Parent Directory
| 
Revision Log
Revision 1.2 - (view) (download)
| 1 : | chl | 1.1 | /************************************************************************** |
| 2 : | * | ||
| 3 : | * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | * multithreaded motion estimation | ||
| 5 : | * | ||
| 6 : | * This program is an implementation of a part of one or more MPEG-4 | ||
| 7 : | * Video tools as specified in ISO/IEC 14496-2 standard. Those intending | ||
| 8 : | * to use this software module in hardware or software products are | ||
| 9 : | * advised that its use may infringe existing patents or copyrights, and | ||
| 10 : | * any such use would be at such party's own risk. The original | ||
| 11 : | * developer of this software module and his/her company, and subsequent | ||
| 12 : | * editors and their companies, will have no liability for use of this | ||
| 13 : | * software or modifications or derivatives thereof. | ||
| 14 : | * | ||
| 15 : | * This program is free software; you can redistribute it and/or modify | ||
| 16 : | * it under the terms of the GNU General Public License as published by | ||
| 17 : | * the Free Software Foundation; either version 2 of the License, or | ||
| 18 : | * (at your option) any later version. | ||
| 19 : | * | ||
| 20 : | * This program is distributed in the hope that it will be useful, | ||
| 21 : | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 22 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 23 : | * GNU General Public License for more details. | ||
| 24 : | * | ||
| 25 : | * You should have received a copy of the GNU General Public License | ||
| 26 : | * along with this program; if not, write to the Free Software | ||
| 27 : | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | ||
| 28 : | * | ||
| 29 : | *************************************************************************/ | ||
| 30 : | |||
| 31 : | |||
| 32 : | /* WARNING ! | ||
| 33 : | * This is early alpha code, only published for debugging and testing | ||
| 34 : | * I cannot garantee that it will not break compiliing, encoding | ||
| 35 : | * or your marriage etc. YOU HAVE BEEN WARNED! | ||
| 36 : | */ | ||
| 37 : | |||
| 38 : | #ifdef _SMP | ||
| 39 : | |||
| 40 : | #include <stdio.h> | ||
| 41 : | #include <stdlib.h> | ||
| 42 : | |||
| 43 : | #include <pthread.h> | ||
| 44 : | #include <signal.h> | ||
| 45 : | #include <unistd.h> | ||
| 46 : | |||
| 47 : | #include "../encoder.h" | ||
| 48 : | #include "../utils/mbfunctions.h" | ||
| 49 : | #include "../prediction/mbprediction.h" | ||
| 50 : | #include "../global.h" | ||
| 51 : | #include "../utils/timer.h" | ||
| 52 : | #include "motion.h" | ||
| 53 : | #include "sad.h" | ||
| 54 : | |||
| 55 : | #include "smp_motion_est.h" | ||
| 56 : | |||
| 57 : | /* global variables for SMP search control, the are not needed anywhere else than here */ | ||
| 58 : | |||
| 59 : | pthread_mutex_t me_mutex = PTHREAD_MUTEX_INITIALIZER; | ||
| 60 : | pthread_cond_t me_inqueue_cond = PTHREAD_COND_INITIALIZER; | ||
| 61 : | pthread_cond_t me_corrqueue_cond = PTHREAD_COND_INITIALIZER; | ||
| 62 : | pthread_cond_t me_outqueue_cond = PTHREAD_COND_INITIALIZER; | ||
| 63 : | |||
| 64 : | |||
| 65 : | int me_iIntra=0; | ||
| 66 : | int me_inqueue=0; // input queue | ||
| 67 : | int me_corrqueue=0; // correction queue | ||
| 68 : | int me_outqueue=0; // output queue | ||
| 69 : | |||
| 70 : | |||
| 71 : | void SMP_correct_pmv(int x, int y, int iWcount, MACROBLOCK* pMBs) | ||
| 72 : | { | ||
| 73 : | MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; | ||
| 74 : | VECTOR pmv; | ||
| 75 : | int k; | ||
| 76 : | |||
| 77 : | switch (pMB->mode) { | ||
| 78 : | |||
| 79 : | case MODE_INTER: | ||
| 80 : | case MODE_INTER_Q: | ||
| 81 : | pmv = get_pmv(pMBs, x, y, iWcount, 0); | ||
| 82 : | pMB->pmvs[0].x = pMB->mvs[0].x - pmv.x; | ||
| 83 : | pMB->pmvs[0].y = pMB->mvs[0].y - pmv.y; | ||
| 84 : | break; | ||
| 85 : | |||
| 86 : | case MODE_INTER4V: | ||
| 87 : | for (k=0;k<4;k++) { | ||
| 88 : | pmv = get_pmv(pMBs, x, y, iWcount, k); | ||
| 89 : | pMB->pmvs[k].x = pMB->mvs[k].x - pmv.x; | ||
| 90 : | pMB->pmvs[k].y = pMB->mvs[k].y - pmv.y; | ||
| 91 : | } | ||
| 92 : | break; | ||
| 93 : | |||
| 94 : | default: | ||
| 95 : | break; /* e.g. everything without prediction, e.g. MODE_INTRA */ | ||
| 96 : | } | ||
| 97 : | return; | ||
| 98 : | } | ||
| 99 : | |||
| 100 : | void SMP_MotionEstimationWorker(jobdata *arg) | ||
| 101 : | { | ||
| 102 : | const VECTOR zeroMV = { 0, 0 }; | ||
| 103 : | |||
| 104 : | // long long time; | ||
| 105 : | int32_t x, y; | ||
| 106 : | VECTOR pmv; | ||
| 107 : | |||
| 108 : | globaldata* gdata; | ||
| 109 : | |||
| 110 : | MBParam * pParam; | ||
| 111 : | FRAMEINFO * current; | ||
| 112 : | FRAMEINFO * reference; | ||
| 113 : | IMAGE * pRefH; | ||
| 114 : | IMAGE * pRefV; | ||
| 115 : | IMAGE * pRefHV; | ||
| 116 : | // uint32_t iLimit; | ||
| 117 : | |||
| 118 : | uint32_t iWcount; | ||
| 119 : | uint32_t iHcount; | ||
| 120 : | MACROBLOCK * pMBs; | ||
| 121 : | MACROBLOCK * prevMBs; | ||
| 122 : | IMAGE * pCurrent; | ||
| 123 : | IMAGE * pRef; | ||
| 124 : | |||
| 125 : | int minx = arg->minx; | ||
| 126 : | int maxx = arg->maxx; | ||
| 127 : | int miny = arg->miny; | ||
| 128 : | int maxy = arg->maxy; | ||
| 129 : | |||
| 130 : | // int run=0; | ||
| 131 : | int iIntra; | ||
| 132 : | |||
| 133 : | pthread_mutex_lock(&me_mutex); | ||
| 134 : | while (1) | ||
| 135 : | { | ||
| 136 : | // run++; | ||
| 137 : | iIntra = 0; | ||
| 138 : | |||
| 139 : | // fprintf(stderr,"[%d,%d] wait inqueue %d init\n",arg->id,run,me_inqueue); | ||
| 140 : | while (!me_inqueue) | ||
| 141 : | pthread_cond_wait(&me_inqueue_cond,&me_mutex); | ||
| 142 : | |||
| 143 : | // fprintf(stderr,"[%d,%d] wait inqueue %d done\n",arg->id,run,me_inqueue); | ||
| 144 : | |||
| 145 : | me_inqueue--; | ||
| 146 : | pthread_mutex_unlock(&me_mutex); | ||
| 147 : | |||
| 148 : | gdata = arg->gdata; | ||
| 149 : | pParam = gdata->pParam; | ||
| 150 : | current = gdata->current; | ||
| 151 : | reference = gdata->reference; | ||
| 152 : | pRefH = gdata->pRefH; | ||
| 153 : | pRefV = gdata->pRefV; | ||
| 154 : | pRefHV = gdata->pRefHV; | ||
| 155 : | // iLimit = gdata->iLimit; | ||
| 156 : | |||
| 157 : | iWcount = pParam->mb_width; | ||
| 158 : | iHcount = pParam->mb_height; | ||
| 159 : | pMBs = current->mbs; | ||
| 160 : | prevMBs = reference->mbs; | ||
| 161 : | pCurrent = ¤t->image; | ||
| 162 : | pRef = &reference->image; | ||
| 163 : | |||
| 164 : | chl | 1.2 | // time = read_counter(); |
| 165 : | chl | 1.1 | |
| 166 : | for (y = miny; y < maxy; y++) { | ||
| 167 : | for (x = minx; x < maxx; x++) { | ||
| 168 : | |||
| 169 : | MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; | ||
| 170 : | |||
| 171 : | pMB->sad16 = | ||
| 172 : | SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, | ||
| 173 : | x, y, current->motion_flags, current->quant, | ||
| 174 : | current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, | ||
| 175 : | &pMB->pmvs[0]); | ||
| 176 : | |||
| 177 : | if (0 < (pMB->sad16 - MV16_INTER_BIAS)) { | ||
| 178 : | int32_t deviation; | ||
| 179 : | |||
| 180 : | deviation = | ||
| 181 : | dev16(pCurrent->y + x * 16 + y * 16 * pParam->edged_width, | ||
| 182 : | pParam->edged_width); | ||
| 183 : | |||
| 184 : | if (deviation < (pMB->sad16 - MV16_INTER_BIAS)) { | ||
| 185 : | pMB->mode = MODE_INTRA; | ||
| 186 : | pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = | ||
| 187 : | pMB->mvs[3] = zeroMV; | ||
| 188 : | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = | ||
| 189 : | pMB->sad8[3] = 0; | ||
| 190 : | |||
| 191 : | iIntra++; | ||
| 192 : | continue; | ||
| 193 : | } | ||
| 194 : | } | ||
| 195 : | |||
| 196 : | pmv = pMB->pmvs[0]; | ||
| 197 : | if (current->global_flags & XVID_INTER4V) | ||
| 198 : | if ((!(current->global_flags & XVID_LUMIMASKING) || | ||
| 199 : | pMB->dquant == NO_CHANGE)) { | ||
| 200 : | int32_t sad8 = IMV16X16 * current->quant; | ||
| 201 : | |||
| 202 : | if (sad8 < pMB->sad16) | ||
| 203 : | |||
| 204 : | sad8 += pMB->sad8[0] = | ||
| 205 : | SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, | ||
| 206 : | pCurrent, 2 * x, 2 * y, pMB->mv16.x, | ||
| 207 : | pMB->mv16.y, current->motion_flags, | ||
| 208 : | current->quant, current->fcode, pParam, | ||
| 209 : | pMBs, prevMBs, &pMB->mvs[0], | ||
| 210 : | &pMB->pmvs[0]); | ||
| 211 : | |||
| 212 : | if (sad8 < pMB->sad16) | ||
| 213 : | sad8 += pMB->sad8[1] = | ||
| 214 : | SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, | ||
| 215 : | pCurrent, 2 * x + 1, 2 * y, pMB->mv16.x, | ||
| 216 : | pMB->mv16.y, current->motion_flags, | ||
| 217 : | current->quant, current->fcode, pParam, | ||
| 218 : | pMBs, prevMBs, &pMB->mvs[1], | ||
| 219 : | &pMB->pmvs[1]); | ||
| 220 : | |||
| 221 : | if (sad8 < pMB->sad16) | ||
| 222 : | sad8 += pMB->sad8[2] = | ||
| 223 : | SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, | ||
| 224 : | pCurrent, 2 * x, 2 * y + 1, pMB->mv16.x, | ||
| 225 : | pMB->mv16.y, current->motion_flags, | ||
| 226 : | current->quant, current->fcode, pParam, | ||
| 227 : | pMBs, prevMBs, &pMB->mvs[2], | ||
| 228 : | &pMB->pmvs[2]); | ||
| 229 : | |||
| 230 : | if (sad8 < pMB->sad16) | ||
| 231 : | sad8 += pMB->sad8[3] = | ||
| 232 : | SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, | ||
| 233 : | pCurrent, 2 * x + 1, 2 * y + 1, | ||
| 234 : | pMB->mv16.x, pMB->mv16.y, | ||
| 235 : | current->motion_flags, current->quant, | ||
| 236 : | current->fcode, pParam, pMBs, prevMBs, | ||
| 237 : | &pMB->mvs[3], &pMB->pmvs[3]); | ||
| 238 : | |||
| 239 : | /* decide: MODE_INTER or MODE_INTER4V | ||
| 240 : | mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v | ||
| 241 : | */ | ||
| 242 : | |||
| 243 : | if (sad8 < pMB->sad16) { | ||
| 244 : | pMB->mode = MODE_INTER4V; | ||
| 245 : | pMB->sad8[0] *= 4; | ||
| 246 : | pMB->sad8[1] *= 4; | ||
| 247 : | pMB->sad8[2] *= 4; | ||
| 248 : | pMB->sad8[3] *= 4; | ||
| 249 : | continue; | ||
| 250 : | } | ||
| 251 : | |||
| 252 : | } | ||
| 253 : | |||
| 254 : | pMB->mode = MODE_INTER; | ||
| 255 : | pMB->pmvs[0] = pmv; /* pMB->pmvs[1] = pMB->pmvs[2] = pMB->pmvs[3] are not needed for INTER */ | ||
| 256 : | pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; | ||
| 257 : | pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = pMB->sad16; | ||
| 258 : | |||
| 259 : | } | ||
| 260 : | } /* end of x/y loop */ | ||
| 261 : | |||
| 262 : | chl | 1.2 | // fprintf(stderr,"[%d,%d] Full ME %lld ticks \n",arg->id,run,read_counter()-time); |
| 263 : | chl | 1.1 | |
| 264 : | pthread_mutex_lock(&me_mutex); | ||
| 265 : | |||
| 266 : | me_corrqueue--; // the last to finish wakes the others to start correction | ||
| 267 : | me_iIntra += iIntra; | ||
| 268 : | |||
| 269 : | if (me_corrqueue==0) | ||
| 270 : | { | ||
| 271 : | // fprintf(stderr,"[%d,%d] corrqueue %d waking neighbours\n",arg->id,run,me_corrqueue); | ||
| 272 : | pthread_cond_broadcast(&me_corrqueue_cond); | ||
| 273 : | } | ||
| 274 : | |||
| 275 : | // fprintf(stderr,"[%d,%d] wait corrqueue %d init\n",arg->id,run,me_corrqueue); | ||
| 276 : | |||
| 277 : | while (me_corrqueue) | ||
| 278 : | pthread_cond_wait(&me_corrqueue_cond,&me_mutex); | ||
| 279 : | |||
| 280 : | // fprintf(stderr,"[%d,%d] wait corrqueue %d done\n",arg->id,run,me_corrqueue); | ||
| 281 : | |||
| 282 : | chl | 1.2 | // time = read_counter(); |
| 283 : | chl | 1.1 | |
| 284 : | // if (me_iIntra <= iLimit) | ||
| 285 : | // { | ||
| 286 : | // pthread_mutex_unlock(&me_mutex); | ||
| 287 : | |||
| 288 : | if (minx!=0) | ||
| 289 : | for (y=miny; y<maxy; y++) | ||
| 290 : | SMP_correct_pmv(minx, y,iWcount,pMBs); | ||
| 291 : | if (maxx!=iWcount) | ||
| 292 : | for (y=miny; y<maxy; y++) | ||
| 293 : | SMP_correct_pmv(maxx-1,y,iWcount,pMBs); | ||
| 294 : | |||
| 295 : | if (miny!=0) | ||
| 296 : | for (x=minx;x<maxx; x++) | ||
| 297 : | SMP_correct_pmv(x,miny, iWcount,pMBs); | ||
| 298 : | |||
| 299 : | if (maxy!=iHcount) | ||
| 300 : | for (x=minx;x<maxx; x++) | ||
| 301 : | SMP_correct_pmv(x,maxy-1,iWcount,pMBs); | ||
| 302 : | |||
| 303 : | // pthread_mutex_lock(&me_mutex); | ||
| 304 : | // } | ||
| 305 : | |||
| 306 : | chl | 1.2 | // fprintf(stderr,"[%d,%d] Full CORR %lld ticks \n",arg->id,run,read_counter()-time); |
| 307 : | chl | 1.1 | |
| 308 : | me_outqueue--; | ||
| 309 : | |||
| 310 : | if (me_outqueue==0) // the last to finish wakes the master | ||
| 311 : | pthread_cond_signal(&me_outqueue_cond); | ||
| 312 : | |||
| 313 : | // fprintf(stderr,"[%d,%d] wait outqueue %d init\n",arg->id,run,me_outqueue); | ||
| 314 : | |||
| 315 : | // while (me_outqueue) | ||
| 316 : | // pthread_cond_wait(&me_outqueue_cond,&me_mutex); | ||
| 317 : | |||
| 318 : | // fprintf(stderr,"[%d,%d] wait outqueue %d done\n",arg->id,run,me_outqueue); | ||
| 319 : | |||
| 320 : | } /* end of while(1) */ | ||
| 321 : | pthread_mutex_unlock(&me_mutex); | ||
| 322 : | } | ||
| 323 : | |||
| 324 : | bool | ||
| 325 : | SMP_MotionEstimation(MBParam * const pParam, | ||
| 326 : | FRAMEINFO * const current, | ||
| 327 : | FRAMEINFO * const reference, | ||
| 328 : | const IMAGE * const pRefH, | ||
| 329 : | const IMAGE * const pRefV, | ||
| 330 : | const IMAGE * const pRefHV, | ||
| 331 : | const uint32_t iLimit) | ||
| 332 : | { | ||
| 333 : | int i; | ||
| 334 : | static int threadscreated=0; | ||
| 335 : | |||
| 336 : | const int iWcount = pParam->mb_width; | ||
| 337 : | const int iHcount = pParam->mb_height; | ||
| 338 : | |||
| 339 : | static globaldata gdata; | ||
| 340 : | static jobdata jdata[MAXNUMTHREADS]; | ||
| 341 : | static pthread_t me_thread[MAXNUMTHREADS]; | ||
| 342 : | |||
| 343 : | gdata.pParam = pParam; | ||
| 344 : | gdata.reference = reference; | ||
| 345 : | gdata.current = current; | ||
| 346 : | gdata.pRefH = pRefH; | ||
| 347 : | gdata.pRefV = pRefV; | ||
| 348 : | gdata.pRefHV = pRefHV; | ||
| 349 : | // gdata.iLimit = iLimit; | ||
| 350 : | |||
| 351 : | if (sadInit) | ||
| 352 : | (*sadInit) (); | ||
| 353 : | |||
| 354 : | pthread_mutex_lock(&me_mutex); | ||
| 355 : | me_iIntra=0; | ||
| 356 : | me_inqueue=NUMTHREADS; | ||
| 357 : | me_corrqueue=NUMTHREADS; | ||
| 358 : | me_outqueue=NUMTHREADS; | ||
| 359 : | |||
| 360 : | if (!threadscreated) | ||
| 361 : | { | ||
| 362 : | for (i=0;i<NUMTHREADS;i++) { /* split domain into NUMTHREADS parts */ | ||
| 363 : | |||
| 364 : | jdata[i].id = i; | ||
| 365 : | jdata[i].minx = i*iWcount/NUMTHREADS; | ||
| 366 : | jdata[i].maxx = (i+1)*iWcount/NUMTHREADS; | ||
| 367 : | jdata[i].miny = 0; | ||
| 368 : | jdata[i].maxy = iHcount; | ||
| 369 : | jdata[i].gdata = &gdata; | ||
| 370 : | |||
| 371 : | pthread_create(&me_thread[i],NULL, | ||
| 372 : | (void*)SMP_MotionEstimationWorker,(void*)&jdata[i]); | ||
| 373 : | } | ||
| 374 : | threadscreated=1; | ||
| 375 : | } | ||
| 376 : | |||
| 377 : | pthread_cond_broadcast(&me_inqueue_cond); // start working | ||
| 378 : | |||
| 379 : | while (me_outqueue) | ||
| 380 : | pthread_cond_wait(&me_outqueue_cond,&me_mutex); | ||
| 381 : | |||
| 382 : | if (me_iIntra > iLimit) | ||
| 383 : | { | ||
| 384 : | pthread_mutex_unlock(&me_mutex); | ||
| 385 : | return 1; | ||
| 386 : | } | ||
| 387 : | |||
| 388 : | pthread_mutex_unlock(&me_mutex); | ||
| 389 : | return 0; | ||
| 390 : | } | ||
| 391 : | |||
| 392 : | #endif |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |