jccolext-sse2.asm (19069B)
1 ; 2 ; jccolext.asm - colorspace conversion (SSE2) 3 ; 4 ; Copyright (C) 2016, 2024, D. R. Commander. 5 ; 6 ; Based on the x86 SIMD extension for IJG JPEG library 7 ; Copyright (C) 1999-2006, MIYASAKA Masaru. 8 ; For conditions of distribution and use, see copyright notice in jsimdext.inc 9 ; 10 ; This file should be assembled with NASM (Netwide Assembler) or Yasm. 11 12 %include "jcolsamp.inc" 13 14 ; -------------------------------------------------------------------------- 15 ; 16 ; Convert some rows of samples to the output colorspace. 17 ; 18 ; GLOBAL(void) 19 ; jsimd_rgb_ycc_convert_sse2(JDIMENSION img_width, JSAMPARRAY input_buf, 20 ; JSAMPIMAGE output_buf, JDIMENSION output_row, 21 ; int num_rows); 22 ; 23 24 %define img_width(b) (b) + 8 ; JDIMENSION img_width 25 %define input_buf(b) (b) + 12 ; JSAMPARRAY input_buf 26 %define output_buf(b) (b) + 16 ; JSAMPIMAGE output_buf 27 %define output_row(b) (b) + 20 ; JDIMENSION output_row 28 %define num_rows(b) (b) + 24 ; int num_rows 29 30 %define original_ebp ebp + 0 31 %define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD 32 ; xmmword wk[WK_NUM] 33 %define WK_NUM 8 34 %define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr 35 36 align 32 37 GLOBAL_FUNCTION(jsimd_rgb_ycc_convert_sse2) 38 39 EXTN(jsimd_rgb_ycc_convert_sse2): 40 push ebp 41 mov eax, esp ; eax = original ebp 42 sub esp, byte 4 43 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits 44 mov [esp], eax 45 mov ebp, esp ; ebp = aligned ebp 46 lea esp, [wk(0)] 47 PUSHPIC eax ; make a room for GOT address 48 push ebx 49 ; push ecx ; need not be preserved 50 ; push edx ; need not be preserved 51 push esi 52 push edi 53 54 GET_GOT ebx ; get GOT address 55 MOVPIC POINTER [gotptr], ebx ; save GOT address 56 57 mov ecx, JDIMENSION [img_width(eax)] 58 test ecx, ecx 59 jz near .return 60 61 push ecx 62 63 mov esi, JSAMPIMAGE [output_buf(eax)] 64 mov ecx, JDIMENSION [output_row(eax)] 65 mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY] 66 mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY] 67 mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY] 68 lea edi, [edi+ecx*SIZEOF_JSAMPROW] 69 lea ebx, [ebx+ecx*SIZEOF_JSAMPROW] 70 lea edx, [edx+ecx*SIZEOF_JSAMPROW] 71 72 pop ecx 73 74 mov esi, JSAMPARRAY [input_buf(eax)] 75 mov eax, INT [num_rows(eax)] 76 test eax, eax 77 jle near .return 78 ALIGNX 16, 7 79 .rowloop: 80 PUSHPIC eax 81 push edx 82 push ebx 83 push edi 84 push esi 85 push ecx ; col 86 87 mov esi, JSAMPROW [esi] ; inptr 88 mov edi, JSAMPROW [edi] ; outptr0 89 mov ebx, JSAMPROW [ebx] ; outptr1 90 mov edx, JSAMPROW [edx] ; outptr2 91 MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) 92 93 cmp ecx, byte SIZEOF_XMMWORD 94 jae near .columnloop 95 ALIGNX 16, 7 96 97 %if RGB_PIXELSIZE == 3 ; --------------- 98 99 .column_ld1: 100 push eax 101 push edx 102 lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE 103 test cl, SIZEOF_BYTE 104 jz short .column_ld2 105 sub ecx, byte SIZEOF_BYTE 106 movzx eax, byte [esi+ecx] 107 .column_ld2: 108 test cl, SIZEOF_WORD 109 jz short .column_ld4 110 sub ecx, byte SIZEOF_WORD 111 movzx edx, word [esi+ecx] 112 shl eax, WORD_BIT 113 or eax, edx 114 .column_ld4: 115 movd xmmA, eax 116 pop edx 117 pop eax 118 test cl, SIZEOF_DWORD 119 jz short .column_ld8 120 sub ecx, byte SIZEOF_DWORD 121 movd xmmF, XMM_DWORD [esi+ecx] 122 pslldq xmmA, SIZEOF_DWORD 123 por xmmA, xmmF 124 .column_ld8: 125 test cl, SIZEOF_MMWORD 126 jz short .column_ld16 127 sub ecx, byte SIZEOF_MMWORD 128 movq xmmB, XMM_MMWORD [esi+ecx] 129 pslldq xmmA, SIZEOF_MMWORD 130 por xmmA, xmmB 131 .column_ld16: 132 test cl, SIZEOF_XMMWORD 133 jz short .column_ld32 134 movdqa xmmF, xmmA 135 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] 136 mov ecx, SIZEOF_XMMWORD 137 jmp short .rgb_ycc_cnv 138 .column_ld32: 139 test cl, 2*SIZEOF_XMMWORD 140 mov ecx, SIZEOF_XMMWORD 141 jz short .rgb_ycc_cnv 142 movdqa xmmB, xmmA 143 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] 144 movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD] 145 jmp short .rgb_ycc_cnv 146 ALIGNX 16, 7 147 148 .columnloop: 149 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] 150 movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD] 151 movdqu xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD] 152 153 .rgb_ycc_cnv: 154 ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) 155 ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) 156 ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) 157 158 movdqa xmmG, xmmA 159 pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12) 160 psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --) 161 162 punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A) 163 pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27) 164 165 punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D) 166 punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F) 167 168 movdqa xmmD, xmmA 169 pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09) 170 psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --) 171 172 punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D) 173 pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B) 174 175 punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E) 176 punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F) 177 178 movdqa xmmE, xmmA 179 pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C) 180 psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --) 181 182 punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) 183 pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D) 184 185 punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F) 186 punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F) 187 188 pxor xmmH, xmmH 189 190 movdqa xmmC, xmmA 191 punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E) 192 punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E) 193 194 movdqa xmmB, xmmE 195 punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E) 196 punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F) 197 198 movdqa xmmF, xmmD 199 punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F) 200 punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F) 201 202 %else ; RGB_PIXELSIZE == 4 ; ----------- 203 204 .column_ld1: 205 test cl, SIZEOF_XMMWORD/16 206 jz short .column_ld2 207 sub ecx, byte SIZEOF_XMMWORD/16 208 movd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE] 209 .column_ld2: 210 test cl, SIZEOF_XMMWORD/8 211 jz short .column_ld4 212 sub ecx, byte SIZEOF_XMMWORD/8 213 movq xmmE, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE] 214 pslldq xmmA, SIZEOF_MMWORD 215 por xmmA, xmmE 216 .column_ld4: 217 test cl, SIZEOF_XMMWORD/4 218 jz short .column_ld8 219 sub ecx, byte SIZEOF_XMMWORD/4 220 movdqa xmmE, xmmA 221 movdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE] 222 .column_ld8: 223 test cl, SIZEOF_XMMWORD/2 224 mov ecx, SIZEOF_XMMWORD 225 jz short .rgb_ycc_cnv 226 movdqa xmmF, xmmA 227 movdqa xmmH, xmmE 228 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] 229 movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD] 230 jmp short .rgb_ycc_cnv 231 ALIGNX 16, 7 232 233 .columnloop: 234 movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] 235 movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD] 236 movdqu xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD] 237 movdqu xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD] 238 239 .rgb_ycc_cnv: 240 ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) 241 ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) 242 ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) 243 ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) 244 245 movdqa xmmD, xmmA 246 punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35) 247 punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37) 248 249 movdqa xmmC, xmmF 250 punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D) 251 punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F) 252 253 movdqa xmmB, xmmA 254 punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C) 255 punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D) 256 257 movdqa xmmG, xmmD 258 punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E) 259 punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F) 260 261 movdqa xmmE, xmmA 262 punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) 263 punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E) 264 265 movdqa xmmH, xmmB 266 punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F) 267 punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F) 268 269 pxor xmmF, xmmF 270 271 movdqa xmmC, xmmA 272 punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E) 273 punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E) 274 275 movdqa xmmD, xmmB 276 punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F) 277 punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F) 278 279 movdqa xmmG, xmmE 280 punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E) 281 punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E) 282 283 punpcklbw xmmF, xmmH 284 punpckhbw xmmH, xmmH 285 psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F) 286 psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F) 287 288 %endif ; RGB_PIXELSIZE ; --------------- 289 290 ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE 291 ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO 292 293 ; (Original) 294 ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B 295 ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE 296 ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE 297 ; 298 ; (This implementation) 299 ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G 300 ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE 301 ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE 302 303 movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=RE 304 movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=RO 305 movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=BE 306 movdqa XMMWORD [wk(3)], xmm5 ; wk(3)=BO 307 308 movdqa xmm6, xmm1 309 punpcklwd xmm1, xmm3 310 punpckhwd xmm6, xmm3 311 movdqa xmm7, xmm1 312 movdqa xmm4, xmm6 313 pmaddwd xmm1, [GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337) 314 pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337) 315 pmaddwd xmm7, [GOTOFF(eax,PW_MF016_MF033)] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331) 316 pmaddwd xmm4, [GOTOFF(eax,PW_MF016_MF033)] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331) 317 318 movdqa XMMWORD [wk(4)], xmm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337) 319 movdqa XMMWORD [wk(5)], xmm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337) 320 321 pxor xmm1, xmm1 322 pxor xmm6, xmm6 323 punpcklwd xmm1, xmm5 ; xmm1=BOL 324 punpckhwd xmm6, xmm5 ; xmm6=BOH 325 psrld xmm1, 1 ; xmm1=BOL*FIX(0.500) 326 psrld xmm6, 1 ; xmm6=BOH*FIX(0.500) 327 328 movdqa xmm5, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm5=[PD_ONEHALFM1_CJ] 329 330 paddd xmm7, xmm1 331 paddd xmm4, xmm6 332 paddd xmm7, xmm5 333 paddd xmm4, xmm5 334 psrld xmm7, SCALEBITS ; xmm7=CbOL 335 psrld xmm4, SCALEBITS ; xmm4=CbOH 336 packssdw xmm7, xmm4 ; xmm7=CbO 337 338 movdqa xmm1, XMMWORD [wk(2)] ; xmm1=BE 339 340 movdqa xmm6, xmm0 341 punpcklwd xmm0, xmm2 342 punpckhwd xmm6, xmm2 343 movdqa xmm5, xmm0 344 movdqa xmm4, xmm6 345 pmaddwd xmm0, [GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337) 346 pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337) 347 pmaddwd xmm5, [GOTOFF(eax,PW_MF016_MF033)] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331) 348 pmaddwd xmm4, [GOTOFF(eax,PW_MF016_MF033)] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331) 349 350 movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337) 351 movdqa XMMWORD [wk(7)], xmm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337) 352 353 pxor xmm0, xmm0 354 pxor xmm6, xmm6 355 punpcklwd xmm0, xmm1 ; xmm0=BEL 356 punpckhwd xmm6, xmm1 ; xmm6=BEH 357 psrld xmm0, 1 ; xmm0=BEL*FIX(0.500) 358 psrld xmm6, 1 ; xmm6=BEH*FIX(0.500) 359 360 movdqa xmm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ] 361 362 paddd xmm5, xmm0 363 paddd xmm4, xmm6 364 paddd xmm5, xmm1 365 paddd xmm4, xmm1 366 psrld xmm5, SCALEBITS ; xmm5=CbEL 367 psrld xmm4, SCALEBITS ; xmm4=CbEH 368 packssdw xmm5, xmm4 ; xmm5=CbE 369 370 psllw xmm7, BYTE_BIT 371 por xmm5, xmm7 ; xmm5=Cb 372 movdqa XMMWORD [ebx], xmm5 ; Save Cb 373 374 movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO 375 movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE 376 movdqa xmm1, XMMWORD [wk(1)] ; xmm1=RO 377 378 movdqa xmm4, xmm0 379 punpcklwd xmm0, xmm3 380 punpckhwd xmm4, xmm3 381 movdqa xmm7, xmm0 382 movdqa xmm5, xmm4 383 pmaddwd xmm0, [GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250) 384 pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250) 385 pmaddwd xmm7, [GOTOFF(eax,PW_MF008_MF041)] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418) 386 pmaddwd xmm5, [GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418) 387 388 movdqa xmm3, [GOTOFF(eax,PD_ONEHALF)] ; xmm3=[PD_ONEHALF] 389 390 paddd xmm0, XMMWORD [wk(4)] 391 paddd xmm4, XMMWORD [wk(5)] 392 paddd xmm0, xmm3 393 paddd xmm4, xmm3 394 psrld xmm0, SCALEBITS ; xmm0=YOL 395 psrld xmm4, SCALEBITS ; xmm4=YOH 396 packssdw xmm0, xmm4 ; xmm0=YO 397 398 pxor xmm3, xmm3 399 pxor xmm4, xmm4 400 punpcklwd xmm3, xmm1 ; xmm3=ROL 401 punpckhwd xmm4, xmm1 ; xmm4=ROH 402 psrld xmm3, 1 ; xmm3=ROL*FIX(0.500) 403 psrld xmm4, 1 ; xmm4=ROH*FIX(0.500) 404 405 movdqa xmm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ] 406 407 paddd xmm7, xmm3 408 paddd xmm5, xmm4 409 paddd xmm7, xmm1 410 paddd xmm5, xmm1 411 psrld xmm7, SCALEBITS ; xmm7=CrOL 412 psrld xmm5, SCALEBITS ; xmm5=CrOH 413 packssdw xmm7, xmm5 ; xmm7=CrO 414 415 movdqa xmm3, XMMWORD [wk(0)] ; xmm3=RE 416 417 movdqa xmm4, xmm6 418 punpcklwd xmm6, xmm2 419 punpckhwd xmm4, xmm2 420 movdqa xmm1, xmm6 421 movdqa xmm5, xmm4 422 pmaddwd xmm6, [GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250) 423 pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250) 424 pmaddwd xmm1, [GOTOFF(eax,PW_MF008_MF041)] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418) 425 pmaddwd xmm5, [GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418) 426 427 movdqa xmm2, [GOTOFF(eax,PD_ONEHALF)] ; xmm2=[PD_ONEHALF] 428 429 paddd xmm6, XMMWORD [wk(6)] 430 paddd xmm4, XMMWORD [wk(7)] 431 paddd xmm6, xmm2 432 paddd xmm4, xmm2 433 psrld xmm6, SCALEBITS ; xmm6=YEL 434 psrld xmm4, SCALEBITS ; xmm4=YEH 435 packssdw xmm6, xmm4 ; xmm6=YE 436 437 psllw xmm0, BYTE_BIT 438 por xmm6, xmm0 ; xmm6=Y 439 movdqa XMMWORD [edi], xmm6 ; Save Y 440 441 pxor xmm2, xmm2 442 pxor xmm4, xmm4 443 punpcklwd xmm2, xmm3 ; xmm2=REL 444 punpckhwd xmm4, xmm3 ; xmm4=REH 445 psrld xmm2, 1 ; xmm2=REL*FIX(0.500) 446 psrld xmm4, 1 ; xmm4=REH*FIX(0.500) 447 448 movdqa xmm0, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm0=[PD_ONEHALFM1_CJ] 449 450 paddd xmm1, xmm2 451 paddd xmm5, xmm4 452 paddd xmm1, xmm0 453 paddd xmm5, xmm0 454 psrld xmm1, SCALEBITS ; xmm1=CrEL 455 psrld xmm5, SCALEBITS ; xmm5=CrEH 456 packssdw xmm1, xmm5 ; xmm1=CrE 457 458 psllw xmm7, BYTE_BIT 459 por xmm1, xmm7 ; xmm1=Cr 460 movdqa XMMWORD [edx], xmm1 ; Save Cr 461 462 sub ecx, byte SIZEOF_XMMWORD 463 add esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr 464 add edi, byte SIZEOF_XMMWORD ; outptr0 465 add ebx, byte SIZEOF_XMMWORD ; outptr1 466 add edx, byte SIZEOF_XMMWORD ; outptr2 467 cmp ecx, byte SIZEOF_XMMWORD 468 jae near .columnloop 469 test ecx, ecx 470 jnz near .column_ld1 471 472 pop ecx ; col 473 pop esi 474 pop edi 475 pop ebx 476 pop edx 477 POPPIC eax 478 479 add esi, byte SIZEOF_JSAMPROW ; input_buf 480 add edi, byte SIZEOF_JSAMPROW 481 add ebx, byte SIZEOF_JSAMPROW 482 add edx, byte SIZEOF_JSAMPROW 483 dec eax ; num_rows 484 jg near .rowloop 485 486 .return: 487 pop edi 488 pop esi 489 ; pop edx ; need not be preserved 490 ; pop ecx ; need not be preserved 491 pop ebx 492 mov esp, ebp ; esp <- aligned ebp 493 pop esp ; esp <- original ebp 494 pop ebp 495 ret 496 497 ; For some reason, the OS X linker does not honor the request to align the 498 ; segment unless we do this. 499 align 32