File: lstm_unit_cpu_avx2.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 (123 lines) | stat: -rw-r--r-- 2,568 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
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
#include "caffe2/perfkernels/lstm_unit_cpu-impl.h"

namespace caffe2 {
namespace perfkernels {
namespace {
// Explicit initialize for float and AVX2 vectorization
template void LstmUnitImpl<float>(
    const int N,
    const int D,
    const int t,
    const float* H_prev,
    const float* C_prev,
    const float* X,
    const int32_t* seqLengths,
    const bool drop_states,
    float* C,
    float* H,
    const float forget_bias);

template void LstmUnitGradientImpl<float>(
    int N,
    int D,
    int t,
    const float* C_prev,
    const float* X,
    const int32_t* seqLengths,
    const float* C,
    const float* H,
    const float* C_diff,
    const float* H_diff,
    bool drop_states,
    float* H_prev_diff,
    float* C_prev_diff,
    float* X_diff,
    const float forget_bias);
} // namespace

// Define templated implementation fo LSTM kernels on CPU supporting AVX2
template <typename T>
void LstmUnitImpl__avx2_fma(
    const int N,
    const int D,
    const int t,
    const T* H_prev,
    const T* C_prev,
    const T* X,
    const int32_t* seqLengths,
    const bool drop_states,
    T* C,
    T* H,
    const float forget_bias) {
  LstmUnitImpl(
      N, D, t, H_prev, C_prev, X, seqLengths, drop_states, C, H, forget_bias);
}

template <typename T>
void LstmUnitGradientImpl__avx2_fma(
    int N,
    int D,
    int t,
    const T* C_prev,
    const T* X,
    const int32_t* seqLengths,
    const T* C,
    const T* H,
    const T* C_diff,
    const T* H_diff,
    bool drop_states,
    T* H_prev_diff,
    T* C_prev_diff,
    T* X_diff,
    const float forget_bias) {
  LstmUnitGradientImpl(
      N,
      D,
      t,
      C_prev,
      X,
      seqLengths,
      C,
      H,
      C_diff,
      H_diff,
      drop_states,
      H_prev_diff,
      C_prev_diff,
      X_diff,
      forget_bias);
}

// Explicit initialize for float
template void LstmUnitImpl__avx2_fma<float>(
    const int N,
    const int D,
    const int t,
    const float* H_prev,
    const float* C_prev,
    const float* X,
    const int32_t* seqLengths,
    const bool drop_states,
    float* C,
    float* H,
    const float forget_bias);

template void LstmUnitGradientImpl__avx2_fma<float>(
    int N,
    int D,
    int t,
    const float* C_prev,
    const float* X,
    const int32_t* seqLengths,
    const float* C,
    const float* H,
    const float* C_diff,
    const float* H_diff,
    bool drop_states,
    float* H_prev_diff,
    float* C_prev_diff,
    float* X_diff,
    const float forget_bias);

} // namespace perfkernels
} // namespace caffe2