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
|
/*
* Copyright (c) 2017-2018 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.
*/
#ifndef __ARM_COMPUTE_NEGEMMWINOGRADLAYERKERNEL_H__
#define __ARM_COMPUTE_NEGEMMWINOGRADLAYERKERNEL_H__
#include "arm_compute/core/NEON/INEKernel.h"
#include "arm_compute/core/NEON/kernels/winograd/convolution.hpp"
#include "arm_compute/core/NEON/kernels/winograd/tensor.hpp"
namespace arm_compute
{
class ITensor;
class NEWinogradLayerKernel;
class Winograd3x3F32 final
{
public:
friend class NEWinogradLayerKernel;
Winograd3x3F32(
const int n_batches, /** Number of batches in the input and output tensors. */
const int n_input_channels, /** Number of feature maps in a batch of the input tensor. */
const int n_input_rows, /** Number of rows in a feature map of the input tensor. */
const int n_input_cols, /** Number of columns in a feature map of the input tensor. */
const int n_output_channels, /** Number of feature maps in the output tensor. */
const bool same_padding, /** Use "SAME" padding, otherwise use "VALID". */
const float *const weights, /** Pointer to weight tensor in spatial domain. Must be ordered as "Height x Rows x Input Feature Maps x Output Feature Maps. */
float *const weights_storage, /** Pointer to storage for weight tensor in the Winograd domain. Must be at least the size returned by `get_weight_storage_size`. */
const float *const input, /** Pointer to NHWC ordered input tensor, in the spatial domain. */
float *const winograd_input, /** Pointer to working space for the input tensor in the Winograd domain. Must be at least the size returned by `get_input_storage_size`. */
float *const output, /** Pointer to NHWC ordered output tensor, in the spatial domain. */
float *const winograd_output /** Pointer to working space for the output tensor in the Winograd domain. Must be at least the size returned by `get_output_storage_size`. */
);
~Winograd3x3F32();
void transform_weights();
void transform_input();
void transform_output();
private:
class Private;
std::unique_ptr<Private> _pimpl;
};
class NEWinogradLayerKernel : public INEKernel
{
public:
/** Constructor */
NEWinogradLayerKernel();
/** Prevent instances of this class from being copied (As this class contains pointers) */
NEWinogradLayerKernel(const NEWinogradLayerKernel &) = delete;
/** Prevent instances of this class from being copied (As this class contains pointers) */
NEWinogradLayerKernel &operator=(const NEWinogradLayerKernel &) = delete;
/** Allow instances of this class to be moved */
NEWinogradLayerKernel(NEWinogradLayerKernel &&) = default;
/** Allow instances of this class to be moved */
NEWinogradLayerKernel &operator=(NEWinogradLayerKernel &&) = default;
virtual ~NEWinogradLayerKernel() = default;
/** Initialise the kernel
*
* @param[in] convolver A pointer to the winograd convolver, this object must have been configured and is ready to execute 16 GEMMS .
*/
void configure(Winograd3x3F32 *convolver);
// Inherited methods overridden:
void run(const Window &window, const ThreadInfo &info) override;
/* Get the memory required to instantiate a new Winograd operator.
*/
static size_t get_weight_storage_size(
const int n_output_channels, /** Number of output feature maps. */
const int n_input_channels /** Number of input feature maps. */
);
static unsigned int get_input_storage_size(
const int n_batches, /** Number of batches in the input tensor. */
const int n_channels, /** Number of feature maps in the input tensor. */
const int n_rows, /** Number of rows in each feature map. */
const int n_cols, /** Number of columns in each feature map. */
const bool same_padding /** Use "SAME" padding, otherwise use "VALID". */
);
/** Determine how much memory (in units of TOut) to allocate for the
* (Winograd domain) output.
*/
static unsigned int get_output_storage_size(
const int n_batches, /** Number of batches in the output tensor. */
const int n_rows, /** Number of rows in each feature map of the input tensor. */
const int n_cols, /** Number of columns in each feature map of the input tensor. */
const int n_output_channels, /** Number of feature maps in the output tensor. */
const bool same_padding /** Use "SAME" padding, otherwise use "VALID". */
);
protected:
Winograd3x3F32 *_convolver;
};
} // namespace arm_compute
#endif /*__ARM_COMPUTE_NEGEMMWINOGRADLAYERKERNEL_H__*/
|
