-
Notifications
You must be signed in to change notification settings - Fork 0
/
euclideanDistance.hpp
272 lines (240 loc) · 7.63 KB
/
euclideanDistance.hpp
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
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
#pragma once
#include <stdexcept>
#include "DistanceFunction.hpp"
namespace chm {
static float euclideanDistance(
const float* node, const float* query, const size_t dim,
const size_t, const size_t, const size_t
) {
auto res = 0.f;
for(size_t i = 0; i < dim; i++) {
const auto diff = node[i] - query[i];
res += diff * diff;
}
return res;
}
FunctionInfo euc(euclideanDistance, "euc");
#if defined(AVX_CAPABLE)
static float euclideanDistance16AVX(
const float* node, const float* query, const size_t,
const size_t, const size_t dim16, const size_t
) {
__m256 diff, v1, v2;
const float* end = node + dim16;
float PORTABLE_ALIGN32 tmp[8];
__m256 sum = _mm256_set1_ps(0);
while (node < end) {
v1 = _mm256_loadu_ps(node);
node += 8;
v2 = _mm256_loadu_ps(query);
query += 8;
diff = _mm256_sub_ps(v1, v2);
sum = _mm256_add_ps(sum, _mm256_mul_ps(diff, diff));
v1 = _mm256_loadu_ps(node);
node += 8;
v2 = _mm256_loadu_ps(query);
query += 8;
diff = _mm256_sub_ps(v1, v2);
sum = _mm256_add_ps(sum, _mm256_mul_ps(diff, diff));
}
_mm256_store_ps(tmp, sum);
return
tmp[0] + tmp[1] + tmp[2] + tmp[3] +
tmp[4] + tmp[5] + tmp[6] + tmp[7];
}
static float euclideanDistance16ResidualAVX(
const float* node, const float* query, const size_t,
const size_t dim4, const size_t dim16, const size_t dimLeft
) {
const float front = euclideanDistance16AVX(node, query, 0, dim4, dim16, 0);
const float back = euclideanDistance(node + dim16, query + dim16, dimLeft, 0, 0, 0);
return front + back;
}
FunctionInfo euc16AVX(euclideanDistance16AVX, "euc16AVX");
FunctionInfo euc16RAVX(euclideanDistance16ResidualAVX, "euc16RAVX");
#endif
#if defined(AVX512_CAPABLE)
static float euclideanDistance16AVX512(
const float* node, const float* query, const size_t,
const size_t, const size_t dim16, const size_t
) {
__m512 diff, v1, v2;
const float* end = node + dim16;
__m512 sum = _mm512_set1_ps(0);
float PORTABLE_ALIGN64 tmp[16];
while (node < end) {
v1 = _mm512_loadu_ps(node);
node += 16;
v2 = _mm512_loadu_ps(query);
query += 16;
diff = _mm512_sub_ps(v1, v2);
sum = _mm512_add_ps(sum, _mm512_mul_ps(diff, diff));
}
_mm512_store_ps(tmp, sum);
return
tmp[0] + tmp[1] + tmp[2] + tmp[3] +
tmp[4] + tmp[5] + tmp[6] + tmp[7] +
tmp[8] + tmp[9] + tmp[10] + tmp[11] +
tmp[12] + tmp[13] + tmp[14] + tmp[15];
}
static float euclideanDistance16ResidualAVX512(
const float* node, const float* query, const size_t,
const size_t dim4, const size_t dim16, const size_t dimLeft
) {
const float front = euclideanDistance16AVX512(node, query, 0, dim4, dim16, 0);
const float back = euclideanDistance(node + dim16, query + dim16, dimLeft, 0, 0, 0);
return front + back;
}
FunctionInfo euc16AVX512(euclideanDistance16AVX512, "euc16AVX512");
FunctionInfo euc16RAVX512(euclideanDistance16ResidualAVX512, "euc16RAVX512");
#endif
#if defined(SSE_CAPABLE)
static float euclideanDistance16SSE(
const float* node, const float* query, const size_t,
const size_t, const size_t dim16, const size_t
) {
__m128 diff, v1, v2;
const float* end = node + dim16;
__m128 sum = _mm_set1_ps(0);
float PORTABLE_ALIGN32 tmp[8];
while (node < end) {
v1 = _mm_loadu_ps(node);
node += 4;
v2 = _mm_loadu_ps(query);
query += 4;
diff = _mm_sub_ps(v1, v2);
sum = _mm_add_ps(sum, _mm_mul_ps(diff, diff));
v1 = _mm_loadu_ps(node);
node += 4;
v2 = _mm_loadu_ps(query);
query += 4;
diff = _mm_sub_ps(v1, v2);
sum = _mm_add_ps(sum, _mm_mul_ps(diff, diff));
v1 = _mm_loadu_ps(node);
node += 4;
v2 = _mm_loadu_ps(query);
query += 4;
diff = _mm_sub_ps(v1, v2);
sum = _mm_add_ps(sum, _mm_mul_ps(diff, diff));
v1 = _mm_loadu_ps(node);
node += 4;
v2 = _mm_loadu_ps(query);
query += 4;
diff = _mm_sub_ps(v1, v2);
sum = _mm_add_ps(sum, _mm_mul_ps(diff, diff));
}
_mm_store_ps(tmp, sum);
return tmp[0] + tmp[1] + tmp[2] + tmp[3];
}
static float euclideanDistance4SSE(
const float* node, const float* query, const size_t,
const size_t dim4, const size_t, const size_t
) {
__m128 diff, v1, v2;
const float* end = node + dim4;
__m128 sum = _mm_set1_ps(0);
float PORTABLE_ALIGN32 tmp[8];
while (node < end) {
v1 = _mm_loadu_ps(node);
node += 4;
v2 = _mm_loadu_ps(query);
query += 4;
diff = _mm_sub_ps(v1, v2);
sum = _mm_add_ps(sum, _mm_mul_ps(diff, diff));
}
_mm_store_ps(tmp, sum);
return tmp[0] + tmp[1] + tmp[2] + tmp[3];
}
static float euclideanDistance4ResidualSSE(
const float* node, const float* query, const size_t,
const size_t dim4, const size_t dim16, const size_t dimLeft
) {
const float front = euclideanDistance4SSE(node, query, 0, dim4, dim16, 0);
const float back = euclideanDistance(node + dim4, query + dim4, dimLeft, 0, 0, 0);
return front + back;
}
static float euclideanDistance16ResidualSSE(
const float* node, const float* query, const size_t,
const size_t dim4, const size_t dim16, const size_t dimLeft
) {
const float front = euclideanDistance16SSE(node, query, 0, dim4, dim16, 0);
const float back = euclideanDistance(node + dim16, query + dim16, dimLeft, 0, 0, 0);
return front + back;
}
FunctionInfo euc16SSE(euclideanDistance16SSE, "euc16SSE");
FunctionInfo euc4SSE(euclideanDistance4SSE, "euc4SSE");
FunctionInfo euc4RSSE(euclideanDistance4ResidualSSE, "euc4RSSE");
FunctionInfo euc16RSSE(euclideanDistance16ResidualSSE, "euc16RSSE");
#endif
inline DistanceInfo getEuclideanInfo(
const size_t dim, const size_t dim4, const size_t dim16, SIMDType type
) {
#if defined(SIMD_CAPABLE)
if(type == SIMDType::NONE)
return DistanceInfo(0, euc);
if(type == SIMDType::BEST)
type = getBestSIMDType();
if(dim % 16 == 0)
switch(type) {
case SIMDType::AVX:
#if defined(AVX_CAPABLE)
return DistanceInfo(0, euc16AVX);
#else
throw std::runtime_error("This CPU doesn't support AVX.");
#endif
case SIMDType::AVX512:
#if defined(AVX512_CAPABLE)
return DistanceInfo(0, euc16AVX512);
#else
throw std::runtime_error("This CPU doesn't support AVX512.");
#endif
case SIMDType::SSE:
#if defined(SSE_CAPABLE)
return DistanceInfo(0, euc16SSE);
#else
throw std::runtime_error("This CPU doesn't support SSE.");
#endif
default:
throw std::runtime_error("Unknown SIMD type.");
}
else if(dim % 4 == 0)
#if defined(SSE_CAPABLE)
return DistanceInfo(0, euc4SSE);
#else
throw std::runtime_error("This CPU doesn't support SSE.");
#endif
else if (dim > 16) {
const auto dimLeft = dim - dim16;
switch(type) {
case SIMDType::AVX:
#if defined(AVX_CAPABLE)
return DistanceInfo(dimLeft, euc16RAVX);
#else
throw std::runtime_error("This CPU doesn't support AVX.");
#endif
case SIMDType::AVX512:
#if defined(AVX512_CAPABLE)
return DistanceInfo(dimLeft, euc16RAVX512);
#else
throw std::runtime_error("This CPU doesn't support AVX512.");
#endif
case SIMDType::SSE:
#if defined(SSE_CAPABLE)
return DistanceInfo(dimLeft, euc16RSSE);
#else
throw std::runtime_error("This CPU doesn't support SSE.");
#endif
default:
throw std::runtime_error("Unknown SIMD type.");
}
}
else if (dim > 4)
#if defined(SSE_CAPABLE)
return DistanceInfo(dim - dim4, euc4RSSE);
#else
throw std::runtime_error("This CPU doesn't support SSE.");
#endif
#endif
return DistanceInfo(0, euc);
}
}