1 /*
2 * Copyright (c) 2017-2020 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24 #include "ConvolutionLayer.h"
25
26 #include "tests/validation/Helpers.h"
27 #include "tests/validation/reference/Convolution3d.h"
28 #include "tests/validation/reference/Permute.h"
29 #include "tests/validation/reference/Utils.h"
30 #include "tests/validation/reference/UtilsQuantizedAsymm.h"
31
32 #include "tests/framework/Asserts.h"
33
34 #include "arm_compute/core/utils/quantization/AsymmHelpers.h"
35
36 namespace arm_compute
37 {
38 namespace test
39 {
40 namespace validation
41 {
42 namespace reference
43 {
44 template <typename T, typename TW, typename TB>
convolution_layer_nchw(const SimpleTensor<T> & src,const SimpleTensor<TW> & weights,const SimpleTensor<TB> & bias,SimpleTensor<T> & dst,const PadStrideInfo & info,const Size2D & dilation,unsigned int num_groups)45 SimpleTensor<T> convolution_layer_nchw(const SimpleTensor<T> &src, const SimpleTensor<TW> &weights, const SimpleTensor<TB> &bias, SimpleTensor<T> &dst, const PadStrideInfo &info,
46 const Size2D &dilation, unsigned int num_groups)
47 {
48 ARM_COMPUTE_ERROR_ON((src.shape()[2] / num_groups) != weights.shape()[2]);
49
50 // Compute reference
51 const int width_in = src.shape().x();
52 const int height_in = src.shape().y();
53 const int depth_in = src.shape().z();
54 const int width_out = dst.shape().x();
55 const int height_out = dst.shape().y();
56 const int depth_out = dst.shape().z();
57 const int width_weights = weights.shape().x();
58 const int height_weights = weights.shape().y();
59 const int depth_weights = weights.shape().z();
60 const int pad_left = info.pad_left();
61 const int pad_top = info.pad_top();
62 const int stride_xi = info.stride().first;
63 const int stride_yi = info.stride().second;
64
65 auto output_wh = scaled_dimensions(width_in, height_in, width_weights, height_weights, info, dilation);
66
67 const int start_xi = (dilation.x() * (width_weights - 1) + 1) / 2 - pad_left;
68 const int start_yi = (dilation.y() * (height_weights - 1) + 1) / 2 - pad_top;
69 const int end_xi = output_wh.first * stride_xi;
70 const int end_yi = output_wh.second * stride_yi;
71 const int num_batches = src.shape().total_size() / (width_in * height_in * depth_in);
72
73 #if defined(_OPENMP) && !( defined(__arm__) && defined(__ANDROID__))
74 #pragma omp parallel for collapse(5)
75 #endif /* _OPENMP */
76 for(int r = 0; r < num_batches; ++r)
77 {
78 for(int yi = start_yi; yi < start_yi + end_yi; yi += stride_yi)
79 {
80 for(int xi = start_xi; xi < start_xi + end_xi; xi += stride_xi)
81 {
82 for(int group = 0; group < static_cast<int>(num_groups); ++group)
83 {
84 for(int ofm = 0; ofm < static_cast<int>(depth_out / num_groups); ++ofm)
85 {
86 // Compute input and output offsets
87 const int offset_in = r * width_in * height_in * depth_in + (group * (depth_in / num_groups) * width_in * height_in);
88 const int xo = (xi - start_xi) / stride_xi;
89 const int yo = (yi - start_yi) / stride_yi;
90 const int offset_out = xo + yo * width_out + ((ofm + group * (depth_out / num_groups)) * width_out * height_out) + (r * width_out * height_out * depth_out);
91 const int offset_w = (ofm + group * (depth_out / num_groups)) * width_weights * height_weights * depth_weights;
92 const int offset_b = (ofm + group * (depth_out / num_groups));
93
94 ARM_COMPUTE_ASSERT(xo < width_out);
95 ARM_COMPUTE_ASSERT(yo < height_out);
96
97 // Compute 3D convolution
98 convolution_3d::detail::convolution3d(src, weights, bias, dst,
99 offset_in, offset_w, offset_b, offset_out,
100 xi, yi,
101 width_in, height_in, (depth_in / num_groups),
102 width_weights, height_weights, dilation.x(), dilation.y(), ofm);
103 }
104 }
105 }
106 }
107 }
108 return dst;
109 }
110 template <typename T, typename TW, typename TB>
convolution_layer(const SimpleTensor<T> & src,const SimpleTensor<TW> & weights,const SimpleTensor<TB> & bias,const TensorShape & output_shape,const PadStrideInfo & info,const Size2D & dilation,unsigned int num_groups,QuantizationInfo out_quant_info)111 SimpleTensor<T> convolution_layer(const SimpleTensor<T> &src, const SimpleTensor<TW> &weights, const SimpleTensor<TB> &bias, const TensorShape &output_shape, const PadStrideInfo &info,
112 const Size2D &dilation, unsigned int num_groups, QuantizationInfo out_quant_info)
113 {
114 // if no explicit quantization has been set you the same as src
115 if(out_quant_info == QuantizationInfo())
116 {
117 out_quant_info = src.quantization_info();
118 }
119 // Create reference
120 SimpleTensor<T> dst{ output_shape, src.data_type(), 1, out_quant_info };
121
122 return convolution_layer_nchw(src, weights, bias, dst, info, dilation, num_groups);
123 }
124
125 template SimpleTensor<float> convolution_layer(const SimpleTensor<float> &src, const SimpleTensor<float> &weights, const SimpleTensor<float> &bias, const TensorShape &output_shape,
126 const PadStrideInfo &info, const Size2D &dilation, unsigned int num_groups, QuantizationInfo out_quant_info);
127 template SimpleTensor<half> convolution_layer(const SimpleTensor<half> &src, const SimpleTensor<half> &weights, const SimpleTensor<half> &bias, const TensorShape &output_shape,
128 const PadStrideInfo &info, const Size2D &dilation, unsigned int num_groups, QuantizationInfo out_quant_info);
129 template SimpleTensor<uint8_t> convolution_layer(const SimpleTensor<uint8_t> &src, const SimpleTensor<uint8_t> &weights, const SimpleTensor<int32_t> &bias, const TensorShape &output_shape,
130 const PadStrideInfo &info, const Size2D &dilation, unsigned int num_groups, QuantizationInfo out_quant_info);
131 template SimpleTensor<uint8_t> convolution_layer(const SimpleTensor<uint8_t> &src, const SimpleTensor<int8_t> &weights, const SimpleTensor<int32_t> &bias, const TensorShape &output_shape,
132 const PadStrideInfo &info, const Size2D &dilation, unsigned int num_groups, QuantizationInfo out_quant_info);
133 template SimpleTensor<int8_t> convolution_layer(const SimpleTensor<int8_t> &src, const SimpleTensor<int8_t> &weights, const SimpleTensor<int32_t> &bias, const TensorShape &output_shape,
134 const PadStrideInfo &info, const Size2D &dilation, unsigned int num_groups, QuantizationInfo out_quant_info);
135 } // namespace reference
136 } // namespace validation
137 } // namespace test
138 } // namespace arm_compute
139