1 | /******************************************************************************* |
2 | * Copyright 2020-2022 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 | /// @example binary.cpp |
18 | /// > Annotated version: @ref binary_example_cpp |
19 | /// |
20 | /// @page binary_example_cpp_short |
21 | /// |
22 | /// This C++ API example demonstrates how to create and execute a |
23 | /// [Binary](@ref dev_guide_binary) primitive. |
24 | /// |
25 | /// Key optimizations included in this example: |
26 | /// - In-place primitive execution; |
27 | /// - Primitive attributes with fused post-ops. |
28 | /// |
29 | /// @page binary_example_cpp Binary Primitive Example |
30 | /// @copydetails binary_example_cpp_short |
31 | /// |
32 | /// @include binary.cpp |
33 | |
34 | #include <algorithm> |
35 | #include <cmath> |
36 | #include <iostream> |
37 | #include <string> |
38 | #include <vector> |
39 | |
40 | #include "example_utils.hpp" |
41 | #include "oneapi/dnnl/dnnl.hpp" |
42 | |
43 | using namespace dnnl; |
44 | |
45 | using tag = memory::format_tag; |
46 | using dt = memory::data_type; |
47 | |
48 | void binary_example(dnnl::engine::kind engine_kind) { |
49 | |
50 | // Create execution dnnl::engine. |
51 | dnnl::engine engine(engine_kind, 0); |
52 | |
53 | // Create dnnl::stream. |
54 | dnnl::stream engine_stream(engine); |
55 | |
56 | // Tensor dimensions. |
57 | const memory::dim N = 3, // batch size |
58 | IC = 3, // channels |
59 | IH = 150, // tensor height |
60 | IW = 150; // tensor width |
61 | |
62 | // Source (src_0 and src_1) and destination (dst) tensors dimensions. |
63 | memory::dims src_0_dims = {N, IC, IH, IW}; |
64 | memory::dims src_1_dims = {N, IC, IH, 1}; |
65 | |
66 | // Allocate buffers. |
67 | std::vector<float> src_0_data(product(src_0_dims)); |
68 | std::vector<float> src_1_data(product(src_1_dims)); |
69 | |
70 | // Initialize src_0 and src_1 (src). |
71 | std::generate(src_0_data.begin(), src_0_data.end(), []() { |
72 | static int i = 0; |
73 | return std::cos(i++ / 10.f); |
74 | }); |
75 | std::generate(src_1_data.begin(), src_1_data.end(), []() { |
76 | static int i = 0; |
77 | return std::sin(i++ * 2.f); |
78 | }); |
79 | |
80 | // Create src and dst memory descriptors. |
81 | auto src_0_md = memory::desc(src_0_dims, dt::f32, tag::nchw); |
82 | auto src_1_md = memory::desc(src_1_dims, dt::f32, tag::nchw); |
83 | auto dst_md = memory::desc(src_0_dims, dt::f32, tag::nchw); |
84 | |
85 | // Create src memory objects. |
86 | auto src_0_mem = memory(src_0_md, engine); |
87 | auto src_1_mem = memory(src_1_md, engine); |
88 | |
89 | // Write data to memory object's handle. |
90 | write_to_dnnl_memory(src_0_data.data(), src_0_mem); |
91 | write_to_dnnl_memory(src_1_data.data(), src_1_mem); |
92 | |
93 | // Create primitive post-ops (ReLU). |
94 | const float alpha = 0.f; |
95 | const float beta = 0.f; |
96 | post_ops binary_ops; |
97 | binary_ops.append_eltwise(algorithm::eltwise_relu, alpha, beta); |
98 | primitive_attr binary_attr; |
99 | binary_attr.set_post_ops(binary_ops); |
100 | |
101 | // Create primitive descriptor. |
102 | auto binary_pd = binary::primitive_desc(engine, algorithm::binary_mul, |
103 | src_0_md, src_1_md, dst_md, binary_attr); |
104 | |
105 | // Create the primitive. |
106 | auto binary_prim = binary(binary_pd); |
107 | |
108 | // Primitive arguments. Set up in-place execution by assigning src_0 as DST. |
109 | std::unordered_map<int, memory> binary_args; |
110 | binary_args.insert({DNNL_ARG_SRC_0, src_0_mem}); |
111 | binary_args.insert({DNNL_ARG_SRC_1, src_1_mem}); |
112 | binary_args.insert({DNNL_ARG_DST, src_0_mem}); |
113 | |
114 | // Primitive execution: binary with ReLU. |
115 | binary_prim.execute(engine_stream, binary_args); |
116 | |
117 | // Wait for the computation to finalize. |
118 | engine_stream.wait(); |
119 | |
120 | // Read data from memory object's handle. |
121 | read_from_dnnl_memory(src_0_data.data(), src_0_mem); |
122 | } |
123 | |
124 | int main(int argc, char **argv) { |
125 | return handle_example_errors(binary_example, parse_engine_kind(argc, argv)); |
126 | } |
127 | |