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
|
/*
* 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/CLCastKernel.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;
CLCastKernel::CLCastKernel() : _input(nullptr), _output(nullptr) {}
void CLCastKernel::configure(const ICLTensor *input, ICLTensor *output, SubDataType input_subtype)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
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_ERROR_ON_MISMATCHING_SHAPES(input, output);
_input = input;
_output = output;
constexpr unsigned int num_elems_processed_per_iteration = 16;
// Set kernel build options
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE_IN=" + get_cl_type_from_data_type(input->info()->data_type()));
build_opts.add_option("-DDATA_TYPE_OUT=" +
get_cl_type_from_data_type(output->info()->data_type()));
build_opts.add_option(
("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration)));
// Create kernel
if (is_data_type_quantized_asymmetric(input->info()->data_type()))
{
const float scale_in = input->info()->quantization_info().scale;
const int offset_in = input->info()->quantization_info().offset;
build_opts.add_option("-DSCALE=" + float_to_string_with_full_precision(scale_in));
build_opts.add_option("-DOFFSET=" + support::cpp11::to_string(offset_in));
_kernel = static_cast<cl::Kernel>(
CLKernelLibraryEx::get().create_kernel("cast_qasymm_in", build_opts.options()));
}
else if (is_data_type_quantized_asymmetric(output->info()->data_type()))
{
const float scale_in = output->info()->quantization_info().scale;
const int offset_in = output->info()->quantization_info().offset;
build_opts.add_option("-DSCALE=" + float_to_string_with_full_precision(scale_in));
build_opts.add_option("-DOFFSET=" + support::cpp11::to_string(offset_in));
_kernel = static_cast<cl::Kernel>(
CLKernelLibraryEx::get().create_kernel("cast_qasymm_out", build_opts.options()));
}
else
{
build_opts.add_option_if(input_subtype == SubDataType::BOOL, "-DBOOL_INPUT");
_kernel = static_cast<cl::Kernel>(
CLKernelLibraryEx::get().create_kernel("cast", build_opts.options()));
}
// Configure kernel window
Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration));
AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration);
AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration);
update_window_and_padding(win, input_access, output_access);
output_access.set_valid_region(win, input->info()->valid_region());
ICLKernel::configure_internal(win);
}
void CLCastKernel::run(const Window &window, cl::CommandQueue &queue)
{
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
Window slice = collapsed.first_slice_window_3D();
do
{
unsigned int idx = 0;
add_3D_tensor_argument(idx, _input, slice);
add_3D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
} while (collapsed.slide_window_slice_3D(slice));
}
|
