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
|
/*
* Copyright 2013 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for Visual C x86.
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
// Offsets for source bytes 0 to 9
static uvec8 kShuf0 =
{ 0, 1, 3, 4, 5, 7, 8, 9, 128, 128, 128, 128, 128, 128, 128, 128 };
// Offsets for source bytes 11 to 20 with 8 subtracted = 3 to 12.
static uvec8 kShuf1 =
{ 3, 4, 5, 7, 8, 9, 11, 12, 128, 128, 128, 128, 128, 128, 128, 128 };
// Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31.
static uvec8 kShuf2 =
{ 5, 7, 8, 9, 11, 12, 13, 15, 128, 128, 128, 128, 128, 128, 128, 128 };
// Offsets for source bytes 0 to 10
static uvec8 kShuf01 =
{ 0, 1, 1, 2, 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10 };
// Offsets for source bytes 10 to 21 with 8 subtracted = 3 to 13.
static uvec8 kShuf11 =
{ 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13 };
// Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31.
static uvec8 kShuf21 =
{ 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13, 13, 14, 14, 15 };
// Coefficients for source bytes 0 to 10
static uvec8 kMadd01 =
{ 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2 };
// Coefficients for source bytes 10 to 21
static uvec8 kMadd11 =
{ 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1 };
// Coefficients for source bytes 21 to 31
static uvec8 kMadd21 =
{ 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3 };
// Coefficients for source bytes 21 to 31
static vec16 kRound34 =
{ 2, 2, 2, 2, 2, 2, 2, 2 };
static uvec8 kShuf38a =
{ 0, 3, 6, 8, 11, 14, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 };
static uvec8 kShuf38b =
{ 128, 128, 128, 128, 128, 128, 0, 3, 6, 8, 11, 14, 128, 128, 128, 128 };
// Arrange words 0,3,6 into 0,1,2
static uvec8 kShufAc =
{ 0, 1, 6, 7, 12, 13, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 };
// Arrange words 0,3,6 into 3,4,5
static uvec8 kShufAc3 =
{ 128, 128, 128, 128, 128, 128, 0, 1, 6, 7, 12, 13, 128, 128, 128, 128 };
// Scaling values for boxes of 3x3 and 2x3
static uvec16 kScaleAc33 =
{ 65536 / 9, 65536 / 9, 65536 / 6, 65536 / 9, 65536 / 9, 65536 / 6, 0, 0 };
// Arrange first value for pixels 0,1,2,3,4,5
static uvec8 kShufAb0 =
{ 0, 128, 3, 128, 6, 128, 8, 128, 11, 128, 14, 128, 128, 128, 128, 128 };
// Arrange second value for pixels 0,1,2,3,4,5
static uvec8 kShufAb1 =
{ 1, 128, 4, 128, 7, 128, 9, 128, 12, 128, 15, 128, 128, 128, 128, 128 };
// Arrange third value for pixels 0,1,2,3,4,5
static uvec8 kShufAb2 =
{ 2, 128, 5, 128, 128, 128, 10, 128, 13, 128, 128, 128, 128, 128, 128, 128 };
// Scaling values for boxes of 3x2 and 2x2
static uvec16 kScaleAb2 =
{ 65536 / 3, 65536 / 3, 65536 / 2, 65536 / 3, 65536 / 3, 65536 / 2, 0, 0 };
// Reads 32 pixels, throws half away and writes 16 pixels.
// Alignment requirement: src_ptr 16 byte aligned, dst_ptr 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleRowDown2_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
mov eax, [esp + 4] // src_ptr
// src_stride ignored
mov edx, [esp + 12] // dst_ptr
mov ecx, [esp + 16] // dst_width
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
lea eax, [eax + 32]
psrlw xmm0, 8 // isolate odd pixels.
psrlw xmm1, 8
packuswb xmm0, xmm1
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 16
jg wloop
ret
}
}
// Blends 32x1 rectangle to 16x1.
// Alignment requirement: src_ptr 16 byte aligned, dst_ptr 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleRowDown2Linear_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
mov eax, [esp + 4] // src_ptr
// src_stride
mov edx, [esp + 12] // dst_ptr
mov ecx, [esp + 16] // dst_width
pcmpeqb xmm5, xmm5 // generate mask 0x00ff00ff
psrlw xmm5, 8
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
lea eax, [eax + 32]
movdqa xmm2, xmm0 // average columns (32 to 16 pixels)
psrlw xmm0, 8
movdqa xmm3, xmm1
psrlw xmm1, 8
pand xmm2, xmm5
pand xmm3, xmm5
pavgw xmm0, xmm2
pavgw xmm1, xmm3
packuswb xmm0, xmm1
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 16
jg wloop
ret
}
}
// Blends 32x2 rectangle to 16x1.
// Alignment requirement: src_ptr 16 byte aligned, dst_ptr 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleRowDown2Box_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
push esi
mov eax, [esp + 4 + 4] // src_ptr
mov esi, [esp + 4 + 8] // src_stride
mov edx, [esp + 4 + 12] // dst_ptr
mov ecx, [esp + 4 + 16] // dst_width
pcmpeqb xmm5, xmm5 // generate mask 0x00ff00ff
psrlw xmm5, 8
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
movdqu xmm2, [eax + esi]
movdqu xmm3, [eax + esi + 16]
lea eax, [eax + 32]
pavgb xmm0, xmm2 // average rows
pavgb xmm1, xmm3
movdqa xmm2, xmm0 // average columns (32 to 16 pixels)
psrlw xmm0, 8
movdqa xmm3, xmm1
psrlw xmm1, 8
pand xmm2, xmm5
pand xmm3, xmm5
pavgw xmm0, xmm2
pavgw xmm1, xmm3
packuswb xmm0, xmm1
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 16
jg wloop
pop esi
ret
}
}
// Point samples 32 pixels to 8 pixels.
// Alignment requirement: src_ptr 16 byte aligned, dst_ptr 8 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleRowDown4_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
mov eax, [esp + 4] // src_ptr
// src_stride ignored
mov edx, [esp + 12] // dst_ptr
mov ecx, [esp + 16] // dst_width
pcmpeqb xmm5, xmm5 // generate mask 0x00ff0000
psrld xmm5, 24
pslld xmm5, 16
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
lea eax, [eax + 32]
pand xmm0, xmm5
pand xmm1, xmm5
packuswb xmm0, xmm1
psrlw xmm0, 8
packuswb xmm0, xmm0
movq qword ptr [edx], xmm0
lea edx, [edx + 8]
sub ecx, 8
jg wloop
ret
}
}
// Blends 32x4 rectangle to 8x1.
// Alignment requirement: src_ptr 16 byte aligned, dst_ptr 8 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleRowDown4Box_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
push esi
push edi
mov eax, [esp + 8 + 4] // src_ptr
mov esi, [esp + 8 + 8] // src_stride
mov edx, [esp + 8 + 12] // dst_ptr
mov ecx, [esp + 8 + 16] // dst_width
lea edi, [esi + esi * 2] // src_stride * 3
pcmpeqb xmm7, xmm7 // generate mask 0x00ff00ff
psrlw xmm7, 8
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
movdqu xmm2, [eax + esi]
movdqu xmm3, [eax + esi + 16]
pavgb xmm0, xmm2 // average rows
pavgb xmm1, xmm3
movdqu xmm2, [eax + esi * 2]
movdqu xmm3, [eax + esi * 2 + 16]
movdqu xmm4, [eax + edi]
movdqu xmm5, [eax + edi + 16]
lea eax, [eax + 32]
pavgb xmm2, xmm4
pavgb xmm3, xmm5
pavgb xmm0, xmm2
pavgb xmm1, xmm3
movdqa xmm2, xmm0 // average columns (32 to 16 pixels)
psrlw xmm0, 8
movdqa xmm3, xmm1
psrlw xmm1, 8
pand xmm2, xmm7
pand xmm3, xmm7
pavgw xmm0, xmm2
pavgw xmm1, xmm3
packuswb xmm0, xmm1
movdqa xmm2, xmm0 // average columns (16 to 8 pixels)
psrlw xmm0, 8
pand xmm2, xmm7
pavgw xmm0, xmm2
packuswb xmm0, xmm0
movq qword ptr [edx], xmm0
lea edx, [edx + 8]
sub ecx, 8
jg wloop
pop edi
pop esi
ret
}
}
// Point samples 32 pixels to 24 pixels.
// Produces three 8 byte values. For each 8 bytes, 16 bytes are read.
// Then shuffled to do the scaling.
// Note that movdqa+palign may be better than movdqu.
// Alignment requirement: src_ptr 16 byte aligned, dst_ptr 8 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleRowDown34_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
mov eax, [esp + 4] // src_ptr
// src_stride ignored
mov edx, [esp + 12] // dst_ptr
mov ecx, [esp + 16] // dst_width
movdqa xmm3, kShuf0
movdqa xmm4, kShuf1
movdqa xmm5, kShuf2
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
lea eax, [eax + 32]
movdqa xmm2, xmm1
palignr xmm1, xmm0, 8
pshufb xmm0, xmm3
pshufb xmm1, xmm4
pshufb xmm2, xmm5
movq qword ptr [edx], xmm0
movq qword ptr [edx + 8], xmm1
movq qword ptr [edx + 16], xmm2
lea edx, [edx + 24]
sub ecx, 24
jg wloop
ret
}
}
// Blends 32x2 rectangle to 24x1
// Produces three 8 byte values. For each 8 bytes, 16 bytes are read.
// Then shuffled to do the scaling.
// Register usage:
// xmm0 src_row 0
// xmm1 src_row 1
// xmm2 shuf 0
// xmm3 shuf 1
// xmm4 shuf 2
// xmm5 madd 0
// xmm6 madd 1
// xmm7 kRound34
// Note that movdqa+palign may be better than movdqu.
// Alignment requirement: src_ptr 16 byte aligned, dst_ptr 8 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleRowDown34_1_Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
push esi
mov eax, [esp + 4 + 4] // src_ptr
mov esi, [esp + 4 + 8] // src_stride
mov edx, [esp + 4 + 12] // dst_ptr
mov ecx, [esp + 4 + 16] // dst_width
movdqa xmm2, kShuf01
movdqa xmm3, kShuf11
movdqa xmm4, kShuf21
movdqa xmm5, kMadd01
movdqa xmm6, kMadd11
movdqa xmm7, kRound34
wloop:
movdqu xmm0, [eax] // pixels 0..7
movdqu xmm1, [eax + esi]
pavgb xmm0, xmm1
pshufb xmm0, xmm2
pmaddubsw xmm0, xmm5
paddsw xmm0, xmm7
psrlw xmm0, 2
packuswb xmm0, xmm0
movq qword ptr [edx], xmm0
movdqu xmm0, [eax + 8] // pixels 8..15
movdqu xmm1, [eax + esi + 8]
pavgb xmm0, xmm1
pshufb xmm0, xmm3
pmaddubsw xmm0, xmm6
paddsw xmm0, xmm7
psrlw xmm0, 2
packuswb xmm0, xmm0
movq qword ptr [edx + 8], xmm0
movdqu xmm0, [eax + 16] // pixels 16..23
movdqu xmm1, [eax + esi + 16]
lea eax, [eax + 32]
pavgb xmm0, xmm1
pshufb xmm0, xmm4
movdqa xmm1, kMadd21
pmaddubsw xmm0, xmm1
paddsw xmm0, xmm7
psrlw xmm0, 2
packuswb xmm0, xmm0
movq qword ptr [edx + 16], xmm0
lea edx, [edx + 24]
sub ecx, 24
jg wloop
pop esi
ret
}
}
// Note that movdqa+palign may be better than movdqu.
// Alignment requirement: src_ptr 16 byte aligned, dst_ptr 8 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleRowDown34_0_Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
push esi
mov eax, [esp + 4 + 4] // src_ptr
mov esi, [esp + 4 + 8] // src_stride
mov edx, [esp + 4 + 12] // dst_ptr
mov ecx, [esp + 4 + 16] // dst_width
movdqa xmm2, kShuf01
movdqa xmm3, kShuf11
movdqa xmm4, kShuf21
movdqa xmm5, kMadd01
movdqa xmm6, kMadd11
movdqa xmm7, kRound34
wloop:
movdqu xmm0, [eax] // pixels 0..7
movdqu xmm1, [eax + esi]
pavgb xmm1, xmm0
pavgb xmm0, xmm1
pshufb xmm0, xmm2
pmaddubsw xmm0, xmm5
paddsw xmm0, xmm7
psrlw xmm0, 2
packuswb xmm0, xmm0
movq qword ptr [edx], xmm0
movdqu xmm0, [eax + 8] // pixels 8..15
movdqu xmm1, [eax + esi + 8]
pavgb xmm1, xmm0
pavgb xmm0, xmm1
pshufb xmm0, xmm3
pmaddubsw xmm0, xmm6
paddsw xmm0, xmm7
psrlw xmm0, 2
packuswb xmm0, xmm0
movq qword ptr [edx + 8], xmm0
movdqu xmm0, [eax + 16] // pixels 16..23
movdqu xmm1, [eax + esi + 16]
lea eax, [eax + 32]
pavgb xmm1, xmm0
pavgb xmm0, xmm1
pshufb xmm0, xmm4
movdqa xmm1, kMadd21
pmaddubsw xmm0, xmm1
paddsw xmm0, xmm7
psrlw xmm0, 2
packuswb xmm0, xmm0
movq qword ptr [edx + 16], xmm0
lea edx, [edx+24]
sub ecx, 24
jg wloop
pop esi
ret
}
}
// 3/8 point sampler
// Scale 32 pixels to 12
__declspec(naked) __declspec(align(16))
void ScaleRowDown38_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
mov eax, [esp + 4] // src_ptr
// src_stride ignored
mov edx, [esp + 12] // dst_ptr
mov ecx, [esp + 16] // dst_width
movdqa xmm4, kShuf38a
movdqa xmm5, kShuf38b
xloop:
movdqu xmm0, [eax] // 16 pixels -> 0,1,2,3,4,5
movdqu xmm1, [eax + 16] // 16 pixels -> 6,7,8,9,10,11
lea eax, [eax + 32]
pshufb xmm0, xmm4
pshufb xmm1, xmm5
paddusb xmm0, xmm1
movq qword ptr [edx], xmm0 // write 12 pixels
movhlps xmm1, xmm0
movd [edx + 8], xmm1
lea edx, [edx + 12]
sub ecx, 12
jg xloop
ret
}
}
// Scale 16x3 pixels to 6x1 with interpolation
__declspec(naked) __declspec(align(16))
void ScaleRowDown38_3_Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
push esi
mov eax, [esp + 4 + 4] // src_ptr
mov esi, [esp + 4 + 8] // src_stride
mov edx, [esp + 4 + 12] // dst_ptr
mov ecx, [esp + 4 + 16] // dst_width
movdqa xmm2, kShufAc
movdqa xmm3, kShufAc3
movdqa xmm4, kScaleAc33
pxor xmm5, xmm5
xloop:
movdqu xmm0, [eax] // sum up 3 rows into xmm0/1
movdqu xmm6, [eax + esi]
movhlps xmm1, xmm0
movhlps xmm7, xmm6
punpcklbw xmm0, xmm5
punpcklbw xmm1, xmm5
punpcklbw xmm6, xmm5
punpcklbw xmm7, xmm5
paddusw xmm0, xmm6
paddusw xmm1, xmm7
movdqu xmm6, [eax + esi * 2]
lea eax, [eax + 16]
movhlps xmm7, xmm6
punpcklbw xmm6, xmm5
punpcklbw xmm7, xmm5
paddusw xmm0, xmm6
paddusw xmm1, xmm7
movdqa xmm6, xmm0 // 8 pixels -> 0,1,2 of xmm6
psrldq xmm0, 2
paddusw xmm6, xmm0
psrldq xmm0, 2
paddusw xmm6, xmm0
pshufb xmm6, xmm2
movdqa xmm7, xmm1 // 8 pixels -> 3,4,5 of xmm6
psrldq xmm1, 2
paddusw xmm7, xmm1
psrldq xmm1, 2
paddusw xmm7, xmm1
pshufb xmm7, xmm3
paddusw xmm6, xmm7
pmulhuw xmm6, xmm4 // divide by 9,9,6, 9,9,6
packuswb xmm6, xmm6
movd [edx], xmm6 // write 6 pixels
psrlq xmm6, 16
movd [edx + 2], xmm6
lea edx, [edx + 6]
sub ecx, 6
jg xloop
pop esi
ret
}
}
// Scale 16x2 pixels to 6x1 with interpolation
__declspec(naked) __declspec(align(16))
void ScaleRowDown38_2_Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
__asm {
push esi
mov eax, [esp + 4 + 4] // src_ptr
mov esi, [esp + 4 + 8] // src_stride
mov edx, [esp + 4 + 12] // dst_ptr
mov ecx, [esp + 4 + 16] // dst_width
movdqa xmm2, kShufAb0
movdqa xmm3, kShufAb1
movdqa xmm4, kShufAb2
movdqa xmm5, kScaleAb2
xloop:
movdqu xmm0, [eax] // average 2 rows into xmm0
movdqu xmm1, [eax + esi]
lea eax, [eax + 16]
pavgb xmm0, xmm1
movdqa xmm1, xmm0 // 16 pixels -> 0,1,2,3,4,5 of xmm1
pshufb xmm1, xmm2
movdqa xmm6, xmm0
pshufb xmm6, xmm3
paddusw xmm1, xmm6
pshufb xmm0, xmm4
paddusw xmm1, xmm0
pmulhuw xmm1, xmm5 // divide by 3,3,2, 3,3,2
packuswb xmm1, xmm1
movd [edx], xmm1 // write 6 pixels
psrlq xmm1, 16
movd [edx + 2], xmm1
lea edx, [edx + 6]
sub ecx, 6
jg xloop
pop esi
ret
}
}
// Reads 16xN bytes and produces 16 shorts at a time.
// TODO(fbarchard): Make this handle 4xN bytes for any width ARGB.
__declspec(naked) __declspec(align(16))
void ScaleAddRows_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
uint16* dst_ptr, int src_width,
int src_height) {
__asm {
push esi
push edi
push ebx
push ebp
mov esi, [esp + 16 + 4] // src_ptr
mov edx, [esp + 16 + 8] // src_stride
mov edi, [esp + 16 + 12] // dst_ptr
mov ecx, [esp + 16 + 16] // dst_width
mov ebx, [esp + 16 + 20] // height
pxor xmm4, xmm4
dec ebx
xloop:
// first row
movdqu xmm0, [esi]
lea eax, [esi + edx]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm4
punpckhbw xmm1, xmm4
lea esi, [esi + 16]
mov ebp, ebx
test ebp, ebp
je ydone
// sum remaining rows
yloop:
movdqu xmm2, [eax] // read 16 pixels
lea eax, [eax + edx] // advance to next row
movdqa xmm3, xmm2
punpcklbw xmm2, xmm4
punpckhbw xmm3, xmm4
paddusw xmm0, xmm2 // sum 16 words
paddusw xmm1, xmm3
sub ebp, 1
jg yloop
ydone:
movdqu [edi], xmm0
movdqu [edi + 16], xmm1
lea edi, [edi + 32]
sub ecx, 16
jg xloop
pop ebp
pop ebx
pop edi
pop esi
ret
}
}
// Bilinear column filtering. SSSE3 version.
// TODO(fbarchard): Port to Neon
// TODO(fbarchard): Switch the following:
// xor ebx, ebx
// mov bx, word ptr [esi + eax] // 2 source x0 pixels
// To
// movzx ebx, word ptr [esi + eax] // 2 source x0 pixels
// when drmemory bug fixed.
// https://code.google.com/p/drmemory/issues/detail?id=1396
__declspec(naked) __declspec(align(16))
void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x, int dx) {
__asm {
push ebx
push esi
push edi
mov edi, [esp + 12 + 4] // dst_ptr
mov esi, [esp + 12 + 8] // src_ptr
mov ecx, [esp + 12 + 12] // dst_width
movd xmm2, [esp + 12 + 16] // x
movd xmm3, [esp + 12 + 20] // dx
mov eax, 0x04040000 // shuffle to line up fractions with pixel.
movd xmm5, eax
pcmpeqb xmm6, xmm6 // generate 0x007f for inverting fraction.
psrlw xmm6, 9
pextrw eax, xmm2, 1 // get x0 integer. preroll
sub ecx, 2
jl xloop29
movdqa xmm0, xmm2 // x1 = x0 + dx
paddd xmm0, xmm3
punpckldq xmm2, xmm0 // x0 x1
punpckldq xmm3, xmm3 // dx dx
paddd xmm3, xmm3 // dx * 2, dx * 2
pextrw edx, xmm2, 3 // get x1 integer. preroll
// 2 Pixel loop.
xloop2:
movdqa xmm1, xmm2 // x0, x1 fractions.
paddd xmm2, xmm3 // x += dx
movzx ebx, word ptr [esi + eax] // 2 source x0 pixels
movd xmm0, ebx
psrlw xmm1, 9 // 7 bit fractions.
movzx ebx, word ptr [esi + edx] // 2 source x1 pixels
movd xmm4, ebx
pshufb xmm1, xmm5 // 0011
punpcklwd xmm0, xmm4
pxor xmm1, xmm6 // 0..7f and 7f..0
pmaddubsw xmm0, xmm1 // 16 bit, 2 pixels.
pextrw eax, xmm2, 1 // get x0 integer. next iteration.
pextrw edx, xmm2, 3 // get x1 integer. next iteration.
psrlw xmm0, 7 // 8.7 fixed point to low 8 bits.
packuswb xmm0, xmm0 // 8 bits, 2 pixels.
movd ebx, xmm0
mov [edi], bx
lea edi, [edi + 2]
sub ecx, 2 // 2 pixels
jge xloop2
xloop29:
add ecx, 2 - 1
jl xloop99
// 1 pixel remainder
movzx ebx, word ptr [esi + eax] // 2 source x0 pixels
movd xmm0, ebx
psrlw xmm2, 9 // 7 bit fractions.
pshufb xmm2, xmm5 // 0011
pxor xmm2, xmm6 // 0..7f and 7f..0
pmaddubsw xmm0, xmm2 // 16 bit
psrlw xmm0, 7 // 8.7 fixed point to low 8 bits.
packuswb xmm0, xmm0 // 8 bits
movd ebx, xmm0
mov [edi], bl
xloop99:
pop edi
pop esi
pop ebx
ret
}
}
// Reads 16 pixels, duplicates them and writes 32 pixels.
// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleColsUp2_SSE2(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x, int dx) {
__asm {
mov edx, [esp + 4] // dst_ptr
mov eax, [esp + 8] // src_ptr
mov ecx, [esp + 12] // dst_width
wloop:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm0
punpckhbw xmm1, xmm1
movdqu [edx], xmm0
movdqu [edx + 16], xmm1
lea edx, [edx + 32]
sub ecx, 32
jg wloop
ret
}
}
// Reads 8 pixels, throws half away and writes 4 even pixels (0, 2, 4, 6)
// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleARGBRowDown2_SSE2(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
__asm {
mov eax, [esp + 4] // src_argb
// src_stride ignored
mov edx, [esp + 12] // dst_argb
mov ecx, [esp + 16] // dst_width
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
lea eax, [eax + 32]
shufps xmm0, xmm1, 0xdd
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
jg wloop
ret
}
}
// Blends 8x1 rectangle to 4x1.
// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleARGBRowDown2Linear_SSE2(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
__asm {
mov eax, [esp + 4] // src_argb
// src_stride ignored
mov edx, [esp + 12] // dst_argb
mov ecx, [esp + 16] // dst_width
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
lea eax, [eax + 32]
movdqa xmm2, xmm0
shufps xmm0, xmm1, 0x88 // even pixels
shufps xmm2, xmm1, 0xdd // odd pixels
pavgb xmm0, xmm2
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
jg wloop
ret
}
}
// Blends 8x2 rectangle to 4x1.
// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleARGBRowDown2Box_SSE2(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
__asm {
push esi
mov eax, [esp + 4 + 4] // src_argb
mov esi, [esp + 4 + 8] // src_stride
mov edx, [esp + 4 + 12] // dst_argb
mov ecx, [esp + 4 + 16] // dst_width
wloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + 16]
movdqu xmm2, [eax + esi]
movdqu xmm3, [eax + esi + 16]
lea eax, [eax + 32]
pavgb xmm0, xmm2 // average rows
pavgb xmm1, xmm3
movdqa xmm2, xmm0 // average columns (8 to 4 pixels)
shufps xmm0, xmm1, 0x88 // even pixels
shufps xmm2, xmm1, 0xdd // odd pixels
pavgb xmm0, xmm2
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
jg wloop
pop esi
ret
}
}
// Reads 4 pixels at a time.
// Alignment requirement: dst_argb 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleARGBRowDownEven_SSE2(const uint8* src_argb, ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb, int dst_width) {
__asm {
push ebx
push edi
mov eax, [esp + 8 + 4] // src_argb
// src_stride ignored
mov ebx, [esp + 8 + 12] // src_stepx
mov edx, [esp + 8 + 16] // dst_argb
mov ecx, [esp + 8 + 20] // dst_width
lea ebx, [ebx * 4]
lea edi, [ebx + ebx * 2]
wloop:
movd xmm0, [eax]
movd xmm1, [eax + ebx]
punpckldq xmm0, xmm1
movd xmm2, [eax + ebx * 2]
movd xmm3, [eax + edi]
lea eax, [eax + ebx * 4]
punpckldq xmm2, xmm3
punpcklqdq xmm0, xmm2
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
jg wloop
pop edi
pop ebx
ret
}
}
// Blends four 2x2 to 4x1.
// Alignment requirement: dst_argb 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleARGBRowDownEvenBox_SSE2(const uint8* src_argb,
ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb, int dst_width) {
__asm {
push ebx
push esi
push edi
mov eax, [esp + 12 + 4] // src_argb
mov esi, [esp + 12 + 8] // src_stride
mov ebx, [esp + 12 + 12] // src_stepx
mov edx, [esp + 12 + 16] // dst_argb
mov ecx, [esp + 12 + 20] // dst_width
lea esi, [eax + esi] // row1 pointer
lea ebx, [ebx * 4]
lea edi, [ebx + ebx * 2]
wloop:
movq xmm0, qword ptr [eax] // row0 4 pairs
movhps xmm0, qword ptr [eax + ebx]
movq xmm1, qword ptr [eax + ebx * 2]
movhps xmm1, qword ptr [eax + edi]
lea eax, [eax + ebx * 4]
movq xmm2, qword ptr [esi] // row1 4 pairs
movhps xmm2, qword ptr [esi + ebx]
movq xmm3, qword ptr [esi + ebx * 2]
movhps xmm3, qword ptr [esi + edi]
lea esi, [esi + ebx * 4]
pavgb xmm0, xmm2 // average rows
pavgb xmm1, xmm3
movdqa xmm2, xmm0 // average columns (8 to 4 pixels)
shufps xmm0, xmm1, 0x88 // even pixels
shufps xmm2, xmm1, 0xdd // odd pixels
pavgb xmm0, xmm2
movdqu [edx], xmm0
lea edx, [edx + 16]
sub ecx, 4
jg wloop
pop edi
pop esi
pop ebx
ret
}
}
// Column scaling unfiltered. SSE2 version.
__declspec(naked) __declspec(align(16))
void ScaleARGBCols_SSE2(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
__asm {
push edi
push esi
mov edi, [esp + 8 + 4] // dst_argb
mov esi, [esp + 8 + 8] // src_argb
mov ecx, [esp + 8 + 12] // dst_width
movd xmm2, [esp + 8 + 16] // x
movd xmm3, [esp + 8 + 20] // dx
pshufd xmm2, xmm2, 0 // x0 x0 x0 x0
pshufd xmm0, xmm3, 0x11 // dx 0 dx 0
paddd xmm2, xmm0
paddd xmm3, xmm3 // 0, 0, 0, dx * 2
pshufd xmm0, xmm3, 0x05 // dx * 2, dx * 2, 0, 0
paddd xmm2, xmm0 // x3 x2 x1 x0
paddd xmm3, xmm3 // 0, 0, 0, dx * 4
pshufd xmm3, xmm3, 0 // dx * 4, dx * 4, dx * 4, dx * 4
pextrw eax, xmm2, 1 // get x0 integer.
pextrw edx, xmm2, 3 // get x1 integer.
cmp ecx, 0
jle xloop99
sub ecx, 4
jl xloop49
// 4 Pixel loop.
xloop4:
movd xmm0, [esi + eax * 4] // 1 source x0 pixels
movd xmm1, [esi + edx * 4] // 1 source x1 pixels
pextrw eax, xmm2, 5 // get x2 integer.
pextrw edx, xmm2, 7 // get x3 integer.
paddd xmm2, xmm3 // x += dx
punpckldq xmm0, xmm1 // x0 x1
movd xmm1, [esi + eax * 4] // 1 source x2 pixels
movd xmm4, [esi + edx * 4] // 1 source x3 pixels
pextrw eax, xmm2, 1 // get x0 integer. next iteration.
pextrw edx, xmm2, 3 // get x1 integer. next iteration.
punpckldq xmm1, xmm4 // x2 x3
punpcklqdq xmm0, xmm1 // x0 x1 x2 x3
movdqu [edi], xmm0
lea edi, [edi + 16]
sub ecx, 4 // 4 pixels
jge xloop4
xloop49:
test ecx, 2
je xloop29
// 2 Pixels.
movd xmm0, [esi + eax * 4] // 1 source x0 pixels
movd xmm1, [esi + edx * 4] // 1 source x1 pixels
pextrw eax, xmm2, 5 // get x2 integer.
punpckldq xmm0, xmm1 // x0 x1
movq qword ptr [edi], xmm0
lea edi, [edi + 8]
xloop29:
test ecx, 1
je xloop99
// 1 Pixels.
movd xmm0, [esi + eax * 4] // 1 source x2 pixels
movd dword ptr [edi], xmm0
xloop99:
pop esi
pop edi
ret
}
}
// Bilinear row filtering combines 2x1 -> 1x1. SSSE3 version.
// TODO(fbarchard): Port to Neon
// Shuffle table for arranging 2 pixels into pairs for pmaddubsw
static uvec8 kShuffleColARGB = {
0u, 4u, 1u, 5u, 2u, 6u, 3u, 7u, // bbggrraa 1st pixel
8u, 12u, 9u, 13u, 10u, 14u, 11u, 15u // bbggrraa 2nd pixel
};
// Shuffle table for duplicating 2 fractions into 8 bytes each
static uvec8 kShuffleFractions = {
0u, 0u, 0u, 0u, 0u, 0u, 0u, 0u, 4u, 4u, 4u, 4u, 4u, 4u, 4u, 4u,
};
__declspec(naked) __declspec(align(16))
void ScaleARGBFilterCols_SSSE3(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
__asm {
push esi
push edi
mov edi, [esp + 8 + 4] // dst_argb
mov esi, [esp + 8 + 8] // src_argb
mov ecx, [esp + 8 + 12] // dst_width
movd xmm2, [esp + 8 + 16] // x
movd xmm3, [esp + 8 + 20] // dx
movdqa xmm4, kShuffleColARGB
movdqa xmm5, kShuffleFractions
pcmpeqb xmm6, xmm6 // generate 0x007f for inverting fraction.
psrlw xmm6, 9
pextrw eax, xmm2, 1 // get x0 integer. preroll
sub ecx, 2
jl xloop29
movdqa xmm0, xmm2 // x1 = x0 + dx
paddd xmm0, xmm3
punpckldq xmm2, xmm0 // x0 x1
punpckldq xmm3, xmm3 // dx dx
paddd xmm3, xmm3 // dx * 2, dx * 2
pextrw edx, xmm2, 3 // get x1 integer. preroll
// 2 Pixel loop.
xloop2:
movdqa xmm1, xmm2 // x0, x1 fractions.
paddd xmm2, xmm3 // x += dx
movq xmm0, qword ptr [esi + eax * 4] // 2 source x0 pixels
psrlw xmm1, 9 // 7 bit fractions.
movhps xmm0, qword ptr [esi + edx * 4] // 2 source x1 pixels
pshufb xmm1, xmm5 // 0000000011111111
pshufb xmm0, xmm4 // arrange pixels into pairs
pxor xmm1, xmm6 // 0..7f and 7f..0
pmaddubsw xmm0, xmm1 // argb_argb 16 bit, 2 pixels.
pextrw eax, xmm2, 1 // get x0 integer. next iteration.
pextrw edx, xmm2, 3 // get x1 integer. next iteration.
psrlw xmm0, 7 // argb 8.7 fixed point to low 8 bits.
packuswb xmm0, xmm0 // argb_argb 8 bits, 2 pixels.
movq qword ptr [edi], xmm0
lea edi, [edi + 8]
sub ecx, 2 // 2 pixels
jge xloop2
xloop29:
add ecx, 2 - 1
jl xloop99
// 1 pixel remainder
psrlw xmm2, 9 // 7 bit fractions.
movq xmm0, qword ptr [esi + eax * 4] // 2 source x0 pixels
pshufb xmm2, xmm5 // 00000000
pshufb xmm0, xmm4 // arrange pixels into pairs
pxor xmm2, xmm6 // 0..7f and 7f..0
pmaddubsw xmm0, xmm2 // argb 16 bit, 1 pixel.
psrlw xmm0, 7
packuswb xmm0, xmm0 // argb 8 bits, 1 pixel.
movd [edi], xmm0
xloop99:
pop edi
pop esi
ret
}
}
// Reads 4 pixels, duplicates them and writes 8 pixels.
// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
__declspec(naked) __declspec(align(16))
void ScaleARGBColsUp2_SSE2(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
__asm {
mov edx, [esp + 4] // dst_argb
mov eax, [esp + 8] // src_argb
mov ecx, [esp + 12] // dst_width
wloop:
movdqu xmm0, [eax]
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpckldq xmm0, xmm0
punpckhdq xmm1, xmm1
movdqu [edx], xmm0
movdqu [edx + 16], xmm1
lea edx, [edx + 32]
sub ecx, 8
jg wloop
ret
}
}
// Divide num by div and return as 16.16 fixed point result.
__declspec(naked) __declspec(align(16))
int FixedDiv_X86(int num, int div) {
__asm {
mov eax, [esp + 4] // num
cdq // extend num to 64 bits
shld edx, eax, 16 // 32.16
shl eax, 16
idiv dword ptr [esp + 8]
ret
}
}
// Divide num by div and return as 16.16 fixed point result.
__declspec(naked) __declspec(align(16))
int FixedDiv1_X86(int num, int div) {
__asm {
mov eax, [esp + 4] // num
mov ecx, [esp + 8] // denom
cdq // extend num to 64 bits
shld edx, eax, 16 // 32.16
shl eax, 16
sub eax, 0x00010001
sbb edx, 0
sub ecx, 1
idiv ecx
ret
}
}
#endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
|