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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
|
/*
* Copyright (c) 2018, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <smmintrin.h> // SSE4.1
#include <immintrin.h> // AVX2
#include <assert.h>
#include "aom/aom_integer.h"
#include "aom_ports/mem.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/x86/synonyms.h"
#include "aom_dsp/x86/synonyms_avx2.h"
#include "aom_dsp/x86/blend_sse4.h"
#include "aom_dsp/x86/blend_mask_sse4.h"
#include "config/aom_dsp_rtcd.h"
static INLINE void blend_a64_d16_mask_w16_avx2(
uint8_t *dst, const CONV_BUF_TYPE *src0, const CONV_BUF_TYPE *src1,
const __m256i *m0, const __m256i *v_round_offset, const __m256i *v_maxval,
int shift) {
const __m256i max_minus_m0 = _mm256_sub_epi16(*v_maxval, *m0);
const __m256i s0_0 = yy_loadu_256(src0);
const __m256i s1_0 = yy_loadu_256(src1);
__m256i res0_lo = _mm256_madd_epi16(_mm256_unpacklo_epi16(s0_0, s1_0),
_mm256_unpacklo_epi16(*m0, max_minus_m0));
__m256i res0_hi = _mm256_madd_epi16(_mm256_unpackhi_epi16(s0_0, s1_0),
_mm256_unpackhi_epi16(*m0, max_minus_m0));
res0_lo =
_mm256_srai_epi32(_mm256_sub_epi32(res0_lo, *v_round_offset), shift);
res0_hi =
_mm256_srai_epi32(_mm256_sub_epi32(res0_hi, *v_round_offset), shift);
const __m256i res0 = _mm256_packs_epi32(res0_lo, res0_hi);
__m256i res = _mm256_packus_epi16(res0, res0);
res = _mm256_permute4x64_epi64(res, 0xd8);
_mm_storeu_si128((__m128i *)(dst), _mm256_castsi256_si128(res));
}
static INLINE void blend_a64_d16_mask_w32_avx2(
uint8_t *dst, const CONV_BUF_TYPE *src0, const CONV_BUF_TYPE *src1,
const __m256i *m0, const __m256i *m1, const __m256i *v_round_offset,
const __m256i *v_maxval, int shift) {
const __m256i max_minus_m0 = _mm256_sub_epi16(*v_maxval, *m0);
const __m256i max_minus_m1 = _mm256_sub_epi16(*v_maxval, *m1);
const __m256i s0_0 = yy_loadu_256(src0);
const __m256i s0_1 = yy_loadu_256(src0 + 16);
const __m256i s1_0 = yy_loadu_256(src1);
const __m256i s1_1 = yy_loadu_256(src1 + 16);
__m256i res0_lo = _mm256_madd_epi16(_mm256_unpacklo_epi16(s0_0, s1_0),
_mm256_unpacklo_epi16(*m0, max_minus_m0));
__m256i res0_hi = _mm256_madd_epi16(_mm256_unpackhi_epi16(s0_0, s1_0),
_mm256_unpackhi_epi16(*m0, max_minus_m0));
__m256i res1_lo = _mm256_madd_epi16(_mm256_unpacklo_epi16(s0_1, s1_1),
_mm256_unpacklo_epi16(*m1, max_minus_m1));
__m256i res1_hi = _mm256_madd_epi16(_mm256_unpackhi_epi16(s0_1, s1_1),
_mm256_unpackhi_epi16(*m1, max_minus_m1));
res0_lo =
_mm256_srai_epi32(_mm256_sub_epi32(res0_lo, *v_round_offset), shift);
res0_hi =
_mm256_srai_epi32(_mm256_sub_epi32(res0_hi, *v_round_offset), shift);
res1_lo =
_mm256_srai_epi32(_mm256_sub_epi32(res1_lo, *v_round_offset), shift);
res1_hi =
_mm256_srai_epi32(_mm256_sub_epi32(res1_hi, *v_round_offset), shift);
const __m256i res0 = _mm256_packs_epi32(res0_lo, res0_hi);
const __m256i res1 = _mm256_packs_epi32(res1_lo, res1_hi);
__m256i res = _mm256_packus_epi16(res0, res1);
res = _mm256_permute4x64_epi64(res, 0xd8);
_mm256_storeu_si256((__m256i *)(dst), res);
}
static INLINE void lowbd_blend_a64_d16_mask_subw0_subh0_w16_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h,
const __m256i *round_offset, int shift) {
const __m256i v_maxval = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
for (int i = 0; i < h; ++i) {
const __m128i m = xx_loadu_128(mask);
const __m256i m0 = _mm256_cvtepu8_epi16(m);
blend_a64_d16_mask_w16_avx2(dst, src0, src1, &m0, round_offset, &v_maxval,
shift);
mask += mask_stride;
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
}
}
static INLINE void lowbd_blend_a64_d16_mask_subw0_subh0_w32_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h, int w,
const __m256i *round_offset, int shift) {
const __m256i v_maxval = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; j += 32) {
const __m256i m = yy_loadu_256(mask + j);
const __m256i m0 = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(m));
const __m256i m1 = _mm256_cvtepu8_epi16(_mm256_extracti128_si256(m, 1));
blend_a64_d16_mask_w32_avx2(dst + j, src0 + j, src1 + j, &m0, &m1,
round_offset, &v_maxval, shift);
}
mask += mask_stride;
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
}
}
static INLINE void lowbd_blend_a64_d16_mask_subw1_subh1_w16_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h,
const __m256i *round_offset, int shift) {
const __m256i v_maxval = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
const __m256i one_b = _mm256_set1_epi8(1);
const __m256i two_w = _mm256_set1_epi16(2);
for (int i = 0; i < h; ++i) {
const __m256i m_i00 = yy_loadu_256(mask);
const __m256i m_i10 = yy_loadu_256(mask + mask_stride);
const __m256i m0_ac = _mm256_adds_epu8(m_i00, m_i10);
const __m256i m0_acbd = _mm256_maddubs_epi16(m0_ac, one_b);
const __m256i m0 = _mm256_srli_epi16(_mm256_add_epi16(m0_acbd, two_w), 2);
blend_a64_d16_mask_w16_avx2(dst, src0, src1, &m0, round_offset, &v_maxval,
shift);
mask += mask_stride << 1;
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
}
}
static INLINE void lowbd_blend_a64_d16_mask_subw1_subh1_w32_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h, int w,
const __m256i *round_offset, int shift) {
const __m256i v_maxval = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
const __m256i one_b = _mm256_set1_epi8(1);
const __m256i two_w = _mm256_set1_epi16(2);
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; j += 32) {
const __m256i m_i00 = yy_loadu_256(mask + 2 * j);
const __m256i m_i01 = yy_loadu_256(mask + 2 * j + 32);
const __m256i m_i10 = yy_loadu_256(mask + mask_stride + 2 * j);
const __m256i m_i11 = yy_loadu_256(mask + mask_stride + 2 * j + 32);
const __m256i m0_ac = _mm256_adds_epu8(m_i00, m_i10);
const __m256i m1_ac = _mm256_adds_epu8(m_i01, m_i11);
const __m256i m0_acbd = _mm256_maddubs_epi16(m0_ac, one_b);
const __m256i m1_acbd = _mm256_maddubs_epi16(m1_ac, one_b);
const __m256i m0 = _mm256_srli_epi16(_mm256_add_epi16(m0_acbd, two_w), 2);
const __m256i m1 = _mm256_srli_epi16(_mm256_add_epi16(m1_acbd, two_w), 2);
blend_a64_d16_mask_w32_avx2(dst + j, src0 + j, src1 + j, &m0, &m1,
round_offset, &v_maxval, shift);
}
mask += mask_stride << 1;
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
}
}
static INLINE void lowbd_blend_a64_d16_mask_subw1_subh0_w16_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h, int w,
const __m256i *round_offset, int shift) {
const __m256i v_maxval = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
const __m256i one_b = _mm256_set1_epi8(1);
const __m256i zeros = _mm256_setzero_si256();
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; j += 16) {
const __m256i m_i00 = yy_loadu_256(mask + 2 * j);
const __m256i m0_ac = _mm256_maddubs_epi16(m_i00, one_b);
const __m256i m0 = _mm256_avg_epu16(m0_ac, zeros);
blend_a64_d16_mask_w16_avx2(dst + j, src0 + j, src1 + j, &m0,
round_offset, &v_maxval, shift);
}
mask += mask_stride;
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
}
}
static INLINE void lowbd_blend_a64_d16_mask_subw1_subh0_w32_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h, int w,
const __m256i *round_offset, int shift) {
const __m256i v_maxval = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
const __m256i one_b = _mm256_set1_epi8(1);
const __m256i zeros = _mm256_setzero_si256();
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; j += 32) {
const __m256i m_i00 = yy_loadu_256(mask + 2 * j);
const __m256i m_i01 = yy_loadu_256(mask + 2 * j + 32);
const __m256i m0_ac = _mm256_maddubs_epi16(m_i00, one_b);
const __m256i m1_ac = _mm256_maddubs_epi16(m_i01, one_b);
const __m256i m0 = _mm256_avg_epu16(m0_ac, zeros);
const __m256i m1 = _mm256_avg_epu16(m1_ac, zeros);
blend_a64_d16_mask_w32_avx2(dst + j, src0 + j, src1 + j, &m0, &m1,
round_offset, &v_maxval, shift);
}
mask += mask_stride;
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
}
}
static INLINE void lowbd_blend_a64_d16_mask_subw0_subh1_w16_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h, int w,
const __m256i *round_offset, int shift) {
const __m256i v_maxval = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
const __m128i zeros = _mm_setzero_si128();
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; j += 16) {
const __m128i m_i00 = xx_loadu_128(mask + j);
const __m128i m_i10 = xx_loadu_128(mask + mask_stride + j);
const __m128i m_ac = _mm_avg_epu8(_mm_adds_epu8(m_i00, m_i10), zeros);
const __m256i m0 = _mm256_cvtepu8_epi16(m_ac);
blend_a64_d16_mask_w16_avx2(dst + j, src0 + j, src1 + j, &m0,
round_offset, &v_maxval, shift);
}
mask += mask_stride << 1;
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
}
}
static INLINE void lowbd_blend_a64_d16_mask_subw0_subh1_w32_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h, int w,
const __m256i *round_offset, int shift) {
const __m256i v_maxval = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
const __m256i zeros = _mm256_setzero_si256();
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; j += 32) {
const __m256i m_i00 = yy_loadu_256(mask + j);
const __m256i m_i10 = yy_loadu_256(mask + mask_stride + j);
const __m256i m_ac =
_mm256_avg_epu8(_mm256_adds_epu8(m_i00, m_i10), zeros);
const __m256i m0 = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(m_ac));
const __m256i m1 =
_mm256_cvtepu8_epi16(_mm256_extracti128_si256(m_ac, 1));
blend_a64_d16_mask_w32_avx2(dst + j, src0 + j, src1 + j, &m0, &m1,
round_offset, &v_maxval, shift);
}
mask += mask_stride << 1;
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
}
}
void aom_lowbd_blend_a64_d16_mask_avx2(
uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h, int subw, int subh,
ConvolveParams *conv_params) {
const int bd = 8;
const int round_bits =
2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
const int round_offset =
((1 << (round_bits + bd)) + (1 << (round_bits + bd - 1)) -
(1 << (round_bits - 1)))
<< AOM_BLEND_A64_ROUND_BITS;
const int shift = round_bits + AOM_BLEND_A64_ROUND_BITS;
assert(IMPLIES((void *)src0 == dst, src0_stride == dst_stride));
assert(IMPLIES((void *)src1 == dst, src1_stride == dst_stride));
assert(h >= 4);
assert(w >= 4);
assert(IS_POWER_OF_TWO(h));
assert(IS_POWER_OF_TWO(w));
const __m128i v_round_offset = _mm_set1_epi32(round_offset);
const __m256i y_round_offset = _mm256_set1_epi32(round_offset);
if (subw == 0 && subh == 0) {
switch (w) {
case 4:
aom_lowbd_blend_a64_d16_mask_subw0_subh0_w4_sse4_1(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift);
break;
case 8:
aom_lowbd_blend_a64_d16_mask_subw0_subh0_w8_sse4_1(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift);
break;
case 16:
lowbd_blend_a64_d16_mask_subw0_subh0_w16_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &y_round_offset, shift);
break;
default:
lowbd_blend_a64_d16_mask_subw0_subh0_w32_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, w, &y_round_offset, shift);
break;
}
} else if (subw == 1 && subh == 1) {
switch (w) {
case 4:
aom_lowbd_blend_a64_d16_mask_subw1_subh1_w4_sse4_1(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift);
break;
case 8:
aom_lowbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift);
break;
case 16:
lowbd_blend_a64_d16_mask_subw1_subh1_w16_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &y_round_offset, shift);
break;
default:
lowbd_blend_a64_d16_mask_subw1_subh1_w32_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, w, &y_round_offset, shift);
break;
}
} else if (subw == 1 && subh == 0) {
switch (w) {
case 4:
aom_lowbd_blend_a64_d16_mask_subw1_subh0_w4_sse4_1(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift);
break;
case 8:
aom_lowbd_blend_a64_d16_mask_subw1_subh0_w8_sse4_1(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift);
break;
case 16:
lowbd_blend_a64_d16_mask_subw1_subh0_w16_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, w, &y_round_offset, shift);
break;
default:
lowbd_blend_a64_d16_mask_subw1_subh0_w32_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, w, &y_round_offset, shift);
break;
}
} else {
switch (w) {
case 4:
aom_lowbd_blend_a64_d16_mask_subw0_subh1_w4_sse4_1(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift);
break;
case 8:
aom_lowbd_blend_a64_d16_mask_subw0_subh1_w8_sse4_1(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift);
break;
case 16:
lowbd_blend_a64_d16_mask_subw0_subh1_w16_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, w, &y_round_offset, shift);
break;
default:
lowbd_blend_a64_d16_mask_subw0_subh1_w32_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, w, &y_round_offset, shift);
break;
}
}
}
static INLINE __m256i blend_16_u8_avx2(const uint8_t *src0, const uint8_t *src1,
const __m256i *v_m0_b,
const __m256i *v_m1_b,
const int32_t bits) {
const __m256i v_s0_b = _mm256_castsi128_si256(xx_loadu_128(src0));
const __m256i v_s1_b = _mm256_castsi128_si256(xx_loadu_128(src1));
const __m256i v_s0_s_b = _mm256_permute4x64_epi64(v_s0_b, 0xd8);
const __m256i v_s1_s_b = _mm256_permute4x64_epi64(v_s1_b, 0xd8);
const __m256i v_p0_w =
_mm256_maddubs_epi16(_mm256_unpacklo_epi8(v_s0_s_b, v_s1_s_b),
_mm256_unpacklo_epi8(*v_m0_b, *v_m1_b));
const __m256i v_res0_w = yy_roundn_epu16(v_p0_w, bits);
const __m256i v_res_b = _mm256_packus_epi16(v_res0_w, v_res0_w);
const __m256i v_res = _mm256_permute4x64_epi64(v_res_b, 0xd8);
return v_res;
}
static INLINE __m256i blend_32_u8_avx2(const uint8_t *src0, const uint8_t *src1,
const __m256i *v_m0_b,
const __m256i *v_m1_b,
const int32_t bits) {
const __m256i v_s0_b = yy_loadu_256(src0);
const __m256i v_s1_b = yy_loadu_256(src1);
const __m256i v_p0_w =
_mm256_maddubs_epi16(_mm256_unpacklo_epi8(v_s0_b, v_s1_b),
_mm256_unpacklo_epi8(*v_m0_b, *v_m1_b));
const __m256i v_p1_w =
_mm256_maddubs_epi16(_mm256_unpackhi_epi8(v_s0_b, v_s1_b),
_mm256_unpackhi_epi8(*v_m0_b, *v_m1_b));
const __m256i v_res0_w = yy_roundn_epu16(v_p0_w, bits);
const __m256i v_res1_w = yy_roundn_epu16(v_p1_w, bits);
const __m256i v_res = _mm256_packus_epi16(v_res0_w, v_res1_w);
return v_res;
}
static INLINE void blend_a64_mask_sx_sy_w16_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h) {
const __m256i v_zmask_b = _mm256_set1_epi16(0xFF);
const __m256i v_maxval_b = _mm256_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
do {
const __m256i v_ral_b = yy_loadu_256(mask);
const __m256i v_rbl_b = yy_loadu_256(mask + mask_stride);
const __m256i v_rvsl_b = _mm256_add_epi8(v_ral_b, v_rbl_b);
const __m256i v_rvsal_w = _mm256_and_si256(v_rvsl_b, v_zmask_b);
const __m256i v_rvsbl_w =
_mm256_and_si256(_mm256_srli_si256(v_rvsl_b, 1), v_zmask_b);
const __m256i v_rsl_w = _mm256_add_epi16(v_rvsal_w, v_rvsbl_w);
const __m256i v_m0_w = yy_roundn_epu16(v_rsl_w, 2);
const __m256i v_m0_b = _mm256_packus_epi16(v_m0_w, v_m0_w);
const __m256i v_m1_b = _mm256_sub_epi8(v_maxval_b, v_m0_b);
const __m256i y_res_b = blend_16_u8_avx2(src0, src1, &v_m0_b, &v_m1_b,
AOM_BLEND_A64_ROUND_BITS);
xx_storeu_128(dst, _mm256_castsi256_si128(y_res_b));
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += 2 * mask_stride;
} while (--h);
}
static INLINE void blend_a64_mask_sx_sy_w32n_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h) {
const __m256i v_maxval_b = _mm256_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
const __m256i v_zmask_b = _mm256_set1_epi16(0xFF);
do {
int c;
for (c = 0; c < w; c += 32) {
const __m256i v_ral_b = yy_loadu_256(mask + 2 * c);
const __m256i v_rah_b = yy_loadu_256(mask + 2 * c + 32);
const __m256i v_rbl_b = yy_loadu_256(mask + mask_stride + 2 * c);
const __m256i v_rbh_b = yy_loadu_256(mask + mask_stride + 2 * c + 32);
const __m256i v_rvsl_b = _mm256_add_epi8(v_ral_b, v_rbl_b);
const __m256i v_rvsh_b = _mm256_add_epi8(v_rah_b, v_rbh_b);
const __m256i v_rvsal_w = _mm256_and_si256(v_rvsl_b, v_zmask_b);
const __m256i v_rvsah_w = _mm256_and_si256(v_rvsh_b, v_zmask_b);
const __m256i v_rvsbl_w =
_mm256_and_si256(_mm256_srli_si256(v_rvsl_b, 1), v_zmask_b);
const __m256i v_rvsbh_w =
_mm256_and_si256(_mm256_srli_si256(v_rvsh_b, 1), v_zmask_b);
const __m256i v_rsl_w = _mm256_add_epi16(v_rvsal_w, v_rvsbl_w);
const __m256i v_rsh_w = _mm256_add_epi16(v_rvsah_w, v_rvsbh_w);
const __m256i v_m0l_w = yy_roundn_epu16(v_rsl_w, 2);
const __m256i v_m0h_w = yy_roundn_epu16(v_rsh_w, 2);
const __m256i v_m0_b =
_mm256_permute4x64_epi64(_mm256_packus_epi16(v_m0l_w, v_m0h_w), 0xd8);
const __m256i v_m1_b = _mm256_sub_epi8(v_maxval_b, v_m0_b);
const __m256i v_res_b = blend_32_u8_avx2(
src0 + c, src1 + c, &v_m0_b, &v_m1_b, AOM_BLEND_A64_ROUND_BITS);
yy_storeu_256(dst + c, v_res_b);
}
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += 2 * mask_stride;
} while (--h);
}
static INLINE void blend_a64_mask_sx_sy_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h) {
const __m128i v_shuffle_b = xx_loadu_128(g_blend_a64_mask_shuffle);
const __m128i v_maxval_b = _mm_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
const __m128i _r = _mm_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_BITS));
switch (w) {
case 4:
do {
const __m128i v_ra_b = xx_loadl_64(mask);
const __m128i v_rb_b = xx_loadl_64(mask + mask_stride);
const __m128i v_rvs_b = _mm_add_epi8(v_ra_b, v_rb_b);
const __m128i v_r_s_b = _mm_shuffle_epi8(v_rvs_b, v_shuffle_b);
const __m128i v_r0_s_w = _mm_cvtepu8_epi16(v_r_s_b);
const __m128i v_r1_s_w = _mm_cvtepu8_epi16(_mm_srli_si128(v_r_s_b, 8));
const __m128i v_rs_w = _mm_add_epi16(v_r0_s_w, v_r1_s_w);
const __m128i v_m0_w = xx_roundn_epu16(v_rs_w, 2);
const __m128i v_m0_b = _mm_packus_epi16(v_m0_w, v_m0_w);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_4_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storel_32(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += 2 * mask_stride;
} while (--h);
break;
case 8:
do {
const __m128i v_ra_b = xx_loadu_128(mask);
const __m128i v_rb_b = xx_loadu_128(mask + mask_stride);
const __m128i v_rvs_b = _mm_add_epi8(v_ra_b, v_rb_b);
const __m128i v_r_s_b = _mm_shuffle_epi8(v_rvs_b, v_shuffle_b);
const __m128i v_r0_s_w = _mm_cvtepu8_epi16(v_r_s_b);
const __m128i v_r1_s_w = _mm_cvtepu8_epi16(_mm_srli_si128(v_r_s_b, 8));
const __m128i v_rs_w = _mm_add_epi16(v_r0_s_w, v_r1_s_w);
const __m128i v_m0_w = xx_roundn_epu16(v_rs_w, 2);
const __m128i v_m0_b = _mm_packus_epi16(v_m0_w, v_m0_w);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_8_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storel_64(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += 2 * mask_stride;
} while (--h);
break;
case 16:
blend_a64_mask_sx_sy_w16_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, h);
break;
default:
blend_a64_mask_sx_sy_w32n_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, w, h);
break;
}
}
static INLINE void blend_a64_mask_sx_w16_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h) {
const __m256i v_maxval_b = _mm256_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
const __m256i v_zmask_b = _mm256_set1_epi16(0xff);
do {
const __m256i v_rl_b = yy_loadu_256(mask);
const __m256i v_al_b =
_mm256_avg_epu8(v_rl_b, _mm256_srli_si256(v_rl_b, 1));
const __m256i v_m0_w = _mm256_and_si256(v_al_b, v_zmask_b);
const __m256i v_m0_b = _mm256_packus_epi16(v_m0_w, _mm256_setzero_si256());
const __m256i v_m1_b = _mm256_sub_epi8(v_maxval_b, v_m0_b);
const __m256i v_res_b = blend_16_u8_avx2(src0, src1, &v_m0_b, &v_m1_b,
AOM_BLEND_A64_ROUND_BITS);
xx_storeu_128(dst, _mm256_castsi256_si128(v_res_b));
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += mask_stride;
} while (--h);
}
static INLINE void blend_a64_mask_sx_w32n_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h) {
const __m256i v_shuffle_b = yy_loadu_256(g_blend_a64_mask_shuffle);
const __m256i v_maxval_b = _mm256_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
do {
int c;
for (c = 0; c < w; c += 32) {
const __m256i v_r0_b = yy_loadu_256(mask + 2 * c);
const __m256i v_r1_b = yy_loadu_256(mask + 2 * c + 32);
const __m256i v_r0_s_b = _mm256_shuffle_epi8(v_r0_b, v_shuffle_b);
const __m256i v_r1_s_b = _mm256_shuffle_epi8(v_r1_b, v_shuffle_b);
const __m256i v_al_b =
_mm256_avg_epu8(v_r0_s_b, _mm256_srli_si256(v_r0_s_b, 8));
const __m256i v_ah_b =
_mm256_avg_epu8(v_r1_s_b, _mm256_srli_si256(v_r1_s_b, 8));
const __m256i v_m0_b =
_mm256_permute4x64_epi64(_mm256_unpacklo_epi64(v_al_b, v_ah_b), 0xd8);
const __m256i v_m1_b = _mm256_sub_epi8(v_maxval_b, v_m0_b);
const __m256i v_res_b = blend_32_u8_avx2(
src0 + c, src1 + c, &v_m0_b, &v_m1_b, AOM_BLEND_A64_ROUND_BITS);
yy_storeu_256(dst + c, v_res_b);
}
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += mask_stride;
} while (--h);
}
static INLINE void blend_a64_mask_sx_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h) {
const __m128i v_shuffle_b = xx_loadu_128(g_blend_a64_mask_shuffle);
const __m128i v_maxval_b = _mm_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
const __m128i _r = _mm_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_BITS));
switch (w) {
case 4:
do {
const __m128i v_r_b = xx_loadl_64(mask);
const __m128i v_r0_s_b = _mm_shuffle_epi8(v_r_b, v_shuffle_b);
const __m128i v_r_lo_b = _mm_unpacklo_epi64(v_r0_s_b, v_r0_s_b);
const __m128i v_r_hi_b = _mm_unpackhi_epi64(v_r0_s_b, v_r0_s_b);
const __m128i v_m0_b = _mm_avg_epu8(v_r_lo_b, v_r_hi_b);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_4_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storel_32(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += mask_stride;
} while (--h);
break;
case 8:
do {
const __m128i v_r_b = xx_loadu_128(mask);
const __m128i v_r0_s_b = _mm_shuffle_epi8(v_r_b, v_shuffle_b);
const __m128i v_r_lo_b = _mm_unpacklo_epi64(v_r0_s_b, v_r0_s_b);
const __m128i v_r_hi_b = _mm_unpackhi_epi64(v_r0_s_b, v_r0_s_b);
const __m128i v_m0_b = _mm_avg_epu8(v_r_lo_b, v_r_hi_b);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_8_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storel_64(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += mask_stride;
} while (--h);
break;
case 16:
blend_a64_mask_sx_w16_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, h);
break;
default:
blend_a64_mask_sx_w32n_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, w, h);
break;
}
}
static INLINE void blend_a64_mask_sy_w16_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h) {
const __m128i _r = _mm_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_BITS));
const __m128i v_maxval_b = _mm_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
do {
const __m128i v_ra_b = xx_loadu_128(mask);
const __m128i v_rb_b = xx_loadu_128(mask + mask_stride);
const __m128i v_m0_b = _mm_avg_epu8(v_ra_b, v_rb_b);
const __m128i v_m1_b = _mm_sub_epi16(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_16_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storeu_128(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += 2 * mask_stride;
} while (--h);
}
static INLINE void blend_a64_mask_sy_w32n_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h) {
const __m256i v_maxval_b = _mm256_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
do {
int c;
for (c = 0; c < w; c += 32) {
const __m256i v_ra_b = yy_loadu_256(mask + c);
const __m256i v_rb_b = yy_loadu_256(mask + c + mask_stride);
const __m256i v_m0_b = _mm256_avg_epu8(v_ra_b, v_rb_b);
const __m256i v_m1_b = _mm256_sub_epi8(v_maxval_b, v_m0_b);
const __m256i v_res_b = blend_32_u8_avx2(
src0 + c, src1 + c, &v_m0_b, &v_m1_b, AOM_BLEND_A64_ROUND_BITS);
yy_storeu_256(dst + c, v_res_b);
}
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += 2 * mask_stride;
} while (--h);
}
static INLINE void blend_a64_mask_sy_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h) {
const __m128i _r = _mm_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_BITS));
const __m128i v_maxval_b = _mm_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
switch (w) {
case 4:
do {
const __m128i v_ra_b = xx_loadl_32(mask);
const __m128i v_rb_b = xx_loadl_32(mask + mask_stride);
const __m128i v_m0_b = _mm_avg_epu8(v_ra_b, v_rb_b);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_4_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storel_32(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += 2 * mask_stride;
} while (--h);
break;
case 8:
do {
const __m128i v_ra_b = xx_loadl_64(mask);
const __m128i v_rb_b = xx_loadl_64(mask + mask_stride);
const __m128i v_m0_b = _mm_avg_epu8(v_ra_b, v_rb_b);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_8_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storel_64(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += 2 * mask_stride;
} while (--h);
break;
case 16:
blend_a64_mask_sy_w16_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, h);
break;
default:
blend_a64_mask_sy_w32n_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, w, h);
}
}
static INLINE void blend_a64_mask_w32n_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h) {
const __m256i v_maxval_b = _mm256_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
do {
int c;
for (c = 0; c < w; c += 32) {
const __m256i v_m0_b = yy_loadu_256(mask + c);
const __m256i v_m1_b = _mm256_sub_epi8(v_maxval_b, v_m0_b);
const __m256i v_res_b = blend_32_u8_avx2(
src0 + c, src1 + c, &v_m0_b, &v_m1_b, AOM_BLEND_A64_ROUND_BITS);
yy_storeu_256(dst + c, v_res_b);
}
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += mask_stride;
} while (--h);
}
static INLINE void blend_a64_mask_avx2(
uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h) {
const __m128i v_maxval_b = _mm_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
const __m128i _r = _mm_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_BITS));
switch (w) {
case 4:
do {
const __m128i v_m0_b = xx_loadl_32(mask);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_4_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storel_32(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += mask_stride;
} while (--h);
break;
case 8:
do {
const __m128i v_m0_b = xx_loadl_64(mask);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_8_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storel_64(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += mask_stride;
} while (--h);
break;
case 16:
do {
const __m128i v_m0_b = xx_loadu_128(mask);
const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);
const __m128i v_res_b = blend_16_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);
xx_storeu_128(dst, v_res_b);
dst += dst_stride;
src0 += src0_stride;
src1 += src1_stride;
mask += mask_stride;
} while (--h);
break;
default:
blend_a64_mask_w32n_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, w, h);
}
}
void aom_blend_a64_mask_avx2(uint8_t *dst, uint32_t dst_stride,
const uint8_t *src0, uint32_t src0_stride,
const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w,
int h, int subw, int subh) {
assert(IMPLIES(src0 == dst, src0_stride == dst_stride));
assert(IMPLIES(src1 == dst, src1_stride == dst_stride));
assert(h >= 1);
assert(w >= 1);
assert(IS_POWER_OF_TWO(h));
assert(IS_POWER_OF_TWO(w));
if (UNLIKELY((h | w) & 3)) { // if (w <= 2 || h <= 2)
aom_blend_a64_mask_c(dst, dst_stride, src0, src0_stride, src1, src1_stride,
mask, mask_stride, w, h, subw, subh);
} else {
if (subw & subh) {
blend_a64_mask_sx_sy_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, w, h);
} else if (subw) {
blend_a64_mask_sx_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, w, h);
} else if (subh) {
blend_a64_mask_sy_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, w, h);
} else {
blend_a64_mask_avx2(dst, dst_stride, src0, src0_stride, src1, src1_stride,
mask, mask_stride, w, h);
}
}
}
#if CONFIG_AV1_HIGHBITDEPTH
//////////////////////////////////////////////////////////////////////////////
// aom_highbd_blend_a64_d16_mask_avx2()
//////////////////////////////////////////////////////////////////////////////
static INLINE void highbd_blend_a64_d16_mask_w4_avx2(
uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
const CONV_BUF_TYPE *src1, int src1_stride, const __m256i *mask0,
const __m256i *round_offset, int shift, const __m256i *clip_low,
const __m256i *clip_high, const __m256i *mask_max) {
// Load 4x u16 pixels from each of 4 rows from each source
const __m256i s0 = _mm256_set_epi64x(*(int64_t *)(src0 + 3 * src0_stride),
*(int64_t *)(src0 + 2 * src0_stride),
*(int64_t *)(src0 + 1 * src0_stride),
*(int64_t *)(src0 + 0 * src0_stride));
const __m256i s1 = _mm256_set_epi64x(*(int64_t *)(src1 + 3 * src1_stride),
*(int64_t *)(src1 + 2 * src1_stride),
*(int64_t *)(src1 + 1 * src1_stride),
*(int64_t *)(src1 + 0 * src1_stride));
// Generate the inverse mask
const __m256i mask1 = _mm256_sub_epi16(*mask_max, *mask0);
// Multiply each mask by the respective source
const __m256i mul0_highs = _mm256_mulhi_epu16(*mask0, s0);
const __m256i mul0_lows = _mm256_mullo_epi16(*mask0, s0);
const __m256i mul0h = _mm256_unpackhi_epi16(mul0_lows, mul0_highs);
const __m256i mul0l = _mm256_unpacklo_epi16(mul0_lows, mul0_highs);
// Note that AVX2 unpack orders 64-bit words as [3 1] [2 0] to keep within
// lanes Later, packs does the same again which cancels this out with no need
// for a permute. The intermediate values being reordered makes no difference
const __m256i mul1_highs = _mm256_mulhi_epu16(mask1, s1);
const __m256i mul1_lows = _mm256_mullo_epi16(mask1, s1);
const __m256i mul1h = _mm256_unpackhi_epi16(mul1_lows, mul1_highs);
const __m256i mul1l = _mm256_unpacklo_epi16(mul1_lows, mul1_highs);
const __m256i sumh = _mm256_add_epi32(mul0h, mul1h);
const __m256i suml = _mm256_add_epi32(mul0l, mul1l);
const __m256i roundh =
_mm256_srai_epi32(_mm256_sub_epi32(sumh, *round_offset), shift);
const __m256i roundl =
_mm256_srai_epi32(_mm256_sub_epi32(suml, *round_offset), shift);
const __m256i pack = _mm256_packs_epi32(roundl, roundh);
const __m256i clip =
_mm256_min_epi16(_mm256_max_epi16(pack, *clip_low), *clip_high);
// _mm256_extract_epi64 doesn't exist on x86, so do it the old-fashioned way:
const __m128i cliph = _mm256_extracti128_si256(clip, 1);
xx_storel_64(dst + 3 * dst_stride, _mm_srli_si128(cliph, 8));
xx_storel_64(dst + 2 * dst_stride, cliph);
const __m128i clipl = _mm256_castsi256_si128(clip);
xx_storel_64(dst + 1 * dst_stride, _mm_srli_si128(clipl, 8));
xx_storel_64(dst + 0 * dst_stride, clipl);
}
static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w4_avx2(
uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h,
const __m256i *round_offset, int shift, const __m256i *clip_low,
const __m256i *clip_high, const __m256i *mask_max) {
do {
// Load 8x u8 pixels from each of 4 rows of the mask, pad each to u16
const __m128i mask08 = _mm_set_epi32(*(int32_t *)(mask + 3 * mask_stride),
*(int32_t *)(mask + 2 * mask_stride),
*(int32_t *)(mask + 1 * mask_stride),
*(int32_t *)(mask + 0 * mask_stride));
const __m256i mask0 = _mm256_cvtepu8_epi16(mask08);
highbd_blend_a64_d16_mask_w4_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, &mask0, round_offset, shift,
clip_low, clip_high, mask_max);
dst += dst_stride * 4;
src0 += src0_stride * 4;
src1 += src1_stride * 4;
mask += mask_stride * 4;
} while (h -= 4);
}
static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w4_avx2(
uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int h,
const __m256i *round_offset, int shift, const __m256i *clip_low,
const __m256i *clip_high, const __m256i *mask_max) {
const __m256i one_b = _mm256_set1_epi8(1);
const __m256i two_w = _mm256_set1_epi16(2);
do {
// Load 8 pixels from each of 8 rows of mask,
// (saturating) add together rows then use madd to add adjacent pixels
// Finally, divide each value by 4 (with rounding)
const __m256i m0246 =
_mm256_set_epi64x(*(int64_t *)(mask + 6 * mask_stride),
*(int64_t *)(mask + 4 * mask_stride),
*(int64_t *)(mask + 2 * mask_stride),
*(int64_t *)(mask + 0 * mask_stride));
const __m256i m1357 =
_mm256_set_epi64x(*(int64_t *)(mask + 7 * mask_stride),
*(int64_t *)(mask + 5 * mask_stride),
*(int64_t *)(mask + 3 * mask_stride),
*(int64_t *)(mask + 1 * mask_stride));
const __m256i addrows = _mm256_adds_epu8(m0246, m1357);
const __m256i adjacent = _mm256_maddubs_epi16(addrows, one_b);
const __m256i mask0 =
_mm256_srli_epi16(_mm256_add_epi16(adjacent, two_w), 2);
highbd_blend_a64_d16_mask_w4_avx2(dst, dst_stride, src0, src0_stride, src1,
src1_stride, &mask0, round_offset, shift,
clip_low, clip_high, mask_max);
dst += dst_stride * 4;
src0 += src0_stride * 4;
src1 += src1_stride * 4;
mask += mask_stride * 8;
} while (h -= 4);
}
static INLINE void highbd_blend_a64_d16_mask_w8_avx2(
uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
const CONV_BUF_TYPE *src1, int src1_stride, const __m256i *mask0a,
const __m256i *mask0b, const __m256i *round_offset, int shift,
const __m256i *clip_low, const __m256i *clip_high,
const __m256i *mask_max) {
// Load 8x u16 pixels from each of 4 rows from each source
const __m256i s0a =
yy_loadu2_128(src0 + 0 * src0_stride, src0 + 1 * src0_stride);
const __m256i s0b =
yy_loadu2_128(src0 + 2 * src0_stride, src0 + 3 * src0_stride);
const __m256i s1a =
yy_loadu2_128(src1 + 0 * src1_stride, src1 + 1 * src1_stride);
const __m256i s1b =
yy_loadu2_128(src1 + 2 * src1_stride, src1 + 3 * src1_stride);
// Generate inverse masks
const __m256i mask1a = _mm256_sub_epi16(*mask_max, *mask0a);
const __m256i mask1b = _mm256_sub_epi16(*mask_max, *mask0b);
// Multiply sources by respective masks
const __m256i mul0a_highs = _mm256_mulhi_epu16(*mask0a, s0a);
const __m256i mul0a_lows = _mm256_mullo_epi16(*mask0a, s0a);
const __m256i mul0ah = _mm256_unpackhi_epi16(mul0a_lows, mul0a_highs);
const __m256i mul0al = _mm256_unpacklo_epi16(mul0a_lows, mul0a_highs);
// Note that AVX2 unpack orders 64-bit words as [3 1] [2 0] to keep within
// lanes Later, packs does the same again which cancels this out with no need
// for a permute. The intermediate values being reordered makes no difference
const __m256i mul1a_highs = _mm256_mulhi_epu16(mask1a, s1a);
const __m256i mul1a_lows = _mm256_mullo_epi16(mask1a, s1a);
const __m256i mul1ah = _mm256_unpackhi_epi16(mul1a_lows, mul1a_highs);
const __m256i mul1al = _mm256_unpacklo_epi16(mul1a_lows, mul1a_highs);
const __m256i sumah = _mm256_add_epi32(mul0ah, mul1ah);
const __m256i sumal = _mm256_add_epi32(mul0al, mul1al);
const __m256i mul0b_highs = _mm256_mulhi_epu16(*mask0b, s0b);
const __m256i mul0b_lows = _mm256_mullo_epi16(*mask0b, s0b);
const __m256i mul0bh = _mm256_unpackhi_epi16(mul0b_lows, mul0b_highs);
const __m256i mul0bl = _mm256_unpacklo_epi16(mul0b_lows, mul0b_highs);
const __m256i mul1b_highs = _mm256_mulhi_epu16(mask1b, s1b);
const __m256i mul1b_lows = _mm256_mullo_epi16(mask1b, s1b);
const __m256i mul1bh = _mm256_unpackhi_epi16(mul1b_lows, mul1b_highs);
const __m256i mul1bl = _mm256_unpacklo_epi16(mul1b_lows, mul1b_highs);
const __m256i sumbh = _mm256_add_epi32(mul0bh, mul1bh);
const __m256i sumbl = _mm256_add_epi32(mul0bl, mul1bl);
// Divide down each result, with rounding
const __m256i roundah =
_mm256_srai_epi32(_mm256_sub_epi32(sumah, *round_offset), shift);
const __m256i roundal =
_mm256_srai_epi32(_mm256_sub_epi32(sumal, *round_offset), shift);
const __m256i roundbh =
_mm256_srai_epi32(_mm256_sub_epi32(sumbh, *round_offset), shift);
const __m256i roundbl =
_mm256_srai_epi32(_mm256_sub_epi32(sumbl, *round_offset), shift);
// Pack each i32 down to an i16 with saturation, then clip to valid range
const __m256i packa = _mm256_packs_epi32(roundal, roundah);
const __m256i clipa =
_mm256_min_epi16(_mm256_max_epi16(packa, *clip_low), *clip_high);
const __m256i packb = _mm256_packs_epi32(roundbl, roundbh);
const __m256i clipb =
_mm256_min_epi16(_mm256_max_epi16(packb, *clip_low), *clip_high);
// Store 8x u16 pixels to each of 4 rows in the destination
yy_storeu2_128(dst + 0 * dst_stride, dst + 1 * dst_stride, clipa);
yy_storeu2_128(dst + 2 * dst_stride, dst + 3 * dst_stride, clipb);
}
static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w8_avx2(
uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
const CONV_BUF_TYPE *src1, int src1_stride, const uint8_t *mask,
int mask_stride, int h, const __m256i *round_offset, int shift,
const __m256i *clip_low, const __m256i *clip_high,
const __m256i *mask_max) {
do {
// Load 8x u8 pixels from each of 4 rows in the mask
const __m128i mask0a8 =
_mm_set_epi64x(*(int64_t *)mask, *(uint64_t *)(mask + mask_stride));
const __m128i mask0b8 =
_mm_set_epi64x(*(int64_t *)(mask + 2 * mask_stride),
*(int64_t *)(mask + 3 * mask_stride));
const __m256i mask0a = _mm256_cvtepu8_epi16(mask0a8);
const __m256i mask0b = _mm256_cvtepu8_epi16(mask0b8);
highbd_blend_a64_d16_mask_w8_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask0a, &mask0b,
round_offset, shift, clip_low, clip_high, mask_max);
dst += dst_stride * 4;
src0 += src0_stride * 4;
src1 += src1_stride * 4;
mask += mask_stride * 4;
} while (h -= 4);
}
static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w8_avx2(
uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
const CONV_BUF_TYPE *src1, int src1_stride, const uint8_t *mask,
int mask_stride, int h, const __m256i *round_offset, int shift,
const __m256i *clip_low, const __m256i *clip_high,
const __m256i *mask_max) {
const __m256i one_b = _mm256_set1_epi8(1);
const __m256i two_w = _mm256_set1_epi16(2);
do {
// Load 16x u8 pixels from each of 8 rows in the mask,
// (saturating) add together rows then use madd to add adjacent pixels
// Finally, divide each value by 4 (with rounding)
const __m256i m02 =
yy_loadu2_128(mask + 0 * mask_stride, mask + 2 * mask_stride);
const __m256i m13 =
yy_loadu2_128(mask + 1 * mask_stride, mask + 3 * mask_stride);
const __m256i m0123 =
_mm256_maddubs_epi16(_mm256_adds_epu8(m02, m13), one_b);
const __m256i mask_0a =
_mm256_srli_epi16(_mm256_add_epi16(m0123, two_w), 2);
const __m256i m46 =
yy_loadu2_128(mask + 4 * mask_stride, mask + 6 * mask_stride);
const __m256i m57 =
yy_loadu2_128(mask + 5 * mask_stride, mask + 7 * mask_stride);
const __m256i m4567 =
_mm256_maddubs_epi16(_mm256_adds_epu8(m46, m57), one_b);
const __m256i mask_0b =
_mm256_srli_epi16(_mm256_add_epi16(m4567, two_w), 2);
highbd_blend_a64_d16_mask_w8_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask_0a,
&mask_0b, round_offset, shift, clip_low, clip_high, mask_max);
dst += dst_stride * 4;
src0 += src0_stride * 4;
src1 += src1_stride * 4;
mask += mask_stride * 8;
} while (h -= 4);
}
static INLINE void highbd_blend_a64_d16_mask_w16_avx2(
uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
const CONV_BUF_TYPE *src1, int src1_stride, const __m256i *mask0a,
const __m256i *mask0b, const __m256i *round_offset, int shift,
const __m256i *clip_low, const __m256i *clip_high,
const __m256i *mask_max) {
// Load 16x pixels from each of 2 rows from each source
const __m256i s0a = yy_loadu_256(src0);
const __m256i s0b = yy_loadu_256(src0 + src0_stride);
const __m256i s1a = yy_loadu_256(src1);
const __m256i s1b = yy_loadu_256(src1 + src1_stride);
// Calculate inverse masks
const __m256i mask1a = _mm256_sub_epi16(*mask_max, *mask0a);
const __m256i mask1b = _mm256_sub_epi16(*mask_max, *mask0b);
// Multiply each source by appropriate mask
const __m256i mul0a_highs = _mm256_mulhi_epu16(*mask0a, s0a);
const __m256i mul0a_lows = _mm256_mullo_epi16(*mask0a, s0a);
const __m256i mul0ah = _mm256_unpackhi_epi16(mul0a_lows, mul0a_highs);
const __m256i mul0al = _mm256_unpacklo_epi16(mul0a_lows, mul0a_highs);
// Note that AVX2 unpack orders 64-bit words as [3 1] [2 0] to keep within
// lanes Later, packs does the same again which cancels this out with no need
// for a permute. The intermediate values being reordered makes no difference
const __m256i mul1a_highs = _mm256_mulhi_epu16(mask1a, s1a);
const __m256i mul1a_lows = _mm256_mullo_epi16(mask1a, s1a);
const __m256i mul1ah = _mm256_unpackhi_epi16(mul1a_lows, mul1a_highs);
const __m256i mul1al = _mm256_unpacklo_epi16(mul1a_lows, mul1a_highs);
const __m256i mulah = _mm256_add_epi32(mul0ah, mul1ah);
const __m256i mulal = _mm256_add_epi32(mul0al, mul1al);
const __m256i mul0b_highs = _mm256_mulhi_epu16(*mask0b, s0b);
const __m256i mul0b_lows = _mm256_mullo_epi16(*mask0b, s0b);
const __m256i mul0bh = _mm256_unpackhi_epi16(mul0b_lows, mul0b_highs);
const __m256i mul0bl = _mm256_unpacklo_epi16(mul0b_lows, mul0b_highs);
const __m256i mul1b_highs = _mm256_mulhi_epu16(mask1b, s1b);
const __m256i mul1b_lows = _mm256_mullo_epi16(mask1b, s1b);
const __m256i mul1bh = _mm256_unpackhi_epi16(mul1b_lows, mul1b_highs);
const __m256i mul1bl = _mm256_unpacklo_epi16(mul1b_lows, mul1b_highs);
const __m256i mulbh = _mm256_add_epi32(mul0bh, mul1bh);
const __m256i mulbl = _mm256_add_epi32(mul0bl, mul1bl);
const __m256i resah =
_mm256_srai_epi32(_mm256_sub_epi32(mulah, *round_offset), shift);
const __m256i resal =
_mm256_srai_epi32(_mm256_sub_epi32(mulal, *round_offset), shift);
const __m256i resbh =
_mm256_srai_epi32(_mm256_sub_epi32(mulbh, *round_offset), shift);
const __m256i resbl =
_mm256_srai_epi32(_mm256_sub_epi32(mulbl, *round_offset), shift);
// Signed saturating pack from i32 to i16:
const __m256i packa = _mm256_packs_epi32(resal, resah);
const __m256i packb = _mm256_packs_epi32(resbl, resbh);
// Clip the values to the valid range
const __m256i clipa =
_mm256_min_epi16(_mm256_max_epi16(packa, *clip_low), *clip_high);
const __m256i clipb =
_mm256_min_epi16(_mm256_max_epi16(packb, *clip_low), *clip_high);
// Store 16 pixels
yy_storeu_256(dst, clipa);
yy_storeu_256(dst + dst_stride, clipb);
}
static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w16_avx2(
uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
const CONV_BUF_TYPE *src1, int src1_stride, const uint8_t *mask,
int mask_stride, int h, int w, const __m256i *round_offset, int shift,
const __m256i *clip_low, const __m256i *clip_high,
const __m256i *mask_max) {
for (int i = 0; i < h; i += 2) {
for (int j = 0; j < w; j += 16) {
// Load 16x u8 alpha-mask values from each of two rows and pad to u16
const __m128i masks_a8 = xx_loadu_128(mask + j);
const __m128i masks_b8 = xx_loadu_128(mask + mask_stride + j);
const __m256i mask0a = _mm256_cvtepu8_epi16(masks_a8);
const __m256i mask0b = _mm256_cvtepu8_epi16(masks_b8);
highbd_blend_a64_d16_mask_w16_avx2(
dst + j, dst_stride, src0 + j, src0_stride, src1 + j, src1_stride,
&mask0a, &mask0b, round_offset, shift, clip_low, clip_high, mask_max);
}
dst += dst_stride * 2;
src0 += src0_stride * 2;
src1 += src1_stride * 2;
mask += mask_stride * 2;
}
}
static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w16_avx2(
uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
const CONV_BUF_TYPE *src1, int src1_stride, const uint8_t *mask,
int mask_stride, int h, int w, const __m256i *round_offset, int shift,
const __m256i *clip_low, const __m256i *clip_high,
const __m256i *mask_max) {
const __m256i one_b = _mm256_set1_epi8(1);
const __m256i two_w = _mm256_set1_epi16(2);
for (int i = 0; i < h; i += 2) {
for (int j = 0; j < w; j += 16) {
// Load 32x u8 alpha-mask values from each of four rows
// (saturating) add pairs of rows, then use madd to add adjacent values
// Finally, divide down each result with rounding
const __m256i m0 = yy_loadu_256(mask + 0 * mask_stride + 2 * j);
const __m256i m1 = yy_loadu_256(mask + 1 * mask_stride + 2 * j);
const __m256i m2 = yy_loadu_256(mask + 2 * mask_stride + 2 * j);
const __m256i m3 = yy_loadu_256(mask + 3 * mask_stride + 2 * j);
const __m256i m01_8 = _mm256_adds_epu8(m0, m1);
const __m256i m23_8 = _mm256_adds_epu8(m2, m3);
const __m256i m01 = _mm256_maddubs_epi16(m01_8, one_b);
const __m256i m23 = _mm256_maddubs_epi16(m23_8, one_b);
const __m256i mask0a = _mm256_srli_epi16(_mm256_add_epi16(m01, two_w), 2);
const __m256i mask0b = _mm256_srli_epi16(_mm256_add_epi16(m23, two_w), 2);
highbd_blend_a64_d16_mask_w16_avx2(
dst + j, dst_stride, src0 + j, src0_stride, src1 + j, src1_stride,
&mask0a, &mask0b, round_offset, shift, clip_low, clip_high, mask_max);
}
dst += dst_stride * 2;
src0 += src0_stride * 2;
src1 += src1_stride * 2;
mask += mask_stride * 4;
}
}
void aom_highbd_blend_a64_d16_mask_avx2(
uint8_t *dst8, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
const uint8_t *mask, uint32_t mask_stride, int w, int h, int subw, int subh,
ConvolveParams *conv_params, const int bd) {
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
const int round_bits =
2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
const int32_t round_offset =
((1 << (round_bits + bd)) + (1 << (round_bits + bd - 1)) -
(1 << (round_bits - 1)))
<< AOM_BLEND_A64_ROUND_BITS;
const __m256i v_round_offset = _mm256_set1_epi32(round_offset);
const int shift = round_bits + AOM_BLEND_A64_ROUND_BITS;
const __m256i clip_low = _mm256_setzero_si256();
const __m256i clip_high = _mm256_set1_epi16((1 << bd) - 1);
const __m256i mask_max = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
assert(IMPLIES((void *)src0 == dst, src0_stride == dst_stride));
assert(IMPLIES((void *)src1 == dst, src1_stride == dst_stride));
assert(h >= 4);
assert(w >= 4);
assert(IS_POWER_OF_TWO(h));
assert(IS_POWER_OF_TWO(w));
if (subw == 0 && subh == 0) {
switch (w) {
case 4:
highbd_blend_a64_d16_mask_subw0_subh0_w4_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
&mask_max);
break;
case 8:
highbd_blend_a64_d16_mask_subw0_subh0_w8_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
&mask_max);
break;
default: // >= 16
highbd_blend_a64_d16_mask_subw0_subh0_w16_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, w, &v_round_offset, shift, &clip_low, &clip_high,
&mask_max);
break;
}
} else if (subw == 1 && subh == 1) {
switch (w) {
case 4:
highbd_blend_a64_d16_mask_subw1_subh1_w4_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
&mask_max);
break;
case 8:
highbd_blend_a64_d16_mask_subw1_subh1_w8_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
&mask_max);
break;
default: // >= 16
highbd_blend_a64_d16_mask_subw1_subh1_w16_avx2(
dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
mask_stride, h, w, &v_round_offset, shift, &clip_low, &clip_high,
&mask_max);
break;
}
} else {
// Sub-sampling in only one axis doesn't seem to happen very much, so fall
// back to the vanilla C implementation instead of having all the optimised
// code for these.
aom_highbd_blend_a64_d16_mask_c(dst8, dst_stride, src0, src0_stride, src1,
src1_stride, mask, mask_stride, w, h, subw,
subh, conv_params, bd);
}
}
#endif // CONFIG_AV1_HIGHBITDEPTH
|