tiny_dnn  1.0.0
A header only, dependency-free deep learning framework in C++11
conv2d_op_nnpack.h
1 /*
2  Copyright (c) 2016, Taiga Nomi, Edgar Riba
3  All rights reserved.
4 
5  Redistribution and use in source and binary forms, with or without
6  modification, are permitted provided that the following conditions are met:
7  * Redistributions of source code must retain the above copyright
8  notice, this list of conditions and the following disclaimer.
9  * Redistributions in binary form must reproduce the above copyright
10  notice, this list of conditions and the following disclaimer in the
11  documentation and/or other materials provided with the distribution.
12  * Neither the name of the <organization> nor the
13  names of its contributors may be used to endorse or promote products
14  derived from this software without specific prior written permission.
15 
16  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY
17  EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
19  DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY
20  DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21  (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
22  LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
23  ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 */
27 #pragma once
28 
29 #include "tiny_dnn/core/params/conv_params.h"
30 
31 #ifdef CNN_USE_NNPACK
32 #include "nnpack.h"
33 
34 inline nnp_convolution_algorithm nnp_algorithm() {
35  return nnp_convolution_algorithm_auto;
36 }
37 
38 inline nnp_convolution_transform_strategy nnp_kts() {
39  return nnp_convolution_transform_strategy_tuple_based;//some algorithm accept tuple based only
40 }
41 #endif
42 
43 namespace tiny_dnn {
44 namespace kernels {
45 
46 inline void
47 conv2d_op_nnpack(const tensor_t& in_data,
48  const vec_t& W,
49  const vec_t& bias,
50  tensor_t& out_data,
51  const core::conv_params& params) {
52 #ifdef CNN_USE_NNPACK
53  nnp_status init_status = nnp_initialize();
54  if (init_status != nnp_status_success) {
55  throw nn_error("Cannot initialize NNPACK.");
56  }
57 
58  // TOOD: use input config
59  const auto algorithm = nnp_algorithm();
60  const auto kernel_transform_strategy = nnp_kts();
61 
62  const serial_size_t input_channels = params.in.depth_;
63  const serial_size_t output_channels = params.out.depth_;
64 
65  //input data passed by convolution layer has been padded already
66  //set input_size to padded size
67  const nnp_size input_size = {
68  static_cast<size_t>(params.in_padded.width_),
69  static_cast<size_t>(params.in_padded.height_)
70  };
71 
72  const nnp_size kernel_size = {
73  static_cast<size_t>(params.weight.width_),
74  static_cast<size_t>(params.weight.height_)
75  };
76 
77  // input padded ,so no need to do padding
78  const float_t dx =0;// params.in_padded.width_ - params.in.width_;
79  const float_t dy =0;// params.in_padded.height_ - params.in.height_;
80 
81  // we'll assume that padding is symmetric
82 
83  const nnp_padding padding = {
84  static_cast<size_t>(dy/2), // top
85  static_cast<size_t>(dx/2), // right
86  static_cast<size_t>(dy/2), // bottom
87  static_cast<size_t>(dx/2) // left
88  };
89 
90  const float* input_ptr = reinterpret_cast<const float*>(in_data[0].data());
91  const float* kernel_ptr = reinterpret_cast<const float*>(W.data());
92  const float* bias_ptr = reinterpret_cast<const float*>(bias.data());
93  const nnp_size stride= {
94  static_cast<size_t>(params.w_stride),
95  static_cast<size_t>(params.h_stride)
96  };
97 
98  float* output_ptr = out_data[0].data();
99 
100  // TODO: embed it into a class
101  const size_t num_mkl_threads = 1;
102  pthreadpool_t threadpool = pthreadpool_create(num_mkl_threads);
103 
104  nnp_profile* profile = nullptr;
105 
106  nnp_status status =
107  nnp_convolution_inference(
108  algorithm,
109  kernel_transform_strategy,
110  input_channels,
111  output_channels,
112  input_size,
113  padding,
114  kernel_size,
115  stride,
116  input_ptr,
117  kernel_ptr,
118  bias_ptr,
119  output_ptr,
120  threadpool,
121  profile);
122 
123  if (status != nnp_status_success) {
124  throw nn_error("Could not succeed with nnp_convolution_inference");
125  }
126 
127  // TODO: embed it into a class
128  pthreadpool_destroy(threadpool);
129 #else
130  throw nn_error("TinyDNN has not been compiled with NNPACK support.");
131 #endif
132 }
133 
134 } // namespace kernels
135 } // namespace tiny_dnn