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
|
/*
* Copyright (c) 2019-2020 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 "arm_compute/core/CL/kernels/CLHeightConcatenateLayerKernel.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/CL/OpenCL.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/IAccessWindow.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/Window.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "support/StringSupport.h"
#include <map>
namespace arm_compute
{
namespace
{
std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, unsigned int &num_elems_processed_per_iteration)
{
num_elems_processed_per_iteration = 4;
// The window needs to be based on input as we copy all the heights of input
Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration));
AccessWindowHorizontal input_access(input, 0, num_elems_processed_per_iteration);
AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
bool window_changed = update_window_and_padding(win, input_access, output_access);
Window win_collapsed = win.collapse(win, Window::DimZ);
Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
return std::make_pair(err, win_collapsed);
}
Status validate_arguments(const ITensorInfo *input, unsigned int height_offset, const ITensorInfo *output)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(Window::DimY) + height_offset > output->dimension(Window::DimY));
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != output->dimension(0));
for(size_t i = 2; i < Coordinates::num_max_dimensions; ++i)
{
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(i) != output->dimension(i));
}
ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
return Status{};
}
} // namespace
CLHeightConcatenateLayerKernel::CLHeightConcatenateLayerKernel()
: _input(nullptr), _output(nullptr), _height_offset(0), _num_elems_processed_per_iteration()
{
}
Status CLHeightConcatenateLayerKernel::validate(const ITensorInfo *input, unsigned int height_offset, const ITensorInfo *output)
{
unsigned int num_elems_processed_per_iteration;
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, height_offset, output));
ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), num_elems_processed_per_iteration).first);
return Status{};
}
void CLHeightConcatenateLayerKernel::configure(const ICLTensor *input, unsigned int height_offset, ICLTensor *output)
{
configure(CLKernelLibrary::get().get_compile_context(), input, height_offset, output);
}
void CLHeightConcatenateLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int height_offset, ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), height_offset, output->info()));
_input = input;
_output = output;
_height_offset = height_offset;
auto win_config = validate_and_configure_window(input->info(), output->info(), _num_elems_processed_per_iteration);
// Add build options
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->info()->element_size()));
build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(_num_elems_processed_per_iteration));
build_opts.add_option("-DHEIGHT_OFFSET=" + support::cpp11::to_string(_height_offset));
build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input->info()->dimension(2)));
if(is_data_type_quantized_asymmetric(input->info()->data_type()) && input->info()->quantization_info() != output->info()->quantization_info())
{
const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform();
const UniformQuantizationInfo oq_info = output->info()->quantization_info().uniform();
build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq_info.offset));
build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq_info.scale));
build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oq_info.scale));
}
// Create kernel
_kernel = create_kernel(compile_context, "concatenate_height", build_opts.options());
// Configure kernel window
ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
ICLKernel::configure_internal(std::get<1>(win_config));
// Set output valid region
output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
}
void CLHeightConcatenateLayerKernel::run(const Window &window, cl::CommandQueue &queue)
{
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
unsigned int idx = 0;
add_4D_tensor_argument(idx, _input, window);
add_4D_tensor_argument(idx, _output, window);
enqueue(queue, *this, window, lws_hint());
}
} // namespace arm_compute
|
