File: math_lp.cc

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 (42 lines) | stat: -rw-r--r-- 845 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
32
33
34
35
36
37
38
39
40
41
42
#include "caffe2/sgd/math_lp.h"

namespace caffe2 {

namespace internal {

template <>
void dot<float, float, float>(
    const int N,
    const float* x,
    const float* y,
    float* z,
    CPUContext* ctx) {
  math::Dot<float, CPUContext>(N, x, y, z, ctx);
}

template <>
void dot<float, at::Half, float>(
    const int N,
    const float* x,
    const at::Half* y,
    float* z,
    CPUContext* ctx) {
#ifdef _MSC_VER
  std::vector<float> tmp_y_vec(N);
  float* tmp_y = tmp_y_vec.data();
#else
  // NOLINTNEXTLINE(cppcoreguidelines-avoid-c-arrays,modernize-avoid-c-arrays)
  float tmp_y[N];
#endif
  for (int i = 0; i < N; i++) {
#ifdef __F16C__
    tmp_y[i] = _cvtss_sh(y[i], 0); // TODO: vectorize
#else
    tmp_y[i] = y[i];
#endif
  }
  math::Dot<float, CPUContext>(N, x, tmp_y, z, ctx);
}

} // namespace internal
} // namespace caffe2