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
|
;
; Copyright (c) 2016, 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 "third_party/x86inc/x86inc.asm"
SECTION_RODATA
pw_64: times 8 dw 64
even_byte_mask: times 8 dw 0x00ff
; %define USE_PMULHRSW
; NOTE: pmulhrsw has a latency of 5 cycles. Tests showed a performance loss
; when using this instruction.
;
; The add order below (based on ffav1) must be followed to prevent outranges.
; x = k0k1 + k4k5
; y = k2k3 + k6k7
; z = signed SAT(x + y)
SECTION .text
%define LOCAL_VARS_SIZE 16*6
%macro SETUP_LOCAL_VARS 0
; TODO(slavarnway): using xmm registers for these on AOM_ARCH_X86_64 +
; pmaddubsw has a higher latency on some platforms, this might be eased by
; interleaving the instructions.
%define k0k1 [rsp + 16*0]
%define k2k3 [rsp + 16*1]
%define k4k5 [rsp + 16*2]
%define k6k7 [rsp + 16*3]
packsswb m4, m4
; TODO(slavarnway): multiple pshufb instructions had a higher latency on
; some platforms.
pshuflw m0, m4, 0b ;k0_k1
pshuflw m1, m4, 01010101b ;k2_k3
pshuflw m2, m4, 10101010b ;k4_k5
pshuflw m3, m4, 11111111b ;k6_k7
punpcklqdq m0, m0
punpcklqdq m1, m1
punpcklqdq m2, m2
punpcklqdq m3, m3
mova k0k1, m0
mova k2k3, m1
mova k4k5, m2
mova k6k7, m3
%if AOM_ARCH_X86_64
%define krd m12
%define tmp0 [rsp + 16*4]
%define tmp1 [rsp + 16*5]
mova krd, [GLOBAL(pw_64)]
%else
%define krd [rsp + 16*4]
%if CONFIG_PIC=0
mova m6, [GLOBAL(pw_64)]
%else
; build constants without accessing global memory
pcmpeqb m6, m6 ;all ones
psrlw m6, 15
psllw m6, 6 ;aka pw_64
%endif
mova krd, m6
%endif
%endm
;-------------------------------------------------------------------------------
%if AOM_ARCH_X86_64
%define LOCAL_VARS_SIZE_H4 0
%else
%define LOCAL_VARS_SIZE_H4 16*4
%endif
%macro SUBPIX_HFILTER4 1
cglobal filter_block1d4_%1, 6, 6, 11, LOCAL_VARS_SIZE_H4, \
src, sstride, dst, dstride, height, filter
mova m4, [filterq]
packsswb m4, m4
%if AOM_ARCH_X86_64
%define k0k1k4k5 m8
%define k2k3k6k7 m9
%define krd m10
mova krd, [GLOBAL(pw_64)]
pshuflw k0k1k4k5, m4, 0b ;k0_k1
pshufhw k0k1k4k5, k0k1k4k5, 10101010b ;k0_k1_k4_k5
pshuflw k2k3k6k7, m4, 01010101b ;k2_k3
pshufhw k2k3k6k7, k2k3k6k7, 11111111b ;k2_k3_k6_k7
%else
%define k0k1k4k5 [rsp + 16*0]
%define k2k3k6k7 [rsp + 16*1]
%define krd [rsp + 16*2]
pshuflw m6, m4, 0b ;k0_k1
pshufhw m6, m6, 10101010b ;k0_k1_k4_k5
pshuflw m7, m4, 01010101b ;k2_k3
pshufhw m7, m7, 11111111b ;k2_k3_k6_k7
%if CONFIG_PIC=0
mova m1, [GLOBAL(pw_64)]
%else
; build constants without accessing global memory
pcmpeqb m1, m1 ;all ones
psrlw m1, 15
psllw m1, 6 ;aka pw_64
%endif
mova k0k1k4k5, m6
mova k2k3k6k7, m7
mova krd, m1
%endif
dec heightd
.loop:
;Do two rows at once
movu m4, [srcq - 3]
movu m5, [srcq + sstrideq - 3]
punpckhbw m1, m4, m4
punpcklbw m4, m4
punpckhbw m3, m5, m5
punpcklbw m5, m5
palignr m0, m1, m4, 1
pmaddubsw m0, k0k1k4k5
palignr m1, m4, 5
pmaddubsw m1, k2k3k6k7
palignr m2, m3, m5, 1
pmaddubsw m2, k0k1k4k5
palignr m3, m5, 5
pmaddubsw m3, k2k3k6k7
punpckhqdq m4, m0, m2
punpcklqdq m0, m2
punpckhqdq m5, m1, m3
punpcklqdq m1, m3
paddsw m0, m4
paddsw m1, m5
%ifidn %1, h8_avg
movd m4, [dstq]
movd m5, [dstq + dstrideq]
%endif
paddsw m0, m1
paddsw m0, krd
psraw m0, 7
%ifidn %1, h8_add_src
pxor m3, m3
movu m4, [srcq]
movu m5, [srcq + sstrideq]
punpckldq m4, m5 ; Bytes 0,1,2,3 from row 0, then 0,1,2,3 from row 2
punpcklbw m4, m3
paddsw m0, m4
%endif
packuswb m0, m0
psrldq m1, m0, 4
%ifidn %1, h8_avg
pavgb m0, m4
pavgb m1, m5
%endif
movd [dstq], m0
movd [dstq + dstrideq], m1
lea srcq, [srcq + sstrideq ]
prefetcht0 [srcq + 4 * sstrideq - 3]
lea srcq, [srcq + sstrideq ]
lea dstq, [dstq + 2 * dstrideq ]
prefetcht0 [srcq + 2 * sstrideq - 3]
sub heightd, 2
jg .loop
; Do last row if output_height is odd
jne .done
movu m4, [srcq - 3]
punpckhbw m1, m4, m4
punpcklbw m4, m4
palignr m0, m1, m4, 1
palignr m1, m4, 5
pmaddubsw m0, k0k1k4k5
pmaddubsw m1, k2k3k6k7
psrldq m2, m0, 8
psrldq m3, m1, 8
paddsw m0, m2
paddsw m1, m3
paddsw m0, m1
paddsw m0, krd
psraw m0, 7
%ifidn %1, h8_add_src
pxor m3, m3
movu m4, [srcq]
punpcklbw m4, m3
paddsw m0, m4
%endif
packuswb m0, m0
%ifidn %1, h8_avg
movd m4, [dstq]
pavgb m0, m4
%endif
movd [dstq], m0
.done:
REP_RET
%endm
;-------------------------------------------------------------------------------
%macro SUBPIX_HFILTER8 1
cglobal filter_block1d8_%1, 6, 6, 14, LOCAL_VARS_SIZE, \
src, sstride, dst, dstride, height, filter
mova m4, [filterq]
SETUP_LOCAL_VARS
dec heightd
.loop:
;Do two rows at once
movu m0, [srcq - 3]
movu m4, [srcq + sstrideq - 3]
punpckhbw m1, m0, m0
punpcklbw m0, m0
palignr m5, m1, m0, 13
pmaddubsw m5, k6k7
palignr m2, m1, m0, 5
palignr m3, m1, m0, 9
palignr m1, m0, 1
pmaddubsw m1, k0k1
punpckhbw m6, m4, m4
punpcklbw m4, m4
pmaddubsw m2, k2k3
pmaddubsw m3, k4k5
palignr m7, m6, m4, 13
palignr m0, m6, m4, 5
pmaddubsw m7, k6k7
paddsw m1, m3
paddsw m2, m5
paddsw m1, m2
%ifidn %1, h8_avg
movh m2, [dstq]
movhps m2, [dstq + dstrideq]
%endif
palignr m5, m6, m4, 9
palignr m6, m4, 1
pmaddubsw m0, k2k3
pmaddubsw m6, k0k1
paddsw m1, krd
pmaddubsw m5, k4k5
psraw m1, 7
paddsw m0, m7
paddsw m6, m5
paddsw m6, m0
paddsw m6, krd
psraw m6, 7
%ifidn %1, h8_add_src
pxor m3, m3
movu m4, [srcq]
movu m5, [srcq + sstrideq]
punpcklbw m4, m3
punpcklbw m5, m3
paddsw m1, m4
paddsw m6, m5
%endif
packuswb m1, m6
%ifidn %1, h8_avg
pavgb m1, m2
%endif
movh [dstq], m1
movhps [dstq + dstrideq], m1
lea srcq, [srcq + sstrideq ]
prefetcht0 [srcq + 4 * sstrideq - 3]
lea srcq, [srcq + sstrideq ]
lea dstq, [dstq + 2 * dstrideq ]
prefetcht0 [srcq + 2 * sstrideq - 3]
sub heightd, 2
jg .loop
; Do last row if output_height is odd
jne .done
movu m0, [srcq - 3]
punpckhbw m3, m0, m0
punpcklbw m0, m0
palignr m1, m3, m0, 1
palignr m2, m3, m0, 5
palignr m4, m3, m0, 13
palignr m3, m0, 9
pmaddubsw m1, k0k1
pmaddubsw m2, k2k3
pmaddubsw m3, k4k5
pmaddubsw m4, k6k7
paddsw m1, m3
paddsw m4, m2
paddsw m1, m4
paddsw m1, krd
psraw m1, 7
%ifidn %1, h8_add_src
pxor m6, m6
movu m5, [srcq]
punpcklbw m5, m6
paddsw m1, m5
%endif
packuswb m1, m1
%ifidn %1, h8_avg
movh m0, [dstq]
pavgb m1, m0
%endif
movh [dstq], m1
.done:
REP_RET
%endm
;-------------------------------------------------------------------------------
%macro SUBPIX_HFILTER16 1
cglobal filter_block1d16_%1, 6, 6, 14, LOCAL_VARS_SIZE, \
src, sstride, dst, dstride, height, filter
mova m4, [filterq]
SETUP_LOCAL_VARS
.loop:
prefetcht0 [srcq + 2 * sstrideq -3]
movu m0, [srcq - 3]
movu m4, [srcq - 2]
pmaddubsw m0, k0k1
pmaddubsw m4, k0k1
movu m1, [srcq - 1]
movu m5, [srcq + 0]
pmaddubsw m1, k2k3
pmaddubsw m5, k2k3
movu m2, [srcq + 1]
movu m6, [srcq + 2]
pmaddubsw m2, k4k5
pmaddubsw m6, k4k5
movu m3, [srcq + 3]
movu m7, [srcq + 4]
pmaddubsw m3, k6k7
pmaddubsw m7, k6k7
paddsw m0, m2
paddsw m1, m3
paddsw m0, m1
paddsw m4, m6
paddsw m5, m7
paddsw m4, m5
paddsw m0, krd
paddsw m4, krd
psraw m0, 7
psraw m4, 7
%ifidn %1, h8_add_src
%if AOM_ARCH_X86=1 && CONFIG_PIC=1
pcmpeqb m2, m2 ;all ones
psrlw m2, 8 ;even_byte_mask
%else
mova m2, [GLOBAL(even_byte_mask)]
%endif
movu m5, [srcq]
mova m7, m5
pand m5, m2
psrlw m7, 8
paddsw m0, m5
paddsw m4, m7
%endif
packuswb m0, m0
packuswb m4, m4
punpcklbw m0, m4
%ifidn %1, h8_avg
pavgb m0, [dstq]
%endif
lea srcq, [srcq + sstrideq]
mova [dstq], m0
lea dstq, [dstq + dstrideq]
dec heightd
jnz .loop
REP_RET
%endm
INIT_XMM ssse3
SUBPIX_HFILTER16 h8
SUBPIX_HFILTER8 h8
SUBPIX_HFILTER4 h8
;-------------------------------------------------------------------------------
; TODO(Linfeng): Detect cpu type and choose the code with better performance.
%define X86_SUBPIX_VFILTER_PREFER_SLOW_CELERON 1
%if AOM_ARCH_X86_64 && X86_SUBPIX_VFILTER_PREFER_SLOW_CELERON
%define NUM_GENERAL_REG_USED 9
%else
%define NUM_GENERAL_REG_USED 6
%endif
%macro SUBPIX_VFILTER 2
cglobal filter_block1d%2_%1, 6, NUM_GENERAL_REG_USED, 15, LOCAL_VARS_SIZE, \
src, sstride, dst, dstride, height, filter
mova m4, [filterq]
SETUP_LOCAL_VARS
%ifidn %2, 8
%define movx movh
%else
%define movx movd
%endif
dec heightd
%if AOM_ARCH_X86 || X86_SUBPIX_VFILTER_PREFER_SLOW_CELERON
%if AOM_ARCH_X86_64
%define src1q r7
%define sstride6q r8
%define dst_stride dstrideq
%else
%define src1q filterq
%define sstride6q dstrideq
%define dst_stride dstridemp
%endif
mov src1q, srcq
add src1q, sstrideq
lea sstride6q, [sstrideq + sstrideq * 4]
add sstride6q, sstrideq ;pitch * 6
.loop:
;Do two rows at once
movx m0, [srcq ] ;A
movx m1, [src1q ] ;B
punpcklbw m0, m1 ;A B
movx m2, [srcq + sstrideq * 2 ] ;C
pmaddubsw m0, k0k1
mova m6, m2
movx m3, [src1q + sstrideq * 2] ;D
punpcklbw m2, m3 ;C D
pmaddubsw m2, k2k3
movx m4, [srcq + sstrideq * 4 ] ;E
mova m7, m4
movx m5, [src1q + sstrideq * 4] ;F
punpcklbw m4, m5 ;E F
pmaddubsw m4, k4k5
punpcklbw m1, m6 ;A B next iter
movx m6, [srcq + sstride6q ] ;G
punpcklbw m5, m6 ;E F next iter
punpcklbw m3, m7 ;C D next iter
pmaddubsw m5, k4k5
movx m7, [src1q + sstride6q ] ;H
punpcklbw m6, m7 ;G H
pmaddubsw m6, k6k7
pmaddubsw m3, k2k3
pmaddubsw m1, k0k1
paddsw m0, m4
paddsw m2, m6
movx m6, [srcq + sstrideq * 8 ] ;H next iter
punpcklbw m7, m6
pmaddubsw m7, k6k7
paddsw m0, m2
paddsw m0, krd
psraw m0, 7
paddsw m1, m5
%ifidn %1, v8_add_src
pxor m6, m6
movu m4, [srcq]
punpcklbw m4, m6
paddsw m0, m4
%endif
packuswb m0, m0
paddsw m3, m7
paddsw m1, m3
paddsw m1, krd
psraw m1, 7
%ifidn %1, v8_add_src
movu m4, [src1q]
punpcklbw m4, m6
paddsw m1, m4
%endif
lea srcq, [srcq + sstrideq * 2 ]
lea src1q, [src1q + sstrideq * 2]
packuswb m1, m1
%ifidn %1, v8_avg
movx m2, [dstq]
pavgb m0, m2
%endif
movx [dstq], m0
add dstq, dst_stride
%ifidn %1, v8_avg
movx m3, [dstq]
pavgb m1, m3
%endif
movx [dstq], m1
add dstq, dst_stride
sub heightd, 2
jg .loop
; Do last row if output_height is odd
jne .done
movx m0, [srcq ] ;A
movx m1, [srcq + sstrideq ] ;B
movx m6, [srcq + sstride6q ] ;G
punpcklbw m0, m1 ;A B
movx m7, [src1q + sstride6q ] ;H
pmaddubsw m0, k0k1
movx m2, [srcq + sstrideq * 2 ] ;C
punpcklbw m6, m7 ;G H
movx m3, [src1q + sstrideq * 2] ;D
pmaddubsw m6, k6k7
movx m4, [srcq + sstrideq * 4 ] ;E
punpcklbw m2, m3 ;C D
movx m5, [src1q + sstrideq * 4] ;F
punpcklbw m4, m5 ;E F
pmaddubsw m2, k2k3
pmaddubsw m4, k4k5
paddsw m2, m6
paddsw m0, m4
paddsw m0, m2
paddsw m0, krd
psraw m0, 7
%ifidn %1, v8_add_src
pxor m6, m6
movu m4, [srcq]
punpcklbw m4, m6
paddsw m0, m4
%endif
packuswb m0, m0
%ifidn %1, v8_avg
movx m1, [dstq]
pavgb m0, m1
%endif
movx [dstq], m0
%else
; AOM_ARCH_X86_64
movx m0, [srcq ] ;A
movx m1, [srcq + sstrideq ] ;B
lea srcq, [srcq + sstrideq * 2 ]
movx m2, [srcq] ;C
movx m3, [srcq + sstrideq] ;D
lea srcq, [srcq + sstrideq * 2 ]
movx m4, [srcq] ;E
movx m5, [srcq + sstrideq] ;F
lea srcq, [srcq + sstrideq * 2 ]
movx m6, [srcq] ;G
punpcklbw m0, m1 ;A B
punpcklbw m1, m2 ;A B next iter
punpcklbw m2, m3 ;C D
punpcklbw m3, m4 ;C D next iter
punpcklbw m4, m5 ;E F
punpcklbw m5, m6 ;E F next iter
.loop:
;Do two rows at once
movx m7, [srcq + sstrideq] ;H
lea srcq, [srcq + sstrideq * 2 ]
movx m14, [srcq] ;H next iter
punpcklbw m6, m7 ;G H
punpcklbw m7, m14 ;G H next iter
pmaddubsw m8, m0, k0k1
pmaddubsw m9, m1, k0k1
mova m0, m2
mova m1, m3
pmaddubsw m10, m2, k2k3
pmaddubsw m11, m3, k2k3
mova m2, m4
mova m3, m5
pmaddubsw m4, k4k5
pmaddubsw m5, k4k5
paddsw m8, m4
paddsw m9, m5
mova m4, m6
mova m5, m7
pmaddubsw m6, k6k7
pmaddubsw m7, k6k7
paddsw m10, m6
paddsw m11, m7
paddsw m8, m10
paddsw m9, m11
mova m6, m14
paddsw m8, krd
paddsw m9, krd
psraw m8, 7
psraw m9, 7
%ifidn %2, 4
packuswb m8, m8
packuswb m9, m9
%else
packuswb m8, m9
%endif
%ifidn %1, v8_avg
movx m7, [dstq]
%ifidn %2, 4
movx m10, [dstq + dstrideq]
pavgb m9, m10
%else
movhpd m7, [dstq + dstrideq]
%endif
pavgb m8, m7
%endif
movx [dstq], m8
%ifidn %2, 4
movx [dstq + dstrideq], m9
%else
movhpd [dstq + dstrideq], m8
%endif
lea dstq, [dstq + dstrideq * 2 ]
sub heightd, 2
jg .loop
; Do last row if output_height is odd
jne .done
movx m7, [srcq + sstrideq] ;H
punpcklbw m6, m7 ;G H
pmaddubsw m0, k0k1
pmaddubsw m2, k2k3
pmaddubsw m4, k4k5
pmaddubsw m6, k6k7
paddsw m0, m4
paddsw m2, m6
paddsw m0, m2
paddsw m0, krd
psraw m0, 7
packuswb m0, m0
%ifidn %1, v8_avg
movx m1, [dstq]
pavgb m0, m1
%endif
movx [dstq], m0
%endif ; AOM_ARCH_X86_64
.done:
REP_RET
%endm
;-------------------------------------------------------------------------------
%macro SUBPIX_VFILTER16 1
cglobal filter_block1d16_%1, 6, NUM_GENERAL_REG_USED, 16, LOCAL_VARS_SIZE, \
src, sstride, dst, dstride, height, filter
mova m4, [filterq]
SETUP_LOCAL_VARS
%if AOM_ARCH_X86 || X86_SUBPIX_VFILTER_PREFER_SLOW_CELERON
%if AOM_ARCH_X86_64
%define src1q r7
%define sstride6q r8
%define dst_stride dstrideq
%else
%define src1q filterq
%define sstride6q dstrideq
%define dst_stride dstridemp
%endif
lea src1q, [srcq + sstrideq]
lea sstride6q, [sstrideq + sstrideq * 4]
add sstride6q, sstrideq ;pitch * 6
.loop:
movh m0, [srcq ] ;A
movh m1, [src1q ] ;B
movh m2, [srcq + sstrideq * 2 ] ;C
movh m3, [src1q + sstrideq * 2] ;D
movh m4, [srcq + sstrideq * 4 ] ;E
movh m5, [src1q + sstrideq * 4] ;F
punpcklbw m0, m1 ;A B
movh m6, [srcq + sstride6q] ;G
punpcklbw m2, m3 ;C D
movh m7, [src1q + sstride6q] ;H
punpcklbw m4, m5 ;E F
pmaddubsw m0, k0k1
movh m3, [srcq + 8] ;A
pmaddubsw m2, k2k3
punpcklbw m6, m7 ;G H
movh m5, [srcq + sstrideq + 8] ;B
pmaddubsw m4, k4k5
punpcklbw m3, m5 ;A B
movh m7, [srcq + sstrideq * 2 + 8] ;C
pmaddubsw m6, k6k7
movh m5, [src1q + sstrideq * 2 + 8] ;D
punpcklbw m7, m5 ;C D
paddsw m2, m6
pmaddubsw m3, k0k1
movh m1, [srcq + sstrideq * 4 + 8] ;E
paddsw m0, m4
pmaddubsw m7, k2k3
movh m6, [src1q + sstrideq * 4 + 8] ;F
punpcklbw m1, m6 ;E F
paddsw m0, m2
paddsw m0, krd
movh m2, [srcq + sstride6q + 8] ;G
pmaddubsw m1, k4k5
movh m5, [src1q + sstride6q + 8] ;H
psraw m0, 7
punpcklbw m2, m5 ;G H
pmaddubsw m2, k6k7
paddsw m7, m2
paddsw m3, m1
paddsw m3, m7
paddsw m3, krd
psraw m3, 7
%ifidn %1, v8_add_src
pxor m6, m6
movu m4, [src1q + 2 * sstrideq] ; Fetch from 3 rows down
mova m5, m4
punpcklbw m4, m6
punpckhbw m5, m6
paddsw m0, m4
paddsw m3, m5
%endif
packuswb m0, m3
add srcq, sstrideq
add src1q, sstrideq
%ifidn %1, v8_avg
pavgb m0, [dstq]
%endif
mova [dstq], m0
add dstq, dst_stride
dec heightd
jnz .loop
REP_RET
%else
; AOM_ARCH_X86_64
dec heightd
movu m1, [srcq ] ;A
movu m3, [srcq + sstrideq ] ;B
lea srcq, [srcq + sstrideq * 2]
punpcklbw m0, m1, m3 ;A B
punpckhbw m1, m3 ;A B
movu m5, [srcq] ;C
punpcklbw m2, m3, m5 ;A B next iter
punpckhbw m3, m5 ;A B next iter
mova tmp0, m2 ;store to stack
mova tmp1, m3 ;store to stack
movu m7, [srcq + sstrideq] ;D
lea srcq, [srcq + sstrideq * 2]
punpcklbw m4, m5, m7 ;C D
punpckhbw m5, m7 ;C D
movu m9, [srcq] ;E
punpcklbw m6, m7, m9 ;C D next iter
punpckhbw m7, m9 ;C D next iter
movu m11, [srcq + sstrideq] ;F
lea srcq, [srcq + sstrideq * 2]
punpcklbw m8, m9, m11 ;E F
punpckhbw m9, m11 ;E F
movu m2, [srcq] ;G
punpcklbw m10, m11, m2 ;E F next iter
punpckhbw m11, m2 ;E F next iter
.loop:
;Do two rows at once
pmaddubsw m13, m0, k0k1
mova m0, m4
pmaddubsw m14, m8, k4k5
pmaddubsw m15, m4, k2k3
mova m4, m8
paddsw m13, m14
movu m3, [srcq + sstrideq] ;H
lea srcq, [srcq + sstrideq * 2]
punpcklbw m14, m2, m3 ;G H
mova m8, m14
pmaddubsw m14, k6k7
paddsw m15, m14
paddsw m13, m15
paddsw m13, krd
psraw m13, 7
pmaddubsw m14, m1, k0k1
pmaddubsw m1, m9, k4k5
pmaddubsw m15, m5, k2k3
paddsw m14, m1
mova m1, m5
mova m5, m9
punpckhbw m2, m3 ;G H
mova m9, m2
pmaddubsw m2, k6k7
paddsw m15, m2
paddsw m14, m15
paddsw m14, krd
psraw m14, 7
packuswb m13, m14
%ifidn %1, v8_avg
pavgb m13, [dstq]
%endif
mova [dstq], m13
; next iter
pmaddubsw m15, tmp0, k0k1
pmaddubsw m14, m10, k4k5
pmaddubsw m13, m6, k2k3
paddsw m15, m14
mova tmp0, m6
mova m6, m10
movu m2, [srcq] ;G next iter
punpcklbw m14, m3, m2 ;G H next iter
mova m10, m14
pmaddubsw m14, k6k7
paddsw m13, m14
paddsw m15, m13
paddsw m15, krd
psraw m15, 7
pmaddubsw m14, tmp1, k0k1
mova tmp1, m7
pmaddubsw m13, m7, k2k3
mova m7, m11
pmaddubsw m11, k4k5
paddsw m14, m11
punpckhbw m3, m2 ;G H next iter
mova m11, m3
pmaddubsw m3, k6k7
paddsw m13, m3
paddsw m14, m13
paddsw m14, krd
psraw m14, 7
packuswb m15, m14
%ifidn %1, v8_avg
pavgb m15, [dstq + dstrideq]
%endif
mova [dstq + dstrideq], m15
lea dstq, [dstq + dstrideq * 2]
sub heightd, 2
jg .loop
; Do last row if output_height is odd
jne .done
movu m3, [srcq + sstrideq] ;H
punpcklbw m6, m2, m3 ;G H
punpckhbw m2, m3 ;G H
pmaddubsw m0, k0k1
pmaddubsw m1, k0k1
pmaddubsw m4, k2k3
pmaddubsw m5, k2k3
pmaddubsw m8, k4k5
pmaddubsw m9, k4k5
pmaddubsw m6, k6k7
pmaddubsw m2, k6k7
paddsw m0, m8
paddsw m1, m9
paddsw m4, m6
paddsw m5, m2
paddsw m0, m4
paddsw m1, m5
paddsw m0, krd
paddsw m1, krd
psraw m0, 7
psraw m1, 7
packuswb m0, m1
%ifidn %1, v8_avg
pavgb m0, [dstq]
%endif
mova [dstq], m0
.done:
REP_RET
%endif ; AOM_ARCH_X86_64
%endm
INIT_XMM ssse3
SUBPIX_VFILTER16 v8
SUBPIX_VFILTER v8, 8
SUBPIX_VFILTER v8, 4
|