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
|
/*
* Copyright (c) 2017-2022, 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 "src/cpu/kernels/CpuQuantizeKernel.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/Window.h"
#include "src/core/common/Registrars.h"
#include "src/core/CPP/Validate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "src/core/NEON/wrapper/wrapper.h"
#include "src/cpu/kernels/quantize/generic/neon/list.h"
#include <arm_neon.h>
#include <map>
namespace arm_compute
{
namespace cpu
{
namespace kernels
{
namespace
{
Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(src);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON(dst->tensor_shape().total_size() == 0);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(dst, 1, DataType::QSYMM8, DataType::QASYMM8,
DataType::QASYMM8_SIGNED, DataType::QASYMM16);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src, dst);
return Status{};
}
} // namespace
void CpuQuantizeKernel::configure(const ITensorInfo *src, ITensorInfo *dst)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst));
static const std::map<std::string, QuantizeFunctionExecutorPtr> quant_map = {
{"op_QASYMM8_QASYMM8", REGISTER_INTEGER_NEON(u8_u8_run_quantize_qasymm8)},
{"op_QASYMM8_QASYMM8_SIGNED", REGISTER_INTEGER_NEON(u8_i8_run_quantize_qasymm8)},
{"op_QASYMM8_QASYMM16", REGISTER_INTEGER_NEON(u8_run_quantize_qasymm16)},
{"op_QASYMM8_SIGNED_QASYMM8", REGISTER_INTEGER_NEON(i8_u8_run_quantize_qasymm8)},
{"op_QASYMM8_SIGNED_QASYMM8_SIGNED", REGISTER_INTEGER_NEON(i8_i8_run_quantize_qasymm8)},
{"op_QASYMM8_SIGNED_QASYMM16", REGISTER_INTEGER_NEON(i8_run_quantize_qasymm16)},
// Functions for offset only requantization
{"op_OFFSET_ONLY_QASYMM8_QASYMM8", REGISTER_INTEGER_NEON(u8_u8_run_requantize_offset_only)},
{"op_OFFSET_ONLY_QASYMM8_QASYMM8_SIGNED", REGISTER_INTEGER_NEON(u8_i8_run_requantize_offset_only)},
{"op_OFFSET_ONLY_QASYMM8_SIGNED_QASYMM8", REGISTER_INTEGER_NEON(i8_u8_run_requantize_offset_only)},
{"op_OFFSET_ONLY_QASYMM8_SIGNED_QASYMM8_SIGNED", REGISTER_INTEGER_NEON(i8_i8_run_requantize_offset_only)},
// Functions for offset uint8 to int8 and vice versa quantization (no scale changes)
{"op_OFFSET_ONLY_CONVERT_QASYMM8_SIGNED_QASYMM8",
REGISTER_INTEGER_NEON(i8_u8_run_requantize_offset_only_convert)},
{"op_OFFSET_ONLY_CONVERT_QASYMM8_QASYMM8_SIGNED",
REGISTER_INTEGER_NEON(u8_i8_run_requantize_offset_only_convert)},
{"op_F32_QSYMM8", REGISTER_FP32_NEON(fp32_i8_run_quantize_qsymm8)},
{"op_F32_QASYMM8", REGISTER_FP32_NEON(fp32_u8_run_quantize_qasymm8)},
{"op_F32_QASYMM8_SIGNED", REGISTER_FP32_NEON(fp32_i8_run_quantize_qasymm8)},
{"op_F32_QASYMM16", REGISTER_FP32_NEON(fp32_run_quantize_qasymm16)},
#ifdef ARM_COMPUTE_ENABLE_FP16
{"op_F16_QASYMM8", REGISTER_FP16_NEON(fp16_u8_run_quantize_qasymm8)},
{"op_F16_QASYMM8_SIGNED", REGISTER_FP16_NEON(fp16_i8_run_quantize_qasymm8)},
{"op_F16_QASYMM16", REGISTER_FP16_NEON(fp16_run_quantize_qasymm16)},
#endif /* ARM_COMPUTE_ENABLE_FP16 */
};
std::string function_to_call("op_");
// For offset only functions - must be 8-bit and have identical scale values.
if (src->quantization_info().scale() == dst->quantization_info().scale() &&
(is_data_type_quantized_asymmetric_char(src->data_type()) &&
is_data_type_quantized_asymmetric_char(dst->data_type())))
{
function_to_call += "OFFSET_ONLY_";
// For optimized datatype conversion 8-bit re-quantization offset only functions.
// These must have an offset of exactly 128 to match requirements - has specific circumstances to match use case.
auto uqinfo =
compute_requantization_scale_offset(src->quantization_info().uniform(), dst->quantization_info().uniform());
const auto src_dt = src->data_type();
if (src->data_type() != dst->data_type() && ((src_dt == DataType::QASYMM8_SIGNED && uqinfo.offset == 128) ||
(src_dt == DataType::QASYMM8 && uqinfo.offset == -128)))
{
function_to_call += "CONVERT_";
}
}
// Specify datatype for function
function_to_call += string_from_data_type(src->data_type()) + "_";
function_to_call += string_from_data_type(dst->data_type());
auto it = quant_map.find(function_to_call);
if (it == quant_map.end())
{
ARM_COMPUTE_ERROR("Unsupported combination of input and output data types");
}
_func = it->second;
// Calculate window. Squash if possible.
Window win;
std::tie(win, _split_dimension) = calculate_squashed_or_max_window(*src);
ICpuKernel::configure(win);
}
Status CpuQuantizeKernel::validate(const ITensorInfo *src, const ITensorInfo *dst)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst));
return Status{};
}
void CpuQuantizeKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);
ARM_COMPUTE_ERROR_ON(_func == nullptr);
const auto src = tensors.get_const_tensor(TensorType::ACL_SRC);
auto dst = tensors.get_tensor(TensorType::ACL_DST);
(*_func)(src, dst, window);
}
const char *CpuQuantizeKernel::name() const
{
return "CpuQuantizeKernel";
}
} // namespace kernels
} // namespace cpu
} // namespace arm_compute
|
