File: fakefp16_transform.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 (25 lines) | stat: -rw-r--r-- 584 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
#pragma once

#include <string>
#include <unordered_map>
#include <vector>

#include <caffe2/core/net.h>
#include <caffe2/core/workspace.h>
#include <caffe2/proto/caffe2_pb.h>

namespace caffe2 {
namespace opt {

// Mapping from fp32 ops to fakefp16 ops
TORCH_API std::unordered_map<std::string, std::string> getFakeFp16OpMapping(
    bool use_fp16_acc = false,
    bool use_nnpi = false);

TORCH_API void fakeFp16FuseOps(NetDef* net);

// Transform normal fp32 operators to fakefp16 operators.
TORCH_API void fakeFp16Transform(NetDef* net);

} // namespace opt
} // namespace caffe2