1 | /******************************************************************************* |
2 | * Copyright 2021-2022 Intel Corporation |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. |
6 | * You may obtain a copy of the License at |
7 | * |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * |
10 | * Unless required by applicable law or agreed to in writing, software |
11 | * distributed under the License is distributed on an "AS IS" BASIS, |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
13 | * See the License for the specific language governing permissions and |
14 | * limitations under the License. |
15 | *******************************************************************************/ |
16 | |
17 | #include "gpu/gpu_impl_list.hpp" |
18 | |
19 | #include "gpu/jit/binary_format.hpp" |
20 | |
21 | #include "gpu/jit/conv/gen_convolution.hpp" |
22 | #include "gpu/ocl/gen9_convolution.hpp" |
23 | #include "gpu/ocl/gen9_wino_convolution.hpp" |
24 | #include "gpu/ocl/ref_convolution.hpp" |
25 | #include "gpu/ocl/xe_lp_x8s8x_1x1_convolution.hpp" |
26 | #include "gpu/ocl/xe_lp_x8s8x_convolution.hpp" |
27 | |
28 | namespace dnnl { |
29 | namespace impl { |
30 | namespace gpu { |
31 | |
32 | namespace { |
33 | using namespace dnnl::impl::prop_kind; |
34 | |
35 | // clang-format off |
36 | const std::map<pk_impl_key_t, std::vector<impl_list_item_t>> |
37 | impl_list_map REG_CONV_P({ |
38 | {{forward}, { |
39 | INSTANCE(jit::gen_convolution_fwd_t) |
40 | INSTANCE(ocl::xe_lp_x8s8x_1x1_convolution_fwd_t) |
41 | INSTANCE(ocl::xe_lp_x8s8x_convolution_fwd_t) |
42 | INSTANCE(ocl::gen9_wino_convolution_fwd_t) |
43 | INSTANCE(ocl::gen9_convolution_fwd_t) |
44 | INSTANCE(ocl::ref_convolution_fwd_t) |
45 | nullptr, |
46 | }}, |
47 | {{backward_data}, REG_BWD_D_PK({ |
48 | INSTANCE(jit::gen_convolution_bwd_data_t) |
49 | INSTANCE(ocl::xe_lp_x8s8x_convolution_bwd_data_t) |
50 | INSTANCE(ocl::gen9_convolution_bwd_data_t) |
51 | INSTANCE(ocl::ref_convolution_bwd_data_t) |
52 | nullptr, |
53 | })}, |
54 | {{backward_weights}, REG_BWD_PK({ |
55 | INSTANCE(jit::gen_convolution_bwd_weights_t) |
56 | INSTANCE(ocl::gen9_convolution_bwd_weights_t) |
57 | INSTANCE(ocl::ref_convolution_bwd_weights_t) |
58 | nullptr, |
59 | })}, |
60 | }); |
61 | // clang-format on |
62 | } // namespace |
63 | |
64 | const impl_list_item_t *get_convolution_impl_list( |
65 | const convolution_desc_t *desc) { |
66 | static const impl_list_item_t empty_list[] = {nullptr}; |
67 | |
68 | const bool is_fwd = utils::one_of( |
69 | desc->prop_kind, forward_training, forward_inference); |
70 | prop_kind_t prop_kind = is_fwd ? forward : desc->prop_kind; |
71 | |
72 | const auto impl_list_it = impl_list_map.find({prop_kind}); |
73 | return impl_list_it != impl_list_map.cend() ? impl_list_it->second.data() |
74 | : empty_list; |
75 | } |
76 | |
77 | } // namespace gpu |
78 | } // namespace impl |
79 | } // namespace dnnl |
80 | |