Parent Directory
| 
Revision Log
Revision 1877 - (view) (download)
| 1 : | edgomez | 1382 | ;/***************************************************************************** |
| 2 : | Isibaar | 262 | ; * |
| 3 : | edgomez | 1382 | ; * XVID MPEG-4 VIDEO CODEC |
| 4 : | ; * - mmx 8x8 block-based halfpel interpolation - | ||
| 5 : | Isibaar | 262 | ; * |
| 6 : | Isibaar | 1795 | ; * Copyright(C) 2002-2008 Michael Militzer <michael@xvid.org> |
| 7 : | edgomez | 1382 | ; * 2002 Pascal Massimino <skal@planet-d.net> |
| 8 : | Isibaar | 262 | ; * |
| 9 : | edgomez | 1382 | ; * This program is free software ; you can redistribute it and/or modify |
| 10 : | ; * it under the terms of the GNU General Public License as published by | ||
| 11 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
| 12 : | ; * (at your option) any later version. | ||
| 13 : | Isibaar | 262 | ; * |
| 14 : | edgomez | 1382 | ; * This program is distributed in the hope that it will be useful, |
| 15 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
| 16 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 17 : | ; * GNU General Public License for more details. | ||
| 18 : | Isibaar | 262 | ; * |
| 19 : | edgomez | 1382 | ; * You should have received a copy of the GNU General Public License |
| 20 : | ; * along with this program ; if not, write to the Free Software | ||
| 21 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 22 : | Isibaar | 262 | ; * |
| 23 : | edgomez | 1382 | ; ****************************************************************************/ |
| 24 : | Isibaar | 262 | |
| 25 : | Isibaar | 1795 | %include "nasm.inc" |
| 26 : | edgomez | 851 | |
| 27 : | edgomez | 1382 | ;============================================================================= |
| 28 : | ; Read only data | ||
| 29 : | ;============================================================================= | ||
| 30 : | Isibaar | 262 | |
| 31 : | Isibaar | 1795 | DATA |
| 32 : | Isibaar | 262 | |
| 33 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 34 : | edgomez | 1382 | mmx_one: |
| 35 : | times 8 db 1 | ||
| 36 : | Isibaar | 262 | |
| 37 : | Isibaar | 1844 | TEXT |
| 38 : | Isibaar | 262 | |
| 39 : | cglobal interpolate8x8_halfpel_h_xmm | ||
| 40 : | cglobal interpolate8x8_halfpel_v_xmm | ||
| 41 : | cglobal interpolate8x8_halfpel_hv_xmm | ||
| 42 : | |||
| 43 : | suxen_drol | 1632 | cglobal interpolate8x4_halfpel_h_xmm |
| 44 : | cglobal interpolate8x4_halfpel_v_xmm | ||
| 45 : | cglobal interpolate8x4_halfpel_hv_xmm | ||
| 46 : | |||
| 47 : | edgomez | 1530 | cglobal interpolate8x8_halfpel_add_xmm |
| 48 : | cglobal interpolate8x8_halfpel_h_add_xmm | ||
| 49 : | cglobal interpolate8x8_halfpel_v_add_xmm | ||
| 50 : | cglobal interpolate8x8_halfpel_hv_add_xmm | ||
| 51 : | |||
| 52 : | Isibaar | 262 | ;=========================================================================== |
| 53 : | ; | ||
| 54 : | ; void interpolate8x8_halfpel_h_xmm(uint8_t * const dst, | ||
| 55 : | ; const uint8_t * const src, | ||
| 56 : | ; const uint32_t stride, | ||
| 57 : | ; const uint32_t rounding); | ||
| 58 : | ; | ||
| 59 : | ;=========================================================================== | ||
| 60 : | |||
| 61 : | %macro COPY_H_SSE_RND0 0 | ||
| 62 : | Isibaar | 1795 | movq mm0, [_EAX] |
| 63 : | pavgb mm0, [_EAX+1] | ||
| 64 : | movq mm1, [_EAX+TMP1] | ||
| 65 : | pavgb mm1, [_EAX+TMP1+1] | ||
| 66 : | lea _EAX,[_EAX+2*TMP1] | ||
| 67 : | movq [TMP0],mm0 | ||
| 68 : | movq [TMP0+TMP1],mm1 | ||
| 69 : | Isibaar | 262 | %endmacro |
| 70 : | |||
| 71 : | %macro COPY_H_SSE_RND1 0 | ||
| 72 : | Isibaar | 1795 | movq mm0, [_EAX] |
| 73 : | movq mm1, [_EAX+TMP1] | ||
| 74 : | Isibaar | 262 | movq mm4, mm0 |
| 75 : | movq mm5, mm1 | ||
| 76 : | Isibaar | 1795 | movq mm2, [_EAX+1] |
| 77 : | movq mm3, [_EAX+TMP1+1] | ||
| 78 : | Isibaar | 262 | pavgb mm0, mm2 |
| 79 : | pxor mm2, mm4 | ||
| 80 : | pavgb mm1, mm3 | ||
| 81 : | Isibaar | 1795 | lea _EAX, [_EAX+2*TMP1] |
| 82 : | Isibaar | 262 | pxor mm3, mm5 |
| 83 : | pand mm2, mm7 | ||
| 84 : | pand mm3, mm7 | ||
| 85 : | psubb mm0, mm2 | ||
| 86 : | Isibaar | 1795 | movq [TMP0], mm0 |
| 87 : | Isibaar | 262 | psubb mm1, mm3 |
| 88 : | Isibaar | 1795 | movq [TMP0+TMP1], mm1 |
| 89 : | Isibaar | 262 | %endmacro |
| 90 : | |||
| 91 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 92 : | Isibaar | 262 | interpolate8x8_halfpel_h_xmm: |
| 93 : | |||
| 94 : | Isibaar | 1795 | mov _EAX, prm4 ; rounding |
| 95 : | mov TMP0, prm1 ; Dst | ||
| 96 : | test _EAX,_EAX | ||
| 97 : | mov _EAX, prm2 ; Src | ||
| 98 : | mov TMP1, prm3 ; stride | ||
| 99 : | Isibaar | 262 | |
| 100 : | jnz near .rounding1 | ||
| 101 : | |||
| 102 : | COPY_H_SSE_RND0 | ||
| 103 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 104 : | Isibaar | 262 | COPY_H_SSE_RND0 |
| 105 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 106 : | Isibaar | 262 | COPY_H_SSE_RND0 |
| 107 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 108 : | Isibaar | 262 | COPY_H_SSE_RND0 |
| 109 : | ret | ||
| 110 : | |||
| 111 : | Isibaar | 1793 | .rounding1: |
| 112 : | edgomez | 1382 | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 |
| 113 : | Isibaar | 262 | movq mm7, [mmx_one] |
| 114 : | COPY_H_SSE_RND1 | ||
| 115 : | Isibaar | 1795 | lea TMP0, [TMP0+2*TMP1] |
| 116 : | Isibaar | 262 | COPY_H_SSE_RND1 |
| 117 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 118 : | Isibaar | 262 | COPY_H_SSE_RND1 |
| 119 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 120 : | Isibaar | 262 | COPY_H_SSE_RND1 |
| 121 : | ret | ||
| 122 : | Isibaar | 1793 | ENDFUNC |
| 123 : | Isibaar | 262 | |
| 124 : | ;=========================================================================== | ||
| 125 : | ; | ||
| 126 : | ; void interpolate8x8_halfpel_v_xmm(uint8_t * const dst, | ||
| 127 : | edgomez | 1382 | ; const uint8_t * const src, |
| 128 : | ; const uint32_t stride, | ||
| 129 : | ; const uint32_t rounding); | ||
| 130 : | Isibaar | 262 | ; |
| 131 : | ;=========================================================================== | ||
| 132 : | |||
| 133 : | %macro COPY_V_SSE_RND0 0 | ||
| 134 : | Isibaar | 1795 | movq mm0, [_EAX] |
| 135 : | movq mm1, [_EAX+TMP1] | ||
| 136 : | Isibaar | 262 | pavgb mm0, mm1 |
| 137 : | Isibaar | 1795 | pavgb mm1, [_EAX+2*TMP1] |
| 138 : | lea _EAX, [_EAX+2*TMP1] | ||
| 139 : | movq [TMP0], mm0 | ||
| 140 : | movq [TMP0+TMP1],mm1 | ||
| 141 : | Isibaar | 262 | %endmacro |
| 142 : | |||
| 143 : | %macro COPY_V_SSE_RND1 0 | ||
| 144 : | movq mm0, mm2 | ||
| 145 : | Isibaar | 1795 | movq mm1, [_EAX] |
| 146 : | movq mm2, [_EAX+TMP1] | ||
| 147 : | lea _EAX,[_EAX+2*TMP1] | ||
| 148 : | Isibaar | 262 | movq mm4, mm0 |
| 149 : | movq mm5, mm1 | ||
| 150 : | pavgb mm0, mm1 | ||
| 151 : | edgomez | 1382 | pxor mm4, mm1 |
| 152 : | Isibaar | 262 | pavgb mm1, mm2 |
| 153 : | pxor mm5, mm2 | ||
| 154 : | edgomez | 1382 | pand mm4, mm7 ; lsb's of (i^j)... |
| 155 : | pand mm5, mm7 ; lsb's of (i^j)... | ||
| 156 : | psubb mm0, mm4 ; ...are substracted from result of pavgb | ||
| 157 : | Isibaar | 1795 | movq [TMP0], mm0 |
| 158 : | edgomez | 1382 | psubb mm1, mm5 ; ...are substracted from result of pavgb |
| 159 : | Isibaar | 1795 | movq [TMP0+TMP1], mm1 |
| 160 : | Isibaar | 262 | %endmacro |
| 161 : | |||
| 162 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 163 : | Isibaar | 262 | interpolate8x8_halfpel_v_xmm: |
| 164 : | |||
| 165 : | Isibaar | 1795 | mov _EAX, prm4 ; rounding |
| 166 : | mov TMP0, prm1 ; Dst | ||
| 167 : | test _EAX,_EAX | ||
| 168 : | mov _EAX, prm2 ; Src | ||
| 169 : | mov TMP1, prm3 ; stride | ||
| 170 : | Isibaar | 262 | |
| 171 : | edgomez | 1382 | ; we process 2 line at a time |
| 172 : | Isibaar | 262 | jnz near .rounding1 |
| 173 : | |||
| 174 : | COPY_V_SSE_RND0 | ||
| 175 : | Isibaar | 1795 | lea TMP0, [TMP0+2*TMP1] |
| 176 : | Isibaar | 262 | COPY_V_SSE_RND0 |
| 177 : | Isibaar | 1795 | lea TMP0, [TMP0+2*TMP1] |
| 178 : | Isibaar | 262 | COPY_V_SSE_RND0 |
| 179 : | Isibaar | 1795 | lea TMP0, [TMP0+2*TMP1] |
| 180 : | Isibaar | 262 | COPY_V_SSE_RND0 |
| 181 : | ret | ||
| 182 : | |||
| 183 : | Isibaar | 1793 | .rounding1: |
| 184 : | edgomez | 1382 | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 |
| 185 : | Isibaar | 262 | movq mm7, [mmx_one] |
| 186 : | Isibaar | 1795 | movq mm2, [_EAX] ; loop invariant |
| 187 : | add _EAX, TMP1 | ||
| 188 : | Isibaar | 262 | |
| 189 : | COPY_V_SSE_RND1 | ||
| 190 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 191 : | Isibaar | 262 | COPY_V_SSE_RND1 |
| 192 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 193 : | Isibaar | 262 | COPY_V_SSE_RND1 |
| 194 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 195 : | Isibaar | 262 | COPY_V_SSE_RND1 |
| 196 : | ret | ||
| 197 : | Isibaar | 1793 | ENDFUNC |
| 198 : | Isibaar | 262 | |
| 199 : | ;=========================================================================== | ||
| 200 : | ; | ||
| 201 : | ; void interpolate8x8_halfpel_hv_xmm(uint8_t * const dst, | ||
| 202 : | edgomez | 1382 | ; const uint8_t * const src, |
| 203 : | ; const uint32_t stride, | ||
| 204 : | ; const uint32_t rounding); | ||
| 205 : | Isibaar | 262 | ; |
| 206 : | ; | ||
| 207 : | ;=========================================================================== | ||
| 208 : | |||
| 209 : | ; The trick is to correct the result of 'pavgb' with some combination of the | ||
| 210 : | ; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgb' (s and t). | ||
| 211 : | ; The boolean relations are: | ||
| 212 : | edgomez | 1382 | ; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st |
| 213 : | Isibaar | 262 | ; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st |
| 214 : | ; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st | ||
| 215 : | ; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st | ||
| 216 : | ; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. | ||
| 217 : | |||
| 218 : | ; Moreover, we process 2 lines at a times, for better overlapping (~15% faster). | ||
| 219 : | |||
| 220 : | %macro COPY_HV_SSE_RND0 0 | ||
| 221 : | Isibaar | 1795 | lea _EAX, [_EAX+TMP1] |
| 222 : | Isibaar | 262 | |
| 223 : | Isibaar | 1795 | movq mm0, [_EAX] |
| 224 : | movq mm1, [_EAX+1] | ||
| 225 : | Isibaar | 262 | |
| 226 : | edgomez | 1382 | movq mm6, mm0 |
| 227 : | pavgb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step | ||
| 228 : | Isibaar | 1795 | lea _EAX, [_EAX+TMP1] |
| 229 : | edgomez | 1382 | pxor mm1, mm6 ; mm1=(j^k). preserved for next step |
| 230 : | Isibaar | 262 | |
| 231 : | edgomez | 1382 | por mm3, mm1 ; ij |= jk |
| 232 : | movq mm6, mm2 | ||
| 233 : | pxor mm6, mm0 ; mm6 = s^t | ||
| 234 : | pand mm3, mm6 ; (ij|jk) &= st | ||
| 235 : | pavgb mm2, mm0 ; mm2 = (s+t+1)/2 | ||
| 236 : | pand mm3, mm7 ; mask lsb | ||
| 237 : | psubb mm2, mm3 ; apply. | ||
| 238 : | Isibaar | 262 | |
| 239 : | Isibaar | 1795 | movq [TMP0], mm2 |
| 240 : | Isibaar | 262 | |
| 241 : | Isibaar | 1795 | movq mm2, [_EAX] |
| 242 : | movq mm3, [_EAX+1] | ||
| 243 : | edgomez | 1382 | movq mm6, mm2 |
| 244 : | pavgb mm2, mm3 ; preserved for next iteration | ||
| 245 : | Isibaar | 1795 | lea TMP0,[TMP0+TMP1] |
| 246 : | edgomez | 1382 | pxor mm3, mm6 ; preserved for next iteration |
| 247 : | Isibaar | 262 | |
| 248 : | edgomez | 1382 | por mm1, mm3 |
| 249 : | movq mm6, mm0 | ||
| 250 : | pxor mm6, mm2 | ||
| 251 : | pand mm1, mm6 | ||
| 252 : | pavgb mm0, mm2 | ||
| 253 : | Isibaar | 262 | |
| 254 : | edgomez | 1382 | pand mm1, mm7 |
| 255 : | psubb mm0, mm1 | ||
| 256 : | Isibaar | 262 | |
| 257 : | Isibaar | 1795 | movq [TMP0], mm0 |
| 258 : | Isibaar | 262 | %endmacro |
| 259 : | |||
| 260 : | %macro COPY_HV_SSE_RND1 0 | ||
| 261 : | Isibaar | 1795 | lea _EAX, [_EAX+TMP1] |
| 262 : | Isibaar | 262 | |
| 263 : | Isibaar | 1795 | movq mm0, [_EAX] |
| 264 : | movq mm1, [_EAX+1] | ||
| 265 : | Isibaar | 262 | |
| 266 : | edgomez | 1382 | movq mm6, mm0 |
| 267 : | pavgb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step | ||
| 268 : | Isibaar | 1795 | lea _EAX, [_EAX+TMP1] |
| 269 : | edgomez | 1382 | pxor mm1, mm6 ; mm1=(j^k). preserved for next step |
| 270 : | Isibaar | 262 | |
| 271 : | edgomez | 1382 | pand mm3, mm1 |
| 272 : | movq mm6, mm2 | ||
| 273 : | pxor mm6, mm0 | ||
| 274 : | por mm3, mm6 | ||
| 275 : | pavgb mm2, mm0 | ||
| 276 : | pand mm3, mm7 | ||
| 277 : | psubb mm2, mm3 | ||
| 278 : | Isibaar | 262 | |
| 279 : | Isibaar | 1795 | movq [TMP0], mm2 |
| 280 : | Isibaar | 262 | |
| 281 : | Isibaar | 1795 | movq mm2, [_EAX] |
| 282 : | movq mm3, [_EAX+1] | ||
| 283 : | edgomez | 1382 | movq mm6, mm2 |
| 284 : | pavgb mm2, mm3 ; preserved for next iteration | ||
| 285 : | Isibaar | 1795 | lea TMP0,[TMP0+TMP1] |
| 286 : | edgomez | 1382 | pxor mm3, mm6 ; preserved for next iteration |
| 287 : | Isibaar | 262 | |
| 288 : | edgomez | 1382 | pand mm1, mm3 |
| 289 : | movq mm6, mm0 | ||
| 290 : | pxor mm6, mm2 | ||
| 291 : | por mm1, mm6 | ||
| 292 : | pavgb mm0, mm2 | ||
| 293 : | pand mm1, mm7 | ||
| 294 : | psubb mm0, mm1 | ||
| 295 : | Isibaar | 262 | |
| 296 : | Isibaar | 1795 | movq [TMP0], mm0 |
| 297 : | Isibaar | 262 | %endmacro |
| 298 : | |||
| 299 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 300 : | Isibaar | 262 | interpolate8x8_halfpel_hv_xmm: |
| 301 : | Isibaar | 1795 | mov _EAX, prm4 ; rounding |
| 302 : | mov TMP0, prm1 ; Dst | ||
| 303 : | test _EAX, _EAX | ||
| 304 : | mov _EAX, prm2 ; Src | ||
| 305 : | mov TMP1, prm3 ; stride | ||
| 306 : | Isibaar | 262 | |
| 307 : | movq mm7, [mmx_one] | ||
| 308 : | |||
| 309 : | ; loop invariants: mm2=(i+j+1)/2 and mm3= i^j | ||
| 310 : | Isibaar | 1795 | movq mm2, [_EAX] |
| 311 : | movq mm3, [_EAX+1] | ||
| 312 : | Isibaar | 262 | movq mm6, mm2 |
| 313 : | pavgb mm2, mm3 | ||
| 314 : | edgomez | 1382 | pxor mm3, mm6 ; mm2/mm3 ready |
| 315 : | Isibaar | 262 | |
| 316 : | jnz near .rounding1 | ||
| 317 : | |||
| 318 : | COPY_HV_SSE_RND0 | ||
| 319 : | Isibaar | 1795 | add TMP0, TMP1 |
| 320 : | Isibaar | 262 | COPY_HV_SSE_RND0 |
| 321 : | Isibaar | 1795 | add TMP0, TMP1 |
| 322 : | Isibaar | 262 | COPY_HV_SSE_RND0 |
| 323 : | Isibaar | 1795 | add TMP0, TMP1 |
| 324 : | Isibaar | 262 | COPY_HV_SSE_RND0 |
| 325 : | ret | ||
| 326 : | |||
| 327 : | Isibaar | 1793 | .rounding1: |
| 328 : | Isibaar | 262 | COPY_HV_SSE_RND1 |
| 329 : | Isibaar | 1795 | add TMP0, TMP1 |
| 330 : | Isibaar | 262 | COPY_HV_SSE_RND1 |
| 331 : | Isibaar | 1795 | add TMP0, TMP1 |
| 332 : | Isibaar | 262 | COPY_HV_SSE_RND1 |
| 333 : | Isibaar | 1795 | add TMP0, TMP1 |
| 334 : | Isibaar | 262 | COPY_HV_SSE_RND1 |
| 335 : | edgomez | 1382 | ret |
| 336 : | Isibaar | 1793 | ENDFUNC |
| 337 : | edgomez | 1530 | |
| 338 : | ;=========================================================================== | ||
| 339 : | ; | ||
| 340 : | suxen_drol | 1632 | ; void interpolate8x4_halfpel_h_xmm(uint8_t * const dst, |
| 341 : | ; const uint8_t * const src, | ||
| 342 : | ; const uint32_t stride, | ||
| 343 : | ; const uint32_t rounding); | ||
| 344 : | ; | ||
| 345 : | ;=========================================================================== | ||
| 346 : | |||
| 347 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 348 : | suxen_drol | 1632 | interpolate8x4_halfpel_h_xmm: |
| 349 : | |||
| 350 : | Isibaar | 1795 | mov _EAX, prm4 ; rounding |
| 351 : | mov TMP0, prm1 ; Dst | ||
| 352 : | test _EAX,_EAX | ||
| 353 : | mov _EAX, prm2 ; Src | ||
| 354 : | mov TMP1, prm3 ; stride | ||
| 355 : | suxen_drol | 1632 | |
| 356 : | jnz near .rounding1 | ||
| 357 : | |||
| 358 : | COPY_H_SSE_RND0 | ||
| 359 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 360 : | suxen_drol | 1632 | COPY_H_SSE_RND0 |
| 361 : | ret | ||
| 362 : | |||
| 363 : | Isibaar | 1793 | .rounding1: |
| 364 : | suxen_drol | 1632 | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 |
| 365 : | movq mm7, [mmx_one] | ||
| 366 : | COPY_H_SSE_RND1 | ||
| 367 : | Isibaar | 1795 | lea TMP0, [TMP0+2*TMP1] |
| 368 : | suxen_drol | 1632 | COPY_H_SSE_RND1 |
| 369 : | ret | ||
| 370 : | Isibaar | 1793 | ENDFUNC |
| 371 : | suxen_drol | 1632 | |
| 372 : | ;=========================================================================== | ||
| 373 : | ; | ||
| 374 : | ; void interpolate8x4_halfpel_v_xmm(uint8_t * const dst, | ||
| 375 : | ; const uint8_t * const src, | ||
| 376 : | ; const uint32_t stride, | ||
| 377 : | ; const uint32_t rounding); | ||
| 378 : | ; | ||
| 379 : | ;=========================================================================== | ||
| 380 : | |||
| 381 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 382 : | suxen_drol | 1632 | interpolate8x4_halfpel_v_xmm: |
| 383 : | |||
| 384 : | Isibaar | 1795 | mov _EAX, prm4 ; rounding |
| 385 : | mov TMP0, prm1 ; Dst | ||
| 386 : | test _EAX,_EAX | ||
| 387 : | mov _EAX, prm2 ; Src | ||
| 388 : | mov TMP1, prm3 ; stride | ||
| 389 : | suxen_drol | 1632 | |
| 390 : | ; we process 2 line at a time | ||
| 391 : | jnz near .rounding1 | ||
| 392 : | |||
| 393 : | COPY_V_SSE_RND0 | ||
| 394 : | Isibaar | 1795 | lea TMP0, [TMP0+2*TMP1] |
| 395 : | suxen_drol | 1632 | COPY_V_SSE_RND0 |
| 396 : | ret | ||
| 397 : | |||
| 398 : | Isibaar | 1793 | .rounding1: |
| 399 : | suxen_drol | 1632 | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 |
| 400 : | movq mm7, [mmx_one] | ||
| 401 : | Isibaar | 1795 | movq mm2, [_EAX] ; loop invariant |
| 402 : | add _EAX, TMP1 | ||
| 403 : | suxen_drol | 1632 | |
| 404 : | COPY_V_SSE_RND1 | ||
| 405 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 406 : | suxen_drol | 1632 | COPY_V_SSE_RND1 |
| 407 : | ret | ||
| 408 : | Isibaar | 1793 | ENDFUNC |
| 409 : | suxen_drol | 1632 | |
| 410 : | ;=========================================================================== | ||
| 411 : | ; | ||
| 412 : | ; void interpolate8x4_halfpel_hv_xmm(uint8_t * const dst, | ||
| 413 : | ; const uint8_t * const src, | ||
| 414 : | ; const uint32_t stride, | ||
| 415 : | ; const uint32_t rounding); | ||
| 416 : | ; | ||
| 417 : | ; | ||
| 418 : | ;=========================================================================== | ||
| 419 : | |||
| 420 : | ; The trick is to correct the result of 'pavgb' with some combination of the | ||
| 421 : | ; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgb' (s and t). | ||
| 422 : | ; The boolean relations are: | ||
| 423 : | ; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st | ||
| 424 : | ; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st | ||
| 425 : | ; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st | ||
| 426 : | ; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st | ||
| 427 : | ; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. | ||
| 428 : | |||
| 429 : | ; Moreover, we process 2 lines at a times, for better overlapping (~15% faster). | ||
| 430 : | |||
| 431 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 432 : | suxen_drol | 1632 | interpolate8x4_halfpel_hv_xmm: |
| 433 : | Isibaar | 1795 | mov _EAX, prm4 ; rounding |
| 434 : | mov TMP0, prm1 ; Dst | ||
| 435 : | test _EAX, _EAX | ||
| 436 : | mov _EAX, prm2 ; Src | ||
| 437 : | mov TMP1, prm3 ; stride | ||
| 438 : | suxen_drol | 1632 | |
| 439 : | movq mm7, [mmx_one] | ||
| 440 : | |||
| 441 : | ; loop invariants: mm2=(i+j+1)/2 and mm3= i^j | ||
| 442 : | Isibaar | 1795 | movq mm2, [_EAX] |
| 443 : | movq mm3, [_EAX+1] | ||
| 444 : | suxen_drol | 1632 | movq mm6, mm2 |
| 445 : | pavgb mm2, mm3 | ||
| 446 : | pxor mm3, mm6 ; mm2/mm3 ready | ||
| 447 : | |||
| 448 : | jnz near .rounding1 | ||
| 449 : | |||
| 450 : | COPY_HV_SSE_RND0 | ||
| 451 : | Isibaar | 1795 | add TMP0, TMP1 |
| 452 : | suxen_drol | 1632 | COPY_HV_SSE_RND0 |
| 453 : | ret | ||
| 454 : | |||
| 455 : | Isibaar | 1793 | .rounding1: |
| 456 : | suxen_drol | 1632 | COPY_HV_SSE_RND1 |
| 457 : | Isibaar | 1795 | add TMP0, TMP1 |
| 458 : | suxen_drol | 1632 | COPY_HV_SSE_RND1 |
| 459 : | ret | ||
| 460 : | Isibaar | 1793 | ENDFUNC |
| 461 : | suxen_drol | 1632 | |
| 462 : | ;=========================================================================== | ||
| 463 : | ; | ||
| 464 : | edgomez | 1530 | ; The next functions combine both source halfpel interpolation step and the |
| 465 : | ; averaging (with rouding) step to avoid wasting memory bandwidth computing | ||
| 466 : | ; intermediate halfpel images and then averaging them. | ||
| 467 : | ; | ||
| 468 : | ;=========================================================================== | ||
| 469 : | |||
| 470 : | %macro PROLOG0 0 | ||
| 471 : | Isibaar | 1795 | mov TMP0, prm1 ; Dst |
| 472 : | mov _EAX, prm2 ; Src | ||
| 473 : | mov TMP1, prm3 ; BpS | ||
| 474 : | edgomez | 1530 | %endmacro |
| 475 : | %macro PROLOG1 0 | ||
| 476 : | PROLOG0 | ||
| 477 : | Isibaar | 1795 | test prm4d, 1; Rounding? |
| 478 : | edgomez | 1530 | %endmacro |
| 479 : | %macro EPILOG 0 | ||
| 480 : | ret | ||
| 481 : | %endmacro | ||
| 482 : | |||
| 483 : | ;=========================================================================== | ||
| 484 : | ; | ||
| 485 : | ; void interpolate8x8_halfpel_add_xmm(uint8_t * const dst, | ||
| 486 : | ; const uint8_t * const src, | ||
| 487 : | ; const uint32_t stride, | ||
| 488 : | ; const uint32_t rounding); | ||
| 489 : | ; | ||
| 490 : | ; | ||
| 491 : | ;=========================================================================== | ||
| 492 : | |||
| 493 : | %macro ADD_FF 2 | ||
| 494 : | Isibaar | 1795 | movq mm0, [_EAX+%1] |
| 495 : | movq mm1, [_EAX+%2] | ||
| 496 : | edgomez | 1530 | ;;--- |
| 497 : | ;; movq mm2, mm0 | ||
| 498 : | ;; movq mm3, mm1 | ||
| 499 : | ;;--- | ||
| 500 : | Isibaar | 1795 | pavgb mm0, [TMP0+%1] |
| 501 : | pavgb mm1, [TMP0+%2] | ||
| 502 : | edgomez | 1530 | ;;-- |
| 503 : | Isibaar | 1795 | ;; por mm2, [TMP0+%1] |
| 504 : | ;; por mm3, [TMP0+%2] | ||
| 505 : | edgomez | 1530 | ;; pand mm2, [mmx_one] |
| 506 : | ;; pand mm3, [mmx_one] | ||
| 507 : | ;; psubsb mm0, mm2 | ||
| 508 : | ;; psubsb mm1, mm3 | ||
| 509 : | ;;-- | ||
| 510 : | Isibaar | 1795 | movq [TMP0+%1], mm0 |
| 511 : | movq [TMP0+%2], mm1 | ||
| 512 : | edgomez | 1530 | %endmacro |
| 513 : | |||
| 514 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 515 : | edgomez | 1530 | interpolate8x8_halfpel_add_xmm: ; 23c |
| 516 : | PROLOG1 | ||
| 517 : | Isibaar | 1795 | ADD_FF 0, TMP1 |
| 518 : | lea _EAX,[_EAX+2*TMP1] | ||
| 519 : | lea TMP0,[TMP0+2*TMP1] | ||
| 520 : | ADD_FF 0, TMP1 | ||
| 521 : | lea _EAX,[_EAX+2*TMP1] | ||
| 522 : | lea TMP0,[TMP0+2*TMP1] | ||
| 523 : | ADD_FF 0, TMP1 | ||
| 524 : | lea _EAX,[_EAX+2*TMP1] | ||
| 525 : | lea TMP0,[TMP0+2*TMP1] | ||
| 526 : | ADD_FF 0, TMP1 | ||
| 527 : | edgomez | 1530 | EPILOG |
| 528 : | Isibaar | 1793 | ENDFUNC |
| 529 : | edgomez | 1530 | |
| 530 : | ;=========================================================================== | ||
| 531 : | ; | ||
| 532 : | ; void interpolate8x8_halfpel_h_add_xmm(uint8_t * const dst, | ||
| 533 : | ; const uint8_t * const src, | ||
| 534 : | ; const uint32_t stride, | ||
| 535 : | ; const uint32_t rounding); | ||
| 536 : | ; | ||
| 537 : | ; | ||
| 538 : | ;=========================================================================== | ||
| 539 : | |||
| 540 : | |||
| 541 : | %macro ADD_FH_RND0 2 | ||
| 542 : | Isibaar | 1795 | movq mm0, [_EAX+%1] |
| 543 : | movq mm1, [_EAX+%2] | ||
| 544 : | pavgb mm0, [_EAX+%1+1] | ||
| 545 : | pavgb mm1, [_EAX+%2+1] | ||
| 546 : | pavgb mm0, [TMP0+%1] | ||
| 547 : | pavgb mm1, [TMP0+%2] | ||
| 548 : | movq [TMP0+%1],mm0 | ||
| 549 : | movq [TMP0+%2],mm1 | ||
| 550 : | edgomez | 1530 | %endmacro |
| 551 : | |||
| 552 : | %macro ADD_FH_RND1 2 | ||
| 553 : | Isibaar | 1795 | movq mm0, [_EAX+%1] |
| 554 : | movq mm1, [_EAX+%2] | ||
| 555 : | edgomez | 1530 | movq mm4, mm0 |
| 556 : | movq mm5, mm1 | ||
| 557 : | Isibaar | 1795 | movq mm2, [_EAX+%1+1] |
| 558 : | movq mm3, [_EAX+%2+1] | ||
| 559 : | edgomez | 1530 | pavgb mm0, mm2 |
| 560 : | ; lea ?? | ||
| 561 : | pxor mm2, mm4 | ||
| 562 : | pavgb mm1, mm3 | ||
| 563 : | pxor mm3, mm5 | ||
| 564 : | pand mm2, [mmx_one] | ||
| 565 : | pand mm3, [mmx_one] | ||
| 566 : | psubb mm0, mm2 | ||
| 567 : | psubb mm1, mm3 | ||
| 568 : | Isibaar | 1795 | pavgb mm0, [TMP0+%1] |
| 569 : | pavgb mm1, [TMP0+%2] | ||
| 570 : | movq [TMP0+%1],mm0 | ||
| 571 : | movq [TMP0+%2],mm1 | ||
| 572 : | edgomez | 1530 | %endmacro |
| 573 : | |||
| 574 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 575 : | edgomez | 1530 | interpolate8x8_halfpel_h_add_xmm: ; 32c |
| 576 : | PROLOG1 | ||
| 577 : | jnz near .Loop1 | ||
| 578 : | Isibaar | 1795 | ADD_FH_RND0 0, TMP1 |
| 579 : | lea _EAX,[_EAX+2*TMP1] | ||
| 580 : | lea TMP0,[TMP0+2*TMP1] | ||
| 581 : | ADD_FH_RND0 0, TMP1 | ||
| 582 : | lea _EAX,[_EAX+2*TMP1] | ||
| 583 : | lea TMP0,[TMP0+2*TMP1] | ||
| 584 : | ADD_FH_RND0 0, TMP1 | ||
| 585 : | lea _EAX,[_EAX+2*TMP1] | ||
| 586 : | lea TMP0,[TMP0+2*TMP1] | ||
| 587 : | ADD_FH_RND0 0, TMP1 | ||
| 588 : | edgomez | 1530 | EPILOG |
| 589 : | |||
| 590 : | Isibaar | 1793 | .Loop1: |
| 591 : | edgomez | 1530 | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 |
| 592 : | ; movq mm7, [mmx_one] | ||
| 593 : | Isibaar | 1795 | ADD_FH_RND1 0, TMP1 |
| 594 : | lea _EAX,[_EAX+2*TMP1] | ||
| 595 : | lea TMP0,[TMP0+2*TMP1] | ||
| 596 : | ADD_FH_RND1 0, TMP1 | ||
| 597 : | lea _EAX,[_EAX+2*TMP1] | ||
| 598 : | lea TMP0,[TMP0+2*TMP1] | ||
| 599 : | ADD_FH_RND1 0, TMP1 | ||
| 600 : | lea _EAX,[_EAX+2*TMP1] | ||
| 601 : | lea TMP0,[TMP0+2*TMP1] | ||
| 602 : | ADD_FH_RND1 0, TMP1 | ||
| 603 : | edgomez | 1530 | EPILOG |
| 604 : | Isibaar | 1793 | ENDFUNC |
| 605 : | edgomez | 1530 | |
| 606 : | |||
| 607 : | ;=========================================================================== | ||
| 608 : | ; | ||
| 609 : | ; void interpolate8x8_halfpel_v_add_xmm(uint8_t * const dst, | ||
| 610 : | ; const uint8_t * const src, | ||
| 611 : | ; const uint32_t stride, | ||
| 612 : | ; const uint32_t rounding); | ||
| 613 : | ; | ||
| 614 : | ; | ||
| 615 : | ;=========================================================================== | ||
| 616 : | |||
| 617 : | %macro ADD_8_HF_RND0 0 | ||
| 618 : | Isibaar | 1795 | movq mm0, [_EAX] |
| 619 : | movq mm1, [_EAX+TMP1] | ||
| 620 : | edgomez | 1530 | pavgb mm0, mm1 |
| 621 : | Isibaar | 1795 | pavgb mm1, [_EAX+2*TMP1] |
| 622 : | lea _EAX,[_EAX+2*TMP1] | ||
| 623 : | pavgb mm0, [TMP0] | ||
| 624 : | pavgb mm1, [TMP0+TMP1] | ||
| 625 : | movq [TMP0],mm0 | ||
| 626 : | movq [TMP0+TMP1],mm1 | ||
| 627 : | edgomez | 1530 | %endmacro |
| 628 : | |||
| 629 : | %macro ADD_8_HF_RND1 0 | ||
| 630 : | Isibaar | 1795 | movq mm1, [_EAX+TMP1] |
| 631 : | movq mm2, [_EAX+2*TMP1] | ||
| 632 : | lea _EAX,[_EAX+2*TMP1] | ||
| 633 : | edgomez | 1530 | movq mm4, mm0 |
| 634 : | movq mm5, mm1 | ||
| 635 : | pavgb mm0, mm1 | ||
| 636 : | pxor mm4, mm1 | ||
| 637 : | pavgb mm1, mm2 | ||
| 638 : | pxor mm5, mm2 | ||
| 639 : | pand mm4, mm7 ; lsb's of (i^j)... | ||
| 640 : | pand mm5, mm7 ; lsb's of (i^j)... | ||
| 641 : | psubb mm0, mm4 ; ...are substracted from result of pavgb | ||
| 642 : | Isibaar | 1795 | pavgb mm0, [TMP0] |
| 643 : | movq [TMP0], mm0 | ||
| 644 : | edgomez | 1530 | psubb mm1, mm5 ; ...are substracted from result of pavgb |
| 645 : | Isibaar | 1795 | pavgb mm1, [TMP0+TMP1] |
| 646 : | movq [TMP0+TMP1], mm1 | ||
| 647 : | edgomez | 1530 | %endmacro |
| 648 : | |||
| 649 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 650 : | edgomez | 1530 | interpolate8x8_halfpel_v_add_xmm: |
| 651 : | PROLOG1 | ||
| 652 : | |||
| 653 : | jnz near .Loop1 | ||
| 654 : | pxor mm7, mm7 ; this is a NOP | ||
| 655 : | |||
| 656 : | ADD_8_HF_RND0 | ||
| 657 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 658 : | edgomez | 1530 | ADD_8_HF_RND0 |
| 659 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 660 : | edgomez | 1530 | ADD_8_HF_RND0 |
| 661 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 662 : | edgomez | 1530 | ADD_8_HF_RND0 |
| 663 : | EPILOG | ||
| 664 : | |||
| 665 : | Isibaar | 1793 | .Loop1: |
| 666 : | Isibaar | 1795 | movq mm0, [_EAX] ; loop invariant |
| 667 : | edgomez | 1530 | movq mm7, [mmx_one] |
| 668 : | |||
| 669 : | ADD_8_HF_RND1 | ||
| 670 : | movq mm0, mm2 | ||
| 671 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 672 : | edgomez | 1530 | ADD_8_HF_RND1 |
| 673 : | movq mm0, mm2 | ||
| 674 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 675 : | edgomez | 1530 | ADD_8_HF_RND1 |
| 676 : | movq mm0, mm2 | ||
| 677 : | Isibaar | 1795 | lea TMP0,[TMP0+2*TMP1] |
| 678 : | edgomez | 1530 | ADD_8_HF_RND1 |
| 679 : | EPILOG | ||
| 680 : | Isibaar | 1793 | ENDFUNC |
| 681 : | edgomez | 1530 | |
| 682 : | ; The trick is to correct the result of 'pavgb' with some combination of the | ||
| 683 : | ; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgb' (s and t). | ||
| 684 : | ; The boolean relations are: | ||
| 685 : | ; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st | ||
| 686 : | ; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st | ||
| 687 : | ; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st | ||
| 688 : | ; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st | ||
| 689 : | ; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. | ||
| 690 : | |||
| 691 : | ; Moreover, we process 2 lines at a times, for better overlapping (~15% faster). | ||
| 692 : | |||
| 693 : | ;=========================================================================== | ||
| 694 : | ; | ||
| 695 : | ; void interpolate8x8_halfpel_hv_add_xmm(uint8_t * const dst, | ||
| 696 : | ; const uint8_t * const src, | ||
| 697 : | ; const uint32_t stride, | ||
| 698 : | ; const uint32_t rounding); | ||
| 699 : | ; | ||
| 700 : | ; | ||
| 701 : | ;=========================================================================== | ||
| 702 : | |||
| 703 : | %macro ADD_HH_RND0 0 | ||
| 704 : | Isibaar | 1795 | lea _EAX,[_EAX+TMP1] |
| 705 : | edgomez | 1530 | |
| 706 : | Isibaar | 1795 | movq mm0, [_EAX] |
| 707 : | movq mm1, [_EAX+1] | ||
| 708 : | edgomez | 1530 | |
| 709 : | movq mm6, mm0 | ||
| 710 : | pavgb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step | ||
| 711 : | Isibaar | 1795 | lea _EAX,[_EAX+TMP1] |
| 712 : | edgomez | 1530 | pxor mm1, mm6 ; mm1=(j^k). preserved for next step |
| 713 : | |||
| 714 : | por mm3, mm1 ; ij |= jk | ||
| 715 : | movq mm6, mm2 | ||
| 716 : | pxor mm6, mm0 ; mm6 = s^t | ||
| 717 : | pand mm3, mm6 ; (ij|jk) &= st | ||
| 718 : | pavgb mm2, mm0 ; mm2 = (s+t+1)/2 | ||
| 719 : | pand mm3, mm7 ; mask lsb | ||
| 720 : | psubb mm2, mm3 ; apply. | ||
| 721 : | |||
| 722 : | Isibaar | 1795 | pavgb mm2, [TMP0] |
| 723 : | movq [TMP0], mm2 | ||
| 724 : | edgomez | 1530 | |
| 725 : | Isibaar | 1795 | movq mm2, [_EAX] |
| 726 : | movq mm3, [_EAX+1] | ||
| 727 : | edgomez | 1530 | movq mm6, mm2 |
| 728 : | pavgb mm2, mm3 ; preserved for next iteration | ||
| 729 : | Isibaar | 1795 | lea TMP0,[TMP0+TMP1] |
| 730 : | edgomez | 1530 | pxor mm3, mm6 ; preserved for next iteration |
| 731 : | |||
| 732 : | por mm1, mm3 | ||
| 733 : | movq mm6, mm0 | ||
| 734 : | pxor mm6, mm2 | ||
| 735 : | pand mm1, mm6 | ||
| 736 : | pavgb mm0, mm2 | ||
| 737 : | |||
| 738 : | pand mm1, mm7 | ||
| 739 : | psubb mm0, mm1 | ||
| 740 : | |||
| 741 : | Isibaar | 1795 | pavgb mm0, [TMP0] |
| 742 : | movq [TMP0], mm0 | ||
| 743 : | edgomez | 1530 | %endmacro |
| 744 : | |||
| 745 : | %macro ADD_HH_RND1 0 | ||
| 746 : | Isibaar | 1795 | lea _EAX,[_EAX+TMP1] |
| 747 : | edgomez | 1530 | |
| 748 : | Isibaar | 1795 | movq mm0, [_EAX] |
| 749 : | movq mm1, [_EAX+1] | ||
| 750 : | edgomez | 1530 | |
| 751 : | movq mm6, mm0 | ||
| 752 : | pavgb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step | ||
| 753 : | Isibaar | 1795 | lea _EAX,[_EAX+TMP1] |
| 754 : | edgomez | 1530 | pxor mm1, mm6 ; mm1=(j^k). preserved for next step |
| 755 : | |||
| 756 : | pand mm3, mm1 | ||
| 757 : | movq mm6, mm2 | ||
| 758 : | pxor mm6, mm0 | ||
| 759 : | por mm3, mm6 | ||
| 760 : | pavgb mm2, mm0 | ||
| 761 : | pand mm3, mm7 | ||
| 762 : | psubb mm2, mm3 | ||
| 763 : | |||
| 764 : | Isibaar | 1795 | pavgb mm2, [TMP0] |
| 765 : | movq [TMP0], mm2 | ||
| 766 : | edgomez | 1530 | |
| 767 : | Isibaar | 1795 | movq mm2, [_EAX] |
| 768 : | movq mm3, [_EAX+1] | ||
| 769 : | edgomez | 1530 | movq mm6, mm2 |
| 770 : | pavgb mm2, mm3 ; preserved for next iteration | ||
| 771 : | Isibaar | 1795 | lea TMP0,[TMP0+TMP1] |
| 772 : | edgomez | 1530 | pxor mm3, mm6 ; preserved for next iteration |
| 773 : | |||
| 774 : | pand mm1, mm3 | ||
| 775 : | movq mm6, mm0 | ||
| 776 : | pxor mm6, mm2 | ||
| 777 : | por mm1, mm6 | ||
| 778 : | pavgb mm0, mm2 | ||
| 779 : | pand mm1, mm7 | ||
| 780 : | psubb mm0, mm1 | ||
| 781 : | |||
| 782 : | Isibaar | 1795 | pavgb mm0, [TMP0] |
| 783 : | movq [TMP0], mm0 | ||
| 784 : | edgomez | 1530 | %endmacro |
| 785 : | |||
| 786 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 787 : | edgomez | 1530 | interpolate8x8_halfpel_hv_add_xmm: |
| 788 : | PROLOG1 | ||
| 789 : | |||
| 790 : | movq mm7, [mmx_one] | ||
| 791 : | |||
| 792 : | ; loop invariants: mm2=(i+j+1)/2 and mm3= i^j | ||
| 793 : | Isibaar | 1795 | movq mm2, [_EAX] |
| 794 : | movq mm3, [_EAX+1] | ||
| 795 : | edgomez | 1530 | movq mm6, mm2 |
| 796 : | pavgb mm2, mm3 | ||
| 797 : | pxor mm3, mm6 ; mm2/mm3 ready | ||
| 798 : | |||
| 799 : | jnz near .Loop1 | ||
| 800 : | |||
| 801 : | ADD_HH_RND0 | ||
| 802 : | Isibaar | 1795 | add TMP0, TMP1 |
| 803 : | edgomez | 1530 | ADD_HH_RND0 |
| 804 : | Isibaar | 1795 | add TMP0, TMP1 |
| 805 : | edgomez | 1530 | ADD_HH_RND0 |
| 806 : | Isibaar | 1795 | add TMP0, TMP1 |
| 807 : | edgomez | 1530 | ADD_HH_RND0 |
| 808 : | EPILOG | ||
| 809 : | |||
| 810 : | Isibaar | 1793 | .Loop1: |
| 811 : | edgomez | 1530 | ADD_HH_RND1 |
| 812 : | Isibaar | 1795 | add TMP0, TMP1 |
| 813 : | edgomez | 1530 | ADD_HH_RND1 |
| 814 : | Isibaar | 1795 | add TMP0, TMP1 |
| 815 : | edgomez | 1530 | ADD_HH_RND1 |
| 816 : | Isibaar | 1795 | add TMP0, TMP1 |
| 817 : | edgomez | 1530 | ADD_HH_RND1 |
| 818 : | |||
| 819 : | EPILOG | ||
| 820 : | Isibaar | 1793 | ENDFUNC |
| 821 : | edgomez | 1540 | |
| 822 : | Isibaar | 1877 | NON_EXEC_STACK |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |