1 ; 2 ; jdclrss2-64.asm - colorspace conversion (64-bit SSE2) 3 ; 4 ; Copyright 2009, 2012 Pierre Ossman <ossman (a] cendio.se> for Cendio AB 5 ; Copyright 2009, 2012 D. R. Commander 6 ; 7 ; Based on 8 ; x86 SIMD extension for IJG JPEG library 9 ; Copyright (C) 1999-2006, MIYASAKA Masaru. 10 ; For conditions of distribution and use, see copyright notice in jsimdext.inc 11 ; 12 ; This file should be assembled with NASM (Netwide Assembler), 13 ; can *not* be assembled with Microsoft's MASM or any compatible 14 ; assembler (including Borland's Turbo Assembler). 15 ; NASM is available from http://nasm.sourceforge.net/ or 16 ; http://sourceforge.net/project/showfiles.php?group_id=6208 17 ; 18 ; [TAB8] 19 20 %include "jcolsamp.inc" 21 22 ; -------------------------------------------------------------------------- 23 ; 24 ; Convert some rows of samples to the output colorspace. 25 ; 26 ; GLOBAL(void) 27 ; jsimd_ycc_rgb_convert_sse2 (JDIMENSION out_width, 28 ; JSAMPIMAGE input_buf, JDIMENSION input_row, 29 ; JSAMPARRAY output_buf, int num_rows) 30 ; 31 32 ; r10 = JDIMENSION out_width 33 ; r11 = JSAMPIMAGE input_buf 34 ; r12 = JDIMENSION input_row 35 ; r13 = JSAMPARRAY output_buf 36 ; r14 = int num_rows 37 38 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] 39 %define WK_NUM 2 40 41 align 16 42 global EXTN(jsimd_ycc_rgb_convert_sse2) PRIVATE 43 44 EXTN(jsimd_ycc_rgb_convert_sse2): 45 push rbp 46 mov rax,rsp ; rax = original rbp 47 sub rsp, byte 4 48 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits 49 mov [rsp],rax 50 mov rbp,rsp ; rbp = aligned rbp 51 lea rsp, [wk(0)] 52 collect_args 53 push rbx 54 55 mov rcx, r10 ; num_cols 56 test rcx,rcx 57 jz near .return 58 59 push rcx 60 61 mov rdi, r11 62 mov rcx, r12 63 mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] 64 mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] 65 mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] 66 lea rsi, [rsi+rcx*SIZEOF_JSAMPROW] 67 lea rbx, [rbx+rcx*SIZEOF_JSAMPROW] 68 lea rdx, [rdx+rcx*SIZEOF_JSAMPROW] 69 70 pop rcx 71 72 mov rdi, r13 73 mov eax, r14d 74 test rax,rax 75 jle near .return 76 .rowloop: 77 push rax 78 push rdi 79 push rdx 80 push rbx 81 push rsi 82 push rcx ; col 83 84 mov rsi, JSAMPROW [rsi] ; inptr0 85 mov rbx, JSAMPROW [rbx] ; inptr1 86 mov rdx, JSAMPROW [rdx] ; inptr2 87 mov rdi, JSAMPROW [rdi] ; outptr 88 .columnloop: 89 90 movdqa xmm5, XMMWORD [rbx] ; xmm5=Cb(0123456789ABCDEF) 91 movdqa xmm1, XMMWORD [rdx] ; xmm1=Cr(0123456789ABCDEF) 92 93 pcmpeqw xmm4,xmm4 94 pcmpeqw xmm7,xmm7 95 psrlw xmm4,BYTE_BIT 96 psllw xmm7,7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} 97 movdqa xmm0,xmm4 ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..} 98 99 pand xmm4,xmm5 ; xmm4=Cb(02468ACE)=CbE 100 psrlw xmm5,BYTE_BIT ; xmm5=Cb(13579BDF)=CbO 101 pand xmm0,xmm1 ; xmm0=Cr(02468ACE)=CrE 102 psrlw xmm1,BYTE_BIT ; xmm1=Cr(13579BDF)=CrO 103 104 paddw xmm4,xmm7 105 paddw xmm5,xmm7 106 paddw xmm0,xmm7 107 paddw xmm1,xmm7 108 109 ; (Original) 110 ; R = Y + 1.40200 * Cr 111 ; G = Y - 0.34414 * Cb - 0.71414 * Cr 112 ; B = Y + 1.77200 * Cb 113 ; 114 ; (This implementation) 115 ; R = Y + 0.40200 * Cr + Cr 116 ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr 117 ; B = Y - 0.22800 * Cb + Cb + Cb 118 119 movdqa xmm2,xmm4 ; xmm2=CbE 120 movdqa xmm3,xmm5 ; xmm3=CbO 121 paddw xmm4,xmm4 ; xmm4=2*CbE 122 paddw xmm5,xmm5 ; xmm5=2*CbO 123 movdqa xmm6,xmm0 ; xmm6=CrE 124 movdqa xmm7,xmm1 ; xmm7=CrO 125 paddw xmm0,xmm0 ; xmm0=2*CrE 126 paddw xmm1,xmm1 ; xmm1=2*CrO 127 128 pmulhw xmm4,[rel PW_MF0228] ; xmm4=(2*CbE * -FIX(0.22800)) 129 pmulhw xmm5,[rel PW_MF0228] ; xmm5=(2*CbO * -FIX(0.22800)) 130 pmulhw xmm0,[rel PW_F0402] ; xmm0=(2*CrE * FIX(0.40200)) 131 pmulhw xmm1,[rel PW_F0402] ; xmm1=(2*CrO * FIX(0.40200)) 132 133 paddw xmm4,[rel PW_ONE] 134 paddw xmm5,[rel PW_ONE] 135 psraw xmm4,1 ; xmm4=(CbE * -FIX(0.22800)) 136 psraw xmm5,1 ; xmm5=(CbO * -FIX(0.22800)) 137 paddw xmm0,[rel PW_ONE] 138 paddw xmm1,[rel PW_ONE] 139 psraw xmm0,1 ; xmm0=(CrE * FIX(0.40200)) 140 psraw xmm1,1 ; xmm1=(CrO * FIX(0.40200)) 141 142 paddw xmm4,xmm2 143 paddw xmm5,xmm3 144 paddw xmm4,xmm2 ; xmm4=(CbE * FIX(1.77200))=(B-Y)E 145 paddw xmm5,xmm3 ; xmm5=(CbO * FIX(1.77200))=(B-Y)O 146 paddw xmm0,xmm6 ; xmm0=(CrE * FIX(1.40200))=(R-Y)E 147 paddw xmm1,xmm7 ; xmm1=(CrO * FIX(1.40200))=(R-Y)O 148 149 movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=(B-Y)E 150 movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(B-Y)O 151 152 movdqa xmm4,xmm2 153 movdqa xmm5,xmm3 154 punpcklwd xmm2,xmm6 155 punpckhwd xmm4,xmm6 156 pmaddwd xmm2,[rel PW_MF0344_F0285] 157 pmaddwd xmm4,[rel PW_MF0344_F0285] 158 punpcklwd xmm3,xmm7 159 punpckhwd xmm5,xmm7 160 pmaddwd xmm3,[rel PW_MF0344_F0285] 161 pmaddwd xmm5,[rel PW_MF0344_F0285] 162 163 paddd xmm2,[rel PD_ONEHALF] 164 paddd xmm4,[rel PD_ONEHALF] 165 psrad xmm2,SCALEBITS 166 psrad xmm4,SCALEBITS 167 paddd xmm3,[rel PD_ONEHALF] 168 paddd xmm5,[rel PD_ONEHALF] 169 psrad xmm3,SCALEBITS 170 psrad xmm5,SCALEBITS 171 172 packssdw xmm2,xmm4 ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285) 173 packssdw xmm3,xmm5 ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285) 174 psubw xmm2,xmm6 ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E 175 psubw xmm3,xmm7 ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O 176 177 movdqa xmm5, XMMWORD [rsi] ; xmm5=Y(0123456789ABCDEF) 178 179 pcmpeqw xmm4,xmm4 180 psrlw xmm4,BYTE_BIT ; xmm4={0xFF 0x00 0xFF 0x00 ..} 181 pand xmm4,xmm5 ; xmm4=Y(02468ACE)=YE 182 psrlw xmm5,BYTE_BIT ; xmm5=Y(13579BDF)=YO 183 184 paddw xmm0,xmm4 ; xmm0=((R-Y)E+YE)=RE=R(02468ACE) 185 paddw xmm1,xmm5 ; xmm1=((R-Y)O+YO)=RO=R(13579BDF) 186 packuswb xmm0,xmm0 ; xmm0=R(02468ACE********) 187 packuswb xmm1,xmm1 ; xmm1=R(13579BDF********) 188 189 paddw xmm2,xmm4 ; xmm2=((G-Y)E+YE)=GE=G(02468ACE) 190 paddw xmm3,xmm5 ; xmm3=((G-Y)O+YO)=GO=G(13579BDF) 191 packuswb xmm2,xmm2 ; xmm2=G(02468ACE********) 192 packuswb xmm3,xmm3 ; xmm3=G(13579BDF********) 193 194 paddw xmm4, XMMWORD [wk(0)] ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE) 195 paddw xmm5, XMMWORD [wk(1)] ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF) 196 packuswb xmm4,xmm4 ; xmm4=B(02468ACE********) 197 packuswb xmm5,xmm5 ; xmm5=B(13579BDF********) 198 199 %if RGB_PIXELSIZE == 3 ; --------------- 200 201 ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) 202 ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) 203 ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) 204 ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **) 205 206 punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) 207 punpcklbw xmmE,xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F) 208 punpcklbw xmmD,xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F) 209 210 movdqa xmmG,xmmA 211 movdqa xmmH,xmmA 212 punpcklwd xmmA,xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07) 213 punpckhwd xmmG,xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F) 214 215 psrldq xmmH,2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --) 216 psrldq xmmE,2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --) 217 218 movdqa xmmC,xmmD 219 movdqa xmmB,xmmD 220 punpcklwd xmmD,xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18) 221 punpckhwd xmmC,xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --) 222 223 psrldq xmmB,2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --) 224 225 movdqa xmmF,xmmE 226 punpcklwd xmmE,xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29) 227 punpckhwd xmmF,xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --) 228 229 pshufd xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03) 230 movdqa xmmB,xmmE 231 punpckldq xmmA,xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14) 232 punpckldq xmmE,xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07) 233 punpckhdq xmmD,xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29) 234 235 pshufd xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B) 236 movdqa xmmB,xmmF 237 punpckldq xmmG,xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C) 238 punpckldq xmmF,xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F) 239 punpckhdq xmmC,xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --) 240 241 punpcklqdq xmmA,xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) 242 punpcklqdq xmmD,xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) 243 punpcklqdq xmmF,xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) 244 245 cmp rcx, byte SIZEOF_XMMWORD 246 jb short .column_st32 247 248 test rdi, SIZEOF_XMMWORD-1 249 jnz short .out1 250 ; --(aligned)------------------- 251 movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA 252 movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD 253 movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF 254 jmp short .out0 255 .out1: ; --(unaligned)----------------- 256 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA 257 movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD 258 movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF 259 .out0: 260 add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr 261 sub rcx, byte SIZEOF_XMMWORD 262 jz near .nextrow 263 264 add rsi, byte SIZEOF_XMMWORD ; inptr0 265 add rbx, byte SIZEOF_XMMWORD ; inptr1 266 add rdx, byte SIZEOF_XMMWORD ; inptr2 267 jmp near .columnloop 268 269 .column_st32: 270 lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE 271 cmp rcx, byte 2*SIZEOF_XMMWORD 272 jb short .column_st16 273 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA 274 movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD 275 add rdi, byte 2*SIZEOF_XMMWORD ; outptr 276 movdqa xmmA,xmmF 277 sub rcx, byte 2*SIZEOF_XMMWORD 278 jmp short .column_st15 279 .column_st16: 280 cmp rcx, byte SIZEOF_XMMWORD 281 jb short .column_st15 282 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA 283 add rdi, byte SIZEOF_XMMWORD ; outptr 284 movdqa xmmA,xmmD 285 sub rcx, byte SIZEOF_XMMWORD 286 .column_st15: 287 ; Store the lower 8 bytes of xmmA to the output when it has enough 288 ; space. 289 cmp rcx, byte SIZEOF_MMWORD 290 jb short .column_st7 291 movq XMM_MMWORD [rdi], xmmA 292 add rdi, byte SIZEOF_MMWORD 293 sub rcx, byte SIZEOF_MMWORD 294 psrldq xmmA, SIZEOF_MMWORD 295 .column_st7: 296 ; Store the lower 4 bytes of xmmA to the output when it has enough 297 ; space. 298 cmp rcx, byte SIZEOF_DWORD 299 jb short .column_st3 300 movd XMM_DWORD [rdi], xmmA 301 add rdi, byte SIZEOF_DWORD 302 sub rcx, byte SIZEOF_DWORD 303 psrldq xmmA, SIZEOF_DWORD 304 .column_st3: 305 ; Store the lower 2 bytes of rax to the output when it has enough 306 ; space. 307 movd eax, xmmA 308 cmp rcx, byte SIZEOF_WORD 309 jb short .column_st1 310 mov WORD [rdi], ax 311 add rdi, byte SIZEOF_WORD 312 sub rcx, byte SIZEOF_WORD 313 shr rax, 16 314 .column_st1: 315 ; Store the lower 1 byte of rax to the output when it has enough 316 ; space. 317 test rcx, rcx 318 jz short .nextrow 319 mov BYTE [rdi], al 320 321 %else ; RGB_PIXELSIZE == 4 ; ----------- 322 323 %ifdef RGBX_FILLER_0XFF 324 pcmpeqb xmm6,xmm6 ; xmm6=XE=X(02468ACE********) 325 pcmpeqb xmm7,xmm7 ; xmm7=XO=X(13579BDF********) 326 %else 327 pxor xmm6,xmm6 ; xmm6=XE=X(02468ACE********) 328 pxor xmm7,xmm7 ; xmm7=XO=X(13579BDF********) 329 %endif 330 ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) 331 ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) 332 ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) 333 ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **) 334 335 punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) 336 punpcklbw xmmE,xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E) 337 punpcklbw xmmB,xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F) 338 punpcklbw xmmF,xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F) 339 340 movdqa xmmC,xmmA 341 punpcklwd xmmA,xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36) 342 punpckhwd xmmC,xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E) 343 movdqa xmmG,xmmB 344 punpcklwd xmmB,xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37) 345 punpckhwd xmmG,xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F) 346 347 movdqa xmmD,xmmA 348 punpckldq xmmA,xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) 349 punpckhdq xmmD,xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) 350 movdqa xmmH,xmmC 351 punpckldq xmmC,xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) 352 punpckhdq xmmH,xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) 353 354 cmp rcx, byte SIZEOF_XMMWORD 355 jb short .column_st32 356 357 test rdi, SIZEOF_XMMWORD-1 358 jnz short .out1 359 ; --(aligned)------------------- 360 movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA 361 movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD 362 movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC 363 movntdq XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH 364 jmp short .out0 365 .out1: ; --(unaligned)----------------- 366 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA 367 movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD 368 movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC 369 movdqu XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH 370 .out0: 371 add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr 372 sub rcx, byte SIZEOF_XMMWORD 373 jz near .nextrow 374 375 add rsi, byte SIZEOF_XMMWORD ; inptr0 376 add rbx, byte SIZEOF_XMMWORD ; inptr1 377 add rdx, byte SIZEOF_XMMWORD ; inptr2 378 jmp near .columnloop 379 380 .column_st32: 381 cmp rcx, byte SIZEOF_XMMWORD/2 382 jb short .column_st16 383 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA 384 movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD 385 add rdi, byte 2*SIZEOF_XMMWORD ; outptr 386 movdqa xmmA,xmmC 387 movdqa xmmD,xmmH 388 sub rcx, byte SIZEOF_XMMWORD/2 389 .column_st16: 390 cmp rcx, byte SIZEOF_XMMWORD/4 391 jb short .column_st15 392 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA 393 add rdi, byte SIZEOF_XMMWORD ; outptr 394 movdqa xmmA,xmmD 395 sub rcx, byte SIZEOF_XMMWORD/4 396 .column_st15: 397 ; Store two pixels (8 bytes) of xmmA to the output when it has enough 398 ; space. 399 cmp rcx, byte SIZEOF_XMMWORD/8 400 jb short .column_st7 401 movq MMWORD [rdi], xmmA 402 add rdi, byte SIZEOF_XMMWORD/8*4 403 sub rcx, byte SIZEOF_XMMWORD/8 404 psrldq xmmA, SIZEOF_XMMWORD/8*4 405 .column_st7: 406 ; Store one pixel (4 bytes) of xmmA to the output when it has enough 407 ; space. 408 test rcx, rcx 409 jz short .nextrow 410 movd XMM_DWORD [rdi], xmmA 411 412 %endif ; RGB_PIXELSIZE ; --------------- 413 414 .nextrow: 415 pop rcx 416 pop rsi 417 pop rbx 418 pop rdx 419 pop rdi 420 pop rax 421 422 add rsi, byte SIZEOF_JSAMPROW 423 add rbx, byte SIZEOF_JSAMPROW 424 add rdx, byte SIZEOF_JSAMPROW 425 add rdi, byte SIZEOF_JSAMPROW ; output_buf 426 dec rax ; num_rows 427 jg near .rowloop 428 429 sfence ; flush the write buffer 430 431 .return: 432 pop rbx 433 uncollect_args 434 mov rsp,rbp ; rsp <- aligned rbp 435 pop rsp ; rsp <- original rbp 436 pop rbp 437 ret 438 439 ; For some reason, the OS X linker does not honor the request to align the 440 ; segment unless we do this. 441 align 16 442
