dimension = dimension, size = size, step = step](const at::Tensor & base, int64_t mutated_view_idx) -> at::Tensor { |
| 21418 | if (reapply_views) { |
| 21419 | return at::_ops::unfold::call(base, dimension, size, step); |
| 21420 | } else { |
| 21421 | return at::_ops::unfold_copy::call(base, dimension, size, step); |
| 21422 | } |
| 21423 | }, |
| 21424 | [reapply_views = reapply_views, dimension = dimension, size = size, step = step](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx) -> at::Tensor { |
| 21425 | return at::functionalization::FunctionalInverses::unfold_copy_inverse(base, mutated_view, reapply_views, dimension, size, step); |
| 21426 | } |
| 21427 | ); |
| 21428 | auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, view_meta); |
| 21429 | // See Note [Propagating strides in the functionalization pass] |
| 21430 | if (compute_reference_meta) { |
| 21431 | at::functionalization::impl::set_sizes_strides_offset(out, reference_tensor_output); |
| 21432 | } |
| 21433 | return out; |
| 21434 | } |
| 21435 | |
| 21436 | } // namespace functionalization |
| 21437 | |
| 21438 | namespace { |
| 21439 | |
| 21440 | TORCH_LIBRARY_IMPL(aten, Functionalize, m) { |
| 21441 | m.impl("_cudnn_rnn_flatten_weight.out", TORCH_FN(functionalization::_cudnn_rnn_flatten_weight_out_out)); |
| 21442 | m.impl("abs.out", TORCH_FN(functionalization::abs_out_out)); |
| 21443 | m.impl("abs_", TORCH_FN(functionalization::abs_)); |
| 21444 | m.impl("absolute", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::absolute)); |
| 21445 | m.impl("absolute.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Tensor & out)>(at::native::absolute_out)); |
| 21446 | m.impl("absolute_", static_cast<at::Tensor & (*)(at::Tensor & self)>(at::native::absolute_)); |
| 21447 | m.impl("_conj_physical.out", TORCH_FN(functionalization::_conj_physical_out_out)); |
| 21448 | m.impl("acos.out", TORCH_FN(functionalization::acos_out_out)); |
| 21449 | m.impl("acos_", TORCH_FN(functionalization::acos_)); |
| 21450 | m.impl("arccos", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::arccos)); |
| 21451 | m.impl("arccos.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Tensor & out)>(at::native::arccos_out)); |
| 21452 | m.impl("arccos_", static_cast<at::Tensor & (*)(at::Tensor & self)>(at::native::arccos_)); |
| 21453 | m.impl("affine_grid_generator.out", TORCH_FN(functionalization::affine_grid_generator_out_out)); |
| 21454 | m.impl("arange.out", TORCH_FN(functionalization::arange_out_out)); |
| 21455 | m.impl("arange.start_out", TORCH_FN(functionalization::arange_out_start_out)); |
| 21456 | m.impl("arccosh", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::arccosh)); |
| 21457 | m.impl("arccosh.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Tensor & out)>(at::native::arccosh_out)); |
| 21458 | m.impl("arccosh_", static_cast<at::Tensor & (*)(at::Tensor & self)>(at::native::arccosh_)); |
| 21459 | m.impl("arcsinh", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::arcsinh)); |
| 21460 | m.impl("arcsinh.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Tensor & out)>(at::native::arcsinh_out)); |
| 21461 | m.impl("arcsinh_", static_cast<at::Tensor & (*)(at::Tensor & self)>(at::native::arcsinh_)); |
| 21462 | m.impl("atanh.out", TORCH_FN(functionalization::atanh_out_out)); |
| 21463 | m.impl("atanh_", TORCH_FN(functionalization::atanh_)); |
| 21464 | m.impl("asin.out", TORCH_FN(functionalization::asin_out_out)); |
| 21465 | m.impl("asin_", TORCH_FN(functionalization::asin_)); |
| 21466 | m.impl("binary_cross_entropy_backward.grad_input", TORCH_FN(functionalization::binary_cross_entropy_backward_out_grad_input)); |
| 21467 | m.impl("binary_cross_entropy_with_logits.out", TORCH_FN(functionalization::binary_cross_entropy_with_logits_out_out)); |
| 21468 | m.impl("logical_not.out", TORCH_FN(functionalization::logical_not_out_out)); |
| 21469 | m.impl("logical_not_", TORCH_FN(functionalization::logical_not_)); |
| 21470 | m.impl("logical_and.out", TORCH_FN(functionalization::logical_and_out_out)); |
| 21471 | m.impl("logical_and_", TORCH_FN(functionalization::logical_and_)); |
| 21472 | m.impl("concatenate", static_cast<at::Tensor (*)(at::TensorList tensors, int64_t dim)>(at::native::concatenate)); |
| 21473 | m.impl("concatenate.out", static_cast<at::Tensor & (*)(at::TensorList tensors, int64_t dim, at::Tensor & out)>(at::native::concatenate_out)); |
| 21474 | m.impl("concatenate.names", static_cast<at::Tensor (*)(at::TensorList tensors, at::Dimname dim)>(at::native::concatenate)); |
| 21475 | m.impl("concatenate.names_out", static_cast<at::Tensor & (*)(at::TensorList tensors, at::Dimname dim, at::Tensor & out)>(at::native::concatenate_out)); |
| 21476 | m.impl("block_diag.out", TORCH_FN(functionalization::block_diag_out_out)); |
| 21477 | m.impl("chain_matmul", static_cast<at::Tensor (*)(at::TensorList matrices)>(at::native::chain_matmul)); |
| 21478 | m.impl("chain_matmul.out", static_cast<at::Tensor & (*)(at::TensorList matrices, at::Tensor & out)>(at::native::chain_matmul_out)); |
| 21479 | m.impl("convolution_backward.out", TORCH_FN(functionalization::convolution_backward_out_out)); |
| 21480 | m.impl("_copy_from.out", TORCH_FN(functionalization::_copy_from_out_out)); |
| 21481 | m.impl("cosh.out", TORCH_FN(functionalization::cosh_out_out)); |
| 21482 | m.impl("cosh_", TORCH_FN(functionalization::cosh_)); |
| 21483 | m.impl("cudnn_convolution_transpose.out", TORCH_FN(functionalization::cudnn_convolution_transpose_out_out)); |
| 21484 | m.impl("_mps_convolution_transpose.out", TORCH_FN(functionalization::_mps_convolution_transpose_out_out)); |
| 21485 | m.impl("cudnn_grid_sampler.out", TORCH_FN(functionalization::cudnn_grid_sampler_out_out)); |
| 21486 | m.impl("cumprod.out", TORCH_FN(functionalization::cumprod_out_out)); |
| 21487 | m.impl("cumprod_", TORCH_FN(functionalization::cumprod_)); |
| 21488 | m.impl("cumprod.dimname", static_cast<at::Tensor (*)(const at::Tensor & self, at::Dimname dim, c10::optional<at::ScalarType> dtype)>(at::native::cumprod)); |
| 21489 | m.impl("cumprod.dimname_out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Dimname dim, c10::optional<at::ScalarType> dtype, at::Tensor & out)>(at::native::cumprod_out)); |
| 21490 | m.impl("cumprod_.dimname", static_cast<at::Tensor & (*)(at::Tensor & self, at::Dimname dim, c10::optional<at::ScalarType> dtype)>(at::native::cumprod_)); |
| 21491 | m.impl("diag_embed.out", TORCH_FN(functionalization::diag_embed_out_out)); |
| 21492 | m.impl("diagonal_backward.out", TORCH_FN(functionalization::diagonal_backward_out_out)); |
| 21493 | m.impl("div.out", TORCH_FN(functionalization::div_out_out)); |
| 21494 | m.impl("div_.Tensor", TORCH_FN(functionalization::div__Tensor)); |
| 21495 | m.impl("div.out_mode", TORCH_FN(functionalization::div_out_out_mode)); |
| 21496 | m.impl("div_.Tensor_mode", TORCH_FN(functionalization::div__Tensor_mode)); |
| 21497 | m.impl("div.Scalar_out", TORCH_FN(functionalization::div_out_Scalar_out)); |
| 21498 | m.impl("div_.Scalar", TORCH_FN(functionalization::div__Scalar)); |
| 21499 | m.impl("div.Scalar_mode_out", TORCH_FN(functionalization::div_out_Scalar_mode_out)); |
| 21500 | m.impl("div_.Scalar_mode", TORCH_FN(functionalization::div__Scalar_mode)); |
| 21501 | m.impl("_embedding_bag.out", TORCH_FN(functionalization::_embedding_bag_out_out)); |
| 21502 | m.impl("_embedding_bag_per_sample_weights_backward.out", TORCH_FN(functionalization::_embedding_bag_per_sample_weights_backward_out_out)); |
| 21503 | m.impl("new_full.out", TORCH_FN(functionalization::new_full_out_out)); |
| 21504 | m.impl("empty_quantized.out", TORCH_FN(functionalization::empty_quantized_out_out)); |
| 21505 | m.impl("empty_strided.out", TORCH_FN(functionalization::empty_strided_out_out)); |
| 21506 | m.impl("exp.out", TORCH_FN(functionalization::exp_out_out)); |
| 21507 | m.impl("exp_", TORCH_FN(functionalization::exp_)); |
| 21508 | m.impl("expm1.out", TORCH_FN(functionalization::expm1_out_out)); |
| 21509 | m.impl("expm1_", TORCH_FN(functionalization::expm1_)); |
| 21510 | m.impl("fill.Scalar_out", TORCH_FN(functionalization::fill_out_Scalar_out)); |
| 21511 | m.impl("fill_.Scalar", TORCH_FN(functionalization::fill__Scalar)); |
| 21512 | m.impl("fill.Tensor_out", TORCH_FN(functionalization::fill_out_Tensor_out)); |
| 21513 | m.impl("fill_.Tensor", TORCH_FN(functionalization::fill__Tensor)); |
| 21514 | m.impl("floor.out", TORCH_FN(functionalization::floor_out_out)); |
| 21515 | m.impl("floor_", TORCH_FN(functionalization::floor_)); |
| 21516 | m.impl("floor_divide.out", TORCH_FN(functionalization::floor_divide_out_out)); |
| 21517 | m.impl("floor_divide_.Tensor", TORCH_FN(functionalization::floor_divide__Tensor)); |
| 21518 | m.impl("full.names_out", TORCH_FN(functionalization::full_out_names_out)); |
| 21519 | m.impl("full.out", TORCH_FN(functionalization::full_out_out)); |
| 21520 | m.impl("from_file.out", TORCH_FN(functionalization::from_file_out_out)); |
| 21521 | m.impl("grid_sampler_2d.out", TORCH_FN(functionalization::grid_sampler_2d_out_out)); |
| 21522 | m.impl("grid_sampler_3d.out", TORCH_FN(functionalization::grid_sampler_3d_out_out)); |
| 21523 | m.impl("hamming_window.out", TORCH_FN(functionalization::hamming_window_out_out)); |
| 21524 | m.impl("hamming_window.periodic_out", TORCH_FN(functionalization::hamming_window_out_periodic_out)); |
| 21525 | m.impl("hamming_window.periodic_alpha_out", TORCH_FN(functionalization::hamming_window_out_periodic_alpha_out)); |
| 21526 | m.impl("hamming_window.periodic_alpha_beta_out", TORCH_FN(functionalization::hamming_window_out_periodic_alpha_beta_out)); |
| 21527 | m.impl("native_group_norm.out", TORCH_FN(functionalization::native_group_norm_out_out)); |
| 21528 | m.impl("_fft_c2r.out", TORCH_FN(functionalization::_fft_c2r_out_out)); |
| 21529 | m.impl("isnan.out", TORCH_FN(functionalization::isnan_out_out)); |
| 21530 | m.impl("ldexp.Tensor", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & other)>(at::native::ldexp)); |
| 21531 | m.impl("ldexp.out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Tensor & other, at::Tensor & out)>(at::native::ldexp_out)); |
| 21532 | m.impl("ldexp_", static_cast<at::Tensor & (*)(at::Tensor & self, const at::Tensor & other)>(at::native::ldexp_)); |
| 21533 | m.impl("log2.out", TORCH_FN(functionalization::log2_out_out)); |
| 21534 | m.impl("log2_", TORCH_FN(functionalization::log2_)); |
| 21535 | m.impl("logaddexp2.out", TORCH_FN(functionalization::logaddexp2_out_out)); |
| 21536 | m.impl("xlogy.OutTensor", TORCH_FN(functionalization::xlogy_out_OutTensor)); |
| 21537 | m.impl("xlogy_.Tensor", TORCH_FN(functionalization::xlogy__Tensor)); |
| 21538 | m.impl("xlogy.OutScalar_Self", TORCH_FN(functionalization::xlogy_out_OutScalar_Self)); |
| 21539 | m.impl("xlogy.OutScalar_Other", TORCH_FN(functionalization::xlogy_out_OutScalar_Other)); |
| 21540 | m.impl("xlogy_.Scalar_Other", TORCH_FN(functionalization::xlogy__Scalar_Other)); |
| 21541 | m.impl("logspace.out", TORCH_FN(functionalization::logspace_out_out)); |
| 21542 | m.impl("matrix_power", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t n)>(at::native::matrix_power)); |
| 21543 | m.impl("matrix_power.out", static_cast<at::Tensor & (*)(const at::Tensor & self, int64_t n, at::Tensor & out)>(at::native::matrix_power_out)); |
| 21544 | m.impl("_aminmax.out", TORCH_FN(functionalization::_aminmax_out_out)); |
| 21545 | m.impl("_aminmax.dim_out", TORCH_FN(functionalization::_aminmax_out_dim_out)); |
| 21546 | m.impl("aminmax.out", TORCH_FN(functionalization::aminmax_out_out)); |
| 21547 | m.impl("_compute_linear_combination.out", TORCH_FN(functionalization::_compute_linear_combination_out_out)); |
| 21548 | m.impl("_mps_max_pool2d.out", TORCH_FN(functionalization::_mps_max_pool2d_out_out)); |
| 21549 | m.impl("mkldnn_max_pool3d_backward.out", TORCH_FN(functionalization::mkldnn_max_pool3d_backward_out_out)); |
| 21550 | m.impl("min.dim_min", TORCH_FN(functionalization::min_out_dim_min)); |
| 21551 | m.impl("min.names_dim", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & self, at::Dimname dim, bool keepdim)>(at::native::min)); |
| 21552 | m.impl("min.names_dim_min", static_cast<::std::tuple<at::Tensor &,at::Tensor &> (*)(const at::Tensor & self, at::Dimname dim, bool keepdim, at::Tensor & min, at::Tensor & min_indices)>(at::native::min_out)); |
| 21553 | m.impl("mps_convolution_backward.out", TORCH_FN(functionalization::mps_convolution_backward_out_out)); |
| 21554 | m.impl("miopen_rnn.out", TORCH_FN(functionalization::miopen_rnn_out_out)); |
| 21555 | m.impl("mv.out", TORCH_FN(functionalization::mv_out_out)); |
| 21556 | m.impl("_native_batch_norm_legit.out", TORCH_FN(functionalization::_native_batch_norm_legit_out_out)); |
| 21557 | m.impl("_native_batch_norm_legit", TORCH_FN(functionalization::_native_batch_norm_legit)); |
| 21558 | m.impl("_native_batch_norm_legit.no_stats_out", TORCH_FN(functionalization::_native_batch_norm_legit_out_no_stats_out)); |
| 21559 | m.impl("batch_norm_stats.out", TORCH_FN(functionalization::batch_norm_stats_out_out)); |
| 21560 | m.impl("batch_norm_backward_elemt.out", TORCH_FN(functionalization::batch_norm_backward_elemt_out_out)); |
| 21561 | m.impl("_euclidean_dist.out", TORCH_FN(functionalization::_euclidean_dist_out_out)); |
| 21562 | m.impl("_cdist_forward.out", TORCH_FN(functionalization::_cdist_forward_out_out)); |
| 21563 | m.impl("_cdist_backward.out", TORCH_FN(functionalization::_cdist_backward_out_out)); |
| 21564 | m.impl("pixel_unshuffle.out", TORCH_FN(functionalization::pixel_unshuffle_out_out)); |
| 21565 | m.impl("rad2deg.out", TORCH_FN(functionalization::rad2deg_out_out)); |
| 21566 | m.impl("rad2deg_", TORCH_FN(functionalization::rad2deg_)); |
| 21567 | m.impl("scalar_tensor.out", TORCH_FN(functionalization::scalar_tensor_out_out)); |
| 21568 | m.impl("rand.names_out", TORCH_FN(functionalization::rand_out_names_out)); |
| 21569 | m.impl("rand.generator_with_names_out", TORCH_FN(functionalization::rand_out_generator_with_names_out)); |
| 21570 | m.impl("rand.out", TORCH_FN(functionalization::rand_out_out)); |
| 21571 | m.impl("rand.generator_out", static_cast<at::Tensor & (*)(at::IntArrayRef size, c10::optional<at::Generator> generator, at::Tensor & out)>(at::native::rand_out)); |
| 21572 | m.impl("rand_like.out", TORCH_FN(functionalization::rand_like_out_out)); |
| 21573 | m.impl("relu.out", TORCH_FN(functionalization::relu_out_out)); |
| 21574 | m.impl("relu_", TORCH_FN(functionalization::relu_)); |
| 21575 | m.impl("logit.out", TORCH_FN(functionalization::logit_out_out)); |
| 21576 | m.impl("logit_", TORCH_FN(functionalization::logit_)); |
| 21577 | m.impl("select_scatter.out", TORCH_FN(functionalization::select_scatter_out_out)); |
| 21578 | m.impl("softmax.int", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t dim, c10::optional<at::ScalarType> dtype)>(at::native::softmax)); |
| 21579 | m.impl("softmax.int_out", TORCH_FN(functionalization::softmax_out_int_out)); |
| 21580 | m.impl("stack.out", TORCH_FN(functionalization::stack_out_out)); |
| 21581 | m.impl("vstack", static_cast<at::Tensor (*)(at::TensorList tensors)>(at::native::vstack)); |
| 21582 | m.impl("vstack.out", static_cast<at::Tensor & (*)(at::TensorList tensors, at::Tensor & out)>(at::native::vstack_out)); |
| 21583 | m.impl("nansum.out", TORCH_FN(functionalization::nansum_out_out)); |
| 21584 | m.impl("sqrt.out", TORCH_FN(functionalization::sqrt_out_out)); |
| 21585 | m.impl("sqrt_", TORCH_FN(functionalization::sqrt_)); |
| 21586 | m.impl("prod.out", TORCH_FN(functionalization::prod_out_out)); |
| 21587 | m.impl("prod.int_out", TORCH_FN(functionalization::prod_out_int_out)); |
| 21588 | m.impl("prod.dim_Dimname", static_cast<at::Tensor (*)(const at::Tensor & self, at::Dimname dim, bool keepdim, c10::optional<at::ScalarType> dtype)>(at::native::prod)); |
| 21589 | m.impl("prod.Dimname_out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Dimname dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out)>(at::native::prod_out)); |
| 21590 | m.impl("threshold_backward.grad_input", TORCH_FN(functionalization::threshold_backward_out_grad_input)); |
| 21591 | m.impl("_transform_bias_rescale_qkv.out", TORCH_FN(functionalization::_transform_bias_rescale_qkv_out_out)); |
| 21592 | m.impl("_nested_from_padded.out", TORCH_FN(functionalization::_nested_from_padded_out_out)); |
| 21593 | m.impl("_nested_tensor_size.out", TORCH_FN(functionalization::_nested_tensor_size_out_out)); |
| 21594 | m.impl("_nested_view_from_buffer_copy.out", TORCH_FN(functionalization::_nested_view_from_buffer_copy_out_out)); |
| 21595 | m.impl("trunc.out", TORCH_FN(functionalization::trunc_out_out)); |
| 21596 | m.impl("trunc_", TORCH_FN(functionalization::trunc_)); |
| 21597 | m.impl("unique_dim_consecutive.out", TORCH_FN(functionalization::unique_dim_consecutive_out_out)); |
| 21598 | m.impl("where.self_out", TORCH_FN(functionalization::where_out_self_out)); |
| 21599 | m.impl("_weight_norm_interface_backward.out", TORCH_FN(functionalization::_weight_norm_interface_backward_out_out)); |
| 21600 | m.impl("_sample_dirichlet.out", TORCH_FN(functionalization::_sample_dirichlet_out_out)); |
| 21601 | m.impl("binomial.out", TORCH_FN(functionalization::binomial_out_out)); |
| 21602 | m.impl("native_norm.out", TORCH_FN(functionalization::native_norm_out_out)); |
| 21603 | m.impl("native_norm.ScalarOpt_dim_dtype_out", TORCH_FN(functionalization::native_norm_out_ScalarOpt_dim_dtype_out)); |
| 21604 | m.impl("_sparse_sum.dim_out", TORCH_FN(functionalization::_sparse_sum_out_dim_out)); |
| 21605 | m.impl("_sparse_sum_backward.out", TORCH_FN(functionalization::_sparse_sum_backward_out_out)); |
| 21606 | m.impl("_sparse_softmax.out", TORCH_FN(functionalization::_sparse_softmax_out_out)); |
| 21607 | m.impl("clone.out", TORCH_FN(functionalization::clone_out_out)); |
| 21608 | m.impl("resize_as.out", TORCH_FN(functionalization::resize_as_out_out)); |
| 21609 | m.impl("resize_as_", TORCH_FN(functionalization::resize_as_)); |
| 21610 | m.impl("zero.out", TORCH_FN(functionalization::zero_out_out)); |
| 21611 | m.impl("zero_", TORCH_FN(functionalization::zero_)); |
| 21612 | m.impl("heaviside.out", TORCH_FN(functionalization::heaviside_out_out)); |
| 21613 | m.impl("heaviside_", TORCH_FN(functionalization::heaviside_)); |
| 21614 | m.impl("addmm.out", TORCH_FN(functionalization::addmm_out_out)); |
| 21615 | m.impl("addmm_", TORCH_FN(functionalization::addmm_)); |
| 21616 | m.impl("_sparse_coo_tensor_with_dims.out", TORCH_FN(functionalization::_sparse_coo_tensor_with_dims_out_out)); |
| 21617 | m.impl("_sparse_coo_tensor_with_dims_and_tensors.out", TORCH_FN(functionalization::_sparse_coo_tensor_with_dims_and_tensors_out_out)); |
| 21618 | m.impl("sparse_resize_and_clear.out", TORCH_FN(functionalization::sparse_resize_and_clear_out_out)); |
| 21619 | m.impl("sparse_resize_and_clear_", TORCH_FN(functionalization::sparse_resize_and_clear_)); |
| 21620 | m.impl("hspmm.out", TORCH_FN(functionalization::hspmm_out_out)); |
| 21621 | m.impl("to_sparse.sparse_dim_out", TORCH_FN(functionalization::to_sparse_out_sparse_dim_out)); |
| 21622 | m.impl("to_sparse.out", TORCH_FN(functionalization::to_sparse_out_out)); |
| 21623 | m.impl("to_sparse_bsr.out", TORCH_FN(functionalization::to_sparse_bsr_out_out)); |
| 21624 | m.impl("to_mkldnn.out", TORCH_FN(functionalization::to_mkldnn_out_out)); |
| 21625 | m.impl("mkldnn_reorder_conv3d_weight.out", TORCH_FN(functionalization::mkldnn_reorder_conv3d_weight_out_out)); |
| 21626 | m.impl("q_per_channel_scales.out", TORCH_FN(functionalization::q_per_channel_scales_out_out)); |
| 21627 | m.impl("int_repr.out", TORCH_FN(functionalization::int_repr_out_out)); |
| 21628 | m.impl("_make_per_tensor_quantized_tensor.out", TORCH_FN(functionalization::_make_per_tensor_quantized_tensor_out_out)); |
| 21629 | m.impl("_thnn_fused_lstm_cell.out", TORCH_FN(functionalization::_thnn_fused_lstm_cell_out_out)); |
| 21630 | m.impl("_thnn_fused_lstm_cell_backward_impl.out", TORCH_FN(functionalization::_thnn_fused_lstm_cell_backward_impl_out_out)); |
| 21631 | m.impl("masked_fill.Scalar_out", TORCH_FN(functionalization::masked_fill_out_Scalar_out)); |
| 21632 | m.impl("masked_fill_.Scalar", TORCH_FN(functionalization::masked_fill__Scalar)); |
| 21633 | m.impl("masked_fill.Tensor_out", TORCH_FN(functionalization::masked_fill_out_Tensor_out)); |
| 21634 | m.impl("masked_fill_.Tensor", TORCH_FN(functionalization::masked_fill__Tensor)); |
| 21635 | m.impl("_masked_softmax.out", TORCH_FN(functionalization::_masked_softmax_out_out)); |
| 21636 | m.impl("bitwise_right_shift.Tensor_out", TORCH_FN(functionalization::bitwise_right_shift_out_Tensor_out)); |
| 21637 | m.impl("bitwise_right_shift_.Tensor", TORCH_FN(functionalization::bitwise_right_shift__Tensor)); |
| 21638 | m.impl("bitwise_right_shift.Tensor_Scalar_out", TORCH_FN(functionalization::bitwise_right_shift_out_Tensor_Scalar_out)); |
| 21639 | m.impl("bitwise_right_shift_.Tensor_Scalar", TORCH_FN(functionalization::bitwise_right_shift__Tensor_Scalar)); |
| 21640 | m.impl("bitwise_right_shift.Scalar_Tensor_out", TORCH_FN(functionalization::bitwise_right_shift_out_Scalar_Tensor_out)); |
| 21641 | m.impl("cauchy.out", TORCH_FN(functionalization::cauchy_out_out)); |
| 21642 | m.impl("cauchy_", TORCH_FN(functionalization::cauchy_)); |
| 21643 | m.impl("log_normal.out", TORCH_FN(functionalization::log_normal_out_out)); |
| 21644 | m.impl("log_normal_", TORCH_FN(functionalization::log_normal_)); |
| 21645 | m.impl("diag", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t diagonal)>(at::native::diag)); |
| 21646 | m.impl("diag.out", static_cast<at::Tensor & (*)(const at::Tensor & self, int64_t diagonal, at::Tensor & out)>(at::native::diag_out)); |
| 21647 | m.impl("ne.Scalar_out", TORCH_FN(functionalization::ne_out_Scalar_out)); |
| 21648 | m.impl("ne_.Scalar", TORCH_FN(functionalization::ne__Scalar)); |
| 21649 | m.impl("ne.Tensor_out", TORCH_FN(functionalization::ne_out_Tensor_out)); |
| 21650 | m.impl("ne_.Tensor", TORCH_FN(functionalization::ne__Tensor)); |
| 21651 | m.impl("not_equal.Scalar", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Scalar & other)>(at::native::not_equal)); |
| 21652 | m.impl("not_equal.Scalar_out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Scalar & other, at::Tensor & out)>(at::native::not_equal_out)); |
| 21653 | m.impl("not_equal_.Scalar", static_cast<at::Tensor & (*)(at::Tensor & self, const at::Scalar & other)>(at::native::not_equal_)); |
| 21654 | m.impl("not_equal.Tensor", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & other)>(at::native::not_equal)); |
| 21655 | m.impl("not_equal.Tensor_out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Tensor & other, at::Tensor & out)>(at::native::not_equal_out)); |
| 21656 | m.impl("not_equal_.Tensor", static_cast<at::Tensor & (*)(at::Tensor & self, const at::Tensor & other)>(at::native::not_equal_)); |
| 21657 | m.impl("addcmul.out", TORCH_FN(functionalization::addcmul_out_out)); |
| 21658 | m.impl("addcmul_", TORCH_FN(functionalization::addcmul_)); |
| 21659 | m.impl("ormqr.out", TORCH_FN(functionalization::ormqr_out_out)); |
| 21660 | m.impl("lu_unpack.out", TORCH_FN(functionalization::lu_unpack_out_out)); |
| 21661 | m.impl("dist.out", TORCH_FN(functionalization::dist_out_out)); |
| 21662 | m.impl("arctan2", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & other)>(at::native::arctan2)); |
| 21663 | m.impl("arctan2.out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Tensor & other, at::Tensor & out)>(at::native::arctan2_out)); |
| 21664 | m.impl("arctan2_", static_cast<at::Tensor & (*)(at::Tensor & self, const at::Tensor & other)>(at::native::arctan2_)); |
| 21665 | m.impl("histc.out", TORCH_FN(functionalization::histc_out_out)); |
| 21666 | m.impl("_histogramdd_from_bin_cts.out", TORCH_FN(functionalization::_histogramdd_from_bin_cts_out_out)); |
| 21667 | m.impl("hypot.out", TORCH_FN(functionalization::hypot_out_out)); |
| 21668 | m.impl("hypot_", TORCH_FN(functionalization::hypot_)); |
| 21669 | m.impl("min.other", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & other)>(at::native::min)); |
| 21670 | m.impl("min.out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Tensor & other, at::Tensor & out)>(at::native::min_out)); |
| 21671 | m.impl("msort", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::msort)); |
| 21672 | m.impl("msort.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Tensor & out)>(at::native::msort_out)); |
| 21673 | m.impl("_foreach_add.Scalar_out", TORCH_FN(functionalization::_foreach_add_out_Scalar_out)); |
| 21674 | m.impl("_foreach_add_.Scalar", TORCH_FN(functionalization::_foreach_add__Scalar)); |
| 21675 | m.impl("_foreach_clamp_min.Scalar_out", TORCH_FN(functionalization::_foreach_clamp_min_out_Scalar_out)); |
| 21676 | m.impl("_foreach_clamp_min_.Scalar", TORCH_FN(functionalization::_foreach_clamp_min__Scalar)); |
| 21677 | m.impl("_foreach_add.List_out", TORCH_FN(functionalization::_foreach_add_out_List_out)); |
| 21678 | m.impl("_foreach_add_.List", TORCH_FN(functionalization::_foreach_add__List)); |
| 21679 | m.impl("_foreach_clamp_min.List_out", TORCH_FN(functionalization::_foreach_clamp_min_out_List_out)); |
| 21680 | m.impl("_foreach_clamp_min_.List", TORCH_FN(functionalization::_foreach_clamp_min__List)); |
| 21681 | m.impl("_foreach_add.ScalarList_out", TORCH_FN(functionalization::_foreach_add_out_ScalarList_out)); |
| 21682 | m.impl("_foreach_add_.ScalarList", TORCH_FN(functionalization::_foreach_add__ScalarList)); |
| 21683 | m.impl("_foreach_clamp_min.ScalarList_out", TORCH_FN(functionalization::_foreach_clamp_min_out_ScalarList_out)); |
| 21684 | m.impl("_foreach_clamp_min_.ScalarList", TORCH_FN(functionalization::_foreach_clamp_min__ScalarList)); |
| 21685 | m.impl("_foreach_zero.out", TORCH_FN(functionalization::_foreach_zero_out_out)); |
| 21686 | m.impl("_foreach_zero_", TORCH_FN(functionalization::_foreach_zero_)); |
| 21687 | m.impl("_foreach_asin.out", TORCH_FN(functionalization::_foreach_asin_out_out)); |
| 21688 | m.impl("_foreach_asin_", TORCH_FN(functionalization::_foreach_asin_)); |
| 21689 | m.impl("_foreach_ceil.out", TORCH_FN(functionalization::_foreach_ceil_out_out)); |
| 21690 | m.impl("_foreach_ceil_", TORCH_FN(functionalization::_foreach_ceil_)); |
| 21691 | m.impl("_foreach_cosh.out", TORCH_FN(functionalization::_foreach_cosh_out_out)); |
| 21692 | m.impl("_foreach_cosh_", TORCH_FN(functionalization::_foreach_cosh_)); |
| 21693 | m.impl("_foreach_log1p.out", TORCH_FN(functionalization::_foreach_log1p_out_out)); |
| 21694 | m.impl("_foreach_log1p_", TORCH_FN(functionalization::_foreach_log1p_)); |
| 21695 | m.impl("_foreach_log2.out", TORCH_FN(functionalization::_foreach_log2_out_out)); |
| 21696 | m.impl("_foreach_log2_", TORCH_FN(functionalization::_foreach_log2_)); |
| 21697 | m.impl("_foreach_round.out", TORCH_FN(functionalization::_foreach_round_out_out)); |
| 21698 | m.impl("_foreach_round_", TORCH_FN(functionalization::_foreach_round_)); |
| 21699 | m.impl("_foreach_addcdiv.Scalar_out", TORCH_FN(functionalization::_foreach_addcdiv_out_Scalar_out)); |
| 21700 | m.impl("_foreach_addcdiv_.Scalar", TORCH_FN(functionalization::_foreach_addcdiv__Scalar)); |
| 21701 | m.impl("_foreach_addcmul.Scalar_out", TORCH_FN(functionalization::_foreach_addcmul_out_Scalar_out)); |
| 21702 | m.impl("_foreach_addcmul_.Scalar", TORCH_FN(functionalization::_foreach_addcmul__Scalar)); |
| 21703 | m.impl("_foreach_addcdiv.ScalarList_out", TORCH_FN(functionalization::_foreach_addcdiv_out_ScalarList_out)); |
| 21704 | m.impl("_foreach_addcdiv_.ScalarList", TORCH_FN(functionalization::_foreach_addcdiv__ScalarList)); |
| 21705 | m.impl("_foreach_addcdiv.Tensor_out", TORCH_FN(functionalization::_foreach_addcdiv_out_Tensor_out)); |
| 21706 | m.impl("_foreach_addcdiv_.Tensor", TORCH_FN(functionalization::_foreach_addcdiv__Tensor)); |
| 21707 | m.impl("_foreach_addcmul.ScalarList_out", TORCH_FN(functionalization::_foreach_addcmul_out_ScalarList_out)); |
| 21708 | m.impl("_foreach_addcmul_.ScalarList", TORCH_FN(functionalization::_foreach_addcmul__ScalarList)); |
| 21709 | m.impl("_foreach_addcmul.Tensor_out", TORCH_FN(functionalization::_foreach_addcmul_out_Tensor_out)); |
| 21710 | m.impl("_foreach_addcmul_.Tensor", TORCH_FN(functionalization::_foreach_addcmul__Tensor)); |
| 21711 | m.impl("bucketize.Tensor_out", TORCH_FN(functionalization::bucketize_out_Tensor_out)); |
| 21712 | m.impl("bucketize.Scalar_out", TORCH_FN(functionalization::bucketize_out_Scalar_out)); |
| 21713 | m.impl("mse_loss.out", TORCH_FN(functionalization::mse_loss_out_out)); |
| 21714 | m.impl("multi_margin_loss.out", TORCH_FN(functionalization::multi_margin_loss_out_out)); |
| 21715 | m.impl("multilabel_margin_loss", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & target, int64_t reduction)>(at::native::multilabel_margin_loss)); |
| 21716 | m.impl("multilabel_margin_loss.out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Tensor & target, int64_t reduction, at::Tensor & out)>(at::native::multilabel_margin_loss_out)); |
| 21717 | m.impl("multilabel_margin_loss_forward.output", TORCH_FN(functionalization::multilabel_margin_loss_forward_out_output)); |
| 21718 | m.impl("nll_loss2d_forward.output", TORCH_FN(functionalization::nll_loss2d_forward_out_output)); |
| 21719 | m.impl("soft_margin_loss.out", TORCH_FN(functionalization::soft_margin_loss_out_out)); |
| 21720 | m.impl("glu_backward.grad_input", TORCH_FN(functionalization::glu_backward_out_grad_input)); |
| 21721 | m.impl("glu_backward_jvp.out", TORCH_FN(functionalization::glu_backward_jvp_out_out)); |
| 21722 | m.impl("hardtanh.out", TORCH_FN(functionalization::hardtanh_out_out)); |
| 21723 | m.impl("hardtanh_", TORCH_FN(functionalization::hardtanh_)); |
| 21724 | m.impl("hardtanh_backward.grad_input", TORCH_FN(functionalization::hardtanh_backward_out_grad_input)); |
| 21725 | m.impl("leaky_relu.out", TORCH_FN(functionalization::leaky_relu_out_out)); |
| 21726 | m.impl("leaky_relu_", TORCH_FN(functionalization::leaky_relu_)); |
| 21727 | m.impl("log_sigmoid_backward.grad_input", TORCH_FN(functionalization::log_sigmoid_backward_out_grad_input)); |
| 21728 | m.impl("softplus.out", TORCH_FN(functionalization::softplus_out_out)); |
| 21729 | m.impl("adaptive_avg_pool2d", static_cast<at::Tensor (*)(const at::Tensor & self, c10::SymIntArrayRef output_size)>(at::native::adaptive_avg_pool2d_symint)); |
| 21730 | m.impl("adaptive_avg_pool2d.out", TORCH_FN(functionalization::adaptive_avg_pool2d_out_out)); |
| 21731 | m.impl("_adaptive_avg_pool3d_backward.out", TORCH_FN(functionalization::_adaptive_avg_pool3d_backward_out_out)); |
| 21732 | m.impl("adaptive_max_pool2d_backward.grad_input", TORCH_FN(functionalization::adaptive_max_pool2d_backward_out_grad_input)); |
| 21733 | m.impl("fractional_max_pool2d_backward.grad_input", TORCH_FN(functionalization::fractional_max_pool2d_backward_out_grad_input)); |
| 21734 | m.impl("fractional_max_pool3d_backward.grad_input", TORCH_FN(functionalization::fractional_max_pool3d_backward_out_grad_input)); |
| 21735 | m.impl("max_pool3d_with_indices.out", TORCH_FN(functionalization::max_pool3d_with_indices_out_out)); |
| 21736 | m.impl("max_pool3d_with_indices_backward.grad_input", TORCH_FN(functionalization::max_pool3d_with_indices_backward_out_grad_input)); |
| 21737 | m.impl("max_unpool2d.out", TORCH_FN(functionalization::max_unpool2d_out_out)); |
| 21738 | m.impl("reflection_pad2d_backward.grad_input", TORCH_FN(functionalization::reflection_pad2d_backward_out_grad_input)); |
| 21739 | m.impl("upsample_bilinear2d.out", TORCH_FN(functionalization::upsample_bilinear2d_out_out)); |
| 21740 | m.impl("upsample_bilinear2d_backward.grad_input", TORCH_FN(functionalization::upsample_bilinear2d_backward_out_grad_input)); |
| 21741 | m.impl("_upsample_bilinear2d_aa.out", TORCH_FN(functionalization::_upsample_bilinear2d_aa_out_out)); |
| 21742 | m.impl("upsample_trilinear3d_backward.grad_input", TORCH_FN(functionalization::upsample_trilinear3d_backward_out_grad_input)); |
| 21743 | m.impl("_upsample_nearest_exact3d.out", TORCH_FN(functionalization::_upsample_nearest_exact3d_out_out)); |
| 21744 | m.impl("upsample_nearest3d_backward.grad_input", TORCH_FN(functionalization::upsample_nearest3d_backward_out_grad_input)); |
| 21745 | m.impl("logit_backward.grad_input", TORCH_FN(functionalization::logit_backward_out_grad_input)); |
| 21746 | m.impl("thnn_conv2d", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding)>(at::native::thnn_conv2d)); |
| 21747 | m.impl("thnn_conv2d.out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::Tensor & out)>(at::native::thnn_conv2d_out)); |
| 21748 | m.impl("_slow_conv2d_backward.output_mask_out", TORCH_FN(functionalization::_slow_conv2d_backward_out_output_mask_out)); |
| 21749 | m.impl("slow_conv3d", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding)>(at::native::slow_conv3d)); |
| 21750 | m.impl("slow_conv3d.out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Tensor & weight, at::IntArrayRef kernel_size, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::Tensor & out)>(at::native::slow_conv3d_out)); |
| 21751 | m.impl("slow_conv3d_forward.output", TORCH_FN(functionalization::slow_conv3d_forward_out_output)); |
| 21752 | m.impl("slow_conv_dilated3d.out", TORCH_FN(functionalization::slow_conv_dilated3d_out_out)); |
| 21753 | m.impl("special_log_ndtr.out", TORCH_FN(functionalization::special_log_ndtr_out_out)); |
| 21754 | m.impl("special_exp2", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::special_exp2)); |
| 21755 | m.impl("special_exp2.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Tensor & out)>(at::native::special_exp2_out)); |
| 21756 | m.impl("special_digamma", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::special_digamma)); |
| 21757 | m.impl("special_digamma.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Tensor & out)>(at::native::special_digamma_out)); |
| 21758 | m.impl("special_gammaln", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::special_gammaln)); |
| 21759 | m.impl("special_gammaln.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::Tensor & out)>(at::native::special_gammaln_out)); |
| 21760 | m.impl("special_erfcx.out", TORCH_FN(functionalization::special_erfcx_out_out)); |
| 21761 | m.impl("special_xlog1py.out", TORCH_FN(functionalization::special_xlog1py_out_out)); |
| 21762 | m.impl("special_xlog1py.self_scalar_out", TORCH_FN(functionalization::special_xlog1py_out_self_scalar_out)); |
| 21763 | m.impl("special_xlog1py.other_scalar_out", TORCH_FN(functionalization::special_xlog1py_out_other_scalar_out)); |
| 21764 | m.impl("special_i1.out", TORCH_FN(functionalization::special_i1_out_out)); |
| 21765 | m.impl("special_i1e.out", TORCH_FN(functionalization::special_i1e_out_out)); |
| 21766 | m.impl("fft_fft", static_cast<at::Tensor (*)(const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm)>(at::native::fft_fft)); |
| 21767 | m.impl("fft_fft.out", static_cast<at::Tensor & (*)(const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm, at::Tensor & out)>(at::native::fft_fft_out)); |
| 21768 | m.impl("fft_rfft", static_cast<at::Tensor (*)(const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm)>(at::native::fft_rfft)); |
| 21769 | m.impl("fft_rfft.out", static_cast<at::Tensor & (*)(const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm, at::Tensor & out)>(at::native::fft_rfft_out)); |
| 21770 | m.impl("fft_hfft", static_cast<at::Tensor (*)(const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm)>(at::native::fft_hfft)); |
| 21771 | m.impl("fft_hfft.out", static_cast<at::Tensor & (*)(const at::Tensor & self, c10::optional<int64_t> n, int64_t dim, c10::optional<c10::string_view> norm, at::Tensor & out)>(at::native::fft_hfft_out)); |
| 21772 | m.impl("fft_hfft2", static_cast<at::Tensor (*)(const at::Tensor & self, at::OptionalIntArrayRef s, at::IntArrayRef dim, c10::optional<c10::string_view> norm)>(at::native::fft_hfft2)); |
| 21773 | m.impl("fft_hfft2.out", static_cast<const at::Tensor & (*)(const at::Tensor & self, at::OptionalIntArrayRef s, at::IntArrayRef dim, c10::optional<c10::string_view> norm, const at::Tensor & out)>(at::native::fft_hfft2_out)); |
| 21774 | m.impl("fft_ifftn", static_cast<at::Tensor (*)(const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm)>(at::native::fft_ifftn)); |
| 21775 | m.impl("fft_ifftn.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm, at::Tensor & out)>(at::native::fft_ifftn_out)); |
| 21776 | m.impl("fft_rfftn", static_cast<at::Tensor (*)(const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm)>(at::native::fft_rfftn)); |
| 21777 | m.impl("fft_rfftn.out", static_cast<at::Tensor & (*)(const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm, at::Tensor & out)>(at::native::fft_rfftn_out)); |
| 21778 | m.impl("fft_hfftn", static_cast<at::Tensor (*)(const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm)>(at::native::fft_hfftn)); |
| 21779 | m.impl("fft_hfftn.out", static_cast<const at::Tensor & (*)(const at::Tensor & self, at::OptionalIntArrayRef s, at::OptionalIntArrayRef dim, c10::optional<c10::string_view> norm, const at::Tensor & out)>(at::native::fft_hfftn_out)); |
| 21780 | m.impl("fft_fftfreq.out", TORCH_FN(functionalization::fft_fftfreq_out_out)); |
| 21781 | m.impl("linalg_det", static_cast<at::Tensor (*)(const at::Tensor & A)>(at::native::linalg_det)); |
| 21782 | m.impl("linalg_det.out", static_cast<at::Tensor & (*)(const at::Tensor & A, at::Tensor & out)>(at::native::linalg_det_out)); |
| 21783 | m.impl("linalg_ldl_factor_ex.out", TORCH_FN(functionalization::linalg_ldl_factor_ex_out_out)); |
| 21784 | m.impl("linalg_lstsq.out", TORCH_FN(functionalization::linalg_lstsq_out_out)); |
| 21785 | m.impl("linalg_matrix_exp.out", TORCH_FN(functionalization::linalg_matrix_exp_out_out)); |
| 21786 | m.impl("linalg_slogdet", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & A)>(at::native::linalg_slogdet)); |
| 21787 | m.impl("linalg_slogdet.out", static_cast<::std::tuple<at::Tensor &,at::Tensor &> (*)(const at::Tensor & A, at::Tensor & sign, at::Tensor & logabsdet)>(at::native::linalg_slogdet_out)); |
| 21788 | m.impl("_linalg_eigh.eigenvalues", TORCH_FN(functionalization::_linalg_eigh_out_eigenvalues)); |
| 21789 | m.impl("inner", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & other)>(at::native::inner)); |
| 21790 | m.impl("inner.out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Tensor & other, at::Tensor & out)>(at::native::inner_out)); |
| 21791 | m.impl("linalg_matrix_norm", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Scalar & ord, at::IntArrayRef dim, bool keepdim, c10::optional<at::ScalarType> dtype)>(at::native::linalg_matrix_norm)); |
| 21792 | m.impl("linalg_matrix_norm.out", static_cast<at::Tensor & (*)(const at::Tensor & self, const at::Scalar & ord, at::IntArrayRef dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out)>(at::native::linalg_matrix_norm_out)); |
| 21793 | m.impl("linalg_matrix_norm.str_ord", static_cast<at::Tensor (*)(const at::Tensor & self, c10::string_view ord, at::IntArrayRef dim, bool keepdim, c10::optional<at::ScalarType> dtype)>(at::native::linalg_matrix_norm)); |
| 21794 | m.impl("linalg_matrix_norm.str_ord_out", static_cast<at::Tensor & (*)(const at::Tensor & self, c10::string_view ord, at::IntArrayRef dim, bool keepdim, c10::optional<at::ScalarType> dtype, at::Tensor & out)>(at::native::linalg_matrix_norm_out)); |
| 21795 | m.impl("linalg_tensorinv", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t ind)>(at::native::linalg_tensorinv)); |
| 21796 | m.impl("linalg_tensorinv.out", static_cast<at::Tensor & (*)(const at::Tensor & self, int64_t ind, at::Tensor & out)>(at::native::linalg_tensorinv_out)); |
| 21797 | m.impl("linalg_matrix_power", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t n)>(at::native::linalg_matrix_power)); |
| 21798 | m.impl("linalg_matrix_power.out", static_cast<at::Tensor & (*)(const at::Tensor & self, int64_t n, at::Tensor & out)>(at::native::linalg_matrix_power_out)); |
| 21799 | m.impl("_make_dual_copy.out", TORCH_FN(functionalization::_make_dual_copy_out_out)); |
| 21800 | m.impl("view_as_real_copy.out", TORCH_FN(functionalization::view_as_real_copy_out_out)); |
| 21801 | m.impl("view_as_complex_copy.out", TORCH_FN(functionalization::view_as_complex_copy_out_out)); |
| 21802 | m.impl("_conj_copy.out", TORCH_FN(functionalization::_conj_copy_out_out)); |
| 21803 | m.impl("_neg_view_copy.out", TORCH_FN(functionalization::_neg_view_copy_out_out)); |
| 21804 | m.impl("permute_copy.out", TORCH_FN(functionalization::permute_copy_out_out)); |
| 21805 | m.impl("split_copy.Tensor_out", TORCH_FN(functionalization::split_copy_out_Tensor_out)); |
| 21806 | m.impl("t_copy.out", TORCH_FN(functionalization::t_copy_out_out)); |
| 21807 | m.impl("ccol_indices_copy.out", TORCH_FN(functionalization::ccol_indices_copy_out_out)); |
| 21808 | m.impl("unfold_copy.out", TORCH_FN(functionalization::unfold_copy_out_out)); |
| 21809 | m.impl("_transformer_encoder_layer_fwd.out", TORCH_FN(functionalization::_transformer_encoder_layer_fwd_out_out)); |
| 21810 | m.impl("_native_multi_head_attention.out", TORCH_FN(functionalization::_native_multi_head_attention_out_out)); |
| 21811 | m.impl("_transformer_decoder_only_layer_fwd.out", TORCH_FN(functionalization::_transformer_decoder_only_layer_fwd_out_out)); |
| 21812 | m.impl("_native_decoder_only_multi_head_attention.out", TORCH_FN(functionalization::_native_decoder_only_multi_head_attention_out_out)); |
| 21813 | m.impl("special_modified_bessel_i1.out", TORCH_FN(functionalization::special_modified_bessel_i1_out_out)); |
| 21814 | m.impl("special_modified_bessel_k0.out", TORCH_FN(functionalization::special_modified_bessel_k0_out_out)); |
| 21815 | m.impl("_foobar.out", TORCH_FN(functionalization::_foobar_out_out)); |
| 21816 | m.impl("view_as_real", TORCH_FN(functionalization::view_as_real)); |
| 21817 | m.impl("broadcast_to", static_cast<at::Tensor (*)(const at::Tensor & self, c10::SymIntArrayRef size)>(at::native::broadcast_to_symint)); |
| 21818 | m.impl("contiguous", static_cast<at::Tensor (*)(const at::Tensor & self, at::MemoryFormat memory_format)>(at::native::contiguous)); |
| 21819 | m.impl("flatten.using_ints", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t start_dim, int64_t end_dim)>(at::native::flatten)); |
| 21820 | m.impl("flatten.named_out_dim", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t start_dim, int64_t end_dim, at::Dimname out_dim)>(at::native::flatten)); |
| 21821 | m.impl("flatten.using_names", static_cast<at::Tensor (*)(const at::Tensor & self, at::Dimname start_dim, at::Dimname end_dim, at::Dimname out_dim)>(at::native::flatten)); |
| 21822 | m.impl("flatten.DimnameList", static_cast<at::Tensor (*)(const at::Tensor & self, at::DimnameList dims, at::Dimname out_dim)>(at::native::flatten)); |
| 21823 | m.impl("moveaxis.intlist", static_cast<at::Tensor (*)(const at::Tensor & self, at::IntArrayRef source, at::IntArrayRef destination)>(at::native::moveaxis)); |
| 21824 | m.impl("moveaxis.int", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t source, int64_t destination)>(at::native::moveaxis)); |
| 21825 | m.impl("mT", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::mT)); |
| 21826 | m.impl("ravel", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::ravel)); |
| 21827 | m.impl("reshape", static_cast<at::Tensor (*)(const at::Tensor & self, c10::SymIntArrayRef shape)>(at::native::reshape_symint)); |
| 21828 | m.impl("slice.Tensor", TORCH_FN(functionalization::slice_Tensor)); |
| 21829 | m.impl("hsplit.int", static_cast<::std::vector<at::Tensor> (*)(const at::Tensor & self, int64_t sections)>(at::native::hsplit)); |
| 21830 | m.impl("hsplit.array", static_cast<::std::vector<at::Tensor> (*)(const at::Tensor & self, at::IntArrayRef indices)>(at::native::hsplit)); |
| 21831 | m.impl("t", TORCH_FN(functionalization::t)); |
| 21832 | m.impl("t_", TORCH_FN(functionalization::t_)); |
| 21833 | m.impl("coalesce", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::coalesce)); |
| 21834 | m.impl("_indices", TORCH_FN(functionalization::_indices)); |
| 21835 | m.impl("indices", TORCH_FN(functionalization::indices)); |
| 21836 | m.impl("crow_indices", TORCH_FN(functionalization::crow_indices)); |
| 21837 | m.impl("col_indices", TORCH_FN(functionalization::col_indices)); |
| 21838 | m.impl("_autocast_to_reduced_precision", static_cast<at::Tensor (*)(const at::Tensor & self, bool cuda_enabled, bool cpu_enabled, at::ScalarType cuda_dtype, at::ScalarType cpu_dtype)>(at::native::_autocast_to_reduced_precision)); |
| 21839 | m.impl("swapaxes", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t axis0, int64_t axis1)>(at::native::swapaxes)); |
| 21840 | m.impl("swapaxes_", static_cast<at::Tensor & (*)(at::Tensor & self, int64_t axis0, int64_t axis1)>(at::native::swapaxes_)); |
| 21841 | m.impl("unfold", TORCH_FN(functionalization::unfold)); |
| 21842 | m.impl("_cast_Byte", static_cast<at::Tensor (*)(const at::Tensor & self, bool non_blocking)>(at::native::_cast_Byte)); |
| 21843 | m.impl("_cast_Float", static_cast<at::Tensor (*)(const at::Tensor & self, bool non_blocking)>(at::native::_cast_Float)); |
| 21844 | m.impl("is_leaf", static_cast<bool (*)(const at::Tensor & self)>(at::native::is_leaf)); |
| 21845 | m.impl("_reshape_from_tensor", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & shape)>(at::native::_reshape_from_tensor)); |
| 21846 | m.impl("dropout", static_cast<at::Tensor (*)(const at::Tensor & input, double p, bool train)>(at::native::dropout)); |
| 21847 | m.impl("dropout_", static_cast<at::Tensor & (*)(at::Tensor & self, double p, bool train)>(at::native::dropout_)); |
| 21848 | m.impl("adaptive_max_pool1d", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & self, at::IntArrayRef output_size)>(at::native::adaptive_max_pool1d)); |
| 21849 | m.impl("atleast_2d", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::atleast_2d)); |
| 21850 | m.impl("atleast_2d.Sequence", static_cast<::std::vector<at::Tensor> (*)(at::TensorList tensors)>(at::native::atleast_2d)); |
| 21851 | m.impl("_batch_norm_impl_index_backward", static_cast<::std::tuple<at::Tensor,at::Tensor,at::Tensor> (*)(int64_t impl_index, const at::Tensor & input, const at::Tensor & grad_output, const c10::optional<at::Tensor> & weight, const c10::optional<at::Tensor> & running_mean, const c10::optional<at::Tensor> & running_var, const c10::optional<at::Tensor> & save_mean, const c10::optional<at::Tensor> & save_var_transform, bool train, double eps, ::std::array<bool,3> output_mask, const at::Tensor & reservedSpace)>(at::native::_batch_norm_impl_index_backward)); |
| 21852 | m.impl("_convolution_double_backward", static_cast<::std::tuple<at::Tensor,at::Tensor,at::Tensor> (*)(const c10::optional<at::Tensor> & ggI, const c10::optional<at::Tensor> & ggW, const c10::optional<at::Tensor> & ggb, const at::Tensor & gO, const at::Tensor & weight, const at::Tensor & self, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, bool transposed, at::IntArrayRef output_padding, int64_t groups, ::std::array<bool,3> output_mask)>(at::native::_convolution_double_backward)); |
| 21853 | m.impl("conv1d", static_cast<at::Tensor (*)(const at::Tensor & input, const at::Tensor & weight, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, int64_t groups)>(at::native::conv1d)); |
| 21854 | m.impl("conv1d.padding", static_cast<at::Tensor (*)(const at::Tensor & input, const at::Tensor & weight, const c10::optional<at::Tensor> & bias, at::IntArrayRef stride, c10::string_view padding, at::IntArrayRef dilation, int64_t groups)>(at::native::conv1d)); |
| 21855 | m.impl("corrcoef", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::corrcoef)); |
| 21856 | m.impl("cummaxmin_backward", static_cast<at::Tensor (*)(const at::Tensor & grad, const at::Tensor & input, const at::Tensor & indices, int64_t dim)>(at::native::cummaxmin_backward)); |
| 21857 | m.impl("cumprod_backward", static_cast<at::Tensor (*)(const at::Tensor & grad, const at::Tensor & input, int64_t dim, const at::Tensor & output)>(at::native::cumprod_backward)); |
| 21858 | m.impl("embedding_sparse_backward", static_cast<at::Tensor (*)(const at::Tensor & grad, const at::Tensor & indices, int64_t num_weights, int64_t padding_idx, bool scale_grad_by_freq)>(at::native::embedding_sparse_backward)); |
| 21859 | m.impl("_rowwise_prune", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & weight, const at::Tensor & mask, at::ScalarType compressed_indices_dtype)>(at::native::_rowwise_prune)); |
| 21860 | m.impl("_embedding_bag_sparse_backward", static_cast<at::Tensor (*)(const at::Tensor & grad, const at::Tensor & indices, const at::Tensor & offsets, const at::Tensor & offset2bag, const at::Tensor & bag_size, c10::SymInt num_weights, bool scale_grad_by_freq, int64_t mode, const c10::optional<at::Tensor> & per_sample_weights, int64_t padding_idx)>(at::native::_embedding_bag_sparse_backward_symint)); |
| 21861 | m.impl("floor_divide.Scalar", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Scalar & other)>(at::native::floor_divide)); |
| 21862 | m.impl("floor_divide_.Scalar", static_cast<at::Tensor & (*)(at::Tensor & self, const at::Scalar & other)>(at::native::floor_divide_)); |
| 21863 | m.impl("group_norm", static_cast<at::Tensor (*)(const at::Tensor & input, int64_t num_groups, const c10::optional<at::Tensor> & weight, const c10::optional<at::Tensor> & bias, double eps, bool cudnn_enabled)>(at::native::group_norm)); |
| 21864 | m.impl("is_distributed", static_cast<bool (*)(const at::Tensor & self)>(at::native::is_distributed)); |
| 21865 | m.impl("is_neg", static_cast<bool (*)(const at::Tensor & self)>(at::native::is_neg)); |
| 21866 | m.impl("is_signed", static_cast<bool (*)(const at::Tensor & self)>(at::native::is_signed)); |
| 21867 | m.impl("fbgemm_linear_fp16_weight_fp32_activation", static_cast<at::Tensor (*)(const at::Tensor & input, const at::Tensor & packed_weight, const at::Tensor & bias)>(at::native::fbgemm_linear_fp16_weight_fp32_activation)); |
| 21868 | m.impl("matrix_exp_backward", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & grad)>(at::native::matrix_exp_backward)); |
| 21869 | m.impl("max_pool2d", static_cast<at::Tensor (*)(const at::Tensor & self, at::IntArrayRef kernel_size, at::IntArrayRef stride, at::IntArrayRef padding, at::IntArrayRef dilation, bool ceil_mode)>(at::native::max_pool2d)); |
| 21870 | m.impl("cosine_similarity", static_cast<at::Tensor (*)(const at::Tensor & x1, const at::Tensor & x2, int64_t dim, double eps)>(at::native::cosine_similarity)); |
| 21871 | m.impl("native_channel_shuffle", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t groups)>(at::native::math_channel_shuffle)); |
| 21872 | m.impl("pinverse", static_cast<at::Tensor (*)(const at::Tensor & self, double rcond)>(at::native::pinverse)); |
| 21873 | m.impl("rrelu", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Scalar & lower, const at::Scalar & upper, bool training, c10::optional<at::Generator> generator)>(at::native::rrelu)); |
| 21874 | m.impl("rrelu_", static_cast<at::Tensor & (*)(at::Tensor & self, const at::Scalar & lower, const at::Scalar & upper, bool training, c10::optional<at::Generator> generator)>(at::native::rrelu_)); |
| 21875 | m.impl("prelu", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & weight)>(at::native::prelu)); |
| 21876 | m.impl("softmax.Dimname", static_cast<at::Tensor (*)(const at::Tensor & self, at::Dimname dim, c10::optional<at::ScalarType> dtype)>(at::native::softmax)); |
| 21877 | m.impl("stft", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t n_fft, c10::optional<int64_t> hop_length, c10::optional<int64_t> win_length, const c10::optional<at::Tensor> & window, bool normalized, c10::optional<bool> onesided, c10::optional<bool> return_complex)>(at::native::stft)); |
| 21878 | m.impl("stft.center", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t n_fft, c10::optional<int64_t> hop_length, c10::optional<int64_t> win_length, const c10::optional<at::Tensor> & window, bool center, c10::string_view pad_mode, bool normalized, c10::optional<bool> onesided, c10::optional<bool> return_complex)>(at::native::stft)); |
| 21879 | m.impl("tile", static_cast<at::Tensor (*)(const at::Tensor & self, at::IntArrayRef dims)>(at::native::tile)); |
| 21880 | m.impl("trapezoid.x", static_cast<at::Tensor (*)(const at::Tensor & y, const at::Tensor & x, int64_t dim)>(at::native::trapezoid)); |
| 21881 | m.impl("trapezoid.dx", static_cast<at::Tensor (*)(const at::Tensor & y, const at::Scalar & dx, int64_t dim)>(at::native::trapezoid)); |
| 21882 | m.impl("trapz.x", static_cast<at::Tensor (*)(const at::Tensor & y, const at::Tensor & x, int64_t dim)>(at::native::trapz)); |
| 21883 | m.impl("trapz.dx", static_cast<at::Tensor (*)(const at::Tensor & y, double dx, int64_t dim)>(at::native::trapz)); |
| 21884 | m.impl("_has_compatible_shallow_copy_type", static_cast<bool (*)(const at::Tensor & self, const at::Tensor & from)>(at::native::_has_compatible_shallow_copy_type)); |
| 21885 | m.impl("vander", static_cast<at::Tensor (*)(const at::Tensor & x, c10::optional<int64_t> N, bool increasing)>(at::native::vander)); |
| 21886 | m.impl("where.ScalarSelf", static_cast<at::Tensor (*)(const at::Tensor & condition, const at::Scalar & self, const at::Tensor & other)>(at::native::where)); |
| 21887 | m.impl("where.ScalarOther", static_cast<at::Tensor (*)(const at::Tensor & condition, const at::Tensor & self, const at::Scalar & other)>(at::native::where)); |
| 21888 | m.impl("where.Scalar", static_cast<at::Tensor (*)(const at::Tensor & condition, const at::Scalar & self, const at::Scalar & other)>(at::native::where)); |
| 21889 | m.impl("where", static_cast<::std::vector<at::Tensor> (*)(const at::Tensor & condition)>(at::native::where)); |
| 21890 | m.impl("_weight_norm", static_cast<at::Tensor (*)(const at::Tensor & v, const at::Tensor & g, int64_t dim)>(at::native::_weight_norm)); |
| 21891 | m.impl("_weight_norm_differentiable_backward", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & grad_w, const at::Tensor & saved_v, const at::Tensor & saved_g, const at::Tensor & saved_norms, int64_t dim)>(at::native::_weight_norm_differentiable_backward)); |
| 21892 | m.impl("_sparse_sum", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::_sparse_sum)); |
| 21893 | m.impl("_sparse_sum.dtype", static_cast<at::Tensor (*)(const at::Tensor & self, at::ScalarType dtype)>(at::native::_sparse_sum)); |
| 21894 | m.impl("_sparse_sum.dim_dtype", static_cast<at::Tensor (*)(const at::Tensor & self, at::IntArrayRef dim, at::ScalarType dtype)>(at::native::_sparse_sum)); |
| 21895 | m.impl("_sparse_softmax.int", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t dim, c10::optional<at::ScalarType> dtype)>(at::native::_sparse_softmax)); |
| 21896 | m.impl("_sparse_softmax.Dimname", static_cast<at::Tensor (*)(const at::Tensor & self, at::Dimname dim, c10::optional<at::ScalarType> dtype)>(at::native::_sparse_softmax)); |
| 21897 | m.impl("sparse_csr_tensor.crow_col_value_size", static_cast<at::Tensor (*)(const at::Tensor & crow_indices, const at::Tensor & col_indices, const at::Tensor & values, at::IntArrayRef size, c10::optional<at::ScalarType> dtype, c10::optional<at::Layout> layout, c10::optional<at::Device> device, c10::optional<bool> pin_memory)>(at::native::sparse_csr_tensor)); |
| 21898 | m.impl("sparse_bsc_tensor.ccol_row_value_size", static_cast<at::Tensor (*)(const at::Tensor & ccol_indices, const at::Tensor & row_indices, const at::Tensor & values, at::IntArrayRef size, c10::optional<at::ScalarType> dtype, c10::optional<at::Layout> layout, c10::optional<at::Device> device, c10::optional<bool> pin_memory)>(at::native::sparse_bsc_tensor)); |
| 21899 | m.impl("sparse_csr_tensor.crow_col_value", static_cast<at::Tensor (*)(const at::Tensor & crow_indices, const at::Tensor & col_indices, const at::Tensor & values, c10::optional<at::ScalarType> dtype, c10::optional<at::Layout> layout, c10::optional<at::Device> device, c10::optional<bool> pin_memory)>(at::native::sparse_csr_tensor)); |
| 21900 | m.impl("sparse_bsc_tensor.ccol_row_value", static_cast<at::Tensor (*)(const at::Tensor & ccol_indices, const at::Tensor & row_indices, const at::Tensor & values, c10::optional<at::ScalarType> dtype, c10::optional<at::Layout> layout, c10::optional<at::Device> device, c10::optional<bool> pin_memory)>(at::native::sparse_bsc_tensor)); |
| 21901 | m.impl("_sparse_csr_tensor_unsafe", static_cast<at::Tensor (*)(const at::Tensor & crow_indices, const at::Tensor & col_indices, const at::Tensor & values, at::IntArrayRef size, c10::optional<at::ScalarType> dtype, c10::optional<at::Layout> layout, c10::optional<at::Device> device, c10::optional<bool> pin_memory)>(at::native::_sparse_csr_tensor_unsafe)); |
| 21902 | m.impl("_sparse_bsc_tensor_unsafe", static_cast<at::Tensor (*)(const at::Tensor & ccol_indices, const at::Tensor & row_indices, const at::Tensor & values, at::IntArrayRef size, c10::optional<at::ScalarType> dtype, c10::optional<at::Layout> layout, c10::optional<at::Device> device, c10::optional<bool> pin_memory)>(at::native::_sparse_bsc_tensor_unsafe)); |
| 21903 | m.impl("_validate_sparse_bsr_tensor_args", static_cast<void (*)(const at::Tensor & crow_indices, const at::Tensor & col_indices, const at::Tensor & values, at::IntArrayRef size)>(at::native::_validate_sparse_bsr_tensor_args)); |
| 21904 | m.impl("_saturate_weight_to_fp16", static_cast<at::Tensor (*)(const at::Tensor & weight)>(at::native::_saturate_weight_to_fp16)); |
| 21905 | m.impl("item", static_cast<at::Scalar (*)(const at::Tensor & self)>(at::native::item)); |
| 21906 | m.impl("_thnn_fused_lstm_cell_backward", static_cast<::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> (*)(const c10::optional<at::Tensor> & grad_hy, const c10::optional<at::Tensor> & grad_cy, const at::Tensor & cx, const at::Tensor & cy, const at::Tensor & workspace, bool has_bias)>(at::native::_thnn_fused_lstm_cell_backward)); |
| 21907 | m.impl("_thnn_differentiable_lstm_cell_backward", static_cast<::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> (*)(const c10::optional<at::Tensor> & grad_hy, const c10::optional<at::Tensor> & grad_cy, const at::Tensor & input_gates, const at::Tensor & hidden_gates, const c10::optional<at::Tensor> & input_bias, const c10::optional<at::Tensor> & hidden_bias, const at::Tensor & cx, const at::Tensor & cy)>(at::native::_thnn_differentiable_lstm_cell_backward)); |
| 21908 | m.impl("rnn_relu.input", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & input, const at::Tensor & hx, at::TensorList params, bool has_biases, int64_t num_layers, double dropout, bool train, bool bidirectional, bool batch_first)>(at::native::rnn_relu)); |
| 21909 | m.impl("rnn_relu.data", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & data, const at::Tensor & batch_sizes, const at::Tensor & hx, at::TensorList params, bool has_biases, int64_t num_layers, double dropout, bool train, bool bidirectional)>(at::native::rnn_relu)); |
| 21910 | m.impl("quantized_lstm_cell", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & input, at::TensorList hx, const at::Tensor & w_ih, const at::Tensor & w_hh, const at::Tensor & b_ih, const at::Tensor & b_hh, const at::Tensor & packed_ih, const at::Tensor & packed_hh, const at::Tensor & col_offsets_ih, const at::Tensor & col_offsets_hh, const at::Scalar & scale_ih, const at::Scalar & scale_hh, const at::Scalar & zero_point_ih, const at::Scalar & zero_point_hh)>(at::native::quantized_lstm_cell)); |
| 21911 | m.impl("quantized_rnn_relu_cell", static_cast<at::Tensor (*)(const at::Tensor & input, const at::Tensor & hx, const at::Tensor & w_ih, const at::Tensor & w_hh, const at::Tensor & b_ih, const at::Tensor & b_hh, const at::Tensor & packed_ih, const at::Tensor & packed_hh, const at::Tensor & col_offsets_ih, const at::Tensor & col_offsets_hh, const at::Scalar & scale_ih, const at::Scalar & scale_hh, const at::Scalar & zero_point_ih, const at::Scalar & zero_point_hh)>(at::native::quantized_rnn_relu_cell)); |
| 21912 | m.impl("_pack_padded_sequence_backward", static_cast<at::Tensor (*)(const at::Tensor & grad, c10::SymIntArrayRef input_size, const at::Tensor & batch_sizes, bool batch_first)>(at::native::_pack_padded_sequence_backward_symint)); |
| 21913 | m.impl("histogramdd", static_cast<::std::tuple<at::Tensor,::std::vector<at::Tensor>> (*)(const at::Tensor & self, at::IntArrayRef bins, c10::optional<at::ArrayRef<double>> range, const c10::optional<at::Tensor> & weight, bool density)>(at::native::histogramdd)); |
| 21914 | m.impl("histogramdd.int_bins", static_cast<::std::tuple<at::Tensor,::std::vector<at::Tensor>> (*)(const at::Tensor & self, int64_t bins, c10::optional<at::ArrayRef<double>> range, const c10::optional<at::Tensor> & weight, bool density)>(at::native::histogramdd)); |
| 21915 | m.impl("histogramdd.TensorList_bins", static_cast<::std::tuple<at::Tensor,::std::vector<at::Tensor>> (*)(const at::Tensor & self, at::TensorList bins, c10::optional<at::ArrayRef<double>> range, const c10::optional<at::Tensor> & weight, bool density)>(at::native::histogramdd)); |
| 21916 | m.impl("l1_loss", static_cast<at::Tensor (*)(const at::Tensor & self, const at::Tensor & target, int64_t reduction)>(at::native::l1_loss)); |
| 21917 | m.impl("upsample_bilinear2d.vec", static_cast<at::Tensor (*)(const at::Tensor & input, at::OptionalIntArrayRef output_size, bool align_corners, c10::optional<at::ArrayRef<double>> scale_factors)>(at::native::upsample_bilinear2d)); |
| 21918 | m.impl("_upsample_bilinear2d_aa.vec", static_cast<at::Tensor (*)(const at::Tensor & input, at::OptionalIntArrayRef output_size, bool align_corners, c10::optional<at::ArrayRef<double>> scale_factors)>(at::native::_upsample_bilinear2d_aa)); |
| 21919 | m.impl("_upsample_nearest_exact3d.vec", static_cast<at::Tensor (*)(const at::Tensor & input, at::OptionalIntArrayRef output_size, c10::optional<at::ArrayRef<double>> scale_factors)>(at::native::_upsample_nearest_exact3d)); |
| 21920 | m.impl("isfinite", static_cast<at::Tensor (*)(const at::Tensor & self)>(at::native::isfinite)); |
| 21921 | m.impl("special_log_softmax", static_cast<at::Tensor (*)(const at::Tensor & self, int64_t dim, c10::optional<at::ScalarType> dtype)>(at::native::special_log_softmax)); |
| 21922 | m.impl("scaled_dot_product_attention", static_cast<at::Tensor (*)(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const c10::optional<at::Tensor> & attn_mask, double dropout_p, bool is_causal)>(at::native::scaled_dot_product_attention)); |
| 21923 | m.impl("_scaled_dot_product_attention_math", static_cast<::std::tuple<at::Tensor,at::Tensor> (*)(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const c10::optional<at::Tensor> & attn_mask, double dropout_p, bool is_causal, const c10::optional<at::Tensor> & dropout_mask)>(at::native::_scaled_dot_product_attention_math));; |
| 21924 | } |
| 21925 | |
| 21926 | } // namespace |
| 21927 | |
| 21928 | } // namespace at |
| 21929 |
Generated on 2023-Feb-12 from project pytorch revision 2c76838d7ff
Powered by 
Code Browser 2.1
Generator usage only permitted with license.