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
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
|
/*
* Copyright (c) 2017-2020 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/runtime/NEON/functions/NEGEMMLowpOutputStage.h"
#include "arm_compute/core/ITensor.h"
#include "arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ScaleKernel.h"
#include "arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h"
#include "arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h"
#include "arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h"
#include "arm_compute/core/Validate.h"
#include "support/MemorySupport.h"
namespace arm_compute
{
void NEGEMMLowpQuantizeDownInt32ToUint8Scale::configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_offset, int result_mult_int, int result_shift, int min, int max)
{
GEMMLowpOutputStageInfo info = GEMMLowpOutputStageInfo();
info.gemmlowp_offset = result_offset;
info.gemmlowp_multiplier = result_mult_int;
info.gemmlowp_shift = result_shift;
info.gemmlowp_min_bound = min;
info.gemmlowp_max_bound = max;
auto k = arm_compute::support::cpp14::make_unique<NEGEMMLowpQuantizeDownInt32ScaleKernel>();
k->configure(input, bias, output, &info);
_kernel = std::move(k);
}
Status NEGEMMLowpQuantizeDownInt32ToUint8Scale::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
{
GEMMLowpOutputStageInfo info = GEMMLowpOutputStageInfo();
info.gemmlowp_min_bound = min;
info.gemmlowp_max_bound = max;
return NEGEMMLowpQuantizeDownInt32ScaleKernel::validate(input, bias, output, &info);
}
void NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint::configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift,
int result_offset_after_shift, int min, int max)
{
auto k = arm_compute::support::cpp14::make_unique<NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel>();
k->configure(input, bias, output, result_fixedpoint_multiplier, result_shift, result_offset_after_shift, min, max);
_kernel = std::move(k);
}
Status NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
{
return NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::validate(input, bias, output, min, max);
}
void NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPoint::configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift,
int result_offset_after_shift, int min, int max)
{
auto k = arm_compute::support::cpp14::make_unique<NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel>();
k->configure(input, bias, output, result_fixedpoint_multiplier, result_shift, result_offset_after_shift, min, max);
_kernel = std::move(k);
}
Status NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPoint::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
{
return NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::validate(input, bias, output, min, max);
}
void NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint::configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int min, int max)
{
auto k = arm_compute::support::cpp14::make_unique<NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel>();
k->configure(input, bias, output, result_fixedpoint_multiplier, result_shift, min, max);
_kernel = std::move(k);
}
Status NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min, int max)
{
return NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::validate(input, bias, output, min, max);
}
void NEGEMMLowpOutputStage::configure(const ITensor *input, const ITensor *bias, ITensor *output, const GEMMLowpOutputStageInfo &info)
{
// Perform validate step
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(NEGEMMLowpOutputStage::validate(input->info(), bias != nullptr ? bias->info() : nullptr, output->info(), info));
switch(info.type)
{
case GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT:
{
switch(info.output_data_type)
{
case DataType::QASYMM8:
{
auto k = arm_compute::support::cpp14::make_unique<NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel>();
k->configure(input, bias, output, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_offset, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
_kernel = std::move(k);
break;
}
case DataType::QASYMM8_SIGNED:
{
auto k = arm_compute::support::cpp14::make_unique<NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel>();
k->configure(input, bias, output, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_offset, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
_kernel = std::move(k);
break;
}
case DataType::QSYMM16:
{
auto k = arm_compute::support::cpp14::make_unique<NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel>();
k->configure(input, bias, output, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
_kernel = std::move(k);
break;
}
default:
{
ARM_COMPUTE_ERROR("Unsupported output data type.");
break;
}
}
break;
}
case GEMMLowpOutputStageType::QUANTIZE_DOWN:
{
switch(info.output_data_type)
{
case DataType::QASYMM8:
case DataType::QASYMM8_SIGNED:
{
auto k = arm_compute::support::cpp14::make_unique<NEGEMMLowpQuantizeDownInt32ScaleKernel>();
k->configure(input, bias, output, &info);
_kernel = std::move(k);
break;
}
default:
{
ARM_COMPUTE_ERROR("Unsupported output data type.");
break;
}
}
break;
}
default:
ARM_COMPUTE_ERROR("Unsupported GEMMLowpOutputStage type.");
}
}
Status NEGEMMLowpOutputStage::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo &info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->data_type() == DataType::UNKNOWN, "NEGEMMLowpQuantizeDownScaleByFixedPoint cannot be used with UNKNOWN output data type.");
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QSYMM16);
ARM_COMPUTE_RETURN_ERROR_ON((info.type != GEMMLowpOutputStageType::QUANTIZE_DOWN) && (info.type != GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT));
switch(info.type)
{
case GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT:
{
switch(output->data_type())
{
case DataType::QASYMM8:
return NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::validate(input, bias, output, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
case DataType::QASYMM8_SIGNED:
return NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::validate(input, bias, output, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
case DataType::QSYMM16:
return NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::validate(input, bias, output, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
default:
return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported output data type.");
}
}
case GEMMLowpOutputStageType::QUANTIZE_DOWN:
{
switch(output->data_type())
{
case DataType::QASYMM8:
case DataType::QASYMM8_SIGNED:
return NEGEMMLowpQuantizeDownInt32ScaleKernel::validate(input, bias, output, &info);
default:
return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported output data type.");
}
}
default:
return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported GEMMLowpOutputStage type.");
}
}
} // namespace arm_compute
|
