Parent Directory
| 
Revision Log
Revision 262 -
(view)
(download)
Original Path: trunk/xvidcore/src/motion/x86_asm/sad_mmx.asm
| 1 : | Isibaar | 262 | ;/************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * mmx sum of absolute difference | ||
| 5 : | ; * | ||
| 6 : | ; * This program is free software; you can redistribute it and/or modify | ||
| 7 : | ; * it under the terms of the GNU General Public License as published by | ||
| 8 : | ; * the Free Software Foundation; either version 2 of the License, or | ||
| 9 : | ; * (at your option) any later version. | ||
| 10 : | ; * | ||
| 11 : | ; * This program is distributed in the hope that it will be useful, | ||
| 12 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 14 : | ; * GNU General Public License for more details. | ||
| 15 : | ; * | ||
| 16 : | ; * You should have received a copy of the GNU General Public License | ||
| 17 : | ; * along with this program; if not, write to the Free Software | ||
| 18 : | ; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | ||
| 19 : | ; * | ||
| 20 : | ; *************************************************************************/ | ||
| 21 : | |||
| 22 : | ;/************************************************************************** | ||
| 23 : | ; * | ||
| 24 : | ; * History: | ||
| 25 : | ; * | ||
| 26 : | ; * 23.07.2002 sad[16,8]bi_xmm; <pross@xvid.org> | ||
| 27 : | ; * 04.06.2002 cleanup -Skal- | ||
| 28 : | ; * 12.11.2001 inital version; (c)2001 peter ross <pross@cs.rmit.edu.au> | ||
| 29 : | ; * | ||
| 30 : | ; *************************************************************************/ | ||
| 31 : | |||
| 32 : | bits 32 | ||
| 33 : | |||
| 34 : | %macro cglobal 1 | ||
| 35 : | global _%1 | ||
| 36 : | %define %1 _%1 | ||
| 37 : | %else | ||
| 38 : | global %1 | ||
| 39 : | %endif | ||
| 40 : | %endmacro | ||
| 41 : | |||
| 42 : | section .data | ||
| 43 : | |||
| 44 : | align 16 | ||
| 45 : | mmx_one times 4 dw 1 | ||
| 46 : | |||
| 47 : | section .text | ||
| 48 : | |||
| 49 : | cglobal sad16_mmx | ||
| 50 : | cglobal sad8_mmx | ||
| 51 : | cglobal sad16bi_mmx | ||
| 52 : | cglobal sad8bi_mmx | ||
| 53 : | cglobal dev16_mmx | ||
| 54 : | |||
| 55 : | ;=========================================================================== | ||
| 56 : | ; | ||
| 57 : | ; uint32_t sad16_mmx(const uint8_t * const cur, | ||
| 58 : | ; const uint8_t * const ref, | ||
| 59 : | ; const uint32_t stride, | ||
| 60 : | ; (early termination ignore; slows this down) | ||
| 61 : | ; | ||
| 62 : | ;=========================================================================== | ||
| 63 : | |||
| 64 : | %macro SAD_16x16_MMX 0 | ||
| 65 : | movq mm0, [eax] | ||
| 66 : | movq mm1, [edx] | ||
| 67 : | |||
| 68 : | movq mm2, [eax+8] | ||
| 69 : | movq mm3, [edx+8] | ||
| 70 : | |||
| 71 : | movq mm4, mm0 | ||
| 72 : | psubusb mm0, mm1 | ||
| 73 : | lea eax,[eax+ecx] | ||
| 74 : | movq mm5, mm2 | ||
| 75 : | psubusb mm2, mm3 | ||
| 76 : | lea edx,[edx+ecx] | ||
| 77 : | |||
| 78 : | psubusb mm1, mm4 | ||
| 79 : | por mm0, mm1 | ||
| 80 : | psubusb mm3, mm5 | ||
| 81 : | por mm2, mm3 | ||
| 82 : | |||
| 83 : | movq mm1,mm0 | ||
| 84 : | movq mm3,mm2 | ||
| 85 : | |||
| 86 : | punpcklbw mm0,mm7 | ||
| 87 : | punpckhbw mm1,mm7 | ||
| 88 : | punpcklbw mm2,mm7 | ||
| 89 : | punpckhbw mm3,mm7 | ||
| 90 : | |||
| 91 : | paddusw mm0,mm1 | ||
| 92 : | paddusw mm6,mm0 | ||
| 93 : | paddusw mm2,mm3 | ||
| 94 : | paddusw mm6,mm2 | ||
| 95 : | %endmacro | ||
| 96 : | |||
| 97 : | align 16 | ||
| 98 : | sad16_mmx: | ||
| 99 : | |||
| 100 : | mov eax, [esp+ 4] ; Src1 | ||
| 101 : | mov edx, [esp+ 8] ; Src2 | ||
| 102 : | mov ecx, [esp+12] ; Stride | ||
| 103 : | |||
| 104 : | pxor mm6, mm6 ; accum | ||
| 105 : | pxor mm7, mm7 ; zero | ||
| 106 : | |||
| 107 : | SAD_16x16_MMX | ||
| 108 : | SAD_16x16_MMX | ||
| 109 : | SAD_16x16_MMX | ||
| 110 : | SAD_16x16_MMX | ||
| 111 : | SAD_16x16_MMX | ||
| 112 : | SAD_16x16_MMX | ||
| 113 : | SAD_16x16_MMX | ||
| 114 : | SAD_16x16_MMX | ||
| 115 : | |||
| 116 : | SAD_16x16_MMX | ||
| 117 : | SAD_16x16_MMX | ||
| 118 : | SAD_16x16_MMX | ||
| 119 : | SAD_16x16_MMX | ||
| 120 : | SAD_16x16_MMX | ||
| 121 : | SAD_16x16_MMX | ||
| 122 : | SAD_16x16_MMX | ||
| 123 : | SAD_16x16_MMX | ||
| 124 : | |||
| 125 : | pmaddwd mm6, [mmx_one] ; collapse | ||
| 126 : | movq mm7, mm6 | ||
| 127 : | psrlq mm7, 32 | ||
| 128 : | paddd mm6, mm7 | ||
| 129 : | |||
| 130 : | movd eax, mm6 | ||
| 131 : | |||
| 132 : | ret | ||
| 133 : | |||
| 134 : | |||
| 135 : | ;=========================================================================== | ||
| 136 : | ; | ||
| 137 : | ; uint32_t sad8_mmx(const uint8_t * const cur, | ||
| 138 : | ; const uint8_t * const ref, | ||
| 139 : | ; const uint32_t stride); | ||
| 140 : | ; | ||
| 141 : | ;=========================================================================== | ||
| 142 : | |||
| 143 : | %macro SAD_8x8_MMX 0 | ||
| 144 : | movq mm0, [eax] | ||
| 145 : | movq mm1, [edx] | ||
| 146 : | |||
| 147 : | movq mm2, [eax+ecx] | ||
| 148 : | movq mm3, [edx+ecx] | ||
| 149 : | |||
| 150 : | lea eax,[eax+2*ecx] | ||
| 151 : | lea edx,[edx+2*ecx] | ||
| 152 : | |||
| 153 : | movq mm4, mm0 | ||
| 154 : | psubusb mm0, mm1 | ||
| 155 : | movq mm5, mm2 | ||
| 156 : | psubusb mm2, mm3 | ||
| 157 : | |||
| 158 : | psubusb mm1, mm4 | ||
| 159 : | por mm0, mm1 | ||
| 160 : | psubusb mm3, mm5 | ||
| 161 : | por mm2, mm3 | ||
| 162 : | |||
| 163 : | movq mm1,mm0 | ||
| 164 : | movq mm3,mm2 | ||
| 165 : | |||
| 166 : | punpcklbw mm0,mm7 | ||
| 167 : | punpckhbw mm1,mm7 | ||
| 168 : | punpcklbw mm2,mm7 | ||
| 169 : | punpckhbw mm3,mm7 | ||
| 170 : | |||
| 171 : | paddusw mm0,mm1 | ||
| 172 : | paddusw mm6,mm0 | ||
| 173 : | paddusw mm2,mm3 | ||
| 174 : | paddusw mm6,mm2 | ||
| 175 : | %endmacro | ||
| 176 : | |||
| 177 : | align 16 | ||
| 178 : | sad8_mmx: | ||
| 179 : | |||
| 180 : | mov eax, [esp+ 4] ; Src1 | ||
| 181 : | mov edx, [esp+ 8] ; Src2 | ||
| 182 : | mov ecx, [esp+12] ; Stride | ||
| 183 : | |||
| 184 : | pxor mm6, mm6 ; accum | ||
| 185 : | pxor mm7, mm7 ; zero | ||
| 186 : | |||
| 187 : | SAD_8x8_MMX | ||
| 188 : | SAD_8x8_MMX | ||
| 189 : | SAD_8x8_MMX | ||
| 190 : | SAD_8x8_MMX | ||
| 191 : | |||
| 192 : | pmaddwd mm6, [mmx_one] ; collapse | ||
| 193 : | movq mm7, mm6 | ||
| 194 : | psrlq mm7, 32 | ||
| 195 : | paddd mm6, mm7 | ||
| 196 : | |||
| 197 : | movd eax, mm6 | ||
| 198 : | |||
| 199 : | ret | ||
| 200 : | |||
| 201 : | |||
| 202 : | |||
| 203 : | |||
| 204 : | |||
| 205 : | ;=========================================================================== | ||
| 206 : | ; | ||
| 207 : | ; uint32_t sad16bi_mmx(const uint8_t * const cur, | ||
| 208 : | ; const uint8_t * const ref1, | ||
| 209 : | movq mm0, [eax] | ||
| 210 : | movq mm2, [eax+8] | ||
| 211 : | lea eax,[eax+ecx] | ||
| 212 : | movq mm1, mm0 | ||
| 213 : | movq mm3, mm2 | ||
| 214 : | punpcklbw mm0,mm7 | ||
| 215 : | punpcklbw mm2,mm7 | ||
| 216 : | punpckhbw mm1,mm7 | ||
| 217 : | punpckhbw mm3,mm7 | ||
| 218 : | paddw mm5, mm0 | ||
| 219 : | paddw mm6, mm1 | ||
| 220 : | paddw mm5, mm2 | ||
| 221 : | paddw mm6, mm3 | ||
| 222 : | %endmacro | ||
| 223 : | |||
| 224 : | %macro ABS_16x16_MMX 0 | ||
| 225 : | movq mm0, [eax] | ||
| 226 : | movq mm2, [eax+8] | ||
| 227 : | lea eax,[eax+ecx] | ||
| 228 : | movq mm1, mm0 | ||
| 229 : | movq mm3, mm2 | ||
| 230 : | punpcklbw mm0, mm7 | ||
| 231 : | punpcklbw mm2, mm7 | ||
| 232 : | punpckhbw mm1, mm7 | ||
| 233 : | punpckhbw mm3, mm7 | ||
| 234 : | movq mm4, mm6 | ||
| 235 : | psubusw mm4, mm0 | ||
| 236 : | |||
| 237 : | psubusw mm0, mm6 | ||
| 238 : | por mm0, mm4 | ||
| 239 : | movq mm4, mm6 | ||
| 240 : | psubusw mm4, mm1 | ||
| 241 : | psubusw mm1, mm6 | ||
| 242 : | por mm1, mm4 | ||
| 243 : | |||
| 244 : | movq mm4, mm6 | ||
| 245 : | psubusw mm4, mm2 | ||
| 246 : | psubusw mm2, mm6 | ||
| 247 : | por mm2, mm4 | ||
| 248 : | movq mm4, mm6 | ||
| 249 : | psubusw mm4, mm3 | ||
| 250 : | psubusw mm3, mm6 | ||
| 251 : | por mm3, mm4 | ||
| 252 : | |||
| 253 : | paddw mm0, mm1 | ||
| 254 : | paddw mm2, mm3 | ||
| 255 : | paddw mm5, mm0 | ||
| 256 : | paddw mm5, mm2 | ||
| 257 : | %endmacro | ||
| 258 : | |||
| 259 : | align 16 | ||
| 260 : | dev16_mmx: | ||
| 261 : | mov eax, [esp+ 4] ; Src | ||
| 262 : | mov ecx, [esp+ 8] ; Stride | ||
| 263 : | |||
| 264 : | pxor mm7, mm7 ; zero | ||
| 265 : | pxor mm5, mm5 ; accum1 | ||
| 266 : | pxor mm6, mm6 ; accum2 | ||
| 267 : | |||
| 268 : | MEAN_16x16_MMX | ||
| 269 : | MEAN_16x16_MMX | ||
| 270 : | MEAN_16x16_MMX | ||
| 271 : | MEAN_16x16_MMX | ||
| 272 : | MEAN_16x16_MMX | ||
| 273 : | MEAN_16x16_MMX | ||
| 274 : | MEAN_16x16_MMX | ||
| 275 : | MEAN_16x16_MMX | ||
| 276 : | |||
| 277 : | MEAN_16x16_MMX | ||
| 278 : | MEAN_16x16_MMX | ||
| 279 : | MEAN_16x16_MMX | ||
| 280 : | MEAN_16x16_MMX | ||
| 281 : | MEAN_16x16_MMX | ||
| 282 : | MEAN_16x16_MMX | ||
| 283 : | MEAN_16x16_MMX | ||
| 284 : | MEAN_16x16_MMX | ||
| 285 : | |||
| 286 : | paddusw mm6, mm5 | ||
| 287 : | pmaddwd mm6, [mmx_one] ; collapse | ||
| 288 : | movq mm5, mm6 | ||
| 289 : | psrlq mm5, 32 | ||
| 290 : | paddd mm6, mm5 | ||
| 291 : | |||
| 292 : | psllq mm6, 32 ; blank upper dword | ||
| 293 : | psrlq mm6, 32 + 8 ; /= (16*16) | ||
| 294 : | |||
| 295 : | punpckldq mm6, mm6 | ||
| 296 : | packssdw mm6, mm6 | ||
| 297 : | |||
| 298 : | ; mm6 contains the mean | ||
| 299 : | ; mm5 is the new accum | ||
| 300 : | |||
| 301 : | pxor mm5, mm5 | ||
| 302 : | mov eax, [esp+ 4] ; Src | ||
| 303 : | |||
| 304 : | ABS_16x16_MMX | ||
| 305 : | ABS_16x16_MMX | ||
| 306 : | ABS_16x16_MMX | ||
| 307 : | ABS_16x16_MMX | ||
| 308 : | ABS_16x16_MMX | ||
| 309 : | ABS_16x16_MMX | ||
| 310 : | ABS_16x16_MMX | ||
| 311 : | ABS_16x16_MMX | ||
| 312 : | |||
| 313 : | ABS_16x16_MMX | ||
| 314 : | ABS_16x16_MMX | ||
| 315 : | ABS_16x16_MMX | ||
| 316 : | ABS_16x16_MMX | ||
| 317 : | ABS_16x16_MMX | ||
| 318 : | ABS_16x16_MMX | ||
| 319 : | ABS_16x16_MMX | ||
| 320 : | ABS_16x16_MMX | ||
| 321 : | |||
| 322 : | pmaddwd mm5, [mmx_one] ; collapse | ||
| 323 : | movq mm6, mm5 | ||
| 324 : | psrlq mm6, 32 | ||
| 325 : | paddd mm6, mm5 | ||
| 326 : | |||
| 327 : | movd eax, mm6 | ||
| 328 : | ret |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |