Parent Directory
| 
Revision Log
Revision 1730 - (view) (download)
| 1 : | Skal | 1730 | ;/**************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - optimized SSIM routines - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright(C) 2006 Johannes Reinhardt <johannes.reinhardt@gmx.de> | ||
| 7 : | ; * | ||
| 8 : | ; * This program is free software; you can redistribute it and/or modify it | ||
| 9 : | ; * under the terms of the GNU General Public License as published by | ||
| 10 : | ; * the Free Software Foundation; either version 2 of the License, or | ||
| 11 : | ; * (at your option) any later version. | ||
| 12 : | ; * | ||
| 13 : | ; * This program is distributed in the hope that it will be useful, | ||
| 14 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 15 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 16 : | ; * GNU General Public License for more details. | ||
| 17 : | ; * | ||
| 18 : | ; * You should have received a copy of the GNU General Public License | ||
| 19 : | ; * along with this program; if not, write to the Free Software | ||
| 20 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 21 : | ; * | ||
| 22 : | ; * | ||
| 23 : | ; ***************************************************************************/ | ||
| 24 : | |||
| 25 : | BITS 32 | ||
| 26 : | |||
| 27 : | %macro cglobal 1 | ||
| 28 : | %ifdef PREFIX | ||
| 29 : | %ifdef MARK_FUNCS | ||
| 30 : | global _%1:function %1.endfunc-%1 | ||
| 31 : | %define %1 _%1:function %1.endfunc-%1 | ||
| 32 : | %else | ||
| 33 : | global _%1 | ||
| 34 : | %define %1 _%1 | ||
| 35 : | %endif | ||
| 36 : | %else | ||
| 37 : | %ifdef MARK_FUNCS | ||
| 38 : | global %1:function %1.endfunc-%1 | ||
| 39 : | %else | ||
| 40 : | global %1 | ||
| 41 : | %endif | ||
| 42 : | %endif | ||
| 43 : | %endmacro | ||
| 44 : | |||
| 45 : | %macro ACC_ROW 1 | ||
| 46 : | movq %1,[ecx] | ||
| 47 : | psadbw %1,mm0 | ||
| 48 : | add ecx, edx | ||
| 49 : | %endmacro | ||
| 50 : | |||
| 51 : | ;load a dq from mem to a xmm reg | ||
| 52 : | %macro LOAD_XMM 2 | ||
| 53 : | movdqu %1,[%2] | ||
| 54 : | ;movhps %1,[%2+8] | ||
| 55 : | %endmacro | ||
| 56 : | |||
| 57 : | %macro WRITE_XMM 2 | ||
| 58 : | ;movlps [%1],%2 | ||
| 59 : | ;movhps [%1+8],%2 | ||
| 60 : | movdqu [%1],%2 | ||
| 61 : | %endmacro | ||
| 62 : | |||
| 63 : | %macro CONSIM_1x8_SSE2 0 | ||
| 64 : | LOAD_XMM xmm0,ecx | ||
| 65 : | LOAD_XMM xmm1,edx | ||
| 66 : | pxor xmm2,xmm2 | ||
| 67 : | |||
| 68 : | ;unpack to words | ||
| 69 : | punpcklbw xmm0,xmm2 | ||
| 70 : | punpcklbw xmm1,xmm2 | ||
| 71 : | |||
| 72 : | ;devo | ||
| 73 : | psubw xmm0,xmm6 | ||
| 74 : | movaps xmm2,xmm0 | ||
| 75 : | pmaddwd xmm2,xmm0 | ||
| 76 : | paddd xmm3,xmm2 | ||
| 77 : | |||
| 78 : | ;devc | ||
| 79 : | psubw xmm1,xmm7 | ||
| 80 : | movaps xmm2,xmm1 | ||
| 81 : | pmaddwd xmm2,xmm1 | ||
| 82 : | paddd xmm4,xmm2 | ||
| 83 : | |||
| 84 : | ;corr | ||
| 85 : | pmaddwd xmm1,xmm0 | ||
| 86 : | paddd xmm5,xmm1 | ||
| 87 : | %endmacro | ||
| 88 : | |||
| 89 : | |||
| 90 : | %macro CONSIM_1x8_MMX 0 | ||
| 91 : | movq mm0,[ecx];orig | ||
| 92 : | movq mm1,[edx];comp | ||
| 93 : | pxor mm2,mm2;null vector | ||
| 94 : | |||
| 95 : | ;unpack low half of qw to words | ||
| 96 : | punpcklbw mm0,mm2 | ||
| 97 : | punpcklbw mm1,mm2 | ||
| 98 : | |||
| 99 : | ;devo | ||
| 100 : | psubw mm0,mm6 | ||
| 101 : | movq mm2,mm0 | ||
| 102 : | pmaddwd mm2,mm0 | ||
| 103 : | paddd mm3,mm2; | ||
| 104 : | |||
| 105 : | ;devc | ||
| 106 : | psubw mm1,mm7 | ||
| 107 : | movq mm2,mm1 | ||
| 108 : | pmaddwd mm2,mm1 | ||
| 109 : | paddd mm4,mm2 | ||
| 110 : | |||
| 111 : | ;corr | ||
| 112 : | pmaddwd mm1,mm0 | ||
| 113 : | paddd mm5,mm1 | ||
| 114 : | |||
| 115 : | movq mm0,[ecx] | ||
| 116 : | movq mm1,[edx] | ||
| 117 : | pxor mm2,mm2;null vector | ||
| 118 : | |||
| 119 : | ;unpack high half of qw to words | ||
| 120 : | punpckhbw mm0,mm2 | ||
| 121 : | punpckhbw mm1,mm2 | ||
| 122 : | |||
| 123 : | ;devo | ||
| 124 : | psubw mm0,mm6 | ||
| 125 : | movq mm2,mm0 | ||
| 126 : | pmaddwd mm2,mm0 | ||
| 127 : | paddd mm3,mm2; | ||
| 128 : | |||
| 129 : | ;devc | ||
| 130 : | psubw mm1,mm7 | ||
| 131 : | movq mm2,mm1 | ||
| 132 : | pmaddwd mm2,mm1 | ||
| 133 : | paddd mm4,mm2 | ||
| 134 : | |||
| 135 : | ;corr | ||
| 136 : | pmaddwd mm1,mm0 | ||
| 137 : | paddd mm5,mm1 | ||
| 138 : | %endmacro | ||
| 139 : | |||
| 140 : | |||
| 141 : | |||
| 142 : | SECTION .text | ||
| 143 : | |||
| 144 : | cglobal lum_8x8_mmx | ||
| 145 : | cglobal consim_sse2 | ||
| 146 : | cglobal consim_mmx | ||
| 147 : | |||
| 148 : | ;int lum_8x8_c(uint8_t* ptr, uint32_t stride) | ||
| 149 : | |||
| 150 : | ALIGN 16 | ||
| 151 : | lum_8x8_mmx: | ||
| 152 : | mov ecx, [esp + 4] ;ptr | ||
| 153 : | mov edx, [esp + 8];stride | ||
| 154 : | |||
| 155 : | pxor mm0,mm0 | ||
| 156 : | |||
| 157 : | ACC_ROW mm1 | ||
| 158 : | ACC_ROW mm2 | ||
| 159 : | paddw mm1 ,mm2 | ||
| 160 : | |||
| 161 : | ACC_ROW mm3 | ||
| 162 : | ACC_ROW mm4 | ||
| 163 : | paddw mm3 ,mm4 | ||
| 164 : | |||
| 165 : | ACC_ROW mm5 | ||
| 166 : | ACC_ROW mm6 | ||
| 167 : | paddw mm5, mm6 | ||
| 168 : | |||
| 169 : | ACC_ROW mm7 | ||
| 170 : | ACC_ROW mm4 | ||
| 171 : | paddw mm7, mm4 | ||
| 172 : | |||
| 173 : | paddw mm1, mm3 | ||
| 174 : | paddw mm5, mm7 | ||
| 175 : | paddw mm1, mm5 | ||
| 176 : | |||
| 177 : | movd eax,mm1 | ||
| 178 : | emms | ||
| 179 : | ret | ||
| 180 : | .endfunc | ||
| 181 : | |||
| 182 : | ALIGN 16 | ||
| 183 : | consim_mmx: | ||
| 184 : | mov ecx,[esp+4] ;ptro | ||
| 185 : | pxor mm6,mm6; | ||
| 186 : | |||
| 187 : | mov edx,[esp+8] ;ptrc | ||
| 188 : | pxor mm3,mm3;devo | ||
| 189 : | pxor mm4,mm4;devc | ||
| 190 : | movd mm6,[esp + 16];lumo | ||
| 191 : | pxor mm7,mm7 | ||
| 192 : | mov eax,[esp+12];stride | ||
| 193 : | movd mm7,[esp + 20];lumc | ||
| 194 : | pshufw mm6,mm6,00000000b | ||
| 195 : | pxor mm5,mm5;corr | ||
| 196 : | pshufw mm7,mm7,00000000b | ||
| 197 : | |||
| 198 : | CONSIM_1x8_MMX | ||
| 199 : | add ecx,eax | ||
| 200 : | add edx,eax | ||
| 201 : | CONSIM_1x8_MMX | ||
| 202 : | add ecx,eax | ||
| 203 : | add edx,eax | ||
| 204 : | CONSIM_1x8_MMX | ||
| 205 : | add ecx,eax | ||
| 206 : | add edx,eax | ||
| 207 : | CONSIM_1x8_MMX | ||
| 208 : | add ecx,eax | ||
| 209 : | add edx,eax | ||
| 210 : | CONSIM_1x8_MMX | ||
| 211 : | add ecx,eax | ||
| 212 : | add edx,eax | ||
| 213 : | CONSIM_1x8_MMX | ||
| 214 : | add ecx,eax | ||
| 215 : | add edx,eax | ||
| 216 : | CONSIM_1x8_MMX | ||
| 217 : | add ecx,eax | ||
| 218 : | add edx,eax | ||
| 219 : | CONSIM_1x8_MMX | ||
| 220 : | |||
| 221 : | pshufw mm0,mm3,01001110b | ||
| 222 : | paddd mm3,mm0 | ||
| 223 : | pshufw mm1,mm4,01001110b | ||
| 224 : | paddd mm4,mm1 | ||
| 225 : | pshufw mm2,mm5,01001110b | ||
| 226 : | paddd mm5,mm2 | ||
| 227 : | |||
| 228 : | ;load target pointer | ||
| 229 : | mov ecx,[esp + 24]; pdevo | ||
| 230 : | movd [ecx],mm3 | ||
| 231 : | mov edx,[esp + 28]; pdevc | ||
| 232 : | movd [edx],mm4 | ||
| 233 : | mov eax,[esp + 32]; corr | ||
| 234 : | movd [eax],mm5 | ||
| 235 : | emms | ||
| 236 : | ret | ||
| 237 : | .endfunc | ||
| 238 : | |||
| 239 : | consim_sse2: | ||
| 240 : | mov ecx,[esp+4] ;ptro | ||
| 241 : | pxor xmm6,xmm6; | ||
| 242 : | mov edx,[esp+8] ;ptrc | ||
| 243 : | pxor xmm3,xmm3;devo | ||
| 244 : | pxor xmm4,xmm4;devc | ||
| 245 : | movd xmm6,[esp + 16];lumo | ||
| 246 : | pxor xmm7,xmm7 | ||
| 247 : | mov eax,[esp+12];stride | ||
| 248 : | movd xmm7,[esp + 20];lumc | ||
| 249 : | pxor xmm5,xmm5;corr | ||
| 250 : | |||
| 251 : | ;broadcast lumo/c | ||
| 252 : | ;punpcklbw xmm6,xmm6 | ||
| 253 : | punpcklwd xmm6,xmm6 | ||
| 254 : | pshufd xmm6,xmm6,00000000b;or shufps | ||
| 255 : | ;punpcklbw xmm7,xmm7 | ||
| 256 : | punpcklwd xmm7,xmm7 | ||
| 257 : | pshufd xmm7,xmm7,00000000b | ||
| 258 : | |||
| 259 : | CONSIM_1x8_SSE2 | ||
| 260 : | add ecx,eax | ||
| 261 : | add edx,eax | ||
| 262 : | CONSIM_1x8_SSE2 | ||
| 263 : | add ecx,eax | ||
| 264 : | add edx,eax | ||
| 265 : | CONSIM_1x8_SSE2 | ||
| 266 : | add ecx,eax | ||
| 267 : | add edx,eax | ||
| 268 : | CONSIM_1x8_SSE2 | ||
| 269 : | add ecx,eax | ||
| 270 : | add edx,eax | ||
| 271 : | CONSIM_1x8_SSE2 | ||
| 272 : | add ecx,eax | ||
| 273 : | add edx,eax | ||
| 274 : | CONSIM_1x8_SSE2 | ||
| 275 : | add ecx,eax | ||
| 276 : | add edx,eax | ||
| 277 : | CONSIM_1x8_SSE2 | ||
| 278 : | add ecx,eax | ||
| 279 : | add edx,eax | ||
| 280 : | CONSIM_1x8_SSE2 | ||
| 281 : | |||
| 282 : | ;accumulate xmm3-5 | ||
| 283 : | pshufd xmm0, xmm3, 0EH ; Get bit 64-127 from xmm1 (or use movhlps) | ||
| 284 : | paddd xmm3, xmm0 ; Sums are in 2 dwords | ||
| 285 : | pshufd xmm0, xmm3, 01H ; Get bit 32-63 from xmm0 | ||
| 286 : | paddd xmm3, xmm0 ; Sum is in one dword | ||
| 287 : | |||
| 288 : | pshufd xmm1, xmm4, 0EH ; Get bit 64-127 from xmm1 (or use movhlps) | ||
| 289 : | paddd xmm4, xmm1 ; Sums are in 2 dwords | ||
| 290 : | pshufd xmm1, xmm4, 01H ; Get bit 32-63 from xmm0 | ||
| 291 : | paddd xmm4, xmm1 ; Sum is in one dword | ||
| 292 : | |||
| 293 : | pshufd xmm2, xmm5, 0EH ; Get bit 64-127 from xmm1 (or use movhlps) | ||
| 294 : | paddd xmm5, xmm2 ; Sums are in 2 dwords | ||
| 295 : | pshufd xmm2, xmm5, 01H ; Get bit 32-63 from xmm0 | ||
| 296 : | paddd xmm5, xmm2 ; Sum is in one dword | ||
| 297 : | |||
| 298 : | |||
| 299 : | ;load target pointer | ||
| 300 : | mov ecx,[esp + 24]; pdevo | ||
| 301 : | movd [ecx],xmm3 | ||
| 302 : | mov edx,[esp + 28]; pdevc | ||
| 303 : | movd [edx],xmm4 | ||
| 304 : | mov eax,[esp + 32]; corr | ||
| 305 : | movd [eax],xmm5 | ||
| 306 : | ret | ||
| 307 : | .endfunc |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |