summaryrefslogtreecommitdiff
path: root/libavcodec/x86/dsputil.asm
blob: b9990bb1bacc31c83cd4dbac317076c1e50a939b (plain)
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
;******************************************************************************
;* MMX optimized DSP utils
;* Copyright (c) 2008 Loren Merritt
;*
;* This file is part of Libav.
;*
;* Libav is free software; you can redistribute it and/or
;* modify it under the terms of the GNU Lesser General Public
;* License as published by the Free Software Foundation; either
;* version 2.1 of the License, or (at your option) any later version.
;*
;* Libav is distributed in the hope that it will be useful,
;* but WITHOUT ANY WARRANTY; without even the implied warranty of
;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
;* Lesser General Public License for more details.
;*
;* You should have received a copy of the GNU Lesser General Public
;* License along with Libav; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************

%include "libavutil/x86/x86util.asm"

SECTION_RODATA
pb_f: times 16 db 15
pb_zzzzzzzz77777777: times 8 db -1
pb_7: times 8 db 7
pb_zzzz3333zzzzbbbb: db -1,-1,-1,-1,3,3,3,3,-1,-1,-1,-1,11,11,11,11
pb_zz11zz55zz99zzdd: db -1,-1,1,1,-1,-1,5,5,-1,-1,9,9,-1,-1,13,13
pb_revwords: SHUFFLE_MASK_W 7, 6, 5, 4, 3, 2, 1, 0
pd_16384: times 4 dd 16384
pb_bswap32: db 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12

SECTION_TEXT

%macro SCALARPRODUCT 0
; int scalarproduct_int16(int16_t *v1, int16_t *v2, int order)
cglobal scalarproduct_int16, 3,3,3, v1, v2, order
    shl orderq, 1
    add v1q, orderq
    add v2q, orderq
    neg orderq
    pxor    m2, m2
.loop:
    movu    m0, [v1q + orderq]
    movu    m1, [v1q + orderq + mmsize]
    pmaddwd m0, [v2q + orderq]
    pmaddwd m1, [v2q + orderq + mmsize]
    paddd   m2, m0
    paddd   m2, m1
    add     orderq, mmsize*2
    jl .loop
%if mmsize == 16
    movhlps m0, m2
    paddd   m2, m0
    pshuflw m0, m2, 0x4e
%else
    pshufw  m0, m2, 0x4e
%endif
    paddd   m2, m0
    movd   eax, m2
    RET

; int scalarproduct_and_madd_int16(int16_t *v1, int16_t *v2, int16_t *v3, int order, int mul)
cglobal scalarproduct_and_madd_int16, 4,4,8, v1, v2, v3, order, mul
    shl orderq, 1
    movd    m7, mulm
%if mmsize == 16
    pshuflw m7, m7, 0
    punpcklqdq m7, m7
%else
    pshufw  m7, m7, 0
%endif
    pxor    m6, m6
    add v1q, orderq
    add v2q, orderq
    add v3q, orderq
    neg orderq
.loop:
    movu    m0, [v2q + orderq]
    movu    m1, [v2q + orderq + mmsize]
    mova    m4, [v1q + orderq]
    mova    m5, [v1q + orderq + mmsize]
    movu    m2, [v3q + orderq]
    movu    m3, [v3q + orderq + mmsize]
    pmaddwd m0, m4
    pmaddwd m1, m5
    pmullw  m2, m7
    pmullw  m3, m7
    paddd   m6, m0
    paddd   m6, m1
    paddw   m2, m4
    paddw   m3, m5
    mova    [v1q + orderq], m2
    mova    [v1q + orderq + mmsize], m3
    add     orderq, mmsize*2
    jl .loop
%if mmsize == 16
    movhlps m0, m6
    paddd   m6, m0
    pshuflw m0, m6, 0x4e
%else
    pshufw  m0, m6, 0x4e
%endif
    paddd   m6, m0
    movd   eax, m6
    RET
%endmacro

INIT_MMX mmxext
SCALARPRODUCT
INIT_XMM sse2
SCALARPRODUCT

%macro SCALARPRODUCT_LOOP 1
align 16
.loop%1:
    sub     orderq, mmsize*2
%if %1
    mova    m1, m4
    mova    m4, [v2q + orderq]
    mova    m0, [v2q + orderq + mmsize]
    palignr m1, m0, %1
    palignr m0, m4, %1
    mova    m3, m5
    mova    m5, [v3q + orderq]
    mova    m2, [v3q + orderq + mmsize]
    palignr m3, m2, %1
    palignr m2, m5, %1
%else
    mova    m0, [v2q + orderq]
    mova    m1, [v2q + orderq + mmsize]
    mova    m2, [v3q + orderq]
    mova    m3, [v3q + orderq + mmsize]
%endif
    %define t0  [v1q + orderq]
    %define t1  [v1q + orderq + mmsize]
%if ARCH_X86_64
    mova    m8, t0
    mova    m9, t1
    %define t0  m8
    %define t1  m9
%endif
    pmaddwd m0, t0
    pmaddwd m1, t1
    pmullw  m2, m7
    pmullw  m3, m7
    paddw   m2, t0
    paddw   m3, t1
    paddd   m6, m0
    paddd   m6, m1
    mova    [v1q + orderq], m2
    mova    [v1q + orderq + mmsize], m3
    jg .loop%1
%if %1
    jmp .end
%endif
%endmacro

; int scalarproduct_and_madd_int16(int16_t *v1, int16_t *v2, int16_t *v3, int order, int mul)
INIT_XMM ssse3
cglobal scalarproduct_and_madd_int16, 4,5,10, v1, v2, v3, order, mul
    shl orderq, 1
    movd    m7, mulm
    pshuflw m7, m7, 0
    punpcklqdq m7, m7
    pxor    m6, m6
    mov    r4d, v2d
    and    r4d, 15
    and    v2q, ~15
    and    v3q, ~15
    mova    m4, [v2q + orderq]
    mova    m5, [v3q + orderq]
    ; linear is faster than branch tree or jump table, because the branches taken are cyclic (i.e. predictable)
    cmp    r4d, 0
    je .loop0
    cmp    r4d, 2
    je .loop2
    cmp    r4d, 4
    je .loop4
    cmp    r4d, 6
    je .loop6
    cmp    r4d, 8
    je .loop8
    cmp    r4d, 10
    je .loop10
    cmp    r4d, 12
    je .loop12
SCALARPRODUCT_LOOP 14
SCALARPRODUCT_LOOP 12
SCALARPRODUCT_LOOP 10
SCALARPRODUCT_LOOP 8
SCALARPRODUCT_LOOP 6
SCALARPRODUCT_LOOP 4
SCALARPRODUCT_LOOP 2
SCALARPRODUCT_LOOP 0
.end:
    movhlps m0, m6
    paddd   m6, m0
    pshuflw m0, m6, 0x4e
    paddd   m6, m0
    movd   eax, m6
    RET


;-----------------------------------------------------------------------------
; void ff_apply_window_int16(int16_t *output, const int16_t *input,
;                            const int16_t *window, unsigned int len)
;-----------------------------------------------------------------------------

%macro REVERSE_WORDS 1-2
%if cpuflag(ssse3) && notcpuflag(atom)
    pshufb  %1, %2
%elif cpuflag(sse2)
    pshuflw  %1, %1, 0x1B
    pshufhw  %1, %1, 0x1B
    pshufd   %1, %1, 0x4E
%elif cpuflag(mmxext)
    pshufw   %1, %1, 0x1B
%endif
%endmacro

%macro MUL16FIXED 3
%if cpuflag(ssse3) ; dst, src, unused
; dst = ((dst * src) + (1<<14)) >> 15
    pmulhrsw   %1, %2
%elif cpuflag(mmxext) ; dst, src, temp
; dst = (dst * src) >> 15
; pmulhw cuts off the bottom bit, so we have to lshift by 1 and add it back
; in from the pmullw result.
    mova    %3, %1
    pmulhw  %1, %2
    pmullw  %3, %2
    psrlw   %3, 15
    psllw   %1, 1
    por     %1, %3
%endif
%endmacro

%macro APPLY_WINDOW_INT16 1 ; %1 bitexact version
%if %1
cglobal apply_window_int16, 4,5,6, output, input, window, offset, offset2
%else
cglobal apply_window_int16_round, 4,5,6, output, input, window, offset, offset2
%endif
    lea     offset2q, [offsetq-mmsize]
%if cpuflag(ssse3) && notcpuflag(atom)
    mova          m5, [pb_revwords]
    ALIGN 16
%elif %1
    mova          m5, [pd_16384]
%endif
.loop:
%if cpuflag(ssse3)
    ; This version does the 16x16->16 multiplication in-place without expanding
    ; to 32-bit. The ssse3 version is bit-identical.
    mova          m0, [windowq+offset2q]
    mova          m1, [ inputq+offset2q]
    pmulhrsw      m1, m0
    REVERSE_WORDS m0, m5
    pmulhrsw      m0, [ inputq+offsetq ]
    mova  [outputq+offset2q], m1
    mova  [outputq+offsetq ], m0
%elif %1
    ; This version expands 16-bit to 32-bit, multiplies by the window,
    ; adds 16384 for rounding, right shifts 15, then repacks back to words to
    ; save to the output. The window is reversed for the second half.
    mova          m3, [windowq+offset2q]
    mova          m4, [ inputq+offset2q]
    pxor          m0, m0
    punpcklwd     m0, m3
    punpcklwd     m1, m4
    pmaddwd       m0, m1
    paddd         m0, m5
    psrad         m0, 15
    pxor          m2, m2
    punpckhwd     m2, m3
    punpckhwd     m1, m4
    pmaddwd       m2, m1
    paddd         m2, m5
    psrad         m2, 15
    packssdw      m0, m2
    mova  [outputq+offset2q], m0
    REVERSE_WORDS m3
    mova          m4, [ inputq+offsetq]
    pxor          m0, m0
    punpcklwd     m0, m3
    punpcklwd     m1, m4
    pmaddwd       m0, m1
    paddd         m0, m5
    psrad         m0, 15
    pxor          m2, m2
    punpckhwd     m2, m3
    punpckhwd     m1, m4
    pmaddwd       m2, m1
    paddd         m2, m5
    psrad         m2, 15
    packssdw      m0, m2
    mova  [outputq+offsetq], m0
%else
    ; This version does the 16x16->16 multiplication in-place without expanding
    ; to 32-bit. The mmxext and sse2 versions do not use rounding, and
    ; therefore are not bit-identical to the C version.
    mova          m0, [windowq+offset2q]
    mova          m1, [ inputq+offset2q]
    mova          m2, [ inputq+offsetq ]
    MUL16FIXED    m1, m0, m3
    REVERSE_WORDS m0
    MUL16FIXED    m2, m0, m3
    mova  [outputq+offset2q], m1
    mova  [outputq+offsetq ], m2
%endif
    add      offsetd, mmsize
    sub     offset2d, mmsize
    jae .loop
    REP_RET
%endmacro

INIT_MMX mmxext
APPLY_WINDOW_INT16 0
INIT_XMM sse2
APPLY_WINDOW_INT16 0

INIT_MMX mmxext
APPLY_WINDOW_INT16 1
INIT_XMM sse2
APPLY_WINDOW_INT16 1
INIT_XMM ssse3
APPLY_WINDOW_INT16 1
INIT_XMM ssse3, atom
APPLY_WINDOW_INT16 1


; void add_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top)
INIT_MMX mmxext
cglobal add_hfyu_median_prediction, 6,6,0, dst, top, diff, w, left, left_top
    movq    mm0, [topq]
    movq    mm2, mm0
    movd    mm4, [left_topq]
    psllq   mm2, 8
    movq    mm1, mm0
    por     mm4, mm2
    movd    mm3, [leftq]
    psubb   mm0, mm4 ; t-tl
    add    dstq, wq
    add    topq, wq
    add   diffq, wq
    neg      wq
    jmp .skip
.loop:
    movq    mm4, [topq+wq]
    movq    mm0, mm4
    psllq   mm4, 8
    por     mm4, mm1
    movq    mm1, mm0 ; t
    psubb   mm0, mm4 ; t-tl
.skip:
    movq    mm2, [diffq+wq]
%assign i 0
%rep 8
    movq    mm4, mm0
    paddb   mm4, mm3 ; t-tl+l
    movq    mm5, mm3
    pmaxub  mm3, mm1
    pminub  mm5, mm1
    pminub  mm3, mm4
    pmaxub  mm3, mm5 ; median
    paddb   mm3, mm2 ; +residual
%if i==0
    movq    mm7, mm3
    psllq   mm7, 56
%else
    movq    mm6, mm3
    psrlq   mm7, 8
    psllq   mm6, 56
    por     mm7, mm6
%endif
%if i<7
    psrlq   mm0, 8
    psrlq   mm1, 8
    psrlq   mm2, 8
%endif
%assign i i+1
%endrep
    movq [dstq+wq], mm7
    add      wq, 8
    jl .loop
    movzx   r2d, byte [dstq-1]
    mov [leftq], r2d
    movzx   r2d, byte [topq-1]
    mov [left_topq], r2d
    RET


%macro ADD_HFYU_LEFT_LOOP 2 ; %1 = dst_is_aligned, %2 = src_is_aligned
    add     srcq, wq
    add     dstq, wq
    neg     wq
%%.loop:
%if %2
    mova    m1, [srcq+wq]
%else
    movu    m1, [srcq+wq]
%endif
    mova    m2, m1
    psllw   m1, 8
    paddb   m1, m2
    mova    m2, m1
    pshufb  m1, m3
    paddb   m1, m2
    pshufb  m0, m5
    mova    m2, m1
    pshufb  m1, m4
    paddb   m1, m2
%if mmsize == 16
    mova    m2, m1
    pshufb  m1, m6
    paddb   m1, m2
%endif
    paddb   m0, m1
%if %1
    mova    [dstq+wq], m0
%else
    movq    [dstq+wq], m0
    movhps  [dstq+wq+8], m0
%endif
    add     wq, mmsize
    jl %%.loop
    mov     eax, mmsize-1
    sub     eax, wd
    movd    m1, eax
    pshufb  m0, m1
    movd    eax, m0
    RET
%endmacro

; int add_hfyu_left_prediction(uint8_t *dst, const uint8_t *src, int w, int left)
INIT_MMX ssse3
cglobal add_hfyu_left_prediction, 3,3,7, dst, src, w, left
.skip_prologue:
    mova    m5, [pb_7]
    mova    m4, [pb_zzzz3333zzzzbbbb]
    mova    m3, [pb_zz11zz55zz99zzdd]
    movd    m0, leftm
    psllq   m0, 56
    ADD_HFYU_LEFT_LOOP 1, 1

INIT_XMM sse4
cglobal add_hfyu_left_prediction, 3,3,7, dst, src, w, left
    mova    m5, [pb_f]
    mova    m6, [pb_zzzzzzzz77777777]
    mova    m4, [pb_zzzz3333zzzzbbbb]
    mova    m3, [pb_zz11zz55zz99zzdd]
    movd    m0, leftm
    pslldq  m0, 15
    test    srcq, 15
    jnz .src_unaligned
    test    dstq, 15
    jnz .dst_unaligned
    ADD_HFYU_LEFT_LOOP 1, 1
.dst_unaligned:
    ADD_HFYU_LEFT_LOOP 0, 1
.src_unaligned:
    ADD_HFYU_LEFT_LOOP 0, 0


; float scalarproduct_float_sse(const float *v1, const float *v2, int len)
INIT_XMM sse
cglobal scalarproduct_float, 3,3,2, v1, v2, offset
    neg offsetq
    shl offsetq, 2
    sub v1q, offsetq
    sub v2q, offsetq
    xorps xmm0, xmm0
    .loop:
        movaps   xmm1, [v1q+offsetq]
        mulps    xmm1, [v2q+offsetq]
        addps    xmm0, xmm1
        add      offsetq, 16
        js       .loop
    movhlps xmm1, xmm0
    addps   xmm0, xmm1
    movss   xmm1, xmm0
    shufps  xmm0, xmm0, 1
    addss   xmm0, xmm1
%if ARCH_X86_64 == 0
    movss   r0m,  xmm0
    fld     dword r0m
%endif
    RET

;-----------------------------------------------------------------------------
; void ff_vector_clip_int32(int32_t *dst, const int32_t *src, int32_t min,
;                           int32_t max, unsigned int len)
;-----------------------------------------------------------------------------

; %1 = number of xmm registers used
; %2 = number of inline load/process/store loops per asm loop
; %3 = process 4*mmsize (%3=0) or 8*mmsize (%3=1) bytes per loop
; %4 = CLIPD function takes min/max as float instead of int (CLIPD_SSE2)
; %5 = suffix
%macro VECTOR_CLIP_INT32 4-5
cglobal vector_clip_int32%5, 5,5,%1, dst, src, min, max, len
%if %4
    cvtsi2ss  m4, minm
    cvtsi2ss  m5, maxm
%else
    movd      m4, minm
    movd      m5, maxm
%endif
    SPLATD    m4
    SPLATD    m5
.loop:
%assign %%i 1
%rep %2
    mova      m0,  [srcq+mmsize*0*%%i]
    mova      m1,  [srcq+mmsize*1*%%i]
    mova      m2,  [srcq+mmsize*2*%%i]
    mova      m3,  [srcq+mmsize*3*%%i]
%if %3
    mova      m7,  [srcq+mmsize*4*%%i]
    mova      m8,  [srcq+mmsize*5*%%i]
    mova      m9,  [srcq+mmsize*6*%%i]
    mova      m10, [srcq+mmsize*7*%%i]
%endif
    CLIPD  m0,  m4, m5, m6
    CLIPD  m1,  m4, m5, m6
    CLIPD  m2,  m4, m5, m6
    CLIPD  m3,  m4, m5, m6
%if %3
    CLIPD  m7,  m4, m5, m6
    CLIPD  m8,  m4, m5, m6
    CLIPD  m9,  m4, m5, m6
    CLIPD  m10, m4, m5, m6
%endif
    mova  [dstq+mmsize*0*%%i], m0
    mova  [dstq+mmsize*1*%%i], m1
    mova  [dstq+mmsize*2*%%i], m2
    mova  [dstq+mmsize*3*%%i], m3
%if %3
    mova  [dstq+mmsize*4*%%i], m7
    mova  [dstq+mmsize*5*%%i], m8
    mova  [dstq+mmsize*6*%%i], m9
    mova  [dstq+mmsize*7*%%i], m10
%endif
%assign %%i %%i+1
%endrep
    add     srcq, mmsize*4*(%2+%3)
    add     dstq, mmsize*4*(%2+%3)
    sub     lend, mmsize*(%2+%3)
    jg .loop
    REP_RET
%endmacro

INIT_MMX mmx
%define CLIPD CLIPD_MMX
VECTOR_CLIP_INT32 0, 1, 0, 0
INIT_XMM sse2
VECTOR_CLIP_INT32 6, 1, 0, 0, _int
%define CLIPD CLIPD_SSE2
VECTOR_CLIP_INT32 6, 2, 0, 1
INIT_XMM sse4
%define CLIPD CLIPD_SSE41
%ifdef m8
VECTOR_CLIP_INT32 11, 1, 1, 0
%else
VECTOR_CLIP_INT32 6, 1, 0, 0
%endif

;-----------------------------------------------------------------------------
; void vector_fmul_reverse(float *dst, const float *src0, const float *src1,
;                          int len)
;-----------------------------------------------------------------------------
%macro VECTOR_FMUL_REVERSE 0
cglobal vector_fmul_reverse, 4,4,2, dst, src0, src1, len
    lea       lenq, [lend*4 - 2*mmsize]
ALIGN 16
.loop:
%if cpuflag(avx)
    vmovaps     xmm0, [src1q + 16]
    vinsertf128 m0, m0, [src1q], 1
    vshufps     m0, m0, m0, q0123
    vmovaps     xmm1, [src1q + mmsize + 16]
    vinsertf128 m1, m1, [src1q + mmsize], 1
    vshufps     m1, m1, m1, q0123
%else
    mova    m0, [src1q]
    mova    m1, [src1q + mmsize]
    shufps  m0, m0, q0123
    shufps  m1, m1, q0123
%endif
    mulps   m0, m0, [src0q + lenq + mmsize]
    mulps   m1, m1, [src0q + lenq]
    mova    [dstq + lenq + mmsize], m0
    mova    [dstq + lenq], m1
    add     src1q, 2*mmsize
    sub     lenq,  2*mmsize
    jge     .loop
    REP_RET
%endmacro

INIT_XMM sse
VECTOR_FMUL_REVERSE
INIT_YMM avx
VECTOR_FMUL_REVERSE

;-----------------------------------------------------------------------------
; vector_fmul_add(float *dst, const float *src0, const float *src1,
;                 const float *src2, int len)
;-----------------------------------------------------------------------------
%macro VECTOR_FMUL_ADD 0
cglobal vector_fmul_add, 5,5,2, dst, src0, src1, src2, len
    lea       lenq, [lend*4 - 2*mmsize]
ALIGN 16
.loop:
    mova    m0,   [src0q + lenq]
    mova    m1,   [src0q + lenq + mmsize]
    mulps   m0, m0, [src1q + lenq]
    mulps   m1, m1, [src1q + lenq + mmsize]
    addps   m0, m0, [src2q + lenq]
    addps   m1, m1, [src2q + lenq + mmsize]
    mova    [dstq + lenq], m0
    mova    [dstq + lenq + mmsize], m1

    sub     lenq,   2*mmsize
    jge     .loop
    REP_RET
%endmacro

INIT_XMM sse
VECTOR_FMUL_ADD
INIT_YMM avx
VECTOR_FMUL_ADD

;-----------------------------------------------------------------------------
; void ff_butterflies_float_interleave(float *dst, const float *src0,
;                                      const float *src1, int len);
;-----------------------------------------------------------------------------

%macro BUTTERFLIES_FLOAT_INTERLEAVE 0
cglobal butterflies_float_interleave, 4,4,3, dst, src0, src1, len
%if ARCH_X86_64
    movsxd    lenq, lend
%endif
    test      lenq, lenq
    jz .end
    shl       lenq, 2
    lea      src0q, [src0q +   lenq]
    lea      src1q, [src1q +   lenq]
    lea       dstq, [ dstq + 2*lenq]
    neg       lenq
.loop:
    mova        m0, [src0q + lenq]
    mova        m1, [src1q + lenq]
    subps       m2, m0, m1
    addps       m0, m0, m1
    unpcklps    m1, m0, m2
    unpckhps    m0, m0, m2
%if cpuflag(avx)
    vextractf128 [dstq + 2*lenq     ], m1, 0
    vextractf128 [dstq + 2*lenq + 16], m0, 0
    vextractf128 [dstq + 2*lenq + 32], m1, 1
    vextractf128 [dstq + 2*lenq + 48], m0, 1
%else
    mova [dstq + 2*lenq         ], m1
    mova [dstq + 2*lenq + mmsize], m0
%endif
    add       lenq, mmsize
    jl .loop
.end:
    REP_RET
%endmacro

INIT_XMM sse
BUTTERFLIES_FLOAT_INTERLEAVE
INIT_YMM avx
BUTTERFLIES_FLOAT_INTERLEAVE

; %1 = aligned/unaligned
%macro BSWAP_LOOPS  1
    mov      r3, r2
    sar      r2, 3
    jz       .left4_%1
.loop8_%1:
    mov%1    m0, [r1 +  0]
    mov%1    m1, [r1 + 16]
%if cpuflag(ssse3)
    pshufb   m0, m2
    pshufb   m1, m2
    mova     [r0 +  0], m0
    mova     [r0 + 16], m1
%else
    pshuflw  m0, m0, 10110001b
    pshuflw  m1, m1, 10110001b
    pshufhw  m0, m0, 10110001b
    pshufhw  m1, m1, 10110001b
    mova     m2, m0
    mova     m3, m1
    psllw    m0, 8
    psllw    m1, 8
    psrlw    m2, 8
    psrlw    m3, 8
    por      m2, m0
    por      m3, m1
    mova     [r0 +  0], m2
    mova     [r0 + 16], m3
%endif
    add      r0, 32
    add      r1, 32
    dec      r2
    jnz      .loop8_%1
.left4_%1:
    mov      r2, r3
    and      r3, 4
    jz       .left
    mov%1    m0, [r1]
%if cpuflag(ssse3)
    pshufb   m0, m2
    mova     [r0], m0
%else
    pshuflw  m0, m0, 10110001b
    pshufhw  m0, m0, 10110001b
    mova     m2, m0
    psllw    m0, 8
    psrlw    m2, 8
    por      m2, m0
    mova     [r0], m2
%endif
    add      r1, 16
    add      r0, 16
%endmacro

; void bswap_buf(uint32_t *dst, const uint32_t *src, int w);
%macro BSWAP32_BUF 0
%if cpuflag(ssse3)
cglobal bswap32_buf, 3,4,3
    mov      r3, r1
    mova     m2, [pb_bswap32]
%else
cglobal bswap32_buf, 3,4,5
    mov      r3, r1
%endif
    and      r3, 15
    jz       .start_align
    BSWAP_LOOPS  u
    jmp      .left
.start_align:
    BSWAP_LOOPS  a
.left:
%if cpuflag(ssse3)
    mov      r3, r2
    and      r2, 2
    jz       .left1
    movq     m0, [r1]
    pshufb   m0, m2
    movq     [r0], m0
    add      r1, 8
    add      r0, 8
.left1:
    and      r3, 1
    jz       .end
    mov      r2d, [r1]
    bswap    r2d
    mov      [r0], r2d
%else
    and      r2, 3
    jz       .end
.loop2:
    mov      r3d, [r1]
    bswap    r3d
    mov      [r0], r3d
    add      r1, 4
    add      r0, 4
    dec      r2
    jnz      .loop2
%endif
.end:
    RET
%endmacro

INIT_XMM sse2
BSWAP32_BUF

INIT_XMM ssse3
BSWAP32_BUF

%macro op_avgh 3
    movh   %3, %2
    pavgb  %1, %3
    movh   %2, %1
%endmacro

%macro op_avg 2
    pavgb  %1, %2
    mova   %2, %1
%endmacro

%macro op_puth 2-3
    movh   %2, %1
%endmacro

%macro op_put 2
    mova   %2, %1
%endmacro

; void pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
%macro PIXELS4_L2 1
%define OP op_%1h
cglobal %1_pixels4_l2, 6,6
    movsxdifnidn r3, r3d
    movsxdifnidn r4, r4d
    test        r5d, 1
    je        .loop
    movd         m0, [r1]
    movd         m1, [r2]
    add          r1, r4
    add          r2, 4
    pavgb        m0, m1
    OP           m0, [r0], m3
    add          r0, r3
    dec         r5d
.loop:
    mova         m0, [r1]
    mova         m1, [r1+r4]
    lea          r1, [r1+2*r4]
    pavgb        m0, [r2]
    pavgb        m1, [r2+4]
    OP           m0, [r0], m3
    OP           m1, [r0+r3], m3
    lea          r0, [r0+2*r3]
    mova         m0, [r1]
    mova         m1, [r1+r4]
    lea          r1, [r1+2*r4]
    pavgb        m0, [r2+8]
    pavgb        m1, [r2+12]
    OP           m0, [r0], m3
    OP           m1, [r0+r3], m3
    lea          r0, [r0+2*r3]
    add          r2, 16
    sub         r5d, 4
    jne       .loop
    REP_RET
%endmacro

INIT_MMX mmxext
PIXELS4_L2 put
PIXELS4_L2 avg

; void pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
%macro PIXELS8_L2 1
%define OP op_%1
cglobal %1_pixels8_l2, 6,6
    movsxdifnidn r3, r3d
    movsxdifnidn r4, r4d
    test        r5d, 1
    je        .loop
    mova         m0, [r1]
    mova         m1, [r2]
    add          r1, r4
    add          r2, 8
    pavgb        m0, m1
    OP           m0, [r0]
    add          r0, r3
    dec         r5d
.loop:
    mova         m0, [r1]
    mova         m1, [r1+r4]
    lea          r1, [r1+2*r4]
    pavgb        m0, [r2]
    pavgb        m1, [r2+8]
    OP           m0, [r0]
    OP           m1, [r0+r3]
    lea          r0, [r0+2*r3]
    mova         m0, [r1]
    mova         m1, [r1+r4]
    lea          r1, [r1+2*r4]
    pavgb        m0, [r2+16]
    pavgb        m1, [r2+24]
    OP           m0, [r0]
    OP           m1, [r0+r3]
    lea          r0, [r0+2*r3]
    add          r2, 32
    sub         r5d, 4
    jne       .loop
    REP_RET
%endmacro

INIT_MMX mmxext
PIXELS8_L2 put
PIXELS8_L2 avg

; void pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
%macro PIXELS16_L2 1
%define OP op_%1
cglobal %1_pixels16_l2, 6,6
    movsxdifnidn r3, r3d
    movsxdifnidn r4, r4d
    test        r5d, 1
    je        .loop
    mova         m0, [r1]
    mova         m1, [r1+8]
    pavgb        m0, [r2]
    pavgb        m1, [r2+8]
    add          r1, r4
    add          r2, 16
    OP           m0, [r0]
    OP           m1, [r0+8]
    add          r0, r3
    dec         r5d
.loop:
    mova         m0, [r1]
    mova         m1, [r1+8]
    add          r1, r4
    pavgb        m0, [r2]
    pavgb        m1, [r2+8]
    OP           m0, [r0]
    OP           m1, [r0+8]
    add          r0, r3
    mova         m0, [r1]
    mova         m1, [r1+8]
    add          r1, r4
    pavgb        m0, [r2+16]
    pavgb        m1, [r2+24]
    OP           m0, [r0]
    OP           m1, [r0+8]
    add          r0, r3
    add          r2, 32
    sub         r5d, 2
    jne       .loop
    REP_RET
%endmacro

INIT_MMX mmxext
PIXELS16_L2 put
PIXELS16_L2 avg

INIT_MMX mmxext
; void pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)
%macro PIXELS48 2
%if %2 == 4
%define OP movh
%else
%define OP mova
%endif
cglobal %1_pixels%2, 4,5
    movsxdifnidn r2, r2d
    lea          r4, [r2*3]
.loop:
    OP           m0, [r1]
    OP           m1, [r1+r2]
    OP           m2, [r1+r2*2]
    OP           m3, [r1+r4]
    lea          r1, [r1+r2*4]
%ifidn %1, avg
    pavgb        m0, [r0]
    pavgb        m1, [r0+r2]
    pavgb        m2, [r0+r2*2]
    pavgb        m3, [r0+r4]
%endif
    OP         [r0], m0
    OP      [r0+r2], m1
    OP    [r0+r2*2], m2
    OP      [r0+r4], m3
    sub         r3d, 4
    lea          r0, [r0+r2*4]
    jne       .loop
    RET
%endmacro

PIXELS48 put, 4
PIXELS48 avg, 4
PIXELS48 put, 8
PIXELS48 avg, 8

INIT_XMM sse2
; void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
cglobal put_pixels16, 4,5,4
    movsxdifnidn r2, r2d
    lea          r4, [r2*3]
.loop:
    movu         m0, [r1]
    movu         m1, [r1+r2]
    movu         m2, [r1+r2*2]
    movu         m3, [r1+r4]
    lea          r1, [r1+r2*4]
    mova       [r0], m0
    mova    [r0+r2], m1
    mova  [r0+r2*2], m2
    mova    [r0+r4], m3
    sub         r3d, 4
    lea          r0, [r0+r2*4]
    jnz       .loop
    REP_RET

; void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
cglobal avg_pixels16, 4,5,4
    movsxdifnidn r2, r2d
    lea          r4, [r2*3]
.loop:
    movu         m0, [r1]
    movu         m1, [r1+r2]
    movu         m2, [r1+r2*2]
    movu         m3, [r1+r4]
    lea          r1, [r1+r2*4]
    pavgb        m0, [r0]
    pavgb        m1, [r0+r2]
    pavgb        m2, [r0+r2*2]
    pavgb        m3, [r0+r4]
    mova       [r0], m0
    mova    [r0+r2], m1
    mova  [r0+r2*2], m2
    mova    [r0+r4], m3
    sub         r3d, 4
    lea          r0, [r0+r2*4]
    jnz       .loop
    REP_RET