-
Notifications
You must be signed in to change notification settings - Fork 2
/
smolscale-avx2.c
3112 lines (2638 loc) · 111 KB
/
smolscale-avx2.c
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
/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
/* Copyright © 2019-2023 Hans Petter Jansson. See COPYING for details. */
#include <assert.h> /* assert */
#include <stdlib.h> /* malloc, free, alloca */
#include <stddef.h> /* ptrdiff_t */
#include <string.h> /* memset */
#include <limits.h>
#include <immintrin.h>
#include "smolscale-private.h"
/* ---------------------- *
* Context initialization *
* ---------------------- */
/* Number of horizontal pixels to process in a single batch. The define exists for
* clarity and cannot be changed without significant changes to the code elsewhere. */
#define BILIN_HORIZ_BATCH_PIXELS 16
/* Batched precalc array layout:
*
* 16 offsets followed by 16 factors, repeating until epilogue. The epilogue
* has offsets and factors alternating one by one, and will always have fewer
* than 16 o/f pairs:
*
* ooooooooooooooooffffffffffffffffooooooooooooooooffffffffffffffffofofofofof...
*
* 16 offsets layout: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
* 16 factors layout: 0 2 4 6 8 10 12 14 1 3 5 7 9 11 13 15
*/
static uint32_t
array_offset_offset (uint32_t elem_i)
{
return (elem_i / (BILIN_HORIZ_BATCH_PIXELS)) * (BILIN_HORIZ_BATCH_PIXELS * 2)
+ (elem_i % BILIN_HORIZ_BATCH_PIXELS);
}
static uint32_t
array_offset_factor (uint32_t elem_i)
{
uint8_t o [BILIN_HORIZ_BATCH_PIXELS] = { 0, 8, 1, 9, 2, 10, 3, 11, 4, 12, 5, 13, 6, 14, 7, 15 };
return (elem_i / (BILIN_HORIZ_BATCH_PIXELS)) * (BILIN_HORIZ_BATCH_PIXELS * 2)
+ BILIN_HORIZ_BATCH_PIXELS + o [elem_i % BILIN_HORIZ_BATCH_PIXELS];
}
static void
precalc_bilinear_array (uint16_t *array,
uint32_t dim_in,
uint32_t dim_out,
unsigned int make_absolute_offsets,
unsigned int do_batches)
{
uint64_t ofs_stepF, fracF, frac_stepF;
uint16_t last_ofs = 0;
uint32_t i = 0;
if (dim_in > dim_out)
{
/* Minification */
frac_stepF = ofs_stepF = (dim_in * SMOL_BILIN_MULTIPLIER) / dim_out;
fracF = (frac_stepF - SMOL_BILIN_MULTIPLIER) / 2;
}
else
{
/* Magnification */
frac_stepF = ofs_stepF = ((dim_in - 1) * SMOL_BILIN_MULTIPLIER)
/ (dim_out > 1 ? (dim_out - 1) : 1);
fracF = 0;
}
if (do_batches)
{
while (dim_out >= BILIN_HORIZ_BATCH_PIXELS)
{
uint32_t j;
for (j = 0; j < BILIN_HORIZ_BATCH_PIXELS; j++)
{
uint16_t ofs = fracF / SMOL_BILIN_MULTIPLIER;
/* We sample ofs and its neighbor -- prevent out of bounds access
* for the latter by sampling the final pixel at 100%. */
if (ofs >= dim_in - 1)
{
array [array_offset_offset (i)] = make_absolute_offsets ? dim_in - 2 : (dim_in - 2) - last_ofs;
array [array_offset_factor (i)] = 0;
last_ofs = dim_in - 2;
}
else
{
array [array_offset_offset (i)] = make_absolute_offsets ? ofs : ofs - last_ofs;
array [array_offset_factor (i)] = SMOL_SMALL_MUL - ((fracF / (SMOL_BILIN_MULTIPLIER / SMOL_SMALL_MUL))
% SMOL_SMALL_MUL);
fracF += frac_stepF;
last_ofs = ofs;
}
i++;
dim_out--;
}
}
i = (i / BILIN_HORIZ_BATCH_PIXELS) * (BILIN_HORIZ_BATCH_PIXELS * 2);
}
/* Epilogue/unbatched */
while (dim_out)
{
uint16_t ofs = fracF / SMOL_BILIN_MULTIPLIER;
if (ofs >= dim_in - 1)
{
array [i++] = make_absolute_offsets ? dim_in - 2 : (dim_in - 2) - last_ofs;
array [i++] = 0;
last_ofs = dim_in - 2;
}
else
{
array [i++] = make_absolute_offsets ? ofs : ofs - last_ofs;
array [i++] = SMOL_SMALL_MUL - ((fracF / (SMOL_BILIN_MULTIPLIER / SMOL_SMALL_MUL))
% SMOL_SMALL_MUL);
fracF += frac_stepF;
last_ofs = ofs;
}
dim_out--;
}
}
static void
precalc_boxes_array (uint16_t *array,
uint32_t *span_mul,
uint32_t dim_in,
uint32_t dim_out,
unsigned int make_absolute_offsets)
{
uint64_t fracF, frac_stepF;
uint16_t *pu16 = array;
uint16_t ofs, next_ofs;
uint64_t f;
uint64_t stride;
uint64_t a, b;
frac_stepF = ((uint64_t) dim_in * SMOL_BIG_MUL) / (uint64_t) dim_out;
fracF = 0;
ofs = 0;
stride = frac_stepF / (uint64_t) SMOL_BIG_MUL;
f = (frac_stepF / SMOL_SMALL_MUL) % SMOL_SMALL_MUL;
a = (SMOL_BOXES_MULTIPLIER * 255);
b = ((stride * 255) + ((f * 255) / 256));
*span_mul = (a + (b / 2)) / b;
do
{
fracF += frac_stepF;
next_ofs = (uint64_t) fracF / ((uint64_t) SMOL_BIG_MUL);
/* Prevent out of bounds access */
if (ofs >= dim_in - 1)
{
ofs = dim_in - 1;
break;
}
if (next_ofs > dim_in - 1)
{
next_ofs = dim_in - 1;
if (next_ofs <= ofs)
break;
}
stride = next_ofs - ofs - 1;
f = (fracF / SMOL_SMALL_MUL) % SMOL_SMALL_MUL;
/* Fraction is the other way around, since left pixel of each span
* comes first, and it's on the right side of the fractional sample. */
*(pu16++) = make_absolute_offsets ? ofs : stride;
*(pu16++) = f;
ofs = next_ofs;
}
while (--dim_out);
/* Instead of going out of bounds, sample the final pair of pixels with a 100%
* bias towards the last pixel */
while (dim_out)
{
*(pu16++) = make_absolute_offsets ? ofs : 0;
*(pu16++) = 0;
dim_out--;
}
*(pu16++) = make_absolute_offsets ? ofs : 0;
*(pu16++) = 0;
}
static void
init_horizontal (SmolScaleCtx *scale_ctx)
{
if (scale_ctx->filter_h == SMOL_FILTER_ONE
|| scale_ctx->filter_h == SMOL_FILTER_COPY)
{
}
else if (scale_ctx->filter_h == SMOL_FILTER_BOX)
{
precalc_boxes_array (scale_ctx->precalc_x, &scale_ctx->span_mul_x,
scale_ctx->width_in, scale_ctx->width_out,
FALSE);
}
else if (scale_ctx->storage_type == SMOL_STORAGE_64BPP
&& scale_ctx->filter_h >= SMOL_FILTER_BILINEAR_0H
&& scale_ctx->filter_h <= SMOL_FILTER_BILINEAR_2H)
{
precalc_bilinear_array (scale_ctx->precalc_x,
scale_ctx->width_in,
scale_ctx->width_bilin_out,
TRUE, TRUE);
}
else /* SMOL_FILTER_BILINEAR_?H */
{
precalc_bilinear_array (scale_ctx->precalc_x,
scale_ctx->width_in,
scale_ctx->width_bilin_out,
FALSE, FALSE);
}
}
static void
init_vertical (SmolScaleCtx *scale_ctx)
{
if (scale_ctx->filter_v == SMOL_FILTER_ONE
|| scale_ctx->filter_v == SMOL_FILTER_COPY)
{
}
else if (scale_ctx->filter_v == SMOL_FILTER_BOX)
{
precalc_boxes_array (scale_ctx->precalc_y, &scale_ctx->span_mul_y,
scale_ctx->height_in, scale_ctx->height_out,
TRUE);
}
else /* SMOL_FILTER_BILINEAR_?H */
{
precalc_bilinear_array (scale_ctx->precalc_y,
scale_ctx->height_in,
scale_ctx->height_bilin_out,
TRUE, FALSE);
}
}
/* ----------------- *
* Premultiplication *
* ----------------- */
static SMOL_INLINE void
premul_u_to_p8_128bpp (uint64_t * SMOL_RESTRICT inout,
uint8_t alpha)
{
inout [0] = (((inout [0] + 0x0000000100000001) * ((uint16_t) alpha + 1) - 0x0000000100000001)
>> 8) & 0x000000ff000000ff;
inout [1] = (((inout [1] + 0x0000000100000001) * ((uint16_t) alpha + 1) - 0x0000000100000001)
>> 8) & 0x000000ff000000ff;
}
static SMOL_INLINE void
unpremul_p8_to_u_128bpp (const uint64_t *in,
uint64_t *out,
uint8_t alpha)
{
out [0] = ((in [0] * _smol_inv_div_p8_lut [alpha])
>> INVERTED_DIV_SHIFT_P8) & 0x000000ff000000ff;
out [1] = ((in [1] * _smol_inv_div_p8_lut [alpha])
>> INVERTED_DIV_SHIFT_P8) & 0x000000ff000000ff;
}
static SMOL_INLINE uint64_t
premul_u_to_p8_64bpp (const uint64_t in,
uint8_t alpha)
{
return (((in + 0x0001000100010001) * ((uint16_t) alpha + 1) - 0x0001000100010001)
>> 8) & 0x00ff00ff00ff00ff;
}
static SMOL_INLINE uint64_t
unpremul_p8_to_u_64bpp (const uint64_t in,
uint8_t alpha)
{
uint64_t in_128bpp [2];
uint64_t out_128bpp [2];
in_128bpp [0] = (in & 0x000000ff000000ff);
in_128bpp [1] = (in & 0x00ff000000ff0000) >> 16;
unpremul_p8_to_u_128bpp (in_128bpp, out_128bpp, alpha);
return out_128bpp [0] | (out_128bpp [1] << 16);
}
static SMOL_INLINE void
premul_u_to_p16_128bpp (uint64_t *inout,
uint8_t alpha)
{
inout [0] = inout [0] * alpha;
inout [1] = inout [1] * alpha;
}
static SMOL_INLINE void
unpremul_p16_to_u_128bpp (const uint64_t * SMOL_RESTRICT in,
uint64_t * SMOL_RESTRICT out,
uint8_t alpha)
{
out [0] = ((in [0] * _smol_inv_div_p16_lut [alpha])
>> INVERTED_DIV_SHIFT_P16) & 0x000000ff000000ffULL;
out [1] = ((in [1] * _smol_inv_div_p16_lut [alpha])
>> INVERTED_DIV_SHIFT_P16) & 0x000000ff000000ffULL;
}
/* --------- *
* Repacking *
* --------- */
/* PACK_SHUF_MM256_EPI8_32_TO_128()
*
* Generates a shuffling register for packing 8bpc pixel channels in the
* provided order. The order (1, 2, 3, 4) is neutral and corresponds to
*
* _mm256_set_epi8 (13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2,
* 13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2);
*/
#define SHUF_ORDER_32_TO_128 0x01000302U
#define SHUF_CH_32_TO_128(n) ((char) (SHUF_ORDER_32_TO_128 >> ((4 - (n)) * 8)))
#define SHUF_QUAD_CH_32_TO_128(q, n) (4 * (q) + SHUF_CH_32_TO_128 (n))
#define SHUF_QUAD_32_TO_128(q, a, b, c, d) \
SHUF_QUAD_CH_32_TO_128 ((q), (a)), \
SHUF_QUAD_CH_32_TO_128 ((q), (b)), \
SHUF_QUAD_CH_32_TO_128 ((q), (c)), \
SHUF_QUAD_CH_32_TO_128 ((q), (d))
#define PACK_SHUF_EPI8_LANE_32_TO_128(a, b, c, d) \
SHUF_QUAD_32_TO_128 (3, (a), (b), (c), (d)), \
SHUF_QUAD_32_TO_128 (2, (a), (b), (c), (d)), \
SHUF_QUAD_32_TO_128 (1, (a), (b), (c), (d)), \
SHUF_QUAD_32_TO_128 (0, (a), (b), (c), (d))
#define PACK_SHUF_MM256_EPI8_32_TO_128(a, b, c, d) _mm256_set_epi8 ( \
PACK_SHUF_EPI8_LANE_32_TO_128 ((a), (b), (c), (d)), \
PACK_SHUF_EPI8_LANE_32_TO_128 ((a), (b), (c), (d)))
/* PACK_SHUF_MM256_EPI8_32_TO_64()
*
* 64bpp version. Packs only once, so fewer contortions required. */
#define SHUF_CH_32_TO_64(n) ((char) (4 - (n)))
#define SHUF_QUAD_CH_32_TO_64(q, n) (4 * (q) + SHUF_CH_32_TO_64 (n))
#define SHUF_QUAD_32_TO_64(q, a, b, c, d) \
SHUF_QUAD_CH_32_TO_64 ((q), (a)), \
SHUF_QUAD_CH_32_TO_64 ((q), (b)), \
SHUF_QUAD_CH_32_TO_64 ((q), (c)), \
SHUF_QUAD_CH_32_TO_64 ((q), (d))
#define PACK_SHUF_EPI8_LANE_32_TO_64(a, b, c, d) \
SHUF_QUAD_32_TO_64 (3, (a), (b), (c), (d)), \
SHUF_QUAD_32_TO_64 (2, (a), (b), (c), (d)), \
SHUF_QUAD_32_TO_64 (1, (a), (b), (c), (d)), \
SHUF_QUAD_32_TO_64 (0, (a), (b), (c), (d))
#define PACK_SHUF_MM256_EPI8_32_TO_64(a, b, c, d) _mm256_set_epi8 ( \
PACK_SHUF_EPI8_LANE_32_TO_64 ((a), (b), (c), (d)), \
PACK_SHUF_EPI8_LANE_32_TO_64 ((a), (b), (c), (d)))
/* It's nice to be able to shift by a negative amount */
#define SHIFT_S(in, s) ((s >= 0) ? (in) << (s) : (in) >> -(s))
/* This is kind of bulky (~13 x86 insns), but it's about the same as using
* unions, and we don't have to worry about endianness. */
#define PACK_FROM_1234_64BPP(in, a, b, c, d) \
((SHIFT_S ((in), ((a) - 1) * 16 + 8 - 32) & 0xff000000) \
| (SHIFT_S ((in), ((b) - 1) * 16 + 8 - 40) & 0x00ff0000) \
| (SHIFT_S ((in), ((c) - 1) * 16 + 8 - 48) & 0x0000ff00) \
| (SHIFT_S ((in), ((d) - 1) * 16 + 8 - 56) & 0x000000ff))
#define PACK_FROM_1234_128BPP(in, a, b, c, d) \
((SHIFT_S ((in [((a) - 1) >> 1]), (((a) - 1) & 1) * 32 + 24 - 32) & 0xff000000) \
| (SHIFT_S ((in [((b) - 1) >> 1]), (((b) - 1) & 1) * 32 + 24 - 40) & 0x00ff0000) \
| (SHIFT_S ((in [((c) - 1) >> 1]), (((c) - 1) & 1) * 32 + 24 - 48) & 0x0000ff00) \
| (SHIFT_S ((in [((d) - 1) >> 1]), (((d) - 1) & 1) * 32 + 24 - 56) & 0x000000ff))
#define SWAP_2_AND_3(n) ((n) == 2 ? 3 : (n) == 3 ? 2 : n)
#define PACK_FROM_1324_64BPP(in, a, b, c, d) \
((SHIFT_S ((in), (SWAP_2_AND_3 (a) - 1) * 16 + 8 - 32) & 0xff000000) \
| (SHIFT_S ((in), (SWAP_2_AND_3 (b) - 1) * 16 + 8 - 40) & 0x00ff0000) \
| (SHIFT_S ((in), (SWAP_2_AND_3 (c) - 1) * 16 + 8 - 48) & 0x0000ff00) \
| (SHIFT_S ((in), (SWAP_2_AND_3 (d) - 1) * 16 + 8 - 56) & 0x000000ff))
/* ---------------------- *
* Repacking: 24/32 -> 64 *
* ---------------------- */
static void
unpack_8x_1234_p8_to_xxxx_p8_64bpp (const uint32_t * SMOL_RESTRICT *in,
uint64_t * SMOL_RESTRICT *out,
uint64_t *out_max,
const __m256i channel_shuf)
{
const __m256i zero = _mm256_setzero_si256 ();
const __m256i * SMOL_RESTRICT my_in = (const __m256i * SMOL_RESTRICT) *in;
__m256i * SMOL_RESTRICT my_out = (__m256i * SMOL_RESTRICT) *out;
__m256i m0, m1, m2;
SMOL_ASSUME_ALIGNED (my_out, __m256i * SMOL_RESTRICT);
while ((ptrdiff_t) (my_out + 2) <= (ptrdiff_t) out_max)
{
m0 = _mm256_loadu_si256 (my_in);
my_in++;
m0 = _mm256_shuffle_epi8 (m0, channel_shuf);
m0 = _mm256_permute4x64_epi64 (m0, SMOL_4X2BIT (3, 1, 2, 0));
m1 = _mm256_unpacklo_epi8 (m0, zero);
m2 = _mm256_unpackhi_epi8 (m0, zero);
_mm256_store_si256 (my_out, m1);
my_out++;
_mm256_store_si256 (my_out, m2);
my_out++;
}
*out = (uint64_t * SMOL_RESTRICT) my_out;
*in = (const uint32_t * SMOL_RESTRICT) my_in;
}
static SMOL_INLINE uint64_t
unpack_pixel_123_p8_to_132a_p8_64bpp (const uint8_t *p)
{
return ((uint64_t) p [0] << 48) | ((uint32_t) p [1] << 16)
| ((uint64_t) p [2] << 32) | 0xff;
}
SMOL_REPACK_ROW_DEF (123, 24, 8, PREMUL8, COMPRESSED,
1324, 64, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
*(row_out++) = unpack_pixel_123_p8_to_132a_p8_64bpp (row_in);
row_in += 3;
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE uint64_t
unpack_pixel_1234_p8_to_1324_p8_64bpp (uint32_t p)
{
return (((uint64_t) p & 0xff00ff00) << 24) | (p & 0x00ff00ff);
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, PREMUL8, COMPRESSED,
1324, 64, 64, PREMUL8, COMPRESSED) {
const __m256i channel_shuf = PACK_SHUF_MM256_EPI8_32_TO_64 (1, 3, 2, 4);
unpack_8x_1234_p8_to_xxxx_p8_64bpp (&row_in, &row_out, row_out_max,
channel_shuf);
while (row_out != row_out_max)
{
*(row_out++) = unpack_pixel_1234_p8_to_1324_p8_64bpp (*(row_in++));
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE uint64_t
unpack_pixel_1234_p8_to_3241_p8_64bpp (uint32_t p)
{
return (((uint64_t) p & 0x0000ff00) << 40)
| (((uint64_t) p & 0x00ff00ff) << 16) | (p >> 24);
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, PREMUL8, COMPRESSED,
3241, 64, 64, PREMUL8, COMPRESSED) {
const __m256i channel_shuf = PACK_SHUF_MM256_EPI8_32_TO_64 (3, 2, 4, 1);
unpack_8x_1234_p8_to_xxxx_p8_64bpp (&row_in, &row_out, row_out_max,
channel_shuf);
while (row_out != row_out_max)
{
*(row_out++) = unpack_pixel_1234_p8_to_3241_p8_64bpp (*(row_in++));
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE uint64_t
unpack_pixel_1234_p8_to_2431_p8_64bpp (uint32_t p)
{
uint64_t p64 = p;
return ((p64 & 0x00ff00ff) << 32) | ((p64 & 0x0000ff00) << 8)
| ((p64 & 0xff000000) >> 24);
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, PREMUL8, COMPRESSED,
2431, 64, 64, PREMUL8, COMPRESSED) {
const __m256i channel_shuf = PACK_SHUF_MM256_EPI8_32_TO_64 (2, 4, 3, 1);
unpack_8x_1234_p8_to_xxxx_p8_64bpp (&row_in, &row_out, row_out_max,
channel_shuf);
while (row_out != row_out_max)
{
*(row_out++) = unpack_pixel_1234_p8_to_2431_p8_64bpp (*(row_in++));
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE uint64_t
unpack_pixel_a234_u_to_324a_p8_64bpp (uint32_t p)
{
uint64_t p64 = (((uint64_t) p & 0x0000ff00) << 40) | (((uint64_t) p & 0x00ff00ff) << 16);
uint8_t alpha = p >> 24;
return premul_u_to_p8_64bpp (p64, alpha) | alpha;
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, UNASSOCIATED, COMPRESSED,
3241, 64, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
*(row_out++) = unpack_pixel_a234_u_to_324a_p8_64bpp (*(row_in++));
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE uint64_t
unpack_pixel_1234_u_to_2431_p8_64bpp (uint32_t p)
{
uint64_t p64 = (((uint64_t) p & 0x00ff00ff) << 32) | (((uint64_t) p & 0x0000ff00) << 8);
uint8_t alpha = p >> 24;
return premul_u_to_p8_64bpp (p64, alpha) | alpha;
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, UNASSOCIATED, COMPRESSED,
2431, 64, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
*(row_out++) = unpack_pixel_1234_u_to_2431_p8_64bpp (*(row_in++));
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE uint64_t
unpack_pixel_123a_u_to_132a_p8_64bpp (uint32_t p)
{
uint64_t p64 = (((uint64_t) p & 0xff00ff00) << 24) | (p & 0x00ff0000);
uint8_t alpha = p & 0xff;
return premul_u_to_p8_64bpp (p64, alpha) | alpha;
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, UNASSOCIATED, COMPRESSED,
1324, 64, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
*(row_out++) = unpack_pixel_123a_u_to_132a_p8_64bpp (*(row_in++));
}
} SMOL_REPACK_ROW_DEF_END
/* ----------------------- *
* Repacking: 24/32 -> 128 *
* ----------------------- */
static void
unpack_8x_xxxx_u_to_123a_p16_128bpp (const uint32_t * SMOL_RESTRICT *in,
uint64_t * SMOL_RESTRICT *out,
uint64_t *out_max,
const __m256i channel_shuf)
{
const __m256i zero = _mm256_setzero_si256 ();
const __m256i factor_shuf = _mm256_set_epi8 (
-1, 12, -1, -1, -1, 12, -1, 12, -1, 4, -1, -1, -1, 4, -1, 4,
-1, 12, -1, -1, -1, 12, -1, 12, -1, 4, -1, -1, -1, 4, -1, 4);
const __m256i alpha_mul = _mm256_set_epi16 (
0, 0x100, 0, 0, 0, 0x100, 0, 0,
0, 0x100, 0, 0, 0, 0x100, 0, 0);
const __m256i alpha_add = _mm256_set_epi16 (
0, 0x80, 0, 0, 0, 0x80, 0, 0,
0, 0x80, 0, 0, 0, 0x80, 0, 0);
const __m256i * SMOL_RESTRICT my_in = (const __m256i * SMOL_RESTRICT) *in;
__m256i * SMOL_RESTRICT my_out = (__m256i * SMOL_RESTRICT) *out;
__m256i m0, m1, m2, m3, m4, m5, m6;
__m256i fact1, fact2;
SMOL_ASSUME_ALIGNED (my_out, __m256i * SMOL_RESTRICT);
while ((ptrdiff_t) (my_out + 4) <= (ptrdiff_t) out_max)
{
m0 = _mm256_loadu_si256 (my_in);
my_in++;
m0 = _mm256_shuffle_epi8 (m0, channel_shuf);
m0 = _mm256_permute4x64_epi64 (m0, SMOL_4X2BIT (3, 1, 2, 0));
m1 = _mm256_unpacklo_epi8 (m0, zero);
m2 = _mm256_unpackhi_epi8 (m0, zero);
fact1 = _mm256_shuffle_epi8 (m1, factor_shuf);
fact2 = _mm256_shuffle_epi8 (m2, factor_shuf);
fact1 = _mm256_or_si256 (fact1, alpha_mul);
fact2 = _mm256_or_si256 (fact2, alpha_mul);
m1 = _mm256_mullo_epi16 (m1, fact1);
m2 = _mm256_mullo_epi16 (m2, fact2);
m1 = _mm256_add_epi16 (m1, alpha_add);
m2 = _mm256_add_epi16 (m2, alpha_add);
m1 = _mm256_permute4x64_epi64 (m1, SMOL_4X2BIT (3, 1, 2, 0));
m2 = _mm256_permute4x64_epi64 (m2, SMOL_4X2BIT (3, 1, 2, 0));
m3 = _mm256_unpacklo_epi16 (m1, zero);
m4 = _mm256_unpackhi_epi16 (m1, zero);
m5 = _mm256_unpacklo_epi16 (m2, zero);
m6 = _mm256_unpackhi_epi16 (m2, zero);
_mm256_store_si256 (my_out, m3);
my_out++;
_mm256_store_si256 (my_out, m4);
my_out++;
_mm256_store_si256 (my_out, m5);
my_out++;
_mm256_store_si256 (my_out, m6);
my_out++;
}
*out = (uint64_t * SMOL_RESTRICT) my_out;
*in = (const uint32_t * SMOL_RESTRICT) my_in;
}
static SMOL_INLINE void
unpack_pixel_123_p8_to_123a_p8_128bpp (const uint8_t *in,
uint64_t *out)
{
out [0] = ((uint64_t) in [0] << 32) | in [1];
out [1] = ((uint64_t) in [2] << 32) | 0xff;
}
SMOL_REPACK_ROW_DEF (123, 24, 8, PREMUL8, COMPRESSED,
1234, 128, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
unpack_pixel_123_p8_to_123a_p8_128bpp (row_in, row_out);
row_in += 3;
row_out += 2;
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE void
unpack_pixel_123a_p8_to_123a_p8_128bpp (uint32_t p,
uint64_t *out)
{
uint64_t p64 = p;
out [0] = ((p64 & 0xff000000) << 8) | ((p64 & 0x00ff0000) >> 16);
out [1] = ((p64 & 0x0000ff00) << 24) | (p64 & 0x000000ff);
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, PREMUL8, COMPRESSED,
1234, 128, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
unpack_pixel_123a_p8_to_123a_p8_128bpp (*(row_in++), row_out);
row_out += 2;
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE void
unpack_pixel_a234_p8_to_234a_p8_128bpp (uint32_t p,
uint64_t *out)
{
uint64_t p64 = p;
out [0] = ((p64 & 0x00ff0000) << 16) | ((p64 & 0x0000ff00) >> 8);
out [1] = ((p64 & 0x000000ff) << 32) | ((p64 & 0xff000000) >> 24);
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, PREMUL8, COMPRESSED,
2341, 128, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
unpack_pixel_a234_p8_to_234a_p8_128bpp (*(row_in++), row_out);
row_out += 2;
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE void
unpack_pixel_a234_u_to_234a_p8_128bpp (uint32_t p,
uint64_t *out)
{
uint64_t p64 = (((uint64_t) p & 0x00ff00ff) << 32) | (((uint64_t) p & 0x0000ff00) << 8);
uint8_t alpha = p >> 24;
p64 = premul_u_to_p8_64bpp (p64, alpha) | alpha;
out [0] = (p64 >> 16) & 0x000000ff000000ff;
out [1] = p64 & 0x000000ff000000ff;
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, UNASSOCIATED, COMPRESSED,
2341, 128, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
unpack_pixel_a234_u_to_234a_p8_128bpp (*(row_in++), row_out);
row_out += 2;
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE void
unpack_pixel_a234_u_to_234a_p16_128bpp (uint32_t p,
uint64_t *out)
{
uint64_t p64 = p;
uint64_t alpha = p >> 24;
out [0] = (((((p64 & 0x00ff0000) << 16) | ((p64 & 0x0000ff00) >> 8)) * alpha));
out [1] = (((((p64 & 0x000000ff) << 32) * alpha))) | (alpha << 8) | 0x80;
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, UNASSOCIATED, COMPRESSED,
2341, 128, 64, PREMUL16, COMPRESSED) {
const __m256i channel_shuf = PACK_SHUF_MM256_EPI8_32_TO_128 (2, 3, 4, 1);
unpack_8x_xxxx_u_to_123a_p16_128bpp (&row_in, &row_out, row_out_max,
channel_shuf);
while (row_out != row_out_max)
{
unpack_pixel_a234_u_to_234a_p16_128bpp (*(row_in++), row_out);
row_out += 2;
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE void
unpack_pixel_123a_u_to_123a_p8_128bpp (uint32_t p,
uint64_t *out)
{
uint64_t p64 = (((uint64_t) p & 0xff00ff00) << 24) | (p & 0x00ff0000);
uint8_t alpha = p & 0xff;
p64 = premul_u_to_p8_64bpp (p64, alpha) | ((uint64_t) alpha);
out [0] = (p64 >> 16) & 0x000000ff000000ff;
out [1] = p64 & 0x000000ff000000ff;
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, UNASSOCIATED, COMPRESSED,
1234, 128, 64, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
unpack_pixel_123a_u_to_123a_p8_128bpp (*(row_in++), row_out);
row_out += 2;
}
} SMOL_REPACK_ROW_DEF_END
static SMOL_INLINE void
unpack_pixel_123a_u_to_123a_p16_128bpp (uint32_t p,
uint64_t *out)
{
uint64_t p64 = p;
uint64_t alpha = p & 0xff;
out [0] = (((((p64 & 0xff000000) << 8) | ((p64 & 0x00ff0000) >> 16)) * alpha));
out [1] = (((((p64 & 0x0000ff00) << 24) * alpha))) | (alpha << 8) | 0x80;
}
SMOL_REPACK_ROW_DEF (1234, 32, 32, UNASSOCIATED, COMPRESSED,
1234, 128, 64, PREMUL16, COMPRESSED) {
const __m256i channel_shuf = PACK_SHUF_MM256_EPI8_32_TO_128 (1, 2, 3, 4);
unpack_8x_xxxx_u_to_123a_p16_128bpp (&row_in, &row_out, row_out_max,
channel_shuf);
while (row_out != row_out_max)
{
unpack_pixel_123a_u_to_123a_p16_128bpp (*(row_in++), row_out);
row_out += 2;
}
} SMOL_REPACK_ROW_DEF_END
/* ---------------------- *
* Repacking: 64 -> 24/32 *
* ---------------------- */
static void
pack_8x_1234_p8_to_xxxx_p8_64bpp (const uint64_t * SMOL_RESTRICT *in,
uint32_t * SMOL_RESTRICT *out,
uint32_t * out_max,
const __m256i channel_shuf)
{
const __m256i * SMOL_RESTRICT my_in = (const __m256i * SMOL_RESTRICT) *in;
__m256i * SMOL_RESTRICT my_out = (__m256i * SMOL_RESTRICT) *out;
__m256i m0, m1;
SMOL_ASSUME_ALIGNED (my_in, __m256i * SMOL_RESTRICT);
while ((ptrdiff_t) (my_out + 1) <= (ptrdiff_t) out_max)
{
/* Load inputs */
m0 = _mm256_stream_load_si256 (my_in);
my_in++;
m1 = _mm256_stream_load_si256 (my_in);
my_in++;
/* Pack and store */
m0 = _mm256_packus_epi16 (m0, m1);
m0 = _mm256_shuffle_epi8 (m0, channel_shuf);
m0 = _mm256_permute4x64_epi64 (m0, SMOL_4X2BIT (3, 1, 2, 0));
_mm256_storeu_si256 (my_out, m0);
my_out++;
}
*out = (uint32_t * SMOL_RESTRICT) my_out;
*in = (const uint64_t * SMOL_RESTRICT) my_in;
}
static SMOL_INLINE uint32_t
pack_pixel_1234_p8_to_1324_p8_64bpp (uint64_t in)
{
return in | (in >> 24);
}
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
132, 24, 8, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
uint32_t p = pack_pixel_1234_p8_to_1324_p8_64bpp (*(row_in++));
*(row_out++) = p >> 24;
*(row_out++) = p >> 16;
*(row_out++) = p >> 8;
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
132, 24, 8, UNASSOCIATED, COMPRESSED) {
while (row_out != row_out_max)
{
uint8_t alpha = *row_in;
uint64_t t = (unpremul_p8_to_u_64bpp (*row_in, alpha) & 0xffffffffffffff00ULL) | alpha;
uint32_t p = pack_pixel_1234_p8_to_1324_p8_64bpp (t);
*(row_out++) = p >> 24;
*(row_out++) = p >> 16;
*(row_out++) = p >> 8;
row_in++;
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
231, 24, 8, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
uint32_t p = pack_pixel_1234_p8_to_1324_p8_64bpp (*(row_in++));
*(row_out++) = p >> 8;
*(row_out++) = p >> 16;
*(row_out++) = p >> 24;
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
231, 24, 8, UNASSOCIATED, COMPRESSED) {
while (row_out != row_out_max)
{
uint8_t alpha = *row_in;
uint64_t t = (unpremul_p8_to_u_64bpp (*row_in, alpha) & 0xffffffffffffff00ULL) | alpha;
uint32_t p = pack_pixel_1234_p8_to_1324_p8_64bpp (t);
*(row_out++) = p >> 8;
*(row_out++) = p >> 16;
*(row_out++) = p >> 24;
row_in++;
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
324, 24, 8, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
uint32_t p = pack_pixel_1234_p8_to_1324_p8_64bpp (*(row_in++));
*(row_out++) = p >> 16;
*(row_out++) = p >> 8;
*(row_out++) = p;
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
324, 24, 8, UNASSOCIATED, COMPRESSED) {
while (row_out != row_out_max)
{
uint8_t alpha = *row_in >> 24;
uint64_t t = (unpremul_p8_to_u_64bpp (*row_in, alpha) & 0xffffffffffffff00ULL) | alpha;
uint32_t p = pack_pixel_1234_p8_to_1324_p8_64bpp (t);
*(row_out++) = p >> 16;
*(row_out++) = p >> 8;
*(row_out++) = p;
row_in++;
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
423, 24, 8, PREMUL8, COMPRESSED) {
while (row_out != row_out_max)
{
uint32_t p = pack_pixel_1234_p8_to_1324_p8_64bpp (*(row_in++));
*(row_out++) = p;
*(row_out++) = p >> 8;
*(row_out++) = p >> 16;
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
423, 24, 8, UNASSOCIATED, COMPRESSED) {
while (row_out != row_out_max)
{
uint8_t alpha = *row_in >> 24;
uint64_t t = (unpremul_p8_to_u_64bpp (*row_in, alpha) & 0xffffffffffffff00ULL) | alpha;
uint32_t p = pack_pixel_1234_p8_to_1324_p8_64bpp (t);
*(row_out++) = p;
*(row_out++) = p >> 8;
*(row_out++) = p >> 16;
row_in++;
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
1324, 32, 32, PREMUL8, COMPRESSED) {
const __m256i channel_shuf = PACK_SHUF_MM256_EPI8_32_TO_64 (1, 3, 2, 4);
pack_8x_1234_p8_to_xxxx_p8_64bpp (&row_in, &row_out, row_out_max,
channel_shuf);
while (row_out != row_out_max)
{
*(row_out++) = pack_pixel_1234_p8_to_1324_p8_64bpp (*(row_in++));
}
} SMOL_REPACK_ROW_DEF_END
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED,
1324, 32, 32, UNASSOCIATED, COMPRESSED) {
while (row_out != row_out_max)
{
uint8_t alpha = *row_in;
uint64_t t = (unpremul_p8_to_u_64bpp (*row_in, alpha) & 0xffffffffffffff00ULL) | alpha;
*(row_out++) = pack_pixel_1234_p8_to_1324_p8_64bpp (t);
row_in++;
}
} SMOL_REPACK_ROW_DEF_END
#define DEF_REPACK_FROM_1234_64BPP_TO_32BPP(a, b, c, d) \
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED, \
a##b##c##d, 32, 32, PREMUL8, COMPRESSED) { \
const __m256i channel_shuf = PACK_SHUF_MM256_EPI8_32_TO_64 ((a), (b), (c), (d)); \
pack_8x_1234_p8_to_xxxx_p8_64bpp (&row_in, &row_out, row_out_max, \
channel_shuf); \
while (row_out != row_out_max) \
{ \
*(row_out++) = PACK_FROM_1234_64BPP (*row_in, a, b, c, d); \
row_in++; \
} \
} SMOL_REPACK_ROW_DEF_END \
SMOL_REPACK_ROW_DEF (1234, 64, 64, PREMUL8, COMPRESSED, \
a##b##c##d, 32, 32, UNASSOCIATED, COMPRESSED) { \
while (row_out != row_out_max) \
{ \
uint8_t alpha = *row_in; \
uint64_t t = (unpremul_p8_to_u_64bpp (*row_in, alpha) & 0xffffffffffffff00ULL) | alpha; \
*(row_out++) = PACK_FROM_1234_64BPP (t, a, b, c, d); \
row_in++; \
} \
} SMOL_REPACK_ROW_DEF_END
DEF_REPACK_FROM_1234_64BPP_TO_32BPP (1, 4, 2, 3)
DEF_REPACK_FROM_1234_64BPP_TO_32BPP (2, 3, 1, 4)
DEF_REPACK_FROM_1234_64BPP_TO_32BPP (4, 1, 3, 2)
DEF_REPACK_FROM_1234_64BPP_TO_32BPP (4, 2, 3, 1)
/* ----------------------- *
* Repacking: 128 -> 24/32 *
* ----------------------- */
static void
pack_8x_123a_p16_to_xxxx_u_128bpp (const uint64_t * SMOL_RESTRICT *in,
uint32_t * SMOL_RESTRICT *out,
uint32_t * out_max,
const __m256i channel_shuf)
{
#define ALPHA_MUL (1 << (INVERTED_DIV_SHIFT_P16 - 8))
#define ALPHA_MASK SMOL_8X1BIT (0, 1, 0, 0, 0, 1, 0, 0)
const __m256i ones = _mm256_set_epi32 (
ALPHA_MUL, ALPHA_MUL, ALPHA_MUL, ALPHA_MUL,
ALPHA_MUL, ALPHA_MUL, ALPHA_MUL, ALPHA_MUL);
const __m256i alpha_clean_mask = _mm256_set_epi32 (
0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff,
0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff);
const __m256i * SMOL_RESTRICT my_in = (const __m256i * SMOL_RESTRICT) *in;
__m256i * SMOL_RESTRICT my_out = (__m256i * SMOL_RESTRICT) *out;
__m256i m0, m1, m2, m3, m4, m5, m6, m7, m8;
SMOL_ASSUME_ALIGNED (my_in, __m256i * SMOL_RESTRICT);
while ((ptrdiff_t) (my_out + 1) <= (ptrdiff_t) out_max)
{
/* Load inputs */
m0 = _mm256_stream_load_si256 (my_in);
my_in++;