summaryrefslogtreecommitdiff
path: root/media/libjpeg/simd/x86_64/jdcolext-sse2.asm
blob: e07c8d75188c7f9ddf447f3048c6f1b0e2018557 (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
;
; jdcolext.asm - colorspace conversion (64-bit SSE2)
;
; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright (C) 2009, 2012, 2016, D. R. Commander.
; 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"

; --------------------------------------------------------------------------
;
; Convert some rows of samples to the output colorspace.
;
; GLOBAL(void)
; jsimd_ycc_rgb_convert_sse2(JDIMENSION out_width, JSAMPIMAGE input_buf,
;                            JDIMENSION input_row, JSAMPARRAY output_buf,
;                            int num_rows)
;

; r10d = JDIMENSION out_width
; r11 = JSAMPIMAGE input_buf
; r12d = JDIMENSION input_row
; r13 = JSAMPARRAY output_buf
; r14d = int num_rows

%define wk(i)   rbp - (WK_NUM - (i)) * SIZEOF_XMMWORD  ; xmmword wk[WK_NUM]
%define WK_NUM  2

    align       32
    GLOBAL_FUNCTION(jsimd_ycc_rgb_convert_sse2)

EXTN(jsimd_ycc_rgb_convert_sse2):
    push        rbp
    mov         rax, rsp                     ; rax = original rbp
    sub         rsp, byte 4
    and         rsp, byte (-SIZEOF_XMMWORD)  ; align to 128 bits
    mov         [rsp], rax
    mov         rbp, rsp                     ; rbp = aligned rbp
    lea         rsp, [wk(0)]
    collect_args 5
    push        rbx

    mov         ecx, r10d               ; num_cols
    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]
    lea         rsi, [rsi+rcx*SIZEOF_JSAMPROW]
    lea         rbx, [rbx+rcx*SIZEOF_JSAMPROW]
    lea         rdx, [rdx+rcx*SIZEOF_JSAMPROW]

    pop         rcx

    mov         rdi, r13
    mov         eax, r14d
    test        rax, rax
    jle         near .return
.rowloop:
    push        rax
    push        rdi
    push        rdx
    push        rbx
    push        rsi
    push        rcx                     ; col

    mov         rsip, JSAMPROW [rsi]    ; inptr0
    mov         rbxp, JSAMPROW [rbx]    ; inptr1
    mov         rdxp, JSAMPROW [rdx]    ; inptr2
    mov         rdip, JSAMPROW [rdi]    ; outptr
.columnloop:

    movdqa      xmm5, XMMWORD [rbx]     ; xmm5=Cb(0123456789ABCDEF)
    movdqa      xmm1, XMMWORD [rdx]     ; xmm1=Cr(0123456789ABCDEF)

    pcmpeqw     xmm4, xmm4
    pcmpeqw     xmm7, xmm7
    psrlw       xmm4, BYTE_BIT
    psllw       xmm7, 7                 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
    movdqa      xmm0, xmm4              ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..}

    pand        xmm4, xmm5              ; xmm4=Cb(02468ACE)=CbE
    psrlw       xmm5, BYTE_BIT          ; xmm5=Cb(13579BDF)=CbO
    pand        xmm0, xmm1              ; xmm0=Cr(02468ACE)=CrE
    psrlw       xmm1, BYTE_BIT          ; xmm1=Cr(13579BDF)=CrO

    paddw       xmm4, xmm7
    paddw       xmm5, xmm7
    paddw       xmm0, xmm7
    paddw       xmm1, xmm7

    ; (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

    movdqa      xmm2, xmm4              ; xmm2=CbE
    movdqa      xmm3, xmm5              ; xmm3=CbO
    paddw       xmm4, xmm4              ; xmm4=2*CbE
    paddw       xmm5, xmm5              ; xmm5=2*CbO
    movdqa      xmm6, xmm0              ; xmm6=CrE
    movdqa      xmm7, xmm1              ; xmm7=CrO
    paddw       xmm0, xmm0              ; xmm0=2*CrE
    paddw       xmm1, xmm1              ; xmm1=2*CrO

    pmulhw      xmm4, [rel PW_MF0228]   ; xmm4=(2*CbE * -FIX(0.22800))
    pmulhw      xmm5, [rel PW_MF0228]   ; xmm5=(2*CbO * -FIX(0.22800))
    pmulhw      xmm0, [rel PW_F0402]    ; xmm0=(2*CrE * FIX(0.40200))
    pmulhw      xmm1, [rel PW_F0402]    ; xmm1=(2*CrO * FIX(0.40200))

    paddw       xmm4, [rel PW_ONE]
    paddw       xmm5, [rel PW_ONE]
    psraw       xmm4, 1                 ; xmm4=(CbE * -FIX(0.22800))
    psraw       xmm5, 1                 ; xmm5=(CbO * -FIX(0.22800))
    paddw       xmm0, [rel PW_ONE]
    paddw       xmm1, [rel PW_ONE]
    psraw       xmm0, 1                 ; xmm0=(CrE * FIX(0.40200))
    psraw       xmm1, 1                 ; xmm1=(CrO * FIX(0.40200))

    paddw       xmm4, xmm2
    paddw       xmm5, xmm3
    paddw       xmm4, xmm2              ; xmm4=(CbE * FIX(1.77200))=(B-Y)E
    paddw       xmm5, xmm3              ; xmm5=(CbO * FIX(1.77200))=(B-Y)O
    paddw       xmm0, xmm6              ; xmm0=(CrE * FIX(1.40200))=(R-Y)E
    paddw       xmm1, xmm7              ; xmm1=(CrO * FIX(1.40200))=(R-Y)O

    movdqa      XMMWORD [wk(0)], xmm4   ; wk(0)=(B-Y)E
    movdqa      XMMWORD [wk(1)], xmm5   ; wk(1)=(B-Y)O

    movdqa      xmm4, xmm2
    movdqa      xmm5, xmm3
    punpcklwd   xmm2, xmm6
    punpckhwd   xmm4, xmm6
    pmaddwd     xmm2, [rel PW_MF0344_F0285]
    pmaddwd     xmm4, [rel PW_MF0344_F0285]
    punpcklwd   xmm3, xmm7
    punpckhwd   xmm5, xmm7
    pmaddwd     xmm3, [rel PW_MF0344_F0285]
    pmaddwd     xmm5, [rel PW_MF0344_F0285]

    paddd       xmm2, [rel PD_ONEHALF]
    paddd       xmm4, [rel PD_ONEHALF]
    psrad       xmm2, SCALEBITS
    psrad       xmm4, SCALEBITS
    paddd       xmm3, [rel PD_ONEHALF]
    paddd       xmm5, [rel PD_ONEHALF]
    psrad       xmm3, SCALEBITS
    psrad       xmm5, SCALEBITS

    packssdw    xmm2, xmm4              ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
    packssdw    xmm3, xmm5              ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
    psubw       xmm2, xmm6              ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
    psubw       xmm3, xmm7              ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O

    movdqa      xmm5, XMMWORD [rsi]     ; xmm5=Y(0123456789ABCDEF)

    pcmpeqw     xmm4, xmm4
    psrlw       xmm4, BYTE_BIT          ; xmm4={0xFF 0x00 0xFF 0x00 ..}
    pand        xmm4, xmm5              ; xmm4=Y(02468ACE)=YE
    psrlw       xmm5, BYTE_BIT          ; xmm5=Y(13579BDF)=YO

    paddw       xmm0, xmm4              ; xmm0=((R-Y)E+YE)=RE=R(02468ACE)
    paddw       xmm1, xmm5              ; xmm1=((R-Y)O+YO)=RO=R(13579BDF)
    packuswb    xmm0, xmm0              ; xmm0=R(02468ACE********)
    packuswb    xmm1, xmm1              ; xmm1=R(13579BDF********)

    paddw       xmm2, xmm4              ; xmm2=((G-Y)E+YE)=GE=G(02468ACE)
    paddw       xmm3, xmm5              ; xmm3=((G-Y)O+YO)=GO=G(13579BDF)
    packuswb    xmm2, xmm2              ; xmm2=G(02468ACE********)
    packuswb    xmm3, xmm3              ; xmm3=G(13579BDF********)

    paddw       xmm4, XMMWORD [wk(0)]   ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE)
    paddw       xmm5, XMMWORD [wk(1)]   ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF)
    packuswb    xmm4, xmm4              ; xmm4=B(02468ACE********)
    packuswb    xmm5, xmm5              ; xmm5=B(13579BDF********)

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

    ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
    ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
    ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
    ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)

    punpcklbw   xmmA, xmmC        ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
    punpcklbw   xmmE, xmmB        ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
    punpcklbw   xmmD, xmmF        ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)

    movdqa      xmmG, xmmA
    movdqa      xmmH, xmmA
    punpcklwd   xmmA, xmmE        ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
    punpckhwd   xmmG, xmmE        ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)

    psrldq      xmmH, 2           ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
    psrldq      xmmE, 2           ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)

    movdqa      xmmC, xmmD
    movdqa      xmmB, xmmD
    punpcklwd   xmmD, xmmH        ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
    punpckhwd   xmmC, xmmH        ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)

    psrldq      xmmB, 2           ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)

    movdqa      xmmF, xmmE
    punpcklwd   xmmE, xmmB        ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
    punpckhwd   xmmF, xmmB        ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)

    pshufd      xmmH, xmmA, 0x4E  ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
    movdqa      xmmB, xmmE
    punpckldq   xmmA, xmmD        ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
    punpckldq   xmmE, xmmH        ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
    punpckhdq   xmmD, xmmB        ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)

    pshufd      xmmH, xmmG, 0x4E  ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
    movdqa      xmmB, xmmF
    punpckldq   xmmG, xmmC        ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
    punpckldq   xmmF, xmmH        ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
    punpckhdq   xmmC, xmmB        ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)

    punpcklqdq  xmmA, xmmE        ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
    punpcklqdq  xmmD, xmmG        ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
    punpcklqdq  xmmF, xmmC        ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)

    cmp         rcx, byte SIZEOF_XMMWORD
    jb          short .column_st32

    test        rdi, SIZEOF_XMMWORD-1
    jnz         short .out1
    ; --(aligned)-------------------
    movntdq     XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    movntdq     XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
    movntdq     XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
    jmp         short .out0
.out1:  ; --(unaligned)-----------------
    movdqu      XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    movdqu      XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
    movdqu      XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
.out0:
    add         rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD  ; outptr
    sub         rcx, byte SIZEOF_XMMWORD
    jz          near .nextrow

    add         rsi, byte SIZEOF_XMMWORD  ; inptr0
    add         rbx, byte SIZEOF_XMMWORD  ; inptr1
    add         rdx, byte SIZEOF_XMMWORD  ; inptr2
    jmp         near .columnloop

.column_st32:
    lea         rcx, [rcx+rcx*2]            ; imul ecx, RGB_PIXELSIZE
    cmp         rcx, byte 2*SIZEOF_XMMWORD
    jb          short .column_st16
    movdqu      XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    movdqu      XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
    add         rdi, byte 2*SIZEOF_XMMWORD  ; outptr
    movdqa      xmmA, xmmF
    sub         rcx, byte 2*SIZEOF_XMMWORD
    jmp         short .column_st15
.column_st16:
    cmp         rcx, byte SIZEOF_XMMWORD
    jb          short .column_st15
    movdqu      XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    add         rdi, byte SIZEOF_XMMWORD    ; outptr
    movdqa      xmmA, xmmD
    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
    movq        XMM_MMWORD [rdi], xmmA
    add         rdi, byte SIZEOF_MMWORD
    sub         rcx, byte SIZEOF_MMWORD
    psrldq      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
    movd        XMM_DWORD [rdi], xmmA
    add         rdi, byte SIZEOF_DWORD
    sub         rcx, byte SIZEOF_DWORD
    psrldq      xmmA, SIZEOF_DWORD
.column_st3:
    ; Store the lower 2 bytes of rax to the output when it has enough
    ; space.
    movd        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 .nextrow
    mov         byte [rdi], al

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

%ifdef RGBX_FILLER_0XFF
    pcmpeqb     xmm6, xmm6              ; xmm6=XE=X(02468ACE********)
    pcmpeqb     xmm7, xmm7              ; xmm7=XO=X(13579BDF********)
%else
    pxor        xmm6, xmm6              ; xmm6=XE=X(02468ACE********)
    pxor        xmm7, xmm7              ; xmm7=XO=X(13579BDF********)
%endif
    ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
    ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
    ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
    ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)

    punpcklbw   xmmA, xmmC  ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
    punpcklbw   xmmE, xmmG  ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
    punpcklbw   xmmB, xmmD  ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
    punpcklbw   xmmF, xmmH  ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)

    movdqa      xmmC, xmmA
    punpcklwd   xmmA, xmmE  ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
    punpckhwd   xmmC, xmmE  ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
    movdqa      xmmG, xmmB
    punpcklwd   xmmB, xmmF  ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
    punpckhwd   xmmG, xmmF  ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)

    movdqa      xmmD, xmmA
    punpckldq   xmmA, xmmB  ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
    punpckhdq   xmmD, xmmB  ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
    movdqa      xmmH, xmmC
    punpckldq   xmmC, xmmG  ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
    punpckhdq   xmmH, xmmG  ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)

    cmp         rcx, byte SIZEOF_XMMWORD
    jb          short .column_st32

    test        rdi, SIZEOF_XMMWORD-1
    jnz         short .out1
    ; --(aligned)-------------------
    movntdq     XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    movntdq     XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
    movntdq     XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
    movntdq     XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
    jmp         short .out0
.out1:  ; --(unaligned)-----------------
    movdqu      XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    movdqu      XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
    movdqu      XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
    movdqu      XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
.out0:
    add         rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD  ; outptr
    sub         rcx, byte SIZEOF_XMMWORD
    jz          near .nextrow

    add         rsi, byte SIZEOF_XMMWORD  ; inptr0
    add         rbx, byte SIZEOF_XMMWORD  ; inptr1
    add         rdx, byte SIZEOF_XMMWORD  ; inptr2
    jmp         near .columnloop

.column_st32:
    cmp         rcx, byte SIZEOF_XMMWORD/2
    jb          short .column_st16
    movdqu      XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    movdqu      XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
    add         rdi, byte 2*SIZEOF_XMMWORD  ; outptr
    movdqa      xmmA, xmmC
    movdqa      xmmD, xmmH
    sub         rcx, byte SIZEOF_XMMWORD/2
.column_st16:
    cmp         rcx, byte SIZEOF_XMMWORD/4
    jb          short .column_st15
    movdqu      XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
    add         rdi, byte SIZEOF_XMMWORD    ; outptr
    movdqa      xmmA, xmmD
    sub         rcx, byte SIZEOF_XMMWORD/4
.column_st15:
    ; Store two pixels (8 bytes) of xmmA to the output when it has enough
    ; space.
    cmp         rcx, byte SIZEOF_XMMWORD/8
    jb          short .column_st7
    movq        MMWORD [rdi], xmmA
    add         rdi, byte SIZEOF_XMMWORD/8*4
    sub         rcx, byte SIZEOF_XMMWORD/8
    psrldq      xmmA, SIZEOF_XMMWORD/8*4
.column_st7:
    ; Store one pixel (4 bytes) of xmmA to the output when it has enough
    ; space.
    test        rcx, rcx
    jz          short .nextrow
    movd        XMM_DWORD [rdi], xmmA

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

.nextrow:
    pop         rcx
    pop         rsi
    pop         rbx
    pop         rdx
    pop         rdi
    pop         rax

    add         rsi, byte SIZEOF_JSAMPROW
    add         rbx, byte SIZEOF_JSAMPROW
    add         rdx, byte SIZEOF_JSAMPROW
    add         rdi, byte SIZEOF_JSAMPROW  ; output_buf
    dec         rax                        ; num_rows
    jg          near .rowloop

    sfence                              ; flush the write buffer

.return:
    pop         rbx
    uncollect_args 5
    mov         rsp, rbp                ; rsp <- aligned rbp
    pop         rsp                     ; rsp <- original rbp
    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