1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
|
/*
* Copyright (c) 2017 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include "arm_compute/runtime/NEON/functions/NEDepthwiseConvolutionLayer.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/ITensor.h"
#include "arm_compute/core/PixelValue.h"
#include "arm_compute/runtime/NEON/NEScheduler.h"
#include "support/ToolchainSupport.h"
using namespace arm_compute;
NEDepthwiseConvolutionLayer3x3::NEDepthwiseConvolutionLayer3x3()
: _kernel(), _bias_kernel(), _border_handler(), _has_bias(false)
{
}
void NEDepthwiseConvolutionLayer3x3::configure(ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights);
// Configure kernels
_kernel.configure(input, weights, output, conv_info);
_border_handler.configure(input, _kernel.border_size(), BorderMode::CONSTANT, PixelValue(static_cast<float>(0.f)));
if(biases != nullptr)
{
_bias_kernel.configure(output, biases);
_has_bias = true;
}
}
void NEDepthwiseConvolutionLayer3x3::run()
{
NEScheduler::get().schedule(&_border_handler, Window::DimX);
NEScheduler::get().schedule(&_kernel, Window::DimX);
if(_has_bias)
{
NEScheduler::get().schedule(&_bias_kernel, Window::DimX);
}
}
NEDepthwiseConvolutionLayer::NEDepthwiseConvolutionLayer()
: _im2col_kernel(), _weights_reshape_kernel(), _v2mm_kernel(), _vector_to_tensor_kernel(), _input_reshaped(), _weights_reshaped(), _v2mm_output()
{
}
void NEDepthwiseConvolutionLayer::configure(ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights);
ARM_COMPUTE_ERROR_ON(input->info()->dimension(2) != weights->info()->dimension(2));
const size_t weights_w = weights->info()->dimension(0);
const size_t weights_h = weights->info()->dimension(1);
const size_t weights_z = weights->info()->dimension(2);
bool has_bias = (biases != nullptr);
unsigned int conv_w = 0;
unsigned int conv_h = 0;
std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), weights_w, weights_h, conv_info);
// Set up intermediate tensors
const size_t patch_size = weights_w * weights_h + ((has_bias) ? 1 : 0);
const size_t conv_size = conv_w * conv_h;
// Im2Col configuration
TensorShape shape_im2col = input->info()->tensor_shape();
shape_im2col.set(0, patch_size);
shape_im2col.set(1, conv_size);
shape_im2col.set(2, weights_z);
const TensorInfo info_im2col(shape_im2col, 1, input->info()->data_type(), input->info()->fixed_point_position());
_input_reshaped.allocator()->init(info_im2col);
_im2col_kernel.configure(input, &_input_reshaped, Size2D(weights_w, weights_h), conv_info, has_bias);
// Weights reshape configuration
const TensorShape shape_weights_reshape(patch_size, weights_z);
const TensorInfo info_weights_reshape(shape_weights_reshape, 1, weights->info()->data_type(), weights->info()->fixed_point_position());
_weights_reshaped.allocator()->init(info_weights_reshape);
_weights_reshape_kernel.configure(weights, &_weights_reshaped, biases);
// GEMV configuration
TensorShape shape_v2mm_out = input->info()->tensor_shape();
shape_v2mm_out.set(0, conv_size * weights_z);
shape_v2mm_out.set(1, 1);
shape_v2mm_out.set(2, 1);
const TensorInfo info_v2mm_out(shape_v2mm_out, 1, input->info()->data_type(), input->info()->fixed_point_position());
_v2mm_output.allocator()->init(info_v2mm_out);
_v2mm_kernel.configure(&_input_reshaped, &_weights_reshaped, &_v2mm_output);
_vector_to_tensor_kernel.configure(&_v2mm_output, output, conv_w, conv_h);
// Allocate intermediate tensors
_input_reshaped.allocator()->allocate();
_weights_reshaped.allocator()->allocate();
_v2mm_output.allocator()->allocate();
}
void NEDepthwiseConvolutionLayer::run()
{
NEScheduler::get().schedule(&_im2col_kernel, Window::DimX);
NEScheduler::get().schedule(&_weights_reshape_kernel, Window::DimX);
NEScheduler::get().schedule(&_v2mm_kernel, Window::DimX);
NEScheduler::get().schedule(&_vector_to_tensor_kernel, Window::DimX);
}
|
