media/libjpeg/simd/jcclrmmx.asm

Wed, 31 Dec 2014 06:09:35 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Wed, 31 Dec 2014 06:09:35 +0100
changeset 0
6474c204b198
permissions
-rw-r--r--

Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.

michael@0 1 ;
michael@0 2 ; jcclrmmx.asm - colorspace conversion (MMX)
michael@0 3 ;
michael@0 4 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
michael@0 5 ;
michael@0 6 ; Based on
michael@0 7 ; x86 SIMD extension for IJG JPEG library
michael@0 8 ; Copyright (C) 1999-2006, MIYASAKA Masaru.
michael@0 9 ; For conditions of distribution and use, see copyright notice in jsimdext.inc
michael@0 10 ;
michael@0 11 ; This file should be assembled with NASM (Netwide Assembler),
michael@0 12 ; can *not* be assembled with Microsoft's MASM or any compatible
michael@0 13 ; assembler (including Borland's Turbo Assembler).
michael@0 14 ; NASM is available from http://nasm.sourceforge.net/ or
michael@0 15 ; http://sourceforge.net/project/showfiles.php?group_id=6208
michael@0 16 ;
michael@0 17 ; [TAB8]
michael@0 18
michael@0 19 %include "jcolsamp.inc"
michael@0 20
michael@0 21 ; --------------------------------------------------------------------------
michael@0 22 ;
michael@0 23 ; Convert some rows of samples to the output colorspace.
michael@0 24 ;
michael@0 25 ; GLOBAL(void)
michael@0 26 ; jsimd_rgb_ycc_convert_mmx (JDIMENSION img_width,
michael@0 27 ; JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
michael@0 28 ; JDIMENSION output_row, int num_rows);
michael@0 29 ;
michael@0 30
michael@0 31 %define img_width(b) (b)+8 ; JDIMENSION img_width
michael@0 32 %define input_buf(b) (b)+12 ; JSAMPARRAY input_buf
michael@0 33 %define output_buf(b) (b)+16 ; JSAMPIMAGE output_buf
michael@0 34 %define output_row(b) (b)+20 ; JDIMENSION output_row
michael@0 35 %define num_rows(b) (b)+24 ; int num_rows
michael@0 36
michael@0 37 %define original_ebp ebp+0
michael@0 38 %define wk(i) ebp-(WK_NUM-(i))*SIZEOF_MMWORD ; mmword wk[WK_NUM]
michael@0 39 %define WK_NUM 8
michael@0 40 %define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
michael@0 41
michael@0 42 align 16
michael@0 43 global EXTN(jsimd_rgb_ycc_convert_mmx)
michael@0 44
michael@0 45 EXTN(jsimd_rgb_ycc_convert_mmx):
michael@0 46 push ebp
michael@0 47 mov eax,esp ; eax = original ebp
michael@0 48 sub esp, byte 4
michael@0 49 and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits
michael@0 50 mov [esp],eax
michael@0 51 mov ebp,esp ; ebp = aligned ebp
michael@0 52 lea esp, [wk(0)]
michael@0 53 pushpic eax ; make a room for GOT address
michael@0 54 push ebx
michael@0 55 ; push ecx ; need not be preserved
michael@0 56 ; push edx ; need not be preserved
michael@0 57 push esi
michael@0 58 push edi
michael@0 59
michael@0 60 get_GOT ebx ; get GOT address
michael@0 61 movpic POINTER [gotptr], ebx ; save GOT address
michael@0 62
michael@0 63 mov ecx, JDIMENSION [img_width(eax)] ; num_cols
michael@0 64 test ecx,ecx
michael@0 65 jz near .return
michael@0 66
michael@0 67 push ecx
michael@0 68
michael@0 69 mov esi, JSAMPIMAGE [output_buf(eax)]
michael@0 70 mov ecx, JDIMENSION [output_row(eax)]
michael@0 71 mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
michael@0 72 mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY]
michael@0 73 mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY]
michael@0 74 lea edi, [edi+ecx*SIZEOF_JSAMPROW]
michael@0 75 lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
michael@0 76 lea edx, [edx+ecx*SIZEOF_JSAMPROW]
michael@0 77
michael@0 78 pop ecx
michael@0 79
michael@0 80 mov esi, JSAMPARRAY [input_buf(eax)]
michael@0 81 mov eax, INT [num_rows(eax)]
michael@0 82 test eax,eax
michael@0 83 jle near .return
michael@0 84 alignx 16,7
michael@0 85 .rowloop:
michael@0 86 pushpic eax
michael@0 87 push edx
michael@0 88 push ebx
michael@0 89 push edi
michael@0 90 push esi
michael@0 91 push ecx ; col
michael@0 92
michael@0 93 mov esi, JSAMPROW [esi] ; inptr
michael@0 94 mov edi, JSAMPROW [edi] ; outptr0
michael@0 95 mov ebx, JSAMPROW [ebx] ; outptr1
michael@0 96 mov edx, JSAMPROW [edx] ; outptr2
michael@0 97 movpic eax, POINTER [gotptr] ; load GOT address (eax)
michael@0 98
michael@0 99 cmp ecx, byte SIZEOF_MMWORD
michael@0 100 jae short .columnloop
michael@0 101 alignx 16,7
michael@0 102
michael@0 103 %if RGB_PIXELSIZE == 3 ; ---------------
michael@0 104
michael@0 105 .column_ld1:
michael@0 106 push eax
michael@0 107 push edx
michael@0 108 lea ecx,[ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
michael@0 109 test cl, SIZEOF_BYTE
michael@0 110 jz short .column_ld2
michael@0 111 sub ecx, byte SIZEOF_BYTE
michael@0 112 xor eax,eax
michael@0 113 mov al, BYTE [esi+ecx]
michael@0 114 .column_ld2:
michael@0 115 test cl, SIZEOF_WORD
michael@0 116 jz short .column_ld4
michael@0 117 sub ecx, byte SIZEOF_WORD
michael@0 118 xor edx,edx
michael@0 119 mov dx, WORD [esi+ecx]
michael@0 120 shl eax, WORD_BIT
michael@0 121 or eax,edx
michael@0 122 .column_ld4:
michael@0 123 movd mmA,eax
michael@0 124 pop edx
michael@0 125 pop eax
michael@0 126 test cl, SIZEOF_DWORD
michael@0 127 jz short .column_ld8
michael@0 128 sub ecx, byte SIZEOF_DWORD
michael@0 129 movd mmG, DWORD [esi+ecx]
michael@0 130 psllq mmA, DWORD_BIT
michael@0 131 por mmA,mmG
michael@0 132 .column_ld8:
michael@0 133 test cl, SIZEOF_MMWORD
michael@0 134 jz short .column_ld16
michael@0 135 movq mmG,mmA
michael@0 136 movq mmA, MMWORD [esi+0*SIZEOF_MMWORD]
michael@0 137 mov ecx, SIZEOF_MMWORD
michael@0 138 jmp short .rgb_ycc_cnv
michael@0 139 .column_ld16:
michael@0 140 test cl, 2*SIZEOF_MMWORD
michael@0 141 mov ecx, SIZEOF_MMWORD
michael@0 142 jz short .rgb_ycc_cnv
michael@0 143 movq mmF,mmA
michael@0 144 movq mmA, MMWORD [esi+0*SIZEOF_MMWORD]
michael@0 145 movq mmG, MMWORD [esi+1*SIZEOF_MMWORD]
michael@0 146 jmp short .rgb_ycc_cnv
michael@0 147 alignx 16,7
michael@0 148
michael@0 149 .columnloop:
michael@0 150 movq mmA, MMWORD [esi+0*SIZEOF_MMWORD]
michael@0 151 movq mmG, MMWORD [esi+1*SIZEOF_MMWORD]
michael@0 152 movq mmF, MMWORD [esi+2*SIZEOF_MMWORD]
michael@0 153
michael@0 154 .rgb_ycc_cnv:
michael@0 155 ; mmA=(00 10 20 01 11 21 02 12)
michael@0 156 ; mmG=(22 03 13 23 04 14 24 05)
michael@0 157 ; mmF=(15 25 06 16 26 07 17 27)
michael@0 158
michael@0 159 movq mmD,mmA
michael@0 160 psllq mmA,4*BYTE_BIT ; mmA=(-- -- -- -- 00 10 20 01)
michael@0 161 psrlq mmD,4*BYTE_BIT ; mmD=(11 21 02 12 -- -- -- --)
michael@0 162
michael@0 163 punpckhbw mmA,mmG ; mmA=(00 04 10 14 20 24 01 05)
michael@0 164 psllq mmG,4*BYTE_BIT ; mmG=(-- -- -- -- 22 03 13 23)
michael@0 165
michael@0 166 punpcklbw mmD,mmF ; mmD=(11 15 21 25 02 06 12 16)
michael@0 167 punpckhbw mmG,mmF ; mmG=(22 26 03 07 13 17 23 27)
michael@0 168
michael@0 169 movq mmE,mmA
michael@0 170 psllq mmA,4*BYTE_BIT ; mmA=(-- -- -- -- 00 04 10 14)
michael@0 171 psrlq mmE,4*BYTE_BIT ; mmE=(20 24 01 05 -- -- -- --)
michael@0 172
michael@0 173 punpckhbw mmA,mmD ; mmA=(00 02 04 06 10 12 14 16)
michael@0 174 psllq mmD,4*BYTE_BIT ; mmD=(-- -- -- -- 11 15 21 25)
michael@0 175
michael@0 176 punpcklbw mmE,mmG ; mmE=(20 22 24 26 01 03 05 07)
michael@0 177 punpckhbw mmD,mmG ; mmD=(11 13 15 17 21 23 25 27)
michael@0 178
michael@0 179 pxor mmH,mmH
michael@0 180
michael@0 181 movq mmC,mmA
michael@0 182 punpcklbw mmA,mmH ; mmA=(00 02 04 06)
michael@0 183 punpckhbw mmC,mmH ; mmC=(10 12 14 16)
michael@0 184
michael@0 185 movq mmB,mmE
michael@0 186 punpcklbw mmE,mmH ; mmE=(20 22 24 26)
michael@0 187 punpckhbw mmB,mmH ; mmB=(01 03 05 07)
michael@0 188
michael@0 189 movq mmF,mmD
michael@0 190 punpcklbw mmD,mmH ; mmD=(11 13 15 17)
michael@0 191 punpckhbw mmF,mmH ; mmF=(21 23 25 27)
michael@0 192
michael@0 193 %else ; RGB_PIXELSIZE == 4 ; -----------
michael@0 194
michael@0 195 .column_ld1:
michael@0 196 test cl, SIZEOF_MMWORD/8
michael@0 197 jz short .column_ld2
michael@0 198 sub ecx, byte SIZEOF_MMWORD/8
michael@0 199 movd mmA, DWORD [esi+ecx*RGB_PIXELSIZE]
michael@0 200 .column_ld2:
michael@0 201 test cl, SIZEOF_MMWORD/4
michael@0 202 jz short .column_ld4
michael@0 203 sub ecx, byte SIZEOF_MMWORD/4
michael@0 204 movq mmF,mmA
michael@0 205 movq mmA, MMWORD [esi+ecx*RGB_PIXELSIZE]
michael@0 206 .column_ld4:
michael@0 207 test cl, SIZEOF_MMWORD/2
michael@0 208 mov ecx, SIZEOF_MMWORD
michael@0 209 jz short .rgb_ycc_cnv
michael@0 210 movq mmD,mmA
michael@0 211 movq mmC,mmF
michael@0 212 movq mmA, MMWORD [esi+0*SIZEOF_MMWORD]
michael@0 213 movq mmF, MMWORD [esi+1*SIZEOF_MMWORD]
michael@0 214 jmp short .rgb_ycc_cnv
michael@0 215 alignx 16,7
michael@0 216
michael@0 217 .columnloop:
michael@0 218 movq mmA, MMWORD [esi+0*SIZEOF_MMWORD]
michael@0 219 movq mmF, MMWORD [esi+1*SIZEOF_MMWORD]
michael@0 220 movq mmD, MMWORD [esi+2*SIZEOF_MMWORD]
michael@0 221 movq mmC, MMWORD [esi+3*SIZEOF_MMWORD]
michael@0 222
michael@0 223 .rgb_ycc_cnv:
michael@0 224 ; mmA=(00 10 20 30 01 11 21 31)
michael@0 225 ; mmF=(02 12 22 32 03 13 23 33)
michael@0 226 ; mmD=(04 14 24 34 05 15 25 35)
michael@0 227 ; mmC=(06 16 26 36 07 17 27 37)
michael@0 228
michael@0 229 movq mmB,mmA
michael@0 230 punpcklbw mmA,mmF ; mmA=(00 02 10 12 20 22 30 32)
michael@0 231 punpckhbw mmB,mmF ; mmB=(01 03 11 13 21 23 31 33)
michael@0 232
michael@0 233 movq mmG,mmD
michael@0 234 punpcklbw mmD,mmC ; mmD=(04 06 14 16 24 26 34 36)
michael@0 235 punpckhbw mmG,mmC ; mmG=(05 07 15 17 25 27 35 37)
michael@0 236
michael@0 237 movq mmE,mmA
michael@0 238 punpcklwd mmA,mmD ; mmA=(00 02 04 06 10 12 14 16)
michael@0 239 punpckhwd mmE,mmD ; mmE=(20 22 24 26 30 32 34 36)
michael@0 240
michael@0 241 movq mmH,mmB
michael@0 242 punpcklwd mmB,mmG ; mmB=(01 03 05 07 11 13 15 17)
michael@0 243 punpckhwd mmH,mmG ; mmH=(21 23 25 27 31 33 35 37)
michael@0 244
michael@0 245 pxor mmF,mmF
michael@0 246
michael@0 247 movq mmC,mmA
michael@0 248 punpcklbw mmA,mmF ; mmA=(00 02 04 06)
michael@0 249 punpckhbw mmC,mmF ; mmC=(10 12 14 16)
michael@0 250
michael@0 251 movq mmD,mmB
michael@0 252 punpcklbw mmB,mmF ; mmB=(01 03 05 07)
michael@0 253 punpckhbw mmD,mmF ; mmD=(11 13 15 17)
michael@0 254
michael@0 255 movq mmG,mmE
michael@0 256 punpcklbw mmE,mmF ; mmE=(20 22 24 26)
michael@0 257 punpckhbw mmG,mmF ; mmG=(30 32 34 36)
michael@0 258
michael@0 259 punpcklbw mmF,mmH
michael@0 260 punpckhbw mmH,mmH
michael@0 261 psrlw mmF,BYTE_BIT ; mmF=(21 23 25 27)
michael@0 262 psrlw mmH,BYTE_BIT ; mmH=(31 33 35 37)
michael@0 263
michael@0 264 %endif ; RGB_PIXELSIZE ; ---------------
michael@0 265
michael@0 266 ; mm0=(R0 R2 R4 R6)=RE, mm2=(G0 G2 G4 G6)=GE, mm4=(B0 B2 B4 B6)=BE
michael@0 267 ; mm1=(R1 R3 R5 R7)=RO, mm3=(G1 G3 G5 G7)=GO, mm5=(B1 B3 B5 B7)=BO
michael@0 268
michael@0 269 ; (Original)
michael@0 270 ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
michael@0 271 ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
michael@0 272 ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
michael@0 273 ;
michael@0 274 ; (This implementation)
michael@0 275 ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
michael@0 276 ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE
michael@0 277 ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE
michael@0 278
michael@0 279 movq MMWORD [wk(0)], mm0 ; wk(0)=RE
michael@0 280 movq MMWORD [wk(1)], mm1 ; wk(1)=RO
michael@0 281 movq MMWORD [wk(2)], mm4 ; wk(2)=BE
michael@0 282 movq MMWORD [wk(3)], mm5 ; wk(3)=BO
michael@0 283
michael@0 284 movq mm6,mm1
michael@0 285 punpcklwd mm1,mm3
michael@0 286 punpckhwd mm6,mm3
michael@0 287 movq mm7,mm1
michael@0 288 movq mm4,mm6
michael@0 289 pmaddwd mm1,[GOTOFF(eax,PW_F0299_F0337)] ; mm1=ROL*FIX(0.299)+GOL*FIX(0.337)
michael@0 290 pmaddwd mm6,[GOTOFF(eax,PW_F0299_F0337)] ; mm6=ROH*FIX(0.299)+GOH*FIX(0.337)
michael@0 291 pmaddwd mm7,[GOTOFF(eax,PW_MF016_MF033)] ; mm7=ROL*-FIX(0.168)+GOL*-FIX(0.331)
michael@0 292 pmaddwd mm4,[GOTOFF(eax,PW_MF016_MF033)] ; mm4=ROH*-FIX(0.168)+GOH*-FIX(0.331)
michael@0 293
michael@0 294 movq MMWORD [wk(4)], mm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337)
michael@0 295 movq MMWORD [wk(5)], mm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337)
michael@0 296
michael@0 297 pxor mm1,mm1
michael@0 298 pxor mm6,mm6
michael@0 299 punpcklwd mm1,mm5 ; mm1=BOL
michael@0 300 punpckhwd mm6,mm5 ; mm6=BOH
michael@0 301 psrld mm1,1 ; mm1=BOL*FIX(0.500)
michael@0 302 psrld mm6,1 ; mm6=BOH*FIX(0.500)
michael@0 303
michael@0 304 movq mm5,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm5=[PD_ONEHALFM1_CJ]
michael@0 305
michael@0 306 paddd mm7,mm1
michael@0 307 paddd mm4,mm6
michael@0 308 paddd mm7,mm5
michael@0 309 paddd mm4,mm5
michael@0 310 psrld mm7,SCALEBITS ; mm7=CbOL
michael@0 311 psrld mm4,SCALEBITS ; mm4=CbOH
michael@0 312 packssdw mm7,mm4 ; mm7=CbO
michael@0 313
michael@0 314 movq mm1, MMWORD [wk(2)] ; mm1=BE
michael@0 315
michael@0 316 movq mm6,mm0
michael@0 317 punpcklwd mm0,mm2
michael@0 318 punpckhwd mm6,mm2
michael@0 319 movq mm5,mm0
michael@0 320 movq mm4,mm6
michael@0 321 pmaddwd mm0,[GOTOFF(eax,PW_F0299_F0337)] ; mm0=REL*FIX(0.299)+GEL*FIX(0.337)
michael@0 322 pmaddwd mm6,[GOTOFF(eax,PW_F0299_F0337)] ; mm6=REH*FIX(0.299)+GEH*FIX(0.337)
michael@0 323 pmaddwd mm5,[GOTOFF(eax,PW_MF016_MF033)] ; mm5=REL*-FIX(0.168)+GEL*-FIX(0.331)
michael@0 324 pmaddwd mm4,[GOTOFF(eax,PW_MF016_MF033)] ; mm4=REH*-FIX(0.168)+GEH*-FIX(0.331)
michael@0 325
michael@0 326 movq MMWORD [wk(6)], mm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337)
michael@0 327 movq MMWORD [wk(7)], mm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337)
michael@0 328
michael@0 329 pxor mm0,mm0
michael@0 330 pxor mm6,mm6
michael@0 331 punpcklwd mm0,mm1 ; mm0=BEL
michael@0 332 punpckhwd mm6,mm1 ; mm6=BEH
michael@0 333 psrld mm0,1 ; mm0=BEL*FIX(0.500)
michael@0 334 psrld mm6,1 ; mm6=BEH*FIX(0.500)
michael@0 335
michael@0 336 movq mm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm1=[PD_ONEHALFM1_CJ]
michael@0 337
michael@0 338 paddd mm5,mm0
michael@0 339 paddd mm4,mm6
michael@0 340 paddd mm5,mm1
michael@0 341 paddd mm4,mm1
michael@0 342 psrld mm5,SCALEBITS ; mm5=CbEL
michael@0 343 psrld mm4,SCALEBITS ; mm4=CbEH
michael@0 344 packssdw mm5,mm4 ; mm5=CbE
michael@0 345
michael@0 346 psllw mm7,BYTE_BIT
michael@0 347 por mm5,mm7 ; mm5=Cb
michael@0 348 movq MMWORD [ebx], mm5 ; Save Cb
michael@0 349
michael@0 350 movq mm0, MMWORD [wk(3)] ; mm0=BO
michael@0 351 movq mm6, MMWORD [wk(2)] ; mm6=BE
michael@0 352 movq mm1, MMWORD [wk(1)] ; mm1=RO
michael@0 353
michael@0 354 movq mm4,mm0
michael@0 355 punpcklwd mm0,mm3
michael@0 356 punpckhwd mm4,mm3
michael@0 357 movq mm7,mm0
michael@0 358 movq mm5,mm4
michael@0 359 pmaddwd mm0,[GOTOFF(eax,PW_F0114_F0250)] ; mm0=BOL*FIX(0.114)+GOL*FIX(0.250)
michael@0 360 pmaddwd mm4,[GOTOFF(eax,PW_F0114_F0250)] ; mm4=BOH*FIX(0.114)+GOH*FIX(0.250)
michael@0 361 pmaddwd mm7,[GOTOFF(eax,PW_MF008_MF041)] ; mm7=BOL*-FIX(0.081)+GOL*-FIX(0.418)
michael@0 362 pmaddwd mm5,[GOTOFF(eax,PW_MF008_MF041)] ; mm5=BOH*-FIX(0.081)+GOH*-FIX(0.418)
michael@0 363
michael@0 364 movq mm3,[GOTOFF(eax,PD_ONEHALF)] ; mm3=[PD_ONEHALF]
michael@0 365
michael@0 366 paddd mm0, MMWORD [wk(4)]
michael@0 367 paddd mm4, MMWORD [wk(5)]
michael@0 368 paddd mm0,mm3
michael@0 369 paddd mm4,mm3
michael@0 370 psrld mm0,SCALEBITS ; mm0=YOL
michael@0 371 psrld mm4,SCALEBITS ; mm4=YOH
michael@0 372 packssdw mm0,mm4 ; mm0=YO
michael@0 373
michael@0 374 pxor mm3,mm3
michael@0 375 pxor mm4,mm4
michael@0 376 punpcklwd mm3,mm1 ; mm3=ROL
michael@0 377 punpckhwd mm4,mm1 ; mm4=ROH
michael@0 378 psrld mm3,1 ; mm3=ROL*FIX(0.500)
michael@0 379 psrld mm4,1 ; mm4=ROH*FIX(0.500)
michael@0 380
michael@0 381 movq mm1,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm1=[PD_ONEHALFM1_CJ]
michael@0 382
michael@0 383 paddd mm7,mm3
michael@0 384 paddd mm5,mm4
michael@0 385 paddd mm7,mm1
michael@0 386 paddd mm5,mm1
michael@0 387 psrld mm7,SCALEBITS ; mm7=CrOL
michael@0 388 psrld mm5,SCALEBITS ; mm5=CrOH
michael@0 389 packssdw mm7,mm5 ; mm7=CrO
michael@0 390
michael@0 391 movq mm3, MMWORD [wk(0)] ; mm3=RE
michael@0 392
michael@0 393 movq mm4,mm6
michael@0 394 punpcklwd mm6,mm2
michael@0 395 punpckhwd mm4,mm2
michael@0 396 movq mm1,mm6
michael@0 397 movq mm5,mm4
michael@0 398 pmaddwd mm6,[GOTOFF(eax,PW_F0114_F0250)] ; mm6=BEL*FIX(0.114)+GEL*FIX(0.250)
michael@0 399 pmaddwd mm4,[GOTOFF(eax,PW_F0114_F0250)] ; mm4=BEH*FIX(0.114)+GEH*FIX(0.250)
michael@0 400 pmaddwd mm1,[GOTOFF(eax,PW_MF008_MF041)] ; mm1=BEL*-FIX(0.081)+GEL*-FIX(0.418)
michael@0 401 pmaddwd mm5,[GOTOFF(eax,PW_MF008_MF041)] ; mm5=BEH*-FIX(0.081)+GEH*-FIX(0.418)
michael@0 402
michael@0 403 movq mm2,[GOTOFF(eax,PD_ONEHALF)] ; mm2=[PD_ONEHALF]
michael@0 404
michael@0 405 paddd mm6, MMWORD [wk(6)]
michael@0 406 paddd mm4, MMWORD [wk(7)]
michael@0 407 paddd mm6,mm2
michael@0 408 paddd mm4,mm2
michael@0 409 psrld mm6,SCALEBITS ; mm6=YEL
michael@0 410 psrld mm4,SCALEBITS ; mm4=YEH
michael@0 411 packssdw mm6,mm4 ; mm6=YE
michael@0 412
michael@0 413 psllw mm0,BYTE_BIT
michael@0 414 por mm6,mm0 ; mm6=Y
michael@0 415 movq MMWORD [edi], mm6 ; Save Y
michael@0 416
michael@0 417 pxor mm2,mm2
michael@0 418 pxor mm4,mm4
michael@0 419 punpcklwd mm2,mm3 ; mm2=REL
michael@0 420 punpckhwd mm4,mm3 ; mm4=REH
michael@0 421 psrld mm2,1 ; mm2=REL*FIX(0.500)
michael@0 422 psrld mm4,1 ; mm4=REH*FIX(0.500)
michael@0 423
michael@0 424 movq mm0,[GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm0=[PD_ONEHALFM1_CJ]
michael@0 425
michael@0 426 paddd mm1,mm2
michael@0 427 paddd mm5,mm4
michael@0 428 paddd mm1,mm0
michael@0 429 paddd mm5,mm0
michael@0 430 psrld mm1,SCALEBITS ; mm1=CrEL
michael@0 431 psrld mm5,SCALEBITS ; mm5=CrEH
michael@0 432 packssdw mm1,mm5 ; mm1=CrE
michael@0 433
michael@0 434 psllw mm7,BYTE_BIT
michael@0 435 por mm1,mm7 ; mm1=Cr
michael@0 436 movq MMWORD [edx], mm1 ; Save Cr
michael@0 437
michael@0 438 sub ecx, byte SIZEOF_MMWORD
michael@0 439 add esi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; inptr
michael@0 440 add edi, byte SIZEOF_MMWORD ; outptr0
michael@0 441 add ebx, byte SIZEOF_MMWORD ; outptr1
michael@0 442 add edx, byte SIZEOF_MMWORD ; outptr2
michael@0 443 cmp ecx, byte SIZEOF_MMWORD
michael@0 444 jae near .columnloop
michael@0 445 test ecx,ecx
michael@0 446 jnz near .column_ld1
michael@0 447
michael@0 448 pop ecx ; col
michael@0 449 pop esi
michael@0 450 pop edi
michael@0 451 pop ebx
michael@0 452 pop edx
michael@0 453 poppic eax
michael@0 454
michael@0 455 add esi, byte SIZEOF_JSAMPROW ; input_buf
michael@0 456 add edi, byte SIZEOF_JSAMPROW
michael@0 457 add ebx, byte SIZEOF_JSAMPROW
michael@0 458 add edx, byte SIZEOF_JSAMPROW
michael@0 459 dec eax ; num_rows
michael@0 460 jg near .rowloop
michael@0 461
michael@0 462 emms ; empty MMX state
michael@0 463
michael@0 464 .return:
michael@0 465 pop edi
michael@0 466 pop esi
michael@0 467 ; pop edx ; need not be preserved
michael@0 468 ; pop ecx ; need not be preserved
michael@0 469 pop ebx
michael@0 470 mov esp,ebp ; esp <- aligned ebp
michael@0 471 pop esp ; esp <- original ebp
michael@0 472 pop ebp
michael@0 473 ret
michael@0 474
michael@0 475 ; For some reason, the OS X linker does not honor the request to align the
michael@0 476 ; segment unless we do this.
michael@0 477 align 16

mercurial