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
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
|
/*
* 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_CLIM2COLKERNEL_H__
#define __ARM_COMPUTE_CLIM2COLKERNEL_H__
#include "arm_compute/core/CL/ICLKernel.h"
#include "arm_compute/core/Size2D.h"
namespace arm_compute
{
class ICLTensor;
/** Interface for the im2col reshape kernel.
*
* Rearranges image blocks into columns. It is used to strip out each convolution block to a single column.
* It is used to transform a convolution to a plain matrix multiplication.
*
* For example taking into account the image below and assuming 3x3 image blocks with stride of 1 we have:
* @f[
* \left( \begin{array}{cccc}
* a00 & a01 & a02 & a03 \\
* a10 & a11 & a12 & a13 \\
* a20 & a21 & a22 & a23 \\
* a30 & a31 & a32 & a33 \\
* \end{array} \right)
* =
* \left( \begin{array}{ccccccccc}
* a00 & a01 & a02 & a10 & a11 & a12 & a20 & a21 & a22 \\
* a01 & a02 & a03 & a11 & a12 & a13 & a21 & a22 & a23 \\
* a10 & a11 & a12 & a20 & a21 & a22 & a30 & a31 & a32 \\
* a11 & a12 & a13 & a21 & a22 & a23 & a31 & a32 & a33 \\
* \end{array} \right)
* @f]
*/
class CLIm2ColKernel : public ICLKernel
{
public:
/** Default constructor */
CLIm2ColKernel();
/** Prevent instances of this class from being copied (As this class contains pointers) */
CLIm2ColKernel(const CLIm2ColKernel &) = delete;
/** Prevent instances of this class from being copied (As this class contains pointers) */
CLIm2ColKernel &operator=(const CLIm2ColKernel &) = delete;
/** Allow instances of this class to be moved */
CLIm2ColKernel(CLIm2ColKernel &&) = default;
/** Allow instances of this class to be moved */
CLIm2ColKernel &operator=(CLIm2ColKernel &&) = default;
/** Set the input and output of the kernel.
*
* @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
* while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QASYMM8/QS16/F16/F32
* @param[out] output The output tensor. First 2 lower dimensions represent a transform of each 3D input,
* while every dimension above represents a batch. Data types supported: Same as @p input
* @param[in] kernel_dims The kernel dimensions (width and height).
* @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
* @param[in] has_bias In case biases are provided expands the matrix with 1.
* @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
*/
void configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation = Size2D(1U, 1U));
/** Static function to check if given info will lead to a valid configuration of @ref CLIm2ColKernel
*
* @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
* while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QASYMM8/QS16/F16/F32
* @param[in] output The output tensor. First 2 lower dimensions represent a transform of each 3D input,
* while every dimension above represents a batch. Data types supported: Same as @p input
* @param[in] kernel_dims The kernel dimensions (width and height).
* @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
* @param[in] has_bias In case biases are provided expands the matrix with 1.
* @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
*
* @return a status
*/
static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation = Size2D(1U, 1U));
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
private:
/** Run the reshape kernel optimised for the special case (stride is 1, padding is 0 and kernel's low 3 dimensions are same as input)
*
* @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
* @param[in,out] queue Command queue on which to enqueue the kernel.
*/
void run_reduced(const Window &window, cl::CommandQueue &queue);
/** run the generic convolution layer input reshape kernel
*
* @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
* @param[in,out] queue Command queue on which to enqueue the kernel.
*/
void run_generic(const Window &window, cl::CommandQueue &queue);
/** Chooses and configure the right kernel for the given input arguments.
*
* @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
* while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QASYMM8/QS16/F16/F32
* @param[in] output The output tensor. First 2 lower dimensions represent a transform of each 3D input,
* while every dimension above represents a batch. Data types supported: Same as @p input
* @param[in] kernel_dims The kernel dimensions (width and height).
* @param[in] dilation Dilation, in elements, across x and y. Defaults to (1, 1).
* @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
* @param[in] has_bias In case biases are provided expands the matrix with 1.
* @param[out] build_opts OpenCL buil program options.
*
* @return the name of the kernel chosen
*/
std::string configure_window(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims,
const Size2D &dilation, const PadStrideInfo &conv_info, CLBuildOptions &build_opts);
/** Common signature for the kernel to run */
using Im2ColFunction = void (CLIm2ColKernel::*)(const Window &, cl::CommandQueue &);
public:
const ICLTensor *_input;
ICLTensor *_output;
PadStrideInfo _conv_info;
std::pair<unsigned int, unsigned int> _convolved_dims;
unsigned int _num_elems_processed_per_iteration;
Im2ColFunction _run_func;
Size2D _kernel_dims;
};
} // namespace arm_compute
#endif /*__ARM_COMPUTE_CLIM2COLKERNEL_H__ */
|
