File: int8_transpose_op.h

package info (click to toggle)
pytorch 1.7.1-7
  • links: PTS, VCS
  • area: main
  • in suites: bullseye
  • size: 80,340 kB
  • sloc: cpp: 670,830; python: 343,991; ansic: 67,845; asm: 5,503; sh: 2,924; java: 2,888; xml: 266; makefile: 244; ruby: 148; yacc: 144; objc: 51; lex: 44
file content (38 lines) | stat: -rw-r--r-- 1,120 bytes parent folder | download | duplicates (2)
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
#ifndef CAFFE2_OPERATORS_INT8_TRANSPOSE_OP_H_
#define CAFFE2_OPERATORS_INT8_TRANSPOSE_OP_H_

#include "caffe2/core/context.h"
#include "caffe2/core/operator.h"
#include "caffe2/core/tensor_int8.h"
#include "caffe2/operators/quantized/int8_utils.h"
#include "caffe2/operators/transpose_op.h"

namespace caffe2 {

namespace int8 {

class Int8TransposeOp final : public TransposeOp<CPUContext> {
 public:
  Int8TransposeOp(const OperatorDef& operator_def, Workspace* ws)
      : TransposeOp(operator_def, ws) {}

  bool RunOnDevice() override {
    auto& X = Inputs()[0]->Get<Int8TensorCPU>();
    auto* Y = Outputs()[0]->GetMutable<Int8TensorCPU>();
    int32_t Y_zero_point =
        this->template GetSingleArgument<int>("Y_zero_point", 0);
    auto Y_scale = this->template GetSingleArgument<float>("Y_scale", 1);
    CAFFE_ENFORCE_EQ(Y_zero_point, X.zero_point);
    CAFFE_ENFORCE_EQ(Y_scale, X.scale);
    Y->scale = Y_scale;
    Y->zero_point = Y_zero_point;
    TransposeImpl<uint8_t>(X.t, &Y->t);
    return true;
  }
};

} // namespace int8

} // namespace caffe2

#endif // CAFFE2_OPERATORS_INT8_TRANSPOSE_OP_H_