Thu, 22 Jan 2015 13:21:57 +0100
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
michael@0 | 1 | ; |
michael@0 | 2 | ; jdclrmmx.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_ycc_rgb_convert_mmx (JDIMENSION out_width, |
michael@0 | 27 | ; JSAMPIMAGE input_buf, JDIMENSION input_row, |
michael@0 | 28 | ; JSAMPARRAY output_buf, int num_rows) |
michael@0 | 29 | ; |
michael@0 | 30 | |
michael@0 | 31 | %define out_width(b) (b)+8 ; JDIMENSION out_width |
michael@0 | 32 | %define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf |
michael@0 | 33 | %define input_row(b) (b)+16 ; JDIMENSION input_row |
michael@0 | 34 | %define output_buf(b) (b)+20 ; JSAMPARRAY output_buf |
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 2 |
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_ycc_rgb_convert_mmx) |
michael@0 | 44 | |
michael@0 | 45 | EXTN(jsimd_ycc_rgb_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 [out_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 edi, JSAMPIMAGE [input_buf(eax)] |
michael@0 | 70 | mov ecx, JDIMENSION [input_row(eax)] |
michael@0 | 71 | mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] |
michael@0 | 72 | mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] |
michael@0 | 73 | mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] |
michael@0 | 74 | lea esi, [esi+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 edi, JSAMPARRAY [output_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 | push eax |
michael@0 | 87 | push edi |
michael@0 | 88 | push edx |
michael@0 | 89 | push ebx |
michael@0 | 90 | push esi |
michael@0 | 91 | push ecx ; col |
michael@0 | 92 | |
michael@0 | 93 | mov esi, JSAMPROW [esi] ; inptr0 |
michael@0 | 94 | mov ebx, JSAMPROW [ebx] ; inptr1 |
michael@0 | 95 | mov edx, JSAMPROW [edx] ; inptr2 |
michael@0 | 96 | mov edi, JSAMPROW [edi] ; outptr |
michael@0 | 97 | movpic eax, POINTER [gotptr] ; load GOT address (eax) |
michael@0 | 98 | alignx 16,7 |
michael@0 | 99 | .columnloop: |
michael@0 | 100 | |
michael@0 | 101 | movq mm5, MMWORD [ebx] ; mm5=Cb(01234567) |
michael@0 | 102 | movq mm1, MMWORD [edx] ; mm1=Cr(01234567) |
michael@0 | 103 | |
michael@0 | 104 | pcmpeqw mm4,mm4 |
michael@0 | 105 | pcmpeqw mm7,mm7 |
michael@0 | 106 | psrlw mm4,BYTE_BIT |
michael@0 | 107 | psllw mm7,7 ; mm7={0xFF80 0xFF80 0xFF80 0xFF80} |
michael@0 | 108 | movq mm0,mm4 ; mm0=mm4={0xFF 0x00 0xFF 0x00 ..} |
michael@0 | 109 | |
michael@0 | 110 | pand mm4,mm5 ; mm4=Cb(0246)=CbE |
michael@0 | 111 | psrlw mm5,BYTE_BIT ; mm5=Cb(1357)=CbO |
michael@0 | 112 | pand mm0,mm1 ; mm0=Cr(0246)=CrE |
michael@0 | 113 | psrlw mm1,BYTE_BIT ; mm1=Cr(1357)=CrO |
michael@0 | 114 | |
michael@0 | 115 | paddw mm4,mm7 |
michael@0 | 116 | paddw mm5,mm7 |
michael@0 | 117 | paddw mm0,mm7 |
michael@0 | 118 | paddw mm1,mm7 |
michael@0 | 119 | |
michael@0 | 120 | ; (Original) |
michael@0 | 121 | ; R = Y + 1.40200 * Cr |
michael@0 | 122 | ; G = Y - 0.34414 * Cb - 0.71414 * Cr |
michael@0 | 123 | ; B = Y + 1.77200 * Cb |
michael@0 | 124 | ; |
michael@0 | 125 | ; (This implementation) |
michael@0 | 126 | ; R = Y + 0.40200 * Cr + Cr |
michael@0 | 127 | ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr |
michael@0 | 128 | ; B = Y - 0.22800 * Cb + Cb + Cb |
michael@0 | 129 | |
michael@0 | 130 | movq mm2,mm4 ; mm2=CbE |
michael@0 | 131 | movq mm3,mm5 ; mm3=CbO |
michael@0 | 132 | paddw mm4,mm4 ; mm4=2*CbE |
michael@0 | 133 | paddw mm5,mm5 ; mm5=2*CbO |
michael@0 | 134 | movq mm6,mm0 ; mm6=CrE |
michael@0 | 135 | movq mm7,mm1 ; mm7=CrO |
michael@0 | 136 | paddw mm0,mm0 ; mm0=2*CrE |
michael@0 | 137 | paddw mm1,mm1 ; mm1=2*CrO |
michael@0 | 138 | |
michael@0 | 139 | pmulhw mm4,[GOTOFF(eax,PW_MF0228)] ; mm4=(2*CbE * -FIX(0.22800)) |
michael@0 | 140 | pmulhw mm5,[GOTOFF(eax,PW_MF0228)] ; mm5=(2*CbO * -FIX(0.22800)) |
michael@0 | 141 | pmulhw mm0,[GOTOFF(eax,PW_F0402)] ; mm0=(2*CrE * FIX(0.40200)) |
michael@0 | 142 | pmulhw mm1,[GOTOFF(eax,PW_F0402)] ; mm1=(2*CrO * FIX(0.40200)) |
michael@0 | 143 | |
michael@0 | 144 | paddw mm4,[GOTOFF(eax,PW_ONE)] |
michael@0 | 145 | paddw mm5,[GOTOFF(eax,PW_ONE)] |
michael@0 | 146 | psraw mm4,1 ; mm4=(CbE * -FIX(0.22800)) |
michael@0 | 147 | psraw mm5,1 ; mm5=(CbO * -FIX(0.22800)) |
michael@0 | 148 | paddw mm0,[GOTOFF(eax,PW_ONE)] |
michael@0 | 149 | paddw mm1,[GOTOFF(eax,PW_ONE)] |
michael@0 | 150 | psraw mm0,1 ; mm0=(CrE * FIX(0.40200)) |
michael@0 | 151 | psraw mm1,1 ; mm1=(CrO * FIX(0.40200)) |
michael@0 | 152 | |
michael@0 | 153 | paddw mm4,mm2 |
michael@0 | 154 | paddw mm5,mm3 |
michael@0 | 155 | paddw mm4,mm2 ; mm4=(CbE * FIX(1.77200))=(B-Y)E |
michael@0 | 156 | paddw mm5,mm3 ; mm5=(CbO * FIX(1.77200))=(B-Y)O |
michael@0 | 157 | paddw mm0,mm6 ; mm0=(CrE * FIX(1.40200))=(R-Y)E |
michael@0 | 158 | paddw mm1,mm7 ; mm1=(CrO * FIX(1.40200))=(R-Y)O |
michael@0 | 159 | |
michael@0 | 160 | movq MMWORD [wk(0)], mm4 ; wk(0)=(B-Y)E |
michael@0 | 161 | movq MMWORD [wk(1)], mm5 ; wk(1)=(B-Y)O |
michael@0 | 162 | |
michael@0 | 163 | movq mm4,mm2 |
michael@0 | 164 | movq mm5,mm3 |
michael@0 | 165 | punpcklwd mm2,mm6 |
michael@0 | 166 | punpckhwd mm4,mm6 |
michael@0 | 167 | pmaddwd mm2,[GOTOFF(eax,PW_MF0344_F0285)] |
michael@0 | 168 | pmaddwd mm4,[GOTOFF(eax,PW_MF0344_F0285)] |
michael@0 | 169 | punpcklwd mm3,mm7 |
michael@0 | 170 | punpckhwd mm5,mm7 |
michael@0 | 171 | pmaddwd mm3,[GOTOFF(eax,PW_MF0344_F0285)] |
michael@0 | 172 | pmaddwd mm5,[GOTOFF(eax,PW_MF0344_F0285)] |
michael@0 | 173 | |
michael@0 | 174 | paddd mm2,[GOTOFF(eax,PD_ONEHALF)] |
michael@0 | 175 | paddd mm4,[GOTOFF(eax,PD_ONEHALF)] |
michael@0 | 176 | psrad mm2,SCALEBITS |
michael@0 | 177 | psrad mm4,SCALEBITS |
michael@0 | 178 | paddd mm3,[GOTOFF(eax,PD_ONEHALF)] |
michael@0 | 179 | paddd mm5,[GOTOFF(eax,PD_ONEHALF)] |
michael@0 | 180 | psrad mm3,SCALEBITS |
michael@0 | 181 | psrad mm5,SCALEBITS |
michael@0 | 182 | |
michael@0 | 183 | packssdw mm2,mm4 ; mm2=CbE*-FIX(0.344)+CrE*FIX(0.285) |
michael@0 | 184 | packssdw mm3,mm5 ; mm3=CbO*-FIX(0.344)+CrO*FIX(0.285) |
michael@0 | 185 | psubw mm2,mm6 ; mm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E |
michael@0 | 186 | psubw mm3,mm7 ; mm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O |
michael@0 | 187 | |
michael@0 | 188 | movq mm5, MMWORD [esi] ; mm5=Y(01234567) |
michael@0 | 189 | |
michael@0 | 190 | pcmpeqw mm4,mm4 |
michael@0 | 191 | psrlw mm4,BYTE_BIT ; mm4={0xFF 0x00 0xFF 0x00 ..} |
michael@0 | 192 | pand mm4,mm5 ; mm4=Y(0246)=YE |
michael@0 | 193 | psrlw mm5,BYTE_BIT ; mm5=Y(1357)=YO |
michael@0 | 194 | |
michael@0 | 195 | paddw mm0,mm4 ; mm0=((R-Y)E+YE)=RE=(R0 R2 R4 R6) |
michael@0 | 196 | paddw mm1,mm5 ; mm1=((R-Y)O+YO)=RO=(R1 R3 R5 R7) |
michael@0 | 197 | packuswb mm0,mm0 ; mm0=(R0 R2 R4 R6 ** ** ** **) |
michael@0 | 198 | packuswb mm1,mm1 ; mm1=(R1 R3 R5 R7 ** ** ** **) |
michael@0 | 199 | |
michael@0 | 200 | paddw mm2,mm4 ; mm2=((G-Y)E+YE)=GE=(G0 G2 G4 G6) |
michael@0 | 201 | paddw mm3,mm5 ; mm3=((G-Y)O+YO)=GO=(G1 G3 G5 G7) |
michael@0 | 202 | packuswb mm2,mm2 ; mm2=(G0 G2 G4 G6 ** ** ** **) |
michael@0 | 203 | packuswb mm3,mm3 ; mm3=(G1 G3 G5 G7 ** ** ** **) |
michael@0 | 204 | |
michael@0 | 205 | paddw mm4, MMWORD [wk(0)] ; mm4=(YE+(B-Y)E)=BE=(B0 B2 B4 B6) |
michael@0 | 206 | paddw mm5, MMWORD [wk(1)] ; mm5=(YO+(B-Y)O)=BO=(B1 B3 B5 B7) |
michael@0 | 207 | packuswb mm4,mm4 ; mm4=(B0 B2 B4 B6 ** ** ** **) |
michael@0 | 208 | packuswb mm5,mm5 ; mm5=(B1 B3 B5 B7 ** ** ** **) |
michael@0 | 209 | |
michael@0 | 210 | %if RGB_PIXELSIZE == 3 ; --------------- |
michael@0 | 211 | |
michael@0 | 212 | ; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **) |
michael@0 | 213 | ; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **) |
michael@0 | 214 | ; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **) |
michael@0 | 215 | ; mmG=(** ** ** ** ** ** ** **), mmH=(** ** ** ** ** ** ** **) |
michael@0 | 216 | |
michael@0 | 217 | punpcklbw mmA,mmC ; mmA=(00 10 02 12 04 14 06 16) |
michael@0 | 218 | punpcklbw mmE,mmB ; mmE=(20 01 22 03 24 05 26 07) |
michael@0 | 219 | punpcklbw mmD,mmF ; mmD=(11 21 13 23 15 25 17 27) |
michael@0 | 220 | |
michael@0 | 221 | movq mmG,mmA |
michael@0 | 222 | movq mmH,mmA |
michael@0 | 223 | punpcklwd mmA,mmE ; mmA=(00 10 20 01 02 12 22 03) |
michael@0 | 224 | punpckhwd mmG,mmE ; mmG=(04 14 24 05 06 16 26 07) |
michael@0 | 225 | |
michael@0 | 226 | psrlq mmH,2*BYTE_BIT ; mmH=(02 12 04 14 06 16 -- --) |
michael@0 | 227 | psrlq mmE,2*BYTE_BIT ; mmE=(22 03 24 05 26 07 -- --) |
michael@0 | 228 | |
michael@0 | 229 | movq mmC,mmD |
michael@0 | 230 | movq mmB,mmD |
michael@0 | 231 | punpcklwd mmD,mmH ; mmD=(11 21 02 12 13 23 04 14) |
michael@0 | 232 | punpckhwd mmC,mmH ; mmC=(15 25 06 16 17 27 -- --) |
michael@0 | 233 | |
michael@0 | 234 | psrlq mmB,2*BYTE_BIT ; mmB=(13 23 15 25 17 27 -- --) |
michael@0 | 235 | |
michael@0 | 236 | movq mmF,mmE |
michael@0 | 237 | punpcklwd mmE,mmB ; mmE=(22 03 13 23 24 05 15 25) |
michael@0 | 238 | punpckhwd mmF,mmB ; mmF=(26 07 17 27 -- -- -- --) |
michael@0 | 239 | |
michael@0 | 240 | punpckldq mmA,mmD ; mmA=(00 10 20 01 11 21 02 12) |
michael@0 | 241 | punpckldq mmE,mmG ; mmE=(22 03 13 23 04 14 24 05) |
michael@0 | 242 | punpckldq mmC,mmF ; mmC=(15 25 06 16 26 07 17 27) |
michael@0 | 243 | |
michael@0 | 244 | cmp ecx, byte SIZEOF_MMWORD |
michael@0 | 245 | jb short .column_st16 |
michael@0 | 246 | |
michael@0 | 247 | movq MMWORD [edi+0*SIZEOF_MMWORD], mmA |
michael@0 | 248 | movq MMWORD [edi+1*SIZEOF_MMWORD], mmE |
michael@0 | 249 | movq MMWORD [edi+2*SIZEOF_MMWORD], mmC |
michael@0 | 250 | |
michael@0 | 251 | sub ecx, byte SIZEOF_MMWORD |
michael@0 | 252 | jz short .nextrow |
michael@0 | 253 | |
michael@0 | 254 | add esi, byte SIZEOF_MMWORD ; inptr0 |
michael@0 | 255 | add ebx, byte SIZEOF_MMWORD ; inptr1 |
michael@0 | 256 | add edx, byte SIZEOF_MMWORD ; inptr2 |
michael@0 | 257 | add edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; outptr |
michael@0 | 258 | jmp near .columnloop |
michael@0 | 259 | alignx 16,7 |
michael@0 | 260 | |
michael@0 | 261 | .column_st16: |
michael@0 | 262 | lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE |
michael@0 | 263 | cmp ecx, byte 2*SIZEOF_MMWORD |
michael@0 | 264 | jb short .column_st8 |
michael@0 | 265 | movq MMWORD [edi+0*SIZEOF_MMWORD], mmA |
michael@0 | 266 | movq MMWORD [edi+1*SIZEOF_MMWORD], mmE |
michael@0 | 267 | movq mmA,mmC |
michael@0 | 268 | sub ecx, byte 2*SIZEOF_MMWORD |
michael@0 | 269 | add edi, byte 2*SIZEOF_MMWORD |
michael@0 | 270 | jmp short .column_st4 |
michael@0 | 271 | .column_st8: |
michael@0 | 272 | cmp ecx, byte SIZEOF_MMWORD |
michael@0 | 273 | jb short .column_st4 |
michael@0 | 274 | movq MMWORD [edi+0*SIZEOF_MMWORD], mmA |
michael@0 | 275 | movq mmA,mmE |
michael@0 | 276 | sub ecx, byte SIZEOF_MMWORD |
michael@0 | 277 | add edi, byte SIZEOF_MMWORD |
michael@0 | 278 | .column_st4: |
michael@0 | 279 | movd eax,mmA |
michael@0 | 280 | cmp ecx, byte SIZEOF_DWORD |
michael@0 | 281 | jb short .column_st2 |
michael@0 | 282 | mov DWORD [edi+0*SIZEOF_DWORD], eax |
michael@0 | 283 | psrlq mmA,DWORD_BIT |
michael@0 | 284 | movd eax,mmA |
michael@0 | 285 | sub ecx, byte SIZEOF_DWORD |
michael@0 | 286 | add edi, byte SIZEOF_DWORD |
michael@0 | 287 | .column_st2: |
michael@0 | 288 | cmp ecx, byte SIZEOF_WORD |
michael@0 | 289 | jb short .column_st1 |
michael@0 | 290 | mov WORD [edi+0*SIZEOF_WORD], ax |
michael@0 | 291 | shr eax,WORD_BIT |
michael@0 | 292 | sub ecx, byte SIZEOF_WORD |
michael@0 | 293 | add edi, byte SIZEOF_WORD |
michael@0 | 294 | .column_st1: |
michael@0 | 295 | cmp ecx, byte SIZEOF_BYTE |
michael@0 | 296 | jb short .nextrow |
michael@0 | 297 | mov BYTE [edi+0*SIZEOF_BYTE], al |
michael@0 | 298 | |
michael@0 | 299 | %else ; RGB_PIXELSIZE == 4 ; ----------- |
michael@0 | 300 | |
michael@0 | 301 | %ifdef RGBX_FILLER_0XFF |
michael@0 | 302 | pcmpeqb mm6,mm6 ; mm6=(X0 X2 X4 X6 ** ** ** **) |
michael@0 | 303 | pcmpeqb mm7,mm7 ; mm7=(X1 X3 X5 X7 ** ** ** **) |
michael@0 | 304 | %else |
michael@0 | 305 | pxor mm6,mm6 ; mm6=(X0 X2 X4 X6 ** ** ** **) |
michael@0 | 306 | pxor mm7,mm7 ; mm7=(X1 X3 X5 X7 ** ** ** **) |
michael@0 | 307 | %endif |
michael@0 | 308 | ; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **) |
michael@0 | 309 | ; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **) |
michael@0 | 310 | ; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **) |
michael@0 | 311 | ; mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **) |
michael@0 | 312 | |
michael@0 | 313 | punpcklbw mmA,mmC ; mmA=(00 10 02 12 04 14 06 16) |
michael@0 | 314 | punpcklbw mmE,mmG ; mmE=(20 30 22 32 24 34 26 36) |
michael@0 | 315 | punpcklbw mmB,mmD ; mmB=(01 11 03 13 05 15 07 17) |
michael@0 | 316 | punpcklbw mmF,mmH ; mmF=(21 31 23 33 25 35 27 37) |
michael@0 | 317 | |
michael@0 | 318 | movq mmC,mmA |
michael@0 | 319 | punpcklwd mmA,mmE ; mmA=(00 10 20 30 02 12 22 32) |
michael@0 | 320 | punpckhwd mmC,mmE ; mmC=(04 14 24 34 06 16 26 36) |
michael@0 | 321 | movq mmG,mmB |
michael@0 | 322 | punpcklwd mmB,mmF ; mmB=(01 11 21 31 03 13 23 33) |
michael@0 | 323 | punpckhwd mmG,mmF ; mmG=(05 15 25 35 07 17 27 37) |
michael@0 | 324 | |
michael@0 | 325 | movq mmD,mmA |
michael@0 | 326 | punpckldq mmA,mmB ; mmA=(00 10 20 30 01 11 21 31) |
michael@0 | 327 | punpckhdq mmD,mmB ; mmD=(02 12 22 32 03 13 23 33) |
michael@0 | 328 | movq mmH,mmC |
michael@0 | 329 | punpckldq mmC,mmG ; mmC=(04 14 24 34 05 15 25 35) |
michael@0 | 330 | punpckhdq mmH,mmG ; mmH=(06 16 26 36 07 17 27 37) |
michael@0 | 331 | |
michael@0 | 332 | cmp ecx, byte SIZEOF_MMWORD |
michael@0 | 333 | jb short .column_st16 |
michael@0 | 334 | |
michael@0 | 335 | movq MMWORD [edi+0*SIZEOF_MMWORD], mmA |
michael@0 | 336 | movq MMWORD [edi+1*SIZEOF_MMWORD], mmD |
michael@0 | 337 | movq MMWORD [edi+2*SIZEOF_MMWORD], mmC |
michael@0 | 338 | movq MMWORD [edi+3*SIZEOF_MMWORD], mmH |
michael@0 | 339 | |
michael@0 | 340 | sub ecx, byte SIZEOF_MMWORD |
michael@0 | 341 | jz short .nextrow |
michael@0 | 342 | |
michael@0 | 343 | add esi, byte SIZEOF_MMWORD ; inptr0 |
michael@0 | 344 | add ebx, byte SIZEOF_MMWORD ; inptr1 |
michael@0 | 345 | add edx, byte SIZEOF_MMWORD ; inptr2 |
michael@0 | 346 | add edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; outptr |
michael@0 | 347 | jmp near .columnloop |
michael@0 | 348 | alignx 16,7 |
michael@0 | 349 | |
michael@0 | 350 | .column_st16: |
michael@0 | 351 | cmp ecx, byte SIZEOF_MMWORD/2 |
michael@0 | 352 | jb short .column_st8 |
michael@0 | 353 | movq MMWORD [edi+0*SIZEOF_MMWORD], mmA |
michael@0 | 354 | movq MMWORD [edi+1*SIZEOF_MMWORD], mmD |
michael@0 | 355 | movq mmA,mmC |
michael@0 | 356 | movq mmD,mmH |
michael@0 | 357 | sub ecx, byte SIZEOF_MMWORD/2 |
michael@0 | 358 | add edi, byte 2*SIZEOF_MMWORD |
michael@0 | 359 | .column_st8: |
michael@0 | 360 | cmp ecx, byte SIZEOF_MMWORD/4 |
michael@0 | 361 | jb short .column_st4 |
michael@0 | 362 | movq MMWORD [edi+0*SIZEOF_MMWORD], mmA |
michael@0 | 363 | movq mmA,mmD |
michael@0 | 364 | sub ecx, byte SIZEOF_MMWORD/4 |
michael@0 | 365 | add edi, byte 1*SIZEOF_MMWORD |
michael@0 | 366 | .column_st4: |
michael@0 | 367 | cmp ecx, byte SIZEOF_MMWORD/8 |
michael@0 | 368 | jb short .nextrow |
michael@0 | 369 | movd DWORD [edi+0*SIZEOF_DWORD], mmA |
michael@0 | 370 | |
michael@0 | 371 | %endif ; RGB_PIXELSIZE ; --------------- |
michael@0 | 372 | |
michael@0 | 373 | alignx 16,7 |
michael@0 | 374 | |
michael@0 | 375 | .nextrow: |
michael@0 | 376 | pop ecx |
michael@0 | 377 | pop esi |
michael@0 | 378 | pop ebx |
michael@0 | 379 | pop edx |
michael@0 | 380 | pop edi |
michael@0 | 381 | pop eax |
michael@0 | 382 | |
michael@0 | 383 | add esi, byte SIZEOF_JSAMPROW |
michael@0 | 384 | add ebx, byte SIZEOF_JSAMPROW |
michael@0 | 385 | add edx, byte SIZEOF_JSAMPROW |
michael@0 | 386 | add edi, byte SIZEOF_JSAMPROW ; output_buf |
michael@0 | 387 | dec eax ; num_rows |
michael@0 | 388 | jg near .rowloop |
michael@0 | 389 | |
michael@0 | 390 | emms ; empty MMX state |
michael@0 | 391 | |
michael@0 | 392 | .return: |
michael@0 | 393 | pop edi |
michael@0 | 394 | pop esi |
michael@0 | 395 | ; pop edx ; need not be preserved |
michael@0 | 396 | ; pop ecx ; need not be preserved |
michael@0 | 397 | pop ebx |
michael@0 | 398 | mov esp,ebp ; esp <- aligned ebp |
michael@0 | 399 | pop esp ; esp <- original ebp |
michael@0 | 400 | pop ebp |
michael@0 | 401 | ret |
michael@0 | 402 | |
michael@0 | 403 | ; For some reason, the OS X linker does not honor the request to align the |
michael@0 | 404 | ; segment unless we do this. |
michael@0 | 405 | align 16 |