-
Notifications
You must be signed in to change notification settings - Fork 0
/
test_simd_distance.cpp
212 lines (194 loc) · 7.01 KB
/
test_simd_distance.cpp
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
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
#include <cmath>
#include <iostream>
#include <chrono>
#include <emmintrin.h>
#include <immintrin.h>
#include <xmmintrin.h>
#include <vector>
void _mm256_print_epi8(__m256i vec)
{
char temp[32];
_mm256_storeu_si256((__m256i*)&temp[0], vec);
int i;
for(i=0; i<32; i++)
printf(" %f,", temp[i]);
}
void _mm512_print_epi32(__m512i vec) {
char temp[64];
_mm512_store_si512((__m512i*)&temp[0], vec);
int i;
int32_t* p = (int32_t*)temp;
for (i = 0; i < 16; i++)
printf("%d,", p[i]);
printf("\n");
}
void _mm128_print_epi32(__m128i vec) {
char temp[16];
_mm_store_si128((__m128i*)&temp[0], vec);
int i;
int32_t* p = (int32_t*)temp;
for (i = 0; i < 4; i++)
printf("%d,", p[i]);
printf("\n");
}
inline int
SquareSum(const int8_t* x, int d) {
d = 32;
__m512i sum = _mm512_setzero_si512();
for (int i = 0; i < d; i += 32) {
__m256i xx = _mm256_loadu_si256((__m256i *)(x + i));
__m512i xx_ext = _mm512_cvtepi8_epi16(xx);
sum = _mm512_add_epi32(_mm512_madd_epi16(xx_ext, xx_ext), sum);
}
// reduce for 16 int32 integers in AVX512
__m256i sumh = _mm256_add_epi32(_mm512_extracti32x8_epi32(sum, 0), _mm512_extracti32x8_epi32(sum, 1));
__m128i sumhh = _mm_add_epi32(_mm256_castsi256_si128(sumh), _mm256_extracti128_si256(sumh, 1));
__m128i tmp = _mm_hadd_epi32(sumhh, sumhh);
_mm128_print_epi32(tmp);
return _mm_extract_epi32(tmp, 0) + _mm_extract_epi32(tmp, 1);
}
inline int
IPSQ8(const int8_t* x, const int8_t *y, int d) {
d = 32;
__m512i sum = _mm512_setzero_si512();
for (int i = 0; i < d; i += 32) {
__m256i xx = _mm256_loadu_si256((__m256i *)(x + i));
__m256i yy = _mm256_loadu_si256((__m256i *)(y + i));
__m512i xx_ext = _mm512_cvtepi8_epi16(xx);
__m512i yy_ext = _mm512_cvtepi8_epi16(yy);
sum = _mm512_add_epi32(_mm512_madd_epi16(xx_ext, yy_ext), sum);
}
// reduce for 16 int32 integers in AVX512
__m256i sumh = _mm256_add_epi32(_mm512_extracti32x8_epi32(sum, 0), _mm512_extracti32x8_epi32(sum, 1));
__m128i sumhh = _mm_add_epi32(_mm256_castsi256_si128(sumh), _mm256_extracti128_si256(sumh, 1));
__m128i tmp = _mm_hadd_epi32(sumhh, sumhh);
_mm128_print_epi32(tmp);
return _mm_extract_epi32(tmp, 0) + _mm_extract_epi32(tmp, 1);
}
inline int NormalIP(const int8_t* x, const int8_t* y, int d) {
d = 32;
int32_t res = 0;
for (int i = 0; i < d; i++) {
int32_t tmp = (int32_t)x[i] * (int32_t)y[i];
res += tmp;
}
printf("%d\n", res);
return res;
}
inline int NormalSquareSum(const int8_t* x, int d) {
d = 32;
int32_t res = 0;
for (int i = 0; i < d; i++) {
int32_t tmp = (int32_t)x[i];
res += tmp * tmp;
}
printf("%d\n", res);
return res;
}
inline float
L2SqrSQ8_sym(const int8_t *x, const int8_t *y, int d) {
size_t i;
d = 32;
// __m512i sum = _mm512_setzero_si512();
__m256i sum = _mm256_setzero_si256();
for (int i = 0; i < d; i += 16) {
// convert to 512bit
__m128i xx = _mm_loadu_si128((__m128i *)(x + i));
__m128i yy = _mm_loadu_si128((__m128i *)(y + i));
// __m512i xx_ext = _mm512_cvtepi8_epi32(xx);
// __m512i yy_ext = _mm512_cvtepi8_epi32(yy);
// __m512i sub = _mm512_sub_epi32(xx_ext, yy_ext);
// __m512i square = _mm512_mullo_epi32(sub, sub);
// sum = _mm512_add_epi32(square, sum);
__m256i xx_ext = _mm256_cvtepi8_epi16(xx);
__m256i yy_ext = _mm256_cvtepi8_epi16(yy);
__m256i sub = _mm256_sub_epi16(xx_ext, yy_ext);
sum = _mm256_add_epi32(_mm256_madd_epi16(sub, sub), sum);
}
// reduce for 16 int32 integers in AVX512
// __m256i sumh = _mm256_add_epi32(_mm512_extracti32x8_epi32(sum, 0), _mm512_extracti32x8_epi32(sum, 1));
// __m128i sumhh = _mm_add_epi32(_mm256_castsi256_si128(sumh), _mm256_extracti128_si256(sumh, 1));
// __m128i tmp = _mm_hadd_epi32(sumhh, sumhh);
__m128i sumh = _mm_add_epi32(_mm256_extracti32x4_epi32(sum, 0), _mm256_extracti32x4_epi32(sum, 1));
__m128i tmp = _mm_hadd_epi32(sumh, sumh);
_mm128_print_epi32(tmp);
return _mm_extract_epi32(tmp, 0) + _mm_extract_epi32(tmp, 1);
}
inline float
L2SqrSQ8_sym_512(const int8_t *x, const int8_t *y, int d) {
size_t i;
d = 32;
__m512i sum = _mm512_setzero_si512();
for (int i = 0; i < d; i += 32) {
__m256i xx = _mm256_load_si256((__m256i *)(x + i));
__m256i yy = _mm256_load_si256((__m256i *)(y + i));
__m512i xx_ext = _mm512_cvtepi8_epi16(xx);
__m512i yy_ext = _mm512_cvtepi8_epi16(yy);
__m512i sub = _mm512_sub_epi16(xx_ext, yy_ext);
sum = _mm512_add_epi32(_mm512_madd_epi16(sub, sub), sum);
}
// reduce for 16 int32 integers in AVX512
__m256i sumh = _mm256_add_epi32(_mm512_extracti32x8_epi32(sum, 0), _mm512_extracti32x8_epi32(sum, 1));
__m128i sumhh = _mm_add_epi32(_mm256_castsi256_si128(sumh), _mm256_extracti128_si256(sumh, 1));
__m128i tmp = _mm_hadd_epi32(sumhh, sumhh);
_mm128_print_epi32(tmp);
return _mm_extract_epi32(tmp, 0) + _mm_extract_epi32(tmp, 1);
}
// 1. (x-y)^2 = x^2 + y^2 - 2xy, 分别算三个项，32个int16，然后再分别加和
static float
L2SqrSQ8_decouple(const int8_t* x, const int8_t* y, int d) {
int i;
// (x-y)^2 = x^2 + y^2 - 2xy
__m512i x_sqr = _mm512_setzero_si512();
__m512i y_sqr = _mm512_setzero_si512();
__m512i xy = _mm512_setzero_si512();
int32_t res = 0;
d = 32;
for (int i = 0; i < d; i += 32) {
__m256i xx = _mm256_loadu_si256((__m256i *)(x + i));
__m256i yy = _mm256_loadu_si256((__m256i *)(y + i));
__m512i xx_ext = _mm512_cvtepi8_epi16(xx);
__m512i yy_ext = _mm512_cvtepi8_epi16(yy);
__m512i xx_sqr = _mm512_mullo_epi16(xx_ext, xx_ext);
__m512i yy_sqr = _mm512_mullo_epi16(yy_ext, yy_ext);
__m512i dot = _mm512_mullo_epi16(xx_ext, yy_ext);
x_sqr = _mm512_add_epi16(x_sqr, xx_sqr);
y_sqr = _mm512_add_epi16(y_sqr, yy_sqr);
xy = _mm512_add_epi16(xy, dot);
}
// reduce
}
float normal_l2(float const *a, float const *b, unsigned dim)
{
float r = 0;
for (unsigned i = 0; i < dim; ++i)
{
float v = float(a[i]) - float(a[i]);
v *= v;
r += v;
}
return r;
}
int normal_l2(int8_t const *a, int8_t const *b, unsigned dim)
{
int r = 0;
for (unsigned i = 0; i < dim; ++i)
{
int v = int(a[i]) - int(b[i]);
v *= v;
r += v;
}
return r;
}
int main(int argc, char **argv)
{
int8_t a0[32] = {127, 127, 127, -4, 5, 6, 7, 8, 9, -127, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, -24, 25, 26, 27, 28, 29, 30, 31, 32};
int8_t a1[32] = {-127, -127, 4, 5, 6, 7, 8, 9, 10, 127, 12, -13, 14, 15, 16, 17, 18, 19, -20, 21, 22, 23, 24, -25, 26, 27, 28, 29, 30, 31, 32, 33};
L2SqrSQ8_sym_512((int8_t*)a0, (int8_t*)a1, 32);
printf("%d\n", normal_l2(a0, a1, 32));
// SquareSum(a0, 32);
// NormalSquareSum(a0, 32);
// IPSQ8(a0, a1, 32);
// NormalIP(a0, a1, 32);
return 0;
}