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
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
|
/*
* Copyright (c) 2020 Samsung Electronics Co., Ltd. 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.
*/
#include "BinaryArithmeticLayer.h"
#include <cker/operation/BinaryArithmeticOps.h>
namespace onert
{
namespace backend
{
namespace cpu
{
namespace ops
{
namespace
{
template <nnfw::cker::BinaryArithmeticOpType arithmetic_type, typename T> struct Eval
{
nnfw::cker::Shape _lhs_shape;
nnfw::cker::Shape _rhs_shape;
nnfw::cker::Shape _output_shape;
nnfw::cker::BinaryArithmeticOpParam _op_params;
bool _need_broadcast;
Eval(const IPortableTensor *lhs, const IPortableTensor *rhs, IPortableTensor *output,
nnfw::cker::BinaryArithmeticOpParam op_params)
: _op_params(std::move(op_params)), _need_broadcast(false)
{
if (!output->is_dynamic())
updateCache(lhs, rhs, output);
}
void updateCache(const IPortableTensor *lhs, const IPortableTensor *rhs, IPortableTensor *output)
{
_lhs_shape.ReplaceWith(getShape(lhs));
_rhs_shape.ReplaceWith(getShape(rhs));
_output_shape.ReplaceWith(getShape(output));
_need_broadcast = nnfw::cker::ProcessBroadcastShapes(_lhs_shape, _rhs_shape, &_op_params);
}
void operator()(const IPortableTensor *lhs, const IPortableTensor *rhs, IPortableTensor *output)
{
// Assume dynamic tensors never become static and static ones never change shape since
// configure()
if (output->is_dynamic())
updateCache(lhs, rhs, output);
else
assert(_lhs_shape == getShape(lhs) && _rhs_shape == getShape(rhs) &&
_output_shape == getShape(output));
auto lhs_buffer = getBuffer<T>(lhs);
auto rhs_buffer = getBuffer<T>(rhs);
auto output_buffer = getBuffer<T>(output);
if (_need_broadcast)
{
nnfw::cker::BroadcastBinaryArithmeticOp<arithmetic_type>(
_op_params, _lhs_shape, lhs_buffer, _rhs_shape, rhs_buffer, _output_shape, output_buffer);
}
else
{
nnfw::cker::BinaryArithmeticOp<arithmetic_type>(
_op_params, _lhs_shape, lhs_buffer, _rhs_shape, rhs_buffer, _output_shape, output_buffer);
}
}
};
template <nnfw::cker::BinaryArithmeticOpType arithmetic_type>
std::function<void(const IPortableTensor *, const IPortableTensor *, IPortableTensor *)>
generateKernelGeneric(const IPortableTensor *lhs, const IPortableTensor *rhs,
IPortableTensor *output, const ir::Activation activation,
nnfw::cker::BinaryArithmeticOpParam &op_params)
{
switch (lhs->data_type())
{
case OperandType::FLOAT32:
{
float output_activation_min = 0, output_activation_max = 0;
CalculateActivationRange(activation, &output_activation_min, &output_activation_max);
op_params.float_activation_max = output_activation_max;
op_params.float_activation_min = output_activation_min;
return Eval<arithmetic_type, float>(lhs, rhs, output, op_params);
break;
}
case OperandType::INT32:
{
int32_t output_activation_min = 0, output_activation_max = 0;
CalculateActivationRange(activation, &output_activation_min, &output_activation_max);
op_params.quantized_activation_max = output_activation_max;
op_params.quantized_activation_min = output_activation_min;
return Eval<arithmetic_type, int32_t>(lhs, rhs, output, op_params);
break;
}
default:
throw std::runtime_error{"BinaryArithmetic(generic): Unsupported data type"};
}
}
void setAddOrSubQuant8Params(const IPortableTensor *lhs, const IPortableTensor *rhs,
IPortableTensor *output, ir::Activation activation,
nnfw::cker::BinaryArithmeticOpParam *params)
{
int32_t output_activation_min, output_activation_max;
CalculateActivationRangeQuantized(activation, output, &output_activation_min,
&output_activation_max);
nnfw::cker::BinaryArithmeticOpParam &op_params = *params;
op_params.quantized_activation_max = output_activation_max;
op_params.quantized_activation_min = output_activation_min;
// Parameters for scaled quantized computation
op_params.left_shift = 20;
// Zero-points of input and output tensors
op_params.input1_offset = -lhs->data_zero_point();
op_params.input2_offset = -rhs->data_zero_point();
op_params.output_offset = output->data_zero_point();
// Compute normalized scale for _lhs and _rhs values,
// and represent in 32-bit fixed point
const double norm_max_scale = 2 * std::max(lhs->data_scale(), rhs->data_scale());
const double real_lhs_scale = lhs->data_scale() / norm_max_scale;
const double real_rhs_scale = rhs->data_scale() / norm_max_scale;
// output scale is used to normalize final result, so we invert the scale here
const double real_output_scale =
norm_max_scale / (output->data_scale() * (1 << op_params.left_shift));
// Represent the scales as fixed int32_t multipliers, and int32_t shifts
QuantizeMultiplier(real_lhs_scale, &op_params.input1_multiplier, &op_params.input1_shift);
QuantizeMultiplier(real_rhs_scale, &op_params.input2_multiplier, &op_params.input2_shift);
QuantizeMultiplier(real_output_scale, &op_params.output_multiplier, &op_params.output_shift);
}
void setMulQuant8Params(const IPortableTensor *lhs, const IPortableTensor *rhs,
IPortableTensor *output, ir::Activation activation,
nnfw::cker::BinaryArithmeticOpParam *params)
{
int32_t output_activation_min, output_activation_max;
CalculateActivationRangeQuantized(activation, output, &output_activation_min,
&output_activation_max);
nnfw::cker::BinaryArithmeticOpParam &op_params = *params;
op_params.quantized_activation_max = output_activation_max;
op_params.quantized_activation_min = output_activation_min;
op_params.input1_offset = -lhs->data_zero_point();
op_params.input2_offset = -rhs->data_zero_point();
op_params.output_offset = output->data_zero_point();
double real_multiplier = lhs->data_scale() * rhs->data_scale() / output->data_scale();
QuantizeMultiplier(real_multiplier, &op_params.output_multiplier, &op_params.output_shift);
}
} // namespace
void BinaryArithmeticLayer::configure(const IPortableTensor *lhs, const IPortableTensor *rhs,
IPortableTensor *output, const ir::Activation activation,
const ArithmeticType arithmetic_type)
{
assert(lhs != nullptr);
assert(rhs != nullptr);
assert(output != nullptr);
_lhs = lhs;
_rhs = rhs;
_output = output;
nnfw::cker::BinaryArithmeticOpParam op_params;
switch (arithmetic_type)
{
case ArithmeticType::kAdd:
if (_lhs->data_type() == OperandType::QUANT_UINT8_ASYMM)
{
setAddOrSubQuant8Params(_lhs, _rhs, _output, activation, &op_params);
_kernel =
Eval<nnfw::cker::BinaryArithmeticOpType::ADD, uint8_t>(_lhs, _rhs, _output, op_params);
}
else if (_lhs->data_type() == OperandType::QUANT_INT8_ASYMM)
{
setAddOrSubQuant8Params(_lhs, _rhs, _output, activation, &op_params);
_kernel =
Eval<nnfw::cker::BinaryArithmeticOpType::ADD, int8_t>(_lhs, _rhs, _output, op_params);
}
else
{
_kernel = generateKernelGeneric<nnfw::cker::BinaryArithmeticOpType::ADD>(
_lhs, _rhs, _output, activation, op_params);
}
break;
case ArithmeticType::kSub:
if (_lhs->data_type() == OperandType::QUANT_UINT8_ASYMM)
{
setAddOrSubQuant8Params(_lhs, _rhs, _output, activation, &op_params);
op_params.input2_multiplier *= -1;
_kernel =
Eval<nnfw::cker::BinaryArithmeticOpType::SUB, uint8_t>(_lhs, _rhs, _output, op_params);
}
else if (_lhs->data_type() == OperandType::QUANT_INT8_ASYMM)
{
setAddOrSubQuant8Params(_lhs, _rhs, _output, activation, &op_params);
op_params.input2_multiplier *= -1;
_kernel =
Eval<nnfw::cker::BinaryArithmeticOpType::SUB, int8_t>(_lhs, _rhs, _output, op_params);
}
else
{
_kernel = generateKernelGeneric<nnfw::cker::BinaryArithmeticOpType::SUB>(
_lhs, _rhs, _output, activation, op_params);
}
break;
case ArithmeticType::kMul:
if (_lhs->data_type() == OperandType::QUANT_UINT8_ASYMM)
{
nnfw::cker::BinaryArithmeticOpParam op_params;
setMulQuant8Params(_lhs, _rhs, _output, activation, &op_params);
_kernel =
Eval<nnfw::cker::BinaryArithmeticOpType::MUL, uint8_t>(_lhs, _rhs, _output, op_params);
}
else if (_lhs->data_type() == OperandType::QUANT_INT8_ASYMM)
{
nnfw::cker::BinaryArithmeticOpParam op_params;
setMulQuant8Params(_lhs, _rhs, _output, activation, &op_params);
_kernel =
Eval<nnfw::cker::BinaryArithmeticOpType::MUL, int8_t>(_lhs, _rhs, _output, op_params);
}
else
{
_kernel = generateKernelGeneric<nnfw::cker::BinaryArithmeticOpType::MUL>(
_lhs, _rhs, _output, activation, op_params);
}
break;
case ArithmeticType::kDiv:
if (_lhs->data_type() == OperandType::QUANT_UINT8_ASYMM)
{
throw std::runtime_error{
"BinaryArithmetic(Div): Div operation does not support quantization"};
}
else if (_lhs->data_type() == OperandType::INT32)
{
throw std::runtime_error{"BinaryArithmetic(Div): Unsupported data type"};
}
else
{
_kernel = generateKernelGeneric<nnfw::cker::BinaryArithmeticOpType::DIV>(
_lhs, _rhs, _output, activation, op_params);
}
break;
default:
throw std::runtime_error{"BinaryArithmetic: Unsupported BinaryArithmetic type"};
}
}
void BinaryArithmeticLayer::run() { _kernel(_lhs, _rhs, _output); }
} // namespace ops
} // namespace cpu
} // namespace backend
} // namespace onert
|
