The Tor Browser / annotate

media/libjpeg/simd/jdclrss2.asm@6474c204b198 (annotated)

media/libjpeg/simd/jdclrss2.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 ; jdclrss2.asm - colorspace conversion (SSE2)
michael@0 3 ;
michael@0 4 ; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
michael@0 5 ; Copyright 2012 D. R. Commander
michael@0 6 ;
michael@0 7 ; Based on
michael@0 8 ; x86 SIMD extension for IJG JPEG library
michael@0 9 ; Copyright (C) 1999-2006, MIYASAKA Masaru.
michael@0 10 ; For conditions of distribution and use, see copyright notice in jsimdext.inc
michael@0 11 ;
michael@0 12 ; This file should be assembled with NASM (Netwide Assembler),
michael@0 13 ; can *not* be assembled with Microsoft's MASM or any compatible
michael@0 14 ; assembler (including Borland's Turbo Assembler).
michael@0 15 ; NASM is available from http://nasm.sourceforge.net/ or
michael@0 16 ; http://sourceforge.net/project/showfiles.php?group_id=6208
michael@0 17 ;
michael@0 18 ; [TAB8]
michael@0 19
michael@0 20 %include "jcolsamp.inc"
michael@0 21
michael@0 22 ; --------------------------------------------------------------------------
michael@0 23 ;
michael@0 24 ; Convert some rows of samples to the output colorspace.
michael@0 25 ;
michael@0 26 ; GLOBAL(void)
michael@0 27 ; jsimd_ycc_rgb_convert_sse2 (JDIMENSION out_width,
michael@0 28 ; JSAMPIMAGE input_buf, JDIMENSION input_row,
michael@0 29 ; JSAMPARRAY output_buf, int num_rows)
michael@0 30 ;
michael@0 31
michael@0 32 %define out_width(b) (b)+8 ; JDIMENSION out_width
michael@0 33 %define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
michael@0 34 %define input_row(b) (b)+16 ; JDIMENSION input_row
michael@0 35 %define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
michael@0 36 %define num_rows(b) (b)+24 ; int num_rows
michael@0 37
michael@0 38 %define original_ebp ebp+0
michael@0 39 %define wk(i) ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
michael@0 40 %define WK_NUM 2
michael@0 41 %define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
michael@0 42
michael@0 43 align 16
michael@0 44 global EXTN(jsimd_ycc_rgb_convert_sse2)
michael@0 45
michael@0 46 EXTN(jsimd_ycc_rgb_convert_sse2):
michael@0 47 push ebp
michael@0 48 mov eax,esp ; eax = original ebp
michael@0 49 sub esp, byte 4
michael@0 50 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
michael@0 51 mov [esp],eax
michael@0 52 mov ebp,esp ; ebp = aligned ebp
michael@0 53 lea esp, [wk(0)]
michael@0 54 pushpic eax ; make a room for GOT address
michael@0 55 push ebx
michael@0 56 ; push ecx ; need not be preserved
michael@0 57 ; push edx ; need not be preserved
michael@0 58 push esi
michael@0 59 push edi
michael@0 60
michael@0 61 get_GOT ebx ; get GOT address
michael@0 62 movpic POINTER [gotptr], ebx ; save GOT address
michael@0 63
michael@0 64 mov ecx, JDIMENSION [out_width(eax)] ; num_cols
michael@0 65 test ecx,ecx
michael@0 66 jz near .return
michael@0 67
michael@0 68 push ecx
michael@0 69
michael@0 70 mov edi, JSAMPIMAGE [input_buf(eax)]
michael@0 71 mov ecx, JDIMENSION [input_row(eax)]
michael@0 72 mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
michael@0 73 mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
michael@0 74 mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
michael@0 75 lea esi, [esi+ecx*SIZEOF_JSAMPROW]
michael@0 76 lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
michael@0 77 lea edx, [edx+ecx*SIZEOF_JSAMPROW]
michael@0 78
michael@0 79 pop ecx
michael@0 80
michael@0 81 mov edi, JSAMPARRAY [output_buf(eax)]
michael@0 82 mov eax, INT [num_rows(eax)]
michael@0 83 test eax,eax
michael@0 84 jle near .return
michael@0 85 alignx 16,7
michael@0 86 .rowloop:
michael@0 87 push eax
michael@0 88 push edi
michael@0 89 push edx
michael@0 90 push ebx
michael@0 91 push esi
michael@0 92 push ecx ; col
michael@0 93
michael@0 94 mov esi, JSAMPROW [esi] ; inptr0
michael@0 95 mov ebx, JSAMPROW [ebx] ; inptr1
michael@0 96 mov edx, JSAMPROW [edx] ; inptr2
michael@0 97 mov edi, JSAMPROW [edi] ; outptr
michael@0 98 movpic eax, POINTER [gotptr] ; load GOT address (eax)
michael@0 99 alignx 16,7
michael@0 100 .columnloop:
michael@0 101
michael@0 102 movdqa xmm5, XMMWORD [ebx] ; xmm5=Cb(0123456789ABCDEF)
michael@0 103 movdqa xmm1, XMMWORD [edx] ; xmm1=Cr(0123456789ABCDEF)
michael@0 104
michael@0 105 pcmpeqw xmm4,xmm4
michael@0 106 pcmpeqw xmm7,xmm7
michael@0 107 psrlw xmm4,BYTE_BIT
michael@0 108 psllw xmm7,7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
michael@0 109 movdqa xmm0,xmm4 ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..}
michael@0 110
michael@0 111 pand xmm4,xmm5 ; xmm4=Cb(02468ACE)=CbE
michael@0 112 psrlw xmm5,BYTE_BIT ; xmm5=Cb(13579BDF)=CbO
michael@0 113 pand xmm0,xmm1 ; xmm0=Cr(02468ACE)=CrE
michael@0 114 psrlw xmm1,BYTE_BIT ; xmm1=Cr(13579BDF)=CrO
michael@0 115
michael@0 116 paddw xmm4,xmm7
michael@0 117 paddw xmm5,xmm7
michael@0 118 paddw xmm0,xmm7
michael@0 119 paddw xmm1,xmm7
michael@0 120
michael@0 121 ; (Original)
michael@0 122 ; R = Y + 1.40200 * Cr
michael@0 123 ; G = Y - 0.34414 * Cb - 0.71414 * Cr
michael@0 124 ; B = Y + 1.77200 * Cb
michael@0 125 ;
michael@0 126 ; (This implementation)
michael@0 127 ; R = Y + 0.40200 * Cr + Cr
michael@0 128 ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
michael@0 129 ; B = Y - 0.22800 * Cb + Cb + Cb
michael@0 130
michael@0 131 movdqa xmm2,xmm4 ; xmm2=CbE
michael@0 132 movdqa xmm3,xmm5 ; xmm3=CbO
michael@0 133 paddw xmm4,xmm4 ; xmm4=2*CbE
michael@0 134 paddw xmm5,xmm5 ; xmm5=2*CbO
michael@0 135 movdqa xmm6,xmm0 ; xmm6=CrE
michael@0 136 movdqa xmm7,xmm1 ; xmm7=CrO
michael@0 137 paddw xmm0,xmm0 ; xmm0=2*CrE
michael@0 138 paddw xmm1,xmm1 ; xmm1=2*CrO
michael@0 139
michael@0 140 pmulhw xmm4,[GOTOFF(eax,PW_MF0228)] ; xmm4=(2*CbE * -FIX(0.22800))
michael@0 141 pmulhw xmm5,[GOTOFF(eax,PW_MF0228)] ; xmm5=(2*CbO * -FIX(0.22800))
michael@0 142 pmulhw xmm0,[GOTOFF(eax,PW_F0402)] ; xmm0=(2*CrE * FIX(0.40200))
michael@0 143 pmulhw xmm1,[GOTOFF(eax,PW_F0402)] ; xmm1=(2*CrO * FIX(0.40200))
michael@0 144
michael@0 145 paddw xmm4,[GOTOFF(eax,PW_ONE)]
michael@0 146 paddw xmm5,[GOTOFF(eax,PW_ONE)]
michael@0 147 psraw xmm4,1 ; xmm4=(CbE * -FIX(0.22800))
michael@0 148 psraw xmm5,1 ; xmm5=(CbO * -FIX(0.22800))
michael@0 149 paddw xmm0,[GOTOFF(eax,PW_ONE)]
michael@0 150 paddw xmm1,[GOTOFF(eax,PW_ONE)]
michael@0 151 psraw xmm0,1 ; xmm0=(CrE * FIX(0.40200))
michael@0 152 psraw xmm1,1 ; xmm1=(CrO * FIX(0.40200))
michael@0 153
michael@0 154 paddw xmm4,xmm2
michael@0 155 paddw xmm5,xmm3
michael@0 156 paddw xmm4,xmm2 ; xmm4=(CbE * FIX(1.77200))=(B-Y)E
michael@0 157 paddw xmm5,xmm3 ; xmm5=(CbO * FIX(1.77200))=(B-Y)O
michael@0 158 paddw xmm0,xmm6 ; xmm0=(CrE * FIX(1.40200))=(R-Y)E
michael@0 159 paddw xmm1,xmm7 ; xmm1=(CrO * FIX(1.40200))=(R-Y)O
michael@0 160
michael@0 161 movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=(B-Y)E
michael@0 162 movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(B-Y)O
michael@0 163
michael@0 164 movdqa xmm4,xmm2
michael@0 165 movdqa xmm5,xmm3
michael@0 166 punpcklwd xmm2,xmm6
michael@0 167 punpckhwd xmm4,xmm6
michael@0 168 pmaddwd xmm2,[GOTOFF(eax,PW_MF0344_F0285)]
michael@0 169 pmaddwd xmm4,[GOTOFF(eax,PW_MF0344_F0285)]
michael@0 170 punpcklwd xmm3,xmm7
michael@0 171 punpckhwd xmm5,xmm7
michael@0 172 pmaddwd xmm3,[GOTOFF(eax,PW_MF0344_F0285)]
michael@0 173 pmaddwd xmm5,[GOTOFF(eax,PW_MF0344_F0285)]
michael@0 174
michael@0 175 paddd xmm2,[GOTOFF(eax,PD_ONEHALF)]
michael@0 176 paddd xmm4,[GOTOFF(eax,PD_ONEHALF)]
michael@0 177 psrad xmm2,SCALEBITS
michael@0 178 psrad xmm4,SCALEBITS
michael@0 179 paddd xmm3,[GOTOFF(eax,PD_ONEHALF)]
michael@0 180 paddd xmm5,[GOTOFF(eax,PD_ONEHALF)]
michael@0 181 psrad xmm3,SCALEBITS
michael@0 182 psrad xmm5,SCALEBITS
michael@0 183
michael@0 184 packssdw xmm2,xmm4 ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
michael@0 185 packssdw xmm3,xmm5 ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
michael@0 186 psubw xmm2,xmm6 ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
michael@0 187 psubw xmm3,xmm7 ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
michael@0 188
michael@0 189 movdqa xmm5, XMMWORD [esi] ; xmm5=Y(0123456789ABCDEF)
michael@0 190
michael@0 191 pcmpeqw xmm4,xmm4
michael@0 192 psrlw xmm4,BYTE_BIT ; xmm4={0xFF 0x00 0xFF 0x00 ..}
michael@0 193 pand xmm4,xmm5 ; xmm4=Y(02468ACE)=YE
michael@0 194 psrlw xmm5,BYTE_BIT ; xmm5=Y(13579BDF)=YO
michael@0 195
michael@0 196 paddw xmm0,xmm4 ; xmm0=((R-Y)E+YE)=RE=R(02468ACE)
michael@0 197 paddw xmm1,xmm5 ; xmm1=((R-Y)O+YO)=RO=R(13579BDF)
michael@0 198 packuswb xmm0,xmm0 ; xmm0=R(02468ACE********)
michael@0 199 packuswb xmm1,xmm1 ; xmm1=R(13579BDF********)
michael@0 200
michael@0 201 paddw xmm2,xmm4 ; xmm2=((G-Y)E+YE)=GE=G(02468ACE)
michael@0 202 paddw xmm3,xmm5 ; xmm3=((G-Y)O+YO)=GO=G(13579BDF)
michael@0 203 packuswb xmm2,xmm2 ; xmm2=G(02468ACE********)
michael@0 204 packuswb xmm3,xmm3 ; xmm3=G(13579BDF********)
michael@0 205
michael@0 206 paddw xmm4, XMMWORD [wk(0)] ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE)
michael@0 207 paddw xmm5, XMMWORD [wk(1)] ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF)
michael@0 208 packuswb xmm4,xmm4 ; xmm4=B(02468ACE********)
michael@0 209 packuswb xmm5,xmm5 ; xmm5=B(13579BDF********)
michael@0 210
michael@0 211 %if RGB_PIXELSIZE == 3 ; ---------------
michael@0 212
michael@0 213 ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
michael@0 214 ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
michael@0 215 ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
michael@0 216 ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
michael@0 217
michael@0 218 punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
michael@0 219 punpcklbw xmmE,xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
michael@0 220 punpcklbw xmmD,xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
michael@0 221
michael@0 222 movdqa xmmG,xmmA
michael@0 223 movdqa xmmH,xmmA
michael@0 224 punpcklwd xmmA,xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
michael@0 225 punpckhwd xmmG,xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
michael@0 226
michael@0 227 psrldq xmmH,2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
michael@0 228 psrldq xmmE,2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
michael@0 229
michael@0 230 movdqa xmmC,xmmD
michael@0 231 movdqa xmmB,xmmD
michael@0 232 punpcklwd xmmD,xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
michael@0 233 punpckhwd xmmC,xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
michael@0 234
michael@0 235 psrldq xmmB,2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
michael@0 236
michael@0 237 movdqa xmmF,xmmE
michael@0 238 punpcklwd xmmE,xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
michael@0 239 punpckhwd xmmF,xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
michael@0 240
michael@0 241 pshufd xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
michael@0 242 movdqa xmmB,xmmE
michael@0 243 punpckldq xmmA,xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
michael@0 244 punpckldq xmmE,xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
michael@0 245 punpckhdq xmmD,xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
michael@0 246
michael@0 247 pshufd xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
michael@0 248 movdqa xmmB,xmmF
michael@0 249 punpckldq xmmG,xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
michael@0 250 punpckldq xmmF,xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
michael@0 251 punpckhdq xmmC,xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
michael@0 252
michael@0 253 punpcklqdq xmmA,xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
michael@0 254 punpcklqdq xmmD,xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
michael@0 255 punpcklqdq xmmF,xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
michael@0 256
michael@0 257 cmp ecx, byte SIZEOF_XMMWORD
michael@0 258 jb short .column_st32
michael@0 259
michael@0 260 test edi, SIZEOF_XMMWORD-1
michael@0 261 jnz short .out1
michael@0 262 ; --(aligned)-------------------
michael@0 263 movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
michael@0 264 movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
michael@0 265 movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
michael@0 266 jmp short .out0
michael@0 267 .out1: ; --(unaligned)-----------------
michael@0 268 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
michael@0 269 movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
michael@0 270 movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
michael@0 271 .out0:
michael@0 272 add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
michael@0 273 sub ecx, byte SIZEOF_XMMWORD
michael@0 274 jz near .nextrow
michael@0 275
michael@0 276 add esi, byte SIZEOF_XMMWORD ; inptr0
michael@0 277 add ebx, byte SIZEOF_XMMWORD ; inptr1
michael@0 278 add edx, byte SIZEOF_XMMWORD ; inptr2
michael@0 279 jmp near .columnloop
michael@0 280 alignx 16,7
michael@0 281
michael@0 282 .column_st32:
michael@0 283 lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE
michael@0 284 cmp ecx, byte 2*SIZEOF_XMMWORD
michael@0 285 jb short .column_st16
michael@0 286 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
michael@0 287 movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
michael@0 288 add edi, byte 2*SIZEOF_XMMWORD ; outptr
michael@0 289 movdqa xmmA,xmmF
michael@0 290 sub ecx, byte 2*SIZEOF_XMMWORD
michael@0 291 jmp short .column_st15
michael@0 292 .column_st16:
michael@0 293 cmp ecx, byte SIZEOF_XMMWORD
michael@0 294 jb short .column_st15
michael@0 295 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
michael@0 296 add edi, byte SIZEOF_XMMWORD ; outptr
michael@0 297 movdqa xmmA,xmmD
michael@0 298 sub ecx, byte SIZEOF_XMMWORD
michael@0 299 .column_st15:
michael@0 300 ; Store the lower 8 bytes of xmmA to the output when it has enough
michael@0 301 ; space.
michael@0 302 cmp ecx, byte SIZEOF_MMWORD
michael@0 303 jb short .column_st7
michael@0 304 movq XMM_MMWORD [edi], xmmA
michael@0 305 add edi, byte SIZEOF_MMWORD
michael@0 306 sub ecx, byte SIZEOF_MMWORD
michael@0 307 psrldq xmmA, SIZEOF_MMWORD
michael@0 308 .column_st7:
michael@0 309 ; Store the lower 4 bytes of xmmA to the output when it has enough
michael@0 310 ; space.
michael@0 311 cmp ecx, byte SIZEOF_DWORD
michael@0 312 jb short .column_st3
michael@0 313 movd XMM_DWORD [edi], xmmA
michael@0 314 add edi, byte SIZEOF_DWORD
michael@0 315 sub ecx, byte SIZEOF_DWORD
michael@0 316 psrldq xmmA, SIZEOF_DWORD
michael@0 317 .column_st3:
michael@0 318 ; Store the lower 2 bytes of eax to the output when it has enough
michael@0 319 ; space.
michael@0 320 movd eax, xmmA
michael@0 321 cmp ecx, byte SIZEOF_WORD
michael@0 322 jb short .column_st1
michael@0 323 mov WORD [edi], ax
michael@0 324 add edi, byte SIZEOF_WORD
michael@0 325 sub ecx, byte SIZEOF_WORD
michael@0 326 shr eax, 16
michael@0 327 .column_st1:
michael@0 328 ; Store the lower 1 byte of eax to the output when it has enough
michael@0 329 ; space.
michael@0 330 test ecx, ecx
michael@0 331 jz short .nextrow
michael@0 332 mov BYTE [edi], al
michael@0 333
michael@0 334 %else ; RGB_PIXELSIZE == 4 ; -----------
michael@0 335
michael@0 336 %ifdef RGBX_FILLER_0XFF
michael@0 337 pcmpeqb xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
michael@0 338 pcmpeqb xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
michael@0 339 %else
michael@0 340 pxor xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
michael@0 341 pxor xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
michael@0 342 %endif
michael@0 343 ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
michael@0 344 ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
michael@0 345 ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
michael@0 346 ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
michael@0 347
michael@0 348 punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
michael@0 349 punpcklbw xmmE,xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
michael@0 350 punpcklbw xmmB,xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
michael@0 351 punpcklbw xmmF,xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
michael@0 352
michael@0 353 movdqa xmmC,xmmA
michael@0 354 punpcklwd xmmA,xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
michael@0 355 punpckhwd xmmC,xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
michael@0 356 movdqa xmmG,xmmB
michael@0 357 punpcklwd xmmB,xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
michael@0 358 punpckhwd xmmG,xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
michael@0 359
michael@0 360 movdqa xmmD,xmmA
michael@0 361 punpckldq xmmA,xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
michael@0 362 punpckhdq xmmD,xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
michael@0 363 movdqa xmmH,xmmC
michael@0 364 punpckldq xmmC,xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
michael@0 365 punpckhdq xmmH,xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
michael@0 366
michael@0 367 cmp ecx, byte SIZEOF_XMMWORD
michael@0 368 jb short .column_st32
michael@0 369
michael@0 370 test edi, SIZEOF_XMMWORD-1
michael@0 371 jnz short .out1
michael@0 372 ; --(aligned)-------------------
michael@0 373 movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
michael@0 374 movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
michael@0 375 movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
michael@0 376 movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
michael@0 377 jmp short .out0
michael@0 378 .out1: ; --(unaligned)-----------------
michael@0 379 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
michael@0 380 movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
michael@0 381 movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
michael@0 382 movdqu XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
michael@0 383 .out0:
michael@0 384 add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
michael@0 385 sub ecx, byte SIZEOF_XMMWORD
michael@0 386 jz near .nextrow
michael@0 387
michael@0 388 add esi, byte SIZEOF_XMMWORD ; inptr0
michael@0 389 add ebx, byte SIZEOF_XMMWORD ; inptr1
michael@0 390 add edx, byte SIZEOF_XMMWORD ; inptr2
michael@0 391 jmp near .columnloop
michael@0 392 alignx 16,7
michael@0 393
michael@0 394 .column_st32:
michael@0 395 cmp ecx, byte SIZEOF_XMMWORD/2
michael@0 396 jb short .column_st16
michael@0 397 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
michael@0 398 movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
michael@0 399 add edi, byte 2*SIZEOF_XMMWORD ; outptr
michael@0 400 movdqa xmmA,xmmC
michael@0 401 movdqa xmmD,xmmH
michael@0 402 sub ecx, byte SIZEOF_XMMWORD/2
michael@0 403 .column_st16:
michael@0 404 cmp ecx, byte SIZEOF_XMMWORD/4
michael@0 405 jb short .column_st15
michael@0 406 movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
michael@0 407 add edi, byte SIZEOF_XMMWORD ; outptr
michael@0 408 movdqa xmmA,xmmD
michael@0 409 sub ecx, byte SIZEOF_XMMWORD/4
michael@0 410 .column_st15:
michael@0 411 ; Store two pixels (8 bytes) of xmmA to the output when it has enough
michael@0 412 ; space.
michael@0 413 cmp ecx, byte SIZEOF_XMMWORD/8
michael@0 414 jb short .column_st7
michael@0 415 movq XMM_MMWORD [edi], xmmA
michael@0 416 add edi, byte SIZEOF_XMMWORD/8*4
michael@0 417 sub ecx, byte SIZEOF_XMMWORD/8
michael@0 418 psrldq xmmA, SIZEOF_XMMWORD/8*4
michael@0 419 .column_st7:
michael@0 420 ; Store one pixel (4 bytes) of xmmA to the output when it has enough
michael@0 421 ; space.
michael@0 422 test ecx, ecx
michael@0 423 jz short .nextrow
michael@0 424 movd XMM_DWORD [edi], xmmA
michael@0 425
michael@0 426 %endif ; RGB_PIXELSIZE ; ---------------
michael@0 427
michael@0 428 alignx 16,7
michael@0 429
michael@0 430 .nextrow:
michael@0 431 pop ecx
michael@0 432 pop esi
michael@0 433 pop ebx
michael@0 434 pop edx
michael@0 435 pop edi
michael@0 436 pop eax
michael@0 437
michael@0 438 add esi, byte SIZEOF_JSAMPROW
michael@0 439 add ebx, byte SIZEOF_JSAMPROW
michael@0 440 add edx, byte SIZEOF_JSAMPROW
michael@0 441 add edi, byte SIZEOF_JSAMPROW ; output_buf
michael@0 442 dec eax ; num_rows
michael@0 443 jg near .rowloop
michael@0 444
michael@0 445 sfence ; flush the write buffer
michael@0 446
michael@0 447 .return:
michael@0 448 pop edi
michael@0 449 pop esi
michael@0 450 ; pop edx ; need not be preserved
michael@0 451 ; pop ecx ; need not be preserved
michael@0 452 pop ebx
michael@0 453 mov esp,ebp ; esp <- aligned ebp
michael@0 454 pop esp ; esp <- original ebp
michael@0 455 pop ebp
michael@0 456 ret
michael@0 457
michael@0 458 ; For some reason, the OS X linker does not honor the request to align the
michael@0 459 ; segment unless we do this.
michael@0 460 align 16

Mercurial Repository: The Tor Browser

Europalab SCM Repository Server

mercurial