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 | #ifndef GPU_combined_REDUCTION_HPP |
18 | #define GPU_combined_REDUCTION_HPP |
19 | |
20 | #include "common/c_types_map.hpp" |
21 | #include "common/primitive.hpp" |
22 | #include "common/type_helpers.hpp" |
23 | #include "common/utils.hpp" |
24 | #include "gpu/compute/compute.hpp" |
25 | #include "gpu/gpu_primitive.hpp" |
26 | #include "gpu/gpu_reduction_pd.hpp" |
27 | #include "gpu/gpu_resource.hpp" |
28 | #include "gpu/primitive_conf.hpp" |
29 | |
30 | #include <iostream> |
31 | |
32 | namespace dnnl { |
33 | namespace impl { |
34 | namespace gpu { |
35 | namespace ocl { |
36 | struct combined_reduction_t : public gpu_primitive_t { |
37 | using gpu_primitive_t::gpu_primitive_t; |
38 | struct pd_t : public gpu_reduction_pd_t { |
39 | using gpu_reduction_pd_t::gpu_reduction_pd_t; |
40 | |
41 | DECLARE_COMMON_PD_T("ocl:combined" , combined_reduction_t); |
42 | |
43 | status_t init(engine_t *engine) { |
44 | using smask_t = primitive_attr_t::skip_mask_t; |
45 | bool ok = set_default_params() == status::success |
46 | && attr_.has_default_values(smask_t::gpu_attr) |
47 | && !memory_desc_ndims_ok(src_md(), dst_md()) |
48 | && post_ops_with_binary_ok(attr(), dst_md()->data_type, 5) |
49 | && attr_.set_default_formats(dst_md(0)) == status::success; |
50 | if (!ok) return status::unimplemented; |
51 | |
52 | CHECK(init_conf(engine)); |
53 | init_scratchpad(); |
54 | |
55 | return status::success; |
56 | } |
57 | |
58 | status_t init_conf(engine_t *engine); |
59 | status_t init_kernel_ctx(compute::kernel_ctx_t &kernel_ctx, |
60 | const reduction_phase_t &phase) const; |
61 | void init_scratchpad(); |
62 | |
63 | reduction_conf_t conf; |
64 | }; |
65 | |
66 | status_t init(engine_t *engine) override { |
67 | auto &phases = pd()->conf.phases; |
68 | |
69 | status_t status; |
70 | for (auto &phase : phases) { |
71 | compute::kernel_ctx_t kernel_ctx(pd()->attr()); |
72 | status = pd()->init_kernel_ctx(kernel_ctx, phase); |
73 | CHECK(status); |
74 | compute::kernel_t kernel; |
75 | status = create_kernel( |
76 | engine, &kernel, "combined_reduce" , kernel_ctx); |
77 | CHECK(status); |
78 | kernels.push_back(kernel); |
79 | } |
80 | |
81 | return status::success; |
82 | } |
83 | |
84 | virtual status_t execute(const exec_ctx_t &ctx) const override { |
85 | return execute_combined(ctx); |
86 | } |
87 | |
88 | private: |
89 | status_t execute_combined(const exec_ctx_t &ctx) const; |
90 | const pd_t *pd() const { return (const pd_t *)primitive_t::pd().get(); } |
91 | |
92 | std::vector<compute::kernel_t> kernels; |
93 | }; |
94 | |
95 | } // namespace ocl |
96 | } // namespace gpu |
97 | } // namespace impl |
98 | } // namespace dnnl |
99 | |
100 | #endif |
101 | |