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
|
/*
* Copyright (c) 2019 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright (c) 2017-2019 ARM Limited.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "arm_compute/core/CL/kernels/CLTransposeConvLayerUpsampleKernel.h"
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/CL/CLValidate.h"
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/Window.h"
using namespace arm_compute;
CLTransposeConvLayerUpsampleKernel::CLTransposeConvLayerUpsampleKernel()
: _input(nullptr), _output(nullptr), _inner_border(), _info()
{
}
Status CLTransposeConvLayerUpsampleKernel::validate(const ITensorInfo *input,
const ITensorInfo *output,
const BorderSize &inner_border,
const PadStrideInfo &info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16,
DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
const DataLayout data_layout = input->data_layout();
const size_t idx_w = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
const size_t idx_h = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
const size_t idx_c = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(idx_w) == 0);
ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(idx_h) == 0);
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(idx_c) != output->dimension(idx_c));
for (size_t i = 3; i < Coordinates::num_max_dimensions; ++i)
{
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(i) != output->dimension(i));
}
ARM_COMPUTE_RETURN_ERROR_ON_MSG(inner_border.right > info.stride().first - 1,
"inner_border_right must be smaller that stride_x");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(inner_border.top > info.stride().second - 1,
"inner_border_top must be smaller that stride_y");
return Status{};
}
void CLTransposeConvLayerUpsampleKernel::configure(const ICLTensor *input, ICLTensor *output,
const BorderSize &inner_border,
const PadStrideInfo &info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
_input = input;
_output = output;
_inner_border = inner_border;
_info = info;
// Perform validation step
ARM_COMPUTE_ERROR_THROW_ON(CLTransposeConvLayerUpsampleKernel::validate(
input->info(), output->info(), inner_border, info));
// Create kernel
CLBuildOptions build_opts;
build_opts.add_option(("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type())));
_kernel = static_cast<cl::Kernel>(
CLKernelLibrary::get().create_kernel("deconvolution_upsample", build_opts.options()));
constexpr unsigned int num_elems_processed_per_iteration = 1;
// Configure kernel window
Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration));
AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration);
output_access.set_valid_region(win, ValidRegion(Coordinates(), output->info()->tensor_shape()));
ICLKernel::configure_internal(win);
}
void CLTransposeConvLayerUpsampleKernel::run(const Window &window, cl::CommandQueue &queue)
{
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
const DataLayout data_layout = _input->info()->data_layout();
const size_t idx_w = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
const size_t idx_h = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
const int out_start_x = _info.pad_left();
const int out_end_x = _output->info()->dimension(idx_w) - _inner_border.right -
_info.pad_right() + _info.stride().first - 1;
const int out_step_x = _info.stride().first;
const int out_start_y = _inner_border.top + _info.pad_top();
const int out_end_y =
_output->info()->dimension(idx_h) - _info.pad_bottom() + _info.stride().second - 1;
const int out_step_y = _info.stride().second;
switch (data_layout)
{
case DataLayout::NCHW:
{
Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
Window slice_out = collapsed.first_slice_window_3D();
slice_out.set(Window::DimX, Window::Dimension(out_start_x, out_end_x, out_step_x));
slice_out.set(Window::DimY, Window::Dimension(out_start_y, out_end_y, out_step_y));
Window slice_in = collapsed.first_slice_window_3D();
do
{
unsigned int idx = 0;
add_3D_tensor_argument(idx, _input, slice_in);
add_3D_tensor_argument(idx, _output, slice_out);
enqueue(queue, *this, slice_out);
} while (collapsed.slide_window_slice_3D(slice_in) &&
collapsed.slide_window_slice_3D(slice_out));
break;
}
case DataLayout::NHWC:
{
// NOTE: not collapsing in NHWC
Window slice_out = window.first_slice_window_3D();
slice_out.set(Window::DimY, Window::Dimension(out_start_x, out_end_x, out_step_x));
slice_out.set(Window::DimZ, Window::Dimension(out_start_y, out_end_y, out_step_y));
Window slice_in = window.first_slice_window_3D();
do
{
unsigned int idx = 0;
add_3D_tensor_argument(idx, _input, slice_in);
add_3D_tensor_argument(idx, _output, slice_out);
enqueue(queue, *this, slice_out);
} while (window.slide_window_slice_3D(slice_in) && window.slide_window_slice_3D(slice_out));
break;
}
default:
ARM_COMPUTE_ERROR("Unsupported data layout");
}
}
|
