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
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
|
#include <cstdint>
#include <limits>
#include <immintrin.h>
#ifdef _OPENMP
#include <omp.h>
#endif
#include <fbgemm/QuantUtils.h>
namespace caffe2 {
namespace internal {
template <typename T>
void SpatialBNNHWCAVX2(
const int N,
const int C,
const int HxW,
const int in_zero_point,
const int out_zero_point,
const T* X,
const float* alpha,
const float* beta,
T* Y,
bool relu_fused);
template <bool ReluFused>
void SpatialBNNHWCAVX2_uint8(
const int N,
const int C,
const int HxW,
const int in_zero_point,
const int out_zero_point,
const uint8_t* X,
const float* alpha,
const float* beta,
uint8_t* Y) {
constexpr int kVLen = 8;
const int outer_size = N * HxW;
const __m256i min_v = _mm256_set1_epi32(std::numeric_limits<uint8_t>::min());
const __m256i max_v = _mm256_set1_epi32(std::numeric_limits<uint8_t>::max());
const __m256i shuffle_mask_v = _mm256_set_epi8(
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
0x0c,
0x08,
0x04,
0x00,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
0xff,
0x0c,
0x08,
0x04,
0x00);
const __m256i permute_mask_v =
_mm256_set_epi32(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00);
const __m256i in_zero_point_v = _mm256_set1_epi32(in_zero_point);
const __m256i out_zero_point_v = _mm256_set1_epi32(out_zero_point);
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int i = 0; i < outer_size; ++i) {
int n = C / kVLen * kVLen;
int r = C % kVLen;
const uint8_t* X_ptr = X + i * C;
uint8_t* Y_ptr = Y + i * C;
for (int j = 0; j < n; j += kVLen) {
const __m256i cur_v = _mm256_cvtepu8_epi32(
_mm_loadl_epi64(reinterpret_cast<const __m128i*>(X_ptr + j)));
const __m256 cur_v_float =
_mm256_cvtepi32_ps(_mm256_sub_epi32(cur_v, in_zero_point_v));
const __m256 alpha_v = _mm256_loadu_ps(alpha + j);
const __m256 beta_v = _mm256_loadu_ps(beta + j);
const __m256 result_float_v =
_mm256_fmadd_ps(alpha_v, cur_v_float, beta_v);
const __m256i result_rounded_v = _mm256_cvtps_epi32(result_float_v);
__m256i result_v = _mm256_add_epi32(result_rounded_v, out_zero_point_v);
if (ReluFused) {
result_v = _mm256_max_epi32(result_v, out_zero_point_v);
}
__m256i clipped_v =
_mm256_max_epi32(min_v, _mm256_min_epi32(max_v, result_v));
clipped_v = _mm256_shuffle_epi8(clipped_v, shuffle_mask_v);
clipped_v = _mm256_permutevar8x32_epi32(clipped_v, permute_mask_v);
*reinterpret_cast<int64_t*>(Y_ptr + j) =
_mm256_extract_epi64(clipped_v, 0);
}
for (int j = 0; j < r; ++j) {
long quantized_down = out_zero_point +
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions,bugprone-narrowing-conversions)
std::lrintf(alpha[n + j] * (X_ptr[n + j] - in_zero_point) +
beta[n + j]);
if (ReluFused) { // static if
quantized_down = std::max<long>(quantized_down, out_zero_point);
}
Y_ptr[n + j] = fbgemm::clamp<long, uint8_t>(quantized_down, 8);
}
}
}
template <>
void SpatialBNNHWCAVX2<uint8_t>(
const int N,
const int C,
const int HxW,
const int in_zero_point,
const int out_zero_point,
const uint8_t* X,
const float* alpha,
const float* beta,
uint8_t* Y,
bool relu_fused) {
if (relu_fused) {
SpatialBNNHWCAVX2_uint8<true>(
N, C, HxW, in_zero_point, out_zero_point, X, alpha, beta, Y);
} else {
SpatialBNNHWCAVX2_uint8<false>(
N, C, HxW, in_zero_point, out_zero_point, X, alpha, beta, Y);
}
}
} // namespace internal
} // namespace caffe2
|
