forked from facebookresearch/faiss
-
Notifications
You must be signed in to change notification settings - Fork 0
/
hamming.cpp
777 lines (671 loc) · 19.9 KB
/
hamming.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
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
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
/**
* Copyright (c) 2015-present, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under the BSD+Patents license found in the
* LICENSE file in the root directory of this source tree.
*/
// -*- c++ -*-
/*
* Implementation of Hamming related functions (distances, smallest distance
* selection with regular heap|radix and probabilistic heap|radix.
*
* IMPLEMENTATION NOTES
* Bitvectors are generally assumed to be multiples of 64 bits.
*
* hamdis_t is used for distances because at this time
* it is not clear how we will need to balance
* - flexibility in vector size (unclear more than 2^16 or even 2^8 bitvectors)
* - memory usage
* - cache-misses when dealing with large volumes of data (lower bits is better)
*
* The hamdis_t should optimally be compatibe with one of the Torch Storage
* (Byte,Short,Long) and therefore should be signed for 2-bytes and 4-bytes
*/
#include "hamming.h"
#include <vector>
#include <memory>
#include <stdlib.h>
#include <stdio.h>
#include <math.h>
#include <assert.h>
#include <limits.h>
#include "Heap.h"
#include "FaissAssert.h"
static const size_t BLOCKSIZE_QUERY = 8192;
namespace faiss {
size_t hamming_batch_size = 65536;
static const uint8_t hamdis_tab_ham_bytes[256] = {
0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8
};
/* Elementary Hamming distance computation: unoptimized */
template <size_t nbits, typename T>
T hamming (const uint8_t *bs1,
const uint8_t *bs2)
{
const size_t nbytes = nbits / 8;
size_t i;
T h = 0;
for (i = 0; i < nbytes; i++)
h += (T) hamdis_tab_ham_bytes[bs1[i]^bs2[i]];
return h;
}
/* Hamming distances for multiples of 64 bits */
template <size_t nbits>
hamdis_t hamming (const uint64_t * bs1, const uint64_t * bs2)
{
const size_t nwords = nbits / 64;
size_t i;
hamdis_t h = 0;
for (i = 0; i < nwords; i++)
h += popcount64 (bs1[i] ^ bs2[i]);
return h;
}
/* specialized (optimized) functions */
template <>
hamdis_t hamming<64> (const uint64_t * pa, const uint64_t * pb)
{
return popcount64 (pa[0] ^ pb[0]);
}
template <>
hamdis_t hamming<128> (const uint64_t *pa, const uint64_t *pb)
{
return popcount64 (pa[0] ^ pb[0]) + popcount64(pa[1] ^ pb[1]);
}
template <>
hamdis_t hamming<256> (const uint64_t * pa, const uint64_t * pb)
{
return popcount64 (pa[0] ^ pb[0])
+ popcount64 (pa[1] ^ pb[1])
+ popcount64 (pa[2] ^ pb[2])
+ popcount64 (pa[3] ^ pb[3]);
}
/* Hamming distances for multiple of 64 bits */
hamdis_t hamming (
const uint64_t * bs1,
const uint64_t * bs2,
size_t nwords)
{
size_t i;
hamdis_t h = 0;
for (i = 0; i < nwords; i++)
h += popcount64 (bs1[i] ^ bs2[i]);
return h;
}
template <size_t nbits>
void hammings (
const uint64_t * bs1,
const uint64_t * bs2,
size_t n1, size_t n2,
hamdis_t * dis)
{
size_t i, j;
const size_t nwords = nbits / 64;
for (i = 0; i < n1; i++) {
const uint64_t * __restrict bs1_ = bs1 + i * nwords;
hamdis_t * __restrict dis_ = dis + i * n2;
for (j = 0; j < n2; j++)
dis_[j] = hamming<nbits>(bs1_, bs2 + j * nwords);
}
}
void hammings (
const uint64_t * bs1,
const uint64_t * bs2,
size_t n1,
size_t n2,
size_t nwords,
hamdis_t * __restrict dis)
{
size_t i, j;
n1 *= nwords;
n2 *= nwords;
for (i = 0; i < n1; i+=nwords) {
const uint64_t * bs1_ = bs1+i;
for (j = 0; j < n2; j+=nwords)
dis[j] = hamming (bs1_, bs2+j, nwords);
}
}
/* Count number of matches given a max threshold */
template <size_t nbits>
void hamming_count_thres (
const uint64_t * bs1,
const uint64_t * bs2,
size_t n1,
size_t n2,
hamdis_t ht,
size_t * nptr)
{
const size_t nwords = nbits / 64;
size_t i, j, posm = 0;
const uint64_t * bs2_ = bs2;
for (i = 0; i < n1; i++) {
bs2 = bs2_;
for (j = 0; j < n2; j++) {
/* collect the match only if this satisfies the threshold */
if (hamming <nbits> (bs1, bs2) <= ht)
posm++;
bs2 += nwords;
}
bs1 += nwords; /* next signature */
}
*nptr = posm;
}
template <size_t nbits>
void crosshamming_count_thres (
const uint64_t * dbs,
size_t n,
int ht,
size_t * nptr)
{
const size_t nwords = nbits / 64;
size_t i, j, posm = 0;
const uint64_t * bs1 = dbs;
for (i = 0; i < n; i++) {
const uint64_t * bs2 = bs1 + 2;
for (j = i + 1; j < n; j++) {
/* collect the match only if this satisfies the threshold */
if (hamming <nbits> (bs1, bs2) <= ht)
posm++;
bs2 += nwords;
}
bs1 += nwords;
}
*nptr = posm;
}
template <size_t nbits>
size_t match_hamming_thres (
const uint64_t * bs1,
const uint64_t * bs2,
size_t n1,
size_t n2,
int ht,
long * idx,
hamdis_t * hams)
{
const size_t nwords = nbits / 64;
size_t i, j, posm = 0;
hamdis_t h;
const uint64_t * bs2_ = bs2;
for (i = 0; i < n1; i++) {
bs2 = bs2_;
for (j = 0; j < n2; j++) {
/* Here perform the real work of computing the distance */
h = hamming <nbits> (bs1, bs2);
/* collect the match only if this satisfies the threshold */
if (h <= ht) {
/* Enough space to store another match ? */
*idx = i; idx++;
*idx = j; idx++;
*hams = h;
hams++;
posm++;
}
bs2+=nwords; /* next signature */
}
bs1+=nwords;
}
return posm;
}
/* Return closest neighbors w.r.t Hamming distance, using a heap. */
template <class HammingComputer>
static
void hammings_knn_hc (
int bytes_per_code,
int_maxheap_array_t * ha,
const uint8_t * bs1,
const uint8_t * bs2,
size_t n2,
bool order = true,
bool init_heap = true)
{
size_t k = ha->k;
if (init_heap) ha->heapify ();
const size_t block_size = hamming_batch_size;
for (size_t j0 = 0; j0 < n2; j0 += block_size) {
const size_t j1 = std::min(j0 + block_size, n2);
#pragma omp parallel for
for (size_t i = 0; i < ha->nh; i++) {
HammingComputer hc (bs1 + i * bytes_per_code, bytes_per_code);
const uint8_t * bs2_ = bs2 + j0 * bytes_per_code;
hamdis_t dis;
hamdis_t * __restrict bh_val_ = ha->val + i * k;
long * __restrict bh_ids_ = ha->ids + i * k;
size_t j;
for (j = j0; j < j1; j++, bs2_+= bytes_per_code) {
dis = hc.hamming (bs2_);
if (dis < bh_val_[0]) {
faiss::maxheap_pop<hamdis_t> (k, bh_val_, bh_ids_);
faiss::maxheap_push<hamdis_t> (k, bh_val_, bh_ids_, dis, j);
}
}
}
}
if (order) ha->reorder ();
}
/* Return closest neighbors w.r.t Hamming distance, using max count. */
template <class HammingComputer>
static
void hammings_knn_mc (
int bytes_per_code,
const uint8_t *a,
const uint8_t *b,
size_t na,
size_t nb,
size_t k,
int32_t *distances,
long *labels)
{
const int nBuckets = bytes_per_code * 8 + 1;
std::vector<int> all_counters(na * nBuckets, 0);
std::unique_ptr<long[]> all_ids_per_dis(new long[na * nBuckets * k]);
std::vector<HCounterState<HammingComputer>> cs;
for (size_t i = 0; i < na; ++i) {
cs.push_back(HCounterState<HammingComputer>(
all_counters.data() + i * nBuckets,
all_ids_per_dis.get() + i * nBuckets * k,
a + i * bytes_per_code,
8 * bytes_per_code,
k
));
}
const size_t block_size = hamming_batch_size;
for (size_t j0 = 0; j0 < nb; j0 += block_size) {
const size_t j1 = std::min(j0 + block_size, nb);
#pragma omp parallel for
for (size_t i = 0; i < na; ++i) {
for (size_t j = j0; j < j1; ++j) {
cs[i].update_counter(b + j * bytes_per_code, j);
}
}
}
for (size_t i = 0; i < na; ++i) {
HCounterState<HammingComputer>& csi = cs[i];
int nres = 0;
for (int b = 0; b < nBuckets && nres < k; b++) {
for (int l = 0; l < csi.counters[b] && nres < k; l++) {
labels[i * k + nres] = csi.ids_per_dis[b * k + l];
distances[i * k + nres] = b;
nres++;
}
}
while (nres < k) {
labels[i * k + nres] = -1;
distances[i * k + nres] = std::numeric_limits<int32_t>::max();
++nres;
}
}
}
// works faster than the template version
static
void hammings_knn_hc_1 (
int_maxheap_array_t * ha,
const uint64_t * bs1,
const uint64_t * bs2,
size_t n2,
bool order = true,
bool init_heap = true)
{
const size_t nwords = 1;
size_t k = ha->k;
if (init_heap) {
ha->heapify ();
}
#pragma omp parallel for
for (size_t i = 0; i < ha->nh; i++) {
const uint64_t bs1_ = bs1 [i];
const uint64_t * bs2_ = bs2;
hamdis_t dis;
hamdis_t * bh_val_ = ha->val + i * k;
hamdis_t bh_val_0 = bh_val_[0];
long * bh_ids_ = ha->ids + i * k;
size_t j;
for (j = 0; j < n2; j++, bs2_+= nwords) {
dis = popcount64 (bs1_ ^ *bs2_);
if (dis < bh_val_0) {
faiss::maxheap_pop<hamdis_t> (k, bh_val_, bh_ids_);
faiss::maxheap_push<hamdis_t> (k, bh_val_, bh_ids_, dis, j);
bh_val_0 = bh_val_[0];
}
}
}
if (order) {
ha->reorder ();
}
}
/* Functions to maps vectors to bits. Assume proper allocation done beforehand,
meaning that b should be be able to receive as many bits as x may produce. */
/*
* dimension 0 corresponds to the least significant bit of b[0], or
* equivalently to the lsb of the first byte that is stored.
*/
void fvec2bitvec (const float * x, uint8_t * b, size_t d)
{
for (int i = 0; i < d; i += 8) {
uint8_t w = 0;
uint8_t mask = 1;
int nj = i + 8 <= d ? 8 : d - i;
for (int j = 0; j < nj; j++) {
if (x[i + j] >= 0)
w |= mask;
mask <<= 1;
}
*b = w;
b++;
}
}
/* Same but for n vectors.
Ensure that the ouptut b is byte-aligned (pad with 0s). */
void fvecs2bitvecs (const float * x, uint8_t * b, size_t d, size_t n)
{
const long ncodes = ((d + 7) / 8);
#pragma omp parallel for
for (size_t i = 0; i < n; i++)
fvec2bitvec (x + i * d, b + i * ncodes, d);
}
/* Reverse bit (NOT a optimized function, only used for print purpose) */
static uint64_t uint64_reverse_bits (uint64_t b)
{
int i;
uint64_t revb = 0;
for (i = 0; i < 64; i++) {
revb <<= 1;
revb |= b & 1;
b >>= 1;
}
return revb;
}
/* print the bit vector */
void bitvec_print (const uint8_t * b, size_t d)
{
size_t i, j;
for (i = 0; i < d; ) {
uint64_t brev = uint64_reverse_bits (* (uint64_t *) b);
for (j = 0; j < 64 && i < d; j++, i++) {
printf ("%d", (int) (brev & 1));
brev >>= 1;
}
b += 8;
printf (" ");
}
}
/*----------------------------------------*/
/* Hamming distance computation and k-nn */
#define C64(x) ((uint64_t *)x)
/* Compute a set of Hamming distances */
void hammings (
const uint8_t * a,
const uint8_t * b,
size_t na, size_t nb,
size_t ncodes,
hamdis_t * __restrict dis)
{
FAISS_THROW_IF_NOT (ncodes % 8 == 0);
switch (ncodes) {
case 8:
faiss::hammings <64> (C64(a), C64(b), na, nb, dis); return;
case 16:
faiss::hammings <128> (C64(a), C64(b), na, nb, dis); return;
case 32:
faiss::hammings <256> (C64(a), C64(b), na, nb, dis); return;
case 64:
faiss::hammings <512> (C64(a), C64(b), na, nb, dis); return;
default:
faiss::hammings (C64(a), C64(b), na, nb, ncodes * 8, dis); return;
}
}
void hammings_knn(
int_maxheap_array_t *ha,
const uint8_t *a,
const uint8_t *b,
size_t nb,
size_t ncodes,
int order)
{
hammings_knn_hc(ha, a, b, nb, ncodes, order);
}
void hammings_knn_hc (
int_maxheap_array_t * ha,
const uint8_t * a,
const uint8_t * b,
size_t nb,
size_t ncodes,
int order)
{
switch (ncodes) {
case 4:
hammings_knn_hc<faiss::HammingComputer4>
(4, ha, a, b, nb, order, true);
break;
case 8:
hammings_knn_hc_1 (ha, C64(a), C64(b), nb, order, true);
// hammings_knn_hc<faiss::HammingComputer8>
// (8, ha, a, b, nb, order, true);
break;
case 16:
hammings_knn_hc<faiss::HammingComputer16>
(16, ha, a, b, nb, order, true);
break;
case 32:
hammings_knn_hc<faiss::HammingComputer32>
(32, ha, a, b, nb, order, true);
break;
default:
if(ncodes % 8 == 0) {
hammings_knn_hc<faiss::HammingComputerM8>
(ncodes, ha, a, b, nb, order, true);
} else {
hammings_knn_hc<faiss::HammingComputerDefault>
(ncodes, ha, a, b, nb, order, true);
}
}
}
void hammings_knn_mc(
const uint8_t * a,
const uint8_t * b,
size_t na,
size_t nb,
size_t k,
size_t ncodes,
int32_t *distances,
long *labels)
{
switch (ncodes) {
case 4:
hammings_knn_mc<faiss::HammingComputer4>(
4, a, b, na, nb, k, distances, labels
);
break;
case 8:
// TODO(hoss): Write analog to hammings_knn_hc_1
// hammings_knn_hc_1 (ha, C64(a), C64(b), nb, order, true);
hammings_knn_mc<faiss::HammingComputer8>(
8, a, b, na, nb, k, distances, labels
);
break;
case 16:
hammings_knn_mc<faiss::HammingComputer16>(
16, a, b, na, nb, k, distances, labels
);
break;
case 32:
hammings_knn_mc<faiss::HammingComputer32>(
32, a, b, na, nb, k, distances, labels
);
break;
default:
if(ncodes % 8 == 0) {
hammings_knn_mc<faiss::HammingComputerM8>(
ncodes, a, b, na, nb, k, distances, labels
);
} else {
hammings_knn_mc<faiss::HammingComputerDefault>(
ncodes, a, b, na, nb, k, distances, labels
);
}
}
}
/* Count number of matches given a max threshold */
void hamming_count_thres (
const uint8_t * bs1,
const uint8_t * bs2,
size_t n1,
size_t n2,
hamdis_t ht,
size_t ncodes,
size_t * nptr)
{
switch (ncodes) {
case 8:
faiss::hamming_count_thres <64> (C64(bs1), C64(bs2),
n1, n2, ht, nptr);
return;
case 16:
faiss::hamming_count_thres <128> (C64(bs1), C64(bs2),
n1, n2, ht, nptr);
return;
case 32:
faiss::hamming_count_thres <256> (C64(bs1), C64(bs2),
n1, n2, ht, nptr);
return;
case 64:
faiss::hamming_count_thres <512> (C64(bs1), C64(bs2),
n1, n2, ht, nptr);
return;
default:
FAISS_THROW_FMT ("not implemented for %zu bits", ncodes);
}
}
/* Count number of cross-matches given a threshold */
void crosshamming_count_thres (
const uint8_t * dbs,
size_t n,
hamdis_t ht,
size_t ncodes,
size_t * nptr)
{
switch (ncodes) {
case 8:
faiss::crosshamming_count_thres <64> (C64(dbs), n, ht, nptr);
return;
case 16:
faiss::crosshamming_count_thres <128> (C64(dbs), n, ht, nptr);
return;
case 32:
faiss::crosshamming_count_thres <256> (C64(dbs), n, ht, nptr);
return;
case 64:
faiss::crosshamming_count_thres <512> (C64(dbs), n, ht, nptr);
return;
default:
FAISS_THROW_FMT ("not implemented for %zu bits", ncodes);
}
}
/* Returns all matches given a threshold */
size_t match_hamming_thres (
const uint8_t * bs1,
const uint8_t * bs2,
size_t n1,
size_t n2,
hamdis_t ht,
size_t ncodes,
long * idx,
hamdis_t * dis)
{
switch (ncodes) {
case 8:
return faiss::match_hamming_thres <64> (C64(bs1), C64(bs2),
n1, n2, ht, idx, dis);
case 16:
return faiss::match_hamming_thres <128> (C64(bs1), C64(bs2),
n1, n2, ht, idx, dis);
case 32:
return faiss::match_hamming_thres <256> (C64(bs1), C64(bs2),
n1, n2, ht, idx, dis);
case 64:
return faiss::match_hamming_thres <512> (C64(bs1), C64(bs2),
n1, n2, ht, idx, dis);
default:
FAISS_THROW_FMT ("not implemented for %zu bits", ncodes);
return 0;
}
}
#undef C64
/*************************************
* generalized Hamming distances
************************************/
template <class HammingComputer>
static void hamming_dis_inner_loop (
const uint8_t *ca,
const uint8_t *cb,
size_t nb,
size_t code_size,
int k,
hamdis_t * bh_val_,
long * bh_ids_)
{
HammingComputer hc (ca, code_size);
for (size_t j = 0; j < nb; j++) {
int ndiff = hc.hamming (cb);
cb += code_size;
if (ndiff < bh_val_[0]) {
maxheap_pop<hamdis_t> (k, bh_val_, bh_ids_);
maxheap_push<hamdis_t> (k, bh_val_, bh_ids_, ndiff, j);
}
}
}
void generalized_hammings_knn_hc (
int_maxheap_array_t * ha,
const uint8_t * a,
const uint8_t * b,
size_t nb,
size_t code_size,
int ordered)
{
int na = ha->nh;
int k = ha->k;
if (ordered)
ha->heapify ();
#pragma omp parallel for
for (int i = 0; i < na; i++) {
const uint8_t *ca = a + i * code_size;
const uint8_t *cb = b;
hamdis_t * bh_val_ = ha->val + i * k;
long * bh_ids_ = ha->ids + i * k;
switch (code_size) {
case 8:
hamming_dis_inner_loop<GenHammingComputer8>
(ca, cb, nb, 8, k, bh_val_, bh_ids_);
break;
case 16:
hamming_dis_inner_loop<GenHammingComputer16>
(ca, cb, nb, 16, k, bh_val_, bh_ids_);
break;
case 32:
hamming_dis_inner_loop<GenHammingComputer32>
(ca, cb, nb, 32, k, bh_val_, bh_ids_);
break;
default:
hamming_dis_inner_loop<GenHammingComputerM8>
(ca, cb, nb, code_size, k, bh_val_, bh_ids_);
break;
}
}
if (ordered)
ha->reorder ();
}
} // namespace faiss