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
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
|
// Copyright 2016 The Gemmlowp Authors. All Rights Reserved.
//
// 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.
#ifndef GEMMLOWP_META_TRANSFORM_KERNELS_H_
#define GEMMLOWP_META_TRANSFORM_KERNELS_H_
#include "base.h"
namespace gemmlowp {
namespace meta {
struct Quantize {
float range_min;
float range_offset;
float range_scale;
int count;
};
struct Dequantize {
float range_min;
float range_offset;
float range_scale;
int count;
};
struct Requantize {
float input_range_min;
float input_range_offset;
float input_range_scale;
float output_range_min;
float output_range_offset;
float one_over_output_range_scale;
int count;
};
template <typename Type>
struct MinMax {
Type min;
Type max;
int count;
};
template <typename BiasType>
struct BiasAdd {
float input_range_min;
float input_range_offset;
float input_range_scale;
float bias_range_min;
float bias_range_offset;
float bias_range_scale;
float output_range_min;
float output_range_offset;
float one_over_output_range_scale;
int count;
int rows;
const BiasType* bias;
};
template <typename InType, typename OutType, int kernel_size, int leftovers>
class Transform1DKernel<InType, OutType, Quantize, kernel_size, leftovers> {
public:
static void Transform(const InType* in, const Quantize& params,
OutType* output) {
#ifdef DEBUG
#ifdef DEBUG_METAGEMM_VERBOSE
std::cout << "Quantize::Transform(" << std::string(typeid(InType).name())
<< ", " << std::string(typeid(OutType).name()) << ") -- "
<< kernel_size << "x" << leftovers << std::endl;
#endif
#else
std::cerr << "FATAL: Quantize::Transform not implemented." << std::endl;
std::exit(1);
#endif
}
};
template <typename InType, typename OutType, int kernel_size, int leftovers>
class Transform1DKernel<InType, OutType, Dequantize, kernel_size, leftovers> {
public:
static void Transform(const InType* in, const Dequantize& params,
OutType* output) {
#ifdef DEBUG
#ifdef DEBUG_METAGEMM_VERBOSE
std::cout << "Dequantize::Transform(" << std::string(typeid(InType).name())
<< ", " << std::string(typeid(OutType).name()) << ") -- "
<< kernel_size << "x" << leftovers << std::endl;
#endif
#else
std::cerr << "FATAL: Dequantize::Transform not implemented." << std::endl;
std::exit(1);
#endif
}
};
template <typename InType, typename OutType, int kernel_size, int leftovers>
class Transform1DKernel<InType, OutType, Requantize, kernel_size, leftovers> {
public:
static void Transform(const InType* in, const Requantize& params,
OutType* output) {
#ifdef DEBUG
#ifdef DEBUG_METAGEMM_VERBOSE
std::cout << "Requantize::Transform(" << std::string(typeid(InType).name())
<< ", " << std::string(typeid(OutType).name()) << ") -- "
<< kernel_size << "x" << leftovers << std::endl;
#endif
#else
std::cerr << "FATAL: Requantize::Transform not implemented." << std::endl;
std::exit(1);
#endif
}
};
template <typename InType, typename OutType, int kernel_size, int leftovers,
typename Type>
class Transform1DKernel<InType, OutType, MinMax<Type>, kernel_size, leftovers> {
public:
static void Transform(const InType* in, const MinMax<Type>& params,
OutType* output) {
#ifdef DEBUG
#ifdef DEBUG_METAGEMM_VERBOSE
std::cout << "MinMax::Transform(" << std::string(typeid(InType).name())
<< ", " << std::string(typeid(OutType).name()) << ") -- "
<< kernel_size << "x" << leftovers << std::endl;
#endif
#else
std::cerr << "FATAL: MinMax::Transform not implemented." << std::endl;
std::exit(1);
#endif
}
};
template <typename InType, typename OutType, int kernel_size, int leftovers,
typename Type>
class Transform1DKernel<InType, OutType, BiasAdd<Type>, kernel_size,
leftovers> {
public:
static void Transform(const InType* in, const BiasAdd<Type>& params,
OutType* output) {
#ifdef DEBUG
#ifdef DEBUG_METAGEMM_VERBOSE
std::cout << "BiasAdd::Transform(" << std::string(typeid(InType).name())
<< ", " << std::string(typeid(OutType).name()) << ") -- "
<< kernel_size << "x" << leftovers << std::endl;
#endif
#else
std::cerr << "FATAL: BiasAdd::Transform not implemented." << std::endl;
std::exit(1);
#endif
}
};
template <typename InType, typename OutType>
class Transform1DUtil<InType, OutType, Quantize> {
public:
static int EstimateComputeCost(const Quantize& params) {
return params.count * 8;
}
static const InType* OffsetInput(const Quantize& params, const InType* input,
int offset) {
return input + offset;
}
static OutType* OffsetOutput(const Quantize& params, OutType* output,
int offset) {
return output + offset;
}
};
template <typename InType, typename OutType>
class Transform1DUtil<InType, OutType, Requantize> {
public:
static int EstimateComputeCost(const Requantize& params) {
return params.count * 12;
}
static const InType* OffsetInput(const Requantize& params,
const InType* input, int offset) {
return input + offset;
}
static OutType* OffsetOutput(const Requantize& params, OutType* output,
int offset) {
return output + offset;
}
};
template <typename InType, typename OutType>
class Transform1DUtil<InType, OutType, Dequantize> {
public:
static int EstimateComputeCost(const Dequantize& params) {
return params.count * 12;
}
static const InType* OffsetInput(const Dequantize& params,
const InType* input, int offset) {
return input + offset;
}
static OutType* OffsetOutput(const Dequantize& params, OutType* output,
int offset) {
return output + offset;
}
};
template <typename InType, typename OutType, typename MinMaxType>
class Transform1DUtil<InType, OutType, MinMax<MinMaxType>> {
public:
static int EstimateComputeCost(const MinMax<MinMaxType>& params) {
return params.count * 4;
}
static const InType* OffsetInput(const MinMax<MinMaxType>& params,
const InType* input, int offset) {
return input + offset;
}
static OutType* OffsetOutput(const MinMax<MinMaxType>& params,
OutType* output, int offset) {
return output + offset;
}
};
} // namespace meta
} // namespace gemmlowp
#ifdef GEMMLOWP_NEON_32
#include "transform_kernels_arm_32.h"
#elif defined(GEMMLOWP_NEON_64)
#include "transform_kernels_arm_64.h"
#endif
#endif // GEMMLOWP_META_TRANSFORM_KERNELS_H_
|
