forked from pytorch/pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
byte_weight_dequant_op.h
55 lines (46 loc) · 1.68 KB
/
byte_weight_dequant_op.h
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
#ifndef CAFFE2_OPERATORS_BYTE_WEIGHT_DEQUANT_OP_H_
#define CAFFE2_OPERATORS_BYTE_WEIGHT_DEQUANT_OP_H_
#include "caffe2/core/operator.h"
#include "caffe2/utils/eigen_utils.h"
#include "caffe2/utils/math.h"
namespace caffe2 {
template <typename Context>
class ByteWeightDequantOp : public Operator<Context> {
public:
ByteWeightDequantOp(const OperatorDef& operator_def, Workspace* ws)
: Operator<Context>(operator_def, ws),
min_(this->template GetSingleArgument<float>("min", -3)),
max_(this->template GetSingleArgument<float>("max", 3)),
shape_(this->template GetRepeatedArgument<int64_t>("shape")) {}
USE_OPERATOR_FUNCTIONS(Context);
using Operator<Context>::Operator;
bool RunOnDevice() override {
const auto& WI = Input(0);
auto* Y = Output(0, shape_, at::dtype<float>());
float bin_interval = (max_ - min_) / 255.0;
int total = 1;
for (int i = 0; i < shape_.size(); i++) {
total *= Y->size(i);
}
const uint8_t* Xdata;
if (WI.template IsType<uint8_t>()) {
CAFFE_ENFORCE(total, WI.nbytes());
Xdata = WI.template data<uint8_t>();
} else {
CAFFE_ENFORCE(total, WI.template data<std::string>()[0].size());
Xdata = reinterpret_cast<const uint8_t*>(
WI.template data<std::string>()[0].c_str());
}
auto* Ydata = Y->template mutable_data<float>();
ConstEigenVectorMap<uint8_t> index(&Xdata[0], total);
EigenVectorMap<float> weights(&Ydata[0], total);
weights = (index.cast<float>().array() * bin_interval) + min_;
return true;
}
private:
float min_;
float max_;
std::vector<int64_t> shape_;
};
} // namespace caffe2
#endif // CAFFE2_OPERATORS_BYTE_WEIGHT_DEQUANT_OP_H_