1 | /******************************************************************************* |
2 | * Copyright 2019-2021 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 | #ifndef GPU_OCL_CROSS_ENGINE_REORDER_HPP |
18 | #define GPU_OCL_CROSS_ENGINE_REORDER_HPP |
19 | |
20 | #include "common/c_types_map.hpp" |
21 | #include "common/memory.hpp" |
22 | #include "common/primitive.hpp" |
23 | #include "common/utils.hpp" |
24 | #include "gpu/gpu_primitive.hpp" |
25 | #include "gpu/gpu_reorder_pd.hpp" |
26 | #include "gpu/ocl/ocl_utils.hpp" |
27 | |
28 | namespace dnnl { |
29 | namespace impl { |
30 | namespace gpu { |
31 | namespace ocl { |
32 | |
33 | // Cross-engine reorder manages all reorders between GPU and CPU engines. |
34 | // |
35 | // For CPU -> GPU reorder, it includes 2 steps: |
36 | // 1. CPU -> GPU copying |
37 | // 2. GPU reorder |
38 | // |
39 | // For GPU -> CPU reorder, it includes 2 steps: |
40 | // 1. GPU reorder |
41 | // 2. GPU -> CPU copying |
42 | struct cross_engine_reorder_t : public gpu_primitive_t { |
43 | using gpu_primitive_t::gpu_primitive_t; |
44 | struct pd_t : public gpu_reorder_pd_t { |
45 | using gpu_reorder_pd_t::gpu_reorder_pd_t; |
46 | |
47 | DECLARE_COMMON_PD_T("ocl:cross_engine::any" , cross_engine_reorder_t); |
48 | |
49 | status_t init( |
50 | engine_t *engine, engine_t *src_engine, engine_t *dst_engine); |
51 | |
52 | std::shared_ptr<primitive_desc_t> reorder_pd_; |
53 | engine_kind_t reorder_engine_kind_ = engine_kind::gpu; |
54 | bool do_reorder_ = true; |
55 | |
56 | private: |
57 | void init_scratchpad(); |
58 | DECLARE_GPU_REORDER_CREATE(); |
59 | }; |
60 | |
61 | status_t init(engine_t *engine) override { |
62 | if (!pd()->do_reorder_) return status::success; |
63 | return create_nested_primitive(reorder_, pd()->reorder_pd_, engine); |
64 | } |
65 | |
66 | status_t execute(const exec_ctx_t &ctx) const override; |
67 | |
68 | private: |
69 | const pd_t *pd() const { return (const pd_t *)primitive_t::pd().get(); } |
70 | std::shared_ptr<primitive_t> reorder_; |
71 | }; |
72 | |
73 | } // namespace ocl |
74 | } // namespace gpu |
75 | } // namespace impl |
76 | } // namespace dnnl |
77 | |
78 | #endif |
79 | |