blob: c13f89e2c4baef53179f8624f1d08158651c062b [file] [log] [blame]
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +00001;
2; jccolss2.asm - colorspace conversion (SSE2)
3;
4; x86 SIMD extension for IJG JPEG library
5; Copyright (C) 1999-2006, MIYASAKA Masaru.
6; For conditions of distribution and use, see copyright notice in jsimdext.inc
7;
8; This file should be assembled with NASM (Netwide Assembler),
9; can *not* be assembled with Microsoft's MASM or any compatible
10; assembler (including Borland's Turbo Assembler).
11; NASM is available from http://nasm.sourceforge.net/ or
12; http://sourceforge.net/project/showfiles.php?group_id=6208
13;
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +000014; [TAB8]
15
Pierre Ossman3a65ef42009-03-16 13:34:18 +000016%include "jsimdext.inc"
17%include "jcolsamp.inc"
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +000018
19; --------------------------------------------------------------------------
20
21%define SCALEBITS 16
22
23F_0_081 equ 5329 ; FIX(0.08131)
24F_0_114 equ 7471 ; FIX(0.11400)
25F_0_168 equ 11059 ; FIX(0.16874)
26F_0_250 equ 16384 ; FIX(0.25000)
27F_0_299 equ 19595 ; FIX(0.29900)
28F_0_331 equ 21709 ; FIX(0.33126)
29F_0_418 equ 27439 ; FIX(0.41869)
30F_0_587 equ 38470 ; FIX(0.58700)
31F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000)
32
33; --------------------------------------------------------------------------
34 SECTION SEG_CONST
35
36 alignz 16
37 global EXTN(jconst_rgb_ycc_convert_sse2)
38
39EXTN(jconst_rgb_ycc_convert_sse2):
40
41PW_F0299_F0337 times 4 dw F_0_299, F_0_337
42PW_F0114_F0250 times 4 dw F_0_114, F_0_250
43PW_MF016_MF033 times 4 dw -F_0_168,-F_0_331
44PW_MF008_MF041 times 4 dw -F_0_081,-F_0_418
45PD_ONEHALFM1_CJ times 4 dd (1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)
46PD_ONEHALF times 4 dd (1 << (SCALEBITS-1))
47
48 alignz 16
49
50; --------------------------------------------------------------------------
51 SECTION SEG_TEXT
52 BITS 32
53;
54; Convert some rows of samples to the output colorspace.
55;
56; GLOBAL(void)
Pierre Ossmaneea72152009-03-09 13:34:17 +000057; jsimd_rgb_ycc_convert_sse2 (JDIMENSION img_width,
58; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
59; JDIMENSION output_row, int num_rows);
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +000060;
61
Pierre Ossmaneea72152009-03-09 13:34:17 +000062%define img_width(b) (b)+8 ; JDIMENSION img_width
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +000063%define input_buf(b) (b)+12 ; JSAMPARRAY input_buf
64%define output_buf(b) (b)+16 ; JSAMPIMAGE output_buf
65%define output_row(b) (b)+20 ; JDIMENSION output_row
66%define num_rows(b) (b)+24 ; int num_rows
67
68%define original_ebp ebp+0
69%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
70%define WK_NUM 8
71%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
72
73 align 16
Pierre Ossmaneea72152009-03-09 13:34:17 +000074 global EXTN(jsimd_rgb_ycc_convert_sse2)
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +000075
Pierre Ossmaneea72152009-03-09 13:34:17 +000076EXTN(jsimd_rgb_ycc_convert_sse2):
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +000077 push ebp
78 mov eax,esp ; eax = original ebp
79 sub esp, byte 4
80 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
81 mov [esp],eax
82 mov ebp,esp ; ebp = aligned ebp
83 lea esp, [wk(0)]
84 pushpic eax ; make a room for GOT address
85 push ebx
86; push ecx ; need not be preserved
87; push edx ; need not be preserved
88 push esi
89 push edi
90
91 get_GOT ebx ; get GOT address
92 movpic POINTER [gotptr], ebx ; save GOT address
93
Pierre Ossmaneea72152009-03-09 13:34:17 +000094 mov ecx, JDIMENSION [img_width(eax)]
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +000095 test ecx,ecx
96 jz near .return
97
98 push ecx
99
100 mov esi, JSAMPIMAGE [output_buf(eax)]
101 mov ecx, JDIMENSION [output_row(eax)]
102 mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
103 mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
104 mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
105 lea edi, [edi+ecx*SIZEOF_JSAMPROW]
106 lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
107 lea edx, [edx+ecx*SIZEOF_JSAMPROW]
108
109 pop ecx
110
111 mov esi, JSAMPARRAY [input_buf(eax)]
112 mov eax, INT [num_rows(eax)]
113 test eax,eax
114 jle near .return
115 alignx 16,7
116.rowloop:
117 pushpic eax
118 push edx
119 push ebx
120 push edi
121 push esi
122 push ecx ; col
123
124 mov esi, JSAMPROW [esi] ; inptr
125 mov edi, JSAMPROW [edi] ; outptr0
126 mov ebx, JSAMPROW [ebx] ; outptr1
127 mov edx, JSAMPROW [edx] ; outptr2
128 movpic eax, POINTER [gotptr] ; load GOT address (eax)
129
130 cmp ecx, byte SIZEOF_XMMWORD
131 jae near .columnloop
132 alignx 16,7
133
134%if RGB_PIXELSIZE == 3 ; ---------------
135
136.column_ld1:
137 push eax
138 push edx
139 lea ecx,[ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
140 test cl, SIZEOF_BYTE
141 jz short .column_ld2
142 sub ecx, byte SIZEOF_BYTE
143 movzx eax, BYTE [esi+ecx]
144.column_ld2:
145 test cl, SIZEOF_WORD
146 jz short .column_ld4
147 sub ecx, byte SIZEOF_WORD
148 movzx edx, WORD [esi+ecx]
149 shl eax, WORD_BIT
150 or eax,edx
151.column_ld4:
152 movd xmmA,eax
153 pop edx
154 pop eax
155 test cl, SIZEOF_DWORD
156 jz short .column_ld8
157 sub ecx, byte SIZEOF_DWORD
Pierre Ossmaneea72152009-03-09 13:34:17 +0000158 movd xmmF, XMM_DWORD [esi+ecx]
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +0000159 pslldq xmmA, SIZEOF_DWORD
160 por xmmA,xmmF
161.column_ld8:
162 test cl, SIZEOF_MMWORD
163 jz short .column_ld16
164 sub ecx, byte SIZEOF_MMWORD
Pierre Ossmaneea72152009-03-09 13:34:17 +0000165 movq xmmB, XMM_MMWORD [esi+ecx]
MIYASAKA Masarua2e6a9d2006-02-04 00:00:00 +0000166 pslldq xmmA, SIZEOF_MMWORD
167 por xmmA,xmmB
168.column_ld16:
169 test cl, SIZEOF_XMMWORD
170 jz short .column_ld32
171 movdqa xmmF,xmmA
172 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
173 mov ecx, SIZEOF_XMMWORD
174 jmp short .rgb_ycc_cnv
175.column_ld32:
176 test cl, 2*SIZEOF_XMMWORD
177 mov ecx, SIZEOF_XMMWORD
178 jz short .rgb_ycc_cnv
179 movdqa xmmB,xmmA
180 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
181 movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
182 jmp short .rgb_ycc_cnv
183 alignx 16,7
184
185.columnloop:
186 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
187 movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD]
188 movdqu xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD]
189
190.rgb_ycc_cnv:
191 ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
192 ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
193 ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
194
195 movdqa xmmG,xmmA
196 pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12)
197 psrldq xmmG,8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --)
198
199 punpckhbw xmmA,xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A)
200 pslldq xmmF,8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27)
201
202 punpcklbw xmmG,xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D)
203 punpckhbw xmmF,xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F)
204
205 movdqa xmmD,xmmA
206 pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09)
207 psrldq xmmD,8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --)
208
209 punpckhbw xmmA,xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D)
210 pslldq xmmG,8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B)
211
212 punpcklbw xmmD,xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E)
213 punpckhbw xmmG,xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F)
214
215 movdqa xmmE,xmmA
216 pslldq xmmA,8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C)
217 psrldq xmmE,8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --)
218
219 punpckhbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
220 pslldq xmmD,8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D)
221
222 punpcklbw xmmE,xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F)
223 punpckhbw xmmD,xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F)
224
225 pxor xmmH,xmmH
226
227 movdqa xmmC,xmmA
228 punpcklbw xmmA,xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E)
229 punpckhbw xmmC,xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E)
230
231 movdqa xmmB,xmmE
232 punpcklbw xmmE,xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E)
233 punpckhbw xmmB,xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F)
234
235 movdqa xmmF,xmmD
236 punpcklbw xmmD,xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F)
237 punpckhbw xmmF,xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F)
238
239%else ; RGB_PIXELSIZE == 4 ; -----------
240
241.column_ld1:
242 test cl, SIZEOF_XMMWORD/16
243 jz short .column_ld2
244 sub ecx, byte SIZEOF_XMMWORD/16
245 movd xmmA, _DWORD [esi+ecx*RGB_PIXELSIZE]
246.column_ld2:
247 test cl, SIZEOF_XMMWORD/8
248 jz short .column_ld4
249 sub ecx, byte SIZEOF_XMMWORD/8
250 movq xmmE, _MMWORD [esi+ecx*RGB_PIXELSIZE]
251 pslldq xmmA, SIZEOF_MMWORD
252 por xmmA,xmmE
253.column_ld4:
254 test cl, SIZEOF_XMMWORD/4
255 jz short .column_ld8
256 sub ecx, byte SIZEOF_XMMWORD/4
257 movdqa xmmE,xmmA
258 movdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
259.column_ld8:
260 test cl, SIZEOF_XMMWORD/2
261 mov ecx, SIZEOF_XMMWORD
262 jz short .rgb_ycc_cnv
263 movdqa xmmF,xmmA
264 movdqa xmmH,xmmE
265 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
266 movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
267 jmp short .rgb_ycc_cnv
268 alignx 16,7
269
270.columnloop:
271 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD]
272 movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD]
273 movdqu xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD]
274 movdqu xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD]
275
276.rgb_ycc_cnv:
277 ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
278 ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
279 ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
280 ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
281
282 movdqa xmmD,xmmA
283 punpcklbw xmmA,xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35)
284 punpckhbw xmmD,xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37)
285
286 movdqa xmmC,xmmF
287 punpcklbw xmmF,xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D)
288 punpckhbw xmmC,xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F)
289
290 movdqa xmmB,xmmA
291 punpcklwd xmmA,xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C)
292 punpckhwd xmmB,xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D)
293
294 movdqa xmmG,xmmD
295 punpcklwd xmmD,xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E)
296 punpckhwd xmmG,xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F)
297
298 movdqa xmmE,xmmA
299 punpcklbw xmmA,xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E)
300 punpckhbw xmmE,xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E)
301
302 movdqa xmmH,xmmB
303 punpcklbw xmmB,xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F)
304 punpckhbw xmmH,xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F)
305
306 pxor xmmF,xmmF
307
308 movdqa xmmC,xmmA
309 punpcklbw xmmA,xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E)
310 punpckhbw xmmC,xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E)
311
312 movdqa xmmD,xmmB
313 punpcklbw xmmB,xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F)
314 punpckhbw xmmD,xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F)
315
316 movdqa xmmG,xmmE
317 punpcklbw xmmE,xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E)
318 punpckhbw xmmG,xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E)
319
320 punpcklbw xmmF,xmmH
321 punpckhbw xmmH,xmmH
322 psrlw xmmF,BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F)
323 psrlw xmmH,BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F)
324
325%endif ; RGB_PIXELSIZE ; ---------------
326
327 ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE
328 ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO
329
330 ; (Original)
331 ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
332 ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
333 ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
334 ;
335 ; (This implementation)
336 ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
337 ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
338 ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
339
340 movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=RE
341 movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=RO
342 movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=BE
343 movdqa XMMWORD [wk(3)], xmm5 ; wk(3)=BO
344
345 movdqa xmm6,xmm1
346 punpcklwd xmm1,xmm3
347 punpckhwd xmm6,xmm3
348 movdqa xmm7,xmm1
349 movdqa xmm4,xmm6
350 pmaddwd xmm1,[GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337)
351 pmaddwd xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337)
352 pmaddwd xmm7,[GOTOFF(eax,PW_MF016_MF033)] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
353 pmaddwd xmm4,[GOTOFF(eax,PW_MF016_MF033)] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
354
355 movdqa XMMWORD [wk(4)], xmm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
356 movdqa XMMWORD [wk(5)], xmm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
357
358 pxor xmm1,xmm1
359 pxor xmm6,xmm6
360 punpcklwd xmm1,xmm5 ; xmm1=BOL
361 punpckhwd xmm6,xmm5 ; xmm6=BOH
362 psrld xmm1,1 ; xmm1=BOL*FIX(0.500)
363 psrld xmm6,1 ; xmm6=BOH*FIX(0.500)
364
365 movdqa xmm5,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm5=[PD_ONEHALFM1_CJ]
366
367 paddd xmm7,xmm1
368 paddd xmm4,xmm6
369 paddd xmm7,xmm5
370 paddd xmm4,xmm5
371 psrld xmm7,SCALEBITS ; xmm7=CbOL
372 psrld xmm4,SCALEBITS ; xmm4=CbOH
373 packssdw xmm7,xmm4 ; xmm7=CbO
374
375 movdqa xmm1, XMMWORD [wk(2)] ; xmm1=BE
376
377 movdqa xmm6,xmm0
378 punpcklwd xmm0,xmm2
379 punpckhwd xmm6,xmm2
380 movdqa xmm5,xmm0
381 movdqa xmm4,xmm6
382 pmaddwd xmm0,[GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337)
383 pmaddwd xmm6,[GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337)
384 pmaddwd xmm5,[GOTOFF(eax,PW_MF016_MF033)] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
385 pmaddwd xmm4,[GOTOFF(eax,PW_MF016_MF033)] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
386
387 movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
388 movdqa XMMWORD [wk(7)], xmm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
389
390 pxor xmm0,xmm0
391 pxor xmm6,xmm6
392 punpcklwd xmm0,xmm1 ; xmm0=BEL
393 punpckhwd xmm6,xmm1 ; xmm6=BEH
394 psrld xmm0,1 ; xmm0=BEL*FIX(0.500)
395 psrld xmm6,1 ; xmm6=BEH*FIX(0.500)
396
397 movdqa xmm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
398
399 paddd xmm5,xmm0
400 paddd xmm4,xmm6
401 paddd xmm5,xmm1
402 paddd xmm4,xmm1
403 psrld xmm5,SCALEBITS ; xmm5=CbEL
404 psrld xmm4,SCALEBITS ; xmm4=CbEH
405 packssdw xmm5,xmm4 ; xmm5=CbE
406
407 psllw xmm7,BYTE_BIT
408 por xmm5,xmm7 ; xmm5=Cb
409 movdqa XMMWORD [ebx], xmm5 ; Save Cb
410
411 movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO
412 movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE
413 movdqa xmm1, XMMWORD [wk(1)] ; xmm1=RO
414
415 movdqa xmm4,xmm0
416 punpcklwd xmm0,xmm3
417 punpckhwd xmm4,xmm3
418 movdqa xmm7,xmm0
419 movdqa xmm5,xmm4
420 pmaddwd xmm0,[GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250)
421 pmaddwd xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250)
422 pmaddwd xmm7,[GOTOFF(eax,PW_MF008_MF041)] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
423 pmaddwd xmm5,[GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
424
425 movdqa xmm3,[GOTOFF(eax,PD_ONEHALF)] ; xmm3=[PD_ONEHALF]
426
427 paddd xmm0, XMMWORD [wk(4)]
428 paddd xmm4, XMMWORD [wk(5)]
429 paddd xmm0,xmm3
430 paddd xmm4,xmm3
431 psrld xmm0,SCALEBITS ; xmm0=YOL
432 psrld xmm4,SCALEBITS ; xmm4=YOH
433 packssdw xmm0,xmm4 ; xmm0=YO
434
435 pxor xmm3,xmm3
436 pxor xmm4,xmm4
437 punpcklwd xmm3,xmm1 ; xmm3=ROL
438 punpckhwd xmm4,xmm1 ; xmm4=ROH
439 psrld xmm3,1 ; xmm3=ROL*FIX(0.500)
440 psrld xmm4,1 ; xmm4=ROH*FIX(0.500)
441
442 movdqa xmm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ]
443
444 paddd xmm7,xmm3
445 paddd xmm5,xmm4
446 paddd xmm7,xmm1
447 paddd xmm5,xmm1
448 psrld xmm7,SCALEBITS ; xmm7=CrOL
449 psrld xmm5,SCALEBITS ; xmm5=CrOH
450 packssdw xmm7,xmm5 ; xmm7=CrO
451
452 movdqa xmm3, XMMWORD [wk(0)] ; xmm3=RE
453
454 movdqa xmm4,xmm6
455 punpcklwd xmm6,xmm2
456 punpckhwd xmm4,xmm2
457 movdqa xmm1,xmm6
458 movdqa xmm5,xmm4
459 pmaddwd xmm6,[GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250)
460 pmaddwd xmm4,[GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250)
461 pmaddwd xmm1,[GOTOFF(eax,PW_MF008_MF041)] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
462 pmaddwd xmm5,[GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
463
464 movdqa xmm2,[GOTOFF(eax,PD_ONEHALF)] ; xmm2=[PD_ONEHALF]
465
466 paddd xmm6, XMMWORD [wk(6)]
467 paddd xmm4, XMMWORD [wk(7)]
468 paddd xmm6,xmm2
469 paddd xmm4,xmm2
470 psrld xmm6,SCALEBITS ; xmm6=YEL
471 psrld xmm4,SCALEBITS ; xmm4=YEH
472 packssdw xmm6,xmm4 ; xmm6=YE
473
474 psllw xmm0,BYTE_BIT
475 por xmm6,xmm0 ; xmm6=Y
476 movdqa XMMWORD [edi], xmm6 ; Save Y
477
478 pxor xmm2,xmm2
479 pxor xmm4,xmm4
480 punpcklwd xmm2,xmm3 ; xmm2=REL
481 punpckhwd xmm4,xmm3 ; xmm4=REH
482 psrld xmm2,1 ; xmm2=REL*FIX(0.500)
483 psrld xmm4,1 ; xmm4=REH*FIX(0.500)
484
485 movdqa xmm0,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm0=[PD_ONEHALFM1_CJ]
486
487 paddd xmm1,xmm2
488 paddd xmm5,xmm4
489 paddd xmm1,xmm0
490 paddd xmm5,xmm0
491 psrld xmm1,SCALEBITS ; xmm1=CrEL
492 psrld xmm5,SCALEBITS ; xmm5=CrEH
493 packssdw xmm1,xmm5 ; xmm1=CrE
494
495 psllw xmm7,BYTE_BIT
496 por xmm1,xmm7 ; xmm1=Cr
497 movdqa XMMWORD [edx], xmm1 ; Save Cr
498
499 sub ecx, byte SIZEOF_XMMWORD
500 add esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr
501 add edi, byte SIZEOF_XMMWORD ; outptr0
502 add ebx, byte SIZEOF_XMMWORD ; outptr1
503 add edx, byte SIZEOF_XMMWORD ; outptr2
504 cmp ecx, byte SIZEOF_XMMWORD
505 jae near .columnloop
506 test ecx,ecx
507 jnz near .column_ld1
508
509 pop ecx ; col
510 pop esi
511 pop edi
512 pop ebx
513 pop edx
514 poppic eax
515
516 add esi, byte SIZEOF_JSAMPROW ; input_buf
517 add edi, byte SIZEOF_JSAMPROW
518 add ebx, byte SIZEOF_JSAMPROW
519 add edx, byte SIZEOF_JSAMPROW
520 dec eax ; num_rows
521 jg near .rowloop
522
523.return:
524 pop edi
525 pop esi
526; pop edx ; need not be preserved
527; pop ecx ; need not be preserved
528 pop ebx
529 mov esp,ebp ; esp <- aligned ebp
530 pop esp ; esp <- original ebp
531 pop ebp
532 ret
533