File: relu_dnnlowp_op.h

package info (click to toggle)
pytorch 1.13.1%2Bdfsg-4
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 139,252 kB
  • sloc: cpp: 1,100,274; python: 706,454; ansic: 83,052; asm: 7,618; java: 3,273; sh: 2,841; javascript: 612; makefile: 323; xml: 269; ruby: 185; yacc: 144; objc: 68; lex: 44
file content (31 lines) | stat: -rw-r--r-- 739 bytes parent folder | download
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
#pragma once

#include "caffe2/operators/relu_op.h"

#include "caffe2/core/tensor_int8.h"
#include "caffe2/quantization/server/caffe2_dnnlowp_utils.h"

namespace caffe2 {

template <typename T>
class ReluDNNLowPOp final : public Operator<CPUContext> {
 public:
  USE_OPERATOR_FUNCTIONS(CPUContext);
  ReluDNNLowPOp(const OperatorDef& operator_def, Workspace* ws)
      : Operator<CPUContext>(operator_def, ws),
        qfactory_(dnnlowp::GetQuantizationFactoryOf(this)) {}

  bool RunOnDevice() override;

 private:
  std::unique_ptr<dnnlowp::QuantizationFactory> qfactory_;
};

namespace internal {

template <typename T>
void ReluAVX2(const int N, const int zero_point, const T* X, T* Y);

} // namespace internal

} // namespace caffe2