Line data Source code
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 "vpx_config.h"
12 : #include "vp8_rtcd.h"
13 : #if !defined(WIN32) && CONFIG_OS_SUPPORT == 1
14 : #include <unistd.h>
15 : #endif
16 : #include "onyxd_int.h"
17 : #include "vpx_mem/vpx_mem.h"
18 : #include "vp8/common/threading.h"
19 :
20 : #include "vp8/common/loopfilter.h"
21 : #include "vp8/common/extend.h"
22 : #include "vpx_ports/vpx_timer.h"
23 : #include "detokenize.h"
24 : #include "vp8/common/reconintra4x4.h"
25 : #include "vp8/common/reconinter.h"
26 : #include "vp8/common/reconintra.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) \
34 : do { \
35 : CHECK_MEM_ERROR((p), vpx_memalign((algn), sizeof(*(p)) * (n))); \
36 : memset((p), 0, (n) * sizeof(*(p))); \
37 : } while (0)
38 :
39 : void vp8_mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd);
40 :
41 0 : static void setup_decoding_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd,
42 : MB_ROW_DEC *mbrd, int count) {
43 0 : VP8_COMMON *const pc = &pbi->common;
44 : int i;
45 :
46 0 : for (i = 0; i < count; ++i) {
47 0 : MACROBLOCKD *mbd = &mbrd[i].mbd;
48 0 : mbd->subpixel_predict = xd->subpixel_predict;
49 0 : mbd->subpixel_predict8x4 = xd->subpixel_predict8x4;
50 0 : mbd->subpixel_predict8x8 = xd->subpixel_predict8x8;
51 0 : mbd->subpixel_predict16x16 = xd->subpixel_predict16x16;
52 :
53 0 : mbd->frame_type = pc->frame_type;
54 0 : mbd->pre = xd->pre;
55 0 : mbd->dst = xd->dst;
56 :
57 0 : mbd->segmentation_enabled = xd->segmentation_enabled;
58 0 : mbd->mb_segement_abs_delta = xd->mb_segement_abs_delta;
59 0 : memcpy(mbd->segment_feature_data, xd->segment_feature_data,
60 : sizeof(xd->segment_feature_data));
61 :
62 : /*signed char ref_lf_deltas[MAX_REF_LF_DELTAS];*/
63 0 : memcpy(mbd->ref_lf_deltas, xd->ref_lf_deltas, sizeof(xd->ref_lf_deltas));
64 : /*signed char mode_lf_deltas[MAX_MODE_LF_DELTAS];*/
65 0 : memcpy(mbd->mode_lf_deltas, xd->mode_lf_deltas, sizeof(xd->mode_lf_deltas));
66 : /*unsigned char mode_ref_lf_delta_enabled;
67 : unsigned char mode_ref_lf_delta_update;*/
68 0 : mbd->mode_ref_lf_delta_enabled = xd->mode_ref_lf_delta_enabled;
69 0 : mbd->mode_ref_lf_delta_update = xd->mode_ref_lf_delta_update;
70 :
71 0 : mbd->current_bc = &pbi->mbc[0];
72 :
73 0 : memcpy(mbd->dequant_y1_dc, xd->dequant_y1_dc, sizeof(xd->dequant_y1_dc));
74 0 : memcpy(mbd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1));
75 0 : memcpy(mbd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2));
76 0 : memcpy(mbd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv));
77 :
78 0 : mbd->fullpixel_mask = 0xffffffff;
79 :
80 0 : if (pc->full_pixel) mbd->fullpixel_mask = 0xfffffff8;
81 : }
82 :
83 0 : for (i = 0; i < pc->mb_rows; ++i) pbi->mt_current_mb_col[i] = -1;
84 0 : }
85 :
86 0 : static void mt_decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd,
87 : unsigned int mb_idx) {
88 : MB_PREDICTION_MODE mode;
89 : int i;
90 : #if CONFIG_ERROR_CONCEALMENT
91 : int corruption_detected = 0;
92 : #else
93 : (void)mb_idx;
94 : #endif
95 :
96 0 : if (xd->mode_info_context->mbmi.mb_skip_coeff) {
97 0 : vp8_reset_mb_tokens_context(xd);
98 0 : } else if (!vp8dx_bool_error(xd->current_bc)) {
99 : int eobtotal;
100 0 : eobtotal = vp8_decode_mb_tokens(pbi, xd);
101 :
102 : /* Special case: Force the loopfilter to skip when eobtotal is zero */
103 0 : xd->mode_info_context->mbmi.mb_skip_coeff = (eobtotal == 0);
104 : }
105 :
106 0 : mode = xd->mode_info_context->mbmi.mode;
107 :
108 0 : if (xd->segmentation_enabled) vp8_mb_init_dequantizer(pbi, xd);
109 :
110 : #if CONFIG_ERROR_CONCEALMENT
111 :
112 : if (pbi->ec_active) {
113 : int throw_residual;
114 : /* When we have independent partitions we can apply residual even
115 : * though other partitions within the frame are corrupt.
116 : */
117 : throw_residual =
118 : (!pbi->independent_partitions && pbi->frame_corrupt_residual);
119 : throw_residual = (throw_residual || vp8dx_bool_error(xd->current_bc));
120 :
121 : if ((mb_idx >= pbi->mvs_corrupt_from_mb || throw_residual)) {
122 : /* MB with corrupt residuals or corrupt mode/motion vectors.
123 : * Better to use the predictor as reconstruction.
124 : */
125 : pbi->frame_corrupt_residual = 1;
126 : memset(xd->qcoeff, 0, sizeof(xd->qcoeff));
127 :
128 : corruption_detected = 1;
129 :
130 : /* force idct to be skipped for B_PRED and use the
131 : * prediction only for reconstruction
132 : * */
133 : memset(xd->eobs, 0, 25);
134 : }
135 : }
136 : #endif
137 :
138 : /* do prediction */
139 0 : if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
140 0 : vp8_build_intra_predictors_mbuv_s(
141 : xd, xd->recon_above[1], xd->recon_above[2], xd->recon_left[1],
142 0 : xd->recon_left[2], xd->recon_left_stride[1], xd->dst.u_buffer,
143 0 : xd->dst.v_buffer, xd->dst.uv_stride);
144 :
145 0 : if (mode != B_PRED) {
146 0 : vp8_build_intra_predictors_mby_s(
147 : xd, xd->recon_above[0], xd->recon_left[0], xd->recon_left_stride[0],
148 0 : xd->dst.y_buffer, xd->dst.y_stride);
149 : } else {
150 0 : short *DQC = xd->dequant_y1;
151 0 : int dst_stride = xd->dst.y_stride;
152 :
153 : /* clear out residual eob info */
154 0 : if (xd->mode_info_context->mbmi.mb_skip_coeff) memset(xd->eobs, 0, 25);
155 :
156 0 : intra_prediction_down_copy(xd, xd->recon_above[0] + 16);
157 :
158 0 : for (i = 0; i < 16; ++i) {
159 0 : BLOCKD *b = &xd->block[i];
160 0 : unsigned char *dst = xd->dst.y_buffer + b->offset;
161 0 : B_PREDICTION_MODE b_mode = xd->mode_info_context->bmi[i].as_mode;
162 : unsigned char *Above;
163 : unsigned char *yleft;
164 : int left_stride;
165 : unsigned char top_left;
166 :
167 : /*Caution: For some b_mode, it needs 8 pixels (4 above + 4
168 : * above-right).*/
169 0 : if (i < 4 && pbi->common.filter_level) {
170 0 : Above = xd->recon_above[0] + b->offset;
171 : } else {
172 0 : Above = dst - dst_stride;
173 : }
174 :
175 0 : if (i % 4 == 0 && pbi->common.filter_level) {
176 0 : yleft = xd->recon_left[0] + i;
177 0 : left_stride = 1;
178 : } else {
179 0 : yleft = dst - 1;
180 0 : left_stride = dst_stride;
181 : }
182 :
183 0 : if ((i == 4 || i == 8 || i == 12) && pbi->common.filter_level) {
184 0 : top_left = *(xd->recon_left[0] + i - 1);
185 : } else {
186 0 : top_left = Above[-1];
187 : }
188 :
189 0 : vp8_intra4x4_predict(Above, yleft, left_stride, b_mode, dst, dst_stride,
190 : top_left);
191 :
192 0 : if (xd->eobs[i]) {
193 0 : if (xd->eobs[i] > 1) {
194 0 : vp8_dequant_idct_add(b->qcoeff, DQC, dst, dst_stride);
195 : } else {
196 0 : vp8_dc_only_idct_add(b->qcoeff[0] * DQC[0], dst, dst_stride, dst,
197 : dst_stride);
198 0 : memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0]));
199 : }
200 : }
201 : }
202 : }
203 : } else {
204 0 : vp8_build_inter_predictors_mb(xd);
205 : }
206 :
207 : #if CONFIG_ERROR_CONCEALMENT
208 : if (corruption_detected) {
209 : return;
210 : }
211 : #endif
212 :
213 0 : if (!xd->mode_info_context->mbmi.mb_skip_coeff) {
214 : /* dequantization and idct */
215 0 : if (mode != B_PRED) {
216 0 : short *DQC = xd->dequant_y1;
217 :
218 0 : if (mode != SPLITMV) {
219 0 : BLOCKD *b = &xd->block[24];
220 :
221 : /* do 2nd order transform on the dc block */
222 0 : if (xd->eobs[24] > 1) {
223 0 : vp8_dequantize_b(b, xd->dequant_y2);
224 :
225 0 : vp8_short_inv_walsh4x4(&b->dqcoeff[0], xd->qcoeff);
226 0 : memset(b->qcoeff, 0, 16 * sizeof(b->qcoeff[0]));
227 : } else {
228 0 : b->dqcoeff[0] = b->qcoeff[0] * xd->dequant_y2[0];
229 0 : vp8_short_inv_walsh4x4_1(&b->dqcoeff[0], xd->qcoeff);
230 0 : memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0]));
231 : }
232 :
233 : /* override the dc dequant constant in order to preserve the
234 : * dc components
235 : */
236 0 : DQC = xd->dequant_y1_dc;
237 : }
238 :
239 0 : vp8_dequant_idct_add_y_block(xd->qcoeff, DQC, xd->dst.y_buffer,
240 0 : xd->dst.y_stride, xd->eobs);
241 : }
242 :
243 0 : vp8_dequant_idct_add_uv_block(xd->qcoeff + 16 * 16, xd->dequant_uv,
244 0 : xd->dst.u_buffer, xd->dst.v_buffer,
245 0 : xd->dst.uv_stride, xd->eobs + 16);
246 : }
247 0 : }
248 :
249 0 : static void mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd,
250 : int start_mb_row) {
251 : const int *last_row_current_mb_col;
252 : int *current_mb_col;
253 : int mb_row;
254 0 : VP8_COMMON *pc = &pbi->common;
255 0 : const int nsync = pbi->sync_range;
256 0 : const int first_row_no_sync_above = pc->mb_cols + nsync;
257 0 : int num_part = 1 << pbi->common.multi_token_partition;
258 0 : int last_mb_row = start_mb_row;
259 :
260 0 : YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
261 0 : YV12_BUFFER_CONFIG *yv12_fb_lst = pbi->dec_fb_ref[LAST_FRAME];
262 :
263 0 : int recon_y_stride = yv12_fb_new->y_stride;
264 0 : int recon_uv_stride = yv12_fb_new->uv_stride;
265 :
266 : unsigned char *ref_buffer[MAX_REF_FRAMES][3];
267 : unsigned char *dst_buffer[3];
268 : int i;
269 : int ref_fb_corrupted[MAX_REF_FRAMES];
270 :
271 0 : ref_fb_corrupted[INTRA_FRAME] = 0;
272 :
273 0 : for (i = 1; i < MAX_REF_FRAMES; ++i) {
274 0 : YV12_BUFFER_CONFIG *this_fb = pbi->dec_fb_ref[i];
275 :
276 0 : ref_buffer[i][0] = this_fb->y_buffer;
277 0 : ref_buffer[i][1] = this_fb->u_buffer;
278 0 : ref_buffer[i][2] = this_fb->v_buffer;
279 :
280 0 : ref_fb_corrupted[i] = this_fb->corrupted;
281 : }
282 :
283 0 : dst_buffer[0] = yv12_fb_new->y_buffer;
284 0 : dst_buffer[1] = yv12_fb_new->u_buffer;
285 0 : dst_buffer[2] = yv12_fb_new->v_buffer;
286 :
287 0 : xd->up_available = (start_mb_row != 0);
288 :
289 0 : xd->mode_info_context = pc->mi + pc->mode_info_stride * start_mb_row;
290 0 : xd->mode_info_stride = pc->mode_info_stride;
291 :
292 0 : for (mb_row = start_mb_row; mb_row < pc->mb_rows;
293 0 : mb_row += (pbi->decoding_thread_count + 1)) {
294 : int recon_yoffset, recon_uvoffset;
295 : int mb_col;
296 : int filter_level;
297 0 : loop_filter_info_n *lfi_n = &pc->lf_info;
298 :
299 : /* save last row processed by this thread */
300 0 : last_mb_row = mb_row;
301 : /* select bool coder for current partition */
302 0 : xd->current_bc = &pbi->mbc[mb_row % num_part];
303 :
304 0 : if (mb_row > 0) {
305 0 : last_row_current_mb_col = &pbi->mt_current_mb_col[mb_row - 1];
306 : } else {
307 0 : last_row_current_mb_col = &first_row_no_sync_above;
308 : }
309 :
310 0 : current_mb_col = &pbi->mt_current_mb_col[mb_row];
311 :
312 0 : recon_yoffset = mb_row * recon_y_stride * 16;
313 0 : recon_uvoffset = mb_row * recon_uv_stride * 8;
314 :
315 : /* reset contexts */
316 0 : xd->above_context = pc->above_context;
317 0 : memset(xd->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
318 :
319 0 : xd->left_available = 0;
320 :
321 0 : xd->mb_to_top_edge = -((mb_row * 16) << 3);
322 0 : xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3;
323 :
324 0 : if (pbi->common.filter_level) {
325 0 : xd->recon_above[0] = pbi->mt_yabove_row[mb_row] + 0 * 16 + 32;
326 0 : xd->recon_above[1] = pbi->mt_uabove_row[mb_row] + 0 * 8 + 16;
327 0 : xd->recon_above[2] = pbi->mt_vabove_row[mb_row] + 0 * 8 + 16;
328 :
329 0 : xd->recon_left[0] = pbi->mt_yleft_col[mb_row];
330 0 : xd->recon_left[1] = pbi->mt_uleft_col[mb_row];
331 0 : xd->recon_left[2] = pbi->mt_vleft_col[mb_row];
332 :
333 : /* TODO: move to outside row loop */
334 0 : xd->recon_left_stride[0] = 1;
335 0 : xd->recon_left_stride[1] = 1;
336 : } else {
337 0 : xd->recon_above[0] = dst_buffer[0] + recon_yoffset;
338 0 : xd->recon_above[1] = dst_buffer[1] + recon_uvoffset;
339 0 : xd->recon_above[2] = dst_buffer[2] + recon_uvoffset;
340 :
341 0 : xd->recon_left[0] = xd->recon_above[0] - 1;
342 0 : xd->recon_left[1] = xd->recon_above[1] - 1;
343 0 : xd->recon_left[2] = xd->recon_above[2] - 1;
344 :
345 0 : xd->recon_above[0] -= xd->dst.y_stride;
346 0 : xd->recon_above[1] -= xd->dst.uv_stride;
347 0 : xd->recon_above[2] -= xd->dst.uv_stride;
348 :
349 : /* TODO: move to outside row loop */
350 0 : xd->recon_left_stride[0] = xd->dst.y_stride;
351 0 : xd->recon_left_stride[1] = xd->dst.uv_stride;
352 :
353 0 : setup_intra_recon_left(xd->recon_left[0], xd->recon_left[1],
354 : xd->recon_left[2], xd->dst.y_stride,
355 : xd->dst.uv_stride);
356 : }
357 :
358 0 : for (mb_col = 0; mb_col < pc->mb_cols; ++mb_col) {
359 0 : if (((mb_col - 1) % nsync) == 0) {
360 0 : pthread_mutex_t *mutex = &pbi->pmutex[mb_row];
361 0 : protected_write(mutex, current_mb_col, mb_col - 1);
362 : }
363 :
364 0 : if (mb_row && !(mb_col & (nsync - 1))) {
365 0 : pthread_mutex_t *mutex = &pbi->pmutex[mb_row - 1];
366 0 : sync_read(mutex, mb_col, last_row_current_mb_col, nsync);
367 : }
368 :
369 : /* Distance of MB to the various image edges.
370 : * These are specified to 8th pel as they are always
371 : * compared to values that are in 1/8th pel units.
372 : */
373 0 : xd->mb_to_left_edge = -((mb_col * 16) << 3);
374 0 : xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3;
375 :
376 : #if CONFIG_ERROR_CONCEALMENT
377 : {
378 : int corrupt_residual =
379 : (!pbi->independent_partitions && pbi->frame_corrupt_residual) ||
380 : vp8dx_bool_error(xd->current_bc);
381 : if (pbi->ec_active &&
382 : (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) &&
383 : corrupt_residual) {
384 : /* We have an intra block with corrupt
385 : * coefficients, better to conceal with an inter
386 : * block.
387 : * Interpolate MVs from neighboring MBs
388 : *
389 : * Note that for the first mb with corrupt
390 : * residual in a frame, we might not discover
391 : * that before decoding the residual. That
392 : * happens after this check, and therefore no
393 : * inter concealment will be done.
394 : */
395 : vp8_interpolate_motion(xd, mb_row, mb_col, pc->mb_rows, pc->mb_cols);
396 : }
397 : }
398 : #endif
399 :
400 0 : xd->dst.y_buffer = dst_buffer[0] + recon_yoffset;
401 0 : xd->dst.u_buffer = dst_buffer[1] + recon_uvoffset;
402 0 : xd->dst.v_buffer = dst_buffer[2] + recon_uvoffset;
403 :
404 0 : xd->pre.y_buffer =
405 0 : ref_buffer[xd->mode_info_context->mbmi.ref_frame][0] + recon_yoffset;
406 0 : xd->pre.u_buffer =
407 0 : ref_buffer[xd->mode_info_context->mbmi.ref_frame][1] + recon_uvoffset;
408 0 : xd->pre.v_buffer =
409 0 : ref_buffer[xd->mode_info_context->mbmi.ref_frame][2] + recon_uvoffset;
410 :
411 : /* propagate errors from reference frames */
412 0 : xd->corrupted |= ref_fb_corrupted[xd->mode_info_context->mbmi.ref_frame];
413 :
414 0 : mt_decode_macroblock(pbi, xd, 0);
415 :
416 0 : xd->left_available = 1;
417 :
418 : /* check if the boolean decoder has suffered an error */
419 0 : xd->corrupted |= vp8dx_bool_error(xd->current_bc);
420 :
421 0 : xd->recon_above[0] += 16;
422 0 : xd->recon_above[1] += 8;
423 0 : xd->recon_above[2] += 8;
424 :
425 0 : if (!pbi->common.filter_level) {
426 0 : xd->recon_left[0] += 16;
427 0 : xd->recon_left[1] += 8;
428 0 : xd->recon_left[2] += 8;
429 : }
430 :
431 0 : if (pbi->common.filter_level) {
432 0 : int skip_lf = (xd->mode_info_context->mbmi.mode != B_PRED &&
433 0 : xd->mode_info_context->mbmi.mode != SPLITMV &&
434 0 : xd->mode_info_context->mbmi.mb_skip_coeff);
435 :
436 0 : const int mode_index =
437 0 : lfi_n->mode_lf_lut[xd->mode_info_context->mbmi.mode];
438 0 : const int seg = xd->mode_info_context->mbmi.segment_id;
439 0 : const int ref_frame = xd->mode_info_context->mbmi.ref_frame;
440 :
441 0 : filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
442 :
443 0 : if (mb_row != pc->mb_rows - 1) {
444 : /* Save decoded MB last row data for next-row decoding */
445 0 : memcpy((pbi->mt_yabove_row[mb_row + 1] + 32 + mb_col * 16),
446 0 : (xd->dst.y_buffer + 15 * recon_y_stride), 16);
447 0 : memcpy((pbi->mt_uabove_row[mb_row + 1] + 16 + mb_col * 8),
448 0 : (xd->dst.u_buffer + 7 * recon_uv_stride), 8);
449 0 : memcpy((pbi->mt_vabove_row[mb_row + 1] + 16 + mb_col * 8),
450 0 : (xd->dst.v_buffer + 7 * recon_uv_stride), 8);
451 : }
452 :
453 : /* save left_col for next MB decoding */
454 0 : if (mb_col != pc->mb_cols - 1) {
455 0 : MODE_INFO *next = xd->mode_info_context + 1;
456 :
457 0 : if (next->mbmi.ref_frame == INTRA_FRAME) {
458 0 : for (i = 0; i < 16; ++i) {
459 0 : pbi->mt_yleft_col[mb_row][i] =
460 0 : xd->dst.y_buffer[i * recon_y_stride + 15];
461 : }
462 0 : for (i = 0; i < 8; ++i) {
463 0 : pbi->mt_uleft_col[mb_row][i] =
464 0 : xd->dst.u_buffer[i * recon_uv_stride + 7];
465 0 : pbi->mt_vleft_col[mb_row][i] =
466 0 : xd->dst.v_buffer[i * recon_uv_stride + 7];
467 : }
468 : }
469 : }
470 :
471 : /* loopfilter on this macroblock. */
472 0 : if (filter_level) {
473 0 : if (pc->filter_type == NORMAL_LOOPFILTER) {
474 : loop_filter_info lfi;
475 0 : FRAME_TYPE frame_type = pc->frame_type;
476 0 : const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level];
477 0 : lfi.mblim = lfi_n->mblim[filter_level];
478 0 : lfi.blim = lfi_n->blim[filter_level];
479 0 : lfi.lim = lfi_n->lim[filter_level];
480 0 : lfi.hev_thr = lfi_n->hev_thr[hev_index];
481 :
482 0 : if (mb_col > 0)
483 0 : vp8_loop_filter_mbv(xd->dst.y_buffer, xd->dst.u_buffer,
484 0 : xd->dst.v_buffer, recon_y_stride,
485 : recon_uv_stride, &lfi);
486 :
487 0 : if (!skip_lf)
488 0 : vp8_loop_filter_bv(xd->dst.y_buffer, xd->dst.u_buffer,
489 0 : xd->dst.v_buffer, recon_y_stride,
490 : recon_uv_stride, &lfi);
491 :
492 : /* don't apply across umv border */
493 0 : if (mb_row > 0)
494 0 : vp8_loop_filter_mbh(xd->dst.y_buffer, xd->dst.u_buffer,
495 0 : xd->dst.v_buffer, recon_y_stride,
496 : recon_uv_stride, &lfi);
497 :
498 0 : if (!skip_lf)
499 0 : vp8_loop_filter_bh(xd->dst.y_buffer, xd->dst.u_buffer,
500 0 : xd->dst.v_buffer, recon_y_stride,
501 : recon_uv_stride, &lfi);
502 : } else {
503 0 : if (mb_col > 0)
504 0 : vp8_loop_filter_simple_mbv(xd->dst.y_buffer, recon_y_stride,
505 0 : lfi_n->mblim[filter_level]);
506 :
507 0 : if (!skip_lf)
508 0 : vp8_loop_filter_simple_bv(xd->dst.y_buffer, recon_y_stride,
509 0 : lfi_n->blim[filter_level]);
510 :
511 : /* don't apply across umv border */
512 0 : if (mb_row > 0)
513 0 : vp8_loop_filter_simple_mbh(xd->dst.y_buffer, recon_y_stride,
514 0 : lfi_n->mblim[filter_level]);
515 :
516 0 : if (!skip_lf)
517 0 : vp8_loop_filter_simple_bh(xd->dst.y_buffer, recon_y_stride,
518 0 : lfi_n->blim[filter_level]);
519 : }
520 : }
521 : }
522 :
523 0 : recon_yoffset += 16;
524 0 : recon_uvoffset += 8;
525 :
526 0 : ++xd->mode_info_context; /* next mb */
527 :
528 0 : xd->above_context++;
529 : }
530 :
531 : /* adjust to the next row of mbs */
532 0 : if (pbi->common.filter_level) {
533 0 : if (mb_row != pc->mb_rows - 1) {
534 0 : int lasty = yv12_fb_lst->y_width + VP8BORDERINPIXELS;
535 0 : int lastuv = (yv12_fb_lst->y_width >> 1) + (VP8BORDERINPIXELS >> 1);
536 :
537 0 : for (i = 0; i < 4; ++i) {
538 0 : pbi->mt_yabove_row[mb_row + 1][lasty + i] =
539 0 : pbi->mt_yabove_row[mb_row + 1][lasty - 1];
540 0 : pbi->mt_uabove_row[mb_row + 1][lastuv + i] =
541 0 : pbi->mt_uabove_row[mb_row + 1][lastuv - 1];
542 0 : pbi->mt_vabove_row[mb_row + 1][lastuv + i] =
543 0 : pbi->mt_vabove_row[mb_row + 1][lastuv - 1];
544 : }
545 : }
546 : } else {
547 0 : vp8_extend_mb_row(yv12_fb_new, xd->dst.y_buffer + 16,
548 0 : xd->dst.u_buffer + 8, xd->dst.v_buffer + 8);
549 : }
550 :
551 : /* last MB of row is ready just after extension is done */
552 0 : protected_write(&pbi->pmutex[mb_row], current_mb_col, mb_col + nsync);
553 :
554 0 : ++xd->mode_info_context; /* skip prediction column */
555 0 : xd->up_available = 1;
556 :
557 : /* since we have multithread */
558 0 : xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count;
559 : }
560 :
561 : /* signal end of frame decoding if this thread processed the last mb_row */
562 0 : if (last_mb_row == (pc->mb_rows - 1)) sem_post(&pbi->h_event_end_decoding);
563 0 : }
564 :
565 0 : static THREAD_FUNCTION thread_decoding_proc(void *p_data) {
566 0 : int ithread = ((DECODETHREAD_DATA *)p_data)->ithread;
567 0 : VP8D_COMP *pbi = (VP8D_COMP *)(((DECODETHREAD_DATA *)p_data)->ptr1);
568 0 : MB_ROW_DEC *mbrd = (MB_ROW_DEC *)(((DECODETHREAD_DATA *)p_data)->ptr2);
569 : ENTROPY_CONTEXT_PLANES mb_row_left_context;
570 :
571 : while (1) {
572 0 : if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd) == 0) break;
573 :
574 0 : if (sem_wait(&pbi->h_event_start_decoding[ithread]) == 0) {
575 0 : if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd) == 0) {
576 0 : break;
577 : } else {
578 0 : MACROBLOCKD *xd = &mbrd->mbd;
579 0 : xd->left_context = &mb_row_left_context;
580 :
581 0 : mt_decode_mb_rows(pbi, xd, ithread + 1);
582 : }
583 : }
584 : }
585 :
586 0 : return 0;
587 : }
588 :
589 0 : void vp8_decoder_create_threads(VP8D_COMP *pbi) {
590 0 : int core_count = 0;
591 : unsigned int ithread;
592 :
593 0 : pbi->b_multithreaded_rd = 0;
594 0 : pbi->allocated_decoding_thread_count = 0;
595 0 : pthread_mutex_init(&pbi->mt_mutex, NULL);
596 :
597 : /* limit decoding threads to the max number of token partitions */
598 0 : core_count = (pbi->max_threads > 8) ? 8 : pbi->max_threads;
599 :
600 : /* limit decoding threads to the available cores */
601 0 : if (core_count > pbi->common.processor_core_count) {
602 0 : core_count = pbi->common.processor_core_count;
603 : }
604 :
605 0 : if (core_count > 1) {
606 0 : pbi->b_multithreaded_rd = 1;
607 0 : pbi->decoding_thread_count = core_count - 1;
608 :
609 0 : CALLOC_ARRAY(pbi->h_decoding_thread, pbi->decoding_thread_count);
610 0 : CALLOC_ARRAY(pbi->h_event_start_decoding, pbi->decoding_thread_count);
611 0 : CALLOC_ARRAY_ALIGNED(pbi->mb_row_di, pbi->decoding_thread_count, 32);
612 0 : CALLOC_ARRAY(pbi->de_thread_data, pbi->decoding_thread_count);
613 :
614 0 : if (sem_init(&pbi->h_event_end_decoding, 0, 0)) {
615 0 : vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,
616 : "Failed to initialize semaphore");
617 : }
618 :
619 0 : for (ithread = 0; ithread < pbi->decoding_thread_count; ++ithread) {
620 0 : if (sem_init(&pbi->h_event_start_decoding[ithread], 0, 0)) break;
621 :
622 0 : vp8_setup_block_dptrs(&pbi->mb_row_di[ithread].mbd);
623 :
624 0 : pbi->de_thread_data[ithread].ithread = ithread;
625 0 : pbi->de_thread_data[ithread].ptr1 = (void *)pbi;
626 0 : pbi->de_thread_data[ithread].ptr2 = (void *)&pbi->mb_row_di[ithread];
627 :
628 0 : if (pthread_create(&pbi->h_decoding_thread[ithread], 0,
629 0 : thread_decoding_proc, &pbi->de_thread_data[ithread])) {
630 0 : sem_destroy(&pbi->h_event_start_decoding[ithread]);
631 0 : break;
632 : }
633 : }
634 :
635 0 : pbi->allocated_decoding_thread_count = ithread;
636 0 : if (pbi->allocated_decoding_thread_count !=
637 0 : (int)pbi->decoding_thread_count) {
638 : /* the remainder of cleanup cases will be handled in
639 : * vp8_decoder_remove_threads(). */
640 0 : if (pbi->allocated_decoding_thread_count == 0) {
641 0 : sem_destroy(&pbi->h_event_end_decoding);
642 : }
643 0 : vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,
644 : "Failed to create threads");
645 : }
646 : }
647 0 : }
648 :
649 0 : void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows) {
650 : int i;
651 :
652 : /* De-allocate mutex */
653 0 : if (pbi->pmutex != NULL) {
654 0 : for (i = 0; i < mb_rows; ++i) {
655 0 : pthread_mutex_destroy(&pbi->pmutex[i]);
656 : }
657 :
658 0 : vpx_free(pbi->pmutex);
659 0 : pbi->pmutex = NULL;
660 : }
661 :
662 0 : vpx_free(pbi->mt_current_mb_col);
663 0 : pbi->mt_current_mb_col = NULL;
664 :
665 : /* Free above_row buffers. */
666 0 : if (pbi->mt_yabove_row) {
667 0 : for (i = 0; i < mb_rows; ++i) {
668 0 : vpx_free(pbi->mt_yabove_row[i]);
669 0 : pbi->mt_yabove_row[i] = NULL;
670 : }
671 0 : vpx_free(pbi->mt_yabove_row);
672 0 : pbi->mt_yabove_row = NULL;
673 : }
674 :
675 0 : if (pbi->mt_uabove_row) {
676 0 : for (i = 0; i < mb_rows; ++i) {
677 0 : vpx_free(pbi->mt_uabove_row[i]);
678 0 : pbi->mt_uabove_row[i] = NULL;
679 : }
680 0 : vpx_free(pbi->mt_uabove_row);
681 0 : pbi->mt_uabove_row = NULL;
682 : }
683 :
684 0 : if (pbi->mt_vabove_row) {
685 0 : for (i = 0; i < mb_rows; ++i) {
686 0 : vpx_free(pbi->mt_vabove_row[i]);
687 0 : pbi->mt_vabove_row[i] = NULL;
688 : }
689 0 : vpx_free(pbi->mt_vabove_row);
690 0 : pbi->mt_vabove_row = NULL;
691 : }
692 :
693 : /* Free left_col buffers. */
694 0 : if (pbi->mt_yleft_col) {
695 0 : for (i = 0; i < mb_rows; ++i) {
696 0 : vpx_free(pbi->mt_yleft_col[i]);
697 0 : pbi->mt_yleft_col[i] = NULL;
698 : }
699 0 : vpx_free(pbi->mt_yleft_col);
700 0 : pbi->mt_yleft_col = NULL;
701 : }
702 :
703 0 : if (pbi->mt_uleft_col) {
704 0 : for (i = 0; i < mb_rows; ++i) {
705 0 : vpx_free(pbi->mt_uleft_col[i]);
706 0 : pbi->mt_uleft_col[i] = NULL;
707 : }
708 0 : vpx_free(pbi->mt_uleft_col);
709 0 : pbi->mt_uleft_col = NULL;
710 : }
711 :
712 0 : if (pbi->mt_vleft_col) {
713 0 : for (i = 0; i < mb_rows; ++i) {
714 0 : vpx_free(pbi->mt_vleft_col[i]);
715 0 : pbi->mt_vleft_col[i] = NULL;
716 : }
717 0 : vpx_free(pbi->mt_vleft_col);
718 0 : pbi->mt_vleft_col = NULL;
719 : }
720 0 : }
721 :
722 0 : void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows) {
723 0 : VP8_COMMON *const pc = &pbi->common;
724 : int i;
725 : int uv_width;
726 :
727 0 : if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd)) {
728 0 : vp8mt_de_alloc_temp_buffers(pbi, prev_mb_rows);
729 :
730 : /* our internal buffers are always multiples of 16 */
731 0 : if ((width & 0xf) != 0) width += 16 - (width & 0xf);
732 :
733 0 : if (width < 640) {
734 0 : pbi->sync_range = 1;
735 0 : } else if (width <= 1280) {
736 0 : pbi->sync_range = 8;
737 0 : } else if (width <= 2560) {
738 0 : pbi->sync_range = 16;
739 : } else {
740 0 : pbi->sync_range = 32;
741 : }
742 :
743 0 : uv_width = width >> 1;
744 :
745 : /* Allocate mutex */
746 0 : CHECK_MEM_ERROR(pbi->pmutex,
747 : vpx_malloc(sizeof(*pbi->pmutex) * pc->mb_rows));
748 0 : if (pbi->pmutex) {
749 0 : for (i = 0; i < pc->mb_rows; ++i) {
750 0 : pthread_mutex_init(&pbi->pmutex[i], NULL);
751 : }
752 : }
753 :
754 : /* Allocate an int for each mb row. */
755 0 : CALLOC_ARRAY(pbi->mt_current_mb_col, pc->mb_rows);
756 :
757 : /* Allocate memory for above_row buffers. */
758 0 : CALLOC_ARRAY(pbi->mt_yabove_row, pc->mb_rows);
759 0 : for (i = 0; i < pc->mb_rows; ++i)
760 0 : CHECK_MEM_ERROR(pbi->mt_yabove_row[i],
761 : vpx_memalign(16, sizeof(unsigned char) *
762 : (width + (VP8BORDERINPIXELS << 1))));
763 :
764 0 : CALLOC_ARRAY(pbi->mt_uabove_row, pc->mb_rows);
765 0 : for (i = 0; i < pc->mb_rows; ++i)
766 0 : CHECK_MEM_ERROR(pbi->mt_uabove_row[i],
767 : vpx_memalign(16, sizeof(unsigned char) *
768 : (uv_width + VP8BORDERINPIXELS)));
769 :
770 0 : CALLOC_ARRAY(pbi->mt_vabove_row, pc->mb_rows);
771 0 : for (i = 0; i < pc->mb_rows; ++i)
772 0 : CHECK_MEM_ERROR(pbi->mt_vabove_row[i],
773 : vpx_memalign(16, sizeof(unsigned char) *
774 : (uv_width + VP8BORDERINPIXELS)));
775 :
776 : /* Allocate memory for left_col buffers. */
777 0 : CALLOC_ARRAY(pbi->mt_yleft_col, pc->mb_rows);
778 0 : for (i = 0; i < pc->mb_rows; ++i)
779 0 : CHECK_MEM_ERROR(pbi->mt_yleft_col[i],
780 : vpx_calloc(sizeof(unsigned char) * 16, 1));
781 :
782 0 : CALLOC_ARRAY(pbi->mt_uleft_col, pc->mb_rows);
783 0 : for (i = 0; i < pc->mb_rows; ++i)
784 0 : CHECK_MEM_ERROR(pbi->mt_uleft_col[i],
785 : vpx_calloc(sizeof(unsigned char) * 8, 1));
786 :
787 0 : CALLOC_ARRAY(pbi->mt_vleft_col, pc->mb_rows);
788 0 : for (i = 0; i < pc->mb_rows; ++i)
789 0 : CHECK_MEM_ERROR(pbi->mt_vleft_col[i],
790 : vpx_calloc(sizeof(unsigned char) * 8, 1));
791 : }
792 0 : }
793 :
794 0 : void vp8_decoder_remove_threads(VP8D_COMP *pbi) {
795 : /* shutdown MB Decoding thread; */
796 0 : if (protected_read(&pbi->mt_mutex, &pbi->b_multithreaded_rd)) {
797 : int i;
798 0 : protected_write(&pbi->mt_mutex, &pbi->b_multithreaded_rd, 0);
799 :
800 : /* allow all threads to exit */
801 0 : for (i = 0; i < pbi->allocated_decoding_thread_count; ++i) {
802 0 : sem_post(&pbi->h_event_start_decoding[i]);
803 0 : pthread_join(pbi->h_decoding_thread[i], NULL);
804 : }
805 :
806 0 : for (i = 0; i < pbi->allocated_decoding_thread_count; ++i) {
807 0 : sem_destroy(&pbi->h_event_start_decoding[i]);
808 : }
809 :
810 0 : if (pbi->allocated_decoding_thread_count) {
811 0 : sem_destroy(&pbi->h_event_end_decoding);
812 : }
813 :
814 0 : vpx_free(pbi->h_decoding_thread);
815 0 : pbi->h_decoding_thread = NULL;
816 :
817 0 : vpx_free(pbi->h_event_start_decoding);
818 0 : pbi->h_event_start_decoding = NULL;
819 :
820 0 : vpx_free(pbi->mb_row_di);
821 0 : pbi->mb_row_di = NULL;
822 :
823 0 : vpx_free(pbi->de_thread_data);
824 0 : pbi->de_thread_data = NULL;
825 :
826 0 : vp8mt_de_alloc_temp_buffers(pbi, pbi->common.mb_rows);
827 : }
828 0 : pthread_mutex_destroy(&pbi->mt_mutex);
829 0 : }
830 :
831 0 : void vp8mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd) {
832 0 : VP8_COMMON *pc = &pbi->common;
833 : unsigned int i;
834 : int j;
835 :
836 0 : int filter_level = pc->filter_level;
837 0 : YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
838 :
839 0 : if (filter_level) {
840 : /* Set above_row buffer to 127 for decoding first MB row */
841 0 : memset(pbi->mt_yabove_row[0] + VP8BORDERINPIXELS - 1, 127,
842 0 : yv12_fb_new->y_width + 5);
843 0 : memset(pbi->mt_uabove_row[0] + (VP8BORDERINPIXELS >> 1) - 1, 127,
844 0 : (yv12_fb_new->y_width >> 1) + 5);
845 0 : memset(pbi->mt_vabove_row[0] + (VP8BORDERINPIXELS >> 1) - 1, 127,
846 0 : (yv12_fb_new->y_width >> 1) + 5);
847 :
848 0 : for (j = 1; j < pc->mb_rows; ++j) {
849 0 : memset(pbi->mt_yabove_row[j] + VP8BORDERINPIXELS - 1, (unsigned char)129,
850 : 1);
851 0 : memset(pbi->mt_uabove_row[j] + (VP8BORDERINPIXELS >> 1) - 1,
852 : (unsigned char)129, 1);
853 0 : memset(pbi->mt_vabove_row[j] + (VP8BORDERINPIXELS >> 1) - 1,
854 : (unsigned char)129, 1);
855 : }
856 :
857 : /* Set left_col to 129 initially */
858 0 : for (j = 0; j < pc->mb_rows; ++j) {
859 0 : memset(pbi->mt_yleft_col[j], (unsigned char)129, 16);
860 0 : memset(pbi->mt_uleft_col[j], (unsigned char)129, 8);
861 0 : memset(pbi->mt_vleft_col[j], (unsigned char)129, 8);
862 : }
863 :
864 : /* Initialize the loop filter for this frame. */
865 0 : vp8_loop_filter_frame_init(pc, &pbi->mb, filter_level);
866 : } else {
867 0 : vp8_setup_intra_recon_top_line(yv12_fb_new);
868 : }
869 :
870 0 : setup_decoding_thread_data(pbi, xd, pbi->mb_row_di,
871 0 : pbi->decoding_thread_count);
872 :
873 0 : for (i = 0; i < pbi->decoding_thread_count; ++i) {
874 0 : sem_post(&pbi->h_event_start_decoding[i]);
875 : }
876 :
877 0 : mt_decode_mb_rows(pbi, xd, 0);
878 :
879 0 : sem_wait(&pbi->h_event_end_decoding); /* add back for each frame */
880 0 : }
|
