forked from facebookresearch/faiss
-
Notifications
You must be signed in to change notification settings - Fork 2
/
IndexScalarQuantizer.cpp
1316 lines (1054 loc) · 36.5 KB
/
IndexScalarQuantizer.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
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/**
* 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++ -*-
#include "IndexScalarQuantizer.h"
#include <cstdio>
#include <algorithm>
#include <omp.h>
#ifdef __SSE__
#include <immintrin.h>
#endif
#include "utils.h"
#include "FaissAssert.h"
namespace faiss {
/*******************************************************************
* ScalarQuantizer implementation
*
* The main source of complexity is to support combinations of 4
* variants without incurring runtime tests or virtual function calls:
*
* - 4 / 8 bits per code component
* - uniform / non-uniform
* - IP / L2 distance search
* - scalar / AVX distance computation
*
* The appropriate Quantizer object is returned via select_quantizer
* that hides the template mess.
********************************************************************/
#ifdef __AVX__
#define USE_AVX
#endif
namespace {
typedef Index::idx_t idx_t;
typedef ScalarQuantizer::QuantizerType QuantizerType;
typedef ScalarQuantizer::RangeStat RangeStat;
using DistanceComputer = ScalarQuantizer::DistanceComputer;
/*******************************************************************
* Codec: converts between values in [0, 1] and an index in a code
* array. The "i" parameter is the vector component index (not byte
* index).
*/
struct Codec8bit {
static void encode_component (float x, uint8_t *code, int i) {
code[i] = (int)(255 * x);
}
static float decode_component (const uint8_t *code, int i) {
return (code[i] + 0.5f) / 255.0f;
}
#ifdef USE_AVX
static __m256 decode_8_components (const uint8_t *code, int i) {
uint64_t c8 = *(uint64_t*)(code + i);
__m128i c4lo = _mm_cvtepu8_epi32 (_mm_set1_epi32(c8));
__m128i c4hi = _mm_cvtepu8_epi32 (_mm_set1_epi32(c8 >> 32));
// __m256i i8 = _mm256_set_m128i(c4lo, c4hi);
__m256i i8 = _mm256_castsi128_si256 (c4lo);
i8 = _mm256_insertf128_si256 (i8, c4hi, 1);
__m256 f8 = _mm256_cvtepi32_ps (i8);
__m256 half = _mm256_set1_ps (0.5f);
f8 += half;
__m256 one_255 = _mm256_set1_ps (1.f / 255.f);
return f8 * one_255;
}
#endif
};
struct Codec4bit {
static void encode_component (float x, uint8_t *code, int i) {
code [i / 2] |= (int)(x * 15.0) << ((i & 1) << 2);
}
static float decode_component (const uint8_t *code, int i) {
return (((code[i / 2] >> ((i & 1) << 2)) & 0xf) + 0.5f) / 15.0f;
}
#ifdef USE_AVX
static __m256 decode_8_components (const uint8_t *code, int i) {
uint32_t c4 = *(uint32_t*)(code + (i >> 1));
uint32_t mask = 0x0f0f0f0f;
uint32_t c4ev = c4 & mask;
uint32_t c4od = (c4 >> 4) & mask;
// the 8 lower bytes of c8 contain the values
__m128i c8 = _mm_unpacklo_epi8 (_mm_set1_epi32(c4ev),
_mm_set1_epi32(c4od));
__m128i c4lo = _mm_cvtepu8_epi32 (c8);
__m128i c4hi = _mm_cvtepu8_epi32 (_mm_srli_si128(c8, 4));
__m256i i8 = _mm256_castsi128_si256 (c4lo);
i8 = _mm256_insertf128_si256 (i8, c4hi, 1);
__m256 f8 = _mm256_cvtepi32_ps (i8);
__m256 half = _mm256_set1_ps (0.5f);
f8 += half;
__m256 one_255 = _mm256_set1_ps (1.f / 15.f);
return f8 * one_255;
}
#endif
};
#ifdef USE_AVX
uint16_t encode_fp16 (float x) {
__m128 xf = _mm_set1_ps (x);
__m128i xi = _mm_cvtps_ph (
xf, _MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC);
return _mm_cvtsi128_si32 (xi) & 0xffff;
}
float decode_fp16 (uint16_t x) {
__m128i xi = _mm_set1_epi16 (x);
__m128 xf = _mm_cvtph_ps (xi);
return _mm_cvtss_f32 (xf);
}
#else
// non-intrinsic FP16 <-> FP32 code adapted from
// https://github.com/ispc/ispc/blob/master/stdlib.ispc
float floatbits (uint32_t x) {
void *xptr = &x;
return *(float*)xptr;
}
uint32_t intbits (float f) {
void *fptr = &f;
return *(uint32_t*)fptr;
}
uint16_t encode_fp16 (float f) {
// via Fabian "ryg" Giesen.
// https://gist.github.com/2156668
uint32_t sign_mask = 0x80000000u;
int32_t o;
uint32_t fint = intbits(f);
uint32_t sign = fint & sign_mask;
fint ^= sign;
// NOTE all the integer compares in this function can be safely
// compiled into signed compares since all operands are below
// 0x80000000. Important if you want fast straight SSE2 code (since
// there's no unsigned PCMPGTD).
// Inf or NaN (all exponent bits set)
// NaN->qNaN and Inf->Inf
// unconditional assignment here, will override with right value for
// the regular case below.
uint32_t f32infty = 255u << 23;
o = (fint > f32infty) ? 0x7e00u : 0x7c00u;
// (De)normalized number or zero
// update fint unconditionally to save the blending; we don't need it
// anymore for the Inf/NaN case anyway.
const uint32_t round_mask = ~0xfffu;
const uint32_t magic = 15u << 23;
const uint32_t f16infty = 31u << 23;
// Shift exponent down, denormalize if necessary.
// NOTE This represents half-float denormals using single
// precision denormals. The main reason to do this is that
// there's no shift with per-lane variable shifts in SSE*, which
// we'd otherwise need. It has some funky side effects though:
// - This conversion will actually respect the FTZ (Flush To Zero)
// flag in MXCSR - if it's set, no half-float denormals will be
// generated. I'm honestly not sure whether this is good or
// bad. It's definitely interesting.
// - If the underlying HW doesn't support denormals (not an issue
// with Intel CPUs, but might be a problem on GPUs or PS3 SPUs),
// you will always get flush-to-zero behavior. This is bad,
// unless you're on a CPU where you don't care.
// - Denormals tend to be slow. FP32 denormals are rare in
// practice outside of things like recursive filters in DSP -
// not a typical half-float application. Whether FP16 denormals
// are rare in practice, I don't know. Whatever slow path your
// HW may or may not have for denormals, this may well hit it.
float fscale = floatbits(fint & round_mask) * floatbits(magic);
fscale = std::min(fscale, floatbits((31u << 23) - 0x1000u));
int32_t fint2 = intbits(fscale) - round_mask;
if (fint < f32infty)
o = fint2 >> 13; // Take the bits!
return (o | (sign >> 16));
}
float decode_fp16 (uint16_t h) {
// https://gist.github.com/2144712
// Fabian "ryg" Giesen.
const uint32_t shifted_exp = 0x7c00u << 13; // exponent mask after shift
int32_t o = ((int32_t)(h & 0x7fffu)) << 13; // exponent/mantissa bits
int32_t exp = shifted_exp & o; // just the exponent
o += (int32_t)(127 - 15) << 23; // exponent adjust
int32_t infnan_val = o + ((int32_t)(128 - 16) << 23);
int32_t zerodenorm_val = intbits(
floatbits(o + (1u<<23)) - floatbits(113u << 23));
int32_t reg_val = (exp == 0) ? zerodenorm_val : o;
int32_t sign_bit = ((int32_t)(h & 0x8000u)) << 16;
return floatbits(((exp == shifted_exp) ? infnan_val : reg_val) | sign_bit);
}
#endif
/*******************************************************************
* Quantizer: normalizes scalar vector components, then passes them
* through a codec
*/
struct Quantizer {
virtual void encode_vector(const float *x, uint8_t *code) const = 0;
virtual void decode_vector(const uint8_t *code, float *x) const = 0;
virtual ~Quantizer() {}
};
template<class Codec>
struct QuantizerUniform: Quantizer {
const size_t d;
const float vmin, vdiff;
QuantizerUniform(size_t d, const std::vector<float> &trained):
d(d), vmin(trained[0]), vdiff(trained[1])
{
}
void encode_vector(const float* x, uint8_t* code) const override {
for (size_t i = 0; i < d; i++) {
float xi = (x[i] - vmin) / vdiff;
if (xi < 0)
xi = 0;
if (xi > 1.0)
xi = 1.0;
Codec::encode_component(xi, code, i);
}
}
void decode_vector(const uint8_t* code, float* x) const override {
for (size_t i = 0; i < d; i++) {
float xi = Codec::decode_component(code, i);
x[i] = vmin + xi * vdiff;
}
}
float reconstruct_component (const uint8_t * code, int i) const
{
float xi = Codec::decode_component (code, i);
return vmin + xi * vdiff;
}
};
#ifdef USE_AVX
template<class Codec>
struct QuantizerUniform8: QuantizerUniform<Codec> {
QuantizerUniform8 (size_t d, const std::vector<float> &trained):
QuantizerUniform<Codec> (d, trained) {}
__m256 reconstruct_8_components (const uint8_t * code, int i) const
{
__m256 xi = Codec::decode_8_components (code, i);
return _mm256_set1_ps(this->vmin) + xi * _mm256_set1_ps (this->vdiff);
}
};
#endif
template<class Codec>
struct QuantizerNonUniform: Quantizer {
const size_t d;
const float *vmin, *vdiff;
QuantizerNonUniform(size_t d, const std::vector<float> &trained):
d(d), vmin(trained.data()), vdiff(trained.data() + d) {}
void encode_vector(const float* x, uint8_t* code) const override {
for (size_t i = 0; i < d; i++) {
float xi = (x[i] - vmin[i]) / vdiff[i];
if (xi < 0)
xi = 0;
if (xi > 1.0)
xi = 1.0;
Codec::encode_component(xi, code, i);
}
}
void decode_vector(const uint8_t* code, float* x) const override {
for (size_t i = 0; i < d; i++) {
float xi = Codec::decode_component(code, i);
x[i] = vmin[i] + xi * vdiff[i];
}
}
float reconstruct_component (const uint8_t * code, int i) const
{
float xi = Codec::decode_component (code, i);
return vmin[i] + xi * vdiff[i];
}
};
#ifdef USE_AVX
template<class Codec>
struct QuantizerNonUniform8: QuantizerNonUniform<Codec> {
QuantizerNonUniform8 (size_t d, const std::vector<float> &trained):
QuantizerNonUniform<Codec> (d, trained) {}
__m256 reconstruct_8_components (const uint8_t * code, int i) const
{
__m256 xi = Codec::decode_8_components (code, i);
return _mm256_loadu_ps (this->vmin + i) + xi * _mm256_loadu_ps (this->vdiff + i);
}
};
#endif
struct QuantizerFP16: Quantizer {
const size_t d;
QuantizerFP16(size_t d, const std::vector<float> & /* unused */):
d(d) {}
void encode_vector(const float* x, uint8_t* code) const override {
for (size_t i = 0; i < d; i++) {
((uint16_t*)code)[i] = encode_fp16(x[i]);
}
}
void decode_vector(const uint8_t* code, float* x) const override {
for (size_t i = 0; i < d; i++) {
x[i] = decode_fp16(((uint16_t*)code)[i]);
}
}
float reconstruct_component (const uint8_t * code, int i) const
{
return decode_fp16(((uint16_t*)code)[i]);
}
};
#ifdef USE_AVX
struct QuantizerFP16_8: QuantizerFP16 {
QuantizerFP16_8 (size_t d, const std::vector<float> &trained):
QuantizerFP16 (d, trained) {}
__m256 reconstruct_8_components (const uint8_t * code, int i) const
{
__m128i codei = _mm_loadu_si128 ((const __m128i*)(code + 2 * i));
return _mm256_cvtph_ps (codei);
}
};
#endif
Quantizer *select_quantizer (
QuantizerType qtype,
size_t d, const std::vector<float> & trained)
{
#ifdef USE_AVX
if (d % 8 == 0) {
switch(qtype) {
case ScalarQuantizer::QT_8bit:
return new QuantizerNonUniform8<Codec8bit>(d, trained);
case ScalarQuantizer::QT_4bit:
return new QuantizerNonUniform8<Codec4bit>(d, trained);
case ScalarQuantizer::QT_8bit_uniform:
return new QuantizerUniform8<Codec8bit>(d, trained);
case ScalarQuantizer::QT_4bit_uniform:
return new QuantizerUniform8<Codec4bit>(d, trained);
case ScalarQuantizer::QT_fp16:
return new QuantizerFP16_8 (d, trained);
}
} else
#endif
{
switch(qtype) {
case ScalarQuantizer::QT_8bit:
return new QuantizerNonUniform<Codec8bit>(d, trained);
case ScalarQuantizer::QT_4bit:
return new QuantizerNonUniform<Codec4bit>(d, trained);
case ScalarQuantizer::QT_8bit_uniform:
return new QuantizerUniform<Codec8bit>(d, trained);
case ScalarQuantizer::QT_4bit_uniform:
return new QuantizerUniform<Codec4bit>(d, trained);
case ScalarQuantizer::QT_fp16:
return new QuantizerFP16 (d, trained);
}
}
FAISS_THROW_MSG ("unknown qtype");
return nullptr;
}
Quantizer *select_quantizer (const ScalarQuantizer &sq)
{
return select_quantizer (sq.qtype, sq.d, sq.trained);
}
/*******************************************************************
* Quantizer range training
*/
static float sqr (float x) {
return x * x;
}
void train_Uniform(RangeStat rs, float rs_arg,
idx_t n, int k, const float *x,
std::vector<float> & trained)
{
trained.resize (2);
float & vmin = trained[0];
float & vmax = trained[1];
if (rs == ScalarQuantizer::RS_minmax) {
vmin = HUGE_VAL; vmax = -HUGE_VAL;
for (size_t i = 0; i < n; i++) {
if (x[i] < vmin) vmin = x[i];
if (x[i] > vmax) vmax = x[i];
}
float vexp = (vmax - vmin) * rs_arg;
vmin -= vexp;
vmax += vexp;
} else if (rs == ScalarQuantizer::RS_meanstd) {
double sum = 0, sum2 = 0;
for (size_t i = 0; i < n; i++) {
sum += x[i];
sum2 += x[i] * x[i];
}
float mean = sum / n;
float var = sum2 / n - mean * mean;
float std = var <= 0 ? 1.0 : sqrt(var);
vmin = mean - std * rs_arg ;
vmax = mean + std * rs_arg ;
} else if (rs == ScalarQuantizer::RS_quantiles) {
std::vector<float> x_copy(n);
memcpy(x_copy.data(), x, n * sizeof(*x));
// TODO just do a qucikselect
std::sort(x_copy.begin(), x_copy.end());
int o = int(rs_arg * n);
if (o < 0) o = 0;
if (o > n - o) o = n / 2;
vmin = x_copy[o];
vmax = x_copy[n - 1 - o];
} else if (rs == ScalarQuantizer::RS_optim) {
float a, b;
float sx = 0;
{
vmin = HUGE_VAL, vmax = -HUGE_VAL;
for (size_t i = 0; i < n; i++) {
if (x[i] < vmin) vmin = x[i];
if (x[i] > vmax) vmax = x[i];
sx += x[i];
}
b = vmin;
a = (vmax - vmin) / (k - 1);
}
int verbose = false;
int niter = 2000;
float last_err = -1;
int iter_last_err = 0;
for (int it = 0; it < niter; it++) {
float sn = 0, sn2 = 0, sxn = 0, err1 = 0;
for (idx_t i = 0; i < n; i++) {
float xi = x[i];
float ni = floor ((xi - b) / a + 0.5);
if (ni < 0) ni = 0;
if (ni >= k) ni = k - 1;
err1 += sqr (xi - (ni * a + b));
sn += ni;
sn2 += ni * ni;
sxn += ni * xi;
}
if (err1 == last_err) {
iter_last_err ++;
if (iter_last_err == 16) break;
} else {
last_err = err1;
iter_last_err = 0;
}
float det = sqr (sn) - sn2 * n;
b = (sn * sxn - sn2 * sx) / det;
a = (sn * sx - n * sxn) / det;
if (verbose) {
printf ("it %d, err1=%g \r", it, err1);
fflush(stdout);
}
}
if (verbose) printf("\n");
vmin = b;
vmax = b + a * (k - 1);
} else {
FAISS_THROW_MSG ("Invalid qtype");
}
vmax -= vmin;
}
void train_NonUniform(RangeStat rs, float rs_arg,
idx_t n, int d, int k, const float *x,
std::vector<float> & trained)
{
trained.resize (2 * d);
float * vmin = trained.data();
float * vmax = trained.data() + d;
if (rs == ScalarQuantizer::RS_minmax) {
memcpy (vmin, x, sizeof(*x) * d);
memcpy (vmax, x, sizeof(*x) * d);
for (size_t i = 1; i < n; i++) {
const float *xi = x + i * d;
for (size_t j = 0; j < d; j++) {
if (xi[j] < vmin[j]) vmin[j] = xi[j];
if (xi[j] > vmax[j]) vmax[j] = xi[j];
}
}
float *vdiff = vmax;
for (size_t j = 0; j < d; j++) {
float vexp = (vmax[j] - vmin[j]) * rs_arg;
vmin[j] -= vexp;
vmax[j] += vexp;
vdiff [j] = vmax[j] - vmin[j];
}
} else {
// transpose
std::vector<float> xt(n * d);
for (size_t i = 1; i < n; i++) {
const float *xi = x + i * d;
for (size_t j = 0; j < d; j++) {
xt[j * n + i] = xi[j];
}
}
std::vector<float> trained_d(2);
#pragma omp parallel for
for (size_t j = 0; j < d; j++) {
train_Uniform(rs, rs_arg,
n, k, xt.data() + j * n,
trained_d);
vmin[j] = trained_d[0];
vmax[j] = trained_d[1];
}
}
}
/*******************************************************************
* Similarity: gets vector components and computes a similarity wrt. a
* query vector stored in the object. The data fields just encapsulate
* an accumulator.
*/
struct SimilarityL2 {
const float *y, *yi;
explicit SimilarityL2 (const float * y): y(y) {}
/******* scalar accumulator *******/
float accu;
void begin () {
accu = 0;
yi = y;
}
void add_component (float x) {
float tmp = *yi++ - x;
accu += tmp * tmp;
}
void add_component_2 (float x1, float x2) {
float tmp = x1 - x2;
accu += tmp * tmp;
}
float result () {
return accu;
}
#ifdef USE_AVX
__m256 accu8;
void begin_8 () {
accu8 = _mm256_setzero_ps();
yi = y;
}
void add_8_components (__m256 x) {
__m256 yiv = _mm256_loadu_ps (yi);
yi += 8;
__m256 tmp = yiv - x;
accu8 += tmp * tmp;
}
void add_8_components_2 (__m256 x, __m256 y) {
__m256 tmp = y - x;
accu8 += tmp * tmp;
}
float result_8 () {
__m256 sum = _mm256_hadd_ps(accu8, accu8);
__m256 sum2 = _mm256_hadd_ps(sum, sum);
// now add the 0th and 4th component
return
_mm_cvtss_f32 (_mm256_castps256_ps128(sum2)) +
_mm_cvtss_f32 (_mm256_extractf128_ps(sum2, 1));
}
#endif
};
struct SimilarityIP {
const float *y, *yi;
/******* scalar accumulator *******/
float accu;
explicit SimilarityIP (const float * y):
y (y) {}
void begin () {
accu = 0;
yi = y;
}
void add_component (float x) {
accu += *yi++ * x;
}
void add_component_2 (float x1, float x2) {
accu += x1 * x2;
}
float result () {
return accu;
}
#ifdef USE_AVX
__m256 accu8;
void begin_8 () {
accu8 = _mm256_setzero_ps();
yi = y;
}
void add_8_components (__m256 x) {
__m256 yiv = _mm256_loadu_ps (yi);
yi += 8;
accu8 += yiv * x;
}
void add_8_components_2 (__m256 x1, __m256 x2) {
accu8 += x1 * x2;
}
float result_8 () {
__m256 sum = _mm256_hadd_ps(accu8, accu8);
__m256 sum2 = _mm256_hadd_ps(sum, sum);
// now add the 0th and 4th component
return
_mm_cvtss_f32 (_mm256_castps256_ps128(sum2)) +
_mm_cvtss_f32 (_mm256_extractf128_ps(sum2, 1));
}
#endif
};
/*******************************************************************
* DistanceComputer: combines a similarity and a quantizer to do
* code-to-vector or code-to-code comparisons
*/
template<class Quantizer, class Similarity>
struct DCTemplate : ScalarQuantizer::DistanceComputer {
Quantizer quant;
DCTemplate(size_t d, const std::vector<float> &trained):
quant(d, trained)
{}
float compute_distance (const float *x,
const uint8_t *code) override
{
Similarity sim(x);
sim.begin();
for (size_t i = 0; i < quant.d; i ++) {
float xi = quant.reconstruct_component (code, i);
sim.add_component (xi);
}
return sim.result();
}
float compute_code_distance (const uint8_t *code1,
const uint8_t *code2) override
{
Similarity sim(nullptr);
sim.begin ();
for (size_t i = 0; i < quant.d; i ++) {
float x1 = quant.reconstruct_component (code1, i);
float x2 = quant.reconstruct_component (code2, i);
sim.add_component_2 (x1, x2);
}
return sim.result ();
}
};
#ifdef USE_AVX
template<class Quantizer, class Similarity>
struct DCTemplate_8 : ScalarQuantizer::DistanceComputer {
Quantizer quant;
DCTemplate_8(size_t d, const std::vector<float> &trained):
quant(d, trained)
{}
float compute_distance (const float *x,
const uint8_t *code) override
{
Similarity sim(x);
sim.begin_8();
for (size_t i = 0; i < quant.d; i += 8) {
__m256 xi = quant.reconstruct_8_components (code, i);
sim.add_8_components (xi);
}
return sim.result_8();
}
float compute_code_distance (const uint8_t *code1,
const uint8_t *code2) override
{
Similarity sim(nullptr);
sim.begin_8 ();
for (size_t i = 0; i < quant.d; i += 8) {
__m256 x1 = quant.reconstruct_8_components (code1, i);
__m256 x2 = quant.reconstruct_8_components (code2, i);
sim.add_8_components_2 (x1, x2);
}
return sim.result_8 ();
}
};
#endif
template<class Sim>
DistanceComputer *select_distance_computer (
QuantizerType qtype,
size_t d, const std::vector<float> & trained)
{
#ifdef USE_AVX
if (d % 8 == 0) {
switch(qtype) {
case ScalarQuantizer::QT_8bit:
return new DCTemplate_8<QuantizerNonUniform8
<Codec8bit>, Sim>(d, trained);
case ScalarQuantizer::QT_4bit:
return new DCTemplate_8<QuantizerNonUniform8
<Codec4bit>, Sim>(d, trained);
case ScalarQuantizer::QT_8bit_uniform:
return new DCTemplate_8<QuantizerUniform8
<Codec8bit>, Sim>(d, trained);
case ScalarQuantizer::QT_4bit_uniform:
return new DCTemplate_8<QuantizerUniform8
<Codec4bit>, Sim>(d, trained);
case ScalarQuantizer::QT_fp16:
return new DCTemplate_8<QuantizerFP16_8, Sim>(d, trained);
}
} else
#endif
{
switch(qtype) {
case ScalarQuantizer::QT_8bit:
return new DCTemplate<QuantizerNonUniform
<Codec8bit>, Sim>(d, trained);
case ScalarQuantizer::QT_4bit:
return new DCTemplate<QuantizerNonUniform
<Codec4bit>, Sim>(d, trained);
case ScalarQuantizer::QT_8bit_uniform:
return new DCTemplate<QuantizerUniform
<Codec8bit>, Sim>(d, trained);
case ScalarQuantizer::QT_4bit_uniform:
return new DCTemplate<QuantizerUniform
<Codec4bit>, Sim>(d, trained);
case ScalarQuantizer::QT_fp16:
return new DCTemplate<QuantizerFP16, Sim>(d, trained);
}
}
FAISS_THROW_MSG ("unknown qtype");
return nullptr;
}
} // anonymous namespace
/*******************************************************************
* ScalarQuantizer implementation
********************************************************************/
ScalarQuantizer::ScalarQuantizer
(size_t d, QuantizerType qtype):
qtype (qtype), rangestat(RS_minmax), rangestat_arg(0), d (d)
{
switch (qtype) {
case QT_8bit: case QT_8bit_uniform:
code_size = d;
break;
case QT_4bit: case QT_4bit_uniform:
code_size = (d + 1) / 2;
break;
case QT_fp16:
code_size = d * 2;
break;
}
}
ScalarQuantizer::ScalarQuantizer ():
qtype(QT_8bit),
rangestat(RS_minmax), rangestat_arg(0), d (0), code_size(0)
{}
void ScalarQuantizer::train (size_t n, const float *x)
{
int bit_per_dim =
qtype == QT_4bit_uniform ? 4 :
qtype == QT_4bit ? 4 :
qtype == QT_8bit_uniform ? 8 :
qtype == QT_8bit ? 8 : -1;
switch (qtype) {
case QT_4bit_uniform: case QT_8bit_uniform:
train_Uniform (rangestat, rangestat_arg,
n * d, 1 << bit_per_dim, x, trained);
break;
case QT_4bit: case QT_8bit:
train_NonUniform (rangestat, rangestat_arg,
n, d, 1 << bit_per_dim, x, trained);
break;
case QT_fp16:
// no training necessary
break;
}
}
void ScalarQuantizer::compute_codes (const float * x,
uint8_t * codes,
size_t n) const
{
Quantizer *squant = select_quantizer (*this);
ScopeDeleter1<Quantizer> del(squant);
#pragma omp parallel for
for (size_t i = 0; i < n; i++)
squant->encode_vector (x + i * d, codes + i * code_size);
}
void ScalarQuantizer::decode (const uint8_t *codes, float *x, size_t n) const
{
Quantizer *squant = select_quantizer (*this);
ScopeDeleter1<Quantizer> del(squant);
#pragma omp parallel for
for (size_t i = 0; i < n; i++)
squant->decode_vector (codes + i * code_size, x + i * d);
}
ScalarQuantizer::DistanceComputer *ScalarQuantizer::get_distance_computer (
MetricType metric)
const
{
if (metric == METRIC_L2) {
return select_distance_computer<SimilarityL2>(qtype, d, trained);
} else {
return select_distance_computer<SimilarityIP>(qtype, d, trained);
}
}
/*******************************************************************
* IndexScalarQuantizer implementation
********************************************************************/
IndexScalarQuantizer::IndexScalarQuantizer
(int d, ScalarQuantizer::QuantizerType qtype,
MetricType metric):
Index(d, metric),
sq (d, qtype)
{
is_trained = false;
code_size = sq.code_size;
}
IndexScalarQuantizer::IndexScalarQuantizer ():
IndexScalarQuantizer(0, ScalarQuantizer::QT_8bit)
{}
void IndexScalarQuantizer::train(idx_t n, const float* x)
{
sq.train(n, x);
is_trained = true;
}
void IndexScalarQuantizer::add(idx_t n, const float* x)
{
FAISS_THROW_IF_NOT (is_trained);
codes.resize ((n + ntotal) * code_size);
sq.compute_codes (x, &codes[ntotal * code_size], n);
ntotal += n;
}