1 /* 2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 12 #ifndef WIN32 13 # include <unistd.h> 14 #endif 15 #ifdef __APPLE__ 16 #include <mach/mach_init.h> 17 #endif 18 #include "onyxd_int.h" 19 #include "vpx_mem/vpx_mem.h" 20 #include "threading.h" 21 22 #include "loopfilter.h" 23 #include "extend.h" 24 #include "vpx_ports/vpx_timer.h" 25 26 #define MAX_ROWS 256 27 28 extern void vp8_decode_mb_row(VP8D_COMP *pbi, 29 VP8_COMMON *pc, 30 int mb_row, 31 MACROBLOCKD *xd); 32 33 extern void vp8_build_uvmvs(MACROBLOCKD *x, int fullpixel); 34 extern void vp8_decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd); 35 36 void vp8_thread_loop_filter(VP8D_COMP *pbi, MB_ROW_DEC *mbrd, int ithread); 37 38 void vp8_setup_decoding_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd, MB_ROW_DEC *mbrd, int count) 39 { 40 #if CONFIG_MULTITHREAD 41 VP8_COMMON *const pc = & pbi->common; 42 int i, j; 43 44 for (i = 0; i < count; i++) 45 { 46 MACROBLOCKD *mbd = &mbrd[i].mbd; 47 #if CONFIG_RUNTIME_CPU_DETECT 48 mbd->rtcd = xd->rtcd; 49 #endif 50 mbd->subpixel_predict = xd->subpixel_predict; 51 mbd->subpixel_predict8x4 = xd->subpixel_predict8x4; 52 mbd->subpixel_predict8x8 = xd->subpixel_predict8x8; 53 mbd->subpixel_predict16x16 = xd->subpixel_predict16x16; 54 55 mbd->mode_info_context = pc->mi + pc->mode_info_stride * (i + 1); 56 mbd->mode_info_stride = pc->mode_info_stride; 57 58 mbd->frame_type = pc->frame_type; 59 mbd->frames_since_golden = pc->frames_since_golden; 60 mbd->frames_till_alt_ref_frame = pc->frames_till_alt_ref_frame; 61 62 mbd->pre = pc->yv12_fb[pc->lst_fb_idx]; 63 mbd->dst = pc->yv12_fb[pc->new_fb_idx]; 64 65 vp8_setup_block_dptrs(mbd); 66 vp8_build_block_doffsets(mbd); 67 mbd->segmentation_enabled = xd->segmentation_enabled; 68 mbd->mb_segement_abs_delta = xd->mb_segement_abs_delta; 69 vpx_memcpy(mbd->segment_feature_data, xd->segment_feature_data, sizeof(xd->segment_feature_data)); 70 71 mbd->current_bc = &pbi->bc2; 72 73 for (j = 0; j < 25; j++) 74 { 75 mbd->block[j].dequant = xd->block[j].dequant; 76 } 77 } 78 79 for (i=0; i< pc->mb_rows; i++) 80 pbi->current_mb_col[i]=-1; 81 #else 82 (void) pbi; 83 (void) xd; 84 (void) mbrd; 85 (void) count; 86 #endif 87 } 88 89 void vp8_setup_loop_filter_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd, MB_ROW_DEC *mbrd, int count) 90 { 91 #if CONFIG_MULTITHREAD 92 VP8_COMMON *const pc = & pbi->common; 93 int i, j; 94 95 for (i = 0; i < count; i++) 96 { 97 MACROBLOCKD *mbd = &mbrd[i].mbd; 98 //#if CONFIG_RUNTIME_CPU_DETECT 99 // mbd->rtcd = xd->rtcd; 100 //#endif 101 102 //mbd->subpixel_predict = xd->subpixel_predict; 103 //mbd->subpixel_predict8x4 = xd->subpixel_predict8x4; 104 //mbd->subpixel_predict8x8 = xd->subpixel_predict8x8; 105 //mbd->subpixel_predict16x16 = xd->subpixel_predict16x16; 106 107 mbd->mode_info_context = pc->mi + pc->mode_info_stride * (i + 1); 108 mbd->mode_info_stride = pc->mode_info_stride; 109 110 //mbd->frame_type = pc->frame_type; 111 //mbd->frames_since_golden = pc->frames_since_golden; 112 //mbd->frames_till_alt_ref_frame = pc->frames_till_alt_ref_frame; 113 114 //mbd->pre = pc->yv12_fb[pc->lst_fb_idx]; 115 //mbd->dst = pc->yv12_fb[pc->new_fb_idx]; 116 117 //vp8_setup_block_dptrs(mbd); 118 //vp8_build_block_doffsets(mbd); 119 mbd->segmentation_enabled = xd->segmentation_enabled; // 120 mbd->mb_segement_abs_delta = xd->mb_segement_abs_delta; // 121 vpx_memcpy(mbd->segment_feature_data, xd->segment_feature_data, sizeof(xd->segment_feature_data)); // 122 123 //signed char ref_lf_deltas[MAX_REF_LF_DELTAS]; 124 vpx_memcpy(mbd->ref_lf_deltas, xd->ref_lf_deltas, sizeof(xd->ref_lf_deltas)); 125 //signed char mode_lf_deltas[MAX_MODE_LF_DELTAS]; 126 vpx_memcpy(mbd->mode_lf_deltas, xd->mode_lf_deltas, sizeof(xd->mode_lf_deltas)); 127 //unsigned char mode_ref_lf_delta_enabled; 128 //unsigned char mode_ref_lf_delta_update; 129 mbd->mode_ref_lf_delta_enabled = xd->mode_ref_lf_delta_enabled; 130 mbd->mode_ref_lf_delta_update = xd->mode_ref_lf_delta_update; 131 132 //mbd->mbmi.mode = DC_PRED; 133 //mbd->mbmi.uv_mode = DC_PRED; 134 //mbd->current_bc = &pbi->bc2; 135 136 //for (j = 0; j < 25; j++) 137 //{ 138 // mbd->block[j].dequant = xd->block[j].dequant; 139 //} 140 } 141 142 for (i=0; i< pc->mb_rows; i++) 143 pbi->current_mb_col[i]=-1; 144 #else 145 (void) pbi; 146 (void) xd; 147 (void) mbrd; 148 (void) count; 149 #endif 150 } 151 152 THREAD_FUNCTION vp8_thread_decoding_proc(void *p_data) 153 { 154 #if CONFIG_MULTITHREAD 155 int ithread = ((DECODETHREAD_DATA *)p_data)->ithread; 156 VP8D_COMP *pbi = (VP8D_COMP *)(((DECODETHREAD_DATA *)p_data)->ptr1); 157 MB_ROW_DEC *mbrd = (MB_ROW_DEC *)(((DECODETHREAD_DATA *)p_data)->ptr2); 158 ENTROPY_CONTEXT_PLANES mb_row_left_context; 159 160 while (1) 161 { 162 int current_filter_level = 0; 163 164 if (pbi->b_multithreaded_rd == 0) 165 break; 166 167 //if(WaitForSingleObject(pbi->h_event_start_decoding[ithread], INFINITE) == WAIT_OBJECT_0) 168 if (sem_wait(&pbi->h_event_start_decoding[ithread]) == 0) 169 { 170 if (pbi->b_multithreaded_rd == 0) 171 break; 172 else 173 { 174 VP8_COMMON *pc = &pbi->common; 175 MACROBLOCKD *xd = &mbrd->mbd; 176 177 int mb_row; 178 int num_part = 1 << pbi->common.multi_token_partition; 179 volatile int *last_row_current_mb_col; 180 181 for (mb_row = ithread+1; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1)) 182 { 183 int i; 184 int recon_yoffset, recon_uvoffset; 185 int mb_col; 186 int ref_fb_idx = pc->lst_fb_idx; 187 int dst_fb_idx = pc->new_fb_idx; 188 int recon_y_stride = pc->yv12_fb[ref_fb_idx].y_stride; 189 int recon_uv_stride = pc->yv12_fb[ref_fb_idx].uv_stride; 190 191 pbi->mb_row_di[ithread].mb_row = mb_row; 192 pbi->mb_row_di[ithread].mbd.current_bc = &pbi->mbc[mb_row%num_part]; 193 194 last_row_current_mb_col = &pbi->current_mb_col[mb_row -1]; 195 196 recon_yoffset = mb_row * recon_y_stride * 16; 197 recon_uvoffset = mb_row * recon_uv_stride * 8; 198 // reset above block coeffs 199 200 xd->above_context = pc->above_context; 201 xd->left_context = &mb_row_left_context; 202 vpx_memset(&mb_row_left_context, 0, sizeof(mb_row_left_context)); 203 xd->up_available = (mb_row != 0); 204 205 xd->mb_to_top_edge = -((mb_row * 16)) << 3; 206 xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3; 207 208 for (mb_col = 0; mb_col < pc->mb_cols; mb_col++) 209 { 210 if ((mb_col & 7) == 0) 211 { 212 while (mb_col > (*last_row_current_mb_col - 8) && *last_row_current_mb_col != pc->mb_cols - 1) 213 { 214 x86_pause_hint(); 215 thread_sleep(0); 216 } 217 } 218 219 if (xd->mode_info_context->mbmi.mode == SPLITMV || xd->mode_info_context->mbmi.mode == B_PRED) 220 { 221 for (i = 0; i < 16; i++) 222 { 223 BLOCKD *d = &xd->block[i]; 224 vpx_memcpy(&d->bmi, &xd->mode_info_context->bmi[i], sizeof(B_MODE_INFO)); 225 } 226 } 227 228 // Distance of Mb to the various image edges. 229 // These specified to 8th pel as they are always compared to values that are in 1/8th pel units 230 xd->mb_to_left_edge = -((mb_col * 16) << 3); 231 xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3; 232 233 xd->dst.y_buffer = pc->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset; 234 xd->dst.u_buffer = pc->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset; 235 xd->dst.v_buffer = pc->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset; 236 237 xd->left_available = (mb_col != 0); 238 239 // Select the appropriate reference frame for this MB 240 if (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME) 241 ref_fb_idx = pc->lst_fb_idx; 242 else if (xd->mode_info_context->mbmi.ref_frame == GOLDEN_FRAME) 243 ref_fb_idx = pc->gld_fb_idx; 244 else 245 ref_fb_idx = pc->alt_fb_idx; 246 247 xd->pre.y_buffer = pc->yv12_fb[ref_fb_idx].y_buffer + recon_yoffset; 248 xd->pre.u_buffer = pc->yv12_fb[ref_fb_idx].u_buffer + recon_uvoffset; 249 xd->pre.v_buffer = pc->yv12_fb[ref_fb_idx].v_buffer + recon_uvoffset; 250 251 vp8_build_uvmvs(xd, pc->full_pixel); 252 253 vp8_decode_macroblock(pbi, xd); 254 255 recon_yoffset += 16; 256 recon_uvoffset += 8; 257 258 ++xd->mode_info_context; /* next mb */ 259 260 xd->above_context++; 261 262 //pbi->mb_row_di[ithread].current_mb_col = mb_col; 263 pbi->current_mb_col[mb_row] = mb_col; 264 } 265 266 // adjust to the next row of mbs 267 vp8_extend_mb_row( 268 &pc->yv12_fb[dst_fb_idx], 269 xd->dst.y_buffer + 16, xd->dst.u_buffer + 8, xd->dst.v_buffer + 8 270 ); 271 272 ++xd->mode_info_context; /* skip prediction column */ 273 274 // since we have multithread 275 xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count; 276 277 pbi->last_mb_row_decoded = mb_row; 278 279 } 280 } 281 } 282 283 // If |pbi->common.filter_level| is 0 the value can change in-between 284 // the sem_post and the check to call vp8_thread_loop_filter. 285 current_filter_level = pbi->common.filter_level; 286 287 // add this to each frame 288 if ((mbrd->mb_row == pbi->common.mb_rows-1) || ((mbrd->mb_row == pbi->common.mb_rows-2) && (pbi->common.mb_rows % (pbi->decoding_thread_count+1))==1)) 289 { 290 //SetEvent(pbi->h_event_end_decoding); 291 sem_post(&pbi->h_event_end_decoding); 292 } 293 294 if ((pbi->b_multithreaded_lf) && (current_filter_level)) 295 vp8_thread_loop_filter(pbi, mbrd, ithread); 296 297 } 298 #else 299 (void) p_data; 300 #endif 301 302 return 0 ; 303 } 304 305 306 void vp8_thread_loop_filter(VP8D_COMP *pbi, MB_ROW_DEC *mbrd, int ithread) 307 { 308 #if CONFIG_MULTITHREAD 309 310 if (sem_wait(&pbi->h_event_start_lpf[ithread]) == 0) 311 { 312 // if (pbi->b_multithreaded_lf == 0) // we're shutting down ???? 313 // break; 314 // else 315 { 316 VP8_COMMON *cm = &pbi->common; 317 MACROBLOCKD *mbd = &mbrd->mbd; 318 int default_filt_lvl = pbi->common.filter_level; 319 320 YV12_BUFFER_CONFIG *post = cm->frame_to_show; 321 loop_filter_info *lfi = cm->lf_info; 322 //int frame_type = cm->frame_type; 323 324 int mb_row; 325 int mb_col; 326 327 int filter_level; 328 int alt_flt_enabled = mbd->segmentation_enabled; 329 330 int i; 331 unsigned char *y_ptr, *u_ptr, *v_ptr; 332 333 volatile int *last_row_current_mb_col; 334 335 // Set up the buffer pointers 336 y_ptr = post->y_buffer + post->y_stride * 16 * (ithread +1); 337 u_ptr = post->u_buffer + post->uv_stride * 8 * (ithread +1); 338 v_ptr = post->v_buffer + post->uv_stride * 8 * (ithread +1); 339 340 // vp8_filter each macro block 341 for (mb_row = ithread+1; mb_row < cm->mb_rows; mb_row+= (pbi->decoding_thread_count + 1)) 342 { 343 last_row_current_mb_col = &pbi->current_mb_col[mb_row -1]; 344 345 for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) 346 { 347 int Segment = (alt_flt_enabled) ? mbd->mode_info_context->mbmi.segment_id : 0; 348 349 if ((mb_col & 7) == 0) 350 { 351 while (mb_col > (*last_row_current_mb_col-8) && *last_row_current_mb_col != cm->mb_cols - 1) 352 { 353 x86_pause_hint(); 354 thread_sleep(0); 355 } 356 } 357 358 filter_level = pbi->mt_baseline_filter_level[Segment]; 359 360 // Apply any context driven MB level adjustment 361 vp8_adjust_mb_lf_value(mbd, &filter_level); 362 363 if (filter_level) 364 { 365 if (mb_col > 0) 366 cm->lf_mbv(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf); 367 368 if (mbd->mode_info_context->mbmi.dc_diff > 0) 369 cm->lf_bv(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf); 370 371 // don't apply across umv border 372 if (mb_row > 0) 373 cm->lf_mbh(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf); 374 375 if (mbd->mode_info_context->mbmi.dc_diff > 0) 376 cm->lf_bh(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf); 377 } 378 379 y_ptr += 16; 380 u_ptr += 8; 381 v_ptr += 8; 382 383 mbd->mode_info_context++; // step to next MB 384 pbi->current_mb_col[mb_row] = mb_col; 385 } 386 387 mbd->mode_info_context++; // Skip border mb 388 389 y_ptr += post->y_stride * 16 * (pbi->decoding_thread_count + 1) - post->y_width; 390 u_ptr += post->uv_stride * 8 * (pbi->decoding_thread_count + 1) - post->uv_width; 391 v_ptr += post->uv_stride * 8 * (pbi->decoding_thread_count + 1) - post->uv_width; 392 393 mbd->mode_info_context += pbi->decoding_thread_count * mbd->mode_info_stride; // Skip border mb 394 } 395 } 396 } 397 398 // add this to each frame 399 if ((mbrd->mb_row == pbi->common.mb_rows-1) || ((mbrd->mb_row == pbi->common.mb_rows-2) && (pbi->common.mb_rows % (pbi->decoding_thread_count+1))==1)) 400 { 401 sem_post(&pbi->h_event_end_lpf); 402 } 403 #else 404 (void) pbi; 405 #endif 406 } 407 408 void vp8_decoder_create_threads(VP8D_COMP *pbi) 409 { 410 #if CONFIG_MULTITHREAD 411 int core_count = 0; 412 int ithread; 413 414 pbi->b_multithreaded_rd = 0; 415 pbi->b_multithreaded_lf = 0; 416 pbi->allocated_decoding_thread_count = 0; 417 core_count = (pbi->max_threads > 16) ? 16 : pbi->max_threads; 418 419 if (core_count > 1) 420 { 421 pbi->b_multithreaded_rd = 1; 422 pbi->b_multithreaded_lf = 1; // this can be merged with pbi->b_multithreaded_rd ? 423 pbi->decoding_thread_count = core_count -1; 424 425 CHECK_MEM_ERROR(pbi->h_decoding_thread, vpx_malloc(sizeof(pthread_t) * pbi->decoding_thread_count)); 426 CHECK_MEM_ERROR(pbi->h_event_start_decoding, vpx_malloc(sizeof(sem_t) * pbi->decoding_thread_count)); 427 CHECK_MEM_ERROR(pbi->mb_row_di, vpx_memalign(32, sizeof(MB_ROW_DEC) * pbi->decoding_thread_count)); 428 vpx_memset(pbi->mb_row_di, 0, sizeof(MB_ROW_DEC) * pbi->decoding_thread_count); 429 CHECK_MEM_ERROR(pbi->de_thread_data, vpx_malloc(sizeof(DECODETHREAD_DATA) * pbi->decoding_thread_count)); 430 431 CHECK_MEM_ERROR(pbi->current_mb_col, vpx_malloc(sizeof(int) * MAX_ROWS)); // pc->mb_rows)); 432 CHECK_MEM_ERROR(pbi->h_event_start_lpf, vpx_malloc(sizeof(sem_t) * pbi->decoding_thread_count)); 433 434 for (ithread = 0; ithread < pbi->decoding_thread_count; ithread++) 435 { 436 sem_init(&pbi->h_event_start_decoding[ithread], 0, 0); 437 sem_init(&pbi->h_event_start_lpf[ithread], 0, 0); 438 439 pbi->de_thread_data[ithread].ithread = ithread; 440 pbi->de_thread_data[ithread].ptr1 = (void *)pbi; 441 pbi->de_thread_data[ithread].ptr2 = (void *) &pbi->mb_row_di[ithread]; 442 443 pthread_create(&pbi->h_decoding_thread[ithread], 0, vp8_thread_decoding_proc, (&pbi->de_thread_data[ithread])); 444 } 445 446 sem_init(&pbi->h_event_end_decoding, 0, 0); 447 sem_init(&pbi->h_event_end_lpf, 0, 0); 448 449 pbi->allocated_decoding_thread_count = pbi->decoding_thread_count; 450 } 451 452 #else 453 (void) pbi; 454 #endif 455 } 456 457 void vp8_decoder_remove_threads(VP8D_COMP *pbi) 458 { 459 #if CONFIG_MULTITHREAD 460 461 if (pbi->b_multithreaded_lf) 462 { 463 int i; 464 pbi->b_multithreaded_lf = 0; 465 466 for (i = 0; i < pbi->allocated_decoding_thread_count; i++) 467 sem_destroy(&pbi->h_event_start_lpf[i]); 468 469 sem_destroy(&pbi->h_event_end_lpf); 470 } 471 472 //shutdown MB Decoding thread; 473 if (pbi->b_multithreaded_rd) 474 { 475 int i; 476 477 pbi->b_multithreaded_rd = 0; 478 479 // allow all threads to exit 480 for (i = 0; i < pbi->allocated_decoding_thread_count; i++) 481 { 482 sem_post(&pbi->h_event_start_decoding[i]); 483 pthread_join(pbi->h_decoding_thread[i], NULL); 484 } 485 486 for (i = 0; i < pbi->allocated_decoding_thread_count; i++) 487 { 488 sem_destroy(&pbi->h_event_start_decoding[i]); 489 } 490 491 sem_destroy(&pbi->h_event_end_decoding); 492 493 if (pbi->h_decoding_thread) 494 { 495 vpx_free(pbi->h_decoding_thread); 496 pbi->h_decoding_thread = NULL; 497 } 498 499 if (pbi->h_event_start_decoding) 500 { 501 vpx_free(pbi->h_event_start_decoding); 502 pbi->h_event_start_decoding = NULL; 503 } 504 505 if (pbi->h_event_start_lpf) 506 { 507 vpx_free(pbi->h_event_start_lpf); 508 pbi->h_event_start_lpf = NULL; 509 } 510 511 if (pbi->mb_row_di) 512 { 513 vpx_free(pbi->mb_row_di); 514 pbi->mb_row_di = NULL ; 515 } 516 517 if (pbi->de_thread_data) 518 { 519 vpx_free(pbi->de_thread_data); 520 pbi->de_thread_data = NULL; 521 } 522 523 if (pbi->current_mb_col) 524 { 525 vpx_free(pbi->current_mb_col); 526 pbi->current_mb_col = NULL ; 527 } 528 } 529 #else 530 (void) pbi; 531 #endif 532 } 533 534 535 void vp8_start_lfthread(VP8D_COMP *pbi) 536 { 537 #if CONFIG_MULTITHREAD 538 /* 539 memcpy(&pbi->lpfmb, &pbi->mb, sizeof(pbi->mb)); 540 pbi->last_mb_row_decoded = 0; 541 sem_post(&pbi->h_event_start_lpf); 542 */ 543 (void) pbi; 544 #else 545 (void) pbi; 546 #endif 547 } 548 549 void vp8_stop_lfthread(VP8D_COMP *pbi) 550 { 551 #if CONFIG_MULTITHREAD 552 /* 553 struct vpx_usec_timer timer; 554 555 vpx_usec_timer_start(&timer); 556 557 sem_wait(&pbi->h_event_end_lpf); 558 559 vpx_usec_timer_mark(&timer); 560 pbi->time_loop_filtering += vpx_usec_timer_elapsed(&timer); 561 */ 562 (void) pbi; 563 #else 564 (void) pbi; 565 #endif 566 } 567 568 569 void vp8_mtdecode_mb_rows(VP8D_COMP *pbi, 570 MACROBLOCKD *xd) 571 { 572 #if CONFIG_MULTITHREAD 573 int mb_row; 574 VP8_COMMON *pc = &pbi->common; 575 576 int ibc = 0; 577 int num_part = 1 << pbi->common.multi_token_partition; 578 int i; 579 volatile int *last_row_current_mb_col = NULL; 580 581 vp8_setup_decoding_thread_data(pbi, xd, pbi->mb_row_di, pbi->decoding_thread_count); 582 583 for (i = 0; i < pbi->decoding_thread_count; i++) 584 sem_post(&pbi->h_event_start_decoding[i]); 585 586 for (mb_row = 0; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1)) 587 { 588 int i; 589 590 xd->current_bc = &pbi->mbc[mb_row%num_part]; 591 592 //vp8_decode_mb_row(pbi, pc, mb_row, xd); 593 { 594 int i; 595 int recon_yoffset, recon_uvoffset; 596 int mb_col; 597 int ref_fb_idx = pc->lst_fb_idx; 598 int dst_fb_idx = pc->new_fb_idx; 599 int recon_y_stride = pc->yv12_fb[ref_fb_idx].y_stride; 600 int recon_uv_stride = pc->yv12_fb[ref_fb_idx].uv_stride; 601 602 // volatile int *last_row_current_mb_col = NULL; 603 if (mb_row > 0) 604 last_row_current_mb_col = &pbi->current_mb_col[mb_row -1]; 605 606 vpx_memset(&pc->left_context, 0, sizeof(pc->left_context)); 607 recon_yoffset = mb_row * recon_y_stride * 16; 608 recon_uvoffset = mb_row * recon_uv_stride * 8; 609 // reset above block coeffs 610 611 xd->above_context = pc->above_context; 612 xd->up_available = (mb_row != 0); 613 614 xd->mb_to_top_edge = -((mb_row * 16)) << 3; 615 xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3; 616 617 for (mb_col = 0; mb_col < pc->mb_cols; mb_col++) 618 { 619 if ( mb_row > 0 && (mb_col & 7) == 0){ 620 while (mb_col > (*last_row_current_mb_col - 8) && *last_row_current_mb_col != pc->mb_cols - 1) 621 { 622 x86_pause_hint(); 623 thread_sleep(0); 624 } 625 } 626 627 if (xd->mode_info_context->mbmi.mode == SPLITMV || xd->mode_info_context->mbmi.mode == B_PRED) 628 { 629 for (i = 0; i < 16; i++) 630 { 631 BLOCKD *d = &xd->block[i]; 632 vpx_memcpy(&d->bmi, &xd->mode_info_context->bmi[i], sizeof(B_MODE_INFO)); 633 } 634 } 635 636 // Distance of Mb to the various image edges. 637 // These specified to 8th pel as they are always compared to values that are in 1/8th pel units 638 xd->mb_to_left_edge = -((mb_col * 16) << 3); 639 xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3; 640 641 xd->dst.y_buffer = pc->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset; 642 xd->dst.u_buffer = pc->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset; 643 xd->dst.v_buffer = pc->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset; 644 645 xd->left_available = (mb_col != 0); 646 647 // Select the appropriate reference frame for this MB 648 if (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME) 649 ref_fb_idx = pc->lst_fb_idx; 650 else if (xd->mode_info_context->mbmi.ref_frame == GOLDEN_FRAME) 651 ref_fb_idx = pc->gld_fb_idx; 652 else 653 ref_fb_idx = pc->alt_fb_idx; 654 655 xd->pre.y_buffer = pc->yv12_fb[ref_fb_idx].y_buffer + recon_yoffset; 656 xd->pre.u_buffer = pc->yv12_fb[ref_fb_idx].u_buffer + recon_uvoffset; 657 xd->pre.v_buffer = pc->yv12_fb[ref_fb_idx].v_buffer + recon_uvoffset; 658 659 vp8_build_uvmvs(xd, pc->full_pixel); 660 661 vp8_decode_macroblock(pbi, xd); 662 663 recon_yoffset += 16; 664 recon_uvoffset += 8; 665 666 ++xd->mode_info_context; /* next mb */ 667 668 xd->above_context++; 669 670 //pbi->current_mb_col_main = mb_col; 671 pbi->current_mb_col[mb_row] = mb_col; 672 } 673 674 // adjust to the next row of mbs 675 vp8_extend_mb_row( 676 &pc->yv12_fb[dst_fb_idx], 677 xd->dst.y_buffer + 16, xd->dst.u_buffer + 8, xd->dst.v_buffer + 8 678 ); 679 680 ++xd->mode_info_context; /* skip prediction column */ 681 682 pbi->last_mb_row_decoded = mb_row; 683 } 684 xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count; 685 } 686 687 sem_wait(&pbi->h_event_end_decoding); // add back for each frame 688 #else 689 (void) pbi; 690 (void) xd; 691 #endif 692 } 693 694 695 void vp8_mt_loop_filter_frame( VP8D_COMP *pbi) 696 { 697 #if CONFIG_MULTITHREAD 698 VP8_COMMON *cm = &pbi->common; 699 MACROBLOCKD *mbd = &pbi->mb; 700 int default_filt_lvl = pbi->common.filter_level; 701 702 YV12_BUFFER_CONFIG *post = cm->frame_to_show; 703 loop_filter_info *lfi = cm->lf_info; 704 int frame_type = cm->frame_type; 705 706 int mb_row; 707 int mb_col; 708 709 int filter_level; 710 int alt_flt_enabled = mbd->segmentation_enabled; 711 712 int i; 713 unsigned char *y_ptr, *u_ptr, *v_ptr; 714 715 volatile int *last_row_current_mb_col=NULL; 716 717 vp8_setup_loop_filter_thread_data(pbi, mbd, pbi->mb_row_di, pbi->decoding_thread_count); 718 719 mbd->mode_info_context = cm->mi; // Point at base of Mb MODE_INFO list 720 721 // Note the baseline filter values for each segment 722 if (alt_flt_enabled) 723 { 724 for (i = 0; i < MAX_MB_SEGMENTS; i++) 725 { 726 // Abs value 727 if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA) 728 pbi->mt_baseline_filter_level[i] = mbd->segment_feature_data[MB_LVL_ALT_LF][i]; 729 // Delta Value 730 else 731 { 732 pbi->mt_baseline_filter_level[i] = default_filt_lvl + mbd->segment_feature_data[MB_LVL_ALT_LF][i]; 733 pbi->mt_baseline_filter_level[i] = (pbi->mt_baseline_filter_level[i] >= 0) ? ((pbi->mt_baseline_filter_level[i] <= MAX_LOOP_FILTER) ? pbi->mt_baseline_filter_level[i] : MAX_LOOP_FILTER) : 0; // Clamp to valid range 734 } 735 } 736 } 737 else 738 { 739 for (i = 0; i < MAX_MB_SEGMENTS; i++) 740 pbi->mt_baseline_filter_level[i] = default_filt_lvl; 741 } 742 743 // Initialize the loop filter for this frame. 744 if ((cm->last_filter_type != cm->filter_type) || (cm->last_sharpness_level != cm->sharpness_level)) 745 vp8_init_loop_filter(cm); 746 else if (frame_type != cm->last_frame_type) 747 vp8_frame_init_loop_filter(lfi, frame_type); 748 749 for (i = 0; i < pbi->decoding_thread_count; i++) 750 sem_post(&pbi->h_event_start_lpf[i]); 751 // sem_post(&pbi->h_event_start_lpf); 752 753 // Set up the buffer pointers 754 y_ptr = post->y_buffer; 755 u_ptr = post->u_buffer; 756 v_ptr = post->v_buffer; 757 758 // vp8_filter each macro block 759 for (mb_row = 0; mb_row < cm->mb_rows; mb_row+= (pbi->decoding_thread_count + 1)) 760 { 761 if (mb_row > 0) 762 last_row_current_mb_col = &pbi->current_mb_col[mb_row -1]; 763 764 for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) 765 { 766 int Segment = (alt_flt_enabled) ? mbd->mode_info_context->mbmi.segment_id : 0; 767 768 if ( mb_row > 0 && (mb_col & 7) == 0){ 769 // if ( mb_row > 0 ){ 770 while (mb_col > (*last_row_current_mb_col-8) && *last_row_current_mb_col != cm->mb_cols - 1) 771 { 772 x86_pause_hint(); 773 thread_sleep(0); 774 } 775 } 776 777 filter_level = pbi->mt_baseline_filter_level[Segment]; 778 779 // Distance of Mb to the various image edges. 780 // These specified to 8th pel as they are always compared to values that are in 1/8th pel units 781 // Apply any context driven MB level adjustment 782 vp8_adjust_mb_lf_value(mbd, &filter_level); 783 784 if (filter_level) 785 { 786 if (mb_col > 0) 787 cm->lf_mbv(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf); 788 789 if (mbd->mode_info_context->mbmi.dc_diff > 0) 790 cm->lf_bv(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf); 791 792 // don't apply across umv border 793 if (mb_row > 0) 794 cm->lf_mbh(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf); 795 796 if (mbd->mode_info_context->mbmi.dc_diff > 0) 797 cm->lf_bh(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf); 798 } 799 800 y_ptr += 16; 801 u_ptr += 8; 802 v_ptr += 8; 803 804 mbd->mode_info_context++; // step to next MB 805 pbi->current_mb_col[mb_row] = mb_col; 806 } 807 mbd->mode_info_context++; // Skip border mb 808 809 //update for multi-thread 810 y_ptr += post->y_stride * 16 * (pbi->decoding_thread_count + 1) - post->y_width; 811 u_ptr += post->uv_stride * 8 * (pbi->decoding_thread_count + 1) - post->uv_width; 812 v_ptr += post->uv_stride * 8 * (pbi->decoding_thread_count + 1) - post->uv_width; 813 mbd->mode_info_context += pbi->decoding_thread_count * mbd->mode_info_stride; 814 } 815 816 sem_wait(&pbi->h_event_end_lpf); 817 #else 818 (void) pbi; 819 #endif 820 } 821