The Tor Browser / annotate

media/libvpx/vp8/encoder/ethreading.c@ac0c01689b40 (annotated)

media/libvpx/vp8/encoder/ethreading.c

Thu, 15 Jan 2015 15:59:08 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Thu, 15 Jan 2015 15:59:08 +0100
branch
TOR_BUG_9701
changeset 10
ac0c01689b40
permissions
-rw-r--r--

Implement a real Private Browsing Mode condition by changing the API/ABI;
This solves Tor bug #9701, complying with disk avoidance documented in
https://www.torproject.org/projects/torbrowser/design/#disk-avoidance.

michael@0 1 /*
michael@0 2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0 3 *
michael@0 4 * Use of this source code is governed by a BSD-style license
michael@0 5 * that can be found in the LICENSE file in the root of the source
michael@0 6 * tree. An additional intellectual property rights grant can be found
michael@0 7 * in the file PATENTS. All contributing project authors may
michael@0 8 * be found in the AUTHORS file in the root of the source tree.
michael@0 9 */
michael@0 10
michael@0 11 #include "onyx_int.h"
michael@0 12 #include "vp8/common/threading.h"
michael@0 13 #include "vp8/common/common.h"
michael@0 14 #include "vp8/common/extend.h"
michael@0 15 #include "bitstream.h"
michael@0 16 #include "encodeframe.h"
michael@0 17
michael@0 18 #if CONFIG_MULTITHREAD
michael@0 19
michael@0 20 extern void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip);
michael@0 21
michael@0 22 extern void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm);
michael@0 23
michael@0 24 static THREAD_FUNCTION thread_loopfilter(void *p_data)
michael@0 25 {
michael@0 26 VP8_COMP *cpi = (VP8_COMP *)(((LPFTHREAD_DATA *)p_data)->ptr1);
michael@0 27 VP8_COMMON *cm = &cpi->common;
michael@0 28
michael@0 29 while (1)
michael@0 30 {
michael@0 31 if (cpi->b_multi_threaded == 0)
michael@0 32 break;
michael@0 33
michael@0 34 if (sem_wait(&cpi->h_event_start_lpf) == 0)
michael@0 35 {
michael@0 36 if (cpi->b_multi_threaded == 0) /* we're shutting down */
michael@0 37 break;
michael@0 38
michael@0 39 vp8_loopfilter_frame(cpi, cm);
michael@0 40
michael@0 41 sem_post(&cpi->h_event_end_lpf);
michael@0 42 }
michael@0 43 }
michael@0 44
michael@0 45 return 0;
michael@0 46 }
michael@0 47
michael@0 48 static
michael@0 49 THREAD_FUNCTION thread_encoding_proc(void *p_data)
michael@0 50 {
michael@0 51 int ithread = ((ENCODETHREAD_DATA *)p_data)->ithread;
michael@0 52 VP8_COMP *cpi = (VP8_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr1);
michael@0 53 MB_ROW_COMP *mbri = (MB_ROW_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr2);
michael@0 54 ENTROPY_CONTEXT_PLANES mb_row_left_context;
michael@0 55
michael@0 56 while (1)
michael@0 57 {
michael@0 58 if (cpi->b_multi_threaded == 0)
michael@0 59 break;
michael@0 60
michael@0 61 if (sem_wait(&cpi->h_event_start_encoding[ithread]) == 0)
michael@0 62 {
michael@0 63 const int nsync = cpi->mt_sync_range;
michael@0 64 VP8_COMMON *cm = &cpi->common;
michael@0 65 int mb_row;
michael@0 66 MACROBLOCK *x = &mbri->mb;
michael@0 67 MACROBLOCKD *xd = &x->e_mbd;
michael@0 68 TOKENEXTRA *tp ;
michael@0 69 #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING
michael@0 70 TOKENEXTRA *tp_start = cpi->tok + (1 + ithread) * (16 * 24);
michael@0 71 const int num_part = (1 << cm->multi_token_partition);
michael@0 72 #endif
michael@0 73
michael@0 74 int *segment_counts = mbri->segment_counts;
michael@0 75 int *totalrate = &mbri->totalrate;
michael@0 76
michael@0 77 if (cpi->b_multi_threaded == 0) /* we're shutting down */
michael@0 78 break;
michael@0 79
michael@0 80 for (mb_row = ithread + 1; mb_row < cm->mb_rows; mb_row += (cpi->encoding_thread_count + 1))
michael@0 81 {
michael@0 82
michael@0 83 int recon_yoffset, recon_uvoffset;
michael@0 84 int mb_col;
michael@0 85 int ref_fb_idx = cm->lst_fb_idx;
michael@0 86 int dst_fb_idx = cm->new_fb_idx;
michael@0 87 int recon_y_stride = cm->yv12_fb[ref_fb_idx].y_stride;
michael@0 88 int recon_uv_stride = cm->yv12_fb[ref_fb_idx].uv_stride;
michael@0 89 int map_index = (mb_row * cm->mb_cols);
michael@0 90 volatile const int *last_row_current_mb_col;
michael@0 91 volatile int *current_mb_col = &cpi->mt_current_mb_col[mb_row];
michael@0 92
michael@0 93 #if (CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING)
michael@0 94 vp8_writer *w = &cpi->bc[1 + (mb_row % num_part)];
michael@0 95 #else
michael@0 96 tp = cpi->tok + (mb_row * (cm->mb_cols * 16 * 24));
michael@0 97 cpi->tplist[mb_row].start = tp;
michael@0 98 #endif
michael@0 99
michael@0 100 last_row_current_mb_col = &cpi->mt_current_mb_col[mb_row - 1];
michael@0 101
michael@0 102 /* reset above block coeffs */
michael@0 103 xd->above_context = cm->above_context;
michael@0 104 xd->left_context = &mb_row_left_context;
michael@0 105
michael@0 106 vp8_zero(mb_row_left_context);
michael@0 107
michael@0 108 xd->up_available = (mb_row != 0);
michael@0 109 recon_yoffset = (mb_row * recon_y_stride * 16);
michael@0 110 recon_uvoffset = (mb_row * recon_uv_stride * 8);
michael@0 111
michael@0 112 /* Set the mb activity pointer to the start of the row. */
michael@0 113 x->mb_activity_ptr = &cpi->mb_activity_map[map_index];
michael@0 114
michael@0 115 /* for each macroblock col in image */
michael@0 116 for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
michael@0 117 {
michael@0 118 *current_mb_col = mb_col - 1;
michael@0 119
michael@0 120 if ((mb_col & (nsync - 1)) == 0)
michael@0 121 {
michael@0 122 while (mb_col > (*last_row_current_mb_col - nsync))
michael@0 123 {
michael@0 124 x86_pause_hint();
michael@0 125 thread_sleep(0);
michael@0 126 }
michael@0 127 }
michael@0 128
michael@0 129 #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING
michael@0 130 tp = tp_start;
michael@0 131 #endif
michael@0 132
michael@0 133 /* Distance of Mb to the various image edges.
michael@0 134 * These specified to 8th pel as they are always compared
michael@0 135 * to values that are in 1/8th pel units
michael@0 136 */
michael@0 137 xd->mb_to_left_edge = -((mb_col * 16) << 3);
michael@0 138 xd->mb_to_right_edge = ((cm->mb_cols - 1 - mb_col) * 16) << 3;
michael@0 139 xd->mb_to_top_edge = -((mb_row * 16) << 3);
michael@0 140 xd->mb_to_bottom_edge = ((cm->mb_rows - 1 - mb_row) * 16) << 3;
michael@0 141
michael@0 142 /* Set up limit values for motion vectors used to prevent
michael@0 143 * them extending outside the UMV borders
michael@0 144 */
michael@0 145 x->mv_col_min = -((mb_col * 16) + (VP8BORDERINPIXELS - 16));
michael@0 146 x->mv_col_max = ((cm->mb_cols - 1 - mb_col) * 16) + (VP8BORDERINPIXELS - 16);
michael@0 147 x->mv_row_min = -((mb_row * 16) + (VP8BORDERINPIXELS - 16));
michael@0 148 x->mv_row_max = ((cm->mb_rows - 1 - mb_row) * 16) + (VP8BORDERINPIXELS - 16);
michael@0 149
michael@0 150 xd->dst.y_buffer = cm->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset;
michael@0 151 xd->dst.u_buffer = cm->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset;
michael@0 152 xd->dst.v_buffer = cm->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset;
michael@0 153 xd->left_available = (mb_col != 0);
michael@0 154
michael@0 155 x->rddiv = cpi->RDDIV;
michael@0 156 x->rdmult = cpi->RDMULT;
michael@0 157
michael@0 158 /* Copy current mb to a buffer */
michael@0 159 vp8_copy_mem16x16(x->src.y_buffer, x->src.y_stride, x->thismb, 16);
michael@0 160
michael@0 161 if (cpi->oxcf.tuning == VP8_TUNE_SSIM)
michael@0 162 vp8_activity_masking(cpi, x);
michael@0 163
michael@0 164 /* Is segmentation enabled */
michael@0 165 /* MB level adjustment to quantizer */
michael@0 166 if (xd->segmentation_enabled)
michael@0 167 {
michael@0 168 /* Code to set segment id in xd->mbmi.segment_id for
michael@0 169 * current MB (with range checking)
michael@0 170 */
michael@0 171 if (cpi->segmentation_map[map_index + mb_col] <= 3)
michael@0 172 xd->mode_info_context->mbmi.segment_id = cpi->segmentation_map[map_index + mb_col];
michael@0 173 else
michael@0 174 xd->mode_info_context->mbmi.segment_id = 0;
michael@0 175
michael@0 176 vp8cx_mb_init_quantizer(cpi, x, 1);
michael@0 177 }
michael@0 178 else
michael@0 179 /* Set to Segment 0 by default */
michael@0 180 xd->mode_info_context->mbmi.segment_id = 0;
michael@0 181
michael@0 182 x->active_ptr = cpi->active_map + map_index + mb_col;
michael@0 183
michael@0 184 if (cm->frame_type == KEY_FRAME)
michael@0 185 {
michael@0 186 *totalrate += vp8cx_encode_intra_macroblock(cpi, x, &tp);
michael@0 187 #ifdef MODE_STATS
michael@0 188 y_modes[xd->mbmi.mode] ++;
michael@0 189 #endif
michael@0 190 }
michael@0 191 else
michael@0 192 {
michael@0 193 *totalrate += vp8cx_encode_inter_macroblock(cpi, x, &tp, recon_yoffset, recon_uvoffset, mb_row, mb_col);
michael@0 194
michael@0 195 #ifdef MODE_STATS
michael@0 196 inter_y_modes[xd->mbmi.mode] ++;
michael@0 197
michael@0 198 if (xd->mbmi.mode == SPLITMV)
michael@0 199 {
michael@0 200 int b;
michael@0 201
michael@0 202 for (b = 0; b < xd->mbmi.partition_count; b++)
michael@0 203 {
michael@0 204 inter_b_modes[x->partition->bmi[b].mode] ++;
michael@0 205 }
michael@0 206 }
michael@0 207
michael@0 208 #endif
michael@0 209
michael@0 210 /* Special case code for cyclic refresh
michael@0 211 * If cyclic update enabled then copy
michael@0 212 * xd->mbmi.segment_id; (which may have been updated
michael@0 213 * based on mode during
michael@0 214 * vp8cx_encode_inter_macroblock()) back into the
michael@0 215 * global segmentation map
michael@0 216 */
michael@0 217 if ((cpi->current_layer == 0) &&
michael@0 218 (cpi->cyclic_refresh_mode_enabled &&
michael@0 219 xd->segmentation_enabled))
michael@0 220 {
michael@0 221 const MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
michael@0 222 cpi->segmentation_map[map_index + mb_col] = mbmi->segment_id;
michael@0 223
michael@0 224 /* If the block has been refreshed mark it as clean
michael@0 225 * (the magnitude of the -ve influences how long it
michael@0 226 * will be before we consider another refresh):
michael@0 227 * Else if it was coded (last frame 0,0) and has
michael@0 228 * not already been refreshed then mark it as a
michael@0 229 * candidate for cleanup next time (marked 0) else
michael@0 230 * mark it as dirty (1).
michael@0 231 */
michael@0 232 if (mbmi->segment_id)
michael@0 233 cpi->cyclic_refresh_map[map_index + mb_col] = -1;
michael@0 234 else if ((mbmi->mode == ZEROMV) && (mbmi->ref_frame == LAST_FRAME))
michael@0 235 {
michael@0 236 if (cpi->cyclic_refresh_map[map_index + mb_col] == 1)
michael@0 237 cpi->cyclic_refresh_map[map_index + mb_col] = 0;
michael@0 238 }
michael@0 239 else
michael@0 240 cpi->cyclic_refresh_map[map_index + mb_col] = 1;
michael@0 241
michael@0 242 }
michael@0 243 }
michael@0 244
michael@0 245 #if CONFIG_REALTIME_ONLY & CONFIG_ONTHEFLY_BITPACKING
michael@0 246 /* pack tokens for this MB */
michael@0 247 {
michael@0 248 int tok_count = tp - tp_start;
michael@0 249 pack_tokens(w, tp_start, tok_count);
michael@0 250 }
michael@0 251 #else
michael@0 252 cpi->tplist[mb_row].stop = tp;
michael@0 253 #endif
michael@0 254 /* Increment pointer into gf usage flags structure. */
michael@0 255 x->gf_active_ptr++;
michael@0 256
michael@0 257 /* Increment the activity mask pointers. */
michael@0 258 x->mb_activity_ptr++;
michael@0 259
michael@0 260 /* adjust to the next column of macroblocks */
michael@0 261 x->src.y_buffer += 16;
michael@0 262 x->src.u_buffer += 8;
michael@0 263 x->src.v_buffer += 8;
michael@0 264
michael@0 265 recon_yoffset += 16;
michael@0 266 recon_uvoffset += 8;
michael@0 267
michael@0 268 /* Keep track of segment usage */
michael@0 269 segment_counts[xd->mode_info_context->mbmi.segment_id]++;
michael@0 270
michael@0 271 /* skip to next mb */
michael@0 272 xd->mode_info_context++;
michael@0 273 x->partition_info++;
michael@0 274 xd->above_context++;
michael@0 275 }
michael@0 276
michael@0 277 vp8_extend_mb_row( &cm->yv12_fb[dst_fb_idx],
michael@0 278 xd->dst.y_buffer + 16,
michael@0 279 xd->dst.u_buffer + 8,
michael@0 280 xd->dst.v_buffer + 8);
michael@0 281
michael@0 282 *current_mb_col = mb_col + nsync;
michael@0 283
michael@0 284 /* this is to account for the border */
michael@0 285 xd->mode_info_context++;
michael@0 286 x->partition_info++;
michael@0 287
michael@0 288 x->src.y_buffer += 16 * x->src.y_stride * (cpi->encoding_thread_count + 1) - 16 * cm->mb_cols;
michael@0 289 x->src.u_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
michael@0 290 x->src.v_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
michael@0 291
michael@0 292 xd->mode_info_context += xd->mode_info_stride * cpi->encoding_thread_count;
michael@0 293 x->partition_info += xd->mode_info_stride * cpi->encoding_thread_count;
michael@0 294 x->gf_active_ptr += cm->mb_cols * cpi->encoding_thread_count;
michael@0 295
michael@0 296 if (mb_row == cm->mb_rows - 1)
michael@0 297 {
michael@0 298 sem_post(&cpi->h_event_end_encoding); /* signal frame encoding end */
michael@0 299 }
michael@0 300 }
michael@0 301 }
michael@0 302 }
michael@0 303
michael@0 304 /* printf("exit thread %d\n", ithread); */
michael@0 305 return 0;
michael@0 306 }
michael@0 307
michael@0 308 static void setup_mbby_copy(MACROBLOCK *mbdst, MACROBLOCK *mbsrc)
michael@0 309 {
michael@0 310
michael@0 311 MACROBLOCK *x = mbsrc;
michael@0 312 MACROBLOCK *z = mbdst;
michael@0 313 int i;
michael@0 314
michael@0 315 z->ss = x->ss;
michael@0 316 z->ss_count = x->ss_count;
michael@0 317 z->searches_per_step = x->searches_per_step;
michael@0 318 z->errorperbit = x->errorperbit;
michael@0 319
michael@0 320 z->sadperbit16 = x->sadperbit16;
michael@0 321 z->sadperbit4 = x->sadperbit4;
michael@0 322
michael@0 323 /*
michael@0 324 z->mv_col_min = x->mv_col_min;
michael@0 325 z->mv_col_max = x->mv_col_max;
michael@0 326 z->mv_row_min = x->mv_row_min;
michael@0 327 z->mv_row_max = x->mv_row_max;
michael@0 328 */
michael@0 329
michael@0 330 z->short_fdct4x4 = x->short_fdct4x4;
michael@0 331 z->short_fdct8x4 = x->short_fdct8x4;
michael@0 332 z->short_walsh4x4 = x->short_walsh4x4;
michael@0 333 z->quantize_b = x->quantize_b;
michael@0 334 z->quantize_b_pair = x->quantize_b_pair;
michael@0 335 z->optimize = x->optimize;
michael@0 336
michael@0 337 /*
michael@0 338 z->mvc = x->mvc;
michael@0 339 z->src.y_buffer = x->src.y_buffer;
michael@0 340 z->src.u_buffer = x->src.u_buffer;
michael@0 341 z->src.v_buffer = x->src.v_buffer;
michael@0 342 */
michael@0 343
michael@0 344 z->mvcost[0] = x->mvcost[0];
michael@0 345 z->mvcost[1] = x->mvcost[1];
michael@0 346 z->mvsadcost[0] = x->mvsadcost[0];
michael@0 347 z->mvsadcost[1] = x->mvsadcost[1];
michael@0 348
michael@0 349 z->token_costs = x->token_costs;
michael@0 350 z->inter_bmode_costs = x->inter_bmode_costs;
michael@0 351 z->mbmode_cost = x->mbmode_cost;
michael@0 352 z->intra_uv_mode_cost = x->intra_uv_mode_cost;
michael@0 353 z->bmode_costs = x->bmode_costs;
michael@0 354
michael@0 355 for (i = 0; i < 25; i++)
michael@0 356 {
michael@0 357 z->block[i].quant = x->block[i].quant;
michael@0 358 z->block[i].quant_fast = x->block[i].quant_fast;
michael@0 359 z->block[i].quant_shift = x->block[i].quant_shift;
michael@0 360 z->block[i].zbin = x->block[i].zbin;
michael@0 361 z->block[i].zrun_zbin_boost = x->block[i].zrun_zbin_boost;
michael@0 362 z->block[i].round = x->block[i].round;
michael@0 363 z->block[i].src_stride = x->block[i].src_stride;
michael@0 364 }
michael@0 365
michael@0 366 z->q_index = x->q_index;
michael@0 367 z->act_zbin_adj = x->act_zbin_adj;
michael@0 368 z->last_act_zbin_adj = x->last_act_zbin_adj;
michael@0 369
michael@0 370 {
michael@0 371 MACROBLOCKD *xd = &x->e_mbd;
michael@0 372 MACROBLOCKD *zd = &z->e_mbd;
michael@0 373
michael@0 374 /*
michael@0 375 zd->mode_info_context = xd->mode_info_context;
michael@0 376 zd->mode_info = xd->mode_info;
michael@0 377
michael@0 378 zd->mode_info_stride = xd->mode_info_stride;
michael@0 379 zd->frame_type = xd->frame_type;
michael@0 380 zd->up_available = xd->up_available ;
michael@0 381 zd->left_available = xd->left_available;
michael@0 382 zd->left_context = xd->left_context;
michael@0 383 zd->last_frame_dc = xd->last_frame_dc;
michael@0 384 zd->last_frame_dccons = xd->last_frame_dccons;
michael@0 385 zd->gold_frame_dc = xd->gold_frame_dc;
michael@0 386 zd->gold_frame_dccons = xd->gold_frame_dccons;
michael@0 387 zd->mb_to_left_edge = xd->mb_to_left_edge;
michael@0 388 zd->mb_to_right_edge = xd->mb_to_right_edge;
michael@0 389 zd->mb_to_top_edge = xd->mb_to_top_edge ;
michael@0 390 zd->mb_to_bottom_edge = xd->mb_to_bottom_edge;
michael@0 391 zd->gf_active_ptr = xd->gf_active_ptr;
michael@0 392 zd->frames_since_golden = xd->frames_since_golden;
michael@0 393 zd->frames_till_alt_ref_frame = xd->frames_till_alt_ref_frame;
michael@0 394 */
michael@0 395 zd->subpixel_predict = xd->subpixel_predict;
michael@0 396 zd->subpixel_predict8x4 = xd->subpixel_predict8x4;
michael@0 397 zd->subpixel_predict8x8 = xd->subpixel_predict8x8;
michael@0 398 zd->subpixel_predict16x16 = xd->subpixel_predict16x16;
michael@0 399 zd->segmentation_enabled = xd->segmentation_enabled;
michael@0 400 zd->mb_segement_abs_delta = xd->mb_segement_abs_delta;
michael@0 401 vpx_memcpy(zd->segment_feature_data, xd->segment_feature_data,
michael@0 402 sizeof(xd->segment_feature_data));
michael@0 403
michael@0 404 vpx_memcpy(zd->dequant_y1_dc, xd->dequant_y1_dc,
michael@0 405 sizeof(xd->dequant_y1_dc));
michael@0 406 vpx_memcpy(zd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1));
michael@0 407 vpx_memcpy(zd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2));
michael@0 408 vpx_memcpy(zd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv));
michael@0 409
michael@0 410 #if 1
michael@0 411 /*TODO: Remove dequant from BLOCKD. This is a temporary solution until
michael@0 412 * the quantizer code uses a passed in pointer to the dequant constants.
michael@0 413 * This will also require modifications to the x86 and neon assembly.
michael@0 414 * */
michael@0 415 for (i = 0; i < 16; i++)
michael@0 416 zd->block[i].dequant = zd->dequant_y1;
michael@0 417 for (i = 16; i < 24; i++)
michael@0 418 zd->block[i].dequant = zd->dequant_uv;
michael@0 419 zd->block[24].dequant = zd->dequant_y2;
michael@0 420 #endif
michael@0 421
michael@0 422
michael@0 423 vpx_memcpy(z->rd_threshes, x->rd_threshes, sizeof(x->rd_threshes));
michael@0 424 vpx_memcpy(z->rd_thresh_mult, x->rd_thresh_mult,
michael@0 425 sizeof(x->rd_thresh_mult));
michael@0 426
michael@0 427 z->zbin_over_quant = x->zbin_over_quant;
michael@0 428 z->zbin_mode_boost_enabled = x->zbin_mode_boost_enabled;
michael@0 429 z->zbin_mode_boost = x->zbin_mode_boost;
michael@0 430
michael@0 431 vpx_memset(z->error_bins, 0, sizeof(z->error_bins));
michael@0 432 }
michael@0 433 }
michael@0 434
michael@0 435 void vp8cx_init_mbrthread_data(VP8_COMP *cpi,
michael@0 436 MACROBLOCK *x,
michael@0 437 MB_ROW_COMP *mbr_ei,
michael@0 438 int count
michael@0 439 )
michael@0 440 {
michael@0 441
michael@0 442 VP8_COMMON *const cm = & cpi->common;
michael@0 443 MACROBLOCKD *const xd = & x->e_mbd;
michael@0 444 int i;
michael@0 445
michael@0 446 for (i = 0; i < count; i++)
michael@0 447 {
michael@0 448 MACROBLOCK *mb = & mbr_ei[i].mb;
michael@0 449 MACROBLOCKD *mbd = &mb->e_mbd;
michael@0 450
michael@0 451 mbd->subpixel_predict = xd->subpixel_predict;
michael@0 452 mbd->subpixel_predict8x4 = xd->subpixel_predict8x4;
michael@0 453 mbd->subpixel_predict8x8 = xd->subpixel_predict8x8;
michael@0 454 mbd->subpixel_predict16x16 = xd->subpixel_predict16x16;
michael@0 455 mb->gf_active_ptr = x->gf_active_ptr;
michael@0 456
michael@0 457 vpx_memset(mbr_ei[i].segment_counts, 0, sizeof(mbr_ei[i].segment_counts));
michael@0 458 mbr_ei[i].totalrate = 0;
michael@0 459
michael@0 460 mb->partition_info = x->pi + x->e_mbd.mode_info_stride * (i + 1);
michael@0 461
michael@0 462 mbd->mode_info_context = cm->mi + x->e_mbd.mode_info_stride * (i + 1);
michael@0 463 mbd->mode_info_stride = cm->mode_info_stride;
michael@0 464
michael@0 465 mbd->frame_type = cm->frame_type;
michael@0 466
michael@0 467 mb->src = * cpi->Source;
michael@0 468 mbd->pre = cm->yv12_fb[cm->lst_fb_idx];
michael@0 469 mbd->dst = cm->yv12_fb[cm->new_fb_idx];
michael@0 470
michael@0 471 mb->src.y_buffer += 16 * x->src.y_stride * (i + 1);
michael@0 472 mb->src.u_buffer += 8 * x->src.uv_stride * (i + 1);
michael@0 473 mb->src.v_buffer += 8 * x->src.uv_stride * (i + 1);
michael@0 474
michael@0 475 vp8_build_block_offsets(mb);
michael@0 476
michael@0 477 mbd->left_context = &cm->left_context;
michael@0 478 mb->mvc = cm->fc.mvc;
michael@0 479
michael@0 480 setup_mbby_copy(&mbr_ei[i].mb, x);
michael@0 481
michael@0 482 mbd->fullpixel_mask = 0xffffffff;
michael@0 483 if(cm->full_pixel)
michael@0 484 mbd->fullpixel_mask = 0xfffffff8;
michael@0 485
michael@0 486 vp8_zero(mb->coef_counts);
michael@0 487 vp8_zero(x->ymode_count);
michael@0 488 mb->skip_true_count = 0;
michael@0 489 vp8_zero(mb->MVcount);
michael@0 490 mb->prediction_error = 0;
michael@0 491 mb->intra_error = 0;
michael@0 492 vp8_zero(mb->count_mb_ref_frame_usage);
michael@0 493 mb->mbs_tested_so_far = 0;
michael@0 494 }
michael@0 495 }
michael@0 496
michael@0 497 int vp8cx_create_encoder_threads(VP8_COMP *cpi)
michael@0 498 {
michael@0 499 const VP8_COMMON * cm = &cpi->common;
michael@0 500
michael@0 501 cpi->b_multi_threaded = 0;
michael@0 502 cpi->encoding_thread_count = 0;
michael@0 503 cpi->b_lpf_running = 0;
michael@0 504
michael@0 505 if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
michael@0 506 {
michael@0 507 int ithread;
michael@0 508 int th_count = cpi->oxcf.multi_threaded - 1;
michael@0 509 int rc = 0;
michael@0 510
michael@0 511 /* don't allocate more threads than cores available */
michael@0 512 if (cpi->oxcf.multi_threaded > cm->processor_core_count)
michael@0 513 th_count = cm->processor_core_count - 1;
michael@0 514
michael@0 515 /* we have th_count + 1 (main) threads processing one row each */
michael@0 516 /* no point to have more threads than the sync range allows */
michael@0 517 if(th_count > ((cm->mb_cols / cpi->mt_sync_range) - 1))
michael@0 518 {
michael@0 519 th_count = (cm->mb_cols / cpi->mt_sync_range) - 1;
michael@0 520 }
michael@0 521
michael@0 522 if(th_count == 0)
michael@0 523 return 0;
michael@0 524
michael@0 525 CHECK_MEM_ERROR(cpi->h_encoding_thread,
michael@0 526 vpx_malloc(sizeof(pthread_t) * th_count));
michael@0 527 CHECK_MEM_ERROR(cpi->h_event_start_encoding,
michael@0 528 vpx_malloc(sizeof(sem_t) * th_count));
michael@0 529 CHECK_MEM_ERROR(cpi->mb_row_ei,
michael@0 530 vpx_memalign(32, sizeof(MB_ROW_COMP) * th_count));
michael@0 531 vpx_memset(cpi->mb_row_ei, 0, sizeof(MB_ROW_COMP) * th_count);
michael@0 532 CHECK_MEM_ERROR(cpi->en_thread_data,
michael@0 533 vpx_malloc(sizeof(ENCODETHREAD_DATA) * th_count));
michael@0 534
michael@0 535 sem_init(&cpi->h_event_end_encoding, 0, 0);
michael@0 536
michael@0 537 cpi->b_multi_threaded = 1;
michael@0 538 cpi->encoding_thread_count = th_count;
michael@0 539
michael@0 540 /*
michael@0 541 printf("[VP8:] multi_threaded encoding is enabled with %d threads\n\n",
michael@0 542 (cpi->encoding_thread_count +1));
michael@0 543 */
michael@0 544
michael@0 545 for (ithread = 0; ithread < th_count; ithread++)
michael@0 546 {
michael@0 547 ENCODETHREAD_DATA *ethd = &cpi->en_thread_data[ithread];
michael@0 548
michael@0 549 /* Setup block ptrs and offsets */
michael@0 550 vp8_setup_block_ptrs(&cpi->mb_row_ei[ithread].mb);
michael@0 551 vp8_setup_block_dptrs(&cpi->mb_row_ei[ithread].mb.e_mbd);
michael@0 552
michael@0 553 sem_init(&cpi->h_event_start_encoding[ithread], 0, 0);
michael@0 554
michael@0 555 ethd->ithread = ithread;
michael@0 556 ethd->ptr1 = (void *)cpi;
michael@0 557 ethd->ptr2 = (void *)&cpi->mb_row_ei[ithread];
michael@0 558
michael@0 559 rc = pthread_create(&cpi->h_encoding_thread[ithread], 0,
michael@0 560 thread_encoding_proc, ethd);
michael@0 561 if(rc)
michael@0 562 break;
michael@0 563 }
michael@0 564
michael@0 565 if(rc)
michael@0 566 {
michael@0 567 /* shutdown other threads */
michael@0 568 cpi->b_multi_threaded = 0;
michael@0 569 for(--ithread; ithread >= 0; ithread--)
michael@0 570 {
michael@0 571 pthread_join(cpi->h_encoding_thread[ithread], 0);
michael@0 572 sem_destroy(&cpi->h_event_start_encoding[ithread]);
michael@0 573 }
michael@0 574 sem_destroy(&cpi->h_event_end_encoding);
michael@0 575
michael@0 576 /* free thread related resources */
michael@0 577 vpx_free(cpi->h_event_start_encoding);
michael@0 578 vpx_free(cpi->h_encoding_thread);
michael@0 579 vpx_free(cpi->mb_row_ei);
michael@0 580 vpx_free(cpi->en_thread_data);
michael@0 581
michael@0 582 return -1;
michael@0 583 }
michael@0 584
michael@0 585
michael@0 586 {
michael@0 587 LPFTHREAD_DATA * lpfthd = &cpi->lpf_thread_data;
michael@0 588
michael@0 589 sem_init(&cpi->h_event_start_lpf, 0, 0);
michael@0 590 sem_init(&cpi->h_event_end_lpf, 0, 0);
michael@0 591
michael@0 592 lpfthd->ptr1 = (void *)cpi;
michael@0 593 rc = pthread_create(&cpi->h_filter_thread, 0, thread_loopfilter,
michael@0 594 lpfthd);
michael@0 595
michael@0 596 if(rc)
michael@0 597 {
michael@0 598 /* shutdown other threads */
michael@0 599 cpi->b_multi_threaded = 0;
michael@0 600 for(--ithread; ithread >= 0; ithread--)
michael@0 601 {
michael@0 602 sem_post(&cpi->h_event_start_encoding[ithread]);
michael@0 603 pthread_join(cpi->h_encoding_thread[ithread], 0);
michael@0 604 sem_destroy(&cpi->h_event_start_encoding[ithread]);
michael@0 605 }
michael@0 606 sem_destroy(&cpi->h_event_end_encoding);
michael@0 607 sem_destroy(&cpi->h_event_end_lpf);
michael@0 608 sem_destroy(&cpi->h_event_start_lpf);
michael@0 609
michael@0 610 /* free thread related resources */
michael@0 611 vpx_free(cpi->h_event_start_encoding);
michael@0 612 vpx_free(cpi->h_encoding_thread);
michael@0 613 vpx_free(cpi->mb_row_ei);
michael@0 614 vpx_free(cpi->en_thread_data);
michael@0 615
michael@0 616 return -2;
michael@0 617 }
michael@0 618 }
michael@0 619 }
michael@0 620 return 0;
michael@0 621 }
michael@0 622
michael@0 623 void vp8cx_remove_encoder_threads(VP8_COMP *cpi)
michael@0 624 {
michael@0 625 if (cpi->b_multi_threaded)
michael@0 626 {
michael@0 627 /* shutdown other threads */
michael@0 628 cpi->b_multi_threaded = 0;
michael@0 629 {
michael@0 630 int i;
michael@0 631
michael@0 632 for (i = 0; i < cpi->encoding_thread_count; i++)
michael@0 633 {
michael@0 634 sem_post(&cpi->h_event_start_encoding[i]);
michael@0 635 pthread_join(cpi->h_encoding_thread[i], 0);
michael@0 636
michael@0 637 sem_destroy(&cpi->h_event_start_encoding[i]);
michael@0 638 }
michael@0 639
michael@0 640 sem_post(&cpi->h_event_start_lpf);
michael@0 641 pthread_join(cpi->h_filter_thread, 0);
michael@0 642 }
michael@0 643
michael@0 644 sem_destroy(&cpi->h_event_end_encoding);
michael@0 645 sem_destroy(&cpi->h_event_end_lpf);
michael@0 646 sem_destroy(&cpi->h_event_start_lpf);
michael@0 647
michael@0 648 /* free thread related resources */
michael@0 649 vpx_free(cpi->h_event_start_encoding);
michael@0 650 vpx_free(cpi->h_encoding_thread);
michael@0 651 vpx_free(cpi->mb_row_ei);
michael@0 652 vpx_free(cpi->en_thread_data);
michael@0 653 }
michael@0 654 }
michael@0 655 #endif

Mercurial Repository: The Tor Browser

Europalab SCM Repository Server

mercurial