summaryrefslogtreecommitdiffstats
path: root/media/libvpx/vp9/common/x86/vp9_postproc_sse2.asm
blob: ec8bfdb18fa819f50c8312ab9af3305892d7621d (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
;
;  Copyright (c) 2010 The WebM 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 "vpx_ports/x86_abi_support.asm"

;void vp9_post_proc_down_and_across_xmm
;(
;    unsigned char *src_ptr,
;    unsigned char *dst_ptr,
;    int src_pixels_per_line,
;    int dst_pixels_per_line,
;    int rows,
;    int cols,
;    int flimit
;)
global sym(vp9_post_proc_down_and_across_xmm) PRIVATE
sym(vp9_post_proc_down_and_across_xmm):
    push        rbp
    mov         rbp, rsp
    SHADOW_ARGS_TO_STACK 7
    SAVE_XMM 7
    GET_GOT     rbx
    push        rsi
    push        rdi
    ; end prolog

%if ABI_IS_32BIT=1 && CONFIG_PIC=1
    ALIGN_STACK 16, rax
    ; move the global rd onto the stack, since we don't have enough registers
    ; to do PIC addressing
    movdqa      xmm0, [GLOBAL(rd42)]
    sub         rsp, 16
    movdqa      [rsp], xmm0
%define RD42 [rsp]
%else
%define RD42 [GLOBAL(rd42)]
%endif


        movd        xmm2,       dword ptr arg(6) ;flimit
        punpcklwd   xmm2,       xmm2
        punpckldq   xmm2,       xmm2
        punpcklqdq  xmm2,       xmm2

        mov         rsi,        arg(0) ;src_ptr
        mov         rdi,        arg(1) ;dst_ptr

        movsxd      rcx,        DWORD PTR arg(4) ;rows
        movsxd      rax,        DWORD PTR arg(2) ;src_pixels_per_line ; destination pitch?
        pxor        xmm0,       xmm0              ; mm0 = 00000000

.nextrow:

        xor         rdx,        rdx       ; clear out rdx for use as loop counter
.nextcol:
        movq        xmm3,       QWORD PTR [rsi]         ; mm4 = r0 p0..p7
        punpcklbw   xmm3,       xmm0                    ; mm3 = p0..p3
        movdqa      xmm1,       xmm3                    ; mm1 = p0..p3
        psllw       xmm3,       2                       ;

        movq        xmm5,       QWORD PTR [rsi + rax]   ; mm4 = r1 p0..p7
        punpcklbw   xmm5,       xmm0                    ; mm5 = r1 p0..p3
        paddusw     xmm3,       xmm5                    ; mm3 += mm6

        ; thresholding
        movdqa      xmm7,       xmm1                    ; mm7 = r0 p0..p3
        psubusw     xmm7,       xmm5                    ; mm7 = r0 p0..p3 - r1 p0..p3
        psubusw     xmm5,       xmm1                    ; mm5 = r1 p0..p3 - r0 p0..p3
        paddusw     xmm7,       xmm5                    ; mm7 = abs(r0 p0..p3 - r1 p0..p3)
        pcmpgtw     xmm7,       xmm2

        movq        xmm5,       QWORD PTR [rsi + 2*rax] ; mm4 = r2 p0..p7
        punpcklbw   xmm5,       xmm0                    ; mm5 = r2 p0..p3
        paddusw     xmm3,       xmm5                    ; mm3 += mm5

        ; thresholding
        movdqa      xmm6,       xmm1                    ; mm6 = r0 p0..p3
        psubusw     xmm6,       xmm5                    ; mm6 = r0 p0..p3 - r2 p0..p3
        psubusw     xmm5,       xmm1                    ; mm5 = r2 p0..p3 - r2 p0..p3
        paddusw     xmm6,       xmm5                    ; mm6 = abs(r0 p0..p3 - r2 p0..p3)
        pcmpgtw     xmm6,       xmm2
        por         xmm7,       xmm6                    ; accumulate thresholds


        neg         rax
        movq        xmm5,       QWORD PTR [rsi+2*rax]   ; mm4 = r-2 p0..p7
        punpcklbw   xmm5,       xmm0                    ; mm5 = r-2 p0..p3
        paddusw     xmm3,       xmm5                    ; mm3 += mm5

        ; thresholding
        movdqa      xmm6,       xmm1                    ; mm6 = r0 p0..p3
        psubusw     xmm6,       xmm5                    ; mm6 = p0..p3 - r-2 p0..p3
        psubusw     xmm5,       xmm1                    ; mm5 = r-2 p0..p3 - p0..p3
        paddusw     xmm6,       xmm5                    ; mm6 = abs(r0 p0..p3 - r-2 p0..p3)
        pcmpgtw     xmm6,       xmm2
        por         xmm7,       xmm6                    ; accumulate thresholds

        movq        xmm4,       QWORD PTR [rsi+rax]     ; mm4 = r-1 p0..p7
        punpcklbw   xmm4,       xmm0                    ; mm4 = r-1 p0..p3
        paddusw     xmm3,       xmm4                    ; mm3 += mm5

        ; thresholding
        movdqa      xmm6,       xmm1                    ; mm6 = r0 p0..p3
        psubusw     xmm6,       xmm4                    ; mm6 = p0..p3 - r-2 p0..p3
        psubusw     xmm4,       xmm1                    ; mm5 = r-1 p0..p3 - p0..p3
        paddusw     xmm6,       xmm4                    ; mm6 = abs(r0 p0..p3 - r-1 p0..p3)
        pcmpgtw     xmm6,       xmm2
        por         xmm7,       xmm6                    ; accumulate thresholds


        paddusw     xmm3,       RD42                    ; mm3 += round value
        psraw       xmm3,       3                       ; mm3 /= 8

        pand        xmm1,       xmm7                    ; mm1 select vals > thresh from source
        pandn       xmm7,       xmm3                    ; mm7 select vals < thresh from blurred result
        paddusw     xmm1,       xmm7                    ; combination

        packuswb    xmm1,       xmm0                    ; pack to bytes
        movq        QWORD PTR [rdi], xmm1             ;

        neg         rax                   ; pitch is positive
        add         rsi,        8
        add         rdi,        8

        add         rdx,        8
        cmp         edx,        dword arg(5) ;cols

        jl          .nextcol

        ; done with the all cols, start the across filtering in place
        sub         rsi,        rdx
        sub         rdi,        rdx

        xor         rdx,        rdx
        movq        mm0,        QWORD PTR [rdi-8];

.acrossnextcol:
        movq        xmm7,       QWORD PTR [rdi +rdx -2]
        movd        xmm4,       DWORD PTR [rdi +rdx +6]

        pslldq      xmm4,       8
        por         xmm4,       xmm7

        movdqa      xmm3,       xmm4
        psrldq      xmm3,       2
        punpcklbw   xmm3,       xmm0              ; mm3 = p0..p3
        movdqa      xmm1,       xmm3              ; mm1 = p0..p3
        psllw       xmm3,       2


        movdqa      xmm5,       xmm4
        psrldq      xmm5,       3
        punpcklbw   xmm5,       xmm0              ; mm5 = p1..p4
        paddusw     xmm3,       xmm5              ; mm3 += mm6

        ; thresholding
        movdqa      xmm7,       xmm1              ; mm7 = p0..p3
        psubusw     xmm7,       xmm5              ; mm7 = p0..p3 - p1..p4
        psubusw     xmm5,       xmm1              ; mm5 = p1..p4 - p0..p3
        paddusw     xmm7,       xmm5              ; mm7 = abs(p0..p3 - p1..p4)
        pcmpgtw     xmm7,       xmm2

        movdqa      xmm5,       xmm4
        psrldq      xmm5,       4
        punpcklbw   xmm5,       xmm0              ; mm5 = p2..p5
        paddusw     xmm3,       xmm5              ; mm3 += mm5

        ; thresholding
        movdqa      xmm6,       xmm1              ; mm6 = p0..p3
        psubusw     xmm6,       xmm5              ; mm6 = p0..p3 - p1..p4
        psubusw     xmm5,       xmm1              ; mm5 = p1..p4 - p0..p3
        paddusw     xmm6,       xmm5              ; mm6 = abs(p0..p3 - p1..p4)
        pcmpgtw     xmm6,       xmm2
        por         xmm7,       xmm6              ; accumulate thresholds


        movdqa      xmm5,       xmm4              ; mm5 = p-2..p5
        punpcklbw   xmm5,       xmm0              ; mm5 = p-2..p1
        paddusw     xmm3,       xmm5              ; mm3 += mm5

        ; thresholding
        movdqa      xmm6,       xmm1              ; mm6 = p0..p3
        psubusw     xmm6,       xmm5              ; mm6 = p0..p3 - p1..p4
        psubusw     xmm5,       xmm1              ; mm5 = p1..p4 - p0..p3
        paddusw     xmm6,       xmm5              ; mm6 = abs(p0..p3 - p1..p4)
        pcmpgtw     xmm6,       xmm2
        por         xmm7,       xmm6              ; accumulate thresholds

        psrldq      xmm4,       1                   ; mm4 = p-1..p5
        punpcklbw   xmm4,       xmm0              ; mm4 = p-1..p2
        paddusw     xmm3,       xmm4              ; mm3 += mm5

        ; thresholding
        movdqa      xmm6,       xmm1              ; mm6 = p0..p3
        psubusw     xmm6,       xmm4              ; mm6 = p0..p3 - p1..p4
        psubusw     xmm4,       xmm1              ; mm5 = p1..p4 - p0..p3
        paddusw     xmm6,       xmm4              ; mm6 = abs(p0..p3 - p1..p4)
        pcmpgtw     xmm6,       xmm2
        por         xmm7,       xmm6              ; accumulate thresholds

        paddusw     xmm3,       RD42              ; mm3 += round value
        psraw       xmm3,       3                 ; mm3 /= 8

        pand        xmm1,       xmm7              ; mm1 select vals > thresh from source
        pandn       xmm7,       xmm3              ; mm7 select vals < thresh from blurred result
        paddusw     xmm1,       xmm7              ; combination

        packuswb    xmm1,       xmm0              ; pack to bytes
        movq        QWORD PTR [rdi+rdx-8],  mm0   ; store previous four bytes
        movdq2q     mm0,        xmm1

        add         rdx,        8
        cmp         edx,        dword arg(5) ;cols
        jl          .acrossnextcol;

        ; last 8 pixels
        movq        QWORD PTR [rdi+rdx-8],  mm0

        ; done with this rwo
        add         rsi,rax               ; next line
        mov         eax, dword arg(3) ;dst_pixels_per_line ; destination pitch?
        add         rdi,rax               ; next destination
        mov         eax, dword arg(2) ;src_pixels_per_line ; destination pitch?

        dec         rcx                   ; decrement count
        jnz         .nextrow              ; next row

%if ABI_IS_32BIT=1 && CONFIG_PIC=1
    add rsp,16
    pop rsp
%endif
    ; begin epilog
    pop rdi
    pop rsi
    RESTORE_GOT
    RESTORE_XMM
    UNSHADOW_ARGS
    pop         rbp
    ret
%undef RD42


;void vp9_mbpost_proc_down_xmm(unsigned char *dst,
;                            int pitch, int rows, int cols,int flimit)
extern sym(vp9_rv)
global sym(vp9_mbpost_proc_down_xmm) PRIVATE
sym(vp9_mbpost_proc_down_xmm):
    push        rbp
    mov         rbp, rsp
    SHADOW_ARGS_TO_STACK 5
    SAVE_XMM 7
    GET_GOT     rbx
    push        rsi
    push        rdi
    ; end prolog

    ALIGN_STACK 16, rax
    sub         rsp, 128+16

    ; unsigned char d[16][8] at [rsp]
    ; create flimit2 at [rsp+128]
    mov         eax, dword ptr arg(4) ;flimit
    mov         [rsp+128], eax
    mov         [rsp+128+4], eax
    mov         [rsp+128+8], eax
    mov         [rsp+128+12], eax
%define flimit4 [rsp+128]

%if ABI_IS_32BIT=0
    lea         r8,       [GLOBAL(sym(vp9_rv))]
%endif

    ;rows +=8;
    add         dword arg(2), 8

    ;for(c=0; c<cols; c+=8)
.loop_col:
            mov         rsi,        arg(0) ; s
            pxor        xmm0,       xmm0        ;

            movsxd      rax,        dword ptr arg(1) ;pitch       ;
            neg         rax                                     ; rax = -pitch

            lea         rsi,        [rsi + rax*8];              ; rdi = s[-pitch*8]
            neg         rax


            pxor        xmm5,       xmm5
            pxor        xmm6,       xmm6        ;

            pxor        xmm7,       xmm7        ;
            mov         rdi,        rsi

            mov         rcx,        15          ;

.loop_initvar:
            movq        xmm1,       QWORD PTR [rdi];
            punpcklbw   xmm1,       xmm0        ;

            paddw       xmm5,       xmm1        ;
            pmullw      xmm1,       xmm1        ;

            movdqa      xmm2,       xmm1        ;
            punpcklwd   xmm1,       xmm0        ;

            punpckhwd   xmm2,       xmm0        ;
            paddd       xmm6,       xmm1        ;

            paddd       xmm7,       xmm2        ;
            lea         rdi,        [rdi+rax]   ;

            dec         rcx
            jne         .loop_initvar
            ;save the var and sum
            xor         rdx,        rdx
.loop_row:
            movq        xmm1,       QWORD PTR [rsi]     ; [s-pitch*8]
            movq        xmm2,       QWORD PTR [rdi]     ; [s+pitch*7]

            punpcklbw   xmm1,       xmm0
            punpcklbw   xmm2,       xmm0

            paddw       xmm5,       xmm2
            psubw       xmm5,       xmm1

            pmullw      xmm2,       xmm2
            movdqa      xmm4,       xmm2

            punpcklwd   xmm2,       xmm0
            punpckhwd   xmm4,       xmm0

            paddd       xmm6,       xmm2
            paddd       xmm7,       xmm4

            pmullw      xmm1,       xmm1
            movdqa      xmm2,       xmm1

            punpcklwd   xmm1,       xmm0
            psubd       xmm6,       xmm1

            punpckhwd   xmm2,       xmm0
            psubd       xmm7,       xmm2


            movdqa      xmm3,       xmm6
            pslld       xmm3,       4

            psubd       xmm3,       xmm6
            movdqa      xmm1,       xmm5

            movdqa      xmm4,       xmm5
            pmullw      xmm1,       xmm1

            pmulhw      xmm4,       xmm4
            movdqa      xmm2,       xmm1

            punpcklwd   xmm1,       xmm4
            punpckhwd   xmm2,       xmm4

            movdqa      xmm4,       xmm7
            pslld       xmm4,       4

            psubd       xmm4,       xmm7

            psubd       xmm3,       xmm1
            psubd       xmm4,       xmm2

            psubd       xmm3,       flimit4
            psubd       xmm4,       flimit4

            psrad       xmm3,       31
            psrad       xmm4,       31

            packssdw    xmm3,       xmm4
            packsswb    xmm3,       xmm0

            movq        xmm1,       QWORD PTR [rsi+rax*8]

            movq        xmm2,       xmm1
            punpcklbw   xmm1,       xmm0

            paddw       xmm1,       xmm5
            mov         rcx,        rdx

            and         rcx,        127
%if ABI_IS_32BIT=1 && CONFIG_PIC=1
            push        rax
            lea         rax,        [GLOBAL(sym(vp9_rv))]
            movdqu      xmm4,       [rax + rcx*2] ;vp9_rv[rcx*2]
            pop         rax
%elif ABI_IS_32BIT=0
            movdqu      xmm4,       [r8 + rcx*2] ;vp9_rv[rcx*2]
%else
            movdqu      xmm4,       [sym(vp9_rv) + rcx*2]
%endif

            paddw       xmm1,       xmm4
            ;paddw     xmm1,       eight8s
            psraw       xmm1,       4

            packuswb    xmm1,       xmm0
            pand        xmm1,       xmm3

            pandn       xmm3,       xmm2
            por         xmm1,       xmm3

            and         rcx,        15
            movq        QWORD PTR   [rsp + rcx*8], xmm1 ;d[rcx*8]

            mov         rcx,        rdx
            sub         rcx,        8

            and         rcx,        15
            movq        mm0,        [rsp + rcx*8] ;d[rcx*8]

            movq        [rsi],      mm0
            lea         rsi,        [rsi+rax]

            lea         rdi,        [rdi+rax]
            add         rdx,        1

            cmp         edx,        dword arg(2) ;rows
            jl          .loop_row

        add         dword arg(0), 8 ; s += 8
        sub         dword arg(3), 8 ; cols -= 8
        cmp         dword arg(3), 0
        jg          .loop_col

    add         rsp, 128+16
    pop         rsp

    ; begin epilog
    pop rdi
    pop rsi
    RESTORE_GOT
    RESTORE_XMM
    UNSHADOW_ARGS
    pop         rbp
    ret
%undef flimit4


;void vp9_mbpost_proc_across_ip_xmm(unsigned char *src,
;                                int pitch, int rows, int cols,int flimit)
global sym(vp9_mbpost_proc_across_ip_xmm) PRIVATE
sym(vp9_mbpost_proc_across_ip_xmm):
    push        rbp
    mov         rbp, rsp
    SHADOW_ARGS_TO_STACK 5
    SAVE_XMM 7
    GET_GOT     rbx
    push        rsi
    push        rdi
    ; end prolog

    ALIGN_STACK 16, rax
    sub         rsp, 16

    ; create flimit4 at [rsp]
    mov         eax, dword ptr arg(4) ;flimit
    mov         [rsp], eax
    mov         [rsp+4], eax
    mov         [rsp+8], eax
    mov         [rsp+12], eax
%define flimit4 [rsp]


    ;for(r=0;r<rows;r++)
.ip_row_loop:

        xor         rdx,    rdx ;sumsq=0;
        xor         rcx,    rcx ;sum=0;
        mov         rsi,    arg(0); s
        mov         rdi,    -8
.ip_var_loop:
        ;for(i=-8;i<=6;i++)
        ;{
        ;    sumsq += s[i]*s[i];
        ;    sum   += s[i];
        ;}
        movzx       eax, byte [rsi+rdi]
        add         ecx, eax
        mul         al
        add         edx, eax
        add         rdi, 1
        cmp         rdi, 6
        jle         .ip_var_loop


            ;mov         rax,    sumsq
            ;movd        xmm7,   rax
            movd        xmm7,   edx

            ;mov         rax,    sum
            ;movd        xmm6,   rax
            movd        xmm6,   ecx

            mov         rsi,    arg(0) ;s
            xor         rcx,    rcx

            movsxd      rdx,    dword arg(3) ;cols
            add         rdx,    8
            pxor        mm0,    mm0
            pxor        mm1,    mm1

            pxor        xmm0,   xmm0
.nextcol4:

            movd        xmm1,   DWORD PTR [rsi+rcx-8]   ; -8 -7 -6 -5
            movd        xmm2,   DWORD PTR [rsi+rcx+7]   ; +7 +8 +9 +10

            punpcklbw   xmm1,   xmm0                    ; expanding
            punpcklbw   xmm2,   xmm0                    ; expanding

            punpcklwd   xmm1,   xmm0                    ; expanding to dwords
            punpcklwd   xmm2,   xmm0                    ; expanding to dwords

            psubd       xmm2,   xmm1                    ; 7--8   8--7   9--6 10--5
            paddd       xmm1,   xmm1                    ; -8*2   -7*2   -6*2 -5*2

            paddd       xmm1,   xmm2                    ; 7+-8   8+-7   9+-6 10+-5
            pmaddwd     xmm1,   xmm2                    ; squared of 7+-8   8+-7   9+-6 10+-5

            paddd       xmm6,   xmm2
            paddd       xmm7,   xmm1

            pshufd      xmm6,   xmm6,   0               ; duplicate the last ones
            pshufd      xmm7,   xmm7,   0               ; duplicate the last ones

            psrldq      xmm1,       4                   ; 8--7   9--6 10--5  0000
            psrldq      xmm2,       4                   ; 8--7   9--6 10--5  0000

            pshufd      xmm3,   xmm1,   3               ; 0000  8--7   8--7   8--7 squared
            pshufd      xmm4,   xmm2,   3               ; 0000  8--7   8--7   8--7 squared

            paddd       xmm6,   xmm4
            paddd       xmm7,   xmm3

            pshufd      xmm3,   xmm1,   01011111b       ; 0000  0000   9--6   9--6 squared
            pshufd      xmm4,   xmm2,   01011111b       ; 0000  0000   9--6   9--6 squared

            paddd       xmm7,   xmm3
            paddd       xmm6,   xmm4

            pshufd      xmm3,   xmm1,   10111111b       ; 0000  0000   8--7   8--7 squared
            pshufd      xmm4,   xmm2,   10111111b       ; 0000  0000   8--7   8--7 squared

            paddd       xmm7,   xmm3
            paddd       xmm6,   xmm4

            movdqa      xmm3,   xmm6
            pmaddwd     xmm3,   xmm3

            movdqa      xmm5,   xmm7
            pslld       xmm5,   4

            psubd       xmm5,   xmm7
            psubd       xmm5,   xmm3

            psubd       xmm5,   flimit4
            psrad       xmm5,   31

            packssdw    xmm5,   xmm0
            packsswb    xmm5,   xmm0

            movd        xmm1,   DWORD PTR [rsi+rcx]
            movq        xmm2,   xmm1

            punpcklbw   xmm1,   xmm0
            punpcklwd   xmm1,   xmm0

            paddd       xmm1,   xmm6
            paddd       xmm1,   [GLOBAL(four8s)]

            psrad       xmm1,   4
            packssdw    xmm1,   xmm0

            packuswb    xmm1,   xmm0
            pand        xmm1,   xmm5

            pandn       xmm5,   xmm2
            por         xmm5,   xmm1

            movd        [rsi+rcx-8],  mm0
            movq        mm0,    mm1

            movdq2q     mm1,    xmm5
            psrldq      xmm7,   12

            psrldq      xmm6,   12
            add         rcx,    4

            cmp         rcx,    rdx
            jl          .nextcol4

        ;s+=pitch;
        movsxd rax, dword arg(1)
        add    arg(0), rax

        sub dword arg(2), 1 ;rows-=1
        cmp dword arg(2), 0
        jg .ip_row_loop

    add         rsp, 16
    pop         rsp

    ; begin epilog
    pop rdi
    pop rsi
    RESTORE_GOT
    RESTORE_XMM
    UNSHADOW_ARGS
    pop         rbp
    ret
%undef flimit4


;void vp9_plane_add_noise_wmt (unsigned char *start, unsigned char *noise,
;                            unsigned char blackclamp[16],
;                            unsigned char whiteclamp[16],
;                            unsigned char bothclamp[16],
;                            unsigned int width, unsigned int height, int pitch)
global sym(vp9_plane_add_noise_wmt) PRIVATE
sym(vp9_plane_add_noise_wmt):
    push        rbp
    mov         rbp, rsp
    SHADOW_ARGS_TO_STACK 8
    GET_GOT     rbx
    push        rsi
    push        rdi
    ; end prolog

.addnoise_loop:
    call sym(LIBVPX_RAND) WRT_PLT
    mov     rcx, arg(1) ;noise
    and     rax, 0xff
    add     rcx, rax

    ; we rely on the fact that the clamping vectors are stored contiguously
    ; in black/white/both order. Note that we have to reload this here because
    ; rdx could be trashed by rand()
    mov     rdx, arg(2) ; blackclamp


            mov     rdi, rcx
            movsxd  rcx, dword arg(5) ;[Width]
            mov     rsi, arg(0) ;Pos
            xor         rax,rax

.addnoise_nextset:
            movdqu      xmm1,[rsi+rax]         ; get the source

            psubusb     xmm1, [rdx]    ;blackclamp        ; clamp both sides so we don't outrange adding noise
            paddusb     xmm1, [rdx+32] ;bothclamp
            psubusb     xmm1, [rdx+16] ;whiteclamp

            movdqu      xmm2,[rdi+rax]         ; get the noise for this line
            paddb       xmm1,xmm2              ; add it in
            movdqu      [rsi+rax],xmm1         ; store the result

            add         rax,16                 ; move to the next line

            cmp         rax, rcx
            jl          .addnoise_nextset

    movsxd  rax, dword arg(7) ; Pitch
    add     arg(0), rax ; Start += Pitch
    sub     dword arg(6), 1   ; Height -= 1
    jg      .addnoise_loop

    ; begin epilog
    pop rdi
    pop rsi
    RESTORE_GOT
    UNSHADOW_ARGS
    pop         rbp
    ret


SECTION_RODATA
align 16
rd42:
    times 8 dw 0x04
four8s:
    times 4 dd 8