aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/libjpeg-turbo/simd/x86_64/jdsample-avx2.asm
blob: 1e4979f933e46b98b1d581339421befe6e4bbf5e (plain) (blame)
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
;
; jdsample.asm - upsampling (64-bit AVX2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright (C) 2009, 2016, D. R. Commander.
; Copyright (C) 2015, Intel Corporation.
; Copyright (C) 2018, Matthias Räncker.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
; For conditions of distribution and use, see copyright notice in jsimdext.inc
;
; This file should be assembled with NASM (Netwide Assembler),
; can *not* be assembled with Microsoft's MASM or any compatible
; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208

%include "jsimdext.inc"

; --------------------------------------------------------------------------
    SECTION     SEG_CONST

    alignz      32
    GLOBAL_DATA(jconst_fancy_upsample_avx2)

EXTN(jconst_fancy_upsample_avx2):

PW_ONE   times 16 dw 1
PW_TWO   times 16 dw 2
PW_THREE times 16 dw 3
PW_SEVEN times 16 dw 7
PW_EIGHT times 16 dw 8

    alignz      32

; --------------------------------------------------------------------------
    SECTION     SEG_TEXT
    BITS        64
;
; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
;
; The upsampling algorithm is linear interpolation between pixel centers,
; also known as a "triangle filter".  This is a good compromise between
; speed and visual quality.  The centers of the output pixels are 1/4 and 3/4
; of the way between input pixel centers.
;
; GLOBAL(void)
; jsimd_h2v1_fancy_upsample_avx2(int max_v_samp_factor,
;                                JDIMENSION downsampled_width,
;                                JSAMPARRAY input_data,
;                                JSAMPARRAY *output_data_ptr);
;

; r10 = int max_v_samp_factor
; r11d = JDIMENSION downsampled_width
; r12 = JSAMPARRAY input_data
; r13 = JSAMPARRAY *output_data_ptr

    align       32
    GLOBAL_FUNCTION(jsimd_h2v1_fancy_upsample_avx2)

EXTN(jsimd_h2v1_fancy_upsample_avx2):
    push        rbp
    mov         rax, rsp
    mov         rbp, rsp
    push_xmm    3
    collect_args 4

    mov         eax, r11d               ; colctr
    test        rax, rax
    jz          near .return

    mov         rcx, r10                ; rowctr
    test        rcx, rcx
    jz          near .return

    mov         rsi, r12                ; input_data
    mov         rdi, r13
    mov         rdip, JSAMPARRAY [rdi]  ; output_data

    vpxor       ymm0, ymm0, ymm0                 ; ymm0=(all 0's)
    vpcmpeqb    xmm9, xmm9, xmm9
    vpsrldq     xmm10, xmm9, (SIZEOF_XMMWORD-1)  ; (ff -- -- -- ... -- --) LSB is ff

    vpslldq     xmm9, xmm9, (SIZEOF_XMMWORD-1)
    vperm2i128  ymm9, ymm9, ymm9, 1              ; (---- ---- ... ---- ---- ff) MSB is ff

.rowloop:
    push        rax                     ; colctr
    push        rdi
    push        rsi

    mov         rsip, JSAMPROW [rsi]    ; inptr
    mov         rdip, JSAMPROW [rdi]    ; outptr

    test        rax, SIZEOF_YMMWORD-1
    jz          short .skip
    mov         dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE]
    mov         JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl    ; insert a dummy sample
.skip:
    vpand       ymm7, ymm10, YMMWORD [rsi+0*SIZEOF_YMMWORD]

    add         rax, byte SIZEOF_YMMWORD-1
    and         rax, byte -SIZEOF_YMMWORD
    cmp         rax, byte SIZEOF_YMMWORD
    ja          short .columnloop

.columnloop_last:
    vpand       ymm6, ymm9, YMMWORD [rsi+0*SIZEOF_YMMWORD]
    jmp         short .upsample

.columnloop:
    vmovdqu     ymm6, YMMWORD [rsi+1*SIZEOF_YMMWORD]
    vperm2i128  ymm6, ymm0, ymm6, 0x20
    vpslldq     ymm6, ymm6, 15

.upsample:
    vmovdqu     ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD]  ; ymm1=( 0  1  2 ... 29 30 31)

    vperm2i128  ymm2, ymm0, ymm1, 0x20
    vpalignr    ymm2, ymm1, ymm2, 15            ; ymm2=(--  0  1 ... 28 29 30)
    vperm2i128  ymm4, ymm0, ymm1, 0x03
    vpalignr    ymm3, ymm4, ymm1, 1             ; ymm3=( 1  2  3 ... 30 31 --)

    vpor        ymm2, ymm2, ymm7                ; ymm2=(-1  0  1 ... 28 29 30)
    vpor        ymm3, ymm3, ymm6                ; ymm3=( 1  2  3 ... 30 31 32)

    vpsrldq     ymm7, ymm4, (SIZEOF_XMMWORD-1)  ; ymm7=(31 -- -- ... -- -- --)

    vpunpckhbw  ymm4, ymm1, ymm0                ; ymm4=( 8  9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
    vpunpcklbw  ymm5, ymm1, ymm0                ; ymm5=( 0  1  2  3  4  5  6  7 16 17 18 19 20 21 22 23)
    vperm2i128  ymm1, ymm5, ymm4, 0x20          ; ymm1=( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vperm2i128  ymm4, ymm5, ymm4, 0x31          ; ymm4=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vpunpckhbw  ymm5, ymm2, ymm0                ; ymm5=( 7  8  9 10 11 12 13 14 23 24 25 26 27 28 29 30)
    vpunpcklbw  ymm6, ymm2, ymm0                ; ymm6=(-1  0  1  2  3  4  5  6 15 16 17 18 19 20 21 22)
    vperm2i128  ymm2, ymm6, ymm5, 0x20          ; ymm2=(-1  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14)
    vperm2i128  ymm5, ymm6, ymm5, 0x31          ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)

    vpunpckhbw  ymm6, ymm3, ymm0                ; ymm6=( 1  2  3  4  5  6  7  8 17 18 19 20 21 22 23 24)
    vpunpcklbw  ymm8, ymm3, ymm0                ; ymm8=( 9 10 11 12 13 14 15 16 25 26 27 28 29 30 31 32)
    vperm2i128  ymm3, ymm8, ymm6, 0x20          ; ymm3=( 1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16)
    vperm2i128  ymm6, ymm8, ymm6, 0x31          ; ymm6=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)

    vpmullw     ymm1, ymm1, [rel PW_THREE]
    vpmullw     ymm4, ymm4, [rel PW_THREE]
    vpaddw      ymm2, ymm2, [rel PW_ONE]
    vpaddw      ymm5, ymm5, [rel PW_ONE]
    vpaddw      ymm3, ymm3, [rel PW_TWO]
    vpaddw      ymm6, ymm6, [rel PW_TWO]

    vpaddw      ymm2, ymm2, ymm1
    vpaddw      ymm5, ymm5, ymm4
    vpsrlw      ymm2, ymm2, 2                   ; ymm2=OutLE=( 0  2  4  6  8 10 12 14 16 18 20 22 24 26 28 30)
    vpsrlw      ymm5, ymm5, 2                   ; ymm5=OutHE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
    vpaddw      ymm3, ymm3, ymm1
    vpaddw      ymm6, ymm6, ymm4
    vpsrlw      ymm3, ymm3, 2                   ; ymm3=OutLO=( 1  3  5  7  9 11 13 15 17 19 21 23 25 27 29 31)
    vpsrlw      ymm6, ymm6, 2                   ; ymm6=OutHO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)

    vpsllw      ymm3, ymm3, BYTE_BIT
    vpsllw      ymm6, ymm6, BYTE_BIT
    vpor        ymm2, ymm2, ymm3                ; ymm2=OutL=( 0  1  2 ... 29 30 31)
    vpor        ymm5, ymm5, ymm6                ; ymm5=OutH=(32 33 34 ... 61 62 63)

    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm2
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm5

    sub         rax, byte SIZEOF_YMMWORD
    add         rsi, byte 1*SIZEOF_YMMWORD  ; inptr
    add         rdi, byte 2*SIZEOF_YMMWORD  ; outptr
    cmp         rax, byte SIZEOF_YMMWORD
    ja          near .columnloop
    test        eax, eax
    jnz         near .columnloop_last

    pop         rsi
    pop         rdi
    pop         rax

    add         rsi, byte SIZEOF_JSAMPROW  ; input_data
    add         rdi, byte SIZEOF_JSAMPROW  ; output_data
    dec         rcx                        ; rowctr
    jg          near .rowloop

.return:
    vzeroupper
    uncollect_args 4
    pop_xmm     3
    pop         rbp
    ret

; --------------------------------------------------------------------------
;
; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
; Again a triangle filter; see comments for h2v1 case, above.
;
; GLOBAL(void)
; jsimd_h2v2_fancy_upsample_avx2(int max_v_samp_factor,
;                                JDIMENSION downsampled_width,
;                                JSAMPARRAY input_data,
;                                JSAMPARRAY *output_data_ptr);
;

; r10 = int max_v_samp_factor
; r11d = JDIMENSION downsampled_width
; r12 = JSAMPARRAY input_data
; r13 = JSAMPARRAY *output_data_ptr

%define wk(i)   rbp - (WK_NUM - (i)) * SIZEOF_YMMWORD  ; ymmword wk[WK_NUM]
%define WK_NUM  4

    align       32
    GLOBAL_FUNCTION(jsimd_h2v2_fancy_upsample_avx2)

EXTN(jsimd_h2v2_fancy_upsample_avx2):
    push        rbp
    mov         rax, rsp                     ; rax = original rbp
    sub         rsp, byte 4
    and         rsp, byte (-SIZEOF_YMMWORD)  ; align to 256 bits
    mov         [rsp], rax
    mov         rbp, rsp                     ; rbp = aligned rbp
    lea         rsp, [wk(0)]
    push_xmm    3
    collect_args 4
    push        rbx

    mov         eax, r11d               ; colctr
    test        rax, rax
    jz          near .return

    mov         rcx, r10                ; rowctr
    test        rcx, rcx
    jz          near .return

    mov         rsi, r12                ; input_data
    mov         rdi, r13
    mov         rdip, JSAMPARRAY [rdi]  ; output_data
.rowloop:
    push        rax                     ; colctr
    push        rcx
    push        rdi
    push        rsi

    mov         rcxp, JSAMPROW [rsi-1*SIZEOF_JSAMPROW]  ; inptr1(above)
    mov         rbxp, JSAMPROW [rsi+0*SIZEOF_JSAMPROW]  ; inptr0
    mov         rsip, JSAMPROW [rsi+1*SIZEOF_JSAMPROW]  ; inptr1(below)
    mov         rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]  ; outptr0
    mov         rdip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]  ; outptr1

    vpxor       ymm8, ymm8, ymm8                 ; ymm8=(all 0's)
    vpcmpeqb    xmm9, xmm9, xmm9
    vpsrldq     xmm10, xmm9, (SIZEOF_XMMWORD-2)  ; (ffff ---- ---- ... ---- ----) LSB is ffff
    vpslldq     xmm9, xmm9, (SIZEOF_XMMWORD-2)
    vperm2i128  ymm9, ymm9, ymm9, 1              ; (---- ---- ... ---- ---- ffff) MSB is ffff

    test        rax, SIZEOF_YMMWORD-1
    jz          short .skip
    push        rdx
    mov         dl, JSAMPLE [rcx+(rax-1)*SIZEOF_JSAMPLE]
    mov         JSAMPLE [rcx+rax*SIZEOF_JSAMPLE], dl
    mov         dl, JSAMPLE [rbx+(rax-1)*SIZEOF_JSAMPLE]
    mov         JSAMPLE [rbx+rax*SIZEOF_JSAMPLE], dl
    mov         dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE]
    mov         JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl    ; insert a dummy sample
    pop         rdx
.skip:
    ; -- process the first column block

    vmovdqu     ymm0, YMMWORD [rbx+0*SIZEOF_YMMWORD]  ; ymm0=row[ 0][0]
    vmovdqu     ymm1, YMMWORD [rcx+0*SIZEOF_YMMWORD]  ; ymm1=row[-1][0]
    vmovdqu     ymm2, YMMWORD [rsi+0*SIZEOF_YMMWORD]  ; ymm2=row[+1][0]

    vpunpckhbw  ymm4, ymm0, ymm8        ; ymm4=row[ 0]( 8  9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
    vpunpcklbw  ymm5, ymm0, ymm8        ; ymm5=row[ 0]( 0  1  2  3  4  5  6  7 16 17 18 19 20 21 22 23)
    vperm2i128  ymm0, ymm5, ymm4, 0x20  ; ymm0=row[ 0]( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vperm2i128  ymm4, ymm5, ymm4, 0x31  ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vpunpckhbw  ymm5, ymm1, ymm8        ; ymm5=row[-1]( 8  9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
    vpunpcklbw  ymm6, ymm1, ymm8        ; ymm6=row[-1]( 0  1  2  3  4  5  6  7 16 17 18 19 20 21 22 23)
    vperm2i128  ymm1, ymm6, ymm5, 0x20  ; ymm1=row[-1]( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vperm2i128  ymm5, ymm6, ymm5, 0x31  ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vpunpckhbw  ymm6, ymm2, ymm8        ; ymm6=row[+1]( 8  9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
    vpunpcklbw  ymm3, ymm2, ymm8        ; ymm3=row[+1]( 0  1  2  3  4  5  6  7 16 17 18 19 20 21 22 23)
    vperm2i128  ymm2, ymm3, ymm6, 0x20  ; ymm2=row[+1]( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vperm2i128  ymm6, ymm3, ymm6, 0x31  ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vpmullw     ymm0, ymm0, [rel PW_THREE]
    vpmullw     ymm4, ymm4, [rel PW_THREE]

    vpaddw      ymm1, ymm1, ymm0        ; ymm1=Int0L=( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vpaddw      ymm5, ymm5, ymm4        ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
    vpaddw      ymm2, ymm2, ymm0        ; ymm2=Int1L=( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vpaddw      ymm6, ymm6, ymm4        ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vmovdqu     YMMWORD [rdx+0*SIZEOF_YMMWORD], ymm1  ; temporarily save
    vmovdqu     YMMWORD [rdx+1*SIZEOF_YMMWORD], ymm5  ; the intermediate data
    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm2
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm6

    vpand       ymm1, ymm1, ymm10       ; ymm1=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
    vpand       ymm2, ymm2, ymm10       ; ymm2=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)

    vmovdqa     YMMWORD [wk(0)], ymm1
    vmovdqa     YMMWORD [wk(1)], ymm2

    add         rax, byte SIZEOF_YMMWORD-1
    and         rax, byte -SIZEOF_YMMWORD
    cmp         rax, byte SIZEOF_YMMWORD
    ja          short .columnloop

.columnloop_last:
    ; -- process the last column block

    vpand       ymm1, ymm9, YMMWORD [rdx+1*SIZEOF_YMMWORD]
    vpand       ymm2, ymm9, YMMWORD [rdi+1*SIZEOF_YMMWORD]

    vmovdqa     YMMWORD [wk(2)], ymm1   ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31)
    vmovdqa     YMMWORD [wk(3)], ymm2   ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31)

    jmp         near .upsample

.columnloop:
    ; -- process the next column block

    vmovdqu     ymm0, YMMWORD [rbx+1*SIZEOF_YMMWORD]  ; ymm0=row[ 0][1]
    vmovdqu     ymm1, YMMWORD [rcx+1*SIZEOF_YMMWORD]  ; ymm1=row[-1][1]
    vmovdqu     ymm2, YMMWORD [rsi+1*SIZEOF_YMMWORD]  ; ymm2=row[+1][1]

    vpunpckhbw  ymm4, ymm0, ymm8        ; ymm4=row[ 0]( 8  9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
    vpunpcklbw  ymm5, ymm0, ymm8        ; ymm5=row[ 0]( 0  1  2  3  4  5  6  7 16 17 18 19 20 21 22 23)
    vperm2i128  ymm0, ymm5, ymm4, 0x20  ; ymm0=row[ 0]( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vperm2i128  ymm4, ymm5, ymm4, 0x31  ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vpunpckhbw  ymm5, ymm1, ymm8        ; ymm5=row[-1]( 8  9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
    vpunpcklbw  ymm6, ymm1, ymm8        ; ymm6=row[-1]( 0  1  2  3  4  5  6  7 16 17 18 19 20 21 22 23)
    vperm2i128  ymm1, ymm6, ymm5, 0x20  ; ymm1=row[-1]( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vperm2i128  ymm5, ymm6, ymm5, 0x31  ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vpunpckhbw  ymm6, ymm2, ymm8        ; ymm6=row[+1]( 8  9 10 11 12 13 14 15 24 25 26 27 28 29 30 31)
    vpunpcklbw  ymm7, ymm2, ymm8        ; ymm7=row[+1]( 0  1  2  3  4  5  6  7 16 17 18 19 20 21 22 23)
    vperm2i128  ymm2, ymm7, ymm6, 0x20  ; ymm2=row[+1]( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vperm2i128  ymm6, ymm7, ymm6, 0x31  ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vpmullw     ymm0, ymm0, [rel PW_THREE]
    vpmullw     ymm4, ymm4, [rel PW_THREE]

    vpaddw      ymm1, ymm1, ymm0        ; ymm1=Int0L=( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vpaddw      ymm5, ymm5, ymm4        ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)
    vpaddw      ymm2, ymm2, ymm0        ; ymm2=Int1L=( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vpaddw      ymm6, ymm6, ymm4        ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vmovdqu     YMMWORD [rdx+2*SIZEOF_YMMWORD], ymm1  ; temporarily save
    vmovdqu     YMMWORD [rdx+3*SIZEOF_YMMWORD], ymm5  ; the intermediate data
    vmovdqu     YMMWORD [rdi+2*SIZEOF_YMMWORD], ymm2
    vmovdqu     YMMWORD [rdi+3*SIZEOF_YMMWORD], ymm6

    vperm2i128  ymm1, ymm8, ymm1, 0x20
    vpslldq     ymm1, ymm1, 14          ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- --  0)
    vperm2i128  ymm2, ymm8, ymm2, 0x20
    vpslldq     ymm2, ymm2, 14          ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- --  0)

    vmovdqa     YMMWORD [wk(2)], ymm1
    vmovdqa     YMMWORD [wk(3)], ymm2

.upsample:
    ; -- process the upper row

    vmovdqu     ymm7, YMMWORD [rdx+0*SIZEOF_YMMWORD]  ; ymm7=Int0L=( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vmovdqu     ymm3, YMMWORD [rdx+1*SIZEOF_YMMWORD]  ; ymm3=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vperm2i128  ymm0, ymm8, ymm7, 0x03
    vpalignr    ymm0, ymm0, ymm7, 2     ; ymm0=( 1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 --)
    vperm2i128  ymm4, ymm8, ymm3, 0x20
    vpslldq     ymm4, ymm4, 14          ; ymm4=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16)

    vperm2i128  ymm5, ymm8, ymm7, 0x03
    vpsrldq     ymm5, ymm5, 14          ; ymm5=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
    vperm2i128  ymm6, ymm8, ymm3, 0x20
    vpalignr    ymm6, ymm3, ymm6, 14    ; ymm6=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)

    vpor        ymm0, ymm0, ymm4        ; ymm0=( 1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16)
    vpor        ymm5, ymm5, ymm6        ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)

    vperm2i128  ymm2, ymm8, ymm3, 0x03
    vpalignr    ymm2, ymm2, ymm3, 2     ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --)
    vperm2i128  ymm4, ymm8, ymm3, 0x03
    vpsrldq     ymm4, ymm4, 14          ; ymm4=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
    vperm2i128  ymm1, ymm8, ymm7, 0x20
    vpalignr    ymm1, ymm7, ymm1, 14    ; ymm1=(--  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14)

    vpor        ymm1, ymm1, YMMWORD [wk(0)]  ; ymm1=(-1  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14)
    vpor        ymm2, ymm2, YMMWORD [wk(2)]  ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)

    vmovdqa     YMMWORD [wk(0)], ymm4

    vpmullw     ymm7, ymm7, [rel PW_THREE]
    vpmullw     ymm3, ymm3, [rel PW_THREE]
    vpaddw      ymm1, ymm1, [rel PW_EIGHT]
    vpaddw      ymm5, ymm5, [rel PW_EIGHT]
    vpaddw      ymm0, ymm0, [rel PW_SEVEN]
    vpaddw      ymm2, [rel PW_SEVEN]

    vpaddw      ymm1, ymm1, ymm7
    vpaddw      ymm5, ymm5, ymm3
    vpsrlw      ymm1, ymm1, 4           ; ymm1=Out0LE=( 0  2  4  6  8 10 12 14 16 18 20 22 24 26 28 30)
    vpsrlw      ymm5, ymm5, 4           ; ymm5=Out0HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
    vpaddw      ymm0, ymm0, ymm7
    vpaddw      ymm2, ymm2, ymm3
    vpsrlw      ymm0, ymm0, 4           ; ymm0=Out0LO=( 1  3  5  7  9 11 13 15 17 19 21 23 25 27 29 31)
    vpsrlw      ymm2, ymm2, 4           ; ymm2=Out0HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)

    vpsllw      ymm0, ymm0, BYTE_BIT
    vpsllw      ymm2, ymm2, BYTE_BIT
    vpor        ymm1, ymm1, ymm0        ; ymm1=Out0L=( 0  1  2 ... 29 30 31)
    vpor        ymm5, ymm5, ymm2        ; ymm5=Out0H=(32 33 34 ... 61 62 63)

    vmovdqu     YMMWORD [rdx+0*SIZEOF_YMMWORD], ymm1
    vmovdqu     YMMWORD [rdx+1*SIZEOF_YMMWORD], ymm5

    ; -- process the lower row

    vmovdqu     ymm6, YMMWORD [rdi+0*SIZEOF_YMMWORD]  ; ymm6=Int1L=( 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15)
    vmovdqu     ymm4, YMMWORD [rdi+1*SIZEOF_YMMWORD]  ; ymm4=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31)

    vperm2i128  ymm7, ymm8, ymm6, 0x03
    vpalignr    ymm7, ymm7, ymm6, 2     ; ymm7=( 1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 --)
    vperm2i128  ymm3, ymm8, ymm4, 0x20
    vpslldq     ymm3, ymm3, 14          ; ymm3=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16)

    vperm2i128  ymm0, ymm8, ymm6, 0x03
    vpsrldq     ymm0, ymm0, 14          ; ymm0=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
    vperm2i128  ymm2, ymm8, ymm4, 0x20
    vpalignr    ymm2, ymm4, ymm2, 14    ; ymm2=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)

    vpor        ymm7, ymm7, ymm3        ; ymm7=( 1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16)
    vpor        ymm0, ymm0, ymm2        ; ymm0=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30)

    vperm2i128  ymm5, ymm8, ymm4, 0x03
    vpalignr    ymm5, ymm5, ymm4, 2     ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --)
    vperm2i128  ymm3, ymm8, ymm4, 0x03
    vpsrldq     ymm3, ymm3, 14          ; ymm3=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --)
    vperm2i128  ymm1, ymm8, ymm6, 0x20
    vpalignr    ymm1, ymm6, ymm1, 14    ; ymm1=(--  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14)

    vpor        ymm1, ymm1, YMMWORD [wk(1)]  ; ymm1=(-1  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14)
    vpor        ymm5, ymm5, YMMWORD [wk(3)]  ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32)

    vmovdqa     YMMWORD [wk(1)], ymm3

    vpmullw     ymm6, ymm6, [rel PW_THREE]
    vpmullw     ymm4, ymm4, [rel PW_THREE]
    vpaddw      ymm1, ymm1, [rel PW_EIGHT]
    vpaddw      ymm0, ymm0, [rel PW_EIGHT]
    vpaddw      ymm7, ymm7, [rel PW_SEVEN]
    vpaddw      ymm5, ymm5, [rel PW_SEVEN]

    vpaddw      ymm1, ymm1, ymm6
    vpaddw      ymm0, ymm0, ymm4
    vpsrlw      ymm1, ymm1, 4           ; ymm1=Out1LE=( 0  2  4  6  8 10 12 14 16 18 20 22 24 26 28 30)
    vpsrlw      ymm0, ymm0, 4           ; ymm0=Out1HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62)
    vpaddw      ymm7, ymm7, ymm6
    vpaddw      ymm5, ymm5, ymm4
    vpsrlw      ymm7, ymm7, 4           ; ymm7=Out1LO=( 1  3  5  7  9 11 13 15 17 19 21 23 25 27 29 31)
    vpsrlw      ymm5, ymm5, 4           ; ymm5=Out1HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63)

    vpsllw      ymm7, ymm7, BYTE_BIT
    vpsllw      ymm5, ymm5, BYTE_BIT
    vpor        ymm1, ymm1, ymm7        ; ymm1=Out1L=( 0  1  2 ... 29 30 31)
    vpor        ymm0, ymm0, ymm5        ; ymm0=Out1H=(32 33 34 ... 61 62 63)

    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm1
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm0

    sub         rax, byte SIZEOF_YMMWORD
    add         rcx, byte 1*SIZEOF_YMMWORD  ; inptr1(above)
    add         rbx, byte 1*SIZEOF_YMMWORD  ; inptr0
    add         rsi, byte 1*SIZEOF_YMMWORD  ; inptr1(below)
    add         rdx, byte 2*SIZEOF_YMMWORD  ; outptr0
    add         rdi, byte 2*SIZEOF_YMMWORD  ; outptr1
    cmp         rax, byte SIZEOF_YMMWORD
    ja          near .columnloop
    test        rax, rax
    jnz         near .columnloop_last

    pop         rsi
    pop         rdi
    pop         rcx
    pop         rax

    add         rsi, byte 1*SIZEOF_JSAMPROW  ; input_data
    add         rdi, byte 2*SIZEOF_JSAMPROW  ; output_data
    sub         rcx, byte 2                  ; rowctr
    jg          near .rowloop

.return:
    pop         rbx
    vzeroupper
    uncollect_args 4
    pop_xmm     3
    mov         rsp, rbp                ; rsp <- aligned rbp
    pop         rsp                     ; rsp <- original rbp
    pop         rbp
    ret

; --------------------------------------------------------------------------
;
; Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
; It's still a box filter.
;
; GLOBAL(void)
; jsimd_h2v1_upsample_avx2(int max_v_samp_factor, JDIMENSION output_width,
;                          JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
;

; r10 = int max_v_samp_factor
; r11d = JDIMENSION output_width
; r12 = JSAMPARRAY input_data
; r13 = JSAMPARRAY *output_data_ptr

    align       32
    GLOBAL_FUNCTION(jsimd_h2v1_upsample_avx2)

EXTN(jsimd_h2v1_upsample_avx2):
    push        rbp
    mov         rax, rsp
    mov         rbp, rsp
    collect_args 4

    mov         edx, r11d
    add         rdx, byte (SIZEOF_YMMWORD-1)
    and         rdx, -SIZEOF_YMMWORD
    jz          near .return

    mov         rcx, r10                ; rowctr
    test        rcx, rcx
    jz          short .return

    mov         rsi, r12                ; input_data
    mov         rdi, r13
    mov         rdip, JSAMPARRAY [rdi]  ; output_data
.rowloop:
    push        rdi
    push        rsi

    mov         rsip, JSAMPROW [rsi]    ; inptr
    mov         rdip, JSAMPROW [rdi]    ; outptr
    mov         rax, rdx                ; colctr
.columnloop:

    cmp         rax, byte SIZEOF_YMMWORD
    ja          near .above_16

    vmovdqu     xmm0, XMMWORD [rsi+0*SIZEOF_YMMWORD]
    vpunpckhbw  xmm1, xmm0, xmm0
    vpunpcklbw  xmm0, xmm0, xmm0

    vmovdqu     XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
    vmovdqu     XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1

    jmp         short .nextrow

.above_16:
    vmovdqu     ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]

    vpermq      ymm0, ymm0, 0xd8
    vpunpckhbw  ymm1, ymm0, ymm0
    vpunpcklbw  ymm0, ymm0, ymm0

    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm1

    sub         rax, byte 2*SIZEOF_YMMWORD
    jz          short .nextrow

    add         rsi, byte SIZEOF_YMMWORD    ; inptr
    add         rdi, byte 2*SIZEOF_YMMWORD  ; outptr
    jmp         short .columnloop

.nextrow:
    pop         rsi
    pop         rdi

    add         rsi, byte SIZEOF_JSAMPROW  ; input_data
    add         rdi, byte SIZEOF_JSAMPROW  ; output_data
    dec         rcx                        ; rowctr
    jg          short .rowloop

.return:
    vzeroupper
    uncollect_args 4
    pop         rbp
    ret

; --------------------------------------------------------------------------
;
; Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
; It's still a box filter.
;
; GLOBAL(void)
; jsimd_h2v2_upsample_avx2(int max_v_samp_factor, JDIMENSION output_width,
;                          JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
;

; r10 = int max_v_samp_factor
; r11d = JDIMENSION output_width
; r12 = JSAMPARRAY input_data
; r13 = JSAMPARRAY *output_data_ptr

    align       32
    GLOBAL_FUNCTION(jsimd_h2v2_upsample_avx2)

EXTN(jsimd_h2v2_upsample_avx2):
    push        rbp
    mov         rax, rsp
    mov         rbp, rsp
    collect_args 4
    push        rbx

    mov         edx, r11d
    add         rdx, byte (SIZEOF_YMMWORD-1)
    and         rdx, -SIZEOF_YMMWORD
    jz          near .return

    mov         rcx, r10                ; rowctr
    test        rcx, rcx
    jz          near .return

    mov         rsi, r12                ; input_data
    mov         rdi, r13
    mov         rdip, JSAMPARRAY [rdi]  ; output_data
.rowloop:
    push        rdi
    push        rsi

    mov         rsip, JSAMPROW [rsi]                   ; inptr
    mov         rbxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0
    mov         rdip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1
    mov         rax, rdx                               ; colctr
.columnloop:

    cmp         rax, byte SIZEOF_YMMWORD
    ja          short .above_16

    vmovdqu     xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
    vpunpckhbw  xmm1, xmm0, xmm0
    vpunpcklbw  xmm0, xmm0, xmm0

    vmovdqu     XMMWORD [rbx+0*SIZEOF_XMMWORD], xmm0
    vmovdqu     XMMWORD [rbx+1*SIZEOF_XMMWORD], xmm1
    vmovdqu     XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
    vmovdqu     XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1

    jmp         near .nextrow

.above_16:
    vmovdqu     ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]

    vpermq      ymm0, ymm0, 0xd8
    vpunpckhbw  ymm1, ymm0, ymm0
    vpunpcklbw  ymm0, ymm0, ymm0

    vmovdqu     YMMWORD [rbx+0*SIZEOF_YMMWORD], ymm0
    vmovdqu     YMMWORD [rbx+1*SIZEOF_YMMWORD], ymm1
    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm1

    sub         rax, byte 2*SIZEOF_YMMWORD
    jz          short .nextrow

    add         rsi, byte SIZEOF_YMMWORD  ; inptr
    add         rbx, 2*SIZEOF_YMMWORD     ; outptr0
    add         rdi, 2*SIZEOF_YMMWORD     ; outptr1
    jmp         short .columnloop

.nextrow:
    pop         rsi
    pop         rdi

    add         rsi, byte 1*SIZEOF_JSAMPROW  ; input_data
    add         rdi, byte 2*SIZEOF_JSAMPROW  ; output_data
    sub         rcx, byte 2                  ; rowctr
    jg          near .rowloop

.return:
    pop         rbx
    vzeroupper
    uncollect_args 4
    pop         rbp
    ret

; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
    align       32