The Tor Browser / file comparison
comparison: media/libvpx/vp8/encoder/ethreading.c
media/libvpx/vp8/encoder/ethreading.c
- branch
- TOR_BUG_9701
- changeset 10
- ac0c01689b40
equal
deleted
inserted
replaced
| -1:000000000000 | 0:1c12ef646015 |
|---|---|
| 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 #include "onyx_int.h" | |
| 12 #include "vp8/common/threading.h" | |
| 13 #include "vp8/common/common.h" | |
| 14 #include "vp8/common/extend.h" | |
| 15 #include "bitstream.h" | |
| 16 #include "encodeframe.h" | |
| 17 | |
| 18 #if CONFIG_MULTITHREAD | |
| 19 | |
| 20 extern void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip); | |
| 21 | |
| 22 extern void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm); | |
| 23 | |
| 24 static THREAD_FUNCTION thread_loopfilter(void *p_data) | |
| 25 { | |
| 26 VP8_COMP *cpi = (VP8_COMP *)(((LPFTHREAD_DATA *)p_data)->ptr1); | |
| 27 VP8_COMMON *cm = &cpi->common; | |
| 28 | |
| 29 while (1) | |
| 30 { | |
| 31 if (cpi->b_multi_threaded == 0) | |
| 32 break; | |
| 33 | |
| 34 if (sem_wait(&cpi->h_event_start_lpf) == 0) | |
| 35 { | |
| 36 if (cpi->b_multi_threaded == 0) /* we're shutting down */ | |
| 37 break; | |
| 38 | |
| 39 vp8_loopfilter_frame(cpi, cm); | |
| 40 | |
| 41 sem_post(&cpi->h_event_end_lpf); | |
| 42 } | |
| 43 } | |
| 44 | |
| 45 return 0; | |
| 46 } | |
| 47 | |
| 48 static | |
| 49 THREAD_FUNCTION thread_encoding_proc(void *p_data) | |
| 50 { | |
| 51 int ithread = ((ENCODETHREAD_DATA *)p_data)->ithread; | |
| 52 VP8_COMP *cpi = (VP8_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr1); | |
| 53 MB_ROW_COMP *mbri = (MB_ROW_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr2); | |
| 54 ENTROPY_CONTEXT_PLANES mb_row_left_context; | |
| 55 | |
| 56 while (1) | |
| 57 { | |
| 58 if (cpi->b_multi_threaded == 0) | |
| 59 break; | |
| 60 | |
| 61 if (sem_wait(&cpi->h_event_start_encoding[ithread]) == 0) | |
| 62 { | |
| 63 const int nsync = cpi->mt_sync_range; | |
| 64 VP8_COMMON *cm = &cpi->common; | |
| 65 int mb_row; | |
| 66 MACROBLOCK *x = &mbri->mb; | |
| 67 MACROBLOCKD *xd = &x->e_mbd; | |
| 68 TOKENEXTRA *tp ; | |
| 69 #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING | |
| 70 TOKENEXTRA *tp_start = cpi->tok + (1 + ithread) * (16 * 24); | |
| 71 const int num_part = (1 << cm->multi_token_partition); | |
| 72 #endif | |
| 73 | |
| 74 int *segment_counts = mbri->segment_counts; | |
| 75 int *totalrate = &mbri->totalrate; | |
| 76 | |
| 77 if (cpi->b_multi_threaded == 0) /* we're shutting down */ | |
| 78 break; | |
| 79 | |
| 80 for (mb_row = ithread + 1; mb_row < cm->mb_rows; mb_row += (cpi->encoding_thread_count + 1)) | |
| 81 { | |
| 82 | |
| 83 int recon_yoffset, recon_uvoffset; | |
| 84 int mb_col; | |
| 85 int ref_fb_idx = cm->lst_fb_idx; | |
| 86 int dst_fb_idx = cm->new_fb_idx; | |
| 87 int recon_y_stride = cm->yv12_fb[ref_fb_idx].y_stride; | |
| 88 int recon_uv_stride = cm->yv12_fb[ref_fb_idx].uv_stride; | |
| 89 int map_index = (mb_row * cm->mb_cols); | |
| 90 volatile const int *last_row_current_mb_col; | |
| 91 volatile int *current_mb_col = &cpi->mt_current_mb_col[mb_row]; | |
| 92 | |
| 93 #if (CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING) | |
| 94 vp8_writer *w = &cpi->bc[1 + (mb_row % num_part)]; | |
| 95 #else | |
| 96 tp = cpi->tok + (mb_row * (cm->mb_cols * 16 * 24)); | |
| 97 cpi->tplist[mb_row].start = tp; | |
| 98 #endif | |
| 99 | |
| 100 last_row_current_mb_col = &cpi->mt_current_mb_col[mb_row - 1]; | |
| 101 | |
| 102 /* reset above block coeffs */ | |
| 103 xd->above_context = cm->above_context; | |
| 104 xd->left_context = &mb_row_left_context; | |
| 105 | |
| 106 vp8_zero(mb_row_left_context); | |
| 107 | |
| 108 xd->up_available = (mb_row != 0); | |
| 109 recon_yoffset = (mb_row * recon_y_stride * 16); | |
| 110 recon_uvoffset = (mb_row * recon_uv_stride * 8); | |
| 111 | |
| 112 /* Set the mb activity pointer to the start of the row. */ | |
| 113 x->mb_activity_ptr = &cpi->mb_activity_map[map_index]; | |
| 114 | |
| 115 /* for each macroblock col in image */ | |
| 116 for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) | |
| 117 { | |
| 118 *current_mb_col = mb_col - 1; | |
| 119 | |
| 120 if ((mb_col & (nsync - 1)) == 0) | |
| 121 { | |
| 122 while (mb_col > (*last_row_current_mb_col - nsync)) | |
| 123 { | |
| 124 x86_pause_hint(); | |
| 125 thread_sleep(0); | |
| 126 } | |
| 127 } | |
| 128 | |
| 129 #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING | |
| 130 tp = tp_start; | |
| 131 #endif | |
| 132 | |
| 133 /* Distance of Mb to the various image edges. | |
| 134 * These specified to 8th pel as they are always compared | |
| 135 * to values that are in 1/8th pel units | |
| 136 */ | |
| 137 xd->mb_to_left_edge = -((mb_col * 16) << 3); | |
| 138 xd->mb_to_right_edge = ((cm->mb_cols - 1 - mb_col) * 16) << 3; | |
| 139 xd->mb_to_top_edge = -((mb_row * 16) << 3); | |
| 140 xd->mb_to_bottom_edge = ((cm->mb_rows - 1 - mb_row) * 16) << 3; | |
| 141 | |
| 142 /* Set up limit values for motion vectors used to prevent | |
| 143 * them extending outside the UMV borders | |
| 144 */ | |
| 145 x->mv_col_min = -((mb_col * 16) + (VP8BORDERINPIXELS - 16)); | |
| 146 x->mv_col_max = ((cm->mb_cols - 1 - mb_col) * 16) + (VP8BORDERINPIXELS - 16); | |
| 147 x->mv_row_min = -((mb_row * 16) + (VP8BORDERINPIXELS - 16)); | |
| 148 x->mv_row_max = ((cm->mb_rows - 1 - mb_row) * 16) + (VP8BORDERINPIXELS - 16); | |
| 149 | |
| 150 xd->dst.y_buffer = cm->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset; | |
| 151 xd->dst.u_buffer = cm->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset; | |
| 152 xd->dst.v_buffer = cm->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset; | |
| 153 xd->left_available = (mb_col != 0); | |
| 154 | |
| 155 x->rddiv = cpi->RDDIV; | |
| 156 x->rdmult = cpi->RDMULT; | |
| 157 | |
| 158 /* Copy current mb to a buffer */ | |
| 159 vp8_copy_mem16x16(x->src.y_buffer, x->src.y_stride, x->thismb, 16); | |
| 160 | |
| 161 if (cpi->oxcf.tuning == VP8_TUNE_SSIM) | |
| 162 vp8_activity_masking(cpi, x); | |
| 163 | |
| 164 /* Is segmentation enabled */ | |
| 165 /* MB level adjustment to quantizer */ | |
| 166 if (xd->segmentation_enabled) | |
| 167 { | |
| 168 /* Code to set segment id in xd->mbmi.segment_id for | |
| 169 * current MB (with range checking) | |
| 170 */ | |
| 171 if (cpi->segmentation_map[map_index + mb_col] <= 3) | |
| 172 xd->mode_info_context->mbmi.segment_id = cpi->segmentation_map[map_index + mb_col]; | |
| 173 else | |
| 174 xd->mode_info_context->mbmi.segment_id = 0; | |
| 175 | |
| 176 vp8cx_mb_init_quantizer(cpi, x, 1); | |
| 177 } | |
| 178 else | |
| 179 /* Set to Segment 0 by default */ | |
| 180 xd->mode_info_context->mbmi.segment_id = 0; | |
| 181 | |
| 182 x->active_ptr = cpi->active_map + map_index + mb_col; | |
| 183 | |
| 184 if (cm->frame_type == KEY_FRAME) | |
| 185 { | |
| 186 *totalrate += vp8cx_encode_intra_macroblock(cpi, x, &tp); | |
| 187 #ifdef MODE_STATS | |
| 188 y_modes[xd->mbmi.mode] ++; | |
| 189 #endif | |
| 190 } | |
| 191 else | |
| 192 { | |
| 193 *totalrate += vp8cx_encode_inter_macroblock(cpi, x, &tp, recon_yoffset, recon_uvoffset, mb_row, mb_col); | |
| 194 | |
| 195 #ifdef MODE_STATS | |
| 196 inter_y_modes[xd->mbmi.mode] ++; | |
| 197 | |
| 198 if (xd->mbmi.mode == SPLITMV) | |
| 199 { | |
| 200 int b; | |
| 201 | |
| 202 for (b = 0; b < xd->mbmi.partition_count; b++) | |
| 203 { | |
| 204 inter_b_modes[x->partition->bmi[b].mode] ++; | |
| 205 } | |
| 206 } | |
| 207 | |
| 208 #endif | |
| 209 | |
| 210 /* Special case code for cyclic refresh | |
| 211 * If cyclic update enabled then copy | |
| 212 * xd->mbmi.segment_id; (which may have been updated | |
| 213 * based on mode during | |
| 214 * vp8cx_encode_inter_macroblock()) back into the | |
| 215 * global segmentation map | |
| 216 */ | |
| 217 if ((cpi->current_layer == 0) && | |
| 218 (cpi->cyclic_refresh_mode_enabled && | |
| 219 xd->segmentation_enabled)) | |
| 220 { | |
| 221 const MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi; | |
| 222 cpi->segmentation_map[map_index + mb_col] = mbmi->segment_id; | |
| 223 | |
| 224 /* If the block has been refreshed mark it as clean | |
| 225 * (the magnitude of the -ve influences how long it | |
| 226 * will be before we consider another refresh): | |
| 227 * Else if it was coded (last frame 0,0) and has | |
| 228 * not already been refreshed then mark it as a | |
| 229 * candidate for cleanup next time (marked 0) else | |
| 230 * mark it as dirty (1). | |
| 231 */ | |
| 232 if (mbmi->segment_id) | |
| 233 cpi->cyclic_refresh_map[map_index + mb_col] = -1; | |
| 234 else if ((mbmi->mode == ZEROMV) && (mbmi->ref_frame == LAST_FRAME)) | |
| 235 { | |
| 236 if (cpi->cyclic_refresh_map[map_index + mb_col] == 1) | |
| 237 cpi->cyclic_refresh_map[map_index + mb_col] = 0; | |
| 238 } | |
| 239 else | |
| 240 cpi->cyclic_refresh_map[map_index + mb_col] = 1; | |
| 241 | |
| 242 } | |
| 243 } | |
| 244 | |
| 245 #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING | |
| 246 /* pack tokens for this MB */ | |
| 247 { | |
| 248 int tok_count = tp - tp_start; | |
| 249 pack_tokens(w, tp_start, tok_count); | |
| 250 } | |
| 251 #else | |
| 252 cpi->tplist[mb_row].stop = tp; | |
| 253 #endif | |
| 254 /* Increment pointer into gf usage flags structure. */ | |
| 255 x->gf_active_ptr++; | |
| 256 | |
| 257 /* Increment the activity mask pointers. */ | |
| 258 x->mb_activity_ptr++; | |
| 259 | |
| 260 /* adjust to the next column of macroblocks */ | |
| 261 x->src.y_buffer += 16; | |
| 262 x->src.u_buffer += 8; | |
| 263 x->src.v_buffer += 8; | |
| 264 | |
| 265 recon_yoffset += 16; | |
| 266 recon_uvoffset += 8; | |
| 267 | |
| 268 /* Keep track of segment usage */ | |
| 269 segment_counts[xd->mode_info_context->mbmi.segment_id]++; | |
| 270 | |
| 271 /* skip to next mb */ | |
| 272 xd->mode_info_context++; | |
| 273 x->partition_info++; | |
| 274 xd->above_context++; | |
| 275 } | |
| 276 | |
| 277 vp8_extend_mb_row( &cm->yv12_fb[dst_fb_idx], | |
| 278 xd->dst.y_buffer + 16, | |
| 279 xd->dst.u_buffer + 8, | |
| 280 xd->dst.v_buffer + 8); | |
| 281 | |
| 282 *current_mb_col = mb_col + nsync; | |
| 283 | |
| 284 /* this is to account for the border */ | |
| 285 xd->mode_info_context++; | |
| 286 x->partition_info++; | |
| 287 | |
| 288 x->src.y_buffer += 16 * x->src.y_stride * (cpi->encoding_thread_count + 1) - 16 * cm->mb_cols; | |
| 289 x->src.u_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols; | |
| 290 x->src.v_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols; | |
| 291 | |
| 292 xd->mode_info_context += xd->mode_info_stride * cpi->encoding_thread_count; | |
| 293 x->partition_info += xd->mode_info_stride * cpi->encoding_thread_count; | |
| 294 x->gf_active_ptr += cm->mb_cols * cpi->encoding_thread_count; | |
| 295 | |
| 296 if (mb_row == cm->mb_rows - 1) | |
| 297 { | |
| 298 sem_post(&cpi->h_event_end_encoding); /* signal frame encoding end */ | |
| 299 } | |
| 300 } | |
| 301 } | |
| 302 } | |
| 303 | |
| 304 /* printf("exit thread %d\n", ithread); */ | |
| 305 return 0; | |
| 306 } | |
| 307 | |
| 308 static void setup_mbby_copy(MACROBLOCK *mbdst, MACROBLOCK *mbsrc) | |
| 309 { | |
| 310 | |
| 311 MACROBLOCK *x = mbsrc; | |
| 312 MACROBLOCK *z = mbdst; | |
| 313 int i; | |
| 314 | |
| 315 z->ss = x->ss; | |
| 316 z->ss_count = x->ss_count; | |
| 317 z->searches_per_step = x->searches_per_step; | |
| 318 z->errorperbit = x->errorperbit; | |
| 319 | |
| 320 z->sadperbit16 = x->sadperbit16; | |
| 321 z->sadperbit4 = x->sadperbit4; | |
| 322 | |
| 323 /* | |
| 324 z->mv_col_min = x->mv_col_min; | |
| 325 z->mv_col_max = x->mv_col_max; | |
| 326 z->mv_row_min = x->mv_row_min; | |
| 327 z->mv_row_max = x->mv_row_max; | |
| 328 */ | |
| 329 | |
| 330 z->short_fdct4x4 = x->short_fdct4x4; | |
| 331 z->short_fdct8x4 = x->short_fdct8x4; | |
| 332 z->short_walsh4x4 = x->short_walsh4x4; | |
| 333 z->quantize_b = x->quantize_b; | |
| 334 z->quantize_b_pair = x->quantize_b_pair; | |
| 335 z->optimize = x->optimize; | |
| 336 | |
| 337 /* | |
| 338 z->mvc = x->mvc; | |
| 339 z->src.y_buffer = x->src.y_buffer; | |
| 340 z->src.u_buffer = x->src.u_buffer; | |
| 341 z->src.v_buffer = x->src.v_buffer; | |
| 342 */ | |
| 343 | |
| 344 z->mvcost[0] = x->mvcost[0]; | |
| 345 z->mvcost[1] = x->mvcost[1]; | |
| 346 z->mvsadcost[0] = x->mvsadcost[0]; | |
| 347 z->mvsadcost[1] = x->mvsadcost[1]; | |
| 348 | |
| 349 z->token_costs = x->token_costs; | |
| 350 z->inter_bmode_costs = x->inter_bmode_costs; | |
| 351 z->mbmode_cost = x->mbmode_cost; | |
| 352 z->intra_uv_mode_cost = x->intra_uv_mode_cost; | |
| 353 z->bmode_costs = x->bmode_costs; | |
| 354 | |
| 355 for (i = 0; i < 25; i++) | |
| 356 { | |
| 357 z->block[i].quant = x->block[i].quant; | |
| 358 z->block[i].quant_fast = x->block[i].quant_fast; | |
| 359 z->block[i].quant_shift = x->block[i].quant_shift; | |
| 360 z->block[i].zbin = x->block[i].zbin; | |
| 361 z->block[i].zrun_zbin_boost = x->block[i].zrun_zbin_boost; | |
| 362 z->block[i].round = x->block[i].round; | |
| 363 z->block[i].src_stride = x->block[i].src_stride; | |
| 364 } | |
| 365 | |
| 366 z->q_index = x->q_index; | |
| 367 z->act_zbin_adj = x->act_zbin_adj; | |
| 368 z->last_act_zbin_adj = x->last_act_zbin_adj; | |
| 369 | |
| 370 { | |
| 371 MACROBLOCKD *xd = &x->e_mbd; | |
| 372 MACROBLOCKD *zd = &z->e_mbd; | |
| 373 | |
| 374 /* | |
| 375 zd->mode_info_context = xd->mode_info_context; | |
| 376 zd->mode_info = xd->mode_info; | |
| 377 | |
| 378 zd->mode_info_stride = xd->mode_info_stride; | |
| 379 zd->frame_type = xd->frame_type; | |
| 380 zd->up_available = xd->up_available ; | |
| 381 zd->left_available = xd->left_available; | |
| 382 zd->left_context = xd->left_context; | |
| 383 zd->last_frame_dc = xd->last_frame_dc; | |
| 384 zd->last_frame_dccons = xd->last_frame_dccons; | |
| 385 zd->gold_frame_dc = xd->gold_frame_dc; | |
| 386 zd->gold_frame_dccons = xd->gold_frame_dccons; | |
| 387 zd->mb_to_left_edge = xd->mb_to_left_edge; | |
| 388 zd->mb_to_right_edge = xd->mb_to_right_edge; | |
| 389 zd->mb_to_top_edge = xd->mb_to_top_edge ; | |
| 390 zd->mb_to_bottom_edge = xd->mb_to_bottom_edge; | |
| 391 zd->gf_active_ptr = xd->gf_active_ptr; | |
| 392 zd->frames_since_golden = xd->frames_since_golden; | |
| 393 zd->frames_till_alt_ref_frame = xd->frames_till_alt_ref_frame; | |
| 394 */ | |
| 395 zd->subpixel_predict = xd->subpixel_predict; | |
| 396 zd->subpixel_predict8x4 = xd->subpixel_predict8x4; | |
| 397 zd->subpixel_predict8x8 = xd->subpixel_predict8x8; | |
| 398 zd->subpixel_predict16x16 = xd->subpixel_predict16x16; | |
| 399 zd->segmentation_enabled = xd->segmentation_enabled; | |
| 400 zd->mb_segement_abs_delta = xd->mb_segement_abs_delta; | |
| 401 vpx_memcpy(zd->segment_feature_data, xd->segment_feature_data, | |
| 402 sizeof(xd->segment_feature_data)); | |
| 403 | |
| 404 vpx_memcpy(zd->dequant_y1_dc, xd->dequant_y1_dc, | |
| 405 sizeof(xd->dequant_y1_dc)); | |
| 406 vpx_memcpy(zd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1)); | |
| 407 vpx_memcpy(zd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2)); | |
| 408 vpx_memcpy(zd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv)); | |
| 409 | |
| 410 #if 1 | |
| 411 /*TODO: Remove dequant from BLOCKD. This is a temporary solution until | |
| 412 * the quantizer code uses a passed in pointer to the dequant constants. | |
| 413 * This will also require modifications to the x86 and neon assembly. | |
| 414 * */ | |
| 415 for (i = 0; i < 16; i++) | |
| 416 zd->block[i].dequant = zd->dequant_y1; | |
| 417 for (i = 16; i < 24; i++) | |
| 418 zd->block[i].dequant = zd->dequant_uv; | |
| 419 zd->block[24].dequant = zd->dequant_y2; | |
| 420 #endif | |
| 421 | |
| 422 | |
| 423 vpx_memcpy(z->rd_threshes, x->rd_threshes, sizeof(x->rd_threshes)); | |
| 424 vpx_memcpy(z->rd_thresh_mult, x->rd_thresh_mult, | |
| 425 sizeof(x->rd_thresh_mult)); | |
| 426 | |
| 427 z->zbin_over_quant = x->zbin_over_quant; | |
| 428 z->zbin_mode_boost_enabled = x->zbin_mode_boost_enabled; | |
| 429 z->zbin_mode_boost = x->zbin_mode_boost; | |
| 430 | |
| 431 vpx_memset(z->error_bins, 0, sizeof(z->error_bins)); | |
| 432 } | |
| 433 } | |
| 434 | |
| 435 void vp8cx_init_mbrthread_data(VP8_COMP *cpi, | |
| 436 MACROBLOCK *x, | |
| 437 MB_ROW_COMP *mbr_ei, | |
| 438 int count | |
| 439 ) | |
| 440 { | |
| 441 | |
| 442 VP8_COMMON *const cm = & cpi->common; | |
| 443 MACROBLOCKD *const xd = & x->e_mbd; | |
| 444 int i; | |
| 445 | |
| 446 for (i = 0; i < count; i++) | |
| 447 { | |
| 448 MACROBLOCK *mb = & mbr_ei[i].mb; | |
| 449 MACROBLOCKD *mbd = &mb->e_mbd; | |
| 450 | |
| 451 mbd->subpixel_predict = xd->subpixel_predict; | |
| 452 mbd->subpixel_predict8x4 = xd->subpixel_predict8x4; | |
| 453 mbd->subpixel_predict8x8 = xd->subpixel_predict8x8; | |
| 454 mbd->subpixel_predict16x16 = xd->subpixel_predict16x16; | |
| 455 mb->gf_active_ptr = x->gf_active_ptr; | |
| 456 | |
| 457 vpx_memset(mbr_ei[i].segment_counts, 0, sizeof(mbr_ei[i].segment_counts)); | |
| 458 mbr_ei[i].totalrate = 0; | |
| 459 | |
| 460 mb->partition_info = x->pi + x->e_mbd.mode_info_stride * (i + 1); | |
| 461 | |
| 462 mbd->mode_info_context = cm->mi + x->e_mbd.mode_info_stride * (i + 1); | |
| 463 mbd->mode_info_stride = cm->mode_info_stride; | |
| 464 | |
| 465 mbd->frame_type = cm->frame_type; | |
| 466 | |
| 467 mb->src = * cpi->Source; | |
| 468 mbd->pre = cm->yv12_fb[cm->lst_fb_idx]; | |
| 469 mbd->dst = cm->yv12_fb[cm->new_fb_idx]; | |
| 470 | |
| 471 mb->src.y_buffer += 16 * x->src.y_stride * (i + 1); | |
| 472 mb->src.u_buffer += 8 * x->src.uv_stride * (i + 1); | |
| 473 mb->src.v_buffer += 8 * x->src.uv_stride * (i + 1); | |
| 474 | |
| 475 vp8_build_block_offsets(mb); | |
| 476 | |
| 477 mbd->left_context = &cm->left_context; | |
| 478 mb->mvc = cm->fc.mvc; | |
| 479 | |
| 480 setup_mbby_copy(&mbr_ei[i].mb, x); | |
| 481 | |
| 482 mbd->fullpixel_mask = 0xffffffff; | |
| 483 if(cm->full_pixel) | |
| 484 mbd->fullpixel_mask = 0xfffffff8; | |
| 485 | |
| 486 vp8_zero(mb->coef_counts); | |
| 487 vp8_zero(x->ymode_count); | |
| 488 mb->skip_true_count = 0; | |
| 489 vp8_zero(mb->MVcount); | |
| 490 mb->prediction_error = 0; | |
| 491 mb->intra_error = 0; | |
| 492 vp8_zero(mb->count_mb_ref_frame_usage); | |
| 493 mb->mbs_tested_so_far = 0; | |
| 494 } | |
| 495 } | |
| 496 | |
| 497 int vp8cx_create_encoder_threads(VP8_COMP *cpi) | |
| 498 { | |
| 499 const VP8_COMMON * cm = &cpi->common; | |
| 500 | |
| 501 cpi->b_multi_threaded = 0; | |
| 502 cpi->encoding_thread_count = 0; | |
| 503 cpi->b_lpf_running = 0; | |
| 504 | |
| 505 if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1) | |
| 506 { | |
| 507 int ithread; | |
| 508 int th_count = cpi->oxcf.multi_threaded - 1; | |
| 509 int rc = 0; | |
| 510 | |
| 511 /* don't allocate more threads than cores available */ | |
| 512 if (cpi->oxcf.multi_threaded > cm->processor_core_count) | |
| 513 th_count = cm->processor_core_count - 1; | |
| 514 | |
| 515 /* we have th_count + 1 (main) threads processing one row each */ | |
| 516 /* no point to have more threads than the sync range allows */ | |
| 517 if(th_count > ((cm->mb_cols / cpi->mt_sync_range) - 1)) | |
| 518 { | |
| 519 th_count = (cm->mb_cols / cpi->mt_sync_range) - 1; | |
| 520 } | |
| 521 | |
| 522 if(th_count == 0) | |
| 523 return 0; | |
| 524 | |
| 525 CHECK_MEM_ERROR(cpi->h_encoding_thread, | |
| 526 vpx_malloc(sizeof(pthread_t) * th_count)); | |
| 527 CHECK_MEM_ERROR(cpi->h_event_start_encoding, | |
| 528 vpx_malloc(sizeof(sem_t) * th_count)); | |
| 529 CHECK_MEM_ERROR(cpi->mb_row_ei, | |
| 530 vpx_memalign(32, sizeof(MB_ROW_COMP) * th_count)); | |
| 531 vpx_memset(cpi->mb_row_ei, 0, sizeof(MB_ROW_COMP) * th_count); | |
| 532 CHECK_MEM_ERROR(cpi->en_thread_data, | |
| 533 vpx_malloc(sizeof(ENCODETHREAD_DATA) * th_count)); | |
| 534 | |
| 535 sem_init(&cpi->h_event_end_encoding, 0, 0); | |
| 536 | |
| 537 cpi->b_multi_threaded = 1; | |
| 538 cpi->encoding_thread_count = th_count; | |
| 539 | |
| 540 /* | |
| 541 printf("[VP8:] multi_threaded encoding is enabled with %d threads\n\n", | |
| 542 (cpi->encoding_thread_count +1)); | |
| 543 */ | |
| 544 | |
| 545 for (ithread = 0; ithread < th_count; ithread++) | |
| 546 { | |
| 547 ENCODETHREAD_DATA *ethd = &cpi->en_thread_data[ithread]; | |
| 548 | |
| 549 /* Setup block ptrs and offsets */ | |
| 550 vp8_setup_block_ptrs(&cpi->mb_row_ei[ithread].mb); | |
| 551 vp8_setup_block_dptrs(&cpi->mb_row_ei[ithread].mb.e_mbd); | |
| 552 | |
| 553 sem_init(&cpi->h_event_start_encoding[ithread], 0, 0); | |
| 554 | |
| 555 ethd->ithread = ithread; | |
| 556 ethd->ptr1 = (void *)cpi; | |
| 557 ethd->ptr2 = (void *)&cpi->mb_row_ei[ithread]; | |
| 558 | |
| 559 rc = pthread_create(&cpi->h_encoding_thread[ithread], 0, | |
| 560 thread_encoding_proc, ethd); | |
| 561 if(rc) | |
| 562 break; | |
| 563 } | |
| 564 | |
| 565 if(rc) | |
| 566 { | |
| 567 /* shutdown other threads */ | |
| 568 cpi->b_multi_threaded = 0; | |
| 569 for(--ithread; ithread >= 0; ithread--) | |
| 570 { | |
| 571 pthread_join(cpi->h_encoding_thread[ithread], 0); | |
| 572 sem_destroy(&cpi->h_event_start_encoding[ithread]); | |
| 573 } | |
| 574 sem_destroy(&cpi->h_event_end_encoding); | |
| 575 | |
| 576 /* free thread related resources */ | |
| 577 vpx_free(cpi->h_event_start_encoding); | |
| 578 vpx_free(cpi->h_encoding_thread); | |
| 579 vpx_free(cpi->mb_row_ei); | |
| 580 vpx_free(cpi->en_thread_data); | |
| 581 | |
| 582 return -1; | |
| 583 } | |
| 584 | |
| 585 | |
| 586 { | |
| 587 LPFTHREAD_DATA * lpfthd = &cpi->lpf_thread_data; | |
| 588 | |
| 589 sem_init(&cpi->h_event_start_lpf, 0, 0); | |
| 590 sem_init(&cpi->h_event_end_lpf, 0, 0); | |
| 591 | |
| 592 lpfthd->ptr1 = (void *)cpi; | |
| 593 rc = pthread_create(&cpi->h_filter_thread, 0, thread_loopfilter, | |
| 594 lpfthd); | |
| 595 | |
| 596 if(rc) | |
| 597 { | |
| 598 /* shutdown other threads */ | |
| 599 cpi->b_multi_threaded = 0; | |
| 600 for(--ithread; ithread >= 0; ithread--) | |
| 601 { | |
| 602 sem_post(&cpi->h_event_start_encoding[ithread]); | |
| 603 pthread_join(cpi->h_encoding_thread[ithread], 0); | |
| 604 sem_destroy(&cpi->h_event_start_encoding[ithread]); | |
| 605 } | |
| 606 sem_destroy(&cpi->h_event_end_encoding); | |
| 607 sem_destroy(&cpi->h_event_end_lpf); | |
| 608 sem_destroy(&cpi->h_event_start_lpf); | |
| 609 | |
| 610 /* free thread related resources */ | |
| 611 vpx_free(cpi->h_event_start_encoding); | |
| 612 vpx_free(cpi->h_encoding_thread); | |
| 613 vpx_free(cpi->mb_row_ei); | |
| 614 vpx_free(cpi->en_thread_data); | |
| 615 | |
| 616 return -2; | |
| 617 } | |
| 618 } | |
| 619 } | |
| 620 return 0; | |
| 621 } | |
| 622 | |
| 623 void vp8cx_remove_encoder_threads(VP8_COMP *cpi) | |
| 624 { | |
| 625 if (cpi->b_multi_threaded) | |
| 626 { | |
| 627 /* shutdown other threads */ | |
| 628 cpi->b_multi_threaded = 0; | |
| 629 { | |
| 630 int i; | |
| 631 | |
| 632 for (i = 0; i < cpi->encoding_thread_count; i++) | |
| 633 { | |
| 634 sem_post(&cpi->h_event_start_encoding[i]); | |
| 635 pthread_join(cpi->h_encoding_thread[i], 0); | |
| 636 | |
| 637 sem_destroy(&cpi->h_event_start_encoding[i]); | |
| 638 } | |
| 639 | |
| 640 sem_post(&cpi->h_event_start_lpf); | |
| 641 pthread_join(cpi->h_filter_thread, 0); | |
| 642 } | |
| 643 | |
| 644 sem_destroy(&cpi->h_event_end_encoding); | |
| 645 sem_destroy(&cpi->h_event_end_lpf); | |
| 646 sem_destroy(&cpi->h_event_start_lpf); | |
| 647 | |
| 648 /* free thread related resources */ | |
| 649 vpx_free(cpi->h_event_start_encoding); | |
| 650 vpx_free(cpi->h_encoding_thread); | |
| 651 vpx_free(cpi->mb_row_ei); | |
| 652 vpx_free(cpi->en_thread_data); | |
| 653 } | |
| 654 } | |
| 655 #endif |
