jdmrgext-sse2.asm (20995B)
1 ; 2 ; jdmrgext.asm - merged upsampling/color conversion (SSE2) 3 ; 4 ; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB 5 ; Copyright (C) 2012, 2016, 2024, D. R. Commander. 6 ; 7 ; Based on the x86 SIMD extension for IJG JPEG library 8 ; Copyright (C) 1999-2006, MIYASAKA Masaru. 9 ; For conditions of distribution and use, see copyright notice in jsimdext.inc 10 ; 11 ; This file should be assembled with NASM (Netwide Assembler) or Yasm. 12 13 %include "jcolsamp.inc" 14 15 ; -------------------------------------------------------------------------- 16 ; 17 ; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical. 18 ; 19 ; GLOBAL(void) 20 ; jsimd_h2v1_merged_upsample_sse2(JDIMENSION output_width, 21 ; JSAMPIMAGE input_buf, 22 ; JDIMENSION in_row_group_ctr, 23 ; JSAMPARRAY output_buf); 24 ; 25 26 %define output_width(b) (b) + 8 ; JDIMENSION output_width 27 %define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf 28 %define in_row_group_ctr(b) (b) + 16 ; JDIMENSION in_row_group_ctr 29 %define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf 30 31 %define original_ebp ebp + 0 32 %define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD 33 ; xmmword wk[WK_NUM] 34 %define WK_NUM 3 35 %define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr 36 37 align 32 38 GLOBAL_FUNCTION(jsimd_h2v1_merged_upsample_sse2) 39 40 EXTN(jsimd_h2v1_merged_upsample_sse2): 41 push ebp 42 mov eax, esp ; eax = original ebp 43 sub esp, byte 4 44 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits 45 mov [esp], eax 46 mov ebp, esp ; ebp = aligned ebp 47 lea esp, [wk(0)] 48 PUSHPIC eax ; make a room for GOT address 49 push ebx 50 ; push ecx ; need not be preserved 51 ; push edx ; need not be preserved 52 push esi 53 push edi 54 55 GET_GOT ebx ; get GOT address 56 MOVPIC POINTER [gotptr], ebx ; save GOT address 57 58 mov ecx, JDIMENSION [output_width(eax)] ; col 59 test ecx, ecx 60 jz near .return 61 62 push ecx 63 64 mov edi, JSAMPIMAGE [input_buf(eax)] 65 mov ecx, JDIMENSION [in_row_group_ctr(eax)] 66 mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] 67 mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] 68 mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] 69 mov edi, JSAMPARRAY [output_buf(eax)] 70 mov esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW] ; inptr0 71 mov ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW] ; inptr1 72 mov edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW] ; inptr2 73 mov edi, JSAMPROW [edi] ; outptr 74 75 pop ecx ; col 76 77 ALIGNX 16, 7 78 .columnloop: 79 MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) 80 81 movdqa xmm6, XMMWORD [ebx] ; xmm6=Cb(0123456789ABCDEF) 82 movdqa xmm7, XMMWORD [edx] ; xmm7=Cr(0123456789ABCDEF) 83 84 pxor xmm1, xmm1 ; xmm1=(all 0's) 85 pcmpeqw xmm3, xmm3 86 psllw xmm3, 7 ; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..} 87 88 movdqa xmm4, xmm6 89 punpckhbw xmm6, xmm1 ; xmm6=Cb(89ABCDEF)=CbH 90 punpcklbw xmm4, xmm1 ; xmm4=Cb(01234567)=CbL 91 movdqa xmm0, xmm7 92 punpckhbw xmm7, xmm1 ; xmm7=Cr(89ABCDEF)=CrH 93 punpcklbw xmm0, xmm1 ; xmm0=Cr(01234567)=CrL 94 95 paddw xmm6, xmm3 96 paddw xmm4, xmm3 97 paddw xmm7, xmm3 98 paddw xmm0, xmm3 99 100 ; (Original) 101 ; R = Y + 1.40200 * Cr 102 ; G = Y - 0.34414 * Cb - 0.71414 * Cr 103 ; B = Y + 1.77200 * Cb 104 ; 105 ; (This implementation) 106 ; R = Y + 0.40200 * Cr + Cr 107 ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr 108 ; B = Y - 0.22800 * Cb + Cb + Cb 109 110 movdqa xmm5, xmm6 ; xmm5=CbH 111 movdqa xmm2, xmm4 ; xmm2=CbL 112 paddw xmm6, xmm6 ; xmm6=2*CbH 113 paddw xmm4, xmm4 ; xmm4=2*CbL 114 movdqa xmm1, xmm7 ; xmm1=CrH 115 movdqa xmm3, xmm0 ; xmm3=CrL 116 paddw xmm7, xmm7 ; xmm7=2*CrH 117 paddw xmm0, xmm0 ; xmm0=2*CrL 118 119 pmulhw xmm6, [GOTOFF(eax,PW_MF0228)] ; xmm6=(2*CbH * -FIX(0.22800)) 120 pmulhw xmm4, [GOTOFF(eax,PW_MF0228)] ; xmm4=(2*CbL * -FIX(0.22800)) 121 pmulhw xmm7, [GOTOFF(eax,PW_F0402)] ; xmm7=(2*CrH * FIX(0.40200)) 122 pmulhw xmm0, [GOTOFF(eax,PW_F0402)] ; xmm0=(2*CrL * FIX(0.40200)) 123 124 paddw xmm6, [GOTOFF(eax,PW_ONE)] 125 paddw xmm4, [GOTOFF(eax,PW_ONE)] 126 psraw xmm6, 1 ; xmm6=(CbH * -FIX(0.22800)) 127 psraw xmm4, 1 ; xmm4=(CbL * -FIX(0.22800)) 128 paddw xmm7, [GOTOFF(eax,PW_ONE)] 129 paddw xmm0, [GOTOFF(eax,PW_ONE)] 130 psraw xmm7, 1 ; xmm7=(CrH * FIX(0.40200)) 131 psraw xmm0, 1 ; xmm0=(CrL * FIX(0.40200)) 132 133 paddw xmm6, xmm5 134 paddw xmm4, xmm2 135 paddw xmm6, xmm5 ; xmm6=(CbH * FIX(1.77200))=(B-Y)H 136 paddw xmm4, xmm2 ; xmm4=(CbL * FIX(1.77200))=(B-Y)L 137 paddw xmm7, xmm1 ; xmm7=(CrH * FIX(1.40200))=(R-Y)H 138 paddw xmm0, xmm3 ; xmm0=(CrL * FIX(1.40200))=(R-Y)L 139 140 movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=(B-Y)H 141 movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(R-Y)H 142 143 movdqa xmm6, xmm5 144 movdqa xmm7, xmm2 145 punpcklwd xmm5, xmm1 146 punpckhwd xmm6, xmm1 147 pmaddwd xmm5, [GOTOFF(eax,PW_MF0344_F0285)] 148 pmaddwd xmm6, [GOTOFF(eax,PW_MF0344_F0285)] 149 punpcklwd xmm2, xmm3 150 punpckhwd xmm7, xmm3 151 pmaddwd xmm2, [GOTOFF(eax,PW_MF0344_F0285)] 152 pmaddwd xmm7, [GOTOFF(eax,PW_MF0344_F0285)] 153 154 paddd xmm5, [GOTOFF(eax,PD_ONEHALF)] 155 paddd xmm6, [GOTOFF(eax,PD_ONEHALF)] 156 psrad xmm5, SCALEBITS 157 psrad xmm6, SCALEBITS 158 paddd xmm2, [GOTOFF(eax,PD_ONEHALF)] 159 paddd xmm7, [GOTOFF(eax,PD_ONEHALF)] 160 psrad xmm2, SCALEBITS 161 psrad xmm7, SCALEBITS 162 163 packssdw xmm5, xmm6 ; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285) 164 packssdw xmm2, xmm7 ; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285) 165 psubw xmm5, xmm1 ; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H 166 psubw xmm2, xmm3 ; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L 167 168 movdqa XMMWORD [wk(2)], xmm5 ; wk(2)=(G-Y)H 169 170 mov al, 2 ; Yctr 171 jmp short .Yloop_1st 172 ALIGNX 16, 7 173 174 .Yloop_2nd: 175 movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(R-Y)H 176 movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(G-Y)H 177 movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(B-Y)H 178 ALIGNX 16, 7 179 180 .Yloop_1st: 181 movdqa xmm7, XMMWORD [esi] ; xmm7=Y(0123456789ABCDEF) 182 183 pcmpeqw xmm6, xmm6 184 psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..} 185 pand xmm6, xmm7 ; xmm6=Y(02468ACE)=YE 186 psrlw xmm7, BYTE_BIT ; xmm7=Y(13579BDF)=YO 187 188 movdqa xmm1, xmm0 ; xmm1=xmm0=(R-Y)(L/H) 189 movdqa xmm3, xmm2 ; xmm3=xmm2=(G-Y)(L/H) 190 movdqa xmm5, xmm4 ; xmm5=xmm4=(B-Y)(L/H) 191 192 paddw xmm0, xmm6 ; xmm0=((R-Y)+YE)=RE=R(02468ACE) 193 paddw xmm1, xmm7 ; xmm1=((R-Y)+YO)=RO=R(13579BDF) 194 packuswb xmm0, xmm0 ; xmm0=R(02468ACE********) 195 packuswb xmm1, xmm1 ; xmm1=R(13579BDF********) 196 197 paddw xmm2, xmm6 ; xmm2=((G-Y)+YE)=GE=G(02468ACE) 198 paddw xmm3, xmm7 ; xmm3=((G-Y)+YO)=GO=G(13579BDF) 199 packuswb xmm2, xmm2 ; xmm2=G(02468ACE********) 200 packuswb xmm3, xmm3 ; xmm3=G(13579BDF********) 201 202 paddw xmm4, xmm6 ; xmm4=((B-Y)+YE)=BE=B(02468ACE) 203 paddw xmm5, xmm7 ; xmm5=((B-Y)+YO)=BO=B(13579BDF) 204 packuswb xmm4, xmm4 ; xmm4=B(02468ACE********) 205 packuswb xmm5, xmm5 ; xmm5=B(13579BDF********) 206 207 %if RGB_PIXELSIZE == 3 ; --------------- 208 209 ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) 210 ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) 211 ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) 212 ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **) 213 214 punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) 215 punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F) 216 punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F) 217 218 movdqa xmmG, xmmA 219 movdqa xmmH, xmmA 220 punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07) 221 punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F) 222 223 psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --) 224 psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --) 225 226 movdqa xmmC, xmmD 227 movdqa xmmB, xmmD 228 punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18) 229 punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --) 230 231 psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --) 232 233 movdqa xmmF, xmmE 234 punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29) 235 punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --) 236 237 pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03) 238 movdqa xmmB, xmmE 239 punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14) 240 punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07) 241 punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29) 242 243 pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B) 244 movdqa xmmB, xmmF 245 punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C) 246 punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F) 247 punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --) 248 249 punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) 250 punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) 251 punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) 252 253 cmp ecx, byte SIZEOF_XMMWORD 254 jb short .column_st32 255 256 test edi, SIZEOF_XMMWORD-1 257 jnz short .out1 258 ; --(aligned)------------------- 259 movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA 260 movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD 261 movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF 262 jmp short .out0 263 .out1: ; --(unaligned)----------------- 264 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA 265 movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD 266 movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF 267 .out0: 268 add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr 269 sub ecx, byte SIZEOF_XMMWORD 270 jz near .endcolumn 271 272 add esi, byte SIZEOF_XMMWORD ; inptr0 273 dec al ; Yctr 274 jnz near .Yloop_2nd 275 276 add ebx, byte SIZEOF_XMMWORD ; inptr1 277 add edx, byte SIZEOF_XMMWORD ; inptr2 278 jmp near .columnloop 279 ALIGNX 16, 7 280 281 .column_st32: 282 lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE 283 cmp ecx, byte 2*SIZEOF_XMMWORD 284 jb short .column_st16 285 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA 286 movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD 287 add edi, byte 2*SIZEOF_XMMWORD ; outptr 288 movdqa xmmA, xmmF 289 sub ecx, byte 2*SIZEOF_XMMWORD 290 jmp short .column_st15 291 .column_st16: 292 cmp ecx, byte SIZEOF_XMMWORD 293 jb short .column_st15 294 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA 295 add edi, byte SIZEOF_XMMWORD ; outptr 296 movdqa xmmA, xmmD 297 sub ecx, byte SIZEOF_XMMWORD 298 .column_st15: 299 ; Store the lower 8 bytes of xmmA to the output when it has enough 300 ; space. 301 cmp ecx, byte SIZEOF_MMWORD 302 jb short .column_st7 303 movq XMM_MMWORD [edi], xmmA 304 add edi, byte SIZEOF_MMWORD 305 sub ecx, byte SIZEOF_MMWORD 306 psrldq xmmA, SIZEOF_MMWORD 307 .column_st7: 308 ; Store the lower 4 bytes of xmmA to the output when it has enough 309 ; space. 310 cmp ecx, byte SIZEOF_DWORD 311 jb short .column_st3 312 movd XMM_DWORD [edi], xmmA 313 add edi, byte SIZEOF_DWORD 314 sub ecx, byte SIZEOF_DWORD 315 psrldq xmmA, SIZEOF_DWORD 316 .column_st3: 317 ; Store the lower 2 bytes of eax to the output when it has enough 318 ; space. 319 movd eax, xmmA 320 cmp ecx, byte SIZEOF_WORD 321 jb short .column_st1 322 mov word [edi], ax 323 add edi, byte SIZEOF_WORD 324 sub ecx, byte SIZEOF_WORD 325 shr eax, 16 326 .column_st1: 327 ; Store the lower 1 byte of eax to the output when it has enough 328 ; space. 329 test ecx, ecx 330 jz short .endcolumn 331 mov byte [edi], al 332 333 %else ; RGB_PIXELSIZE == 4 ; ----------- 334 335 %ifdef RGBX_FILLER_0XFF 336 pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********) 337 pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********) 338 %else 339 pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********) 340 pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********) 341 %endif 342 ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) 343 ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) 344 ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) 345 ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **) 346 347 punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) 348 punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E) 349 punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F) 350 punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F) 351 352 movdqa xmmC, xmmA 353 punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36) 354 punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E) 355 movdqa xmmG, xmmB 356 punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37) 357 punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F) 358 359 movdqa xmmD, xmmA 360 punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) 361 punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) 362 movdqa xmmH, xmmC 363 punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) 364 punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) 365 366 cmp ecx, byte SIZEOF_XMMWORD 367 jb short .column_st32 368 369 test edi, SIZEOF_XMMWORD-1 370 jnz short .out1 371 ; --(aligned)------------------- 372 movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA 373 movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD 374 movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC 375 movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH 376 jmp short .out0 377 .out1: ; --(unaligned)----------------- 378 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA 379 movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD 380 movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC 381 movdqu XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH 382 .out0: 383 add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr 384 sub ecx, byte SIZEOF_XMMWORD 385 jz near .endcolumn 386 387 add esi, byte SIZEOF_XMMWORD ; inptr0 388 dec al ; Yctr 389 jnz near .Yloop_2nd 390 391 add ebx, byte SIZEOF_XMMWORD ; inptr1 392 add edx, byte SIZEOF_XMMWORD ; inptr2 393 jmp near .columnloop 394 ALIGNX 16, 7 395 396 .column_st32: 397 cmp ecx, byte SIZEOF_XMMWORD/2 398 jb short .column_st16 399 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA 400 movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD 401 add edi, byte 2*SIZEOF_XMMWORD ; outptr 402 movdqa xmmA, xmmC 403 movdqa xmmD, xmmH 404 sub ecx, byte SIZEOF_XMMWORD/2 405 .column_st16: 406 cmp ecx, byte SIZEOF_XMMWORD/4 407 jb short .column_st15 408 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA 409 add edi, byte SIZEOF_XMMWORD ; outptr 410 movdqa xmmA, xmmD 411 sub ecx, byte SIZEOF_XMMWORD/4 412 .column_st15: 413 ; Store two pixels (8 bytes) of xmmA to the output when it has enough 414 ; space. 415 cmp ecx, byte SIZEOF_XMMWORD/8 416 jb short .column_st7 417 movq XMM_MMWORD [edi], xmmA 418 add edi, byte SIZEOF_XMMWORD/8*4 419 sub ecx, byte SIZEOF_XMMWORD/8 420 psrldq xmmA, SIZEOF_XMMWORD/8*4 421 .column_st7: 422 ; Store one pixel (4 bytes) of xmmA to the output when it has enough 423 ; space. 424 test ecx, ecx 425 jz short .endcolumn 426 movd XMM_DWORD [edi], xmmA 427 428 %endif ; RGB_PIXELSIZE ; --------------- 429 430 .endcolumn: 431 sfence ; flush the write buffer 432 433 .return: 434 pop edi 435 pop esi 436 ; pop edx ; need not be preserved 437 ; pop ecx ; need not be preserved 438 pop ebx 439 mov esp, ebp ; esp <- aligned ebp 440 pop esp ; esp <- original ebp 441 pop ebp 442 ret 443 444 ; -------------------------------------------------------------------------- 445 ; 446 ; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical. 447 ; 448 ; GLOBAL(void) 449 ; jsimd_h2v2_merged_upsample_sse2(JDIMENSION output_width, 450 ; JSAMPIMAGE input_buf, 451 ; JDIMENSION in_row_group_ctr, 452 ; JSAMPARRAY output_buf); 453 ; 454 455 %define output_width(b) (b) + 8 ; JDIMENSION output_width 456 %define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf 457 %define in_row_group_ctr(b) (b) + 16 ; JDIMENSION in_row_group_ctr 458 %define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf 459 460 align 32 461 GLOBAL_FUNCTION(jsimd_h2v2_merged_upsample_sse2) 462 463 EXTN(jsimd_h2v2_merged_upsample_sse2): 464 push ebp 465 mov ebp, esp 466 push ebx 467 ; push ecx ; need not be preserved 468 ; push edx ; need not be preserved 469 push esi 470 push edi 471 472 mov eax, POINTER [output_width(ebp)] 473 474 mov edi, JSAMPIMAGE [input_buf(ebp)] 475 mov ecx, JDIMENSION [in_row_group_ctr(ebp)] 476 mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] 477 mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] 478 mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] 479 mov edi, JSAMPARRAY [output_buf(ebp)] 480 lea esi, [esi+ecx*SIZEOF_JSAMPROW] 481 482 push edx ; inptr2 483 push ebx ; inptr1 484 push esi ; inptr00 485 mov ebx, esp 486 487 push edi ; output_buf (outptr0) 488 push ecx ; in_row_group_ctr 489 push ebx ; input_buf 490 push eax ; output_width 491 492 call near EXTN(jsimd_h2v1_merged_upsample_sse2) 493 494 add esi, byte SIZEOF_JSAMPROW ; inptr01 495 add edi, byte SIZEOF_JSAMPROW ; outptr1 496 mov POINTER [ebx+0*SIZEOF_POINTER], esi 497 mov POINTER [ebx-1*SIZEOF_POINTER], edi 498 499 call near EXTN(jsimd_h2v1_merged_upsample_sse2) 500 501 add esp, byte 7*SIZEOF_DWORD 502 503 pop edi 504 pop esi 505 ; pop edx ; need not be preserved 506 ; pop ecx ; need not be preserved 507 pop ebx 508 pop ebp 509 ret 510 511 ; For some reason, the OS X linker does not honor the request to align the 512 ; segment unless we do this. 513 align 32