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
|
/*
* Copyright (c) 2022-2024 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 "ClComponentDepthwiseConv2d.h"
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/dynamic_fusion/sketch/attributes/DepthwiseConv2dAttributes.h"
#include "src/core/CL/CLValidate.h"
#include "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwDepthwiseConv2d.h"
namespace arm_compute
{
namespace experimental
{
namespace dynamic_fusion
{
using Settings = ClComponentDepthwiseConv2dSettings;
Settings &Settings::export_input_to_cl_image(bool cl_image)
{
_export_input_to_cl_image = cl_image;
return *this;
}
bool Settings::export_input_to_cl_image() const
{
return _export_input_to_cl_image;
}
Settings &Settings::export_weights_to_cl_image(bool cl_image)
{
_export_weights_to_cl_image = cl_image;
return *this;
}
bool Settings::export_weights_to_cl_image() const
{
return _export_weights_to_cl_image;
}
Settings &Settings::fast_relaxed_math(bool fast_relaxed_math)
{
_fast_relaxed_math = fast_relaxed_math;
return *this;
}
bool Settings::fast_relaxed_math() const
{
return _fast_relaxed_math;
}
Settings &Settings::is_fma_available(bool is_fma_available)
{
_is_fma_available = is_fma_available;
return *this;
}
bool Settings::is_fma_available() const
{
return _is_fma_available;
}
Settings &Settings::n0(unsigned int n0)
{
_n0 = n0;
return *this;
}
unsigned int Settings::n0() const
{
return _n0;
}
Settings &Settings::m0(unsigned int m0)
{
_m0 = m0;
return *this;
}
unsigned int Settings::m0() const
{
return _m0;
}
Status ClComponentDepthwiseConv2d::validate(const Properties &properties,
const ArgumentPack<ITensorInfo> &tensors,
const Attributes &attributes,
const Settings &settings)
{
ARM_COMPUTE_UNUSED(properties, settings);
const auto src = tensors.get_const_tensor(TensorType::ACL_SRC_0);
const auto wei = tensors.get_const_tensor(TensorType::ACL_SRC_1);
const auto bia = tensors.get_const_tensor(TensorType::ACL_SRC_2);
const auto dst = tensors.get_const_tensor(TensorType::ACL_DST_0);
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, wei, dst);
// 1. Check validity
// Matching data type
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, wei);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
if (bia != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, bia);
}
// Matching data layout
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, wei);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, dst);
if (bia != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, bia);
}
// All tensor infos are initialized
ARM_COMPUTE_RETURN_ERROR_ON(src->tensor_shape().total_size() == 0);
ARM_COMPUTE_RETURN_ERROR_ON(wei->tensor_shape().total_size() == 0);
ARM_COMPUTE_RETURN_ERROR_ON(dst->tensor_shape().total_size() == 0);
if (bia != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON(bia->tensor_shape().total_size() == 0);
}
// Device requirements are met
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
// wei shape is correct
const DataLayout data_layout = src->data_layout();
const size_t channel_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
ARM_COMPUTE_RETURN_ERROR_ON(wei->dimension(channel_idx) !=
(src->dimension(channel_idx) * attributes.depth_multiplier()));
ARM_COMPUTE_RETURN_ERROR_ON_MSG(wei->num_dimensions() > 3, "Weights can be at most 3 dimensional");
// dst shape is correct
const PadStrideInfo pad_stride_info =
PadStrideInfo(attributes.stride().x(), attributes.stride().y(), attributes.pad().left, attributes.pad().right,
attributes.pad().top, attributes.pad().bottom, attributes.dimension_rounding_type());
const ConvolutionInfo conv_info{pad_stride_info, attributes.depth_multiplier(), ActivationLayerInfo(),
attributes.dilation()};
const TensorShape output_shape = misc::shape_calculator::compute_depthwise_convolution_shape(*src, *wei, conv_info);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), output_shape);
// Check strides and dilation
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.stride().first < 1);
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.stride().second < 1);
ARM_COMPUTE_RETURN_ERROR_ON((conv_info.dilation.x() < 1) || (conv_info.dilation.y() < 1));
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.stride().first > 1 && settings.m0() != 1);
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.dilation.x() > 1 && settings.m0() != 1);
if (conv_info.depth_multiplier > 1 && settings.n0() > 1)
{
ARM_COMPUTE_RETURN_ERROR_ON((conv_info.depth_multiplier % settings.n0()) != 0);
}
// Check export weights to cl image
ARM_COMPUTE_RETURN_ERROR_ON_MSG((settings.export_weights_to_cl_image() == true) &&
(export_to_cl_image(wei) == false),
"Weights cannot be exported to cl_image!");
ARM_COMPUTE_RETURN_ERROR_ON((settings.export_weights_to_cl_image() == true) && ((settings.n0() % 4) != 0));
ARM_COMPUTE_RETURN_ERROR_ON(wei->dimension(channel_idx) !=
(src->dimension(channel_idx) * conv_info.depth_multiplier));
// bia shape is correct
if (bia != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON_MSG(bia->dimension(0) != output_shape[channel_idx],
"Biases size and number of dst feature maps should match");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(bia->num_dimensions() > 1, "Biases should be one dimensional");
}
// 2. Check support level
// Data type
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::F16, DataType::F32);
// Data layout
ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(src, DataLayout::NHWC);
// Texture in the input tensor
ARM_COMPUTE_RETURN_ERROR_ON((settings.export_input_to_cl_image() == true));
return Status{};
}
ClComponentDepthwiseConv2d::ClComponentDepthwiseConv2d(ComponentId id,
const Properties &properties,
const ArgumentPack<ITensorInfo> &tensors,
const Attributes &attributes,
const Settings &settings)
: IGpuKernelComponent{id, properties, tensors},
_component_writer{std::make_unique<GpuCkwDepthwiseConv2d>(id, tensors, attributes, settings)}
{
ARM_COMPUTE_UNUSED(attributes, settings);
}
ClComponentDepthwiseConv2d::~ClComponentDepthwiseConv2d()
{
}
const IGpuCkwComponentDriver *ClComponentDepthwiseConv2d::ckw_component_driver() const
{
return _component_writer.get();
}
} // namespace dynamic_fusion
} // namespace experimental
} // namespace arm_compute
|
