summaryrefslogtreecommitdiff
path: root/media/libjpeg/simd/i386/jdsample-sse2.asm
blob: 4e28d2f4b80293c4780d4f5daeecafdbf002755d (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
;
; jdsample.asm - upsampling (SSE2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright (C) 2016, D. R. Commander.
;
; 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_sse2)

EXTN(jconst_fancy_upsample_sse2):

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

    alignz      32

; --------------------------------------------------------------------------
    SECTION     SEG_TEXT
    BITS        32
;
; 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_sse2(int max_v_samp_factor,
;                                JDIMENSION downsampled_width,
;                                JSAMPARRAY input_data,
;                                JSAMPARRAY *output_data_ptr);
;

%define max_v_samp(b)       (b) + 8     ; int max_v_samp_factor
%define downsamp_width(b)   (b) + 12    ; JDIMENSION downsampled_width
%define input_data(b)       (b) + 16    ; JSAMPARRAY input_data
%define output_data_ptr(b)  (b) + 20    ; JSAMPARRAY *output_data_ptr

    align       32
    GLOBAL_FUNCTION(jsimd_h2v1_fancy_upsample_sse2)

EXTN(jsimd_h2v1_fancy_upsample_sse2):
    push        ebp
    mov         ebp, esp
    pushpic     ebx
;   push        ecx                     ; need not be preserved
;   push        edx                     ; need not be preserved
    push        esi
    push        edi

    get_GOT     ebx                     ; get GOT address

    mov         eax, JDIMENSION [downsamp_width(ebp)]  ; colctr
    test        eax, eax
    jz          near .return

    mov         ecx, INT [max_v_samp(ebp)]  ; rowctr
    test        ecx, ecx
    jz          near .return

    mov         esi, JSAMPARRAY [input_data(ebp)]    ; input_data
    mov         edi, POINTER [output_data_ptr(ebp)]
    mov         edi, JSAMPARRAY [edi]                ; output_data
    alignx      16, 7
.rowloop:
    push        eax                     ; colctr
    push        edi
    push        esi

    mov         esi, JSAMPROW [esi]     ; inptr
    mov         edi, JSAMPROW [edi]     ; outptr

    test        eax, SIZEOF_XMMWORD-1
    jz          short .skip
    mov         dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE]
    mov         JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl    ; insert a dummy sample
.skip:
    pxor        xmm0, xmm0              ; xmm0=(all 0's)
    pcmpeqb     xmm7, xmm7
    psrldq      xmm7, (SIZEOF_XMMWORD-1)
    pand        xmm7, XMMWORD [esi+0*SIZEOF_XMMWORD]

    add         eax, byte SIZEOF_XMMWORD-1
    and         eax, byte -SIZEOF_XMMWORD
    cmp         eax, byte SIZEOF_XMMWORD
    ja          short .columnloop
    alignx      16, 7

.columnloop_last:
    pcmpeqb     xmm6, xmm6
    pslldq      xmm6, (SIZEOF_XMMWORD-1)
    pand        xmm6, XMMWORD [esi+0*SIZEOF_XMMWORD]
    jmp         short .upsample
    alignx      16, 7

.columnloop:
    movdqa      xmm6, XMMWORD [esi+1*SIZEOF_XMMWORD]
    pslldq      xmm6, (SIZEOF_XMMWORD-1)

.upsample:
    movdqa      xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
    movdqa      xmm2, xmm1
    movdqa      xmm3, xmm1                ; xmm1=( 0  1  2 ... 13 14 15)
    pslldq      xmm2, 1                   ; xmm2=(--  0  1 ... 12 13 14)
    psrldq      xmm3, 1                   ; xmm3=( 1  2  3 ... 14 15 --)

    por         xmm2, xmm7                ; xmm2=(-1  0  1 ... 12 13 14)
    por         xmm3, xmm6                ; xmm3=( 1  2  3 ... 14 15 16)

    movdqa      xmm7, xmm1
    psrldq      xmm7, (SIZEOF_XMMWORD-1)  ; xmm7=(15 -- -- ... -- -- --)

    movdqa      xmm4, xmm1
    punpcklbw   xmm1, xmm0                ; xmm1=( 0  1  2  3  4  5  6  7)
    punpckhbw   xmm4, xmm0                ; xmm4=( 8  9 10 11 12 13 14 15)
    movdqa      xmm5, xmm2
    punpcklbw   xmm2, xmm0                ; xmm2=(-1  0  1  2  3  4  5  6)
    punpckhbw   xmm5, xmm0                ; xmm5=( 7  8  9 10 11 12 13 14)
    movdqa      xmm6, xmm3
    punpcklbw   xmm3, xmm0                ; xmm3=( 1  2  3  4  5  6  7  8)
    punpckhbw   xmm6, xmm0                ; xmm6=( 9 10 11 12 13 14 15 16)

    pmullw      xmm1, [GOTOFF(ebx,PW_THREE)]
    pmullw      xmm4, [GOTOFF(ebx,PW_THREE)]
    paddw       xmm2, [GOTOFF(ebx,PW_ONE)]
    paddw       xmm5, [GOTOFF(ebx,PW_ONE)]
    paddw       xmm3, [GOTOFF(ebx,PW_TWO)]
    paddw       xmm6, [GOTOFF(ebx,PW_TWO)]

    paddw       xmm2, xmm1
    paddw       xmm5, xmm4
    psrlw       xmm2, 2                 ; xmm2=OutLE=( 0  2  4  6  8 10 12 14)
    psrlw       xmm5, 2                 ; xmm5=OutHE=(16 18 20 22 24 26 28 30)
    paddw       xmm3, xmm1
    paddw       xmm6, xmm4
    psrlw       xmm3, 2                 ; xmm3=OutLO=( 1  3  5  7  9 11 13 15)
    psrlw       xmm6, 2                 ; xmm6=OutHO=(17 19 21 23 25 27 29 31)

    psllw       xmm3, BYTE_BIT
    psllw       xmm6, BYTE_BIT
    por         xmm2, xmm3              ; xmm2=OutL=( 0  1  2 ... 13 14 15)
    por         xmm5, xmm6              ; xmm5=OutH=(16 17 18 ... 29 30 31)

    movdqa      XMMWORD [edi+0*SIZEOF_XMMWORD], xmm2
    movdqa      XMMWORD [edi+1*SIZEOF_XMMWORD], xmm5

    sub         eax, byte SIZEOF_XMMWORD
    add         esi, byte 1*SIZEOF_XMMWORD  ; inptr
    add         edi, byte 2*SIZEOF_XMMWORD  ; outptr
    cmp         eax, byte SIZEOF_XMMWORD
    ja          near .columnloop
    test        eax, eax
    jnz         near .columnloop_last

    pop         esi
    pop         edi
    pop         eax

    add         esi, byte SIZEOF_JSAMPROW  ; input_data
    add         edi, byte SIZEOF_JSAMPROW  ; output_data
    dec         ecx                        ; rowctr
    jg          near .rowloop

.return:
    pop         edi
    pop         esi
;   pop         edx                     ; need not be preserved
;   pop         ecx                     ; need not be preserved
    poppic      ebx
    pop         ebp
    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_sse2(int max_v_samp_factor,
;                                JDIMENSION downsampled_width,
;                                JSAMPARRAY input_data,
;                                JSAMPARRAY *output_data_ptr);
;

%define max_v_samp(b)       (b) + 8     ; int max_v_samp_factor
%define downsamp_width(b)   (b) + 12    ; JDIMENSION downsampled_width
%define input_data(b)       (b) + 16    ; JSAMPARRAY input_data
%define output_data_ptr(b)  (b) + 20    ; JSAMPARRAY *output_data_ptr

%define original_ebp  ebp + 0
%define wk(i)         ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD
                                        ; xmmword wk[WK_NUM]
%define WK_NUM        4
%define gotptr        wk(0) - SIZEOF_POINTER  ; void *gotptr

    align       32
    GLOBAL_FUNCTION(jsimd_h2v2_fancy_upsample_sse2)

EXTN(jsimd_h2v2_fancy_upsample_sse2):
    push        ebp
    mov         eax, esp                     ; eax = original ebp
    sub         esp, byte 4
    and         esp, byte (-SIZEOF_XMMWORD)  ; align to 128 bits
    mov         [esp], eax
    mov         ebp, esp                     ; ebp = aligned ebp
    lea         esp, [wk(0)]
    pushpic     eax                     ; make a room for GOT address
    push        ebx
;   push        ecx                     ; need not be preserved
;   push        edx                     ; need not be preserved
    push        esi
    push        edi

    get_GOT     ebx                     ; get GOT address
    movpic      POINTER [gotptr], ebx   ; save GOT address

    mov         edx, eax                ; edx = original ebp
    mov         eax, JDIMENSION [downsamp_width(edx)]  ; colctr
    test        eax, eax
    jz          near .return

    mov         ecx, INT [max_v_samp(edx)]  ; rowctr
    test        ecx, ecx
    jz          near .return

    mov         esi, JSAMPARRAY [input_data(edx)]    ; input_data
    mov         edi, POINTER [output_data_ptr(edx)]
    mov         edi, JSAMPARRAY [edi]                ; output_data
    alignx      16, 7
.rowloop:
    push        eax                     ; colctr
    push        ecx
    push        edi
    push        esi

    mov         ecx, JSAMPROW [esi-1*SIZEOF_JSAMPROW]  ; inptr1(above)
    mov         ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW]  ; inptr0
    mov         esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW]  ; inptr1(below)
    mov         edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]  ; outptr0
    mov         edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW]  ; outptr1

    test        eax, SIZEOF_XMMWORD-1
    jz          short .skip
    push        edx
    mov         dl, JSAMPLE [ecx+(eax-1)*SIZEOF_JSAMPLE]
    mov         JSAMPLE [ecx+eax*SIZEOF_JSAMPLE], dl
    mov         dl, JSAMPLE [ebx+(eax-1)*SIZEOF_JSAMPLE]
    mov         JSAMPLE [ebx+eax*SIZEOF_JSAMPLE], dl
    mov         dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE]
    mov         JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl    ; insert a dummy sample
    pop         edx
.skip:
    ; -- process the first column block

    movdqa      xmm0, XMMWORD [ebx+0*SIZEOF_XMMWORD]  ; xmm0=row[ 0][0]
    movdqa      xmm1, XMMWORD [ecx+0*SIZEOF_XMMWORD]  ; xmm1=row[-1][0]
    movdqa      xmm2, XMMWORD [esi+0*SIZEOF_XMMWORD]  ; xmm2=row[+1][0]

    pushpic     ebx
    movpic      ebx, POINTER [gotptr]   ; load GOT address

    pxor        xmm3, xmm3              ; xmm3=(all 0's)
    movdqa      xmm4, xmm0
    punpcklbw   xmm0, xmm3              ; xmm0=row[ 0]( 0  1  2  3  4  5  6  7)
    punpckhbw   xmm4, xmm3              ; xmm4=row[ 0]( 8  9 10 11 12 13 14 15)
    movdqa      xmm5, xmm1
    punpcklbw   xmm1, xmm3              ; xmm1=row[-1]( 0  1  2  3  4  5  6  7)
    punpckhbw   xmm5, xmm3              ; xmm5=row[-1]( 8  9 10 11 12 13 14 15)
    movdqa      xmm6, xmm2
    punpcklbw   xmm2, xmm3              ; xmm2=row[+1]( 0  1  2  3  4  5  6  7)
    punpckhbw   xmm6, xmm3              ; xmm6=row[+1]( 8  9 10 11 12 13 14 15)

    pmullw      xmm0, [GOTOFF(ebx,PW_THREE)]
    pmullw      xmm4, [GOTOFF(ebx,PW_THREE)]

    pcmpeqb     xmm7, xmm7
    psrldq      xmm7, (SIZEOF_XMMWORD-2)

    paddw       xmm1, xmm0              ; xmm1=Int0L=( 0  1  2  3  4  5  6  7)
    paddw       xmm5, xmm4              ; xmm5=Int0H=( 8  9 10 11 12 13 14 15)
    paddw       xmm2, xmm0              ; xmm2=Int1L=( 0  1  2  3  4  5  6  7)
    paddw       xmm6, xmm4              ; xmm6=Int1H=( 8  9 10 11 12 13 14 15)

    movdqa      XMMWORD [edx+0*SIZEOF_XMMWORD], xmm1  ; temporarily save
    movdqa      XMMWORD [edx+1*SIZEOF_XMMWORD], xmm5  ; the intermediate data
    movdqa      XMMWORD [edi+0*SIZEOF_XMMWORD], xmm2
    movdqa      XMMWORD [edi+1*SIZEOF_XMMWORD], xmm6

    pand        xmm1, xmm7              ; xmm1=( 0 -- -- -- -- -- -- --)
    pand        xmm2, xmm7              ; xmm2=( 0 -- -- -- -- -- -- --)

    movdqa      XMMWORD [wk(0)], xmm1
    movdqa      XMMWORD [wk(1)], xmm2

    poppic      ebx

    add         eax, byte SIZEOF_XMMWORD-1
    and         eax, byte -SIZEOF_XMMWORD
    cmp         eax, byte SIZEOF_XMMWORD
    ja          short .columnloop
    alignx      16, 7

.columnloop_last:
    ; -- process the last column block

    pushpic     ebx
    movpic      ebx, POINTER [gotptr]   ; load GOT address

    pcmpeqb     xmm1, xmm1
    pslldq      xmm1, (SIZEOF_XMMWORD-2)
    movdqa      xmm2, xmm1

    pand        xmm1, XMMWORD [edx+1*SIZEOF_XMMWORD]
    pand        xmm2, XMMWORD [edi+1*SIZEOF_XMMWORD]

    movdqa      XMMWORD [wk(2)], xmm1   ; xmm1=(-- -- -- -- -- -- -- 15)
    movdqa      XMMWORD [wk(3)], xmm2   ; xmm2=(-- -- -- -- -- -- -- 15)

    jmp         near .upsample
    alignx      16, 7

.columnloop:
    ; -- process the next column block

    movdqa      xmm0, XMMWORD [ebx+1*SIZEOF_XMMWORD]  ; xmm0=row[ 0][1]
    movdqa      xmm1, XMMWORD [ecx+1*SIZEOF_XMMWORD]  ; xmm1=row[-1][1]
    movdqa      xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD]  ; xmm2=row[+1][1]

    pushpic     ebx
    movpic      ebx, POINTER [gotptr]   ; load GOT address

    pxor        xmm3, xmm3              ; xmm3=(all 0's)
    movdqa      xmm4, xmm0
    punpcklbw   xmm0, xmm3              ; xmm0=row[ 0]( 0  1  2  3  4  5  6  7)
    punpckhbw   xmm4, xmm3              ; xmm4=row[ 0]( 8  9 10 11 12 13 14 15)
    movdqa      xmm5, xmm1
    punpcklbw   xmm1, xmm3              ; xmm1=row[-1]( 0  1  2  3  4  5  6  7)
    punpckhbw   xmm5, xmm3              ; xmm5=row[-1]( 8  9 10 11 12 13 14 15)
    movdqa      xmm6, xmm2
    punpcklbw   xmm2, xmm3              ; xmm2=row[+1]( 0  1  2  3  4  5  6  7)
    punpckhbw   xmm6, xmm3              ; xmm6=row[+1]( 8  9 10 11 12 13 14 15)

    pmullw      xmm0, [GOTOFF(ebx,PW_THREE)]
    pmullw      xmm4, [GOTOFF(ebx,PW_THREE)]

    paddw       xmm1, xmm0              ; xmm1=Int0L=( 0  1  2  3  4  5  6  7)
    paddw       xmm5, xmm4              ; xmm5=Int0H=( 8  9 10 11 12 13 14 15)
    paddw       xmm2, xmm0              ; xmm2=Int1L=( 0  1  2  3  4  5  6  7)
    paddw       xmm6, xmm4              ; xmm6=Int1H=( 8  9 10 11 12 13 14 15)

    movdqa      XMMWORD [edx+2*SIZEOF_XMMWORD], xmm1  ; temporarily save
    movdqa      XMMWORD [edx+3*SIZEOF_XMMWORD], xmm5  ; the intermediate data
    movdqa      XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
    movdqa      XMMWORD [edi+3*SIZEOF_XMMWORD], xmm6

    pslldq      xmm1, (SIZEOF_XMMWORD-2)  ; xmm1=(-- -- -- -- -- -- --  0)
    pslldq      xmm2, (SIZEOF_XMMWORD-2)  ; xmm2=(-- -- -- -- -- -- --  0)

    movdqa      XMMWORD [wk(2)], xmm1
    movdqa      XMMWORD [wk(3)], xmm2

.upsample:
    ; -- process the upper row

    movdqa      xmm7, XMMWORD [edx+0*SIZEOF_XMMWORD]
    movdqa      xmm3, XMMWORD [edx+1*SIZEOF_XMMWORD]

    movdqa      xmm0, xmm7                ; xmm7=Int0L=( 0  1  2  3  4  5  6  7)
    movdqa      xmm4, xmm3                ; xmm3=Int0H=( 8  9 10 11 12 13 14 15)
    psrldq      xmm0, 2                   ; xmm0=( 1  2  3  4  5  6  7 --)
    pslldq      xmm4, (SIZEOF_XMMWORD-2)  ; xmm4=(-- -- -- -- -- -- --  8)
    movdqa      xmm5, xmm7
    movdqa      xmm6, xmm3
    psrldq      xmm5, (SIZEOF_XMMWORD-2)  ; xmm5=( 7 -- -- -- -- -- -- --)
    pslldq      xmm6, 2                   ; xmm6=(--  8  9 10 11 12 13 14)

    por         xmm0, xmm4                ; xmm0=( 1  2  3  4  5  6  7  8)
    por         xmm5, xmm6                ; xmm5=( 7  8  9 10 11 12 13 14)

    movdqa      xmm1, xmm7
    movdqa      xmm2, xmm3
    pslldq      xmm1, 2                   ; xmm1=(--  0  1  2  3  4  5  6)
    psrldq      xmm2, 2                   ; xmm2=( 9 10 11 12 13 14 15 --)
    movdqa      xmm4, xmm3
    psrldq      xmm4, (SIZEOF_XMMWORD-2)  ; xmm4=(15 -- -- -- -- -- -- --)

    por         xmm1, XMMWORD [wk(0)]     ; xmm1=(-1  0  1  2  3  4  5  6)
    por         xmm2, XMMWORD [wk(2)]     ; xmm2=( 9 10 11 12 13 14 15 16)

    movdqa      XMMWORD [wk(0)], xmm4

    pmullw      xmm7, [GOTOFF(ebx,PW_THREE)]
    pmullw      xmm3, [GOTOFF(ebx,PW_THREE)]
    paddw       xmm1, [GOTOFF(ebx,PW_EIGHT)]
    paddw       xmm5, [GOTOFF(ebx,PW_EIGHT)]
    paddw       xmm0, [GOTOFF(ebx,PW_SEVEN)]
    paddw       xmm2, [GOTOFF(ebx,PW_SEVEN)]

    paddw       xmm1, xmm7
    paddw       xmm5, xmm3
    psrlw       xmm1, 4                 ; xmm1=Out0LE=( 0  2  4  6  8 10 12 14)
    psrlw       xmm5, 4                 ; xmm5=Out0HE=(16 18 20 22 24 26 28 30)
    paddw       xmm0, xmm7
    paddw       xmm2, xmm3
    psrlw       xmm0, 4                 ; xmm0=Out0LO=( 1  3  5  7  9 11 13 15)
    psrlw       xmm2, 4                 ; xmm2=Out0HO=(17 19 21 23 25 27 29 31)

    psllw       xmm0, BYTE_BIT
    psllw       xmm2, BYTE_BIT
    por         xmm1, xmm0              ; xmm1=Out0L=( 0  1  2 ... 13 14 15)
    por         xmm5, xmm2              ; xmm5=Out0H=(16 17 18 ... 29 30 31)

    movdqa      XMMWORD [edx+0*SIZEOF_XMMWORD], xmm1
    movdqa      XMMWORD [edx+1*SIZEOF_XMMWORD], xmm5

    ; -- process the lower row

    movdqa      xmm6, XMMWORD [edi+0*SIZEOF_XMMWORD]
    movdqa      xmm4, XMMWORD [edi+1*SIZEOF_XMMWORD]

    movdqa      xmm7, xmm6                ; xmm6=Int1L=( 0  1  2  3  4  5  6  7)
    movdqa      xmm3, xmm4                ; xmm4=Int1H=( 8  9 10 11 12 13 14 15)
    psrldq      xmm7, 2                   ; xmm7=( 1  2  3  4  5  6  7 --)
    pslldq      xmm3, (SIZEOF_XMMWORD-2)  ; xmm3=(-- -- -- -- -- -- --  8)
    movdqa      xmm0, xmm6
    movdqa      xmm2, xmm4
    psrldq      xmm0, (SIZEOF_XMMWORD-2)  ; xmm0=( 7 -- -- -- -- -- -- --)
    pslldq      xmm2, 2                   ; xmm2=(--  8  9 10 11 12 13 14)

    por         xmm7, xmm3                ; xmm7=( 1  2  3  4  5  6  7  8)
    por         xmm0, xmm2                ; xmm0=( 7  8  9 10 11 12 13 14)

    movdqa      xmm1, xmm6
    movdqa      xmm5, xmm4
    pslldq      xmm1, 2                   ; xmm1=(--  0  1  2  3  4  5  6)
    psrldq      xmm5, 2                   ; xmm5=( 9 10 11 12 13 14 15 --)
    movdqa      xmm3, xmm4
    psrldq      xmm3, (SIZEOF_XMMWORD-2)  ; xmm3=(15 -- -- -- -- -- -- --)

    por         xmm1, XMMWORD [wk(1)]     ; xmm1=(-1  0  1  2  3  4  5  6)
    por         xmm5, XMMWORD [wk(3)]     ; xmm5=( 9 10 11 12 13 14 15 16)

    movdqa      XMMWORD [wk(1)], xmm3

    pmullw      xmm6, [GOTOFF(ebx,PW_THREE)]
    pmullw      xmm4, [GOTOFF(ebx,PW_THREE)]
    paddw       xmm1, [GOTOFF(ebx,PW_EIGHT)]
    paddw       xmm0, [GOTOFF(ebx,PW_EIGHT)]
    paddw       xmm7, [GOTOFF(ebx,PW_SEVEN)]
    paddw       xmm5, [GOTOFF(ebx,PW_SEVEN)]

    paddw       xmm1, xmm6
    paddw       xmm0, xmm4
    psrlw       xmm1, 4                 ; xmm1=Out1LE=( 0  2  4  6  8 10 12 14)
    psrlw       xmm0, 4                 ; xmm0=Out1HE=(16 18 20 22 24 26 28 30)
    paddw       xmm7, xmm6
    paddw       xmm5, xmm4
    psrlw       xmm7, 4                 ; xmm7=Out1LO=( 1  3  5  7  9 11 13 15)
    psrlw       xmm5, 4                 ; xmm5=Out1HO=(17 19 21 23 25 27 29 31)

    psllw       xmm7, BYTE_BIT
    psllw       xmm5, BYTE_BIT
    por         xmm1, xmm7              ; xmm1=Out1L=( 0  1  2 ... 13 14 15)
    por         xmm0, xmm5              ; xmm0=Out1H=(16 17 18 ... 29 30 31)

    movdqa      XMMWORD [edi+0*SIZEOF_XMMWORD], xmm1
    movdqa      XMMWORD [edi+1*SIZEOF_XMMWORD], xmm0

    poppic      ebx

    sub         eax, byte SIZEOF_XMMWORD
    add         ecx, byte 1*SIZEOF_XMMWORD  ; inptr1(above)
    add         ebx, byte 1*SIZEOF_XMMWORD  ; inptr0
    add         esi, byte 1*SIZEOF_XMMWORD  ; inptr1(below)
    add         edx, byte 2*SIZEOF_XMMWORD  ; outptr0
    add         edi, byte 2*SIZEOF_XMMWORD  ; outptr1
    cmp         eax, byte SIZEOF_XMMWORD
    ja          near .columnloop
    test        eax, eax
    jnz         near .columnloop_last

    pop         esi
    pop         edi
    pop         ecx
    pop         eax

    add         esi, byte 1*SIZEOF_JSAMPROW  ; input_data
    add         edi, byte 2*SIZEOF_JSAMPROW  ; output_data
    sub         ecx, byte 2                  ; rowctr
    jg          near .rowloop

.return:
    pop         edi
    pop         esi
;   pop         edx                     ; need not be preserved
;   pop         ecx                     ; need not be preserved
    pop         ebx
    mov         esp, ebp                ; esp <- aligned ebp
    pop         esp                     ; esp <- original ebp
    pop         ebp
    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_sse2(int max_v_samp_factor, JDIMENSION output_width,
;                          JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
;

%define max_v_samp(b)       (b) + 8     ; int max_v_samp_factor
%define output_width(b)     (b) + 12    ; JDIMENSION output_width
%define input_data(b)       (b) + 16    ; JSAMPARRAY input_data
%define output_data_ptr(b)  (b) + 20    ; JSAMPARRAY *output_data_ptr

    align       32
    GLOBAL_FUNCTION(jsimd_h2v1_upsample_sse2)

EXTN(jsimd_h2v1_upsample_sse2):
    push        ebp
    mov         ebp, esp
;   push        ebx                     ; unused
;   push        ecx                     ; need not be preserved
;   push        edx                     ; need not be preserved
    push        esi
    push        edi

    mov         edx, JDIMENSION [output_width(ebp)]
    add         edx, byte (2*SIZEOF_XMMWORD)-1
    and         edx, byte -(2*SIZEOF_XMMWORD)
    jz          short .return

    mov         ecx, INT [max_v_samp(ebp)]  ; rowctr
    test        ecx, ecx
    jz          short .return

    mov         esi, JSAMPARRAY [input_data(ebp)]    ; input_data
    mov         edi, POINTER [output_data_ptr(ebp)]
    mov         edi, JSAMPARRAY [edi]                ; output_data
    alignx      16, 7
.rowloop:
    push        edi
    push        esi

    mov         esi, JSAMPROW [esi]     ; inptr
    mov         edi, JSAMPROW [edi]     ; outptr
    mov         eax, edx                ; colctr
    alignx      16, 7
.columnloop:

    movdqa      xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]

    movdqa      xmm1, xmm0
    punpcklbw   xmm0, xmm0
    punpckhbw   xmm1, xmm1

    movdqa      XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
    movdqa      XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1

    sub         eax, byte 2*SIZEOF_XMMWORD
    jz          short .nextrow

    movdqa      xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD]

    movdqa      xmm3, xmm2
    punpcklbw   xmm2, xmm2
    punpckhbw   xmm3, xmm3

    movdqa      XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
    movdqa      XMMWORD [edi+3*SIZEOF_XMMWORD], xmm3

    sub         eax, byte 2*SIZEOF_XMMWORD
    jz          short .nextrow

    add         esi, byte 2*SIZEOF_XMMWORD  ; inptr
    add         edi, byte 4*SIZEOF_XMMWORD  ; outptr
    jmp         short .columnloop
    alignx      16, 7

.nextrow:
    pop         esi
    pop         edi

    add         esi, byte SIZEOF_JSAMPROW  ; input_data
    add         edi, byte SIZEOF_JSAMPROW  ; output_data
    dec         ecx                        ; rowctr
    jg          short .rowloop

.return:
    pop         edi
    pop         esi
;   pop         edx                     ; need not be preserved
;   pop         ecx                     ; need not be preserved
;   pop         ebx                     ; unused
    pop         ebp
    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_sse2(int max_v_samp_factor, JDIMENSION output_width,
;                          JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
;

%define max_v_samp(b)       (b) + 8     ; int max_v_samp_factor
%define output_width(b)     (b) + 12    ; JDIMENSION output_width
%define input_data(b)       (b) + 16    ; JSAMPARRAY input_data
%define output_data_ptr(b)  (b) + 20    ; JSAMPARRAY *output_data_ptr

    align       32
    GLOBAL_FUNCTION(jsimd_h2v2_upsample_sse2)

EXTN(jsimd_h2v2_upsample_sse2):
    push        ebp
    mov         ebp, esp
    push        ebx
;   push        ecx                     ; need not be preserved
;   push        edx                     ; need not be preserved
    push        esi
    push        edi

    mov         edx, JDIMENSION [output_width(ebp)]
    add         edx, byte (2*SIZEOF_XMMWORD)-1
    and         edx, byte -(2*SIZEOF_XMMWORD)
    jz          near .return

    mov         ecx, INT [max_v_samp(ebp)]  ; rowctr
    test        ecx, ecx
    jz          near .return

    mov         esi, JSAMPARRAY [input_data(ebp)]    ; input_data
    mov         edi, POINTER [output_data_ptr(ebp)]
    mov         edi, JSAMPARRAY [edi]                ; output_data
    alignx      16, 7
.rowloop:
    push        edi
    push        esi

    mov         esi, JSAMPROW [esi]                    ; inptr
    mov         ebx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]  ; outptr0
    mov         edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW]  ; outptr1
    mov         eax, edx                               ; colctr
    alignx      16, 7
.columnloop:

    movdqa      xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD]

    movdqa      xmm1, xmm0
    punpcklbw   xmm0, xmm0
    punpckhbw   xmm1, xmm1

    movdqa      XMMWORD [ebx+0*SIZEOF_XMMWORD], xmm0
    movdqa      XMMWORD [ebx+1*SIZEOF_XMMWORD], xmm1
    movdqa      XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0
    movdqa      XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1

    sub         eax, byte 2*SIZEOF_XMMWORD
    jz          short .nextrow

    movdqa      xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD]

    movdqa      xmm3, xmm2
    punpcklbw   xmm2, xmm2
    punpckhbw   xmm3, xmm3

    movdqa      XMMWORD [ebx+2*SIZEOF_XMMWORD], xmm2
    movdqa      XMMWORD [ebx+3*SIZEOF_XMMWORD], xmm3
    movdqa      XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2
    movdqa      XMMWORD [edi+3*SIZEOF_XMMWORD], xmm3

    sub         eax, byte 2*SIZEOF_XMMWORD
    jz          short .nextrow

    add         esi, byte 2*SIZEOF_XMMWORD  ; inptr
    add         ebx, byte 4*SIZEOF_XMMWORD  ; outptr0
    add         edi, byte 4*SIZEOF_XMMWORD  ; outptr1
    jmp         short .columnloop
    alignx      16, 7

.nextrow:
    pop         esi
    pop         edi

    add         esi, byte 1*SIZEOF_JSAMPROW  ; input_data
    add         edi, byte 2*SIZEOF_JSAMPROW  ; output_data
    sub         ecx, byte 2                  ; rowctr
    jg          short .rowloop

.return:
    pop         edi
    pop         esi
;   pop         edx                     ; need not be preserved
;   pop         ecx                     ; need not be preserved
    pop         ebx
    pop         ebp
    ret

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