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 #include "vpx_config.h"
13 #include "vpx_rtcd.h"
14 #if !defined(WIN32) && CONFIG_OS_SUPPORT == 1
15 # include <unistd.h>
16 #endif
17 #include "onyxd_int.h"
18 #include "vpx_mem/vpx_mem.h"
19 #include "vp8/common/threading.h"
20
21 #include "vp8/common/loopfilter.h"
22 #include "vp8/common/extend.h"
23 #include "vpx_ports/vpx_timer.h"
24 #include "detokenize.h"
25 #include "vp8/common/reconintra4x4.h"
26 #include "vp8/common/reconinter.h"
27 #include "vp8/common/setupintrarecon.h"
28 #if CONFIG_ERROR_CONCEALMENT
29 #include "error_concealment.h"
30 #endif
31
32 #define CALLOC_ARRAY(p, n) CHECK_MEM_ERROR((p), vpx_calloc(sizeof(*(p)), (n)))
33 #define CALLOC_ARRAY_ALIGNED(p, n, algn) do { \
34 CHECK_MEM_ERROR((p), vpx_memalign((algn), sizeof(*(p)) * (n))); \
35 memset((p), 0, (n) * sizeof(*(p))); \
36 } while (0)
37
38
39 extern void vp8_mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd);
40
setup_decoding_thread_data(VP8D_COMP * pbi,MACROBLOCKD * xd,MB_ROW_DEC * mbrd,int count)41 static void setup_decoding_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd, MB_ROW_DEC *mbrd, int count)
42 {
43 VP8_COMMON *const pc = & pbi->common;
44 int i;
45
46 for (i = 0; i < count; i++)
47 {
48 MACROBLOCKD *mbd = &mbrd[i].mbd;
49 mbd->subpixel_predict = xd->subpixel_predict;
50 mbd->subpixel_predict8x4 = xd->subpixel_predict8x4;
51 mbd->subpixel_predict8x8 = xd->subpixel_predict8x8;
52 mbd->subpixel_predict16x16 = xd->subpixel_predict16x16;
53
54 mbd->mode_info_context = pc->mi + pc->mode_info_stride * (i + 1);
55 mbd->mode_info_stride = pc->mode_info_stride;
56
57 mbd->frame_type = pc->frame_type;
58 mbd->pre = xd->pre;
59 mbd->dst = xd->dst;
60
61 mbd->segmentation_enabled = xd->segmentation_enabled;
62 mbd->mb_segement_abs_delta = xd->mb_segement_abs_delta;
63 vpx_memcpy(mbd->segment_feature_data, xd->segment_feature_data, sizeof(xd->segment_feature_data));
64
65 /*signed char ref_lf_deltas[MAX_REF_LF_DELTAS];*/
66 vpx_memcpy(mbd->ref_lf_deltas, xd->ref_lf_deltas, sizeof(xd->ref_lf_deltas));
67 /*signed char mode_lf_deltas[MAX_MODE_LF_DELTAS];*/
68 vpx_memcpy(mbd->mode_lf_deltas, xd->mode_lf_deltas, sizeof(xd->mode_lf_deltas));
69 /*unsigned char mode_ref_lf_delta_enabled;
70 unsigned char mode_ref_lf_delta_update;*/
71 mbd->mode_ref_lf_delta_enabled = xd->mode_ref_lf_delta_enabled;
72 mbd->mode_ref_lf_delta_update = xd->mode_ref_lf_delta_update;
73
74 mbd->current_bc = &pbi->mbc[0];
75
76 vpx_memcpy(mbd->dequant_y1_dc, xd->dequant_y1_dc, sizeof(xd->dequant_y1_dc));
77 vpx_memcpy(mbd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1));
78 vpx_memcpy(mbd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2));
79 vpx_memcpy(mbd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv));
80
81 mbd->fullpixel_mask = 0xffffffff;
82
83 if (pc->full_pixel)
84 mbd->fullpixel_mask = 0xfffffff8;
85
86 }
87
88 for (i = 0; i < pc->mb_rows; i++)
89 pbi->mt_current_mb_col[i] = -1;
90 }
91
mt_decode_macroblock(VP8D_COMP * pbi,MACROBLOCKD * xd,unsigned int mb_idx)92 static void mt_decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd,
93 unsigned int mb_idx)
94 {
95 MB_PREDICTION_MODE mode;
96 int i;
97 #if CONFIG_ERROR_CONCEALMENT
98 int corruption_detected = 0;
99 #endif
100
101 if (xd->mode_info_context->mbmi.mb_skip_coeff)
102 {
103 vp8_reset_mb_tokens_context(xd);
104 }
105 else if (!vp8dx_bool_error(xd->current_bc))
106 {
107 int eobtotal;
108 eobtotal = vp8_decode_mb_tokens(pbi, xd);
109
110 /* Special case: Force the loopfilter to skip when eobtotal is zero */
111 xd->mode_info_context->mbmi.mb_skip_coeff = (eobtotal==0);
112 }
113
114 mode = xd->mode_info_context->mbmi.mode;
115
116 if (xd->segmentation_enabled)
117 vp8_mb_init_dequantizer(pbi, xd);
118
119
120 #if CONFIG_ERROR_CONCEALMENT
121
122 if(pbi->ec_active)
123 {
124 int throw_residual;
125 /* When we have independent partitions we can apply residual even
126 * though other partitions within the frame are corrupt.
127 */
128 throw_residual = (!pbi->independent_partitions &&
129 pbi->frame_corrupt_residual);
130 throw_residual = (throw_residual || vp8dx_bool_error(xd->current_bc));
131
132 if ((mb_idx >= pbi->mvs_corrupt_from_mb || throw_residual))
133 {
134 /* MB with corrupt residuals or corrupt mode/motion vectors.
135 * Better to use the predictor as reconstruction.
136 */
137 pbi->frame_corrupt_residual = 1;
138 vpx_memset(xd->qcoeff, 0, sizeof(xd->qcoeff));
139 vp8_conceal_corrupt_mb(xd);
140
141
142 corruption_detected = 1;
143
144 /* force idct to be skipped for B_PRED and use the
145 * prediction only for reconstruction
146 * */
147 vpx_memset(xd->eobs, 0, 25);
148 }
149 }
150 #endif
151
152 /* do prediction */
153 if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME)
154 {
155 vp8_build_intra_predictors_mbuv_s(xd,
156 xd->recon_above[1],
157 xd->recon_above[2],
158 xd->recon_left[1],
159 xd->recon_left[2],
160 xd->recon_left_stride[1],
161 xd->dst.u_buffer, xd->dst.v_buffer,
162 xd->dst.uv_stride);
163
164 if (mode != B_PRED)
165 {
166 vp8_build_intra_predictors_mby_s(xd,
167 xd->recon_above[0],
168 xd->recon_left[0],
169 xd->recon_left_stride[0],
170 xd->dst.y_buffer,
171 xd->dst.y_stride);
172 }
173 else
174 {
175 short *DQC = xd->dequant_y1;
176 int dst_stride = xd->dst.y_stride;
177
178 /* clear out residual eob info */
179 if(xd->mode_info_context->mbmi.mb_skip_coeff)
180 vpx_memset(xd->eobs, 0, 25);
181
182 intra_prediction_down_copy(xd, xd->recon_above[0] + 16);
183
184 for (i = 0; i < 16; i++)
185 {
186 BLOCKD *b = &xd->block[i];
187 unsigned char *dst = xd->dst.y_buffer + b->offset;
188 B_PREDICTION_MODE b_mode =
189 xd->mode_info_context->bmi[i].as_mode;
190 unsigned char *Above;
191 unsigned char *yleft;
192 int left_stride;
193 unsigned char top_left;
194
195 /*Caution: For some b_mode, it needs 8 pixels (4 above + 4 above-right).*/
196 if (i < 4 && pbi->common.filter_level)
197 Above = xd->recon_above[0] + b->offset;
198 else
199 Above = dst - dst_stride;
200
201 if (i%4==0 && pbi->common.filter_level)
202 {
203 yleft = xd->recon_left[0] + i;
204 left_stride = 1;
205 }
206 else
207 {
208 yleft = dst - 1;
209 left_stride = dst_stride;
210 }
211
212 if ((i==4 || i==8 || i==12) && pbi->common.filter_level)
213 top_left = *(xd->recon_left[0] + i - 1);
214 else
215 top_left = Above[-1];
216
217 vp8_intra4x4_predict(Above, yleft, left_stride,
218 b_mode, dst, dst_stride, top_left);
219
220 if (xd->eobs[i] )
221 {
222 if (xd->eobs[i] > 1)
223 {
224 vp8_dequant_idct_add(b->qcoeff, DQC, dst, dst_stride);
225 }
226 else
227 {
228 vp8_dc_only_idct_add(b->qcoeff[0] * DQC[0],
229 dst, dst_stride, dst, dst_stride);
230 ((int *)b->qcoeff)[0] = 0;
231 }
232 }
233 }
234 }
235 }
236 else
237 {
238 vp8_build_inter_predictors_mb(xd);
239 }
240
241
242 #if CONFIG_ERROR_CONCEALMENT
243 if (corruption_detected)
244 {
245 return;
246 }
247 #endif
248
249 if(!xd->mode_info_context->mbmi.mb_skip_coeff)
250 {
251 /* dequantization and idct */
252 if (mode != B_PRED)
253 {
254 short *DQC = xd->dequant_y1;
255
256 if (mode != SPLITMV)
257 {
258 BLOCKD *b = &xd->block[24];
259
260 /* do 2nd order transform on the dc block */
261 if (xd->eobs[24] > 1)
262 {
263 vp8_dequantize_b(b, xd->dequant_y2);
264
265 vp8_short_inv_walsh4x4(&b->dqcoeff[0],
266 xd->qcoeff);
267 ((int *)b->qcoeff)[0] = 0;
268 ((int *)b->qcoeff)[1] = 0;
269 ((int *)b->qcoeff)[2] = 0;
270 ((int *)b->qcoeff)[3] = 0;
271 ((int *)b->qcoeff)[4] = 0;
272 ((int *)b->qcoeff)[5] = 0;
273 ((int *)b->qcoeff)[6] = 0;
274 ((int *)b->qcoeff)[7] = 0;
275 }
276 else
277 {
278 b->dqcoeff[0] = b->qcoeff[0] * xd->dequant_y2[0];
279 vp8_short_inv_walsh4x4_1(&b->dqcoeff[0],
280 xd->qcoeff);
281 ((int *)b->qcoeff)[0] = 0;
282 }
283
284 /* override the dc dequant constant in order to preserve the
285 * dc components
286 */
287 DQC = xd->dequant_y1_dc;
288 }
289
290 vp8_dequant_idct_add_y_block
291 (xd->qcoeff, DQC,
292 xd->dst.y_buffer,
293 xd->dst.y_stride, xd->eobs);
294 }
295
296 vp8_dequant_idct_add_uv_block
297 (xd->qcoeff+16*16, xd->dequant_uv,
298 xd->dst.u_buffer, xd->dst.v_buffer,
299 xd->dst.uv_stride, xd->eobs+16);
300 }
301 }
302
mt_decode_mb_rows(VP8D_COMP * pbi,MACROBLOCKD * xd,int start_mb_row)303 static void mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd, int start_mb_row)
304 {
305 volatile const int *last_row_current_mb_col;
306 volatile int *current_mb_col;
307 int mb_row;
308 VP8_COMMON *pc = &pbi->common;
309 const int nsync = pbi->sync_range;
310 const int first_row_no_sync_above = pc->mb_cols + nsync;
311 int num_part = 1 << pbi->common.multi_token_partition;
312 int last_mb_row = start_mb_row;
313
314 YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
315 YV12_BUFFER_CONFIG *yv12_fb_lst = pbi->dec_fb_ref[LAST_FRAME];
316
317 int recon_y_stride = yv12_fb_new->y_stride;
318 int recon_uv_stride = yv12_fb_new->uv_stride;
319
320 unsigned char *ref_buffer[MAX_REF_FRAMES][3];
321 unsigned char *dst_buffer[3];
322 int i;
323 int ref_fb_corrupted[MAX_REF_FRAMES];
324
325 ref_fb_corrupted[INTRA_FRAME] = 0;
326
327 for(i = 1; i < MAX_REF_FRAMES; i++)
328 {
329 YV12_BUFFER_CONFIG *this_fb = pbi->dec_fb_ref[i];
330
331 ref_buffer[i][0] = this_fb->y_buffer;
332 ref_buffer[i][1] = this_fb->u_buffer;
333 ref_buffer[i][2] = this_fb->v_buffer;
334
335 ref_fb_corrupted[i] = this_fb->corrupted;
336 }
337
338 dst_buffer[0] = yv12_fb_new->y_buffer;
339 dst_buffer[1] = yv12_fb_new->u_buffer;
340 dst_buffer[2] = yv12_fb_new->v_buffer;
341
342 xd->up_available = (start_mb_row != 0);
343
344 for (mb_row = start_mb_row; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1))
345 {
346 int i;
347 int recon_yoffset, recon_uvoffset;
348 int mb_col;
349 int filter_level;
350 loop_filter_info_n *lfi_n = &pc->lf_info;
351
352 /* save last row processed by this thread */
353 last_mb_row = mb_row;
354 /* select bool coder for current partition */
355 xd->current_bc = &pbi->mbc[mb_row%num_part];
356
357 if (mb_row > 0)
358 last_row_current_mb_col = &pbi->mt_current_mb_col[mb_row -1];
359 else
360 last_row_current_mb_col = &first_row_no_sync_above;
361
362 current_mb_col = &pbi->mt_current_mb_col[mb_row];
363
364 recon_yoffset = mb_row * recon_y_stride * 16;
365 recon_uvoffset = mb_row * recon_uv_stride * 8;
366
367 /* reset contexts */
368 xd->above_context = pc->above_context;
369 vpx_memset(xd->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
370
371 xd->left_available = 0;
372
373 xd->mb_to_top_edge = -((mb_row * 16)) << 3;
374 xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3;
375
376 if (pbi->common.filter_level)
377 {
378 xd->recon_above[0] = pbi->mt_yabove_row[mb_row] + 0*16 +32;
379 xd->recon_above[1] = pbi->mt_uabove_row[mb_row] + 0*8 +16;
380 xd->recon_above[2] = pbi->mt_vabove_row[mb_row] + 0*8 +16;
381
382 xd->recon_left[0] = pbi->mt_yleft_col[mb_row];
383 xd->recon_left[1] = pbi->mt_uleft_col[mb_row];
384 xd->recon_left[2] = pbi->mt_vleft_col[mb_row];
385
386 /* TODO: move to outside row loop */
387 xd->recon_left_stride[0] = 1;
388 xd->recon_left_stride[1] = 1;
389 }
390 else
391 {
392 xd->recon_above[0] = dst_buffer[0] + recon_yoffset;
393 xd->recon_above[1] = dst_buffer[1] + recon_uvoffset;
394 xd->recon_above[2] = dst_buffer[2] + recon_uvoffset;
395
396 xd->recon_left[0] = xd->recon_above[0] - 1;
397 xd->recon_left[1] = xd->recon_above[1] - 1;
398 xd->recon_left[2] = xd->recon_above[2] - 1;
399
400 xd->recon_above[0] -= xd->dst.y_stride;
401 xd->recon_above[1] -= xd->dst.uv_stride;
402 xd->recon_above[2] -= xd->dst.uv_stride;
403
404 /* TODO: move to outside row loop */
405 xd->recon_left_stride[0] = xd->dst.y_stride;
406 xd->recon_left_stride[1] = xd->dst.uv_stride;
407
408 setup_intra_recon_left(xd->recon_left[0], xd->recon_left[1],
409 xd->recon_left[2], xd->dst.y_stride,
410 xd->dst.uv_stride);
411 }
412
413 for (mb_col = 0; mb_col < pc->mb_cols; mb_col++)
414 {
415 *current_mb_col = mb_col - 1;
416
417 if ((mb_col & (nsync - 1)) == 0)
418 {
419 while (mb_col > (*last_row_current_mb_col - nsync))
420 {
421 x86_pause_hint();
422 thread_sleep(0);
423 }
424 }
425
426 /* Distance of MB to the various image edges.
427 * These are specified to 8th pel as they are always
428 * compared to values that are in 1/8th pel units.
429 */
430 xd->mb_to_left_edge = -((mb_col * 16) << 3);
431 xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3;
432
433 #if CONFIG_ERROR_CONCEALMENT
434 {
435 int corrupt_residual =
436 (!pbi->independent_partitions &&
437 pbi->frame_corrupt_residual) ||
438 vp8dx_bool_error(xd->current_bc);
439 if (pbi->ec_active &&
440 (xd->mode_info_context->mbmi.ref_frame ==
441 INTRA_FRAME) &&
442 corrupt_residual)
443 {
444 /* We have an intra block with corrupt
445 * coefficients, better to conceal with an inter
446 * block.
447 * Interpolate MVs from neighboring MBs
448 *
449 * Note that for the first mb with corrupt
450 * residual in a frame, we might not discover
451 * that before decoding the residual. That
452 * happens after this check, and therefore no
453 * inter concealment will be done.
454 */
455 vp8_interpolate_motion(xd,
456 mb_row, mb_col,
457 pc->mb_rows, pc->mb_cols,
458 pc->mode_info_stride);
459 }
460 }
461 #endif
462
463
464 xd->dst.y_buffer = dst_buffer[0] + recon_yoffset;
465 xd->dst.u_buffer = dst_buffer[1] + recon_uvoffset;
466 xd->dst.v_buffer = dst_buffer[2] + recon_uvoffset;
467
468 xd->pre.y_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][0] + recon_yoffset;
469 xd->pre.u_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][1] + recon_uvoffset;
470 xd->pre.v_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][2] + recon_uvoffset;
471
472 /* propagate errors from reference frames */
473 xd->corrupted |= ref_fb_corrupted[xd->mode_info_context->mbmi.ref_frame];
474
475 mt_decode_macroblock(pbi, xd, 0);
476
477 xd->left_available = 1;
478
479 /* check if the boolean decoder has suffered an error */
480 xd->corrupted |= vp8dx_bool_error(xd->current_bc);
481
482 xd->recon_above[0] += 16;
483 xd->recon_above[1] += 8;
484 xd->recon_above[2] += 8;
485
486 if (!pbi->common.filter_level)
487 {
488 xd->recon_left[0] += 16;
489 xd->recon_left[1] += 8;
490 xd->recon_left[2] += 8;
491 }
492
493 if (pbi->common.filter_level)
494 {
495 int skip_lf = (xd->mode_info_context->mbmi.mode != B_PRED &&
496 xd->mode_info_context->mbmi.mode != SPLITMV &&
497 xd->mode_info_context->mbmi.mb_skip_coeff);
498
499 const int mode_index = lfi_n->mode_lf_lut[xd->mode_info_context->mbmi.mode];
500 const int seg = xd->mode_info_context->mbmi.segment_id;
501 const int ref_frame = xd->mode_info_context->mbmi.ref_frame;
502
503 filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
504
505 if( mb_row != pc->mb_rows-1 )
506 {
507 /* Save decoded MB last row data for next-row decoding */
508 vpx_memcpy((pbi->mt_yabove_row[mb_row + 1] + 32 + mb_col*16), (xd->dst.y_buffer + 15 * recon_y_stride), 16);
509 vpx_memcpy((pbi->mt_uabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.u_buffer + 7 * recon_uv_stride), 8);
510 vpx_memcpy((pbi->mt_vabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.v_buffer + 7 * recon_uv_stride), 8);
511 }
512
513 /* save left_col for next MB decoding */
514 if(mb_col != pc->mb_cols-1)
515 {
516 MODE_INFO *next = xd->mode_info_context +1;
517
518 if (next->mbmi.ref_frame == INTRA_FRAME)
519 {
520 for (i = 0; i < 16; i++)
521 pbi->mt_yleft_col[mb_row][i] = xd->dst.y_buffer [i* recon_y_stride + 15];
522 for (i = 0; i < 8; i++)
523 {
524 pbi->mt_uleft_col[mb_row][i] = xd->dst.u_buffer [i* recon_uv_stride + 7];
525 pbi->mt_vleft_col[mb_row][i] = xd->dst.v_buffer [i* recon_uv_stride + 7];
526 }
527 }
528 }
529
530 /* loopfilter on this macroblock. */
531 if (filter_level)
532 {
533 if(pc->filter_type == NORMAL_LOOPFILTER)
534 {
535 loop_filter_info lfi;
536 FRAME_TYPE frame_type = pc->frame_type;
537 const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level];
538 lfi.mblim = lfi_n->mblim[filter_level];
539 lfi.blim = lfi_n->blim[filter_level];
540 lfi.lim = lfi_n->lim[filter_level];
541 lfi.hev_thr = lfi_n->hev_thr[hev_index];
542
543 if (mb_col > 0)
544 vp8_loop_filter_mbv
545 (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
546
547 if (!skip_lf)
548 vp8_loop_filter_bv
549 (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
550
551 /* don't apply across umv border */
552 if (mb_row > 0)
553 vp8_loop_filter_mbh
554 (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
555
556 if (!skip_lf)
557 vp8_loop_filter_bh
558 (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
559 }
560 else
561 {
562 if (mb_col > 0)
563 vp8_loop_filter_simple_mbv
564 (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]);
565
566 if (!skip_lf)
567 vp8_loop_filter_simple_bv
568 (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]);
569
570 /* don't apply across umv border */
571 if (mb_row > 0)
572 vp8_loop_filter_simple_mbh
573 (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]);
574
575 if (!skip_lf)
576 vp8_loop_filter_simple_bh
577 (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]);
578 }
579 }
580
581 }
582
583 recon_yoffset += 16;
584 recon_uvoffset += 8;
585
586 ++xd->mode_info_context; /* next mb */
587
588 xd->above_context++;
589 }
590
591 /* adjust to the next row of mbs */
592 if (pbi->common.filter_level)
593 {
594 if(mb_row != pc->mb_rows-1)
595 {
596 int lasty = yv12_fb_lst->y_width + VP8BORDERINPIXELS;
597 int lastuv = (yv12_fb_lst->y_width>>1) + (VP8BORDERINPIXELS>>1);
598
599 for (i = 0; i < 4; i++)
600 {
601 pbi->mt_yabove_row[mb_row +1][lasty + i] = pbi->mt_yabove_row[mb_row +1][lasty -1];
602 pbi->mt_uabove_row[mb_row +1][lastuv + i] = pbi->mt_uabove_row[mb_row +1][lastuv -1];
603 pbi->mt_vabove_row[mb_row +1][lastuv + i] = pbi->mt_vabove_row[mb_row +1][lastuv -1];
604 }
605 }
606 }
607 else
608 vp8_extend_mb_row(yv12_fb_new, xd->dst.y_buffer + 16,
609 xd->dst.u_buffer + 8, xd->dst.v_buffer + 8);
610
611 /* last MB of row is ready just after extension is done */
612 *current_mb_col = mb_col + nsync;
613
614 ++xd->mode_info_context; /* skip prediction column */
615 xd->up_available = 1;
616
617 /* since we have multithread */
618 xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count;
619 }
620
621 /* signal end of frame decoding if this thread processed the last mb_row */
622 if (last_mb_row == (pc->mb_rows - 1))
623 sem_post(&pbi->h_event_end_decoding);
624
625 }
626
627
thread_decoding_proc(void * p_data)628 static THREAD_FUNCTION thread_decoding_proc(void *p_data)
629 {
630 int ithread = ((DECODETHREAD_DATA *)p_data)->ithread;
631 VP8D_COMP *pbi = (VP8D_COMP *)(((DECODETHREAD_DATA *)p_data)->ptr1);
632 MB_ROW_DEC *mbrd = (MB_ROW_DEC *)(((DECODETHREAD_DATA *)p_data)->ptr2);
633 ENTROPY_CONTEXT_PLANES mb_row_left_context;
634
635 while (1)
636 {
637 if (pbi->b_multithreaded_rd == 0)
638 break;
639
640 if (sem_wait(&pbi->h_event_start_decoding[ithread]) == 0)
641 {
642 if (pbi->b_multithreaded_rd == 0)
643 break;
644 else
645 {
646 MACROBLOCKD *xd = &mbrd->mbd;
647 xd->left_context = &mb_row_left_context;
648
649 mt_decode_mb_rows(pbi, xd, ithread+1);
650 }
651 }
652 }
653
654 return 0 ;
655 }
656
657
vp8_decoder_create_threads(VP8D_COMP * pbi)658 void vp8_decoder_create_threads(VP8D_COMP *pbi)
659 {
660 int core_count = 0;
661 unsigned int ithread;
662
663 pbi->b_multithreaded_rd = 0;
664 pbi->allocated_decoding_thread_count = 0;
665
666 /* limit decoding threads to the max number of token partitions */
667 core_count = (pbi->max_threads > 8) ? 8 : pbi->max_threads;
668
669 /* limit decoding threads to the available cores */
670 if (core_count > pbi->common.processor_core_count)
671 core_count = pbi->common.processor_core_count;
672
673 if (core_count > 1)
674 {
675 pbi->b_multithreaded_rd = 1;
676 pbi->decoding_thread_count = core_count - 1;
677
678 CALLOC_ARRAY(pbi->h_decoding_thread, pbi->decoding_thread_count);
679 CALLOC_ARRAY(pbi->h_event_start_decoding, pbi->decoding_thread_count);
680 CALLOC_ARRAY_ALIGNED(pbi->mb_row_di, pbi->decoding_thread_count, 32);
681 CALLOC_ARRAY(pbi->de_thread_data, pbi->decoding_thread_count);
682
683 for (ithread = 0; ithread < pbi->decoding_thread_count; ithread++)
684 {
685 sem_init(&pbi->h_event_start_decoding[ithread], 0, 0);
686
687 vp8_setup_block_dptrs(&pbi->mb_row_di[ithread].mbd);
688
689 pbi->de_thread_data[ithread].ithread = ithread;
690 pbi->de_thread_data[ithread].ptr1 = (void *)pbi;
691 pbi->de_thread_data[ithread].ptr2 = (void *) &pbi->mb_row_di[ithread];
692
693 pthread_create(&pbi->h_decoding_thread[ithread], 0, thread_decoding_proc, (&pbi->de_thread_data[ithread]));
694 }
695
696 sem_init(&pbi->h_event_end_decoding, 0, 0);
697
698 pbi->allocated_decoding_thread_count = pbi->decoding_thread_count;
699 }
700 }
701
702
vp8mt_de_alloc_temp_buffers(VP8D_COMP * pbi,int mb_rows)703 void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows)
704 {
705 int i;
706
707 if (pbi->b_multithreaded_rd)
708 {
709 vpx_free(pbi->mt_current_mb_col);
710 pbi->mt_current_mb_col = NULL ;
711
712 /* Free above_row buffers. */
713 if (pbi->mt_yabove_row)
714 {
715 for (i=0; i< mb_rows; i++)
716 {
717 vpx_free(pbi->mt_yabove_row[i]);
718 pbi->mt_yabove_row[i] = NULL ;
719 }
720 vpx_free(pbi->mt_yabove_row);
721 pbi->mt_yabove_row = NULL ;
722 }
723
724 if (pbi->mt_uabove_row)
725 {
726 for (i=0; i< mb_rows; i++)
727 {
728 vpx_free(pbi->mt_uabove_row[i]);
729 pbi->mt_uabove_row[i] = NULL ;
730 }
731 vpx_free(pbi->mt_uabove_row);
732 pbi->mt_uabove_row = NULL ;
733 }
734
735 if (pbi->mt_vabove_row)
736 {
737 for (i=0; i< mb_rows; i++)
738 {
739 vpx_free(pbi->mt_vabove_row[i]);
740 pbi->mt_vabove_row[i] = NULL ;
741 }
742 vpx_free(pbi->mt_vabove_row);
743 pbi->mt_vabove_row = NULL ;
744 }
745
746 /* Free left_col buffers. */
747 if (pbi->mt_yleft_col)
748 {
749 for (i=0; i< mb_rows; i++)
750 {
751 vpx_free(pbi->mt_yleft_col[i]);
752 pbi->mt_yleft_col[i] = NULL ;
753 }
754 vpx_free(pbi->mt_yleft_col);
755 pbi->mt_yleft_col = NULL ;
756 }
757
758 if (pbi->mt_uleft_col)
759 {
760 for (i=0; i< mb_rows; i++)
761 {
762 vpx_free(pbi->mt_uleft_col[i]);
763 pbi->mt_uleft_col[i] = NULL ;
764 }
765 vpx_free(pbi->mt_uleft_col);
766 pbi->mt_uleft_col = NULL ;
767 }
768
769 if (pbi->mt_vleft_col)
770 {
771 for (i=0; i< mb_rows; i++)
772 {
773 vpx_free(pbi->mt_vleft_col[i]);
774 pbi->mt_vleft_col[i] = NULL ;
775 }
776 vpx_free(pbi->mt_vleft_col);
777 pbi->mt_vleft_col = NULL ;
778 }
779 }
780 }
781
782
vp8mt_alloc_temp_buffers(VP8D_COMP * pbi,int width,int prev_mb_rows)783 void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows)
784 {
785 VP8_COMMON *const pc = & pbi->common;
786 int i;
787 int uv_width;
788
789 if (pbi->b_multithreaded_rd)
790 {
791 vp8mt_de_alloc_temp_buffers(pbi, prev_mb_rows);
792
793 /* our internal buffers are always multiples of 16 */
794 if ((width & 0xf) != 0)
795 width += 16 - (width & 0xf);
796
797 if (width < 640) pbi->sync_range = 1;
798 else if (width <= 1280) pbi->sync_range = 8;
799 else if (width <= 2560) pbi->sync_range =16;
800 else pbi->sync_range = 32;
801
802 uv_width = width >>1;
803
804 /* Allocate an int for each mb row. */
805 CALLOC_ARRAY(pbi->mt_current_mb_col, pc->mb_rows);
806
807 /* Allocate memory for above_row buffers. */
808 CALLOC_ARRAY(pbi->mt_yabove_row, pc->mb_rows);
809 for (i = 0; i < pc->mb_rows; i++)
810 CHECK_MEM_ERROR(pbi->mt_yabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (width + (VP8BORDERINPIXELS<<1))));
811
812 CALLOC_ARRAY(pbi->mt_uabove_row, pc->mb_rows);
813 for (i = 0; i < pc->mb_rows; i++)
814 CHECK_MEM_ERROR(pbi->mt_uabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS)));
815
816 CALLOC_ARRAY(pbi->mt_vabove_row, pc->mb_rows);
817 for (i = 0; i < pc->mb_rows; i++)
818 CHECK_MEM_ERROR(pbi->mt_vabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS)));
819
820 /* Allocate memory for left_col buffers. */
821 CALLOC_ARRAY(pbi->mt_yleft_col, pc->mb_rows);
822 for (i = 0; i < pc->mb_rows; i++)
823 CHECK_MEM_ERROR(pbi->mt_yleft_col[i], vpx_calloc(sizeof(unsigned char) * 16, 1));
824
825 CALLOC_ARRAY(pbi->mt_uleft_col, pc->mb_rows);
826 for (i = 0; i < pc->mb_rows; i++)
827 CHECK_MEM_ERROR(pbi->mt_uleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1));
828
829 CALLOC_ARRAY(pbi->mt_vleft_col, pc->mb_rows);
830 for (i = 0; i < pc->mb_rows; i++)
831 CHECK_MEM_ERROR(pbi->mt_vleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1));
832 }
833 }
834
835
vp8_decoder_remove_threads(VP8D_COMP * pbi)836 void vp8_decoder_remove_threads(VP8D_COMP *pbi)
837 {
838 /* shutdown MB Decoding thread; */
839 if (pbi->b_multithreaded_rd)
840 {
841 int i;
842
843 pbi->b_multithreaded_rd = 0;
844
845 /* allow all threads to exit */
846 for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
847 {
848 sem_post(&pbi->h_event_start_decoding[i]);
849 pthread_join(pbi->h_decoding_thread[i], NULL);
850 }
851
852 for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
853 {
854 sem_destroy(&pbi->h_event_start_decoding[i]);
855 }
856
857 sem_destroy(&pbi->h_event_end_decoding);
858
859 vpx_free(pbi->h_decoding_thread);
860 pbi->h_decoding_thread = NULL;
861
862 vpx_free(pbi->h_event_start_decoding);
863 pbi->h_event_start_decoding = NULL;
864
865 vpx_free(pbi->mb_row_di);
866 pbi->mb_row_di = NULL ;
867
868 vpx_free(pbi->de_thread_data);
869 pbi->de_thread_data = NULL;
870 }
871 }
872
vp8mt_decode_mb_rows(VP8D_COMP * pbi,MACROBLOCKD * xd)873 void vp8mt_decode_mb_rows( VP8D_COMP *pbi, MACROBLOCKD *xd)
874 {
875 VP8_COMMON *pc = &pbi->common;
876 unsigned int i;
877 int j;
878
879 int filter_level = pc->filter_level;
880 YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
881
882 if (filter_level)
883 {
884 /* Set above_row buffer to 127 for decoding first MB row */
885 vpx_memset(pbi->mt_yabove_row[0] + VP8BORDERINPIXELS-1, 127, yv12_fb_new->y_width + 5);
886 vpx_memset(pbi->mt_uabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5);
887 vpx_memset(pbi->mt_vabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5);
888
889 for (j=1; j<pc->mb_rows; j++)
890 {
891 vpx_memset(pbi->mt_yabove_row[j] + VP8BORDERINPIXELS-1, (unsigned char)129, 1);
892 vpx_memset(pbi->mt_uabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1);
893 vpx_memset(pbi->mt_vabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1);
894 }
895
896 /* Set left_col to 129 initially */
897 for (j=0; j<pc->mb_rows; j++)
898 {
899 vpx_memset(pbi->mt_yleft_col[j], (unsigned char)129, 16);
900 vpx_memset(pbi->mt_uleft_col[j], (unsigned char)129, 8);
901 vpx_memset(pbi->mt_vleft_col[j], (unsigned char)129, 8);
902 }
903
904 /* Initialize the loop filter for this frame. */
905 vp8_loop_filter_frame_init(pc, &pbi->mb, filter_level);
906 }
907 else
908 vp8_setup_intra_recon_top_line(yv12_fb_new);
909
910 setup_decoding_thread_data(pbi, xd, pbi->mb_row_di, pbi->decoding_thread_count);
911
912 for (i = 0; i < pbi->decoding_thread_count; i++)
913 sem_post(&pbi->h_event_start_decoding[i]);
914
915 mt_decode_mb_rows(pbi, xd, 0);
916
917 sem_wait(&pbi->h_event_end_decoding); /* add back for each frame */
918 }
919