quantization_utils.cpp
6.47 KB
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
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#include "../precomp.hpp"
#include "layers_common.hpp"
namespace cv
{
namespace dnn
{
// Quantize FP32/FP16 Inputs to INT8
class QuantizeLayerImpl CV_FINAL : public QuantizeLayer
{
public:
QuantizeLayerImpl(const LayerParams& params)
{
scale = params.get<float>("scales", 1.0f);
zeropoint = params.get<int>("zeropoints", 0);
setParamsFrom(params);
}
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
return backendId == DNN_BACKEND_OPENCV;
}
bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
std::vector<MatShape> &internals) const CV_OVERRIDE
{
CV_Assert(inputs.size() == 1);
Layer::getMemoryShapes(inputs, requiredOutputs, outputs, internals);
return false;
}
virtual void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
}
#ifdef HAVE_OPENCL
bool forward_ocl(InputArrayOfArrays inputs_, OutputArrayOfArrays outputs_, OutputArrayOfArrays internals_)
{
std::vector<UMat> inputs, outputs;
inputs_.getUMatVector(inputs);
outputs_.getUMatVector(outputs);
if (inputs_.depth() == CV_16S)
{
UMat inputFp32(shape(inputs[0]), CV_32F);
convertFp16(inputs[0], inputFp32);
inputFp32.copyTo(inputs[0]);
}
inputs[0].convertTo(outputs[0], CV_8S, 1.f/scale, zeropoint);
return true;
}
#endif
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) CV_OVERRIDE
{
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
inputs[0].convertTo(outputs[0], CV_8S, 1.f/scale, zeropoint);
}
};
// Dequantize INT8 Inputs to FP32/FP16
class DequantizeLayerImpl CV_FINAL : public DequantizeLayer
{
public:
DequantizeLayerImpl(const LayerParams& params)
{
scale = params.get<float>("scales", 1.0f);
zeropoint = params.get<int>("zeropoints", 0);
setParamsFrom(params);
}
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
return backendId == DNN_BACKEND_OPENCV;
}
bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
std::vector<MatShape> &internals) const CV_OVERRIDE
{
CV_Assert(inputs.size() == 1);
Layer::getMemoryShapes(inputs, requiredOutputs, outputs, internals);
return false;
}
virtual void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
}
#ifdef HAVE_OPENCL
bool forward_ocl(InputArrayOfArrays inputs_, OutputArrayOfArrays outputs_, OutputArrayOfArrays internals_)
{
std::vector<UMat> inputs, outputs;
inputs_.getUMatVector(inputs);
outputs_.getUMatVector(outputs);
UMat outputFp32(shape(outputs[0]), CV_32F);
inputs[0].convertTo(outputFp32, CV_32F, scale, -(scale*zeropoint));
if (outputs_.depth() == CV_16S)
convertFp16(outputFp32, outputs[0]);
else
outputFp32.copyTo(outputs[0]);
return true;
}
#endif
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) CV_OVERRIDE
{
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
inputs[0].convertTo(outputs[0], CV_32F, scale, -(scale*zeropoint));
}
};
// Rescale/Requantize INT8 Inputs from (scale1, zeropoint1) to (scale2, zeropoint2)
class RequantizeLayerImpl CV_FINAL : public RequantizeLayer
{
public:
RequantizeLayerImpl(const LayerParams& params)
{
scale = params.get<float>("scale", 1.f);
shift = params.get<float>("shift", 0.f);
setParamsFrom(params);
}
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
return backendId == DNN_BACKEND_OPENCV;
}
bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
std::vector<MatShape> &internals) const CV_OVERRIDE
{
CV_Assert(inputs.size() == 1);
Layer::getMemoryShapes(inputs, requiredOutputs, outputs, internals);
return false;
}
virtual void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
}
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) CV_OVERRIDE
{
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
inputs[0].convertTo(outputs[0], CV_8S, scale, shift);
}
};
Ptr<QuantizeLayer> QuantizeLayer::create(const LayerParams& params)
{
return Ptr<QuantizeLayer>(new QuantizeLayerImpl(params));
}
Ptr<DequantizeLayer> DequantizeLayer::create(const LayerParams& params)
{
return Ptr<DequantizeLayer>(new DequantizeLayerImpl(params));
}
Ptr<RequantizeLayer> RequantizeLayer::create(const LayerParams& params)
{
return Ptr<RequantizeLayer>(new RequantizeLayerImpl(params));
}
}
}