1 | /******************************************************************************* |
2 | * Copyright 2019-2022 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_OCL_REF_REORDER_HPP |
18 | #define GPU_OCL_REF_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/gpu_resource.hpp" |
27 | #include "gpu/ocl/ocl_utils.hpp" |
28 | #include "gpu/primitive_conf.hpp" |
29 | |
30 | namespace dnnl { |
31 | namespace impl { |
32 | namespace gpu { |
33 | namespace ocl { |
34 | |
35 | struct ref_reorder_t : public gpu_primitive_t { |
36 | using gpu_primitive_t::gpu_primitive_t; |
37 | struct pd_t : public gpu_reorder_pd_t { |
38 | using gpu_reorder_pd_t::gpu_reorder_pd_t; |
39 | |
40 | DECLARE_COMMON_PD_T("ocl:ref:any" , ref_reorder_t); |
41 | |
42 | status_t init( |
43 | engine_t *engine, engine_t *src_engine, engine_t *dst_engine) { |
44 | bool ok = src_engine == dst_engine |
45 | && !memory_desc_ndims_ok(src_md(), dst_md()) |
46 | && src_engine->kind() == engine_kind::gpu && attr_ok() |
47 | && extra_ok(); |
48 | if (!ok) return status::unimplemented; |
49 | |
50 | if (memory_desc_wrapper(src_md()).has_runtime_dims_or_strides()) |
51 | return status::unimplemented; |
52 | |
53 | auto *compute_engine = utils::downcast<compute::compute_engine_t *>( |
54 | dst_engine->kind() == engine_kind::gpu ? dst_engine |
55 | : src_engine); |
56 | |
57 | ok = ok |
58 | && compute_engine->mayiuse( |
59 | compute::device_ext_t::intel_subgroups) |
60 | && IMPLICATION( |
61 | utils::one_of(data_type::f16, src_md()->data_type, |
62 | dst_md()->data_type), |
63 | compute_engine->mayiuse( |
64 | compute::device_ext_t::khr_fp16) |
65 | && compute_engine->mayiuse( |
66 | compute::device_ext_t:: |
67 | intel_subgroups_short)) |
68 | && IMPLICATION( |
69 | utils::one_of(data_type::f64, src_md()->data_type, |
70 | dst_md()->data_type), |
71 | compute_engine->mayiuse( |
72 | compute::device_ext_t::khr_fp64) |
73 | && attr()->post_ops_.has_default_values()); |
74 | |
75 | if (!ok) return status::unimplemented; |
76 | |
77 | status_t status = init_conf(engine); |
78 | if (status != status::success) return status; |
79 | init_scratchpad(); |
80 | |
81 | return status::success; |
82 | } |
83 | |
84 | status_t init_conf(engine_t *engine); |
85 | void init_scratchpad(); |
86 | status_t init_kernel_ctx(compute::kernel_ctx_t &kernel_ctx) const; |
87 | |
88 | reorder_conf_t conf; |
89 | |
90 | private: |
91 | DECLARE_GPU_REORDER_CREATE(); |
92 | }; |
93 | |
94 | status_t init(engine_t *engine) override { |
95 | compute::kernel_ctx_t kernel_ctx; |
96 | |
97 | auto status = pd()->init_kernel_ctx(kernel_ctx); |
98 | if (status != status::success) return status; |
99 | |
100 | const auto &conf = pd()->conf; |
101 | if (conf.nelems == 0) return status::success; |
102 | |
103 | create_kernel(engine, &kernel_, "ref_reorder" , kernel_ctx); |
104 | if (!kernel_) return status::runtime_error; |
105 | return status::success; |
106 | } |
107 | |
108 | status_t execute(const exec_ctx_t &ctx) const override; |
109 | |
110 | private: |
111 | const pd_t *pd() const { return (const pd_t *)primitive_t::pd().get(); } |
112 | compute::kernel_t kernel_; |
113 | }; |
114 | |
115 | } // namespace ocl |
116 | } // namespace gpu |
117 | } // namespace impl |
118 | } // namespace dnnl |
119 | |
120 | #endif |
121 | |