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 #if !defined(WIN32) && CONFIG_OS_SUPPORT == 1 13 # include <unistd.h> 14 #endif 15 #include "onyxd_int.h" 16 #include "vpx_mem/vpx_mem.h" 17 #include "vp8/common/threading.h" 18 19 #include "vp8/common/loopfilter.h" 20 #include "vp8/common/extend.h" 21 #include "vpx_ports/vpx_timer.h" 22 #include "detokenize.h" 23 #include "vp8/common/reconinter.h" 24 #include "reconintra_mt.h" 25 26 extern void mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd); 27 extern void clamp_mvs(MACROBLOCKD *xd); 28 extern void vp8_build_uvmvs(MACROBLOCKD *x, int fullpixel); 29 30 #if CONFIG_RUNTIME_CPU_DETECT 31 #define RTCD_VTABLE(x) (&(pbi)->common.rtcd.x) 32 #else 33 #define RTCD_VTABLE(x) NULL 34 #endif 35 36 static void setup_decoding_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd, MB_ROW_DEC *mbrd, int count) 37 { 38 VP8_COMMON *const pc = & pbi->common; 39 int i, j; 40 41 for (i = 0; i < count; i++) 42 { 43 MACROBLOCKD *mbd = &mbrd[i].mbd; 44 #if CONFIG_RUNTIME_CPU_DETECT 45 mbd->rtcd = xd->rtcd; 46 #endif 47 mbd->subpixel_predict = xd->subpixel_predict; 48 mbd->subpixel_predict8x4 = xd->subpixel_predict8x4; 49 mbd->subpixel_predict8x8 = xd->subpixel_predict8x8; 50 mbd->subpixel_predict16x16 = xd->subpixel_predict16x16; 51 52 mbd->mode_info_context = pc->mi + pc->mode_info_stride * (i + 1); 53 mbd->mode_info_stride = pc->mode_info_stride; 54 55 mbd->frame_type = pc->frame_type; 56 mbd->frames_since_golden = pc->frames_since_golden; 57 mbd->frames_till_alt_ref_frame = pc->frames_till_alt_ref_frame; 58 59 mbd->pre = pc->yv12_fb[pc->lst_fb_idx]; 60 mbd->dst = pc->yv12_fb[pc->new_fb_idx]; 61 62 vp8_setup_block_dptrs(mbd); 63 vp8_build_block_doffsets(mbd); 64 mbd->segmentation_enabled = xd->segmentation_enabled; 65 mbd->mb_segement_abs_delta = xd->mb_segement_abs_delta; 66 vpx_memcpy(mbd->segment_feature_data, xd->segment_feature_data, sizeof(xd->segment_feature_data)); 67 68 /*signed char ref_lf_deltas[MAX_REF_LF_DELTAS];*/ 69 vpx_memcpy(mbd->ref_lf_deltas, xd->ref_lf_deltas, sizeof(xd->ref_lf_deltas)); 70 /*signed char mode_lf_deltas[MAX_MODE_LF_DELTAS];*/ 71 vpx_memcpy(mbd->mode_lf_deltas, xd->mode_lf_deltas, sizeof(xd->mode_lf_deltas)); 72 /*unsigned char mode_ref_lf_delta_enabled; 73 unsigned char mode_ref_lf_delta_update;*/ 74 mbd->mode_ref_lf_delta_enabled = xd->mode_ref_lf_delta_enabled; 75 mbd->mode_ref_lf_delta_update = xd->mode_ref_lf_delta_update; 76 77 mbd->current_bc = &pbi->bc2; 78 79 for (j = 0; j < 25; j++) 80 { 81 mbd->block[j].dequant = xd->block[j].dequant; 82 } 83 } 84 85 for (i=0; i< pc->mb_rows; i++) 86 pbi->mt_current_mb_col[i]=-1; 87 } 88 89 90 static void decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col) 91 { 92 int eobtotal = 0; 93 int i, do_clamp = xd->mode_info_context->mbmi.need_to_clamp_mvs; 94 VP8_COMMON *pc = &pbi->common; 95 96 if (xd->mode_info_context->mbmi.mb_skip_coeff) 97 { 98 vp8_reset_mb_tokens_context(xd); 99 } 100 else 101 { 102 eobtotal = vp8_decode_mb_tokens(pbi, xd); 103 } 104 105 /* Perform temporary clamping of the MV to be used for prediction */ 106 if (do_clamp) 107 { 108 clamp_mvs(xd); 109 } 110 111 xd->mode_info_context->mbmi.dc_diff = 1; 112 113 if (xd->mode_info_context->mbmi.mode != B_PRED && xd->mode_info_context->mbmi.mode != SPLITMV && eobtotal == 0) 114 { 115 xd->mode_info_context->mbmi.dc_diff = 0; 116 117 /*mt_skip_recon_mb(pbi, xd, mb_row, mb_col);*/ 118 if (xd->frame_type == KEY_FRAME || xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) 119 { 120 vp8mt_build_intra_predictors_mbuv_s(pbi, xd, mb_row, mb_col); 121 vp8mt_build_intra_predictors_mby_s(pbi, xd, mb_row, mb_col); 122 } 123 else 124 { 125 vp8_build_inter_predictors_mb_s(xd); 126 } 127 return; 128 } 129 130 if (xd->segmentation_enabled) 131 mb_init_dequantizer(pbi, xd); 132 133 /* do prediction */ 134 if (xd->frame_type == KEY_FRAME || xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) 135 { 136 vp8mt_build_intra_predictors_mbuv(pbi, xd, mb_row, mb_col); 137 138 if (xd->mode_info_context->mbmi.mode != B_PRED) 139 { 140 vp8mt_build_intra_predictors_mby(pbi, xd, mb_row, mb_col); 141 } else { 142 vp8mt_intra_prediction_down_copy(pbi, xd, mb_row, mb_col); 143 } 144 } 145 else 146 { 147 vp8_build_inter_predictors_mb(xd); 148 } 149 150 /* dequantization and idct */ 151 if (xd->mode_info_context->mbmi.mode != B_PRED && xd->mode_info_context->mbmi.mode != SPLITMV) 152 { 153 BLOCKD *b = &xd->block[24]; 154 DEQUANT_INVOKE(&pbi->dequant, block)(b); 155 156 /* do 2nd order transform on the dc block */ 157 if (xd->eobs[24] > 1) 158 { 159 IDCT_INVOKE(RTCD_VTABLE(idct), iwalsh16)(&b->dqcoeff[0], b->diff); 160 ((int *)b->qcoeff)[0] = 0; 161 ((int *)b->qcoeff)[1] = 0; 162 ((int *)b->qcoeff)[2] = 0; 163 ((int *)b->qcoeff)[3] = 0; 164 ((int *)b->qcoeff)[4] = 0; 165 ((int *)b->qcoeff)[5] = 0; 166 ((int *)b->qcoeff)[6] = 0; 167 ((int *)b->qcoeff)[7] = 0; 168 } 169 else 170 { 171 IDCT_INVOKE(RTCD_VTABLE(idct), iwalsh1)(&b->dqcoeff[0], b->diff); 172 ((int *)b->qcoeff)[0] = 0; 173 } 174 175 DEQUANT_INVOKE (&pbi->dequant, dc_idct_add_y_block) 176 (xd->qcoeff, xd->block[0].dequant, 177 xd->predictor, xd->dst.y_buffer, 178 xd->dst.y_stride, xd->eobs, xd->block[24].diff); 179 } 180 else if ((xd->frame_type == KEY_FRAME || xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) && xd->mode_info_context->mbmi.mode == B_PRED) 181 { 182 for (i = 0; i < 16; i++) 183 { 184 BLOCKD *b = &xd->block[i]; 185 vp8mt_predict_intra4x4(pbi, xd, b->bmi.mode, b->predictor, mb_row, mb_col, i); 186 187 if (xd->eobs[i] > 1) 188 { 189 DEQUANT_INVOKE(&pbi->dequant, idct_add) 190 (b->qcoeff, b->dequant, b->predictor, 191 *(b->base_dst) + b->dst, 16, b->dst_stride); 192 } 193 else 194 { 195 IDCT_INVOKE(RTCD_VTABLE(idct), idct1_scalar_add) 196 (b->qcoeff[0] * b->dequant[0], b->predictor, 197 *(b->base_dst) + b->dst, 16, b->dst_stride); 198 ((int *)b->qcoeff)[0] = 0; 199 } 200 } 201 } 202 else 203 { 204 DEQUANT_INVOKE (&pbi->dequant, idct_add_y_block) 205 (xd->qcoeff, xd->block[0].dequant, 206 xd->predictor, xd->dst.y_buffer, 207 xd->dst.y_stride, xd->eobs); 208 } 209 210 DEQUANT_INVOKE (&pbi->dequant, idct_add_uv_block) 211 (xd->qcoeff+16*16, xd->block[16].dequant, 212 xd->predictor+16*16, xd->dst.u_buffer, xd->dst.v_buffer, 213 xd->dst.uv_stride, xd->eobs+16); 214 } 215 216 217 static THREAD_FUNCTION thread_decoding_proc(void *p_data) 218 { 219 int ithread = ((DECODETHREAD_DATA *)p_data)->ithread; 220 VP8D_COMP *pbi = (VP8D_COMP *)(((DECODETHREAD_DATA *)p_data)->ptr1); 221 MB_ROW_DEC *mbrd = (MB_ROW_DEC *)(((DECODETHREAD_DATA *)p_data)->ptr2); 222 ENTROPY_CONTEXT_PLANES mb_row_left_context; 223 224 while (1) 225 { 226 if (pbi->b_multithreaded_rd == 0) 227 break; 228 229 /*if(WaitForSingleObject(pbi->h_event_start_decoding[ithread], INFINITE) == WAIT_OBJECT_0)*/ 230 if (sem_wait(&pbi->h_event_start_decoding[ithread]) == 0) 231 { 232 if (pbi->b_multithreaded_rd == 0) 233 break; 234 else 235 { 236 VP8_COMMON *pc = &pbi->common; 237 MACROBLOCKD *xd = &mbrd->mbd; 238 239 int mb_row; 240 int num_part = 1 << pbi->common.multi_token_partition; 241 volatile int *last_row_current_mb_col; 242 int nsync = pbi->sync_range; 243 244 for (mb_row = ithread+1; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1)) 245 { 246 int i; 247 int recon_yoffset, recon_uvoffset; 248 int mb_col; 249 int ref_fb_idx = pc->lst_fb_idx; 250 int dst_fb_idx = pc->new_fb_idx; 251 int recon_y_stride = pc->yv12_fb[ref_fb_idx].y_stride; 252 int recon_uv_stride = pc->yv12_fb[ref_fb_idx].uv_stride; 253 254 int filter_level; 255 loop_filter_info *lfi = pc->lf_info; 256 int alt_flt_enabled = xd->segmentation_enabled; 257 int Segment; 258 259 pbi->mb_row_di[ithread].mb_row = mb_row; 260 pbi->mb_row_di[ithread].mbd.current_bc = &pbi->mbc[mb_row%num_part]; 261 262 last_row_current_mb_col = &pbi->mt_current_mb_col[mb_row -1]; 263 264 recon_yoffset = mb_row * recon_y_stride * 16; 265 recon_uvoffset = mb_row * recon_uv_stride * 8; 266 /* reset above block coeffs */ 267 268 xd->above_context = pc->above_context; 269 xd->left_context = &mb_row_left_context; 270 vpx_memset(&mb_row_left_context, 0, sizeof(mb_row_left_context)); 271 xd->up_available = (mb_row != 0); 272 273 xd->mb_to_top_edge = -((mb_row * 16)) << 3; 274 xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3; 275 276 for (mb_col = 0; mb_col < pc->mb_cols; mb_col++) 277 { 278 if ((mb_col & (nsync-1)) == 0) 279 { 280 while (mb_col > (*last_row_current_mb_col - nsync) && *last_row_current_mb_col != pc->mb_cols - 1) 281 { 282 x86_pause_hint(); 283 thread_sleep(0); 284 } 285 } 286 287 if (xd->mode_info_context->mbmi.mode == SPLITMV || xd->mode_info_context->mbmi.mode == B_PRED) 288 { 289 for (i = 0; i < 16; i++) 290 { 291 BLOCKD *d = &xd->block[i]; 292 vpx_memcpy(&d->bmi, &xd->mode_info_context->bmi[i], sizeof(B_MODE_INFO)); 293 } 294 } 295 296 /* Distance of Mb to the various image edges. 297 * These are specified to 8th pel as they are always compared to values that are in 1/8th pel units 298 */ 299 xd->mb_to_left_edge = -((mb_col * 16) << 3); 300 xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3; 301 302 xd->dst.y_buffer = pc->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset; 303 xd->dst.u_buffer = pc->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset; 304 xd->dst.v_buffer = pc->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset; 305 306 xd->left_available = (mb_col != 0); 307 308 /* Select the appropriate reference frame for this MB */ 309 if (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME) 310 ref_fb_idx = pc->lst_fb_idx; 311 else if (xd->mode_info_context->mbmi.ref_frame == GOLDEN_FRAME) 312 ref_fb_idx = pc->gld_fb_idx; 313 else 314 ref_fb_idx = pc->alt_fb_idx; 315 316 xd->pre.y_buffer = pc->yv12_fb[ref_fb_idx].y_buffer + recon_yoffset; 317 xd->pre.u_buffer = pc->yv12_fb[ref_fb_idx].u_buffer + recon_uvoffset; 318 xd->pre.v_buffer = pc->yv12_fb[ref_fb_idx].v_buffer + recon_uvoffset; 319 320 vp8_build_uvmvs(xd, pc->full_pixel); 321 decode_macroblock(pbi, xd, mb_row, mb_col); 322 323 if (pbi->common.filter_level) 324 { 325 if( mb_row != pc->mb_rows-1 ) 326 { 327 /* Save decoded MB last row data for next-row decoding */ 328 vpx_memcpy((pbi->mt_yabove_row[mb_row + 1] + 32 + mb_col*16), (xd->dst.y_buffer + 15 * recon_y_stride), 16); 329 vpx_memcpy((pbi->mt_uabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.u_buffer + 7 * recon_uv_stride), 8); 330 vpx_memcpy((pbi->mt_vabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.v_buffer + 7 * recon_uv_stride), 8); 331 } 332 333 /* save left_col for next MB decoding */ 334 if(mb_col != pc->mb_cols-1) 335 { 336 MODE_INFO *next = xd->mode_info_context +1; 337 338 if (xd->frame_type == KEY_FRAME || next->mbmi.ref_frame == INTRA_FRAME) 339 { 340 for (i = 0; i < 16; i++) 341 pbi->mt_yleft_col[mb_row][i] = xd->dst.y_buffer [i* recon_y_stride + 15]; 342 for (i = 0; i < 8; i++) 343 { 344 pbi->mt_uleft_col[mb_row][i] = xd->dst.u_buffer [i* recon_uv_stride + 7]; 345 pbi->mt_vleft_col[mb_row][i] = xd->dst.v_buffer [i* recon_uv_stride + 7]; 346 } 347 } 348 } 349 350 /* update loopfilter info */ 351 Segment = (alt_flt_enabled) ? xd->mode_info_context->mbmi.segment_id : 0; 352 filter_level = pbi->mt_baseline_filter_level[Segment]; 353 /* Distance of Mb to the various image edges. 354 * These are specified to 8th pel as they are always compared to values that are in 1/8th pel units 355 * Apply any context driven MB level adjustment 356 */ 357 filter_level = vp8_adjust_mb_lf_value(xd, filter_level); 358 359 /* loopfilter on this macroblock. */ 360 if (filter_level) 361 { 362 if (mb_col > 0) 363 pc->lf_mbv(xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi[filter_level], pc->simpler_lpf); 364 365 if (xd->mode_info_context->mbmi.dc_diff > 0) 366 pc->lf_bv(xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi[filter_level], pc->simpler_lpf); 367 368 /* don't apply across umv border */ 369 if (mb_row > 0) 370 pc->lf_mbh(xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi[filter_level], pc->simpler_lpf); 371 372 if (xd->mode_info_context->mbmi.dc_diff > 0) 373 pc->lf_bh(xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi[filter_level], pc->simpler_lpf); 374 } 375 } 376 377 recon_yoffset += 16; 378 recon_uvoffset += 8; 379 380 ++xd->mode_info_context; /* next mb */ 381 382 xd->above_context++; 383 384 /*pbi->mb_row_di[ithread].current_mb_col = mb_col;*/ 385 pbi->mt_current_mb_col[mb_row] = mb_col; 386 } 387 388 /* adjust to the next row of mbs */ 389 if (pbi->common.filter_level) 390 { 391 if(mb_row != pc->mb_rows-1) 392 { 393 int lasty = pc->yv12_fb[ref_fb_idx].y_width + VP8BORDERINPIXELS; 394 int lastuv = (pc->yv12_fb[ref_fb_idx].y_width>>1) + (VP8BORDERINPIXELS>>1); 395 396 for (i = 0; i < 4; i++) 397 { 398 pbi->mt_yabove_row[mb_row +1][lasty + i] = pbi->mt_yabove_row[mb_row +1][lasty -1]; 399 pbi->mt_uabove_row[mb_row +1][lastuv + i] = pbi->mt_uabove_row[mb_row +1][lastuv -1]; 400 pbi->mt_vabove_row[mb_row +1][lastuv + i] = pbi->mt_vabove_row[mb_row +1][lastuv -1]; 401 } 402 } 403 } else 404 vp8_extend_mb_row(&pc->yv12_fb[dst_fb_idx], xd->dst.y_buffer + 16, xd->dst.u_buffer + 8, xd->dst.v_buffer + 8); 405 406 ++xd->mode_info_context; /* skip prediction column */ 407 408 /* since we have multithread */ 409 xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count; 410 } 411 } 412 } 413 /* add this to each frame */ 414 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)) 415 { 416 /*SetEvent(pbi->h_event_end_decoding);*/ 417 sem_post(&pbi->h_event_end_decoding); 418 } 419 } 420 421 return 0 ; 422 } 423 424 425 void vp8_decoder_create_threads(VP8D_COMP *pbi) 426 { 427 int core_count = 0; 428 int ithread; 429 430 pbi->b_multithreaded_rd = 0; 431 pbi->allocated_decoding_thread_count = 0; 432 core_count = (pbi->max_threads > 16) ? 16 : pbi->max_threads; 433 434 if (core_count > 1) 435 { 436 pbi->b_multithreaded_rd = 1; 437 pbi->decoding_thread_count = core_count -1; 438 439 CHECK_MEM_ERROR(pbi->h_decoding_thread, vpx_malloc(sizeof(pthread_t) * pbi->decoding_thread_count)); 440 CHECK_MEM_ERROR(pbi->h_event_start_decoding, vpx_malloc(sizeof(sem_t) * pbi->decoding_thread_count)); 441 CHECK_MEM_ERROR(pbi->mb_row_di, vpx_memalign(32, sizeof(MB_ROW_DEC) * pbi->decoding_thread_count)); 442 vpx_memset(pbi->mb_row_di, 0, sizeof(MB_ROW_DEC) * pbi->decoding_thread_count); 443 CHECK_MEM_ERROR(pbi->de_thread_data, vpx_malloc(sizeof(DECODETHREAD_DATA) * pbi->decoding_thread_count)); 444 445 for (ithread = 0; ithread < pbi->decoding_thread_count; ithread++) 446 { 447 sem_init(&pbi->h_event_start_decoding[ithread], 0, 0); 448 449 pbi->de_thread_data[ithread].ithread = ithread; 450 pbi->de_thread_data[ithread].ptr1 = (void *)pbi; 451 pbi->de_thread_data[ithread].ptr2 = (void *) &pbi->mb_row_di[ithread]; 452 453 pthread_create(&pbi->h_decoding_thread[ithread], 0, thread_decoding_proc, (&pbi->de_thread_data[ithread])); 454 } 455 456 sem_init(&pbi->h_event_end_decoding, 0, 0); 457 458 pbi->allocated_decoding_thread_count = pbi->decoding_thread_count; 459 } 460 } 461 462 463 void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows) 464 { 465 VP8_COMMON *const pc = & pbi->common; 466 int i; 467 468 if (pbi->b_multithreaded_rd) 469 { 470 vpx_free(pbi->mt_current_mb_col); 471 pbi->mt_current_mb_col = NULL ; 472 473 /* Free above_row buffers. */ 474 if (pbi->mt_yabove_row) 475 { 476 for (i=0; i< mb_rows; i++) 477 { 478 vpx_free(pbi->mt_yabove_row[i]); 479 pbi->mt_yabove_row[i] = NULL ; 480 } 481 vpx_free(pbi->mt_yabove_row); 482 pbi->mt_yabove_row = NULL ; 483 } 484 485 if (pbi->mt_uabove_row) 486 { 487 for (i=0; i< mb_rows; i++) 488 { 489 vpx_free(pbi->mt_uabove_row[i]); 490 pbi->mt_uabove_row[i] = NULL ; 491 } 492 vpx_free(pbi->mt_uabove_row); 493 pbi->mt_uabove_row = NULL ; 494 } 495 496 if (pbi->mt_vabove_row) 497 { 498 for (i=0; i< mb_rows; i++) 499 { 500 vpx_free(pbi->mt_vabove_row[i]); 501 pbi->mt_vabove_row[i] = NULL ; 502 } 503 vpx_free(pbi->mt_vabove_row); 504 pbi->mt_vabove_row = NULL ; 505 } 506 507 /* Free left_col buffers. */ 508 if (pbi->mt_yleft_col) 509 { 510 for (i=0; i< mb_rows; i++) 511 { 512 vpx_free(pbi->mt_yleft_col[i]); 513 pbi->mt_yleft_col[i] = NULL ; 514 } 515 vpx_free(pbi->mt_yleft_col); 516 pbi->mt_yleft_col = NULL ; 517 } 518 519 if (pbi->mt_uleft_col) 520 { 521 for (i=0; i< mb_rows; i++) 522 { 523 vpx_free(pbi->mt_uleft_col[i]); 524 pbi->mt_uleft_col[i] = NULL ; 525 } 526 vpx_free(pbi->mt_uleft_col); 527 pbi->mt_uleft_col = NULL ; 528 } 529 530 if (pbi->mt_vleft_col) 531 { 532 for (i=0; i< mb_rows; i++) 533 { 534 vpx_free(pbi->mt_vleft_col[i]); 535 pbi->mt_vleft_col[i] = NULL ; 536 } 537 vpx_free(pbi->mt_vleft_col); 538 pbi->mt_vleft_col = NULL ; 539 } 540 } 541 } 542 543 544 void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows) 545 { 546 VP8_COMMON *const pc = & pbi->common; 547 int i; 548 int uv_width; 549 550 if (pbi->b_multithreaded_rd) 551 { 552 vp8mt_de_alloc_temp_buffers(pbi, prev_mb_rows); 553 554 /* our internal buffers are always multiples of 16 */ 555 if ((width & 0xf) != 0) 556 width += 16 - (width & 0xf); 557 558 if (width < 640) pbi->sync_range = 1; 559 else if (width <= 1280) pbi->sync_range = 8; 560 else if (width <= 2560) pbi->sync_range =16; 561 else pbi->sync_range = 32; 562 563 uv_width = width >>1; 564 565 /* Allocate an int for each mb row. */ 566 CHECK_MEM_ERROR(pbi->mt_current_mb_col, vpx_malloc(sizeof(int) * pc->mb_rows)); 567 568 /* Allocate memory for above_row buffers. */ 569 CHECK_MEM_ERROR(pbi->mt_yabove_row, vpx_malloc(sizeof(unsigned char *) * pc->mb_rows)); 570 for (i=0; i< pc->mb_rows; i++) 571 CHECK_MEM_ERROR(pbi->mt_yabove_row[i], vpx_calloc(sizeof(unsigned char) * (width + (VP8BORDERINPIXELS<<1)), 1)); 572 573 CHECK_MEM_ERROR(pbi->mt_uabove_row, vpx_malloc(sizeof(unsigned char *) * pc->mb_rows)); 574 for (i=0; i< pc->mb_rows; i++) 575 CHECK_MEM_ERROR(pbi->mt_uabove_row[i], vpx_calloc(sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS), 1)); 576 577 CHECK_MEM_ERROR(pbi->mt_vabove_row, vpx_malloc(sizeof(unsigned char *) * pc->mb_rows)); 578 for (i=0; i< pc->mb_rows; i++) 579 CHECK_MEM_ERROR(pbi->mt_vabove_row[i], vpx_calloc(sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS), 1)); 580 581 /* Allocate memory for left_col buffers. */ 582 CHECK_MEM_ERROR(pbi->mt_yleft_col, vpx_malloc(sizeof(unsigned char *) * pc->mb_rows)); 583 for (i=0; i< pc->mb_rows; i++) 584 CHECK_MEM_ERROR(pbi->mt_yleft_col[i], vpx_calloc(sizeof(unsigned char) * 16, 1)); 585 586 CHECK_MEM_ERROR(pbi->mt_uleft_col, vpx_malloc(sizeof(unsigned char *) * pc->mb_rows)); 587 for (i=0; i< pc->mb_rows; i++) 588 CHECK_MEM_ERROR(pbi->mt_uleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1)); 589 590 CHECK_MEM_ERROR(pbi->mt_vleft_col, vpx_malloc(sizeof(unsigned char *) * pc->mb_rows)); 591 for (i=0; i< pc->mb_rows; i++) 592 CHECK_MEM_ERROR(pbi->mt_vleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1)); 593 } 594 } 595 596 597 void vp8_decoder_remove_threads(VP8D_COMP *pbi) 598 { 599 /* shutdown MB Decoding thread; */ 600 if (pbi->b_multithreaded_rd) 601 { 602 int i; 603 604 pbi->b_multithreaded_rd = 0; 605 606 /* allow all threads to exit */ 607 for (i = 0; i < pbi->allocated_decoding_thread_count; i++) 608 { 609 sem_post(&pbi->h_event_start_decoding[i]); 610 pthread_join(pbi->h_decoding_thread[i], NULL); 611 } 612 613 for (i = 0; i < pbi->allocated_decoding_thread_count; i++) 614 { 615 sem_destroy(&pbi->h_event_start_decoding[i]); 616 } 617 618 sem_destroy(&pbi->h_event_end_decoding); 619 620 vpx_free(pbi->h_decoding_thread); 621 pbi->h_decoding_thread = NULL; 622 623 vpx_free(pbi->h_event_start_decoding); 624 pbi->h_event_start_decoding = NULL; 625 626 vpx_free(pbi->mb_row_di); 627 pbi->mb_row_di = NULL ; 628 629 vpx_free(pbi->de_thread_data); 630 pbi->de_thread_data = NULL; 631 } 632 } 633 634 635 static void lpf_init( VP8D_COMP *pbi, int default_filt_lvl) 636 { 637 VP8_COMMON *cm = &pbi->common; 638 MACROBLOCKD *mbd = &pbi->mb; 639 /*YV12_BUFFER_CONFIG *post = &cm->new_frame;*/ /*frame_to_show;*/ 640 loop_filter_info *lfi = cm->lf_info; 641 FRAME_TYPE frame_type = cm->frame_type; 642 643 /*int mb_row; 644 int mb_col; 645 int baseline_filter_level[MAX_MB_SEGMENTS];*/ 646 int alt_flt_enabled = mbd->segmentation_enabled; 647 648 int i; 649 /*unsigned char *y_ptr, *u_ptr, *v_ptr;*/ 650 651 /* Note the baseline filter values for each segment */ 652 if (alt_flt_enabled) 653 { 654 for (i = 0; i < MAX_MB_SEGMENTS; i++) 655 { 656 /* Abs value */ 657 if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA) 658 pbi->mt_baseline_filter_level[i] = mbd->segment_feature_data[MB_LVL_ALT_LF][i]; 659 /* Delta Value */ 660 else 661 { 662 pbi->mt_baseline_filter_level[i] = default_filt_lvl + mbd->segment_feature_data[MB_LVL_ALT_LF][i]; 663 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 */ 664 } 665 } 666 } 667 else 668 { 669 for (i = 0; i < MAX_MB_SEGMENTS; i++) 670 pbi->mt_baseline_filter_level[i] = default_filt_lvl; 671 } 672 673 /* Initialize the loop filter for this frame. */ 674 if ((cm->last_filter_type != cm->filter_type) || (cm->last_sharpness_level != cm->sharpness_level)) 675 vp8_init_loop_filter(cm); 676 else if (frame_type != cm->last_frame_type) 677 vp8_frame_init_loop_filter(lfi, frame_type); 678 } 679 680 681 void vp8mt_decode_mb_rows( VP8D_COMP *pbi, MACROBLOCKD *xd) 682 { 683 int mb_row; 684 VP8_COMMON *pc = &pbi->common; 685 686 int ibc = 0; 687 int num_part = 1 << pbi->common.multi_token_partition; 688 int i; 689 volatile int *last_row_current_mb_col = NULL; 690 int nsync = pbi->sync_range; 691 692 int filter_level; 693 loop_filter_info *lfi = pc->lf_info; 694 int alt_flt_enabled = xd->segmentation_enabled; 695 int Segment; 696 697 if(pbi->common.filter_level) 698 { 699 /* Set above_row buffer to 127 for decoding first MB row */ 700 vpx_memset(pbi->mt_yabove_row[0] + VP8BORDERINPIXELS-1, 127, pc->yv12_fb[pc->lst_fb_idx].y_width + 5); 701 vpx_memset(pbi->mt_uabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (pc->yv12_fb[pc->lst_fb_idx].y_width>>1) +5); 702 vpx_memset(pbi->mt_vabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (pc->yv12_fb[pc->lst_fb_idx].y_width>>1) +5); 703 704 for (i=1; i<pc->mb_rows; i++) 705 { 706 vpx_memset(pbi->mt_yabove_row[i] + VP8BORDERINPIXELS-1, (unsigned char)129, 1); 707 vpx_memset(pbi->mt_uabove_row[i] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1); 708 vpx_memset(pbi->mt_vabove_row[i] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1); 709 } 710 711 /* Set left_col to 129 initially */ 712 for (i=0; i<pc->mb_rows; i++) 713 { 714 vpx_memset(pbi->mt_yleft_col[i], (unsigned char)129, 16); 715 vpx_memset(pbi->mt_uleft_col[i], (unsigned char)129, 8); 716 vpx_memset(pbi->mt_vleft_col[i], (unsigned char)129, 8); 717 } 718 lpf_init(pbi, pc->filter_level); 719 } 720 721 setup_decoding_thread_data(pbi, xd, pbi->mb_row_di, pbi->decoding_thread_count); 722 723 for (i = 0; i < pbi->decoding_thread_count; i++) 724 sem_post(&pbi->h_event_start_decoding[i]); 725 726 for (mb_row = 0; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1)) 727 { 728 729 xd->current_bc = &pbi->mbc[mb_row%num_part]; 730 731 /* vp8_decode_mb_row(pbi, pc, mb_row, xd); */ 732 { 733 int i; 734 int recon_yoffset, recon_uvoffset; 735 int mb_col; 736 int ref_fb_idx = pc->lst_fb_idx; 737 int dst_fb_idx = pc->new_fb_idx; 738 int recon_y_stride = pc->yv12_fb[ref_fb_idx].y_stride; 739 int recon_uv_stride = pc->yv12_fb[ref_fb_idx].uv_stride; 740 741 /* volatile int *last_row_current_mb_col = NULL; */ 742 if (mb_row > 0) 743 last_row_current_mb_col = &pbi->mt_current_mb_col[mb_row -1]; 744 745 vpx_memset(&pc->left_context, 0, sizeof(pc->left_context)); 746 recon_yoffset = mb_row * recon_y_stride * 16; 747 recon_uvoffset = mb_row * recon_uv_stride * 8; 748 /* reset above block coeffs */ 749 750 xd->above_context = pc->above_context; 751 xd->up_available = (mb_row != 0); 752 753 xd->mb_to_top_edge = -((mb_row * 16)) << 3; 754 xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3; 755 756 for (mb_col = 0; mb_col < pc->mb_cols; mb_col++) 757 { 758 if ( mb_row > 0 && (mb_col & (nsync-1)) == 0){ 759 while (mb_col > (*last_row_current_mb_col - nsync) && *last_row_current_mb_col != pc->mb_cols - 1) 760 { 761 x86_pause_hint(); 762 thread_sleep(0); 763 } 764 } 765 766 if (xd->mode_info_context->mbmi.mode == SPLITMV || xd->mode_info_context->mbmi.mode == B_PRED) 767 { 768 for (i = 0; i < 16; i++) 769 { 770 BLOCKD *d = &xd->block[i]; 771 vpx_memcpy(&d->bmi, &xd->mode_info_context->bmi[i], sizeof(B_MODE_INFO)); 772 } 773 } 774 775 /* Distance of Mb to the various image edges. 776 * These are specified to 8th pel as they are always compared to values that are in 1/8th pel units 777 */ 778 xd->mb_to_left_edge = -((mb_col * 16) << 3); 779 xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3; 780 781 xd->dst.y_buffer = pc->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset; 782 xd->dst.u_buffer = pc->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset; 783 xd->dst.v_buffer = pc->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset; 784 785 xd->left_available = (mb_col != 0); 786 787 /* Select the appropriate reference frame for this MB */ 788 if (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME) 789 ref_fb_idx = pc->lst_fb_idx; 790 else if (xd->mode_info_context->mbmi.ref_frame == GOLDEN_FRAME) 791 ref_fb_idx = pc->gld_fb_idx; 792 else 793 ref_fb_idx = pc->alt_fb_idx; 794 795 xd->pre.y_buffer = pc->yv12_fb[ref_fb_idx].y_buffer + recon_yoffset; 796 xd->pre.u_buffer = pc->yv12_fb[ref_fb_idx].u_buffer + recon_uvoffset; 797 xd->pre.v_buffer = pc->yv12_fb[ref_fb_idx].v_buffer + recon_uvoffset; 798 799 if (xd->mode_info_context->mbmi.ref_frame != INTRA_FRAME) 800 { 801 /* propagate errors from reference frames */ 802 xd->corrupted |= pc->yv12_fb[ref_fb_idx].corrupted; 803 } 804 805 vp8_build_uvmvs(xd, pc->full_pixel); 806 decode_macroblock(pbi, xd, mb_row, mb_col); 807 808 /* check if the boolean decoder has suffered an error */ 809 xd->corrupted |= vp8dx_bool_error(xd->current_bc); 810 811 if (pbi->common.filter_level) 812 { 813 /* Save decoded MB last row data for next-row decoding */ 814 if(mb_row != pc->mb_rows-1) 815 { 816 vpx_memcpy((pbi->mt_yabove_row[mb_row +1] + 32 + mb_col*16), (xd->dst.y_buffer + 15 * recon_y_stride), 16); 817 vpx_memcpy((pbi->mt_uabove_row[mb_row +1] + 16 + mb_col*8), (xd->dst.u_buffer + 7 * recon_uv_stride), 8); 818 vpx_memcpy((pbi->mt_vabove_row[mb_row +1] + 16 + mb_col*8), (xd->dst.v_buffer + 7 * recon_uv_stride), 8); 819 } 820 821 /* save left_col for next MB decoding */ 822 if(mb_col != pc->mb_cols-1) 823 { 824 MODE_INFO *next = xd->mode_info_context +1; 825 826 if (xd->frame_type == KEY_FRAME || next->mbmi.ref_frame == INTRA_FRAME) 827 { 828 for (i = 0; i < 16; i++) 829 pbi->mt_yleft_col[mb_row][i] = xd->dst.y_buffer [i* recon_y_stride + 15]; 830 for (i = 0; i < 8; i++) 831 { 832 pbi->mt_uleft_col[mb_row][i] = xd->dst.u_buffer [i* recon_uv_stride + 7]; 833 pbi->mt_vleft_col[mb_row][i] = xd->dst.v_buffer [i* recon_uv_stride + 7]; 834 } 835 } 836 } 837 838 /* update loopfilter info */ 839 Segment = (alt_flt_enabled) ? xd->mode_info_context->mbmi.segment_id : 0; 840 filter_level = pbi->mt_baseline_filter_level[Segment]; 841 /* Distance of Mb to the various image edges. 842 * These are specified to 8th pel as they are always compared to values that are in 1/8th pel units 843 * Apply any context driven MB level adjustment 844 */ 845 filter_level = vp8_adjust_mb_lf_value(xd, filter_level); 846 847 /* loopfilter on this macroblock. */ 848 if (filter_level) 849 { 850 if (mb_col > 0) 851 pc->lf_mbv(xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi[filter_level], pc->simpler_lpf); 852 853 if (xd->mode_info_context->mbmi.dc_diff > 0) 854 pc->lf_bv(xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi[filter_level], pc->simpler_lpf); 855 856 /* don't apply across umv border */ 857 if (mb_row > 0) 858 pc->lf_mbh(xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi[filter_level], pc->simpler_lpf); 859 860 if (xd->mode_info_context->mbmi.dc_diff > 0) 861 pc->lf_bh(xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi[filter_level], pc->simpler_lpf); 862 } 863 } 864 865 recon_yoffset += 16; 866 recon_uvoffset += 8; 867 868 ++xd->mode_info_context; /* next mb */ 869 870 xd->above_context++; 871 872 pbi->mt_current_mb_col[mb_row] = mb_col; 873 } 874 875 /* adjust to the next row of mbs */ 876 if (pbi->common.filter_level) 877 { 878 if(mb_row != pc->mb_rows-1) 879 { 880 int lasty = pc->yv12_fb[ref_fb_idx].y_width + VP8BORDERINPIXELS; 881 int lastuv = (pc->yv12_fb[ref_fb_idx].y_width>>1) + (VP8BORDERINPIXELS>>1); 882 883 for (i = 0; i < 4; i++) 884 { 885 pbi->mt_yabove_row[mb_row +1][lasty + i] = pbi->mt_yabove_row[mb_row +1][lasty -1]; 886 pbi->mt_uabove_row[mb_row +1][lastuv + i] = pbi->mt_uabove_row[mb_row +1][lastuv -1]; 887 pbi->mt_vabove_row[mb_row +1][lastuv + i] = pbi->mt_vabove_row[mb_row +1][lastuv -1]; 888 } 889 } 890 }else 891 vp8_extend_mb_row(&pc->yv12_fb[dst_fb_idx], xd->dst.y_buffer + 16, xd->dst.u_buffer + 8, xd->dst.v_buffer + 8); 892 893 ++xd->mode_info_context; /* skip prediction column */ 894 } 895 xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count; 896 } 897 898 sem_wait(&pbi->h_event_end_decoding); /* add back for each frame */ 899 } 900
