aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/libjpeg-turbo/simd/x86_64/jdmrgext-avx2.asm
blob: 8b264b4f039fb9b22c59348dacd56952bf3e3588 (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
;
; jdmrgext.asm - merged upsampling/color conversion (64-bit AVX2)
;
; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright (C) 2009, 2012, 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 "jcolsamp.inc"

; --------------------------------------------------------------------------
;
; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
;
; GLOBAL(void)
; jsimd_h2v1_merged_upsample_avx2(JDIMENSION output_width,
;                                 JSAMPIMAGE input_buf,
;                                 JDIMENSION in_row_group_ctr,
;                                 JSAMPARRAY output_buf);
;

; r10d = JDIMENSION output_width
; r11 = JSAMPIMAGE input_buf
; r12d = JDIMENSION in_row_group_ctr
; r13 = JSAMPARRAY output_buf

%define wk(i)   rbp - (WK_NUM - (i)) * SIZEOF_YMMWORD  ; ymmword wk[WK_NUM]
%define WK_NUM  3

    align       32
    GLOBAL_FUNCTION(jsimd_h2v1_merged_upsample_avx2)

EXTN(jsimd_h2v1_merged_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)]
    collect_args 4
    push        rbx

    mov         ecx, r10d               ; col
    test        rcx, rcx
    jz          near .return

    push        rcx

    mov         rdi, r11
    mov         ecx, r12d
    mov         rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
    mov         rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
    mov         rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
    mov         rdi, r13
    mov         rsip, JSAMPROW [rsi+rcx*SIZEOF_JSAMPROW]  ; inptr0
    mov         rbxp, JSAMPROW [rbx+rcx*SIZEOF_JSAMPROW]  ; inptr1
    mov         rdxp, JSAMPROW [rdx+rcx*SIZEOF_JSAMPROW]  ; inptr2
    mov         rdip, JSAMPROW [rdi]                      ; outptr

    pop         rcx                     ; col

.columnloop:

    vmovdqu     ymm6, YMMWORD [rbx]     ; ymm6=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV)
    vmovdqu     ymm7, YMMWORD [rdx]     ; ymm7=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV)

    vpxor       ymm1, ymm1, ymm1        ; ymm1=(all 0's)
    vpcmpeqw    ymm3, ymm3, ymm3
    vpsllw      ymm3, ymm3, 7           ; ymm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}

    vpermq      ymm6, ymm6, 0xd8        ; ymm6=Cb(01234567GHIJKLMN89ABCDEFOPQRSTUV)
    vpermq      ymm7, ymm7, 0xd8        ; ymm7=Cr(01234567GHIJKLMN89ABCDEFOPQRSTUV)
    vpunpcklbw  ymm4, ymm6, ymm1        ; ymm4=Cb(0123456789ABCDEF)=CbL
    vpunpckhbw  ymm6, ymm6, ymm1        ; ymm6=Cb(GHIJKLMNOPQRSTUV)=CbH
    vpunpcklbw  ymm0, ymm7, ymm1        ; ymm0=Cr(0123456789ABCDEF)=CrL
    vpunpckhbw  ymm7, ymm7, ymm1        ; ymm7=Cr(GHIJKLMNOPQRSTUV)=CrH

    vpaddw      ymm5, ymm6, ymm3
    vpaddw      ymm2, ymm4, ymm3
    vpaddw      ymm1, ymm7, ymm3
    vpaddw      ymm3, ymm0, ymm3

    ; (Original)
    ; R = Y                + 1.40200 * Cr
    ; G = Y - 0.34414 * Cb - 0.71414 * Cr
    ; B = Y + 1.77200 * Cb
    ;
    ; (This implementation)
    ; R = Y                + 0.40200 * Cr + Cr
    ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
    ; B = Y - 0.22800 * Cb + Cb + Cb

    vpaddw      ymm6, ymm5, ymm5             ; ymm6=2*CbH
    vpaddw      ymm4, ymm2, ymm2             ; ymm4=2*CbL
    vpaddw      ymm7, ymm1, ymm1             ; ymm7=2*CrH
    vpaddw      ymm0, ymm3, ymm3             ; ymm0=2*CrL

    vpmulhw     ymm6, ymm6, [rel PW_MF0228]  ; ymm6=(2*CbH * -FIX(0.22800))
    vpmulhw     ymm4, ymm4, [rel PW_MF0228]  ; ymm4=(2*CbL * -FIX(0.22800))
    vpmulhw     ymm7, ymm7, [rel PW_F0402]   ; ymm7=(2*CrH * FIX(0.40200))
    vpmulhw     ymm0, ymm0, [rel PW_F0402]   ; ymm0=(2*CrL * FIX(0.40200))

    vpaddw      ymm6, ymm6, [rel PW_ONE]
    vpaddw      ymm4, ymm4, [rel PW_ONE]
    vpsraw      ymm6, ymm6, 1                ; ymm6=(CbH * -FIX(0.22800))
    vpsraw      ymm4, ymm4, 1                ; ymm4=(CbL * -FIX(0.22800))
    vpaddw      ymm7, ymm7, [rel PW_ONE]
    vpaddw      ymm0, ymm0, [rel PW_ONE]
    vpsraw      ymm7, ymm7, 1                ; ymm7=(CrH * FIX(0.40200))
    vpsraw      ymm0, ymm0, 1                ; ymm0=(CrL * FIX(0.40200))

    vpaddw      ymm6, ymm6, ymm5
    vpaddw      ymm4, ymm4, ymm2
    vpaddw      ymm6, ymm6, ymm5             ; ymm6=(CbH * FIX(1.77200))=(B-Y)H
    vpaddw      ymm4, ymm4, ymm2             ; ymm4=(CbL * FIX(1.77200))=(B-Y)L
    vpaddw      ymm7, ymm7, ymm1             ; ymm7=(CrH * FIX(1.40200))=(R-Y)H
    vpaddw      ymm0, ymm0, ymm3             ; ymm0=(CrL * FIX(1.40200))=(R-Y)L

    vmovdqa     YMMWORD [wk(0)], ymm6        ; wk(0)=(B-Y)H
    vmovdqa     YMMWORD [wk(1)], ymm7        ; wk(1)=(R-Y)H

    vpunpckhwd  ymm6, ymm5, ymm1
    vpunpcklwd  ymm5, ymm5, ymm1
    vpmaddwd    ymm5, ymm5, [rel PW_MF0344_F0285]
    vpmaddwd    ymm6, ymm6, [rel PW_MF0344_F0285]
    vpunpckhwd  ymm7, ymm2, ymm3
    vpunpcklwd  ymm2, ymm2, ymm3
    vpmaddwd    ymm2, ymm2, [rel PW_MF0344_F0285]
    vpmaddwd    ymm7, ymm7, [rel PW_MF0344_F0285]

    vpaddd      ymm5, ymm5, [rel PD_ONEHALF]
    vpaddd      ymm6, ymm6, [rel PD_ONEHALF]
    vpsrad      ymm5, ymm5, SCALEBITS
    vpsrad      ymm6, ymm6, SCALEBITS
    vpaddd      ymm2, ymm2, [rel PD_ONEHALF]
    vpaddd      ymm7, ymm7, [rel PD_ONEHALF]
    vpsrad      ymm2, ymm2, SCALEBITS
    vpsrad      ymm7, ymm7, SCALEBITS

    vpackssdw   ymm5, ymm5, ymm6        ; ymm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
    vpackssdw   ymm2, ymm2, ymm7        ; ymm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
    vpsubw      ymm5, ymm5, ymm1        ; ymm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
    vpsubw      ymm2, ymm2, ymm3        ; ymm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L

    vmovdqa     YMMWORD [wk(2)], ymm5   ; wk(2)=(G-Y)H

    mov         al, 2                   ; Yctr
    jmp         short .Yloop_1st

.Yloop_2nd:
    vmovdqa     ymm0, YMMWORD [wk(1)]   ; ymm0=(R-Y)H
    vmovdqa     ymm2, YMMWORD [wk(2)]   ; ymm2=(G-Y)H
    vmovdqa     ymm4, YMMWORD [wk(0)]   ; ymm4=(B-Y)H

.Yloop_1st:
    vmovdqu     ymm7, YMMWORD [rsi]     ; ymm7=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV)

    vpcmpeqw    ymm6, ymm6, ymm6
    vpsrlw      ymm6, ymm6, BYTE_BIT    ; ymm6={0xFF 0x00 0xFF 0x00 ..}
    vpand       ymm6, ymm6, ymm7        ; ymm6=Y(02468ACEGIKMOQSU)=YE
    vpsrlw      ymm7, ymm7, BYTE_BIT    ; ymm7=Y(13579BDFHJLNPRTV)=YO

    vmovdqa     ymm1, ymm0              ; ymm1=ymm0=(R-Y)(L/H)
    vmovdqa     ymm3, ymm2              ; ymm3=ymm2=(G-Y)(L/H)
    vmovdqa     ymm5, ymm4              ; ymm5=ymm4=(B-Y)(L/H)

    vpaddw      ymm0, ymm0, ymm6        ; ymm0=((R-Y)+YE)=RE=R(02468ACEGIKMOQSU)
    vpaddw      ymm1, ymm1, ymm7        ; ymm1=((R-Y)+YO)=RO=R(13579BDFHJLNPRTV)
    vpackuswb   ymm0, ymm0, ymm0        ; ymm0=R(02468ACE********GIKMOQSU********)
    vpackuswb   ymm1, ymm1, ymm1        ; ymm1=R(13579BDF********HJLNPRTV********)

    vpaddw      ymm2, ymm2, ymm6        ; ymm2=((G-Y)+YE)=GE=G(02468ACEGIKMOQSU)
    vpaddw      ymm3, ymm3, ymm7        ; ymm3=((G-Y)+YO)=GO=G(13579BDFHJLNPRTV)
    vpackuswb   ymm2, ymm2, ymm2        ; ymm2=G(02468ACE********GIKMOQSU********)
    vpackuswb   ymm3, ymm3, ymm3        ; ymm3=G(13579BDF********HJLNPRTV********)

    vpaddw      ymm4, ymm4, ymm6        ; ymm4=((B-Y)+YE)=BE=B(02468ACEGIKMOQSU)
    vpaddw      ymm5, ymm5, ymm7        ; ymm5=((B-Y)+YO)=BO=B(13579BDFHJLNPRTV)
    vpackuswb   ymm4, ymm4, ymm4        ; ymm4=B(02468ACE********GIKMOQSU********)
    vpackuswb   ymm5, ymm5, ymm5        ; ymm5=B(13579BDF********HJLNPRTV********)

%if RGB_PIXELSIZE == 3  ; ---------------

    ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
    ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
    ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
    ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
    ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
    ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
    ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
    ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)

    vpunpcklbw  ymmA, ymmA, ymmC        ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
                                        ;       0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
    vpunpcklbw  ymmE, ymmE, ymmB        ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F
                                        ;       2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V)
    vpunpcklbw  ymmD, ymmD, ymmF        ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F
                                        ;       1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V)

    vpsrldq     ymmH, ymmA, 2           ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G
                                        ;       0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --)
    vpunpckhwd  ymmG, ymmA, ymmE        ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F
                                        ;       0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V)
    vpunpcklwd  ymmA, ymmA, ymmE        ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07
                                        ;       0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N)

    vpsrldq     ymmE, ymmE, 2           ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H
                                        ;       2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --)

    vpsrldq     ymmB, ymmD, 2           ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H
                                        ;       1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --)
    vpunpckhwd  ymmC, ymmD, ymmH        ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G
                                        ;       1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --)
    vpunpcklwd  ymmD, ymmD, ymmH        ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18
                                        ;       1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O)

    vpunpckhwd  ymmF, ymmE, ymmB        ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H
                                        ;       2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --)
    vpunpcklwd  ymmE, ymmE, ymmB        ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29
                                        ;       2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P)

    vpshufd     ymmH, ymmA, 0x4E        ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03
                                        ;       0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J)
    vpunpckldq  ymmA, ymmA, ymmD        ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14
                                        ;       0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K)
    vpunpckhdq  ymmD, ymmD, ymmE        ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29
                                        ;       1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P)
    vpunpckldq  ymmE, ymmE, ymmH        ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07
                                        ;       2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N)

    vpshufd     ymmH, ymmG, 0x4E        ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B
                                        ;       0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R)
    vpunpckldq  ymmG, ymmG, ymmC        ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C
                                        ;       0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S)
    vpunpckhdq  ymmC, ymmC, ymmF        ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H
                                        ;       1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --)
    vpunpckldq  ymmF, ymmF, ymmH        ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F
                                        ;       2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V)

    vpunpcklqdq ymmH, ymmA, ymmE        ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
                                        ;       0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
    vpunpcklqdq ymmG, ymmD, ymmG        ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
                                        ;       1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
    vpunpcklqdq ymmC, ymmF, ymmC        ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
                                        ;       2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)

    vperm2i128  ymmA, ymmH, ymmG, 0x20  ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
                                        ;       15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
    vperm2i128  ymmD, ymmC, ymmH, 0x30  ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
                                        ;       0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
    vperm2i128  ymmF, ymmG, ymmC, 0x31  ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
                                        ;       2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)

    cmp         rcx, byte SIZEOF_YMMWORD
    jb          short .column_st64

    test        rdi, SIZEOF_YMMWORD-1
    jnz         short .out1
    ; --(aligned)-------------------
    vmovntdq    YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
    vmovntdq    YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
    vmovntdq    YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF
    jmp         short .out0
.out1:  ; --(unaligned)-----------------
    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
    vmovdqu     YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF
.out0:
    add         rdi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD  ; outptr
    sub         rcx, byte SIZEOF_YMMWORD
    jz          near .endcolumn

    add         rsi, byte SIZEOF_YMMWORD  ; inptr0
    dec         al                        ; Yctr
    jnz         near .Yloop_2nd

    add         rbx, byte SIZEOF_YMMWORD  ; inptr1
    add         rdx, byte SIZEOF_YMMWORD  ; inptr2
    jmp         near .columnloop

.column_st64:
    lea         rcx, [rcx+rcx*2]            ; imul ecx, RGB_PIXELSIZE
    cmp         rcx, byte 2*SIZEOF_YMMWORD
    jb          short .column_st32
    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
    add         rdi, byte 2*SIZEOF_YMMWORD  ; outptr
    vmovdqa     ymmA, ymmF
    sub         rcx, byte 2*SIZEOF_YMMWORD
    jmp         short .column_st31
.column_st32:
    cmp         rcx, byte SIZEOF_YMMWORD
    jb          short .column_st31
    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
    add         rdi, byte SIZEOF_YMMWORD    ; outptr
    vmovdqa     ymmA, ymmD
    sub         rcx, byte SIZEOF_YMMWORD
    jmp         short .column_st31
.column_st31:
    cmp         rcx, byte SIZEOF_XMMWORD
    jb          short .column_st15
    vmovdqu     XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    add         rdi, byte SIZEOF_XMMWORD    ; outptr
    vperm2i128  ymmA, ymmA, ymmA, 1
    sub         rcx, byte SIZEOF_XMMWORD
.column_st15:
    ; Store the lower 8 bytes of xmmA to the output when it has enough
    ; space.
    cmp         rcx, byte SIZEOF_MMWORD
    jb          short .column_st7
    vmovq       XMM_MMWORD [rdi], xmmA
    add         rdi, byte SIZEOF_MMWORD
    sub         rcx, byte SIZEOF_MMWORD
    vpsrldq     xmmA, xmmA, SIZEOF_MMWORD
.column_st7:
    ; Store the lower 4 bytes of xmmA to the output when it has enough
    ; space.
    cmp         rcx, byte SIZEOF_DWORD
    jb          short .column_st3
    vmovd       XMM_DWORD [rdi], xmmA
    add         rdi, byte SIZEOF_DWORD
    sub         rcx, byte SIZEOF_DWORD
    vpsrldq     xmmA, xmmA, SIZEOF_DWORD
.column_st3:
    ; Store the lower 2 bytes of rax to the output when it has enough
    ; space.
    vmovd       eax, xmmA
    cmp         rcx, byte SIZEOF_WORD
    jb          short .column_st1
    mov         word [rdi], ax
    add         rdi, byte SIZEOF_WORD
    sub         rcx, byte SIZEOF_WORD
    shr         rax, 16
.column_st1:
    ; Store the lower 1 byte of rax to the output when it has enough
    ; space.
    test        rcx, rcx
    jz          short .endcolumn
    mov         byte [rdi], al

%else  ; RGB_PIXELSIZE == 4 ; -----------

%ifdef RGBX_FILLER_0XFF
    vpcmpeqb    ymm6, ymm6, ymm6        ; ymm6=XE=X(02468ACE********GIKMOQSU********)
    vpcmpeqb    ymm7, ymm7, ymm7        ; ymm7=XO=X(13579BDF********HJLNPRTV********)
%else
    vpxor       ymm6, ymm6, ymm6        ; ymm6=XE=X(02468ACE********GIKMOQSU********)
    vpxor       ymm7, ymm7, ymm7        ; ymm7=XO=X(13579BDF********HJLNPRTV********)
%endif
    ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
    ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
    ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
    ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
    ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
    ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
    ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **)
    ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **)

    vpunpcklbw  ymmA, ymmA, ymmC        ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
                                        ;       0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
    vpunpcklbw  ymmE, ymmE, ymmG        ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E
                                        ;       2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U)
    vpunpcklbw  ymmB, ymmB, ymmD        ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F
                                        ;       0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V)
    vpunpcklbw  ymmF, ymmF, ymmH        ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F
                                        ;       2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V)

    vpunpckhwd  ymmC, ymmA, ymmE        ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E
                                        ;       0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U)
    vpunpcklwd  ymmA, ymmA, ymmE        ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36
                                        ;       0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M)
    vpunpckhwd  ymmG, ymmB, ymmF        ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F
                                        ;       0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V)
    vpunpcklwd  ymmB, ymmB, ymmF        ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37
                                        ;       0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N)

    vpunpckhdq  ymmE, ymmA, ymmB        ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
                                        ;       0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
    vpunpckldq  ymmB, ymmA, ymmB        ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
                                        ;       0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
    vpunpckhdq  ymmF, ymmC, ymmG        ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
                                        ;       0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
    vpunpckldq  ymmG, ymmC, ymmG        ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
                                        ;       0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)

    vperm2i128  ymmA, ymmB, ymmE, 0x20  ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
                                        ;       04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
    vperm2i128  ymmD, ymmG, ymmF, 0x20  ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
                                        ;       0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
    vperm2i128  ymmC, ymmB, ymmE, 0x31  ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
                                        ;       0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
    vperm2i128  ymmH, ymmG, ymmF, 0x31  ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
                                        ;       0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)

    cmp         rcx, byte SIZEOF_YMMWORD
    jb          short .column_st64

    test        rdi, SIZEOF_YMMWORD-1
    jnz         short .out1
    ; --(aligned)-------------------
    vmovntdq    YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
    vmovntdq    YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
    vmovntdq    YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC
    vmovntdq    YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH
    jmp         short .out0
.out1:  ; --(unaligned)-----------------
    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
    vmovdqu     YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC
    vmovdqu     YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH
.out0:
    add         rdi, RGB_PIXELSIZE*SIZEOF_YMMWORD  ; outptr
    sub         rcx, byte SIZEOF_YMMWORD
    jz          near .endcolumn

    add         rsi, byte SIZEOF_YMMWORD  ; inptr0
    dec         al
    jnz         near .Yloop_2nd

    add         rbx, byte SIZEOF_YMMWORD  ; inptr1
    add         rdx, byte SIZEOF_YMMWORD  ; inptr2
    jmp         near .columnloop

.column_st64:
    cmp         rcx, byte SIZEOF_YMMWORD/2
    jb          short .column_st32
    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
    add         rdi, byte 2*SIZEOF_YMMWORD  ; outptr
    vmovdqa     ymmA, ymmC
    vmovdqa     ymmD, ymmH
    sub         rcx, byte SIZEOF_YMMWORD/2
.column_st32:
    cmp         rcx, byte SIZEOF_YMMWORD/4
    jb          short .column_st16
    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
    add         rdi, byte SIZEOF_YMMWORD    ; outptr
    vmovdqa     ymmA, ymmD
    sub         rcx, byte SIZEOF_YMMWORD/4
.column_st16:
    cmp         rcx, byte SIZEOF_YMMWORD/8
    jb          short .column_st15
    vmovdqu     XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    add         rdi, byte SIZEOF_XMMWORD    ; outptr
    vperm2i128  ymmA, ymmA, ymmA, 1
    sub         rcx, byte SIZEOF_YMMWORD/8
.column_st15:
    ; Store two pixels (8 bytes) of ymmA to the output when it has enough
    ; space.
    cmp         rcx, byte SIZEOF_YMMWORD/16
    jb          short .column_st7
    vmovq       MMWORD [rdi], xmmA
    add         rdi, byte SIZEOF_YMMWORD/16*4
    sub         rcx, byte SIZEOF_YMMWORD/16
    vpsrldq     xmmA, SIZEOF_YMMWORD/16*4
.column_st7:
    ; Store one pixel (4 bytes) of ymmA to the output when it has enough
    ; space.
    test        rcx, rcx
    jz          short .endcolumn
    vmovd       XMM_DWORD [rdi], xmmA

%endif  ; RGB_PIXELSIZE ; ---------------

.endcolumn:
    sfence                              ; flush the write buffer

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

; --------------------------------------------------------------------------
;
; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
;
; GLOBAL(void)
; jsimd_h2v2_merged_upsample_avx2(JDIMENSION output_width,
;                                 JSAMPIMAGE input_buf,
;                                 JDIMENSION in_row_group_ctr,
;                                 JSAMPARRAY output_buf);
;

; r10d = JDIMENSION output_width
; r11 = JSAMPIMAGE input_buf
; r12d = JDIMENSION in_row_group_ctr
; r13 = JSAMPARRAY output_buf

    align       32
    GLOBAL_FUNCTION(jsimd_h2v2_merged_upsample_avx2)

EXTN(jsimd_h2v2_merged_upsample_avx2):
    push        rbp
    mov         rax, rsp
    mov         rbp, rsp
    collect_args 4
    push        rbx

    mov         eax, r10d

    mov         rdi, r11
    mov         ecx, r12d
    mov         rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
    mov         rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
    mov         rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
    mov         rdi, r13
    lea         rsi, [rsi+rcx*SIZEOF_JSAMPROW]

    sub         rsp, SIZEOF_JSAMPARRAY*4
    mov         JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY], rsip  ; intpr00
    mov         JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY], rbxp  ; intpr1
    mov         JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY], rdxp  ; intpr2
    mov         rbx, rsp

    push        rdi
    push        rcx
    push        rax

    %ifdef WIN64
    mov         r8, rcx
    mov         r9, rdi
    mov         rcx, rax
    mov         rdx, rbx
    %else
    mov         rdx, rcx
    mov         rcx, rdi
    mov         rdi, rax
    mov         rsi, rbx
    %endif

    call        EXTN(jsimd_h2v1_merged_upsample_avx2)

    pop         rax
    pop         rcx
    pop         rdi
    mov         rsip, JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY]
    mov         rbxp, JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY]
    mov         rdxp, JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY]

    add         rdi, byte SIZEOF_JSAMPROW  ; outptr1
    add         rsi, byte SIZEOF_JSAMPROW  ; inptr01

    mov         JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY], rsip  ; intpr00
    mov         JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY], rbxp  ; intpr1
    mov         JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY], rdxp  ; intpr2
    mov         rbx, rsp

    push        rdi
    push        rcx
    push        rax

    %ifdef WIN64
    mov         r8, rcx
    mov         r9, rdi
    mov         rcx, rax
    mov         rdx, rbx
    %else
    mov         rdx, rcx
    mov         rcx, rdi
    mov         rdi, rax
    mov         rsi, rbx
    %endif

    call        EXTN(jsimd_h2v1_merged_upsample_avx2)

    pop         rax
    pop         rcx
    pop         rdi
    mov         rsip, JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY]
    mov         rbxp, JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY]
    mov         rdxp, JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY]
    add         rsp, SIZEOF_JSAMPARRAY*4

    pop         rbx
    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