-
Notifications
You must be signed in to change notification settings - Fork 259
/
Copy pathloopfilter16_sse.asm
1793 lines (1689 loc) · 47.6 KB
/
loopfilter16_sse.asm
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
; Copyright © 2021, VideoLAN and dav1d authors
; Copyright © 2021, Two Orioles, LLC
; All rights reserved.
;
; Redistribution and use in source and binary forms, with or without
; modification, are permitted provided that the following conditions are met:
;
; 1. Redistributions of source code must retain the above copyright notice, this
; list of conditions and the following disclaimer.
;
; 2. Redistributions in binary form must reproduce the above copyright notice,
; this list of conditions and the following disclaimer in the documentation
; and/or other materials provided with the distribution.
;
; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
; ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
; WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
; DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
; ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
; (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
; ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
%include "config.asm"
%include "ext/x86/x86inc.asm"
SECTION_RODATA 16
%if ARCH_X86_64
%define PIC_sym(a) a
%else
%define PIC_base $$
%define PIC_sym(a) pic_regq+a-PIC_base
%endif
pb_4x1_4x5_4x9_4x13: times 4 db 0, 1
times 4 db 8, 9
pw_1: times 8 dw 1
pw_2: times 8 dw 2
pw_3: times 8 dw 3
; 4 and 16 need to be next to each other since they are used as alternates
; depending on whether bitdepth is 10 or 12
pw_4: times 8 dw 4
pw_16: times 8 dw 16
pw_8: times 8 dw 8
pw_4096: times 8 dw 4096
pb_mask: dd 1, 1, 2, 2
SECTION .text
%if ARCH_X86_32
%if STACK_ALIGNMENT < 16
%define extra_stack 2
%else
%define extra_stack 0
%endif
%endif
%macro RELOC_ARGS 2 ; h/v, off
ASSERT ARCH_X86_32
%if STACK_ALIGNMENT < 16
mov r5d, [rstk + stack_offset + 4*4 + 4]
%define lstridem [esp+%2+0*gprsize]
mov lstridem, r5d
mov r5d, [rstk + stack_offset + 4*5 + 4]
%define lutm [esp+%2+1*gprsize]
mov lutm, r5d
mov r5d, [rstk + stack_offset + 4*6 + 4]
%ifidn %1, v
%define wm [esp+%2+2*gprsize]
mov wm, r5d
mov r5d, [rstk + stack_offset + 4*3 + 4]
%define lm [esp+%2+3*gprsize]
mov lm, r5d
%else ; %1 == h
%define hm [esp+%2+2*gprsize]
mov hm, r5d
%endif ; %1==v
mov r5d, r7m
%define bdmulm [esp+%2+4*gprsize]
mov bdmulm, r5d
%else
%define lstridem r4m
%define lutm r5m
%ifidn %1, v
%define wm r6m
%define lm r3m
%else
%define hm r6m
%endif
%define bdmulm r7m
%endif ; STACK_ALIGNMENT
%endmacro
%macro UNRELOC_ARGS 0
%if ARCH_X86_32
%undef lm
%undef lstridem
%undef wm
%undef hm
%undef lutm
%endif
%endmacro
%macro SPLATD 2
movd %1, %2
pshufd %1, %1, q0000
%endmacro
%macro SPLATW 2
movd %1, %2
pshuflw %1, %1, q0000
punpcklqdq %1, %1
%endmacro
; in: out:
; mm%1 a b c d a e i m
; mm%2 e f g h b f j n
; mm%3 i j k l -> c g k o
; mm%4 m n o p d h l p
%macro TRANSPOSE4X4W 5
punpcklwd m%5, m%1, m%2
punpckhwd m%1, m%2
punpcklwd m%2, m%3, m%4
punpckhwd m%3, m%4
punpckldq m%4, m%5, m%2
punpckhdq m%5, m%2
punpckldq m%2, m%1, m%3
punpckhdq m%1, m%3
SWAP %1, %4
SWAP %2, %5, %3
%endmacro
; in: out:
; m%1 a b c d e f g h a i q y 6 E M U
; m%2 i j k l m n o p b j r z 7 F N V
; m%3 q r s t u v w x c k s 0 8 G O W
; m%4 y z 0 1 2 3 4 5 d l t 1 9 H P X
; m%5 6 7 8 9 A B C D -> e m u 2 A I Q Y
; m%6 E F G H I J K L f n v 3 B J R Z
; m%7 M N O P Q R S T g o w 4 C K S +
; m%8 U V W X Y Z + = h p x 5 D L T =
%if ARCH_X86_64
%macro TRANSPOSE8X8W 9
; m%1 a b c d e f g h a i q y b j r z
; m%2 i j k l m n o p c k s 0 d l t 1
; m%3 q r s t u v w x -> e m u 2 f n v 3
; m%4 y z 0 1 2 3 4 5 g o w 4 h p x 5
TRANSPOSE4X4W %1, %2, %3, %4, %9
; m%5 6 7 8 9 A B C D 6 E M U 7 F N V
; m%6 E F G H I J K L 8 G O W 9 H P X
; m%7 M N O P Q R S T -> A I Q Y B J R Z
; m%8 U V W X Y Z + = C K S + D L T =
TRANSPOSE4X4W %5, %6, %7, %8, %9
; m%1 a i q y b j r z a i q y 6 E M U
; m%2 c k s 0 d l t 1 b j r z 7 F N V
; m%3 e m u 2 f n v 3 c k s 0 8 G O W
; m%4 g o w 4 h p x 5 d l t 1 9 H P X
; m%5 6 E M U 7 F N V -> e m u 2 A I Q Y
; m%6 8 G O W 9 H P X f n v 3 B J R Z
; m%7 A I Q Y B J R Z g o w 4 C K S +
; m%8 C K S + D L T = h p x 5 D L T =
punpckhqdq m%9, m%1, m%5
punpcklqdq m%1, m%5
punpckhqdq m%5, m%2, m%6
punpcklqdq m%2, m%6
punpckhqdq m%6, m%3, m%7
punpcklqdq m%3, m%7
punpckhqdq m%7, m%4, m%8
punpcklqdq m%4, m%8
SWAP %8, %7, %4, %5, %3, %2, %9
%endmacro
%else ; x86-32
; input: 1-7 in registers, 8 in first memory [read-only]
; second memory is scratch, and may overlap with first or third memory
; output: 1-5,7-8 in registers, 6 in third memory [write-only]
%macro TRANSPOSE8X8W 13 ; regs [8x], mem [3x], a/u [in/out alignment [2x]
TRANSPOSE4X4W %1, %2, %3, %4, %8
%ifnidn %9, ""
mov%12 m%8, %9
%else
mova m%8, %10
%endif
mova %10, m%4
TRANSPOSE4X4W %5, %6, %7, %8, %4
punpckhqdq m%4, m%1, m%5
punpcklqdq m%1, m%5
punpckhqdq m%5, m%2, m%6
punpcklqdq m%2, m%6
punpckhqdq m%6, m%3, m%7
punpcklqdq m%3, m%7
mova m%7, %10
%ifnidn %11, ""
mov%13 %11, m%6
%else
mova %10, m%6
%endif
punpckhqdq m%6, m%7, m%8
punpcklqdq m%7, m%8
; 1,4,2,5,3,8,7,6 -> 1,2,3,4,5,6,7,8
SWAP %2, %4, %5, %3
SWAP %6, %8
%endmacro
%endif ; x86-32/64
; transpose and write m8-11, everything else is scratch
%macro TRANSPOSE_8x4_AND_WRITE_4x8 5 ; p1, p0, q0, q1, tmp
; transpose 8x4
punpcklwd %5, %1, %2
punpckhwd %1, %2
punpcklwd %2, %3, %4
punpckhwd %3, %4
punpckldq %4, %5, %2
punpckhdq %5, %2
punpckldq %2, %1, %3
punpckhdq %1, %3
; write out
movq [dstq+strideq*0-4], %4
movhps [dstq+strideq*1-4], %4
movq [dstq+strideq*2-4], %5
movhps [dstq+stride3q -4], %5
lea dstq, [dstq+strideq*4]
movq [dstq+strideq*0-4], %2
movhps [dstq+strideq*1-4], %2
movq [dstq+strideq*2-4], %1
movhps [dstq+stride3q -4], %1
lea dstq, [dstq+strideq*4]
%endmacro
%macro FILTER 2 ; width [4/6/8/16], dir [h/v]
; load data
%ifidn %2, v
%if %1 == 4
%if ARCH_X86_64
%define P1 m8
%define P0 m9
%define Q0 m10
%define Q1 m11
mova P1, [dstq+mstrideq*2] ; p1
mova P0, [dstq+mstrideq*1] ; p0
mova Q0, [dstq+strideq*0] ; q0
mova Q1, [dstq+strideq*1] ; q1
%else ; x86-32
%define P1 [dstq+mstrideq*2]
%define P0 [dstq+mstrideq*1]
%define Q0 [dstq+strideq*0]
%define Q1 [dstq+strideq*1]
%endif ; x86-32/64
%else ; %1 != 4
; load 6-8 pixels, remainder (for wd=16) will be read inline
lea tmpq, [dstq+mstrideq*4]
%if ARCH_X86_64
; we load p3 later
%define P2 m13
%define P1 m8
%define P0 m9
%define Q0 m10
%define Q1 m11
%define Q2 m14
mova P2, [tmpq+strideq*1]
mova P1, [tmpq+strideq*2]
mova P0, [tmpq+stride3q]
mova Q0, [dstq+strideq*0]
mova Q1, [dstq+strideq*1]
mova Q2, [dstq+strideq*2]
%if %1 != 6
%define P3 [tmpq+strideq*0]
%define Q3 m15
mova Q3, [dstq+stride3q]
%endif ; %1 != 6
%else ; x86-32
%define P2 [tmpq+strideq*1]
%define P1 [dstq+mstrideq*2]
%define P0 [dstq+mstrideq*1]
%define Q0 [dstq+strideq*0]
%define Q1 [dstq+strideq*1]
%define Q2 [dstq+strideq*2]
%if %1 != 6
%define P3 [dstq+mstrideq*4]
%define Q3 [dstq+stride3q]
%endif ; %1 != 6
%endif ; x86-32/64
%endif ; %1 ==/!= 4
%else ; %2 != v
; load lines
%if %1 == 4
movq m0, [dstq+strideq*0-4]
movq m2, [dstq+strideq*1-4]
movq m4, [dstq+strideq*2-4]
movq m5, [dstq+stride3q -4]
lea tmpq, [dstq+strideq*4]
movq m3, [tmpq+strideq*0-4]
movq m6, [tmpq+strideq*1-4]
movq m1, [tmpq+strideq*2-4]
movq m7, [tmpq+stride3q -4]
; transpose 4x8
; m0: A-D0
; m2: A-D1
; m4: A-D2
; m5: A-D3
; m3: A-D4
; m6: A-D5
; m1: A-D6
; m7: A-D7
punpcklwd m0, m2
punpcklwd m4, m5
punpcklwd m3, m6
punpcklwd m1, m7
; m0: A0-1,B0-1,C0-1,D0-1
; m4: A2-3,B2-3,C2-3,D2-3
; m3: A4-5,B4-5,C4-5,D4-5
; m1: A6-7,B6-7,C6-7,D6-7
punpckhdq m2, m0, m4
punpckldq m0, m4
punpckhdq m4, m3, m1
punpckldq m3, m1
; m0: A0-3,B0-3
; m2: C0-3,D0-3
; m3: A4-7,B4-7
; m4: C4-7,D4-7
punpckhqdq m1, m0, m3
punpcklqdq m0, m3
punpckhqdq m3, m2, m4
punpcklqdq m2, m4
; m0: A0-7
; m1: B0-7
; m2: C0-7
; m3: D0-7
%if ARCH_X86_64
SWAP 0, 8
SWAP 1, 9
SWAP 2, 10
SWAP 3, 11
%define P1 m8
%define P0 m9
%define Q0 m10
%define Q1 m11
%else
%define P1 [esp+3*mmsize]
%define P0 [esp+4*mmsize]
%define Q0 [esp+5*mmsize]
%define Q1 [esp+6*mmsize]
mova P1, m0
mova P0, m1
mova Q0, m2
mova Q1, m3
%endif
%elif %1 == 6 || %1 == 8
movu m0, [dstq+strideq*0-8]
movu m1, [dstq+strideq*1-8]
movu m2, [dstq+strideq*2-8]
movu m3, [dstq+stride3q -8]
lea tmpq, [dstq+strideq*4]
movu m4, [tmpq+strideq*0-8]
movu m5, [tmpq+strideq*1-8]
movu m6, [tmpq+strideq*2-8]
%if ARCH_X86_64
movu m7, [tmpq+stride3q -8]
%endif
; transpose 8x16
; m0: A-H0,A-H8
; m1: A-H1,A-H9
; m2: A-H2,A-H10
; m3: A-H3,A-H11
; m4: A-H4,A-H12
; m5: A-H5,A-H13
; m6: A-H6,A-H14
; m7: A-H7,A-H15
%if ARCH_X86_64
punpcklwd m8, m0, m1
%else
punpcklwd m7, m0, m1
%endif
punpckhwd m0, m1
punpcklwd m1, m2, m3
punpckhwd m2, m3
punpcklwd m3, m4, m5
punpckhwd m4, m5
%if ARCH_X86_64
punpcklwd m5, m6, m7
punpckhwd m6, m7
%else
mova [rsp+3*16], m4
movu m4, [tmpq+stride3q -8]
punpcklwd m5, m6, m4
punpckhwd m6, m4
%endif
; m8: A0-1,B0-1,C0-1,D0-1 [m7 on x86-32]
; m0: E0-1,F0-1,G0-1,H0-1
; m1: A2-3,B2-3,C2-3,D2-3
; m2: E2-3,F2-3,G2-3,H2-3
; m3: A4-5,B4-5,C4-5,D4-5
; m4: E4-5,F4-5,G4-5,H4-5 [r3 on x86-32]
; m5: A6-7,B6-7,C6-7,D6-7
; m6: E6-7,F6-7,G6-7,H6-7
%if ARCH_X86_64
punpckldq m7, m8, m1
punpckhdq m8, m1
%else
punpckldq m4, m7, m1
punpckhdq m7, m1
%endif
punpckldq m1, m0, m2
punpckhdq m0, m2
punpckldq m2, m3, m5
punpckhdq m3, m5
%if ARCH_X86_64
punpckldq m5, m4, m6
punpckhdq m4, m6
%else
mova [rsp+4*16], m3
mova m3, [rsp+3*16]
punpckldq m5, m3, m6
punpckhdq m3, m6
%endif
; m7: A0-3,B0-3 [m4 on x86-32]
; m8: C0-3,D0-3 [m7 on x86-32]
; m1: E0-3,F0-3
; m0: G0-3,H0-3
; m2: A4-7,B4-7
; m3: C4-7,D4-7 [r4 on x86-32]
; m5: E4-7,F4-7
; m4: G4-7,H4-7 [m3 on x86-32]
%if ARCH_X86_64
%if %1 != 6
punpcklqdq m6, m7, m2
%endif
punpckhqdq m7, m2
punpcklqdq m2, m8, m3
punpckhqdq m8, m3
punpcklqdq m3, m1, m5
punpckhqdq m1, m5
%if %1 != 6
punpckhqdq m5, m0, m4
%endif
punpcklqdq m0, m4
%if %1 == 8
mova [rsp+1*16], m6
%define P3 [rsp+1*16]
%endif
; 7,2,8,3,1,0,5 -> 13,8,9,10,11,14,15
SWAP 7, 13
SWAP 8, 2, 9
SWAP 3, 10
SWAP 1, 11
SWAP 0, 14
SWAP 5, 15
%define P2 m13
%define P1 m8
%define P0 m9
%define Q0 m10
%define Q1 m11
%define Q2 m14
%if %1 == 8
%define Q3 m15
%endif
%else ; x86-32
%if %1 == 8
%define P3 [rsp+ 6*16]
punpcklqdq m6, m4, m2
mova P3, m6
%endif
mova m6, [rsp+4*16]
punpckhqdq m4, m2
punpcklqdq m2, m7, m6
punpckhqdq m7, m6
punpcklqdq m6, m1, m5
punpckhqdq m1, m5
%if %1 == 8
%define Q3 [rsp+24*16]
punpckhqdq m5, m0, m3
mova Q3, m5
%endif
punpcklqdq m0, m3
%if %1 == 8
%define P2 [rsp+18*16]
%define P1 [rsp+19*16]
%define P0 [rsp+20*16]
%define Q0 [rsp+21*16]
%define Q1 [rsp+22*16]
%define Q2 [rsp+23*16]
%else
%define P2 [rsp+3*16]
%define P1 [rsp+4*16]
%define P0 [rsp+5*16]
%define Q0 [rsp+6*16]
%define Q1 [rsp+7*16]
%define Q2 [rsp+8*16]
%endif
mova P2, m4
mova P1, m2
mova P0, m7
mova Q0, m6
mova Q1, m1
mova Q2, m0
%endif ; x86-32/64
%else ; %1 == 16
; We only use 14 pixels but we'll need the remainder at the end for
; the second transpose
mova m0, [dstq+strideq*0-16]
mova m1, [dstq+strideq*1-16]
mova m2, [dstq+strideq*2-16]
mova m3, [dstq+stride3q -16]
lea tmpq, [dstq+strideq*4]
mova m4, [tmpq+strideq*0-16]
mova m5, [tmpq+strideq*1-16]
mova m6, [tmpq+strideq*2-16]
%if ARCH_X86_64
mova m7, [tmpq+stride3q -16]
TRANSPOSE8X8W 0, 1, 2, 3, 4, 5, 6, 7, 8
SWAP 5, 13
SWAP 6, 8
SWAP 7, 9
%define P2 m13
%define P1 m8
%define P0 m9
%else ; x86-32
%define P2 [esp+18*16]
%define P1 [esp+19*16]
%define P0 [esp+20*16]
TRANSPOSE8X8W 0, 1, 2, 3, 4, 5, 6, 7, \
[tmpq+stride3q -16], P2, "", a, a
mova P1, m6
mova P0, m7
%endif ; x86-32/64
mova [rsp+ 7*16], m0
mova [rsp+ 8*16], m1
mova [rsp+ 9*16], m2
mova [rsp+10*16], m3
%define P3 [rsp+6*16]
mova P3, m4
mova m0, [dstq+strideq*0]
mova m1, [dstq+strideq*1]
mova m2, [dstq+strideq*2]
mova m3, [dstq+stride3q ]
lea tmpq, [dstq+strideq*4]
mova m4, [tmpq+strideq*0]
mova m5, [tmpq+strideq*1]
mova m6, [tmpq+strideq*2]
%if ARCH_X86_64
mova m7, [tmpq+stride3q ]
TRANSPOSE8X8W 0, 1, 2, 3, 4, 5, 6, 7, 10
SWAP 0, 10
SWAP 1, 11
SWAP 2, 14
SWAP 3, 15
%define Q0 m10
%define Q1 m11
%define Q2 m14
%define Q3 m15
%else ; x86-32
TRANSPOSE8X8W 0, 1, 2, 3, 4, 5, 6, 7, \
[tmpq+stride3q ], [rsp+12*16], "", a, a
%define Q0 [esp+21*16]
%define Q1 [esp+22*16]
%define Q2 [esp+23*16]
%define Q3 [esp+24*16]
mova Q0, m0
mova Q1, m1
mova Q2, m2
mova Q3, m3
%endif ; x86-32/64
mova [rsp+11*16], m4
%if ARCH_X86_64
mova [rsp+12*16], m5
%endif
mova [rsp+13*16], m6
mova [rsp+14*16], m7
%endif ; %1 == 4/6/8/16
%endif ; %2 ==/!= v
; load L/E/I/H
%if ARCH_X86_32
%define l_strideq r5
mov l_strideq, dword lstridem
%ifidn %2, v
%define lq r3
mov lq, dword lm
%endif
%endif
%ifidn %2, v
%if cpuflag(sse4)
pmovzxbw m1, [lq]
pmovzxbw m0, [lq+l_strideq]
pxor m2, m2
%else ; ssse3
movq m1, [lq]
movq m0, [lq+l_strideq]
pxor m2, m2
REPX {punpcklbw x, m2}, m1, m0
%endif ; ssse3/sse4
%else ; %2 != v
movq m0, [lq] ; l0, l1
movq m1, [lq+l_strideq] ; l2, l3
punpckldq m0, m1 ; l0, l2, l1, l3
pxor m2, m2
punpcklbw m1, m0, m2 ; l0, l2
punpckhbw m0, m2 ; l1, l3
%endif ; %2==/!=v
%if ARCH_X86_32
%ifidn %2, v
%undef lq
mov mstrideq, mstridem
%endif
%endif
pcmpeqw m5, m2, m0
pand m1, m5
por m0, m1 ; l[x][] ? l[x][] : l[x-stride][]
pshufb m0, [PIC_sym(pb_4x1_4x5_4x9_4x13)] ; l[x][1]
pcmpeqw m5, m2, m0 ; !L
psrlw m5, 1
%if ARCH_X86_64
psrlw m2, m0, [lutq+128]
SPLATW m1, [lutq+136]
%else ; x86-32
mov r5, lutm
psrlw m2, m0, [r5+128]
SPLATW m1, [r5+136]
%endif ; x86-32/64
pminsw m2, m1
pmaxsw m2, [PIC_sym(pw_1)] ; I
psrlw m1, m0, 4 ; H
paddw m0, [PIC_sym(pw_2)]
paddw m0, m0
paddw m0, m2 ; E
REPX {pmullw x, [bdmulq]}, m0, m1, m2
%if ARCH_X86_32
%undef l_strideq
lea stride3q, [strideq*3]
%endif
psubw m3, P1, P0 ; p1-p0
psubw m4, Q0, Q1 ; q0-q1
REPX {pabsw x, x}, m3, m4
pmaxsw m3, m5
pmaxsw m3, m4
pcmpgtw m7, m3, m1 ; hev
%if %1 != 4
psubw m4, P2, P0 ; p2-p0
pabsw m4, m4
pmaxsw m4, m3
%if %1 != 6
mova m6, P3 ; p3
psubw m5, m6, P0 ; p3-p0
pabsw m5, m5
pmaxsw m4, m5
%endif ; %1 != 6
psubw m5, Q0, Q2 ; q0-q2
pabsw m5, m5
pmaxsw m4, m5
%if %1 != 6
psubw m5, Q0, Q3 ; q0-q3
pabsw m5, m5
pmaxsw m4, m5
%endif ; %1 != 6
pcmpgtw m4, [bdmulq] ; !flat8in
psubw m5, P2, P1 ; p2-p1
pabsw m5, m5
%if %1 != 6
psubw m6, P2 ; p3-p2
pabsw m6, m6
pmaxsw m5, m6
psubw m6, Q2, Q3 ; q2-q3
pabsw m6, m6
pmaxsw m5, m6
%endif ; %1 != 6
psubw m6, Q2, Q1 ; q2-q1
pabsw m6, m6
pmaxsw m5, m6
%if %1 == 16
SPLATD m6, [maskq+8]
SPLATD m1, [maskq+4]
por m6, m1
pand m6, m12
pcmpeqd m6, m12
pand m5, m6
%else ; %1 != 16
SPLATD m6, [maskq+4]
pand m6, m12
pcmpeqd m6, m12
pand m5, m6 ; only apply fm-wide to wd>4 blocks
%endif ; %1==/!=16
pmaxsw m3, m5
%endif ; %1 != 4
pcmpgtw m3, m2
psubw m5, P1, Q1 ; p1-q1
psubw m6, P0, Q0 ; p0-q0
REPX {pabsw x, x}, m5, m6
paddw m6, m6
psrlw m5, 1
paddw m5, m6 ; abs(p0-q0)*2+(abs(p1-q1)>>1)
pcmpgtw m5, m0 ; abs(p0-q0)*2+(abs(p1-q1)>>1) > E
por m3, m5
%if %1 == 16
%ifidn %2, v
lea tmpq, [dstq+mstrideq*8]
mova m0, [tmpq+strideq*1]
mova m1, [tmpq+strideq*2]
mova m2, [tmpq+stride3q]
%else ; %2 != v
mova m0, [rsp+ 8*16]
mova m1, [rsp+ 9*16]
mova m2, [rsp+10*16]
%endif ; %2==/!=v
REPX {psubw x, P0}, m0, m1, m2
REPX {pabsw x, x}, m0, m1, m2
pmaxsw m1, m0
pmaxsw m1, m2
%ifidn %2, v
lea tmpq, [dstq+strideq*4]
mova m0, [tmpq+strideq*0]
mova m2, [tmpq+strideq*1]
mova m5, [tmpq+strideq*2]
%else ; %2 != v
mova m0, [rsp+11*16]
mova m2, [rsp+12*16]
mova m5, [rsp+13*16]
%endif ; %2==/!=v
REPX {psubw x, Q0}, m0, m2, m5
REPX {pabsw x, x}, m0, m2, m5
pmaxsw m0, m2
pmaxsw m1, m5
pmaxsw m1, m0
pcmpgtw m1, [bdmulq] ; !flat8out
por m1, m4 ; !flat8in | !flat8out
SPLATD m2, [maskq+8]
pand m5, m2, m12
pcmpeqd m5, m12
pandn m1, m5 ; flat16
pandn m5, m3, m1 ; flat16 & fm
SWAP 1, 5
SPLATD m5, [maskq+4]
por m5, m2
pand m2, m5, m12
pcmpeqd m2, m12
pandn m4, m2 ; flat8in
pandn m2, m3, m4
SWAP 2, 4
SPLATD m2, [maskq+0]
por m2, m5
pand m2, m12
pcmpeqd m2, m12
pandn m3, m2
pandn m0, m4, m3 ; fm & !flat8 & !flat16
SWAP 0, 3
pandn m0, m1, m4 ; flat8 & !flat16
SWAP 0, 4
%elif %1 != 4
SPLATD m0, [maskq+4]
pand m2, m0, m12
pcmpeqd m2, m12
pandn m4, m2
pandn m2, m3, m4 ; flat8 & fm
SWAP 2, 4
SPLATD m2, [maskq+0]
por m0, m2
pand m0, m12
pcmpeqd m0, m12
pandn m3, m0
pandn m0, m4, m3 ; fm & !flat8
SWAP 0, 3
%else ; %1 == 4
SPLATD m0, [maskq+0]
pand m0, m12
pcmpeqd m0, m12
pandn m3, m0 ; fm
%endif ; %1==/!=4
; short filter
%if ARCH_X86_64
SPLATW m0, r7m
%else
SPLATW m0, bdmulm
%endif
pcmpeqw m2, m2
psrlw m0, 1 ; 511 or 2047
pxor m2, m0 ; -512 or -2048
psubw m5, Q0, P0 ; q0-p0
paddw m6, m5, m5
paddw m6, m5 ; 3*(q0-p0)
psubw m5, P1, Q1 ; iclip_diff(p1-q1)
pminsw m5, m0
pmaxsw m5, m2
pand m5, m7 ; f=iclip_diff(p1-q1)&hev
paddw m5, m6 ; f=iclip_diff(3*(q0-p0)+f)
pminsw m5, m0
pmaxsw m5, m2
pand m3, m5 ; f&=fm
paddw m5, m3, [PIC_sym(pw_3)]
paddw m3, [PIC_sym(pw_4)]
REPX {pminsw x, m0}, m5, m3
psraw m5, 3 ; f2
psraw m3, 3 ; f1
psubw m0, m2 ; 1023 or 4095
pxor m2, m2
%if ARCH_X86_64
paddw P0, m5
psubw Q0, m3
%else
paddw m5, P0
psubw m6, Q0, m3
REPX {pminsw x, m0}, m5, m6
REPX {pmaxsw x, m2}, m5, m6
%endif
paddw m3, [PIC_sym(pw_1)]
psraw m3, 1 ; f=(f1+1)>>1
pandn m7, m3 ; f&=!hev
SWAP 7, 3
%if ARCH_X86_64
paddw P1, m3
psubw Q1, m3
REPX {pminsw x, m0}, P1, P0, Q0, Q1
REPX {pmaxsw x, m2}, P1, P0, Q0, Q1
%else
psubw m7, Q1, m3
paddw m3, P1
REPX {pminsw x, m0}, m7, m3
REPX {pmaxsw x, m2}, m7, m3
%if %1 > 4
mova P1, m3
mova P0, m5
mova Q0, m6
mova Q1, m7
%endif
%endif
%if %1 == 16
; m8-11 = p1/p0/q0/q1, m4=flat8, m1=flat16
; m12=filter bits mask
; m13-15=p2/q2/q3
; m0,2-3,5-7 = free
; flat16 filter
%ifidn %2, v
lea tmpq, [dstq+mstrideq*8]
mova m0, [tmpq+strideq*1] ; p6
mova m2, [tmpq+strideq*2] ; p5
mova m7, [tmpq+stride3q] ; p4
mova m6, [tmpq+strideq*4] ; p3
lea tmpq, [dstq+mstrideq*4]
%else ; %2 != v
mova m0, [rsp+ 8*16]
mova m2, [rsp+ 9*16]
mova m7, [rsp+10*16]
mova m6, [rsp+ 6*16]
%endif ; %2==/!=v
mova [rsp+ 0*16], m4
; p6*7+p5*2+p4*2+p3+p2+p1+p0+q0
psllw m3, m0, 3 ; p6*8
paddw m3, [PIC_sym(pw_8)]
paddw m5, m2, m7 ; p5+p4
psubw m3, m0
paddw m5, m5 ; (p5+p4)*2
paddw m3, m6 ; p6*7+p3
paddw m5, P2 ; (p5+p4)*2+p2
paddw m3, P1 ; p6*7+p3+p1
paddw m5, P0 ; (p5+p4)*2+p2+p0
paddw m3, Q0 ; p6*7+p3+p1+q0
paddw m3, m5 ; p6*7+p5*2+p4*2+p3+p2+p1+p0+q0
psrlw m5, m3, 4
pand m5, m1
pandn m4, m1, m2
por m5, m4
%ifidn %2, v
mova [tmpq+mstrideq*2], m5 ; p5
%else ; %2 != v
mova [rsp+9*16], m5
%endif ; %2==/!=v
; sub p6*2, add p3/q1
paddw m3, m6
paddw m5, m0, m0
paddw m3, Q1
psubw m3, m5
psrlw m5, m3, 4
pand m5, m1
pandn m4, m1, m7
por m5, m4
%ifidn %2, v
mova [tmpq+mstrideq*1], m5 ; p4
%else ; %2 != v
mova [rsp+10*16], m5
%endif ; %2==/!=v
; sub p6/p5, add p2/q2
psubw m3, m0
paddw m5, P2, Q2
psubw m3, m2
paddw m3, m5
psrlw m5, m3, 4
pand m5, m1
pandn m4, m1, m6
por m5, m4
%ifidn %2, v
mova [tmpq+strideq*0], m5 ; p3
%else ; %2 != v
mova [rsp+6*16], m5
%endif ; %2==/!=v
%define WRITE_IN_PLACE 0
%ifidn %2, v
%if ARCH_X86_64
%define WRITE_IN_PLACE 1
%endif
%endif
; sub p6/p4, add p1/q3
paddw m3, P1
paddw m5, m0, m7
paddw m3, Q3
psubw m3, m5
psrlw m5, m3, 4
pand m5, m1
pandn m4, m1, P2
por m5, m4
%if WRITE_IN_PLACE
mova [tmpq+strideq*1], m5
%else
mova [rsp+1*16], m5 ; don't clobber p2/m13
%endif
; sub p6/p3, add p0/q4
paddw m3, P0
paddw m5, m0, m6
%ifidn %2, v
paddw m3, [dstq+strideq*4]
%else ; %2 != v
paddw m3, [rsp+11*16]
%endif ; %2==/!=v
psubw m3, m5
psrlw m5, m3, 4
pand m5, m1
pandn m4, m1, P1
por m5, m4
%if WRITE_IN_PLACE
mova [dstq+mstrideq*2], m5
%else
mova [rsp+2*16], m5 ; don't clobber p1/m3
%endif
; sub p6/p2, add q0/q5
paddw m3, Q0
paddw m5, m0, P2
%ifidn %2, v
%if ARCH_X86_32
lea r4, P2
%endif
lea tmpq, [dstq+strideq*4]
paddw m3, [tmpq+strideq*1]
%else ; %2 != v
paddw m3, [rsp+12*16]
%endif ; %2==/!=v
psubw m3, m5
psrlw m5, m3, 4
pand m5, m1
pandn m4, m1, P0
por m5, m4
%if WRITE_IN_PLACE
mova [dstq+mstrideq*1], m5
%else
mova [rsp+3*16], m5 ; don't clobber p0/m4
%endif
; sub p6/p1, add q1/q6
paddw m3, Q1
paddw m5, m0, P1
%ifidn %2, v
mova m0, [tmpq+strideq*2] ; q6
%else ; %2 != v
mova m0, [rsp+13*16] ; q6
%endif ; %2==/!=v
paddw m3, m0
psubw m3, m5