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
|
/*
* Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright (c) 2016-2018 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/CLSpaceToDepthKernel.h"
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/CL/CLKernelLibraryEx.h"
#include "arm_compute/core/CL/ICLTensor.h"
using namespace arm_compute;
namespace
{
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output,
const int32_t block_size)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::QASYMM8,
DataType::S16, DataType::S32, DataType::F16,
DataType::F32);
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8, DataType::QASYMM8,
DataType::S16, DataType::S32, DataType::F16,
DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(block_size < 1,
"Block size should be greater than or equal to 1.");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->dimension(3) != output->dimension(3),
"Input batch should be equal to Output batch");
auto layout_out = input->data_layout();
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
auto index_depth = get_data_layout_dimension_index(layout_out, DataLayoutDimension::CHANNEL);
auto index_height = get_data_layout_dimension_index(layout_out, DataLayoutDimension::HEIGHT);
auto index_width = get_data_layout_dimension_index(layout_out, DataLayoutDimension::WIDTH);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(
input->dimension(index_depth) * block_size * block_size != output->dimension(index_depth),
"Output depth should be equal to (input depth * block size *block size)");
ARM_COMPUTE_RETURN_ERROR_ON_MSG((input->dimension(index_width) % block_size) ||
(input->dimension(index_height) % block_size),
"Input height and width should be divisible by block size");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(
(output->dimension(index_width) != (input->dimension(index_width) / block_size)) ||
(output->dimension(index_height) != (input->dimension(index_height) / block_size)),
"Output height and width should be equal to "
"input_height/blocksize and input_width/blocksize respectively");
return Status{};
}
} // namespace
CLSpaceToDepthKernel::CLSpaceToDepthKernel() : _input(nullptr), _output(nullptr) {}
void CLSpaceToDepthKernel::configure(const ICLTensor *input, ICLTensor *output,
const int32_t block_size)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), block_size));
_input = input;
_output = output;
// Set kernel build options
auto layout_out = input->info()->data_layout();
std::set<std::string> build_opts;
build_opts.emplace("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
build_opts.emplace("-DBLOCK_SIZE=" + support::cpp11::to_string(block_size));
auto index_depth = get_data_layout_dimension_index(layout_out, DataLayoutDimension::CHANNEL);
auto depth = input->info()->dimension(index_depth);
build_opts.emplace("-DDEPTH_IN=" + support::cpp11::to_string(depth));
build_opts.emplace("-DZ_IN=" + support::cpp11::to_string(input->info()->tensor_shape().z()));
// Create kernel
_kernel = static_cast<cl::Kernel>(CLKernelLibraryEx::get().create_kernel(
"space_to_depth_" + lower_string(string_from_data_layout(layout_out)), build_opts));
// Configure kernel window
Window win = calculate_max_window(*input->info(), Steps());
Coordinates coord;
coord.set_num_dimensions(output->info()->num_dimensions());
output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
ICLKernel::configure_internal(win);
}
void CLSpaceToDepthKernel::run(const Window &window, cl::CommandQueue &queue)
{
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
Window slice_in = window.first_slice_window_4D().collapse(ICLKernel::window(), 2, 4);
// Setup output slice
Window slice_out(slice_in);
slice_out.set(Window::DimX, Window::Dimension(0, 0, 0));
slice_out.set(Window::DimY, Window::Dimension(0, 0, 0));
slice_out.set(Window::DimZ, Window::Dimension(0, 0, 0));
slice_out.set(3, Window::Dimension(0, 0, 0));
do
{
unsigned int idx = 0;
add_4D_tensor_argument(idx, _input, slice_in);
add_4D_tensor_argument(idx, _output, slice_out);
enqueue(queue, *this, slice_in);
} while (window.slide_window_slice_4D(slice_in) && window.slide_window_slice_4D(slice_out));
}
|
