Parent Directory
| 
Revision Log
Revision 300 - (view) (download)
| 1 : | ia64p | 300 | // ------------------------------------------------------------------------------ |
| 2 : | // * Programmed by | ||
| 3 : | // * Johannes Singler (email@jsingler.de), Daniel Winkler (infostudent@uni.de) | ||
| 4 : | // * | ||
| 5 : | // * Programmed for the IA64 laboratory held at University Karlsruhe 2002 | ||
| 6 : | // * http://www.info.uni-karlsruhe.de/~rubino/ia64p/ | ||
| 7 : | // * | ||
| 8 : | // ------------------------------------------------------------------------------ | ||
| 9 : | // * | ||
| 10 : | // * This is the optimized assembler version of Halfpel8_Refine. This function | ||
| 11 : | // * is worth it to be optimized for the IA-64 architecture because of the huge | ||
| 12 : | // * register set. We can hold all necessary data in general use registers | ||
| 13 : | // * and reuse it. | ||
| 14 : | // * | ||
| 15 : | // * Our approach uses: | ||
| 16 : | // * - The Itanium command psad1, which solves the problem in hardware. | ||
| 17 : | // * - Alignment resolving to avoid memory faults | ||
| 18 : | // * - Massive lopp unrolling | ||
| 19 : | // * | ||
| 20 : | // ------------------------------------------------------------------------------ | ||
| 21 : | // * | ||
| 22 : | // * ------- Half-pixel steps around the center (*) and corresponding | ||
| 23 : | // * |0|1|0| register set parts. | ||
| 24 : | // * ------- | ||
| 25 : | // * |2|*|2| | ||
| 26 : | // * ------- | ||
| 27 : | // * |0|1|0| | ||
| 28 : | // * ------- | ||
| 29 : | // * | ||
| 30 : | // ------------------------------------------------------------------------------ | ||
| 31 : | // * calc_delta is split up in three parts wich are included from | ||
| 32 : | // * | ||
| 33 : | // * calc_delta_1.s | ||
| 34 : | // * calc_delta_2.s | ||
| 35 : | // * calc_delta_3.s | ||
| 36 : | // * | ||
| 37 : | // ------------------------------------------------------------------------------ | ||
| 38 : | // * We assume min_dx <= currX <= max_dx && min_dy <= currY <= max_dy | ||
| 39 : | |||
| 40 : | |||
| 41 : | .sdata | ||
| 42 : | .align 4 | ||
| 43 : | .type lambda_vec8#,@object | ||
| 44 : | .size lambda_vec8#,128 | ||
| 45 : | lambda_vec8: | ||
| 46 : | data4 0 | ||
| 47 : | data4 1 | ||
| 48 : | data4 1 | ||
| 49 : | data4 1 | ||
| 50 : | data4 1 | ||
| 51 : | data4 2 | ||
| 52 : | data4 2 | ||
| 53 : | data4 2 | ||
| 54 : | data4 2 | ||
| 55 : | data4 3 | ||
| 56 : | data4 3 | ||
| 57 : | data4 3 | ||
| 58 : | data4 4 | ||
| 59 : | data4 4 | ||
| 60 : | data4 4 | ||
| 61 : | data4 5 | ||
| 62 : | data4 5 | ||
| 63 : | data4 6 | ||
| 64 : | data4 7 | ||
| 65 : | data4 7 | ||
| 66 : | data4 8 | ||
| 67 : | data4 9 | ||
| 68 : | data4 10 | ||
| 69 : | data4 11 | ||
| 70 : | data4 13 | ||
| 71 : | data4 14 | ||
| 72 : | data4 16 | ||
| 73 : | data4 18 | ||
| 74 : | data4 21 | ||
| 75 : | data4 25 | ||
| 76 : | data4 30 | ||
| 77 : | data4 36 | ||
| 78 : | |||
| 79 : | |||
| 80 : | .type mvtab#,@object | ||
| 81 : | .size mvtab#,132 | ||
| 82 : | mvtab: | ||
| 83 : | data4 1 | ||
| 84 : | data4 2 | ||
| 85 : | data4 3 | ||
| 86 : | data4 4 | ||
| 87 : | data4 6 | ||
| 88 : | data4 7 | ||
| 89 : | data4 7 | ||
| 90 : | data4 7 | ||
| 91 : | data4 9 | ||
| 92 : | data4 9 | ||
| 93 : | data4 9 | ||
| 94 : | data4 10 | ||
| 95 : | data4 10 | ||
| 96 : | data4 10 | ||
| 97 : | data4 10 | ||
| 98 : | data4 10 | ||
| 99 : | data4 10 | ||
| 100 : | data4 10 | ||
| 101 : | data4 10 | ||
| 102 : | data4 10 | ||
| 103 : | data4 10 | ||
| 104 : | data4 10 | ||
| 105 : | data4 10 | ||
| 106 : | data4 10 | ||
| 107 : | data4 10 | ||
| 108 : | data4 11 | ||
| 109 : | data4 11 | ||
| 110 : | data4 11 | ||
| 111 : | data4 11 | ||
| 112 : | data4 11 | ||
| 113 : | data4 11 | ||
| 114 : | data4 12 | ||
| 115 : | data4 12 | ||
| 116 : | .text | ||
| 117 : | .align 16 | ||
| 118 : | .global Halfpel8_Refine_ia64# | ||
| 119 : | .proc Halfpel8_Refine_ia64# | ||
| 120 : | |||
| 121 : | Halfpel8_Refine_ia64: | ||
| 122 : | |||
| 123 : | pfs = r14 | ||
| 124 : | prsave = r15 | ||
| 125 : | |||
| 126 : | // Save important registers | ||
| 127 : | |||
| 128 : | alloc pfs = ar.pfs, 17, 75, 4, 96 | ||
| 129 : | mov prsave = pr | ||
| 130 : | |||
| 131 : | // Naming registers for better readability | ||
| 132 : | |||
| 133 : | pRef = in0 | ||
| 134 : | pRefH = in1 | ||
| 135 : | pRefV = in2 | ||
| 136 : | pRefHV = in3 | ||
| 137 : | cura = in4 | ||
| 138 : | x = in5 | ||
| 139 : | y = in6 | ||
| 140 : | currMV = in7 | ||
| 141 : | iMinSAD = in8 | ||
| 142 : | pmv = in9 | ||
| 143 : | min_dx = in10 | ||
| 144 : | max_dx = in11 | ||
| 145 : | min_dy = in12 | ||
| 146 : | max_dy = in13 | ||
| 147 : | iFcode = in14 | ||
| 148 : | iQuant = in15 | ||
| 149 : | iEdgedWidth = in16 | ||
| 150 : | |||
| 151 : | iSAD = r17 | ||
| 152 : | backupX = r18 | ||
| 153 : | backupY = r19 | ||
| 154 : | currX = r20 | ||
| 155 : | currY = r21 | ||
| 156 : | currYAddress = r22 | ||
| 157 : | bitX0 = r23 | ||
| 158 : | bitY0 = r24 | ||
| 159 : | dxd2 = r25 | ||
| 160 : | dyd2 = r26 | ||
| 161 : | offset = r27 | ||
| 162 : | block = r28 | ||
| 163 : | nob02 = r29 | ||
| 164 : | nob1 = r30 | ||
| 165 : | nob64m02 = r31 | ||
| 166 : | nob64m1 = r127 | ||
| 167 : | const7 = r126 | ||
| 168 : | nob56m02 = r125 | ||
| 169 : | dx = r124 | ||
| 170 : | dy = r123 | ||
| 171 : | oldX = r122 | ||
| 172 : | oldY = r121 | ||
| 173 : | |||
| 174 : | .rotr inregisters[17], refaa[3], refab[3], cur[8], ref0a[9], ref0b[9], ref1a[9], mpr[9], ref2a[8], ref2b[8], component[2], sc[2], tabaddress[2] | ||
| 175 : | |||
| 176 : | fx = f8 | ||
| 177 : | fy = f9 | ||
| 178 : | fblock = f10 | ||
| 179 : | fiEdgedWidth = f11 | ||
| 180 : | fdxd2 = f12 | ||
| 181 : | fdyd2 = f13 | ||
| 182 : | foffset = f14 | ||
| 183 : | fydiEdgedWidth = f15 | ||
| 184 : | fQuant = f16 | ||
| 185 : | fmv = f17 | ||
| 186 : | |||
| 187 : | n = p16 | ||
| 188 : | h = p17 | ||
| 189 : | v = p18 | ||
| 190 : | hv = p19 | ||
| 191 : | l = p20 | ||
| 192 : | r = p21 | ||
| 193 : | t = p22 | ||
| 194 : | b = p23 | ||
| 195 : | lt = p24 | ||
| 196 : | lb = p25 | ||
| 197 : | rt = p26 | ||
| 198 : | rb = p27 | ||
| 199 : | fb = p28 | ||
| 200 : | non0_0 = p30 | ||
| 201 : | non0_1 = p31 | ||
| 202 : | non0_2 = p32 | ||
| 203 : | non0_3 = p33 | ||
| 204 : | neg_0 = p34 | ||
| 205 : | neg_1 = p35 | ||
| 206 : | neg_2 = p36 | ||
| 207 : | neg_3 = p37 | ||
| 208 : | cg32_0 = p29 | ||
| 209 : | cg32_1 = p38 | ||
| 210 : | |||
| 211 : | // Initialize input variables | ||
| 212 : | |||
| 213 : | add sp = 16, sp | ||
| 214 : | ;; | ||
| 215 : | ld4 iMinSAD = [sp], 8 | ||
| 216 : | ;; | ||
| 217 : | sxt4 iMinSAD = iMinSAD | ||
| 218 : | ld8 pmv = [sp], 8 | ||
| 219 : | ;; | ||
| 220 : | ld4 min_dx = [sp], 8 | ||
| 221 : | ;; | ||
| 222 : | sxt4 min_dx = min_dx | ||
| 223 : | |||
| 224 : | ld4 max_dx = [sp], 8 | ||
| 225 : | ;; | ||
| 226 : | sxt4 max_dx = max_dx | ||
| 227 : | |||
| 228 : | ld4 min_dy = [sp], 8 | ||
| 229 : | ;; | ||
| 230 : | sxt4 min_dy = min_dy | ||
| 231 : | |||
| 232 : | ld4 max_dy = [sp], 8 | ||
| 233 : | ;; | ||
| 234 : | sxt4 max_dy = max_dy | ||
| 235 : | |||
| 236 : | ld4 iFcode = [sp], 8 | ||
| 237 : | ;; | ||
| 238 : | sxt4 iFcode = iFcode | ||
| 239 : | |||
| 240 : | ld4 iQuant = [sp], 8 | ||
| 241 : | |||
| 242 : | add tabaddress[0] = @gprel(lambda_vec8#), gp | ||
| 243 : | ;; | ||
| 244 : | shladd tabaddress[0] = iQuant, 2, tabaddress[0] | ||
| 245 : | ;; | ||
| 246 : | ld4 iQuant = [tabaddress[0]] | ||
| 247 : | ;; | ||
| 248 : | sxt4 iQuant = iQuant | ||
| 249 : | ;; | ||
| 250 : | add iFcode = -1, iFcode //only used in decreased version | ||
| 251 : | shl iQuant = iQuant, 1 | ||
| 252 : | ;; | ||
| 253 : | setf.sig fQuant = iQuant | ||
| 254 : | |||
| 255 : | ld4 iEdgedWidth = [sp] | ||
| 256 : | add sp = -80, sp | ||
| 257 : | |||
| 258 : | |||
| 259 : | |||
| 260 : | |||
| 261 : | // Initialize local variables | ||
| 262 : | |||
| 263 : | |||
| 264 : | ld4 currX = [currMV] | ||
| 265 : | add currYAddress = 4, currMV | ||
| 266 : | ;; | ||
| 267 : | sxt4 currX = currX | ||
| 268 : | ld4 currY = [currYAddress] | ||
| 269 : | ;; | ||
| 270 : | sxt4 currY = currY | ||
| 271 : | ;; | ||
| 272 : | // Calculate references | ||
| 273 : | |||
| 274 : | cmp.gt l, p0 = currX, min_dx | ||
| 275 : | cmp.lt r, p0 = currX, max_dx | ||
| 276 : | cmp.gt t, p0 = currY, min_dy | ||
| 277 : | cmp.lt b, p0 = currY, max_dy | ||
| 278 : | add backupX = -1, currX //move to left upper corner of quadrate | ||
| 279 : | add backupY = -1, currY | ||
| 280 : | |||
| 281 : | ;; | ||
| 282 : | (b) cmp.gt.unc lb, p0 = currX, min_dx | ||
| 283 : | (t) cmp.lt.unc rt, p0 = currX, max_dx | ||
| 284 : | (l) cmp.gt.unc lt, p0 = currY, min_dy | ||
| 285 : | (r) cmp.lt.unc rb, p0 = currY, max_dy | ||
| 286 : | |||
| 287 : | and bitX0 = 1, backupX | ||
| 288 : | and bitY0 = 1, backupY | ||
| 289 : | ;; | ||
| 290 : | cmp.eq n, p0 = 0, bitX0 | ||
| 291 : | cmp.eq h, p0 = 1, bitX0 | ||
| 292 : | cmp.eq v, p0 = 0, bitX0 | ||
| 293 : | cmp.eq hv, p0 = 1, bitX0 | ||
| 294 : | ;; | ||
| 295 : | cmp.eq.and n, p0 = 0, bitY0 | ||
| 296 : | cmp.eq.and h, p0 = 0, bitY0 | ||
| 297 : | cmp.eq.and v, p0 = 1, bitY0 | ||
| 298 : | cmp.eq.and hv, p0 = 1, bitY0 | ||
| 299 : | ;; | ||
| 300 : | |||
| 301 : | .pred.rel "mutex", p16, p17, p18, p19 //n, h, v, hv | ||
| 302 : | (n) mov refaa[0] = pRef | ||
| 303 : | (h) mov refaa[0] = pRefH | ||
| 304 : | (v) mov refaa[0] = pRefV | ||
| 305 : | (hv) mov refaa[0] = pRefHV | ||
| 306 : | |||
| 307 : | (n) mov refaa[1] = pRefH | ||
| 308 : | (h) mov refaa[1] = pRef | ||
| 309 : | (v) mov refaa[1] = pRefHV | ||
| 310 : | (hv) mov refaa[1] = pRefV | ||
| 311 : | |||
| 312 : | (n) mov refaa[2] = pRefV | ||
| 313 : | (h) mov refaa[2] = pRefHV | ||
| 314 : | (v) mov refaa[2] = pRef | ||
| 315 : | (hv) mov refaa[2] = pRefH | ||
| 316 : | |||
| 317 : | |||
| 318 : | // Calculate offset (integer multiplication on IA-64 sucks!) | ||
| 319 : | |||
| 320 : | mov block = 8 | ||
| 321 : | |||
| 322 : | shr dxd2 = backupX, 1 | ||
| 323 : | shr dyd2 = backupY, 1 | ||
| 324 : | |||
| 325 : | setf.sig fx = x | ||
| 326 : | setf.sig fy = y | ||
| 327 : | ;; | ||
| 328 : | setf.sig fblock = block | ||
| 329 : | setf.sig fiEdgedWidth = iEdgedWidth | ||
| 330 : | ;; | ||
| 331 : | setf.sig fdxd2 = dxd2 | ||
| 332 : | setf.sig fdyd2 = dyd2 | ||
| 333 : | ;; | ||
| 334 : | xma.l foffset = fx, fblock, fdxd2 | ||
| 335 : | xma.l fydiEdgedWidth = fy, fblock, fdyd2 | ||
| 336 : | ;; | ||
| 337 : | xma.l foffset = fydiEdgedWidth, fiEdgedWidth, foffset | ||
| 338 : | ;; | ||
| 339 : | getf.sig offset = foffset | ||
| 340 : | ;; | ||
| 341 : | add refaa[0] = refaa[0], offset | ||
| 342 : | add refaa[1] = refaa[1], offset | ||
| 343 : | add refaa[2] = refaa[2], offset | ||
| 344 : | ;; | ||
| 345 : | (h) add refaa[1] = 1, refaa[1] | ||
| 346 : | (hv) add refaa[1] = 1, refaa[1] | ||
| 347 : | (v) add refaa[2] = iEdgedWidth, refaa[2] | ||
| 348 : | (hv) add refaa[2] = iEdgedWidth, refaa[2] | ||
| 349 : | |||
| 350 : | // Load respecting misalignment of refx... | ||
| 351 : | |||
| 352 : | mov const7 = 7 | ||
| 353 : | ;; | ||
| 354 : | dep.z nob02 = refaa[0], 3, 3 | ||
| 355 : | dep.z nob1 = refaa[1], 3, 3 | ||
| 356 : | ;; | ||
| 357 : | andcm refaa[0] = refaa[0], const7 // set last 3 bits = 0 | ||
| 358 : | andcm refaa[1] = refaa[1], const7 | ||
| 359 : | andcm refaa[2] = refaa[2], const7 | ||
| 360 : | ;; | ||
| 361 : | add refab[0] = 8, refaa[0] | ||
| 362 : | add refab[1] = 8, refaa[1] | ||
| 363 : | add refab[2] = 8, refaa[2] | ||
| 364 : | ;; | ||
| 365 : | ld8 cur[0] = [cura], iEdgedWidth | ||
| 366 : | ld8 ref0a[0] = [refaa[0]], iEdgedWidth | ||
| 367 : | sub nob64m02 = 64, nob02 // 64 - nob | ||
| 368 : | |||
| 369 : | ld8 ref0b[0] = [refab[0]], iEdgedWidth | ||
| 370 : | ld8 ref1a[0] = [refaa[1]], iEdgedWidth | ||
| 371 : | sub nob56m02 = 56, nob02 // 56 - nob | ||
| 372 : | |||
| 373 : | ld8 mpr[0] = [refab[1]], iEdgedWidth | ||
| 374 : | ld8 ref2a[0] = [refaa[2]], iEdgedWidth | ||
| 375 : | sub nob64m1 = 64, nob1 | ||
| 376 : | |||
| 377 : | ld8 ref2b[0] = [refab[2]], iEdgedWidth | ||
| 378 : | ;; | ||
| 379 : | ld8 cur[1] = [cura], iEdgedWidth | ||
| 380 : | ld8 ref0a[1] = [refaa[0]], iEdgedWidth | ||
| 381 : | ld8 ref0b[1] = [refab[0]], iEdgedWidth | ||
| 382 : | ld8 ref1a[1] = [refaa[1]], iEdgedWidth | ||
| 383 : | ld8 mpr[1] = [refab[1]], iEdgedWidth | ||
| 384 : | ld8 ref2a[1] = [refaa[2]], iEdgedWidth | ||
| 385 : | ld8 ref2b[1] = [refab[2]], iEdgedWidth | ||
| 386 : | ;; | ||
| 387 : | ld8 cur[2] = [cura], iEdgedWidth | ||
| 388 : | ld8 ref0a[2] = [refaa[0]], iEdgedWidth | ||
| 389 : | ld8 ref0b[2] = [refab[0]], iEdgedWidth | ||
| 390 : | ld8 ref1a[2] = [refaa[1]], iEdgedWidth | ||
| 391 : | ld8 mpr[2] = [refab[1]], iEdgedWidth | ||
| 392 : | ld8 ref2a[2] = [refaa[2]], iEdgedWidth | ||
| 393 : | ld8 ref2b[2] = [refab[2]], iEdgedWidth | ||
| 394 : | ;; | ||
| 395 : | ld8 cur[3] = [cura], iEdgedWidth | ||
| 396 : | ld8 ref0a[3] = [refaa[0]], iEdgedWidth | ||
| 397 : | ld8 ref0b[3] = [refab[0]], iEdgedWidth | ||
| 398 : | ld8 ref1a[3] = [refaa[1]], iEdgedWidth | ||
| 399 : | ld8 mpr[3] = [refab[1]], iEdgedWidth | ||
| 400 : | ld8 ref2a[3] = [refaa[2]], iEdgedWidth | ||
| 401 : | ld8 ref2b[3] = [refab[2]], iEdgedWidth | ||
| 402 : | ;; | ||
| 403 : | ld8 cur[4] = [cura], iEdgedWidth | ||
| 404 : | ld8 ref0a[4] = [refaa[0]], iEdgedWidth | ||
| 405 : | ld8 ref0b[4] = [refab[0]], iEdgedWidth | ||
| 406 : | ld8 ref1a[4] = [refaa[1]], iEdgedWidth | ||
| 407 : | ld8 mpr[4] = [refab[1]], iEdgedWidth | ||
| 408 : | ld8 ref2a[4] = [refaa[2]], iEdgedWidth | ||
| 409 : | ld8 ref2b[4] = [refab[2]], iEdgedWidth | ||
| 410 : | ;; | ||
| 411 : | ld8 cur[5] = [cura], iEdgedWidth | ||
| 412 : | ld8 ref0a[5] = [refaa[0]], iEdgedWidth | ||
| 413 : | ld8 ref0b[5] = [refab[0]], iEdgedWidth | ||
| 414 : | ld8 ref1a[5] = [refaa[1]], iEdgedWidth | ||
| 415 : | ld8 mpr[5] = [refab[1]], iEdgedWidth | ||
| 416 : | ld8 ref2a[5] = [refaa[2]], iEdgedWidth | ||
| 417 : | ld8 ref2b[5] = [refab[2]], iEdgedWidth | ||
| 418 : | ;; | ||
| 419 : | ld8 cur[6] = [cura], iEdgedWidth | ||
| 420 : | ld8 ref0a[6] = [refaa[0]], iEdgedWidth | ||
| 421 : | ld8 ref0b[6] = [refab[0]], iEdgedWidth | ||
| 422 : | ld8 ref1a[6] = [refaa[1]], iEdgedWidth | ||
| 423 : | ld8 mpr[6] = [refab[1]], iEdgedWidth | ||
| 424 : | ld8 ref2a[6] = [refaa[2]], iEdgedWidth | ||
| 425 : | ld8 ref2b[6] = [refab[2]], iEdgedWidth | ||
| 426 : | ;; | ||
| 427 : | ld8 cur[7] = [cura] | ||
| 428 : | ld8 ref0a[7] = [refaa[0]], iEdgedWidth | ||
| 429 : | ld8 ref0b[7] = [refab[0]], iEdgedWidth | ||
| 430 : | ld8 ref1a[7] = [refaa[1]], iEdgedWidth | ||
| 431 : | ld8 mpr[7] = [refab[1]], iEdgedWidth | ||
| 432 : | ld8 ref2a[7] = [refaa[2]] | ||
| 433 : | ld8 ref2b[7] = [refab[2]] | ||
| 434 : | ;; | ||
| 435 : | ld8 ref0a[8] = [refaa[0]] | ||
| 436 : | ld8 ref0b[8] = [refab[0]] | ||
| 437 : | ld8 ref1a[8] = [refaa[1]] | ||
| 438 : | ld8 mpr[8] = [refab[1]] | ||
| 439 : | ;; | ||
| 440 : | |||
| 441 : | |||
| 442 : | // Align ref1 | ||
| 443 : | |||
| 444 : | shr.u ref1a[0] = ref1a[0], nob1 | ||
| 445 : | shr.u ref1a[1] = ref1a[1], nob1 | ||
| 446 : | shr.u ref1a[2] = ref1a[2], nob1 | ||
| 447 : | shr.u ref1a[3] = ref1a[3], nob1 | ||
| 448 : | shr.u ref1a[4] = ref1a[4], nob1 | ||
| 449 : | shr.u ref1a[5] = ref1a[5], nob1 | ||
| 450 : | shr.u ref1a[6] = ref1a[6], nob1 | ||
| 451 : | shr.u ref1a[7] = ref1a[7], nob1 | ||
| 452 : | shr.u ref1a[8] = ref1a[8], nob1 | ||
| 453 : | |||
| 454 : | shl mpr[0] = mpr[0], nob64m1 | ||
| 455 : | shl mpr[1] = mpr[1], nob64m1 | ||
| 456 : | shl mpr[2] = mpr[2], nob64m1 | ||
| 457 : | shl mpr[3] = mpr[3], nob64m1 | ||
| 458 : | shl mpr[4] = mpr[4], nob64m1 | ||
| 459 : | shl mpr[5] = mpr[5], nob64m1 | ||
| 460 : | shl mpr[6] = mpr[6], nob64m1 | ||
| 461 : | shl mpr[7] = mpr[7], nob64m1 | ||
| 462 : | shl mpr[8] = mpr[8], nob64m1 | ||
| 463 : | ;; | ||
| 464 : | .explicit | ||
| 465 : | {.mii | ||
| 466 : | or ref1a[0] = ref1a[0], mpr[0] | ||
| 467 : | shr.u ref0a[0] = ref0a[0], nob02 | ||
| 468 : | shr.u ref0a[1] = ref0a[1], nob02 | ||
| 469 : | } | ||
| 470 : | {.mmi | ||
| 471 : | or ref1a[1] = ref1a[1], mpr[1] | ||
| 472 : | or ref1a[2] = ref1a[2], mpr[2] | ||
| 473 : | shr.u ref0a[2] = ref0a[2], nob02 | ||
| 474 : | } | ||
| 475 : | {.mii | ||
| 476 : | or ref1a[3] = ref1a[3], mpr[3] | ||
| 477 : | shr.u ref0a[3] = ref0a[3], nob02 | ||
| 478 : | shr.u ref0a[4] = ref0a[4], nob02 | ||
| 479 : | } | ||
| 480 : | {.mmi | ||
| 481 : | or ref1a[4] = ref1a[4], mpr[4] | ||
| 482 : | or ref1a[5] = ref1a[5], mpr[5] | ||
| 483 : | shr.u ref0a[5] = ref0a[5], nob02 | ||
| 484 : | } | ||
| 485 : | {.mii | ||
| 486 : | or ref1a[6] = ref1a[6], mpr[6] | ||
| 487 : | shr.u ref0a[6] = ref0a[6], nob02 | ||
| 488 : | shr.u ref0a[7] = ref0a[7], nob02 | ||
| 489 : | } | ||
| 490 : | {.mii | ||
| 491 : | or ref1a[7] = ref1a[7], mpr[7] | ||
| 492 : | or ref1a[8] = ref1a[8], mpr[8] | ||
| 493 : | shr.u ref0a[8] = ref0a[8], nob02 | ||
| 494 : | } | ||
| 495 : | .default | ||
| 496 : | // ref1a[] now contains center position values | ||
| 497 : | // mpr[] not used any more | ||
| 498 : | |||
| 499 : | // Align ref0 left | ||
| 500 : | |||
| 501 : | ;; | ||
| 502 : | shl mpr[0] = ref0b[0], nob56m02 | ||
| 503 : | shl mpr[1] = ref0b[1], nob56m02 | ||
| 504 : | shl mpr[2] = ref0b[2], nob56m02 | ||
| 505 : | shl mpr[3] = ref0b[3], nob56m02 | ||
| 506 : | shl mpr[4] = ref0b[4], nob56m02 | ||
| 507 : | shl mpr[5] = ref0b[5], nob56m02 | ||
| 508 : | shl mpr[6] = ref0b[6], nob56m02 | ||
| 509 : | shl mpr[7] = ref0b[7], nob56m02 | ||
| 510 : | shl mpr[8] = ref0b[8], nob56m02 | ||
| 511 : | |||
| 512 : | shl ref0b[0] = ref0b[0], nob64m02 | ||
| 513 : | shl ref0b[1] = ref0b[1], nob64m02 | ||
| 514 : | shl ref0b[2] = ref0b[2], nob64m02 | ||
| 515 : | shl ref0b[3] = ref0b[3], nob64m02 | ||
| 516 : | shl ref0b[4] = ref0b[4], nob64m02 | ||
| 517 : | shl ref0b[5] = ref0b[5], nob64m02 | ||
| 518 : | shl ref0b[6] = ref0b[6], nob64m02 | ||
| 519 : | shl ref0b[7] = ref0b[7], nob64m02 | ||
| 520 : | shl ref0b[8] = ref0b[8], nob64m02 | ||
| 521 : | ;; | ||
| 522 : | or ref0a[0] = ref0a[0], ref0b[0] | ||
| 523 : | or ref0a[1] = ref0a[1], ref0b[1] | ||
| 524 : | or ref0a[2] = ref0a[2], ref0b[2] | ||
| 525 : | or ref0a[3] = ref0a[3], ref0b[3] | ||
| 526 : | or ref0a[4] = ref0a[4], ref0b[4] | ||
| 527 : | or ref0a[5] = ref0a[5], ref0b[5] | ||
| 528 : | or ref0a[6] = ref0a[6], ref0b[6] | ||
| 529 : | or ref0a[7] = ref0a[7], ref0b[7] | ||
| 530 : | or ref0a[8] = ref0a[8], ref0b[8] | ||
| 531 : | ;; | ||
| 532 : | |||
| 533 : | // ref0a[] now contains left position values | ||
| 534 : | // mpr[] contains intermediate result for right position values (former ref0a << 56 - nob02) | ||
| 535 : | |||
| 536 : | // Align ref0 right | ||
| 537 : | |||
| 538 : | // Shift one byte more to the right (seen als big-endian) | ||
| 539 : | shr.u ref0b[0] = ref0a[0], 8 | ||
| 540 : | shr.u ref0b[1] = ref0a[1], 8 | ||
| 541 : | shr.u ref0b[2] = ref0a[2], 8 | ||
| 542 : | shr.u ref0b[3] = ref0a[3], 8 | ||
| 543 : | shr.u ref0b[4] = ref0a[4], 8 | ||
| 544 : | shr.u ref0b[5] = ref0a[5], 8 | ||
| 545 : | shr.u ref0b[6] = ref0a[6], 8 | ||
| 546 : | shr.u ref0b[7] = ref0a[7], 8 | ||
| 547 : | shr.u ref0b[8] = ref0a[8], 8 | ||
| 548 : | ;; | ||
| 549 : | .explicit | ||
| 550 : | {.mii | ||
| 551 : | or ref0b[0] = ref0b[0], mpr[0] | ||
| 552 : | shr.u ref2a[0] = ref2a[0], nob02 | ||
| 553 : | shr.u ref2a[1] = ref2a[1], nob02 | ||
| 554 : | } | ||
| 555 : | {.mmi | ||
| 556 : | or ref0b[1] = ref0b[1], mpr[1] | ||
| 557 : | or ref0b[2] = ref0b[2], mpr[2] | ||
| 558 : | shr.u ref2a[2] = ref2a[2], nob02 | ||
| 559 : | } | ||
| 560 : | {.mii | ||
| 561 : | or ref0b[3] = ref0b[3], mpr[3] | ||
| 562 : | shr.u ref2a[3] = ref2a[3], nob02 | ||
| 563 : | shr.u ref2a[4] = ref2a[4], nob02 | ||
| 564 : | } | ||
| 565 : | {.mmi | ||
| 566 : | or ref0b[4] = ref0b[4], mpr[4] | ||
| 567 : | or ref0b[5] = ref0b[5], mpr[5] | ||
| 568 : | shr.u ref2a[5] = ref2a[5], nob02 | ||
| 569 : | } | ||
| 570 : | {.mii | ||
| 571 : | or ref0b[6] = ref0b[6], mpr[6] | ||
| 572 : | shr.u ref2a[6] = ref2a[6], nob02 | ||
| 573 : | shr.u ref2a[7] = ref2a[7], nob02 | ||
| 574 : | } | ||
| 575 : | .default | ||
| 576 : | or ref0b[7] = ref0b[7], mpr[7] | ||
| 577 : | or ref0b[8] = ref0b[8], mpr[8] | ||
| 578 : | |||
| 579 : | // ref0b[] now contains right position values | ||
| 580 : | // mpr[] not needed any more | ||
| 581 : | |||
| 582 : | |||
| 583 : | // Align ref2 left | ||
| 584 : | |||
| 585 : | ;; | ||
| 586 : | shl mpr[0] = ref2b[0], nob56m02 | ||
| 587 : | shl mpr[1] = ref2b[1], nob56m02 | ||
| 588 : | shl mpr[2] = ref2b[2], nob56m02 | ||
| 589 : | shl mpr[3] = ref2b[3], nob56m02 | ||
| 590 : | shl mpr[4] = ref2b[4], nob56m02 | ||
| 591 : | shl mpr[5] = ref2b[5], nob56m02 | ||
| 592 : | shl mpr[6] = ref2b[6], nob56m02 | ||
| 593 : | shl mpr[7] = ref2b[7], nob56m02 | ||
| 594 : | |||
| 595 : | shl ref2b[0] = ref2b[0], nob64m02 | ||
| 596 : | shl ref2b[1] = ref2b[1], nob64m02 | ||
| 597 : | shl ref2b[2] = ref2b[2], nob64m02 | ||
| 598 : | shl ref2b[3] = ref2b[3], nob64m02 | ||
| 599 : | shl ref2b[4] = ref2b[4], nob64m02 | ||
| 600 : | shl ref2b[5] = ref2b[5], nob64m02 | ||
| 601 : | shl ref2b[6] = ref2b[6], nob64m02 | ||
| 602 : | shl ref2b[7] = ref2b[7], nob64m02 | ||
| 603 : | ;; | ||
| 604 : | or ref2a[0] = ref2a[0], ref2b[0] | ||
| 605 : | or ref2a[1] = ref2a[1], ref2b[1] | ||
| 606 : | or ref2a[2] = ref2a[2], ref2b[2] | ||
| 607 : | or ref2a[3] = ref2a[3], ref2b[3] | ||
| 608 : | or ref2a[4] = ref2a[4], ref2b[4] | ||
| 609 : | or ref2a[5] = ref2a[5], ref2b[5] | ||
| 610 : | or ref2a[6] = ref2a[6], ref2b[6] | ||
| 611 : | or ref2a[7] = ref2a[7], ref2b[7] | ||
| 612 : | ;; | ||
| 613 : | |||
| 614 : | // ref2a[] now contains left position values | ||
| 615 : | // mpr[] contains intermediate result for right position values (former ref2a << 56 - nob02) | ||
| 616 : | |||
| 617 : | // Align ref2 right | ||
| 618 : | |||
| 619 : | // Shift one byte more to the right (seen als big-endian) | ||
| 620 : | shr.u ref2b[0] = ref2a[0], 8 | ||
| 621 : | shr.u ref2b[1] = ref2a[1], 8 | ||
| 622 : | shr.u ref2b[2] = ref2a[2], 8 | ||
| 623 : | shr.u ref2b[3] = ref2a[3], 8 | ||
| 624 : | shr.u ref2b[4] = ref2a[4], 8 | ||
| 625 : | shr.u ref2b[5] = ref2a[5], 8 | ||
| 626 : | shr.u ref2b[6] = ref2a[6], 8 | ||
| 627 : | shr.u ref2b[7] = ref2a[7], 8 | ||
| 628 : | ;; | ||
| 629 : | or ref2b[0] = ref2b[0], mpr[0] | ||
| 630 : | or ref2b[1] = ref2b[1], mpr[1] | ||
| 631 : | or ref2b[2] = ref2b[2], mpr[2] | ||
| 632 : | or ref2b[3] = ref2b[3], mpr[3] | ||
| 633 : | or ref2b[4] = ref2b[4], mpr[4] | ||
| 634 : | or ref2b[5] = ref2b[5], mpr[5] | ||
| 635 : | or ref2b[6] = ref2b[6], mpr[6] | ||
| 636 : | or ref2b[7] = ref2b[7], mpr[7] | ||
| 637 : | |||
| 638 : | |||
| 639 : | // ref2b[] now contains right position values | ||
| 640 : | // mpr[] not needed any more | ||
| 641 : | |||
| 642 : | |||
| 643 : | ld4 dx = [pmv], 4 //+ sizeof(int) | ||
| 644 : | ;; | ||
| 645 : | sxt4 dx = dx | ||
| 646 : | ld4 dy = [pmv] | ||
| 647 : | ;; | ||
| 648 : | sxt4 dy = dy | ||
| 649 : | ;; | ||
| 650 : | |||
| 651 : | |||
| 652 : | // Let's SAD | ||
| 653 : | |||
| 654 : | // Left top corner | ||
| 655 : | |||
| 656 : | |||
| 657 : | sub dx = backupX, dx | ||
| 658 : | psad1 mpr[0] = cur[0], ref0a[0] | ||
| 659 : | psad1 mpr[1] = cur[1], ref0a[1] | ||
| 660 : | |||
| 661 : | sub dy = backupY, dy | ||
| 662 : | psad1 mpr[2] = cur[2], ref0a[2] | ||
| 663 : | psad1 mpr[3] = cur[3], ref0a[3] | ||
| 664 : | psad1 mpr[4] = cur[4], ref0a[4] | ||
| 665 : | psad1 mpr[5] = cur[5], ref0a[5] | ||
| 666 : | psad1 mpr[6] = cur[6], ref0a[6] | ||
| 667 : | psad1 mpr[7] = cur[7], ref0a[7] | ||
| 668 : | ;; | ||
| 669 : | .include "../../src/motion/ia64_asm/calc_delta_1.s" | ||
| 670 : | |||
| 671 : | // Top edge | ||
| 672 : | |||
| 673 : | psad1 mpr[0] = cur[0], ref1a[0] | ||
| 674 : | psad1 mpr[1] = cur[1], ref1a[1] | ||
| 675 : | psad1 mpr[2] = cur[2], ref1a[2] | ||
| 676 : | psad1 mpr[3] = cur[3], ref1a[3] | ||
| 677 : | psad1 mpr[4] = cur[4], ref1a[4] | ||
| 678 : | |||
| 679 : | add dx = 1, dx | ||
| 680 : | psad1 mpr[5] = cur[5], ref1a[5] | ||
| 681 : | psad1 mpr[6] = cur[6], ref1a[6] | ||
| 682 : | |||
| 683 : | psad1 mpr[7] = cur[7], ref1a[7] | ||
| 684 : | ;; | ||
| 685 : | |||
| 686 : | .include "../../src/motion/ia64_asm/calc_delta_2.s" | ||
| 687 : | (lt) cmp.lt.unc fb, p0 = mpr[8], iMinSAD | ||
| 688 : | .include "../../src/motion/ia64_asm/calc_delta_3.s" | ||
| 689 : | |||
| 690 : | // Right top corner | ||
| 691 : | |||
| 692 : | |||
| 693 : | psad1 mpr[0] = cur[0], ref0b[0] | ||
| 694 : | psad1 mpr[1] = cur[1], ref0b[1] | ||
| 695 : | psad1 mpr[2] = cur[2], ref0b[2] | ||
| 696 : | psad1 mpr[3] = cur[3], ref0b[3] | ||
| 697 : | psad1 mpr[4] = cur[4], ref0b[4] | ||
| 698 : | |||
| 699 : | add backupX = 1, backupX | ||
| 700 : | psad1 mpr[5] = cur[5], ref0b[5] | ||
| 701 : | psad1 mpr[6] = cur[6], ref0b[6] | ||
| 702 : | |||
| 703 : | add dx = 1, dx | ||
| 704 : | psad1 mpr[7] = cur[7], ref0b[7] | ||
| 705 : | ;; | ||
| 706 : | |||
| 707 : | .include "../../src/motion/ia64_asm/calc_delta_1.s" | ||
| 708 : | (t) cmp.lt.unc fb, p0 = iSAD, iMinSAD | ||
| 709 : | ;; | ||
| 710 : | |||
| 711 : | // Left edge | ||
| 712 : | |||
| 713 : | (fb) mov iMinSAD = iSAD | ||
| 714 : | psad1 mpr[0] = cur[0], ref2a[0] | ||
| 715 : | |||
| 716 : | (fb) mov currX = backupX | ||
| 717 : | psad1 mpr[1] = cur[1], ref2a[1] | ||
| 718 : | psad1 mpr[2] = cur[2], ref2a[2] | ||
| 719 : | |||
| 720 : | (fb) mov currY = backupY | ||
| 721 : | psad1 mpr[3] = cur[3], ref2a[3] | ||
| 722 : | psad1 mpr[4] = cur[4], ref2a[4] | ||
| 723 : | |||
| 724 : | add backupX = 1, backupX | ||
| 725 : | psad1 mpr[5] = cur[5], ref2a[5] | ||
| 726 : | psad1 mpr[6] = cur[6], ref2a[6] | ||
| 727 : | |||
| 728 : | psad1 mpr[7] = cur[7], ref2a[7] | ||
| 729 : | |||
| 730 : | add dx = -2, dx | ||
| 731 : | add dy = 1, dy | ||
| 732 : | ;; | ||
| 733 : | |||
| 734 : | .include "../../src/motion/ia64_asm/calc_delta_2.s" | ||
| 735 : | (rt) cmp.lt.unc fb, p0 = mpr[8], iMinSAD | ||
| 736 : | .include "../../src/motion/ia64_asm/calc_delta_3.s" | ||
| 737 : | |||
| 738 : | // Right edge | ||
| 739 : | |||
| 740 : | |||
| 741 : | psad1 mpr[0] = cur[0], ref2b[0] | ||
| 742 : | psad1 mpr[1] = cur[1], ref2b[1] | ||
| 743 : | psad1 mpr[2] = cur[2], ref2b[2] | ||
| 744 : | psad1 mpr[3] = cur[3], ref2b[3] | ||
| 745 : | psad1 mpr[4] = cur[4], ref2b[4] | ||
| 746 : | |||
| 747 : | add backupX = -2, backupX | ||
| 748 : | psad1 mpr[5] = cur[5], ref2b[5] | ||
| 749 : | psad1 mpr[6] = cur[6], ref2b[6] | ||
| 750 : | |||
| 751 : | add backupY = 1, backupY | ||
| 752 : | add dx = 2, dx | ||
| 753 : | psad1 mpr[7] = cur[7], ref2b[7] | ||
| 754 : | ;; | ||
| 755 : | |||
| 756 : | .include "../../src/motion/ia64_asm/calc_delta_1.s" | ||
| 757 : | (l) cmp.lt.unc fb, p0 = iSAD, iMinSAD | ||
| 758 : | ;; | ||
| 759 : | |||
| 760 : | // Left bottom corner | ||
| 761 : | |||
| 762 : | (fb) mov iMinSAD = iSAD | ||
| 763 : | psad1 mpr[0] = cur[0], ref0a[1] | ||
| 764 : | |||
| 765 : | (fb) mov currX = backupX | ||
| 766 : | psad1 mpr[1] = cur[1], ref0a[2] | ||
| 767 : | psad1 mpr[2] = cur[2], ref0a[3] | ||
| 768 : | |||
| 769 : | (fb) mov currY = backupY | ||
| 770 : | psad1 mpr[3] = cur[3], ref0a[4] | ||
| 771 : | psad1 mpr[4] = cur[4], ref0a[5] | ||
| 772 : | |||
| 773 : | add backupX = 2, backupX | ||
| 774 : | psad1 mpr[5] = cur[5], ref0a[6] | ||
| 775 : | psad1 mpr[6] = cur[6], ref0a[7] | ||
| 776 : | |||
| 777 : | psad1 mpr[7] = cur[7], ref0a[8] | ||
| 778 : | |||
| 779 : | add dx = -2, dx | ||
| 780 : | add dy = 1, dy | ||
| 781 : | ;; | ||
| 782 : | |||
| 783 : | .include "../../src/motion/ia64_asm/calc_delta_2.s" | ||
| 784 : | (r) cmp.lt.unc fb, p0 = mpr[8], iMinSAD | ||
| 785 : | .include "../../src/motion/ia64_asm/calc_delta_3.s" | ||
| 786 : | |||
| 787 : | // Bottom edge | ||
| 788 : | |||
| 789 : | psad1 mpr[0] = cur[0], ref1a[1] | ||
| 790 : | psad1 mpr[1] = cur[1], ref1a[2] | ||
| 791 : | psad1 mpr[2] = cur[2], ref1a[3] | ||
| 792 : | psad1 mpr[3] = cur[3], ref1a[4] | ||
| 793 : | psad1 mpr[4] = cur[4], ref1a[5] | ||
| 794 : | |||
| 795 : | add backupX = -2, backupX | ||
| 796 : | psad1 mpr[5] = cur[5], ref1a[6] | ||
| 797 : | psad1 mpr[6] = cur[6], ref1a[7] | ||
| 798 : | |||
| 799 : | add backupY = 1, backupY | ||
| 800 : | add dx = 1, dx | ||
| 801 : | psad1 mpr[7] = cur[7], ref1a[8] | ||
| 802 : | ;; | ||
| 803 : | |||
| 804 : | .include "../../src/motion/ia64_asm/calc_delta_1.s" | ||
| 805 : | (lb) cmp.lt.unc fb, p0 = iSAD, iMinSAD | ||
| 806 : | ;; | ||
| 807 : | // Right bottom corner | ||
| 808 : | |||
| 809 : | |||
| 810 : | (fb) mov iMinSAD = iSAD | ||
| 811 : | psad1 mpr[0] = cur[0], ref0b[1] | ||
| 812 : | |||
| 813 : | (fb) mov currX = backupX | ||
| 814 : | psad1 mpr[1] = cur[1], ref0b[2] | ||
| 815 : | psad1 mpr[2] = cur[2], ref0b[3] | ||
| 816 : | |||
| 817 : | (fb) mov currY = backupY | ||
| 818 : | psad1 mpr[3] = cur[3], ref0b[4] | ||
| 819 : | psad1 mpr[4] = cur[4], ref0b[5] | ||
| 820 : | |||
| 821 : | add backupX = 1, backupX | ||
| 822 : | psad1 mpr[5] = cur[5], ref0b[6] | ||
| 823 : | psad1 mpr[6] = cur[6], ref0b[7] | ||
| 824 : | |||
| 825 : | add dx = 1, dx | ||
| 826 : | psad1 mpr[7] = cur[7], ref0b[8] | ||
| 827 : | ;; | ||
| 828 : | |||
| 829 : | .include "../../src/motion/ia64_asm/calc_delta_2.s" | ||
| 830 : | (b) cmp.lt.unc fb, p0 = mpr[8], iMinSAD | ||
| 831 : | .include "../../src/motion/ia64_asm/calc_delta_3.s" | ||
| 832 : | |||
| 833 : | (rb) getf.sig ret0 = fmv | ||
| 834 : | add backupX = 1, backupX | ||
| 835 : | ;; | ||
| 836 : | (rb) add iSAD = iSAD, ret0 | ||
| 837 : | ;; | ||
| 838 : | (rb) cmp.lt.unc fb, p0 = iSAD, iMinSAD | ||
| 839 : | ;; | ||
| 840 : | (fb) mov iMinSAD = iSAD | ||
| 841 : | (fb) mov currX = backupX | ||
| 842 : | (fb) mov currY = backupY | ||
| 843 : | ;; | ||
| 844 : | |||
| 845 : | // Write back result | ||
| 846 : | |||
| 847 : | st4 [currMV] = currX | ||
| 848 : | st4 [currYAddress] = currY | ||
| 849 : | mov ret0 = iMinSAD | ||
| 850 : | |||
| 851 : | // Restore important registers | ||
| 852 : | |||
| 853 : | ;; | ||
| 854 : | mov pr = prsave, -1 | ||
| 855 : | mov ar.pfs = pfs | ||
| 856 : | br.ret.sptk.many b0 | ||
| 857 : | |||
| 858 : | .endp Halfpel8_Refine_ia64# |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |