| 1 | namespace dnnl { |
|---|---|
| 2 | namespace impl { |
| 3 | namespace gpu { |
| 4 | namespace ocl { |
| 5 | const char *gen9_gemm_copy_kernel = R"==(/******************************************************************************* )==" "\n" |
| 6 | R"==(* Copyright 2019-2022 Intel Corporation )==" "\n" |
| 7 | R"==(* )==" "\n" |
| 8 | R"==(* Licensed under the Apache License, Version 2.0 (the "License"); )==" "\n" |
| 9 | R"==(* you may not use this file except in compliance with the License. )==" "\n" |
| 10 | R"==(* You may obtain a copy of the License at )==" "\n" |
| 11 | R"==(* )==" "\n" |
| 12 | R"==(* http: )==" "\n" |
| 13 | R"==(* )==" "\n" |
| 14 | R"==(* Unless required by applicable law or agreed to in writing, software )==" "\n" |
| 15 | R"==(* distributed under the License is distributed on an "AS IS" BASIS, )==" "\n" |
| 16 | R"==(* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. )==" "\n" |
| 17 | R"==(* See the License for the specific language governing permissions and )==" "\n" |
| 18 | R"==(* limitations under the License. )==" "\n" |
| 19 | R"==(*******************************************************************************/ )==" "\n" |
| 20 | R"==(#include "gpu/ocl/gemm/ocl_gemm_attrs.h" )==" "\n" |
| 21 | R"==(#include "gpu/ocl/ocl_types.h" )==" "\n" |
| 22 | R"==(__kernel void gen9_gemm_copy(long m, long n, __global SRC_DATA_T *a, )==" "\n" |
| 23 | R"==(long offseta, long lda, __global float *alpha, __global DATA_T *b, )==" "\n" |
| 24 | R"==(long offsetb) { )==" "\n" |
| 25 | R"==(int idx = get_group_id(0); )==" "\n" |
| 26 | R"==(int idy = get_group_id(1) * COPY_UNROLL; )==" "\n" |
| 27 | R"==(int i; )==" "\n" |
| 28 | R"==(#ifdef USE_TRANS )==" "\n" |
| 29 | R"==(offseta += (idy + idx * lda); )==" "\n" |
| 30 | R"==(#else )==" "\n" |
| 31 | R"==(offseta += (idx + idy * lda); )==" "\n" |
| 32 | R"==(#endif )==" "\n" |
| 33 | R"==(offsetb += (idy * m + idx * COPY_UNROLL); )==" "\n" |
| 34 | R"==(n -= idy; )==" "\n" |
| 35 | R"==(for (i = 0; i < COPY_UNROLL; i++) { )==" "\n" |
| 36 | R"==(b[offsetb] )==" "\n" |
| 37 | R"==(= (i < n) ? (ATTR_ALPHA * SRC_TO_REF(a[offseta])) : DATA_ZERO; )==" "\n" |
| 38 | R"==(#ifdef USE_TRANS )==" "\n" |
| 39 | R"==(offseta++; )==" "\n" |
| 40 | R"==(#else )==" "\n" |
| 41 | R"==(offseta += lda; )==" "\n" |
| 42 | R"==(#endif )==" "\n" |
| 43 | R"==(offsetb++; )==" "\n" |
| 44 | R"==(} )==" "\n" |
| 45 | R"==(} )==" "\n" |
| 46 | R"==()=="; |
| 47 | } |
| 48 | } |
| 49 | } |
| 50 | } |
Generated on 2022-Nov-30 from project oneDNN revision 84bc949a1
Powered by 
Code Browser 2.1
Generator usage only permitted with license.