The Tor Browser / annotate

media/libjpeg/simd/jdmrgss2-64.asm@b8a032363ba2 (annotated)

media/libjpeg/simd/jdmrgss2-64.asm

Thu, 22 Jan 2015 13:21:57 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Thu, 22 Jan 2015 13:21:57 +0100
branch
TOR_BUG_9701
changeset 15
b8a032363ba2
permissions
-rw-r--r--

Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6

michael@0 1 ;
michael@0 2 ; jdmrgss2-64.asm - merged upsampling/color conversion (64-bit SSE2)
michael@0 3 ;
michael@0 4 ; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
michael@0 5 ; Copyright 2009, 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 ; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
michael@0 25 ;
michael@0 26 ; GLOBAL(void)
michael@0 27 ; jsimd_h2v1_merged_upsample_sse2 (JDIMENSION output_width,
michael@0 28 ; JSAMPIMAGE input_buf,
michael@0 29 ; JDIMENSION in_row_group_ctr,
michael@0 30 ; JSAMPARRAY output_buf);
michael@0 31 ;
michael@0 32
michael@0 33 ; r10 = JDIMENSION output_width
michael@0 34 ; r11 = JSAMPIMAGE input_buf
michael@0 35 ; r12 = JDIMENSION in_row_group_ctr
michael@0 36 ; r13 = JSAMPARRAY output_buf
michael@0 37
michael@0 38 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
michael@0 39 %define WK_NUM 3
michael@0 40
michael@0 41 align 16
michael@0 42 global EXTN(jsimd_h2v1_merged_upsample_sse2)
michael@0 43
michael@0 44 EXTN(jsimd_h2v1_merged_upsample_sse2):
michael@0 45 push rbp
michael@0 46 mov rax,rsp ; rax = original rbp
michael@0 47 sub rsp, byte 4
michael@0 48 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
michael@0 49 mov [rsp],rax
michael@0 50 mov rbp,rsp ; rbp = aligned rbp
michael@0 51 lea rsp, [wk(0)]
michael@0 52 collect_args
michael@0 53 push rbx
michael@0 54
michael@0 55 mov rcx, r10 ; col
michael@0 56 test rcx,rcx
michael@0 57 jz near .return
michael@0 58
michael@0 59 push rcx
michael@0 60
michael@0 61 mov rdi, r11
michael@0 62 mov rcx, r12
michael@0 63 mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
michael@0 64 mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
michael@0 65 mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
michael@0 66 mov rdi, r13
michael@0 67 mov rsi, JSAMPROW [rsi+rcx*SIZEOF_JSAMPROW] ; inptr0
michael@0 68 mov rbx, JSAMPROW [rbx+rcx*SIZEOF_JSAMPROW] ; inptr1
michael@0 69 mov rdx, JSAMPROW [rdx+rcx*SIZEOF_JSAMPROW] ; inptr2
michael@0 70 mov rdi, JSAMPROW [rdi] ; outptr
michael@0 71
michael@0 72 pop rcx ; col
michael@0 73
michael@0 74 .columnloop:
michael@0 75
michael@0 76 movdqa xmm6, XMMWORD [rbx] ; xmm6=Cb(0123456789ABCDEF)
michael@0 77 movdqa xmm7, XMMWORD [rdx] ; xmm7=Cr(0123456789ABCDEF)
michael@0 78
michael@0 79 pxor xmm1,xmm1 ; xmm1=(all 0's)
michael@0 80 pcmpeqw xmm3,xmm3
michael@0 81 psllw xmm3,7 ; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
michael@0 82
michael@0 83 movdqa xmm4,xmm6
michael@0 84 punpckhbw xmm6,xmm1 ; xmm6=Cb(89ABCDEF)=CbH
michael@0 85 punpcklbw xmm4,xmm1 ; xmm4=Cb(01234567)=CbL
michael@0 86 movdqa xmm0,xmm7
michael@0 87 punpckhbw xmm7,xmm1 ; xmm7=Cr(89ABCDEF)=CrH
michael@0 88 punpcklbw xmm0,xmm1 ; xmm0=Cr(01234567)=CrL
michael@0 89
michael@0 90 paddw xmm6,xmm3
michael@0 91 paddw xmm4,xmm3
michael@0 92 paddw xmm7,xmm3
michael@0 93 paddw xmm0,xmm3
michael@0 94
michael@0 95 ; (Original)
michael@0 96 ; R = Y + 1.40200 * Cr
michael@0 97 ; G = Y - 0.34414 * Cb - 0.71414 * Cr
michael@0 98 ; B = Y + 1.77200 * Cb
michael@0 99 ;
michael@0 100 ; (This implementation)
michael@0 101 ; R = Y + 0.40200 * Cr + Cr
michael@0 102 ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
michael@0 103 ; B = Y - 0.22800 * Cb + Cb + Cb
michael@0 104
michael@0 105 movdqa xmm5,xmm6 ; xmm5=CbH
michael@0 106 movdqa xmm2,xmm4 ; xmm2=CbL
michael@0 107 paddw xmm6,xmm6 ; xmm6=2*CbH
michael@0 108 paddw xmm4,xmm4 ; xmm4=2*CbL
michael@0 109 movdqa xmm1,xmm7 ; xmm1=CrH
michael@0 110 movdqa xmm3,xmm0 ; xmm3=CrL
michael@0 111 paddw xmm7,xmm7 ; xmm7=2*CrH
michael@0 112 paddw xmm0,xmm0 ; xmm0=2*CrL
michael@0 113
michael@0 114 pmulhw xmm6,[rel PW_MF0228] ; xmm6=(2*CbH * -FIX(0.22800))
michael@0 115 pmulhw xmm4,[rel PW_MF0228] ; xmm4=(2*CbL * -FIX(0.22800))
michael@0 116 pmulhw xmm7,[rel PW_F0402] ; xmm7=(2*CrH * FIX(0.40200))
michael@0 117 pmulhw xmm0,[rel PW_F0402] ; xmm0=(2*CrL * FIX(0.40200))
michael@0 118
michael@0 119 paddw xmm6,[rel PW_ONE]
michael@0 120 paddw xmm4,[rel PW_ONE]
michael@0 121 psraw xmm6,1 ; xmm6=(CbH * -FIX(0.22800))
michael@0 122 psraw xmm4,1 ; xmm4=(CbL * -FIX(0.22800))
michael@0 123 paddw xmm7,[rel PW_ONE]
michael@0 124 paddw xmm0,[rel PW_ONE]
michael@0 125 psraw xmm7,1 ; xmm7=(CrH * FIX(0.40200))
michael@0 126 psraw xmm0,1 ; xmm0=(CrL * FIX(0.40200))
michael@0 127
michael@0 128 paddw xmm6,xmm5
michael@0 129 paddw xmm4,xmm2
michael@0 130 paddw xmm6,xmm5 ; xmm6=(CbH * FIX(1.77200))=(B-Y)H
michael@0 131 paddw xmm4,xmm2 ; xmm4=(CbL * FIX(1.77200))=(B-Y)L
michael@0 132 paddw xmm7,xmm1 ; xmm7=(CrH * FIX(1.40200))=(R-Y)H
michael@0 133 paddw xmm0,xmm3 ; xmm0=(CrL * FIX(1.40200))=(R-Y)L
michael@0 134
michael@0 135 movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=(B-Y)H
michael@0 136 movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(R-Y)H
michael@0 137
michael@0 138 movdqa xmm6,xmm5
michael@0 139 movdqa xmm7,xmm2
michael@0 140 punpcklwd xmm5,xmm1
michael@0 141 punpckhwd xmm6,xmm1
michael@0 142 pmaddwd xmm5,[rel PW_MF0344_F0285]
michael@0 143 pmaddwd xmm6,[rel PW_MF0344_F0285]
michael@0 144 punpcklwd xmm2,xmm3
michael@0 145 punpckhwd xmm7,xmm3
michael@0 146 pmaddwd xmm2,[rel PW_MF0344_F0285]
michael@0 147 pmaddwd xmm7,[rel PW_MF0344_F0285]
michael@0 148
michael@0 149 paddd xmm5,[rel PD_ONEHALF]
michael@0 150 paddd xmm6,[rel PD_ONEHALF]
michael@0 151 psrad xmm5,SCALEBITS
michael@0 152 psrad xmm6,SCALEBITS
michael@0 153 paddd xmm2,[rel PD_ONEHALF]
michael@0 154 paddd xmm7,[rel PD_ONEHALF]
michael@0 155 psrad xmm2,SCALEBITS
michael@0 156 psrad xmm7,SCALEBITS
michael@0 157
michael@0 158 packssdw xmm5,xmm6 ; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
michael@0 159 packssdw xmm2,xmm7 ; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
michael@0 160 psubw xmm5,xmm1 ; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
michael@0 161 psubw xmm2,xmm3 ; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
michael@0 162
michael@0 163 movdqa XMMWORD [wk(2)], xmm5 ; wk(2)=(G-Y)H
michael@0 164
michael@0 165 mov al,2 ; Yctr
michael@0 166 jmp short .Yloop_1st
michael@0 167
michael@0 168 .Yloop_2nd:
michael@0 169 movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(R-Y)H
michael@0 170 movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(G-Y)H
michael@0 171 movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(B-Y)H
michael@0 172
michael@0 173 .Yloop_1st:
michael@0 174 movdqa xmm7, XMMWORD [rsi] ; xmm7=Y(0123456789ABCDEF)
michael@0 175
michael@0 176 pcmpeqw xmm6,xmm6
michael@0 177 psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
michael@0 178 pand xmm6,xmm7 ; xmm6=Y(02468ACE)=YE
michael@0 179 psrlw xmm7,BYTE_BIT ; xmm7=Y(13579BDF)=YO
michael@0 180
michael@0 181 movdqa xmm1,xmm0 ; xmm1=xmm0=(R-Y)(L/H)
michael@0 182 movdqa xmm3,xmm2 ; xmm3=xmm2=(G-Y)(L/H)
michael@0 183 movdqa xmm5,xmm4 ; xmm5=xmm4=(B-Y)(L/H)
michael@0 184
michael@0 185 paddw xmm0,xmm6 ; xmm0=((R-Y)+YE)=RE=R(02468ACE)
michael@0 186 paddw xmm1,xmm7 ; xmm1=((R-Y)+YO)=RO=R(13579BDF)
michael@0 187 packuswb xmm0,xmm0 ; xmm0=R(02468ACE********)
michael@0 188 packuswb xmm1,xmm1 ; xmm1=R(13579BDF********)
michael@0 189
michael@0 190 paddw xmm2,xmm6 ; xmm2=((G-Y)+YE)=GE=G(02468ACE)
michael@0 191 paddw xmm3,xmm7 ; xmm3=((G-Y)+YO)=GO=G(13579BDF)
michael@0 192 packuswb xmm2,xmm2 ; xmm2=G(02468ACE********)
michael@0 193 packuswb xmm3,xmm3 ; xmm3=G(13579BDF********)
michael@0 194
michael@0 195 paddw xmm4,xmm6 ; xmm4=((B-Y)+YE)=BE=B(02468ACE)
michael@0 196 paddw xmm5,xmm7 ; xmm5=((B-Y)+YO)=BO=B(13579BDF)
michael@0 197 packuswb xmm4,xmm4 ; xmm4=B(02468ACE********)
michael@0 198 packuswb xmm5,xmm5 ; xmm5=B(13579BDF********)
michael@0 199
michael@0 200 %if RGB_PIXELSIZE == 3 ; ---------------
michael@0 201
michael@0 202 ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
michael@0 203 ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
michael@0 204 ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
michael@0 205 ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
michael@0 206
michael@0 207 punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
michael@0 208 punpcklbw xmmE,xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
michael@0 209 punpcklbw xmmD,xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
michael@0 210
michael@0 211 movdqa xmmG,xmmA
michael@0 212 movdqa xmmH,xmmA
michael@0 213 punpcklwd xmmA,xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
michael@0 214 punpckhwd xmmG,xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
michael@0 215
michael@0 216 psrldq xmmH,2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
michael@0 217 psrldq xmmE,2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
michael@0 218
michael@0 219 movdqa xmmC,xmmD
michael@0 220 movdqa xmmB,xmmD
michael@0 221 punpcklwd xmmD,xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
michael@0 222 punpckhwd xmmC,xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
michael@0 223
michael@0 224 psrldq xmmB,2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
michael@0 225
michael@0 226 movdqa xmmF,xmmE
michael@0 227 punpcklwd xmmE,xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
michael@0 228 punpckhwd xmmF,xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
michael@0 229
michael@0 230 pshufd xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
michael@0 231 movdqa xmmB,xmmE
michael@0 232 punpckldq xmmA,xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
michael@0 233 punpckldq xmmE,xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
michael@0 234 punpckhdq xmmD,xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
michael@0 235
michael@0 236 pshufd xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
michael@0 237 movdqa xmmB,xmmF
michael@0 238 punpckldq xmmG,xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
michael@0 239 punpckldq xmmF,xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
michael@0 240 punpckhdq xmmC,xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
michael@0 241
michael@0 242 punpcklqdq xmmA,xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
michael@0 243 punpcklqdq xmmD,xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
michael@0 244 punpcklqdq xmmF,xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
michael@0 245
michael@0 246 cmp rcx, byte SIZEOF_XMMWORD
michael@0 247 jb short .column_st32
michael@0 248
michael@0 249 test rdi, SIZEOF_XMMWORD-1
michael@0 250 jnz short .out1
michael@0 251 ; --(aligned)-------------------
michael@0 252 movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
michael@0 253 movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
michael@0 254 movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
michael@0 255 jmp short .out0
michael@0 256 .out1: ; --(unaligned)-----------------
michael@0 257 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
michael@0 258 movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
michael@0 259 movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
michael@0 260 .out0:
michael@0 261 add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
michael@0 262 sub rcx, byte SIZEOF_XMMWORD
michael@0 263 jz near .endcolumn
michael@0 264
michael@0 265 add rsi, byte SIZEOF_XMMWORD ; inptr0
michael@0 266 dec al ; Yctr
michael@0 267 jnz near .Yloop_2nd
michael@0 268
michael@0 269 add rbx, byte SIZEOF_XMMWORD ; inptr1
michael@0 270 add rdx, byte SIZEOF_XMMWORD ; inptr2
michael@0 271 jmp near .columnloop
michael@0 272
michael@0 273 .column_st32:
michael@0 274 lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE
michael@0 275 cmp rcx, byte 2*SIZEOF_XMMWORD
michael@0 276 jb short .column_st16
michael@0 277 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
michael@0 278 movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
michael@0 279 add rdi, byte 2*SIZEOF_XMMWORD ; outptr
michael@0 280 movdqa xmmA,xmmF
michael@0 281 sub rcx, byte 2*SIZEOF_XMMWORD
michael@0 282 jmp short .column_st15
michael@0 283 .column_st16:
michael@0 284 cmp rcx, byte SIZEOF_XMMWORD
michael@0 285 jb short .column_st15
michael@0 286 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
michael@0 287 add rdi, byte SIZEOF_XMMWORD ; outptr
michael@0 288 movdqa xmmA,xmmD
michael@0 289 sub rcx, byte SIZEOF_XMMWORD
michael@0 290 .column_st15:
michael@0 291 ; Store the lower 8 bytes of xmmA to the output when it has enough
michael@0 292 ; space.
michael@0 293 cmp rcx, byte SIZEOF_MMWORD
michael@0 294 jb short .column_st7
michael@0 295 movq XMM_MMWORD [rdi], xmmA
michael@0 296 add rdi, byte SIZEOF_MMWORD
michael@0 297 sub rcx, byte SIZEOF_MMWORD
michael@0 298 psrldq xmmA, SIZEOF_MMWORD
michael@0 299 .column_st7:
michael@0 300 ; Store the lower 4 bytes of xmmA to the output when it has enough
michael@0 301 ; space.
michael@0 302 cmp rcx, byte SIZEOF_DWORD
michael@0 303 jb short .column_st3
michael@0 304 movd XMM_DWORD [rdi], xmmA
michael@0 305 add rdi, byte SIZEOF_DWORD
michael@0 306 sub rcx, byte SIZEOF_DWORD
michael@0 307 psrldq xmmA, SIZEOF_DWORD
michael@0 308 .column_st3:
michael@0 309 ; Store the lower 2 bytes of rax to the output when it has enough
michael@0 310 ; space.
michael@0 311 movd eax, xmmA
michael@0 312 cmp rcx, byte SIZEOF_WORD
michael@0 313 jb short .column_st1
michael@0 314 mov WORD [rdi], ax
michael@0 315 add rdi, byte SIZEOF_WORD
michael@0 316 sub rcx, byte SIZEOF_WORD
michael@0 317 shr rax, 16
michael@0 318 .column_st1:
michael@0 319 ; Store the lower 1 byte of rax to the output when it has enough
michael@0 320 ; space.
michael@0 321 test rcx, rcx
michael@0 322 jz short .endcolumn
michael@0 323 mov BYTE [rdi], al
michael@0 324
michael@0 325 %else ; RGB_PIXELSIZE == 4 ; -----------
michael@0 326
michael@0 327 %ifdef RGBX_FILLER_0XFF
michael@0 328 pcmpeqb xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
michael@0 329 pcmpeqb xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
michael@0 330 %else
michael@0 331 pxor xmm6,xmm6 ; xmm6=XE=X(02468ACE********)
michael@0 332 pxor xmm7,xmm7 ; xmm7=XO=X(13579BDF********)
michael@0 333 %endif
michael@0 334 ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
michael@0 335 ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
michael@0 336 ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
michael@0 337 ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
michael@0 338
michael@0 339 punpcklbw xmmA,xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
michael@0 340 punpcklbw xmmE,xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
michael@0 341 punpcklbw xmmB,xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
michael@0 342 punpcklbw xmmF,xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
michael@0 343
michael@0 344 movdqa xmmC,xmmA
michael@0 345 punpcklwd xmmA,xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
michael@0 346 punpckhwd xmmC,xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
michael@0 347 movdqa xmmG,xmmB
michael@0 348 punpcklwd xmmB,xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
michael@0 349 punpckhwd xmmG,xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
michael@0 350
michael@0 351 movdqa xmmD,xmmA
michael@0 352 punpckldq xmmA,xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
michael@0 353 punpckhdq xmmD,xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
michael@0 354 movdqa xmmH,xmmC
michael@0 355 punpckldq xmmC,xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
michael@0 356 punpckhdq xmmH,xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
michael@0 357
michael@0 358 cmp rcx, byte SIZEOF_XMMWORD
michael@0 359 jb short .column_st32
michael@0 360
michael@0 361 test rdi, SIZEOF_XMMWORD-1
michael@0 362 jnz short .out1
michael@0 363 ; --(aligned)-------------------
michael@0 364 movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
michael@0 365 movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
michael@0 366 movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
michael@0 367 movntdq XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
michael@0 368 jmp short .out0
michael@0 369 .out1: ; --(unaligned)-----------------
michael@0 370 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
michael@0 371 movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
michael@0 372 movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
michael@0 373 movdqu XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
michael@0 374 .out0:
michael@0 375 add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr
michael@0 376 sub rcx, byte SIZEOF_XMMWORD
michael@0 377 jz near .endcolumn
michael@0 378
michael@0 379 add rsi, byte SIZEOF_XMMWORD ; inptr0
michael@0 380 dec al ; Yctr
michael@0 381 jnz near .Yloop_2nd
michael@0 382
michael@0 383 add rbx, byte SIZEOF_XMMWORD ; inptr1
michael@0 384 add rdx, byte SIZEOF_XMMWORD ; inptr2
michael@0 385 jmp near .columnloop
michael@0 386
michael@0 387 .column_st32:
michael@0 388 cmp rcx, byte SIZEOF_XMMWORD/2
michael@0 389 jb short .column_st16
michael@0 390 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
michael@0 391 movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
michael@0 392 add rdi, byte 2*SIZEOF_XMMWORD ; outptr
michael@0 393 movdqa xmmA,xmmC
michael@0 394 movdqa xmmD,xmmH
michael@0 395 sub rcx, byte SIZEOF_XMMWORD/2
michael@0 396 .column_st16:
michael@0 397 cmp rcx, byte SIZEOF_XMMWORD/4
michael@0 398 jb short .column_st15
michael@0 399 movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
michael@0 400 add rdi, byte SIZEOF_XMMWORD ; outptr
michael@0 401 movdqa xmmA,xmmD
michael@0 402 sub rcx, byte SIZEOF_XMMWORD/4
michael@0 403 .column_st15:
michael@0 404 ; Store two pixels (8 bytes) of xmmA to the output when it has enough
michael@0 405 ; space.
michael@0 406 cmp rcx, byte SIZEOF_XMMWORD/8
michael@0 407 jb short .column_st7
michael@0 408 movq XMM_MMWORD [rdi], xmmA
michael@0 409 add rdi, byte SIZEOF_XMMWORD/8*4
michael@0 410 sub rcx, byte SIZEOF_XMMWORD/8
michael@0 411 psrldq xmmA, SIZEOF_XMMWORD/8*4
michael@0 412 .column_st7:
michael@0 413 ; Store one pixel (4 bytes) of xmmA to the output when it has enough
michael@0 414 ; space.
michael@0 415 test rcx, rcx
michael@0 416 jz short .endcolumn
michael@0 417 movd XMM_DWORD [rdi], xmmA
michael@0 418
michael@0 419 %endif ; RGB_PIXELSIZE ; ---------------
michael@0 420
michael@0 421 .endcolumn:
michael@0 422 sfence ; flush the write buffer
michael@0 423
michael@0 424 .return:
michael@0 425 pop rbx
michael@0 426 uncollect_args
michael@0 427 mov rsp,rbp ; rsp <- aligned rbp
michael@0 428 pop rsp ; rsp <- original rbp
michael@0 429 pop rbp
michael@0 430 ret
michael@0 431
michael@0 432 ; --------------------------------------------------------------------------
michael@0 433 ;
michael@0 434 ; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
michael@0 435 ;
michael@0 436 ; GLOBAL(void)
michael@0 437 ; jsimd_h2v2_merged_upsample_sse2 (JDIMENSION output_width,
michael@0 438 ; JSAMPIMAGE input_buf,
michael@0 439 ; JDIMENSION in_row_group_ctr,
michael@0 440 ; JSAMPARRAY output_buf);
michael@0 441 ;
michael@0 442
michael@0 443 ; r10 = JDIMENSION output_width
michael@0 444 ; r11 = JSAMPIMAGE input_buf
michael@0 445 ; r12 = JDIMENSION in_row_group_ctr
michael@0 446 ; r13 = JSAMPARRAY output_buf
michael@0 447
michael@0 448 align 16
michael@0 449 global EXTN(jsimd_h2v2_merged_upsample_sse2)
michael@0 450
michael@0 451 EXTN(jsimd_h2v2_merged_upsample_sse2):
michael@0 452 push rbp
michael@0 453 mov rax,rsp
michael@0 454 mov rbp,rsp
michael@0 455 collect_args
michael@0 456 push rbx
michael@0 457
michael@0 458 mov rax, r10
michael@0 459
michael@0 460 mov rdi, r11
michael@0 461 mov rcx, r12
michael@0 462 mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
michael@0 463 mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
michael@0 464 mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
michael@0 465 mov rdi, r13
michael@0 466 lea rsi, [rsi+rcx*SIZEOF_JSAMPROW]
michael@0 467
michael@0 468 push rdx ; inptr2
michael@0 469 push rbx ; inptr1
michael@0 470 push rsi ; inptr00
michael@0 471 mov rbx,rsp
michael@0 472
michael@0 473 push rdi
michael@0 474 push rcx
michael@0 475 push rax
michael@0 476
michael@0 477 %ifdef WIN64
michael@0 478 mov r8, rcx
michael@0 479 mov r9, rdi
michael@0 480 mov rcx, rax
michael@0 481 mov rdx, rbx
michael@0 482 %else
michael@0 483 mov rdx, rcx
michael@0 484 mov rcx, rdi
michael@0 485 mov rdi, rax
michael@0 486 mov rsi, rbx
michael@0 487 %endif
michael@0 488
michael@0 489 call EXTN(jsimd_h2v1_merged_upsample_sse2)
michael@0 490
michael@0 491 pop rax
michael@0 492 pop rcx
michael@0 493 pop rdi
michael@0 494 pop rsi
michael@0 495 pop rbx
michael@0 496 pop rdx
michael@0 497
michael@0 498 add rdi, byte SIZEOF_JSAMPROW ; outptr1
michael@0 499 add rsi, byte SIZEOF_JSAMPROW ; inptr01
michael@0 500
michael@0 501 push rdx ; inptr2
michael@0 502 push rbx ; inptr1
michael@0 503 push rsi ; inptr00
michael@0 504 mov rbx,rsp
michael@0 505
michael@0 506 push rdi
michael@0 507 push rcx
michael@0 508 push rax
michael@0 509
michael@0 510 %ifdef WIN64
michael@0 511 mov r8, rcx
michael@0 512 mov r9, rdi
michael@0 513 mov rcx, rax
michael@0 514 mov rdx, rbx
michael@0 515 %else
michael@0 516 mov rdx, rcx
michael@0 517 mov rcx, rdi
michael@0 518 mov rdi, rax
michael@0 519 mov rsi, rbx
michael@0 520 %endif
michael@0 521
michael@0 522 call EXTN(jsimd_h2v1_merged_upsample_sse2)
michael@0 523
michael@0 524 pop rax
michael@0 525 pop rcx
michael@0 526 pop rdi
michael@0 527 pop rsi
michael@0 528 pop rbx
michael@0 529 pop rdx
michael@0 530
michael@0 531 pop rbx
michael@0 532 uncollect_args
michael@0 533 pop rbp
michael@0 534 ret
michael@0 535
michael@0 536 ; For some reason, the OS X linker does not honor the request to align the
michael@0 537 ; segment unless we do this.
michael@0 538 align 16

Mercurial Repository: The Tor Browser

Europalab SCM Repository Server

mercurial