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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
|
/*
* Copyright (c) 2017-2021 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_NEDEPTHWISECONVOLUTION_H
#define ARM_COMPUTE_NEDEPTHWISECONVOLUTION_H
#include "arm_compute/runtime/IMemoryManager.h"
#include "arm_compute/runtime/MemoryGroup.h"
#include "arm_compute/runtime/NEON/functions/NEActivationLayer.h"
#include "arm_compute/runtime/NEON/functions/NEPermute.h"
#include <memory>
namespace arm_compute
{
// Forward declarations
class ITensor;
class NEDepthwiseConvolutionLayerNativeKernel;
/** Function to execute a depthwise convolution.
*/
class NEDepthwiseConvolutionLayer : public IFunction
{
public:
/** Default constructor */
NEDepthwiseConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
/** Prevent instances of this class from being copied (As this class contains pointers) */
NEDepthwiseConvolutionLayer(const NEDepthwiseConvolutionLayer &) = delete;
/** Default move constructor */
NEDepthwiseConvolutionLayer(NEDepthwiseConvolutionLayer &&) = default;
/** Prevent instances of this class from being copied (As this class contains pointers) */
NEDepthwiseConvolutionLayer &operator=(const NEDepthwiseConvolutionLayer &) = delete;
/** Default move assignment operator */
NEDepthwiseConvolutionLayer &operator=(NEDepthwiseConvolutionLayer &&) = default;
/** Default destructor */
~NEDepthwiseConvolutionLayer();
/** Initialize the function's source, destination, weights and convolution information.
*
* Valid data layouts:
* - NHWC
* - NCHW
*
* Valid data type configurations:
* |src0 |src1 |src2 |dst |
* |:--------------|:------------------|:------|:--------------|
* |F16 |F16 |F16 |F16 |
* |F32 |F32 |F32 |F32 |
* |QASYMM8 |QASYMM8 |S32 |QASYMM8 |
* |QASYMM8 |QSYMM8_PER_CHANNEL |S32 |QASYMM8 |
* |QASYMM8_SIGNED |QASYMM8_SIGNED |S32 |QASYMM8_SIGNED |
* |QASYMM8_SIGNED |QSYMM8_PER_CHANNEL |S32 |QASYMM8_SIGNED |
*
* @param[in, out] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32
* @param[out] output Destination tensor. Data type supported: same as @p input.
* @param[in] weights Weights tensor. These are 3D tensors with shape [kernel_x, kernel_y, IFM].
* Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED.
* @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed.
* Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED.
* @param[in] conv_info Padding and stride information to use for the convolution.
* @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation.
* @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
*/
void configure(ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info,
unsigned int depth_multiplier = 1, const ActivationLayerInfo &act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U));
/** Static function to check if given info will lead to a valid configuration of @ref NEDepthwiseConvolutionLayer
*
* @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32
* @param[in] output Destination tensor. Data type supported: same as @p input.
* @param[in] weights Weights tensor. These are 3D tensors with shape [kernel_x, kernel_y, IFM].
* Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED.
* @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed.
* Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED.
* @param[in] conv_info Padding and stride information to use for the convolution.
* @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation.
* @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 *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
unsigned int depth_multiplier = 1, const ActivationLayerInfo &act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U));
// Inherited methods overriden:
void run() override;
void prepare() override;
private:
/** Basic function to execute optimized depthwise convolution routines. This function calls the following kernels:
*
* @note At the moment 3x3 and 5x5 convolution of stride 1, 2 are supported
*
* -# @ref NEFillBorderKernel (if pad_x or pad_y > 0) and no assembly kernel implementation is present
* -# @ref NEDepthwiseConvolutionLayer3x3Kernel if 3x3 and no assembly kernel implementation is present
* -# @ref cpu::CpuDepthwiseConvolutionAssemblyDispatch if assembly kernel implementation is present
* -# @ref NEDirectConvolutionLayerOutputStageKernel if re-quantization of output is required
* -# @ref NEActivationLayer if fused activation is required
*
*/
class NEDepthwiseConvolutionLayerOptimizedInternal : public IFunction
{
public:
/** Default constructor */
NEDepthwiseConvolutionLayerOptimizedInternal(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
/** Prevent instances of this class from being copied (As this class contains pointers) */
NEDepthwiseConvolutionLayerOptimizedInternal(const NEDepthwiseConvolutionLayerOptimizedInternal &) = delete;
/** Default move constructor */
NEDepthwiseConvolutionLayerOptimizedInternal(NEDepthwiseConvolutionLayerOptimizedInternal &&) = default;
/** Prevent instances of this class from being copied (As this class contains pointers) */
NEDepthwiseConvolutionLayerOptimizedInternal &operator=(const NEDepthwiseConvolutionLayerOptimizedInternal &) = delete;
/** Default move assignment operator */
NEDepthwiseConvolutionLayerOptimizedInternal &operator=(NEDepthwiseConvolutionLayerOptimizedInternal &&) = default;
/** Default destructor */
~NEDepthwiseConvolutionLayerOptimizedInternal() = default;
/** Initialize the function's source, destination, kernels and border_size.
*
* @param[in, out] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. (Written to only for border filling).
* @param[in] weights Weights tensor. These are 3D tensors with shape [kernel_x, kernel_y, IFM]. Data type supported: Same as @p input.
* @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed.
* Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED.
* @param[out] output Destination tensor. Data type supported: same as @p input.
* @param[in] conv_info Padding and stride information to use for the convolution.
* @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation.
* @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
*/
void configure(ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info,
unsigned int depth_multiplier = 1, const ActivationLayerInfo &act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U));
/** Static function to check if given info will lead to a valid configuration of @ref NEDepthwiseConvolutionLayer3x3
*
* @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. (Written to only for border filling).
* @param[in] weights Weights tensor. These are 3D tensors with shape [kernel_x, kernel_y, IFM]. Data type supported: Same as @p input.
* @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed.
* Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED.
* @param[in] output Destination tensor. Data type supported: same as @p input.
* @param[in] conv_info Padding and stride information to use for the convolution.
* @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation.
* @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 *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
unsigned int depth_multiplier = 1, const ActivationLayerInfo &act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U));
// Inherited methods overriden:
void run() override;
void prepare() override;
private:
MemoryGroup _memory_group;
struct Impl;
std::unique_ptr<Impl> _impl;
};
/** Basic function to execute a generic depthwise convolution. This function calls the following kernel:
*
* -# @ref NEDepthwiseConvolutionLayerNativeKernel
*
*/
class NEDepthwiseConvolutionLayerGeneric : public IFunction
{
public:
/** Default constructor */
NEDepthwiseConvolutionLayerGeneric();
/** Prevent instances of this class from being copied (As this class contains pointers) */
NEDepthwiseConvolutionLayerGeneric(const NEDepthwiseConvolutionLayerGeneric &) = delete;
/** Default move constructor */
NEDepthwiseConvolutionLayerGeneric(NEDepthwiseConvolutionLayerGeneric &&) = default;
/** Prevent instances of this class from being copied (As this class contains pointers) */
NEDepthwiseConvolutionLayerGeneric &operator=(const NEDepthwiseConvolutionLayerGeneric &) = delete;
/** Default move assignment operator */
NEDepthwiseConvolutionLayerGeneric &operator=(NEDepthwiseConvolutionLayerGeneric &&) = default;
/** Default destructor */
~NEDepthwiseConvolutionLayerGeneric() = default;
/** Initialize the function's source, destination, weights and convolution information.
*
* @param[in, out] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. (Written to only for border filling).
* @param[out] output Destination tensor. Data type supported: same as @p input.
* @param[in] weights Weights tensor. These are 3D tensors with shape [kernel_x, kernel_y, IFM].
* Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED.
* @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed.
* Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED.
* @param[in] conv_info Padding and stride information to use for the convolution.
* @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation.
* @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
*/
void configure(ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info,
unsigned int depth_multiplier = 1, const ActivationLayerInfo &act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U));
/** Static function to check if given info will lead to a valid configuration of @ref NEDepthwiseConvolutionLayerGeneric
*
* @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. (Written to only for border filling).
* @param[in] output Destination tensor. Data type supported: same as @p input.
* @param[in] weights Weights tensor. These are 3D tensors with shape [kernel_x, kernel_y, IFM].
* Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8/QASYMM8_SIGNED.
* @param[in] biases Biases tensor. A 1D tensor with shape [IFM]. Must be nullptr if not needed.
* Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED.
* @param[in] conv_info Padding and stride information to use for the convolution.
* @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation.
* @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 *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
unsigned int depth_multiplier = 1, const ActivationLayerInfo &act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U));
// Inherited methods overriden:
void run() override;
private:
struct Impl;
std::unique_ptr<Impl> _impl;
};
MemoryGroup _memory_group;
struct Impl;
std::unique_ptr<Impl> _impl;
};
} // namespace arm_compute
#endif /* ARM_COMPUTE_NEDEPTHWISECONVOLUTION_H */
|
