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 "vp8_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 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 vpx_memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[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 vpx_memset(b->qcoeff, 0, 16 * sizeof(b->qcoeff[0]));
268 }
269 else
270 {
271 b->dqcoeff[0] = b->qcoeff[0] * xd->dequant_y2[0];
272 vp8_short_inv_walsh4x4_1(&b->dqcoeff[0],
273 xd->qcoeff);
274 vpx_memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0]));
275 }
276
277 /* override the dc dequant constant in order to preserve the
278 * dc components
279 */
280 DQC = xd->dequant_y1_dc;
281 }
282
283 vp8_dequant_idct_add_y_block
284 (xd->qcoeff, DQC,
285 xd->dst.y_buffer,
286 xd->dst.y_stride, xd->eobs);
287 }
288
289 vp8_dequant_idct_add_uv_block
290 (xd->qcoeff+16*16, xd->dequant_uv,
291 xd->dst.u_buffer, xd->dst.v_buffer,
292 xd->dst.uv_stride, xd->eobs+16);
293 }
294 }
295
mt_decode_mb_rows(VP8D_COMP * pbi,MACROBLOCKD * xd,int start_mb_row)296 static void mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd, int start_mb_row)
297 {
298 volatile const int *last_row_current_mb_col;
299 volatile int *current_mb_col;
300 int mb_row;
301 VP8_COMMON *pc = &pbi->common;
302 const int nsync = pbi->sync_range;
303 const int first_row_no_sync_above = pc->mb_cols + nsync;
304 int num_part = 1 << pbi->common.multi_token_partition;
305 int last_mb_row = start_mb_row;
306
307 YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
308 YV12_BUFFER_CONFIG *yv12_fb_lst = pbi->dec_fb_ref[LAST_FRAME];
309
310 int recon_y_stride = yv12_fb_new->y_stride;
311 int recon_uv_stride = yv12_fb_new->uv_stride;
312
313 unsigned char *ref_buffer[MAX_REF_FRAMES][3];
314 unsigned char *dst_buffer[3];
315 int i;
316 int ref_fb_corrupted[MAX_REF_FRAMES];
317
318 ref_fb_corrupted[INTRA_FRAME] = 0;
319
320 for(i = 1; i < MAX_REF_FRAMES; i++)
321 {
322 YV12_BUFFER_CONFIG *this_fb = pbi->dec_fb_ref[i];
323
324 ref_buffer[i][0] = this_fb->y_buffer;
325 ref_buffer[i][1] = this_fb->u_buffer;
326 ref_buffer[i][2] = this_fb->v_buffer;
327
328 ref_fb_corrupted[i] = this_fb->corrupted;
329 }
330
331 dst_buffer[0] = yv12_fb_new->y_buffer;
332 dst_buffer[1] = yv12_fb_new->u_buffer;
333 dst_buffer[2] = yv12_fb_new->v_buffer;
334
335 xd->up_available = (start_mb_row != 0);
336
337 for (mb_row = start_mb_row; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1))
338 {
339 int recon_yoffset, recon_uvoffset;
340 int mb_col;
341 int filter_level;
342 loop_filter_info_n *lfi_n = &pc->lf_info;
343
344 /* save last row processed by this thread */
345 last_mb_row = mb_row;
346 /* select bool coder for current partition */
347 xd->current_bc = &pbi->mbc[mb_row%num_part];
348
349 if (mb_row > 0)
350 last_row_current_mb_col = &pbi->mt_current_mb_col[mb_row -1];
351 else
352 last_row_current_mb_col = &first_row_no_sync_above;
353
354 current_mb_col = &pbi->mt_current_mb_col[mb_row];
355
356 recon_yoffset = mb_row * recon_y_stride * 16;
357 recon_uvoffset = mb_row * recon_uv_stride * 8;
358
359 /* reset contexts */
360 xd->above_context = pc->above_context;
361 vpx_memset(xd->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
362
363 xd->left_available = 0;
364
365 xd->mb_to_top_edge = -((mb_row * 16)) << 3;
366 xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3;
367
368 if (pbi->common.filter_level)
369 {
370 xd->recon_above[0] = pbi->mt_yabove_row[mb_row] + 0*16 +32;
371 xd->recon_above[1] = pbi->mt_uabove_row[mb_row] + 0*8 +16;
372 xd->recon_above[2] = pbi->mt_vabove_row[mb_row] + 0*8 +16;
373
374 xd->recon_left[0] = pbi->mt_yleft_col[mb_row];
375 xd->recon_left[1] = pbi->mt_uleft_col[mb_row];
376 xd->recon_left[2] = pbi->mt_vleft_col[mb_row];
377
378 /* TODO: move to outside row loop */
379 xd->recon_left_stride[0] = 1;
380 xd->recon_left_stride[1] = 1;
381 }
382 else
383 {
384 xd->recon_above[0] = dst_buffer[0] + recon_yoffset;
385 xd->recon_above[1] = dst_buffer[1] + recon_uvoffset;
386 xd->recon_above[2] = dst_buffer[2] + recon_uvoffset;
387
388 xd->recon_left[0] = xd->recon_above[0] - 1;
389 xd->recon_left[1] = xd->recon_above[1] - 1;
390 xd->recon_left[2] = xd->recon_above[2] - 1;
391
392 xd->recon_above[0] -= xd->dst.y_stride;
393 xd->recon_above[1] -= xd->dst.uv_stride;
394 xd->recon_above[2] -= xd->dst.uv_stride;
395
396 /* TODO: move to outside row loop */
397 xd->recon_left_stride[0] = xd->dst.y_stride;
398 xd->recon_left_stride[1] = xd->dst.uv_stride;
399
400 setup_intra_recon_left(xd->recon_left[0], xd->recon_left[1],
401 xd->recon_left[2], xd->dst.y_stride,
402 xd->dst.uv_stride);
403 }
404
405 for (mb_col = 0; mb_col < pc->mb_cols; mb_col++)
406 {
407 *current_mb_col = mb_col - 1;
408
409 if ((mb_col & (nsync - 1)) == 0)
410 {
411 while (mb_col > (*last_row_current_mb_col - nsync))
412 {
413 x86_pause_hint();
414 thread_sleep(0);
415 }
416 }
417
418 /* Distance of MB to the various image edges.
419 * These are specified to 8th pel as they are always
420 * compared to values that are in 1/8th pel units.
421 */
422 xd->mb_to_left_edge = -((mb_col * 16) << 3);
423 xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3;
424
425 #if CONFIG_ERROR_CONCEALMENT
426 {
427 int corrupt_residual =
428 (!pbi->independent_partitions &&
429 pbi->frame_corrupt_residual) ||
430 vp8dx_bool_error(xd->current_bc);
431 if (pbi->ec_active &&
432 (xd->mode_info_context->mbmi.ref_frame ==
433 INTRA_FRAME) &&
434 corrupt_residual)
435 {
436 /* We have an intra block with corrupt
437 * coefficients, better to conceal with an inter
438 * block.
439 * Interpolate MVs from neighboring MBs
440 *
441 * Note that for the first mb with corrupt
442 * residual in a frame, we might not discover
443 * that before decoding the residual. That
444 * happens after this check, and therefore no
445 * inter concealment will be done.
446 */
447 vp8_interpolate_motion(xd,
448 mb_row, mb_col,
449 pc->mb_rows, pc->mb_cols,
450 pc->mode_info_stride);
451 }
452 }
453 #endif
454
455
456 xd->dst.y_buffer = dst_buffer[0] + recon_yoffset;
457 xd->dst.u_buffer = dst_buffer[1] + recon_uvoffset;
458 xd->dst.v_buffer = dst_buffer[2] + recon_uvoffset;
459
460 xd->pre.y_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][0] + recon_yoffset;
461 xd->pre.u_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][1] + recon_uvoffset;
462 xd->pre.v_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][2] + recon_uvoffset;
463
464 /* propagate errors from reference frames */
465 xd->corrupted |= ref_fb_corrupted[xd->mode_info_context->mbmi.ref_frame];
466
467 mt_decode_macroblock(pbi, xd, 0);
468
469 xd->left_available = 1;
470
471 /* check if the boolean decoder has suffered an error */
472 xd->corrupted |= vp8dx_bool_error(xd->current_bc);
473
474 xd->recon_above[0] += 16;
475 xd->recon_above[1] += 8;
476 xd->recon_above[2] += 8;
477
478 if (!pbi->common.filter_level)
479 {
480 xd->recon_left[0] += 16;
481 xd->recon_left[1] += 8;
482 xd->recon_left[2] += 8;
483 }
484
485 if (pbi->common.filter_level)
486 {
487 int skip_lf = (xd->mode_info_context->mbmi.mode != B_PRED &&
488 xd->mode_info_context->mbmi.mode != SPLITMV &&
489 xd->mode_info_context->mbmi.mb_skip_coeff);
490
491 const int mode_index = lfi_n->mode_lf_lut[xd->mode_info_context->mbmi.mode];
492 const int seg = xd->mode_info_context->mbmi.segment_id;
493 const int ref_frame = xd->mode_info_context->mbmi.ref_frame;
494
495 filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
496
497 if( mb_row != pc->mb_rows-1 )
498 {
499 /* Save decoded MB last row data for next-row decoding */
500 vpx_memcpy((pbi->mt_yabove_row[mb_row + 1] + 32 + mb_col*16), (xd->dst.y_buffer + 15 * recon_y_stride), 16);
501 vpx_memcpy((pbi->mt_uabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.u_buffer + 7 * recon_uv_stride), 8);
502 vpx_memcpy((pbi->mt_vabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.v_buffer + 7 * recon_uv_stride), 8);
503 }
504
505 /* save left_col for next MB decoding */
506 if(mb_col != pc->mb_cols-1)
507 {
508 MODE_INFO *next = xd->mode_info_context +1;
509
510 if (next->mbmi.ref_frame == INTRA_FRAME)
511 {
512 for (i = 0; i < 16; i++)
513 pbi->mt_yleft_col[mb_row][i] = xd->dst.y_buffer [i* recon_y_stride + 15];
514 for (i = 0; i < 8; i++)
515 {
516 pbi->mt_uleft_col[mb_row][i] = xd->dst.u_buffer [i* recon_uv_stride + 7];
517 pbi->mt_vleft_col[mb_row][i] = xd->dst.v_buffer [i* recon_uv_stride + 7];
518 }
519 }
520 }
521
522 /* loopfilter on this macroblock. */
523 if (filter_level)
524 {
525 if(pc->filter_type == NORMAL_LOOPFILTER)
526 {
527 loop_filter_info lfi;
528 FRAME_TYPE frame_type = pc->frame_type;
529 const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level];
530 lfi.mblim = lfi_n->mblim[filter_level];
531 lfi.blim = lfi_n->blim[filter_level];
532 lfi.lim = lfi_n->lim[filter_level];
533 lfi.hev_thr = lfi_n->hev_thr[hev_index];
534
535 if (mb_col > 0)
536 vp8_loop_filter_mbv
537 (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
538
539 if (!skip_lf)
540 vp8_loop_filter_bv
541 (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
542
543 /* don't apply across umv border */
544 if (mb_row > 0)
545 vp8_loop_filter_mbh
546 (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
547
548 if (!skip_lf)
549 vp8_loop_filter_bh
550 (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
551 }
552 else
553 {
554 if (mb_col > 0)
555 vp8_loop_filter_simple_mbv
556 (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]);
557
558 if (!skip_lf)
559 vp8_loop_filter_simple_bv
560 (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]);
561
562 /* don't apply across umv border */
563 if (mb_row > 0)
564 vp8_loop_filter_simple_mbh
565 (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]);
566
567 if (!skip_lf)
568 vp8_loop_filter_simple_bh
569 (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]);
570 }
571 }
572
573 }
574
575 recon_yoffset += 16;
576 recon_uvoffset += 8;
577
578 ++xd->mode_info_context; /* next mb */
579
580 xd->above_context++;
581 }
582
583 /* adjust to the next row of mbs */
584 if (pbi->common.filter_level)
585 {
586 if(mb_row != pc->mb_rows-1)
587 {
588 int lasty = yv12_fb_lst->y_width + VP8BORDERINPIXELS;
589 int lastuv = (yv12_fb_lst->y_width>>1) + (VP8BORDERINPIXELS>>1);
590
591 for (i = 0; i < 4; i++)
592 {
593 pbi->mt_yabove_row[mb_row +1][lasty + i] = pbi->mt_yabove_row[mb_row +1][lasty -1];
594 pbi->mt_uabove_row[mb_row +1][lastuv + i] = pbi->mt_uabove_row[mb_row +1][lastuv -1];
595 pbi->mt_vabove_row[mb_row +1][lastuv + i] = pbi->mt_vabove_row[mb_row +1][lastuv -1];
596 }
597 }
598 }
599 else
600 vp8_extend_mb_row(yv12_fb_new, xd->dst.y_buffer + 16,
601 xd->dst.u_buffer + 8, xd->dst.v_buffer + 8);
602
603 /* last MB of row is ready just after extension is done */
604 *current_mb_col = mb_col + nsync;
605
606 ++xd->mode_info_context; /* skip prediction column */
607 xd->up_available = 1;
608
609 /* since we have multithread */
610 xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count;
611 }
612
613 /* signal end of frame decoding if this thread processed the last mb_row */
614 if (last_mb_row == (pc->mb_rows - 1))
615 sem_post(&pbi->h_event_end_decoding);
616
617 }
618
619
thread_decoding_proc(void * p_data)620 static THREAD_FUNCTION thread_decoding_proc(void *p_data)
621 {
622 int ithread = ((DECODETHREAD_DATA *)p_data)->ithread;
623 VP8D_COMP *pbi = (VP8D_COMP *)(((DECODETHREAD_DATA *)p_data)->ptr1);
624 MB_ROW_DEC *mbrd = (MB_ROW_DEC *)(((DECODETHREAD_DATA *)p_data)->ptr2);
625 ENTROPY_CONTEXT_PLANES mb_row_left_context;
626
627 while (1)
628 {
629 if (pbi->b_multithreaded_rd == 0)
630 break;
631
632 if (sem_wait(&pbi->h_event_start_decoding[ithread]) == 0)
633 {
634 if (pbi->b_multithreaded_rd == 0)
635 break;
636 else
637 {
638 MACROBLOCKD *xd = &mbrd->mbd;
639 xd->left_context = &mb_row_left_context;
640
641 mt_decode_mb_rows(pbi, xd, ithread+1);
642 }
643 }
644 }
645
646 return 0 ;
647 }
648
649
vp8_decoder_create_threads(VP8D_COMP * pbi)650 void vp8_decoder_create_threads(VP8D_COMP *pbi)
651 {
652 int core_count = 0;
653 unsigned int ithread;
654
655 pbi->b_multithreaded_rd = 0;
656 pbi->allocated_decoding_thread_count = 0;
657
658 /* limit decoding threads to the max number of token partitions */
659 core_count = (pbi->max_threads > 8) ? 8 : pbi->max_threads;
660
661 /* limit decoding threads to the available cores */
662 if (core_count > pbi->common.processor_core_count)
663 core_count = pbi->common.processor_core_count;
664
665 if (core_count > 1)
666 {
667 pbi->b_multithreaded_rd = 1;
668 pbi->decoding_thread_count = core_count - 1;
669
670 CALLOC_ARRAY(pbi->h_decoding_thread, pbi->decoding_thread_count);
671 CALLOC_ARRAY(pbi->h_event_start_decoding, pbi->decoding_thread_count);
672 CALLOC_ARRAY_ALIGNED(pbi->mb_row_di, pbi->decoding_thread_count, 32);
673 CALLOC_ARRAY(pbi->de_thread_data, pbi->decoding_thread_count);
674
675 for (ithread = 0; ithread < pbi->decoding_thread_count; ithread++)
676 {
677 sem_init(&pbi->h_event_start_decoding[ithread], 0, 0);
678
679 vp8_setup_block_dptrs(&pbi->mb_row_di[ithread].mbd);
680
681 pbi->de_thread_data[ithread].ithread = ithread;
682 pbi->de_thread_data[ithread].ptr1 = (void *)pbi;
683 pbi->de_thread_data[ithread].ptr2 = (void *) &pbi->mb_row_di[ithread];
684
685 pthread_create(&pbi->h_decoding_thread[ithread], 0, thread_decoding_proc, (&pbi->de_thread_data[ithread]));
686 }
687
688 sem_init(&pbi->h_event_end_decoding, 0, 0);
689
690 pbi->allocated_decoding_thread_count = pbi->decoding_thread_count;
691 }
692 }
693
694
vp8mt_de_alloc_temp_buffers(VP8D_COMP * pbi,int mb_rows)695 void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows)
696 {
697 int i;
698
699 if (pbi->b_multithreaded_rd)
700 {
701 vpx_free(pbi->mt_current_mb_col);
702 pbi->mt_current_mb_col = NULL ;
703
704 /* Free above_row buffers. */
705 if (pbi->mt_yabove_row)
706 {
707 for (i=0; i< mb_rows; i++)
708 {
709 vpx_free(pbi->mt_yabove_row[i]);
710 pbi->mt_yabove_row[i] = NULL ;
711 }
712 vpx_free(pbi->mt_yabove_row);
713 pbi->mt_yabove_row = NULL ;
714 }
715
716 if (pbi->mt_uabove_row)
717 {
718 for (i=0; i< mb_rows; i++)
719 {
720 vpx_free(pbi->mt_uabove_row[i]);
721 pbi->mt_uabove_row[i] = NULL ;
722 }
723 vpx_free(pbi->mt_uabove_row);
724 pbi->mt_uabove_row = NULL ;
725 }
726
727 if (pbi->mt_vabove_row)
728 {
729 for (i=0; i< mb_rows; i++)
730 {
731 vpx_free(pbi->mt_vabove_row[i]);
732 pbi->mt_vabove_row[i] = NULL ;
733 }
734 vpx_free(pbi->mt_vabove_row);
735 pbi->mt_vabove_row = NULL ;
736 }
737
738 /* Free left_col buffers. */
739 if (pbi->mt_yleft_col)
740 {
741 for (i=0; i< mb_rows; i++)
742 {
743 vpx_free(pbi->mt_yleft_col[i]);
744 pbi->mt_yleft_col[i] = NULL ;
745 }
746 vpx_free(pbi->mt_yleft_col);
747 pbi->mt_yleft_col = NULL ;
748 }
749
750 if (pbi->mt_uleft_col)
751 {
752 for (i=0; i< mb_rows; i++)
753 {
754 vpx_free(pbi->mt_uleft_col[i]);
755 pbi->mt_uleft_col[i] = NULL ;
756 }
757 vpx_free(pbi->mt_uleft_col);
758 pbi->mt_uleft_col = NULL ;
759 }
760
761 if (pbi->mt_vleft_col)
762 {
763 for (i=0; i< mb_rows; i++)
764 {
765 vpx_free(pbi->mt_vleft_col[i]);
766 pbi->mt_vleft_col[i] = NULL ;
767 }
768 vpx_free(pbi->mt_vleft_col);
769 pbi->mt_vleft_col = NULL ;
770 }
771 }
772 }
773
774
vp8mt_alloc_temp_buffers(VP8D_COMP * pbi,int width,int prev_mb_rows)775 void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows)
776 {
777 VP8_COMMON *const pc = & pbi->common;
778 int i;
779 int uv_width;
780
781 if (pbi->b_multithreaded_rd)
782 {
783 vp8mt_de_alloc_temp_buffers(pbi, prev_mb_rows);
784
785 /* our internal buffers are always multiples of 16 */
786 if ((width & 0xf) != 0)
787 width += 16 - (width & 0xf);
788
789 if (width < 640) pbi->sync_range = 1;
790 else if (width <= 1280) pbi->sync_range = 8;
791 else if (width <= 2560) pbi->sync_range =16;
792 else pbi->sync_range = 32;
793
794 uv_width = width >>1;
795
796 /* Allocate an int for each mb row. */
797 CALLOC_ARRAY(pbi->mt_current_mb_col, pc->mb_rows);
798
799 /* Allocate memory for above_row buffers. */
800 CALLOC_ARRAY(pbi->mt_yabove_row, pc->mb_rows);
801 for (i = 0; i < pc->mb_rows; i++)
802 CHECK_MEM_ERROR(pbi->mt_yabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (width + (VP8BORDERINPIXELS<<1))));
803
804 CALLOC_ARRAY(pbi->mt_uabove_row, pc->mb_rows);
805 for (i = 0; i < pc->mb_rows; i++)
806 CHECK_MEM_ERROR(pbi->mt_uabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS)));
807
808 CALLOC_ARRAY(pbi->mt_vabove_row, pc->mb_rows);
809 for (i = 0; i < pc->mb_rows; i++)
810 CHECK_MEM_ERROR(pbi->mt_vabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS)));
811
812 /* Allocate memory for left_col buffers. */
813 CALLOC_ARRAY(pbi->mt_yleft_col, pc->mb_rows);
814 for (i = 0; i < pc->mb_rows; i++)
815 CHECK_MEM_ERROR(pbi->mt_yleft_col[i], vpx_calloc(sizeof(unsigned char) * 16, 1));
816
817 CALLOC_ARRAY(pbi->mt_uleft_col, pc->mb_rows);
818 for (i = 0; i < pc->mb_rows; i++)
819 CHECK_MEM_ERROR(pbi->mt_uleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1));
820
821 CALLOC_ARRAY(pbi->mt_vleft_col, pc->mb_rows);
822 for (i = 0; i < pc->mb_rows; i++)
823 CHECK_MEM_ERROR(pbi->mt_vleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1));
824 }
825 }
826
827
vp8_decoder_remove_threads(VP8D_COMP * pbi)828 void vp8_decoder_remove_threads(VP8D_COMP *pbi)
829 {
830 /* shutdown MB Decoding thread; */
831 if (pbi->b_multithreaded_rd)
832 {
833 int i;
834
835 pbi->b_multithreaded_rd = 0;
836
837 /* allow all threads to exit */
838 for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
839 {
840 sem_post(&pbi->h_event_start_decoding[i]);
841 pthread_join(pbi->h_decoding_thread[i], NULL);
842 }
843
844 for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
845 {
846 sem_destroy(&pbi->h_event_start_decoding[i]);
847 }
848
849 sem_destroy(&pbi->h_event_end_decoding);
850
851 vpx_free(pbi->h_decoding_thread);
852 pbi->h_decoding_thread = NULL;
853
854 vpx_free(pbi->h_event_start_decoding);
855 pbi->h_event_start_decoding = NULL;
856
857 vpx_free(pbi->mb_row_di);
858 pbi->mb_row_di = NULL ;
859
860 vpx_free(pbi->de_thread_data);
861 pbi->de_thread_data = NULL;
862 }
863 }
864
vp8mt_decode_mb_rows(VP8D_COMP * pbi,MACROBLOCKD * xd)865 void vp8mt_decode_mb_rows( VP8D_COMP *pbi, MACROBLOCKD *xd)
866 {
867 VP8_COMMON *pc = &pbi->common;
868 unsigned int i;
869 int j;
870
871 int filter_level = pc->filter_level;
872 YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
873
874 if (filter_level)
875 {
876 /* Set above_row buffer to 127 for decoding first MB row */
877 vpx_memset(pbi->mt_yabove_row[0] + VP8BORDERINPIXELS-1, 127, yv12_fb_new->y_width + 5);
878 vpx_memset(pbi->mt_uabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5);
879 vpx_memset(pbi->mt_vabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5);
880
881 for (j=1; j<pc->mb_rows; j++)
882 {
883 vpx_memset(pbi->mt_yabove_row[j] + VP8BORDERINPIXELS-1, (unsigned char)129, 1);
884 vpx_memset(pbi->mt_uabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1);
885 vpx_memset(pbi->mt_vabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1);
886 }
887
888 /* Set left_col to 129 initially */
889 for (j=0; j<pc->mb_rows; j++)
890 {
891 vpx_memset(pbi->mt_yleft_col[j], (unsigned char)129, 16);
892 vpx_memset(pbi->mt_uleft_col[j], (unsigned char)129, 8);
893 vpx_memset(pbi->mt_vleft_col[j], (unsigned char)129, 8);
894 }
895
896 /* Initialize the loop filter for this frame. */
897 vp8_loop_filter_frame_init(pc, &pbi->mb, filter_level);
898 }
899 else
900 vp8_setup_intra_recon_top_line(yv12_fb_new);
901
902 setup_decoding_thread_data(pbi, xd, pbi->mb_row_di, pbi->decoding_thread_count);
903
904 for (i = 0; i < pbi->decoding_thread_count; i++)
905 sem_post(&pbi->h_event_start_decoding[i]);
906
907 mt_decode_mb_rows(pbi, xd, 0);
908
909 sem_wait(&pbi->h_event_end_decoding); /* add back for each frame */
910 }
911