FFmpeg
vp8.c
Go to the documentation of this file.
1 /*
2  * VP7/VP8 compatible video decoder
3  *
4  * Copyright (C) 2010 David Conrad
5  * Copyright (C) 2010 Ronald S. Bultje
6  * Copyright (C) 2010 Fiona Glaser
7  * Copyright (C) 2012 Daniel Kang
8  * Copyright (C) 2014 Peter Ross
9  *
10  * This file is part of FFmpeg.
11  *
12  * FFmpeg is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU Lesser General Public
14  * License as published by the Free Software Foundation; either
15  * version 2.1 of the License, or (at your option) any later version.
16  *
17  * FFmpeg is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20  * Lesser General Public License for more details.
21  *
22  * You should have received a copy of the GNU Lesser General Public
23  * License along with FFmpeg; if not, write to the Free Software
24  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25  */
26 
27 #include "config_components.h"
28 
29 #include "libavutil/mem_internal.h"
30 
31 #include "avcodec.h"
32 #include "codec_internal.h"
33 #include "decode.h"
34 #include "hwaccel_internal.h"
35 #include "hwconfig.h"
36 #include "mathops.h"
37 #include "refstruct.h"
38 #include "thread.h"
39 #include "threadframe.h"
40 #include "vp8.h"
41 #include "vp89_rac.h"
42 #include "vp8data.h"
43 #include "vpx_rac.h"
44 
45 #if ARCH_ARM
46 # include "arm/vp8.h"
47 #endif
48 
49 // fixme: add 1 bit to all the calls to this?
51 {
52  int v;
53 
54  if (!vp89_rac_get(c))
55  return 0;
56 
57  v = vp89_rac_get_uint(c, bits);
58 
59  if (vp89_rac_get(c))
60  v = -v;
61 
62  return v;
63 }
64 
66 {
67  int v = vp89_rac_get_uint(c, 7) << 1;
68  return v + !v;
69 }
70 
71 // DCTextra
72 static int vp8_rac_get_coeff(VPXRangeCoder *c, const uint8_t *prob)
73 {
74  int v = 0;
75 
76  do {
77  v = (v<<1) + vpx_rac_get_prob(c, *prob++);
78  } while (*prob);
79 
80  return v;
81 }
82 
83 static void free_buffers(VP8Context *s)
84 {
85  int i;
86  if (s->thread_data)
87  for (i = 0; i < MAX_THREADS; i++) {
88 #if HAVE_THREADS
89  pthread_cond_destroy(&s->thread_data[i].cond);
90  pthread_mutex_destroy(&s->thread_data[i].lock);
91 #endif
92  av_freep(&s->thread_data[i].filter_strength);
93  }
94  av_freep(&s->thread_data);
95  av_freep(&s->macroblocks_base);
96  av_freep(&s->intra4x4_pred_mode_top);
97  av_freep(&s->top_nnz);
98  av_freep(&s->top_border);
99 
100  s->macroblocks = NULL;
101 }
102 
104 {
105  int ret;
106  if ((ret = ff_thread_get_ext_buffer(s->avctx, &f->tf,
107  ref ? AV_GET_BUFFER_FLAG_REF : 0)) < 0)
108  return ret;
109  if (!(f->seg_map = ff_refstruct_allocz(s->mb_width * s->mb_height)))
110  goto fail;
111  ret = ff_hwaccel_frame_priv_alloc(s->avctx, &f->hwaccel_picture_private);
112  if (ret < 0)
113  goto fail;
114 
115  return 0;
116 
117 fail:
118  ff_refstruct_unref(&f->seg_map);
120  return ret;
121 }
122 
124 {
125  ff_refstruct_unref(&f->seg_map);
126  ff_refstruct_unref(&f->hwaccel_picture_private);
128 }
129 
130 #if CONFIG_VP8_DECODER
131 static int vp8_ref_frame(VP8Frame *dst, const VP8Frame *src)
132 {
133  int ret;
134 
135  vp8_release_frame(dst);
136 
137  if ((ret = ff_thread_ref_frame(&dst->tf, &src->tf)) < 0)
138  return ret;
139  ff_refstruct_replace(&dst->seg_map, src->seg_map);
141  src->hwaccel_picture_private);
142 
143  return 0;
144 }
145 #endif /* CONFIG_VP8_DECODER */
146 
147 static void vp8_decode_flush_impl(AVCodecContext *avctx, int free_mem)
148 {
149  VP8Context *s = avctx->priv_data;
150  int i;
151 
152  for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++)
153  vp8_release_frame(&s->frames[i]);
154  memset(s->framep, 0, sizeof(s->framep));
155 
156  if (free_mem)
157  free_buffers(s);
158 
159  if (FF_HW_HAS_CB(avctx, flush))
160  FF_HW_SIMPLE_CALL(avctx, flush);
161 }
162 
163 static void vp8_decode_flush(AVCodecContext *avctx)
164 {
165  vp8_decode_flush_impl(avctx, 0);
166 }
167 
169 {
170  VP8Frame *frame = NULL;
171  int i;
172 
173  // find a free buffer
174  for (i = 0; i < 5; i++)
175  if (&s->frames[i] != s->framep[VP8_FRAME_CURRENT] &&
176  &s->frames[i] != s->framep[VP8_FRAME_PREVIOUS] &&
177  &s->frames[i] != s->framep[VP8_FRAME_GOLDEN] &&
178  &s->frames[i] != s->framep[VP8_FRAME_ALTREF]) {
179  frame = &s->frames[i];
180  break;
181  }
182  if (i == 5) {
183  av_log(s->avctx, AV_LOG_FATAL, "Ran out of free frames!\n");
184  abort();
185  }
186  if (frame->tf.f->buf[0])
188 
189  return frame;
190 }
191 
193 {
194  enum AVPixelFormat pix_fmts[] = {
195 #if CONFIG_VP8_VAAPI_HWACCEL
197 #endif
198 #if CONFIG_VP8_NVDEC_HWACCEL
200 #endif
203  };
204 
205  return ff_get_format(s->avctx, pix_fmts);
206 }
207 
208 static av_always_inline
209 int update_dimensions(VP8Context *s, int width, int height, int is_vp7)
210 {
211  AVCodecContext *avctx = s->avctx;
212  int i, ret, dim_reset = 0;
213 
214  if (width != s->avctx->width || ((width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) && s->macroblocks_base ||
215  height != s->avctx->height) {
216  vp8_decode_flush_impl(s->avctx, 1);
217 
218  ret = ff_set_dimensions(s->avctx, width, height);
219  if (ret < 0)
220  return ret;
221 
222  dim_reset = (s->macroblocks_base != NULL);
223  }
224 
225  if ((s->pix_fmt == AV_PIX_FMT_NONE || dim_reset) &&
226  !s->actually_webp && !is_vp7) {
227  s->pix_fmt = get_pixel_format(s);
228  if (s->pix_fmt < 0)
229  return AVERROR(EINVAL);
230  avctx->pix_fmt = s->pix_fmt;
231  }
232 
233  s->mb_width = (s->avctx->coded_width + 15) / 16;
234  s->mb_height = (s->avctx->coded_height + 15) / 16;
235 
236  s->mb_layout = is_vp7 || avctx->active_thread_type == FF_THREAD_SLICE &&
237  avctx->thread_count > 1;
238  if (!s->mb_layout) { // Frame threading and one thread
239  s->macroblocks_base = av_mallocz((s->mb_width + s->mb_height * 2 + 1) *
240  sizeof(*s->macroblocks));
241  s->intra4x4_pred_mode_top = av_mallocz(s->mb_width * 4);
242  } else // Sliced threading
243  s->macroblocks_base = av_mallocz((s->mb_width + 2) * (s->mb_height + 2) *
244  sizeof(*s->macroblocks));
245  s->top_nnz = av_mallocz(s->mb_width * sizeof(*s->top_nnz));
246  s->top_border = av_mallocz((s->mb_width + 1) * sizeof(*s->top_border));
247  s->thread_data = av_mallocz(MAX_THREADS * sizeof(VP8ThreadData));
248 
249  if (!s->macroblocks_base || !s->top_nnz || !s->top_border ||
250  !s->thread_data || (!s->intra4x4_pred_mode_top && !s->mb_layout)) {
251  free_buffers(s);
252  return AVERROR(ENOMEM);
253  }
254 
255  for (i = 0; i < MAX_THREADS; i++) {
256  s->thread_data[i].filter_strength =
257  av_mallocz(s->mb_width * sizeof(*s->thread_data[0].filter_strength));
258  if (!s->thread_data[i].filter_strength) {
259  free_buffers(s);
260  return AVERROR(ENOMEM);
261  }
262 #if HAVE_THREADS
263  pthread_mutex_init(&s->thread_data[i].lock, NULL);
264  pthread_cond_init(&s->thread_data[i].cond, NULL);
265 #endif
266  }
267 
268  s->macroblocks = s->macroblocks_base + 1;
269 
270  return 0;
271 }
272 
274 {
276 }
277 
279 {
281 }
282 
283 
285 {
286  VPXRangeCoder *c = &s->c;
287  int i;
288 
289  s->segmentation.update_map = vp89_rac_get(c);
290  s->segmentation.update_feature_data = vp89_rac_get(c);
291 
292  if (s->segmentation.update_feature_data) {
293  s->segmentation.absolute_vals = vp89_rac_get(c);
294 
295  for (i = 0; i < 4; i++)
296  s->segmentation.base_quant[i] = vp8_rac_get_sint(c, 7);
297 
298  for (i = 0; i < 4; i++)
299  s->segmentation.filter_level[i] = vp8_rac_get_sint(c, 6);
300  }
301  if (s->segmentation.update_map)
302  for (i = 0; i < 3; i++)
303  s->prob->segmentid[i] = vp89_rac_get(c) ? vp89_rac_get_uint(c, 8) : 255;
304 }
305 
307 {
308  VPXRangeCoder *c = &s->c;
309  int i;
310 
311  for (i = 0; i < 4; i++) {
312  if (vp89_rac_get(c)) {
313  s->lf_delta.ref[i] = vp89_rac_get_uint(c, 6);
314 
315  if (vp89_rac_get(c))
316  s->lf_delta.ref[i] = -s->lf_delta.ref[i];
317  }
318  }
319 
320  for (i = MODE_I4x4; i <= VP8_MVMODE_SPLIT; i++) {
321  if (vp89_rac_get(c)) {
322  s->lf_delta.mode[i] = vp89_rac_get_uint(c, 6);
323 
324  if (vp89_rac_get(c))
325  s->lf_delta.mode[i] = -s->lf_delta.mode[i];
326  }
327  }
328 }
329 
330 static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size)
331 {
332  const uint8_t *sizes = buf;
333  int i;
334  int ret;
335 
336  s->num_coeff_partitions = 1 << vp89_rac_get_uint(&s->c, 2);
337 
338  buf += 3 * (s->num_coeff_partitions - 1);
339  buf_size -= 3 * (s->num_coeff_partitions - 1);
340  if (buf_size < 0)
341  return -1;
342 
343  for (i = 0; i < s->num_coeff_partitions - 1; i++) {
344  int size = AV_RL24(sizes + 3 * i);
345  if (buf_size - size < 0)
346  return -1;
347  s->coeff_partition_size[i] = size;
348 
349  ret = ff_vpx_init_range_decoder(&s->coeff_partition[i], buf, size);
350  if (ret < 0)
351  return ret;
352  buf += size;
353  buf_size -= size;
354  }
355 
356  s->coeff_partition_size[i] = buf_size;
357  ff_vpx_init_range_decoder(&s->coeff_partition[i], buf, buf_size);
358 
359  return 0;
360 }
361 
363 {
364  VPXRangeCoder *c = &s->c;
365 
366  int yac_qi = vp89_rac_get_uint(c, 7);
367  int ydc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
368  int y2dc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
369  int y2ac_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
370  int uvdc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
371  int uvac_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
372 
373  s->qmat[0].luma_qmul[0] = vp7_ydc_qlookup[ydc_qi];
374  s->qmat[0].luma_qmul[1] = vp7_yac_qlookup[yac_qi];
375  s->qmat[0].luma_dc_qmul[0] = vp7_y2dc_qlookup[y2dc_qi];
376  s->qmat[0].luma_dc_qmul[1] = vp7_y2ac_qlookup[y2ac_qi];
377  s->qmat[0].chroma_qmul[0] = FFMIN(vp7_ydc_qlookup[uvdc_qi], 132);
378  s->qmat[0].chroma_qmul[1] = vp7_yac_qlookup[uvac_qi];
379 }
380 
382 {
383  VPXRangeCoder *c = &s->c;
384  int i, base_qi;
385 
386  s->quant.yac_qi = vp89_rac_get_uint(c, 7);
387  s->quant.ydc_delta = vp8_rac_get_sint(c, 4);
388  s->quant.y2dc_delta = vp8_rac_get_sint(c, 4);
389  s->quant.y2ac_delta = vp8_rac_get_sint(c, 4);
390  s->quant.uvdc_delta = vp8_rac_get_sint(c, 4);
391  s->quant.uvac_delta = vp8_rac_get_sint(c, 4);
392 
393  for (i = 0; i < 4; i++) {
394  if (s->segmentation.enabled) {
395  base_qi = s->segmentation.base_quant[i];
396  if (!s->segmentation.absolute_vals)
397  base_qi += s->quant.yac_qi;
398  } else
399  base_qi = s->quant.yac_qi;
400 
401  s->qmat[i].luma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.ydc_delta, 7)];
402  s->qmat[i].luma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi, 7)];
403  s->qmat[i].luma_dc_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.y2dc_delta, 7)] * 2;
404  /* 101581>>16 is equivalent to 155/100 */
405  s->qmat[i].luma_dc_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi + s->quant.y2ac_delta, 7)] * 101581 >> 16;
406  s->qmat[i].chroma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.uvdc_delta, 7)];
407  s->qmat[i].chroma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi + s->quant.uvac_delta, 7)];
408 
409  s->qmat[i].luma_dc_qmul[1] = FFMAX(s->qmat[i].luma_dc_qmul[1], 8);
410  s->qmat[i].chroma_qmul[0] = FFMIN(s->qmat[i].chroma_qmul[0], 132);
411  }
412 }
413 
414 /**
415  * Determine which buffers golden and altref should be updated with after this frame.
416  * The spec isn't clear here, so I'm going by my understanding of what libvpx does
417  *
418  * Intra frames update all 3 references
419  * Inter frames update VP8_FRAME_PREVIOUS if the update_last flag is set
420  * If the update (golden|altref) flag is set, it's updated with the current frame
421  * if update_last is set, and VP8_FRAME_PREVIOUS otherwise.
422  * If the flag is not set, the number read means:
423  * 0: no update
424  * 1: VP8_FRAME_PREVIOUS
425  * 2: update golden with altref, or update altref with golden
426  */
428 {
429  VPXRangeCoder *c = &s->c;
430 
431  if (update)
432  return VP8_FRAME_CURRENT;
433 
434  switch (vp89_rac_get_uint(c, 2)) {
435  case 1:
436  return VP8_FRAME_PREVIOUS;
437  case 2:
439  }
440  return VP8_FRAME_NONE;
441 }
442 
444 {
445  int i, j;
446  for (i = 0; i < 4; i++)
447  for (j = 0; j < 16; j++)
448  memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]],
449  sizeof(s->prob->token[i][j]));
450 }
451 
453 {
454  VPXRangeCoder *c = &s->c;
455  int i, j, k, l, m;
456 
457  for (i = 0; i < 4; i++)
458  for (j = 0; j < 8; j++)
459  for (k = 0; k < 3; k++)
460  for (l = 0; l < NUM_DCT_TOKENS-1; l++)
462  int prob = vp89_rac_get_uint(c, 8);
463  for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++)
464  s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob;
465  }
466 }
467 
468 #define VP7_MVC_SIZE 17
469 #define VP8_MVC_SIZE 19
470 
472  int mvc_size)
473 {
474  VPXRangeCoder *c = &s->c;
475  int i, j;
476 
477  if (vp89_rac_get(c))
478  for (i = 0; i < 4; i++)
479  s->prob->pred16x16[i] = vp89_rac_get_uint(c, 8);
480  if (vp89_rac_get(c))
481  for (i = 0; i < 3; i++)
482  s->prob->pred8x8c[i] = vp89_rac_get_uint(c, 8);
483 
484  // 17.2 MV probability update
485  for (i = 0; i < 2; i++)
486  for (j = 0; j < mvc_size; j++)
488  s->prob->mvc[i][j] = vp8_rac_get_nn(c);
489 }
490 
491 static void update_refs(VP8Context *s)
492 {
493  VPXRangeCoder *c = &s->c;
494 
495  int update_golden = vp89_rac_get(c);
496  int update_altref = vp89_rac_get(c);
497 
498  s->update_golden = ref_to_update(s, update_golden, VP8_FRAME_GOLDEN);
499  s->update_altref = ref_to_update(s, update_altref, VP8_FRAME_ALTREF);
500 }
501 
502 static void copy_chroma(AVFrame *dst, const AVFrame *src, int width, int height)
503 {
504  int i, j;
505 
506  for (j = 1; j < 3; j++) {
507  for (i = 0; i < height / 2; i++)
508  memcpy(dst->data[j] + i * dst->linesize[j],
509  src->data[j] + i * src->linesize[j], width / 2);
510  }
511 }
512 
513 static void fade(uint8_t *dst, ptrdiff_t dst_linesize,
514  const uint8_t *src, ptrdiff_t src_linesize,
515  int width, int height,
516  int alpha, int beta)
517 {
518  int i, j;
519  for (j = 0; j < height; j++) {
520  const uint8_t *src2 = src + j * src_linesize;
521  uint8_t *dst2 = dst + j * dst_linesize;
522  for (i = 0; i < width; i++) {
523  uint8_t y = src2[i];
524  dst2[i] = av_clip_uint8(y + ((y * beta) >> 8) + alpha);
525  }
526  }
527 }
528 
529 static int vp7_fade_frame(VP8Context *s, int alpha, int beta)
530 {
531  int ret;
532 
533  if (!s->keyframe && (alpha || beta)) {
534  int width = s->mb_width * 16;
535  int height = s->mb_height * 16;
536  const AVFrame *src;
537  AVFrame *dst;
538 
539  if (!s->framep[VP8_FRAME_PREVIOUS] ||
540  !s->framep[VP8_FRAME_GOLDEN]) {
541  av_log(s->avctx, AV_LOG_WARNING, "Discarding interframe without a prior keyframe!\n");
542  return AVERROR_INVALIDDATA;
543  }
544 
545  src =
546  dst = s->framep[VP8_FRAME_PREVIOUS]->tf.f;
547 
548  /* preserve the golden frame, write a new previous frame */
549  if (s->framep[VP8_FRAME_GOLDEN] == s->framep[VP8_FRAME_PREVIOUS]) {
551  if ((ret = vp8_alloc_frame(s, s->framep[VP8_FRAME_PREVIOUS], 1)) < 0)
552  return ret;
553 
554  dst = s->framep[VP8_FRAME_PREVIOUS]->tf.f;
555 
556  copy_chroma(dst, src, width, height);
557  }
558 
559  fade(dst->data[0], dst->linesize[0],
560  src->data[0], src->linesize[0],
561  width, height, alpha, beta);
562  }
563 
564  return 0;
565 }
566 
567 static int vp7_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
568 {
569  VPXRangeCoder *c = &s->c;
570  int part1_size, hscale, vscale, i, j, ret;
571  int width = s->avctx->width;
572  int height = s->avctx->height;
573  int alpha = 0;
574  int beta = 0;
575  int fade_present = 1;
576 
577  if (buf_size < 4) {
578  return AVERROR_INVALIDDATA;
579  }
580 
581  s->profile = (buf[0] >> 1) & 7;
582  if (s->profile > 1) {
583  avpriv_request_sample(s->avctx, "Unknown profile %d", s->profile);
584  return AVERROR_INVALIDDATA;
585  }
586 
587  s->keyframe = !(buf[0] & 1);
588  s->invisible = 0;
589  part1_size = AV_RL24(buf) >> 4;
590 
591  if (buf_size < 4 - s->profile + part1_size) {
592  av_log(s->avctx, AV_LOG_ERROR, "Buffer size %d is too small, needed : %d\n", buf_size, 4 - s->profile + part1_size);
593  return AVERROR_INVALIDDATA;
594  }
595 
596  buf += 4 - s->profile;
597  buf_size -= 4 - s->profile;
598 
599  memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, sizeof(s->put_pixels_tab));
600 
601  ret = ff_vpx_init_range_decoder(c, buf, part1_size);
602  if (ret < 0)
603  return ret;
604  buf += part1_size;
605  buf_size -= part1_size;
606 
607  /* A. Dimension information (keyframes only) */
608  if (s->keyframe) {
609  width = vp89_rac_get_uint(c, 12);
610  height = vp89_rac_get_uint(c, 12);
611  hscale = vp89_rac_get_uint(c, 2);
612  vscale = vp89_rac_get_uint(c, 2);
613  if (hscale || vscale)
614  avpriv_request_sample(s->avctx, "Upscaling");
615 
616  s->update_golden = s->update_altref = VP8_FRAME_CURRENT;
618  memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter,
619  sizeof(s->prob->pred16x16));
620  memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter,
621  sizeof(s->prob->pred8x8c));
622  for (i = 0; i < 2; i++)
623  memcpy(s->prob->mvc[i], vp7_mv_default_prob[i],
624  sizeof(vp7_mv_default_prob[i]));
625  memset(&s->segmentation, 0, sizeof(s->segmentation));
626  memset(&s->lf_delta, 0, sizeof(s->lf_delta));
627  memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan));
628  }
629 
630  if (s->keyframe || s->profile > 0)
631  memset(s->inter_dc_pred, 0 , sizeof(s->inter_dc_pred));
632 
633  /* B. Decoding information for all four macroblock-level features */
634  for (i = 0; i < 4; i++) {
635  s->feature_enabled[i] = vp89_rac_get(c);
636  if (s->feature_enabled[i]) {
637  s->feature_present_prob[i] = vp89_rac_get_uint(c, 8);
638 
639  for (j = 0; j < 3; j++)
640  s->feature_index_prob[i][j] =
641  vp89_rac_get(c) ? vp89_rac_get_uint(c, 8) : 255;
642 
643  if (vp7_feature_value_size[s->profile][i])
644  for (j = 0; j < 4; j++)
645  s->feature_value[i][j] =
647  }
648  }
649 
650  s->segmentation.enabled = 0;
651  s->segmentation.update_map = 0;
652  s->lf_delta.enabled = 0;
653 
654  s->num_coeff_partitions = 1;
655  ret = ff_vpx_init_range_decoder(&s->coeff_partition[0], buf, buf_size);
656  if (ret < 0)
657  return ret;
658 
659  if (!s->macroblocks_base || /* first frame */
660  width != s->avctx->width || height != s->avctx->height ||
661  (width + 15) / 16 != s->mb_width || (height + 15) / 16 != s->mb_height) {
662  if ((ret = vp7_update_dimensions(s, width, height)) < 0)
663  return ret;
664  }
665 
666  /* C. Dequantization indices */
667  vp7_get_quants(s);
668 
669  /* D. Golden frame update flag (a Flag) for interframes only */
670  if (!s->keyframe) {
671  s->update_golden = vp89_rac_get(c) ? VP8_FRAME_CURRENT : VP8_FRAME_NONE;
672  s->sign_bias[VP8_FRAME_GOLDEN] = 0;
673  }
674 
675  s->update_last = 1;
676  s->update_probabilities = 1;
677 
678  if (s->profile > 0) {
679  s->update_probabilities = vp89_rac_get(c);
680  if (!s->update_probabilities)
681  s->prob[1] = s->prob[0];
682 
683  if (!s->keyframe)
684  fade_present = vp89_rac_get(c);
685  }
686 
687  if (vpx_rac_is_end(c))
688  return AVERROR_INVALIDDATA;
689  /* E. Fading information for previous frame */
690  if (fade_present && vp89_rac_get(c)) {
691  alpha = (int8_t) vp89_rac_get_uint(c, 8);
692  beta = (int8_t) vp89_rac_get_uint(c, 8);
693  }
694 
695  /* F. Loop filter type */
696  if (!s->profile)
697  s->filter.simple = vp89_rac_get(c);
698 
699  /* G. DCT coefficient ordering specification */
700  if (vp89_rac_get(c))
701  for (i = 1; i < 16; i++)
702  s->prob[0].scan[i] = ff_zigzag_scan[vp89_rac_get_uint(c, 4)];
703 
704  /* H. Loop filter levels */
705  if (s->profile > 0)
706  s->filter.simple = vp89_rac_get(c);
707  s->filter.level = vp89_rac_get_uint(c, 6);
708  s->filter.sharpness = vp89_rac_get_uint(c, 3);
709 
710  /* I. DCT coefficient probability update; 13.3 Token Probability Updates */
712 
713  s->mbskip_enabled = 0;
714 
715  /* J. The remaining frame header data occurs ONLY FOR INTERFRAMES */
716  if (!s->keyframe) {
717  s->prob->intra = vp89_rac_get_uint(c, 8);
718  s->prob->last = vp89_rac_get_uint(c, 8);
720  }
721 
722  if (vpx_rac_is_end(c))
723  return AVERROR_INVALIDDATA;
724 
725  if ((ret = vp7_fade_frame(s, alpha, beta)) < 0)
726  return ret;
727 
728  return 0;
729 }
730 
731 static int vp8_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
732 {
733  VPXRangeCoder *c = &s->c;
734  int header_size, hscale, vscale, ret;
735  int width = s->avctx->width;
736  int height = s->avctx->height;
737 
738  if (buf_size < 3) {
739  av_log(s->avctx, AV_LOG_ERROR, "Insufficent data (%d) for header\n", buf_size);
740  return AVERROR_INVALIDDATA;
741  }
742 
743  s->keyframe = !(buf[0] & 1);
744  s->profile = (buf[0]>>1) & 7;
745  s->invisible = !(buf[0] & 0x10);
746  header_size = AV_RL24(buf) >> 5;
747  buf += 3;
748  buf_size -= 3;
749 
750  s->header_partition_size = header_size;
751 
752  if (s->profile > 3)
753  av_log(s->avctx, AV_LOG_WARNING, "Unknown profile %d\n", s->profile);
754 
755  if (!s->profile)
756  memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab,
757  sizeof(s->put_pixels_tab));
758  else // profile 1-3 use bilinear, 4+ aren't defined so whatever
759  memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_bilinear_pixels_tab,
760  sizeof(s->put_pixels_tab));
761 
762  if (header_size > buf_size - 7 * s->keyframe) {
763  av_log(s->avctx, AV_LOG_ERROR, "Header size larger than data provided\n");
764  return AVERROR_INVALIDDATA;
765  }
766 
767  if (s->keyframe) {
768  if (AV_RL24(buf) != 0x2a019d) {
769  av_log(s->avctx, AV_LOG_ERROR,
770  "Invalid start code 0x%x\n", AV_RL24(buf));
771  return AVERROR_INVALIDDATA;
772  }
773  width = AV_RL16(buf + 3) & 0x3fff;
774  height = AV_RL16(buf + 5) & 0x3fff;
775  hscale = buf[4] >> 6;
776  vscale = buf[6] >> 6;
777  buf += 7;
778  buf_size -= 7;
779 
780  if (hscale || vscale)
781  avpriv_request_sample(s->avctx, "Upscaling");
782 
783  s->update_golden = s->update_altref = VP8_FRAME_CURRENT;
785  memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter,
786  sizeof(s->prob->pred16x16));
787  memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter,
788  sizeof(s->prob->pred8x8c));
789  memcpy(s->prob->mvc, vp8_mv_default_prob,
790  sizeof(s->prob->mvc));
791  memset(&s->segmentation, 0, sizeof(s->segmentation));
792  memset(&s->lf_delta, 0, sizeof(s->lf_delta));
793  }
794 
795  ret = ff_vpx_init_range_decoder(c, buf, header_size);
796  if (ret < 0)
797  return ret;
798  buf += header_size;
799  buf_size -= header_size;
800 
801  if (s->keyframe) {
802  s->colorspace = vp89_rac_get(c);
803  if (s->colorspace)
804  av_log(s->avctx, AV_LOG_WARNING, "Unspecified colorspace\n");
805  s->fullrange = vp89_rac_get(c);
806  }
807 
808  if ((s->segmentation.enabled = vp89_rac_get(c)))
810  else
811  s->segmentation.update_map = 0; // FIXME: move this to some init function?
812 
813  s->filter.simple = vp89_rac_get(c);
814  s->filter.level = vp89_rac_get_uint(c, 6);
815  s->filter.sharpness = vp89_rac_get_uint(c, 3);
816 
817  if ((s->lf_delta.enabled = vp89_rac_get(c))) {
818  s->lf_delta.update = vp89_rac_get(c);
819  if (s->lf_delta.update)
821  }
822 
823  if (setup_partitions(s, buf, buf_size)) {
824  av_log(s->avctx, AV_LOG_ERROR, "Invalid partitions\n");
825  return AVERROR_INVALIDDATA;
826  }
827 
828  if (!s->macroblocks_base || /* first frame */
829  width != s->avctx->width || height != s->avctx->height ||
830  (width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height)
831  if ((ret = vp8_update_dimensions(s, width, height)) < 0)
832  return ret;
833 
834  vp8_get_quants(s);
835 
836  if (!s->keyframe) {
837  update_refs(s);
838  s->sign_bias[VP8_FRAME_GOLDEN] = vp89_rac_get(c);
839  s->sign_bias[VP8_FRAME_ALTREF] = vp89_rac_get(c);
840  }
841 
842  // if we aren't saving this frame's probabilities for future frames,
843  // make a copy of the current probabilities
844  if (!(s->update_probabilities = vp89_rac_get(c)))
845  s->prob[1] = s->prob[0];
846 
847  s->update_last = s->keyframe || vp89_rac_get(c);
848 
850 
851  if ((s->mbskip_enabled = vp89_rac_get(c)))
852  s->prob->mbskip = vp89_rac_get_uint(c, 8);
853 
854  if (!s->keyframe) {
855  s->prob->intra = vp89_rac_get_uint(c, 8);
856  s->prob->last = vp89_rac_get_uint(c, 8);
857  s->prob->golden = vp89_rac_get_uint(c, 8);
859  }
860 
861  // Record the entropy coder state here so that hwaccels can use it.
862  s->c.code_word = vpx_rac_renorm(&s->c);
863  s->coder_state_at_header_end.input = s->c.buffer - (-s->c.bits / 8);
864  s->coder_state_at_header_end.range = s->c.high;
865  s->coder_state_at_header_end.value = s->c.code_word >> 16;
866  s->coder_state_at_header_end.bit_count = -s->c.bits % 8;
867 
868  return 0;
869 }
870 
871 static av_always_inline
872 void clamp_mv(const VP8mvbounds *s, VP8mv *dst, const VP8mv *src)
873 {
874  dst->x = av_clip(src->x, av_clip(s->mv_min.x, INT16_MIN, INT16_MAX),
875  av_clip(s->mv_max.x, INT16_MIN, INT16_MAX));
876  dst->y = av_clip(src->y, av_clip(s->mv_min.y, INT16_MIN, INT16_MAX),
877  av_clip(s->mv_max.y, INT16_MIN, INT16_MAX));
878 }
879 
880 /**
881  * Motion vector coding, 17.1.
882  */
883 static av_always_inline int read_mv_component(VPXRangeCoder *c, const uint8_t *p, int vp7)
884 {
885  int bit, x = 0;
886 
887  if (vpx_rac_get_prob_branchy(c, p[0])) {
888  int i;
889 
890  for (i = 0; i < 3; i++)
891  x += vpx_rac_get_prob(c, p[9 + i]) << i;
892  for (i = (vp7 ? 7 : 9); i > 3; i--)
893  x += vpx_rac_get_prob(c, p[9 + i]) << i;
894  if (!(x & (vp7 ? 0xF0 : 0xFFF0)) || vpx_rac_get_prob(c, p[12]))
895  x += 8;
896  } else {
897  // small_mvtree
898  const uint8_t *ps = p + 2;
899  bit = vpx_rac_get_prob(c, *ps);
900  ps += 1 + 3 * bit;
901  x += 4 * bit;
902  bit = vpx_rac_get_prob(c, *ps);
903  ps += 1 + bit;
904  x += 2 * bit;
905  x += vpx_rac_get_prob(c, *ps);
906  }
907 
908  return (x && vpx_rac_get_prob(c, p[1])) ? -x : x;
909 }
910 
911 static int vp7_read_mv_component(VPXRangeCoder *c, const uint8_t *p)
912 {
913  return read_mv_component(c, p, 1);
914 }
915 
916 static int vp8_read_mv_component(VPXRangeCoder *c, const uint8_t *p)
917 {
918  return read_mv_component(c, p, 0);
919 }
920 
921 static av_always_inline
922 const uint8_t *get_submv_prob(uint32_t left, uint32_t top, int is_vp7)
923 {
924  if (is_vp7)
925  return vp7_submv_prob;
926 
927  if (left == top)
928  return vp8_submv_prob[4 - !!left];
929  if (!top)
930  return vp8_submv_prob[2];
931  return vp8_submv_prob[1 - !!left];
932 }
933 
934 /**
935  * Split motion vector prediction, 16.4.
936  * @returns the number of motion vectors parsed (2, 4 or 16)
937  */
938 static av_always_inline
940  int layout, int is_vp7)
941 {
942  int part_idx;
943  int n, num;
944  const VP8Macroblock *top_mb;
945  const VP8Macroblock *left_mb = &mb[-1];
946  const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning];
947  const uint8_t *mbsplits_top, *mbsplits_cur, *firstidx;
948  const VP8mv *top_mv;
949  const VP8mv *left_mv = left_mb->bmv;
950  const VP8mv *cur_mv = mb->bmv;
951 
952  if (!layout) // layout is inlined, s->mb_layout is not
953  top_mb = &mb[2];
954  else
955  top_mb = &mb[-s->mb_width - 1];
956  mbsplits_top = vp8_mbsplits[top_mb->partitioning];
957  top_mv = top_mb->bmv;
958 
962  else
963  part_idx = VP8_SPLITMVMODE_8x8;
964  } else {
965  part_idx = VP8_SPLITMVMODE_4x4;
966  }
967 
968  num = vp8_mbsplit_count[part_idx];
969  mbsplits_cur = vp8_mbsplits[part_idx],
970  firstidx = vp8_mbfirstidx[part_idx];
971  mb->partitioning = part_idx;
972 
973  for (n = 0; n < num; n++) {
974  int k = firstidx[n];
975  uint32_t left, above;
976  const uint8_t *submv_prob;
977 
978  if (!(k & 3))
979  left = AV_RN32A(&left_mv[mbsplits_left[k + 3]]);
980  else
981  left = AV_RN32A(&cur_mv[mbsplits_cur[k - 1]]);
982  if (k <= 3)
983  above = AV_RN32A(&top_mv[mbsplits_top[k + 12]]);
984  else
985  above = AV_RN32A(&cur_mv[mbsplits_cur[k - 4]]);
986 
987  submv_prob = get_submv_prob(left, above, is_vp7);
988 
989  if (vpx_rac_get_prob_branchy(c, submv_prob[0])) {
990  if (vpx_rac_get_prob_branchy(c, submv_prob[1])) {
991  if (vpx_rac_get_prob_branchy(c, submv_prob[2])) {
992  mb->bmv[n].y = mb->mv.y +
993  read_mv_component(c, s->prob->mvc[0], is_vp7);
994  mb->bmv[n].x = mb->mv.x +
995  read_mv_component(c, s->prob->mvc[1], is_vp7);
996  } else {
997  AV_ZERO32(&mb->bmv[n]);
998  }
999  } else {
1000  AV_WN32A(&mb->bmv[n], above);
1001  }
1002  } else {
1003  AV_WN32A(&mb->bmv[n], left);
1004  }
1005  }
1006 
1007  return num;
1008 }
1009 
1010 /**
1011  * The vp7 reference decoder uses a padding macroblock column (added to right
1012  * edge of the frame) to guard against illegal macroblock offsets. The
1013  * algorithm has bugs that permit offsets to straddle the padding column.
1014  * This function replicates those bugs.
1015  *
1016  * @param[out] edge_x macroblock x address
1017  * @param[out] edge_y macroblock y address
1018  *
1019  * @return macroblock offset legal (boolean)
1020  */
1021 static int vp7_calculate_mb_offset(int mb_x, int mb_y, int mb_width,
1022  int xoffset, int yoffset, int boundary,
1023  int *edge_x, int *edge_y)
1024 {
1025  int vwidth = mb_width + 1;
1026  int new = (mb_y + yoffset) * vwidth + mb_x + xoffset;
1027  if (new < boundary || new % vwidth == vwidth - 1)
1028  return 0;
1029  *edge_y = new / vwidth;
1030  *edge_x = new % vwidth;
1031  return 1;
1032 }
1033 
1034 static const VP8mv *get_bmv_ptr(const VP8Macroblock *mb, int subblock)
1035 {
1036  return &mb->bmv[mb->mode == VP8_MVMODE_SPLIT ? vp8_mbsplits[mb->partitioning][subblock] : 0];
1037 }
1038 
1039 static av_always_inline
1041  int mb_x, int mb_y, int layout)
1042 {
1043  enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR };
1044  enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
1045  int idx = CNT_ZERO;
1046  VP8mv near_mv[3];
1047  uint8_t cnt[3] = { 0 };
1048  VPXRangeCoder *c = &s->c;
1049  int i;
1050 
1051  AV_ZERO32(&near_mv[0]);
1052  AV_ZERO32(&near_mv[1]);
1053  AV_ZERO32(&near_mv[2]);
1054 
1055  for (i = 0; i < VP7_MV_PRED_COUNT; i++) {
1056  const VP7MVPred * pred = &vp7_mv_pred[i];
1057  int edge_x, edge_y;
1058 
1059  if (vp7_calculate_mb_offset(mb_x, mb_y, s->mb_width, pred->xoffset,
1060  pred->yoffset, !s->profile, &edge_x, &edge_y)) {
1061  const VP8Macroblock *edge = (s->mb_layout == 1)
1062  ? s->macroblocks_base + 1 + edge_x +
1063  (s->mb_width + 1) * (edge_y + 1)
1064  : s->macroblocks + edge_x +
1065  (s->mb_height - edge_y - 1) * 2;
1066  uint32_t mv = AV_RN32A(get_bmv_ptr(edge, vp7_mv_pred[i].subblock));
1067  if (mv) {
1068  if (AV_RN32A(&near_mv[CNT_NEAREST])) {
1069  if (mv == AV_RN32A(&near_mv[CNT_NEAREST])) {
1070  idx = CNT_NEAREST;
1071  } else if (AV_RN32A(&near_mv[CNT_NEAR])) {
1072  if (mv != AV_RN32A(&near_mv[CNT_NEAR]))
1073  continue;
1074  idx = CNT_NEAR;
1075  } else {
1076  AV_WN32A(&near_mv[CNT_NEAR], mv);
1077  idx = CNT_NEAR;
1078  }
1079  } else {
1080  AV_WN32A(&near_mv[CNT_NEAREST], mv);
1081  idx = CNT_NEAREST;
1082  }
1083  } else {
1084  idx = CNT_ZERO;
1085  }
1086  } else {
1087  idx = CNT_ZERO;
1088  }
1089  cnt[idx] += vp7_mv_pred[i].score;
1090  }
1091 
1092  mb->partitioning = VP8_SPLITMVMODE_NONE;
1093 
1094  if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_ZERO]][0])) {
1095  mb->mode = VP8_MVMODE_MV;
1096 
1097  if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAREST]][1])) {
1098 
1099  if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][2])) {
1100 
1101  if (cnt[CNT_NEAREST] > cnt[CNT_NEAR])
1102  AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAREST] ? 0 : AV_RN32A(&near_mv[CNT_NEAREST]));
1103  else
1104  AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAR] ? 0 : AV_RN32A(&near_mv[CNT_NEAR]));
1105 
1106  if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][3])) {
1107  mb->mode = VP8_MVMODE_SPLIT;
1108  mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP7) - 1];
1109  } else {
1110  mb->mv.y += vp7_read_mv_component(c, s->prob->mvc[0]);
1111  mb->mv.x += vp7_read_mv_component(c, s->prob->mvc[1]);
1112  mb->bmv[0] = mb->mv;
1113  }
1114  } else {
1115  mb->mv = near_mv[CNT_NEAR];
1116  mb->bmv[0] = mb->mv;
1117  }
1118  } else {
1119  mb->mv = near_mv[CNT_NEAREST];
1120  mb->bmv[0] = mb->mv;
1121  }
1122  } else {
1123  mb->mode = VP8_MVMODE_ZERO;
1124  AV_ZERO32(&mb->mv);
1125  mb->bmv[0] = mb->mv;
1126  }
1127 }
1128 
1129 static av_always_inline
1131  int mb_x, int mb_y, int layout)
1132 {
1133  VP8Macroblock *mb_edge[3] = { 0 /* top */,
1134  mb - 1 /* left */,
1135  0 /* top-left */ };
1136  enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV };
1137  enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
1138  int idx = CNT_ZERO;
1139  int cur_sign_bias = s->sign_bias[mb->ref_frame];
1140  const int8_t *sign_bias = s->sign_bias;
1141  VP8mv near_mv[4];
1142  uint8_t cnt[4] = { 0 };
1143  VPXRangeCoder *c = &s->c;
1144 
1145  if (!layout) { // layout is inlined (s->mb_layout is not)
1146  mb_edge[0] = mb + 2;
1147  mb_edge[2] = mb + 1;
1148  } else {
1149  mb_edge[0] = mb - s->mb_width - 1;
1150  mb_edge[2] = mb - s->mb_width - 2;
1151  }
1152 
1153  AV_ZERO32(&near_mv[0]);
1154  AV_ZERO32(&near_mv[1]);
1155  AV_ZERO32(&near_mv[2]);
1156 
1157  /* Process MB on top, left and top-left */
1158 #define MV_EDGE_CHECK(n) \
1159  { \
1160  const VP8Macroblock *edge = mb_edge[n]; \
1161  int edge_ref = edge->ref_frame; \
1162  if (edge_ref != VP8_FRAME_CURRENT) { \
1163  uint32_t mv = AV_RN32A(&edge->mv); \
1164  if (mv) { \
1165  if (cur_sign_bias != sign_bias[edge_ref]) { \
1166  /* SWAR negate of the values in mv. */ \
1167  mv = ~mv; \
1168  mv = ((mv & 0x7fff7fff) + \
1169  0x00010001) ^ (mv & 0x80008000); \
1170  } \
1171  if (!n || mv != AV_RN32A(&near_mv[idx])) \
1172  AV_WN32A(&near_mv[++idx], mv); \
1173  cnt[idx] += 1 + (n != 2); \
1174  } else \
1175  cnt[CNT_ZERO] += 1 + (n != 2); \
1176  } \
1177  }
1178 
1179  MV_EDGE_CHECK(0)
1180  MV_EDGE_CHECK(1)
1181  MV_EDGE_CHECK(2)
1182 
1183  mb->partitioning = VP8_SPLITMVMODE_NONE;
1184  if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_ZERO]][0])) {
1185  mb->mode = VP8_MVMODE_MV;
1186 
1187  /* If we have three distinct MVs, merge first and last if they're the same */
1188  if (cnt[CNT_SPLITMV] &&
1189  AV_RN32A(&near_mv[1 + VP8_EDGE_TOP]) == AV_RN32A(&near_mv[1 + VP8_EDGE_TOPLEFT]))
1190  cnt[CNT_NEAREST] += 1;
1191 
1192  /* Swap near and nearest if necessary */
1193  if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) {
1194  FFSWAP(uint8_t, cnt[CNT_NEAREST], cnt[CNT_NEAR]);
1195  FFSWAP(VP8mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR]);
1196  }
1197 
1198  if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAREST]][1])) {
1199  if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAR]][2])) {
1200  /* Choose the best mv out of 0,0 and the nearest mv */
1201  clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]);
1202  cnt[CNT_SPLITMV] = ((mb_edge[VP8_EDGE_LEFT]->mode == VP8_MVMODE_SPLIT) +
1203  (mb_edge[VP8_EDGE_TOP]->mode == VP8_MVMODE_SPLIT)) * 2 +
1204  (mb_edge[VP8_EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT);
1205 
1206  if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_SPLITMV]][3])) {
1207  mb->mode = VP8_MVMODE_SPLIT;
1208  mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP8) - 1];
1209  } else {
1210  mb->mv.y += vp8_read_mv_component(c, s->prob->mvc[0]);
1211  mb->mv.x += vp8_read_mv_component(c, s->prob->mvc[1]);
1212  mb->bmv[0] = mb->mv;
1213  }
1214  } else {
1215  clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAR]);
1216  mb->bmv[0] = mb->mv;
1217  }
1218  } else {
1219  clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAREST]);
1220  mb->bmv[0] = mb->mv;
1221  }
1222  } else {
1223  mb->mode = VP8_MVMODE_ZERO;
1224  AV_ZERO32(&mb->mv);
1225  mb->bmv[0] = mb->mv;
1226  }
1227 }
1228 
1229 static av_always_inline
1231  int mb_x, int keyframe, int layout)
1232 {
1233  uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
1234 
1235  if (layout) {
1236  VP8Macroblock *mb_top = mb - s->mb_width - 1;
1237  memcpy(mb->intra4x4_pred_mode_top, mb_top->intra4x4_pred_mode_top, 4);
1238  }
1239  if (keyframe) {
1240  int x, y;
1241  uint8_t *top;
1242  uint8_t *const left = s->intra4x4_pred_mode_left;
1243  if (layout)
1244  top = mb->intra4x4_pred_mode_top;
1245  else
1246  top = s->intra4x4_pred_mode_top + 4 * mb_x;
1247  for (y = 0; y < 4; y++) {
1248  for (x = 0; x < 4; x++) {
1249  const uint8_t *ctx;
1250  ctx = vp8_pred4x4_prob_intra[top[x]][left[y]];
1251  *intra4x4 = vp89_rac_get_tree(c, vp8_pred4x4_tree, ctx);
1252  left[y] = top[x] = *intra4x4;
1253  intra4x4++;
1254  }
1255  }
1256  } else {
1257  int i;
1258  for (i = 0; i < 16; i++)
1259  intra4x4[i] = vp89_rac_get_tree(c, vp8_pred4x4_tree,
1261  }
1262 }
1263 
1264 static av_always_inline
1265 void decode_mb_mode(VP8Context *s, const VP8mvbounds *mv_bounds,
1266  VP8Macroblock *mb, int mb_x, int mb_y,
1267  uint8_t *segment, const uint8_t *ref, int layout, int is_vp7)
1268 {
1269  VPXRangeCoder *c = &s->c;
1270  static const char * const vp7_feature_name[] = { "q-index",
1271  "lf-delta",
1272  "partial-golden-update",
1273  "blit-pitch" };
1274  if (is_vp7) {
1275  int i;
1276  *segment = 0;
1277  for (i = 0; i < 4; i++) {
1278  if (s->feature_enabled[i]) {
1279  if (vpx_rac_get_prob_branchy(c, s->feature_present_prob[i])) {
1281  s->feature_index_prob[i]);
1282  av_log(s->avctx, AV_LOG_WARNING,
1283  "Feature %s present in macroblock (value 0x%x)\n",
1284  vp7_feature_name[i], s->feature_value[i][index]);
1285  }
1286  }
1287  }
1288  } else if (s->segmentation.update_map) {
1289  int bit = vpx_rac_get_prob(c, s->prob->segmentid[0]);
1290  *segment = vpx_rac_get_prob(c, s->prob->segmentid[1+bit]) + 2*bit;
1291  } else if (s->segmentation.enabled)
1292  *segment = ref ? *ref : *segment;
1293  mb->segment = *segment;
1294 
1295  mb->skip = s->mbskip_enabled ? vpx_rac_get_prob(c, s->prob->mbskip) : 0;
1296 
1297  if (s->keyframe) {
1300 
1301  if (mb->mode == MODE_I4x4) {
1302  decode_intra4x4_modes(s, c, mb, mb_x, 1, layout);
1303  } else {
1304  const uint32_t modes = (is_vp7 ? vp7_pred4x4_mode
1305  : vp8_pred4x4_mode)[mb->mode] * 0x01010101u;
1306  if (s->mb_layout)
1307  AV_WN32A(mb->intra4x4_pred_mode_top, modes);
1308  else
1309  AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes);
1310  AV_WN32A(s->intra4x4_pred_mode_left, modes);
1311  }
1312 
1313  mb->chroma_pred_mode = vp89_rac_get_tree(c, vp8_pred8x8c_tree,
1315  mb->ref_frame = VP8_FRAME_CURRENT;
1316  } else if (vpx_rac_get_prob_branchy(c, s->prob->intra)) {
1317  // inter MB, 16.2
1318  if (vpx_rac_get_prob_branchy(c, s->prob->last))
1319  mb->ref_frame =
1320  (!is_vp7 && vpx_rac_get_prob(c, s->prob->golden)) ? VP8_FRAME_ALTREF
1321  : VP8_FRAME_GOLDEN;
1322  else
1323  mb->ref_frame = VP8_FRAME_PREVIOUS;
1324  s->ref_count[mb->ref_frame - 1]++;
1325 
1326  // motion vectors, 16.3
1327  if (is_vp7)
1328  vp7_decode_mvs(s, mb, mb_x, mb_y, layout);
1329  else
1330  vp8_decode_mvs(s, mv_bounds, mb, mb_x, mb_y, layout);
1331  } else {
1332  // intra MB, 16.1
1334  s->prob->pred16x16);
1335 
1336  if (mb->mode == MODE_I4x4)
1337  decode_intra4x4_modes(s, c, mb, mb_x, 0, layout);
1338 
1339  mb->chroma_pred_mode = vp89_rac_get_tree(c, vp8_pred8x8c_tree,
1340  s->prob->pred8x8c);
1341  mb->ref_frame = VP8_FRAME_CURRENT;
1342  mb->partitioning = VP8_SPLITMVMODE_NONE;
1343  AV_ZERO32(&mb->bmv[0]);
1344  }
1345 }
1346 
1347 /**
1348  * @param r arithmetic bitstream reader context
1349  * @param block destination for block coefficients
1350  * @param probs probabilities to use when reading trees from the bitstream
1351  * @param i initial coeff index, 0 unless a separate DC block is coded
1352  * @param qmul array holding the dc/ac dequant factor at position 0/1
1353  *
1354  * @return 0 if no coeffs were decoded
1355  * otherwise, the index of the last coeff decoded plus one
1356  */
1357 static av_always_inline
1359  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
1360  int i, const uint8_t *token_prob, const int16_t qmul[2],
1361  const uint8_t scan[16], int vp7)
1362 {
1363  VPXRangeCoder c = *r;
1364  goto skip_eob;
1365  do {
1366  int coeff;
1367 restart:
1368  if (!vpx_rac_get_prob_branchy(&c, token_prob[0])) // DCT_EOB
1369  break;
1370 
1371 skip_eob:
1372  if (!vpx_rac_get_prob_branchy(&c, token_prob[1])) { // DCT_0
1373  if (++i == 16)
1374  break; // invalid input; blocks should end with EOB
1375  token_prob = probs[i][0];
1376  if (vp7)
1377  goto restart;
1378  goto skip_eob;
1379  }
1380 
1381  if (!vpx_rac_get_prob_branchy(&c, token_prob[2])) { // DCT_1
1382  coeff = 1;
1383  token_prob = probs[i + 1][1];
1384  } else {
1385  if (!vpx_rac_get_prob_branchy(&c, token_prob[3])) { // DCT 2,3,4
1386  coeff = vpx_rac_get_prob_branchy(&c, token_prob[4]);
1387  if (coeff)
1388  coeff += vpx_rac_get_prob(&c, token_prob[5]);
1389  coeff += 2;
1390  } else {
1391  // DCT_CAT*
1392  if (!vpx_rac_get_prob_branchy(&c, token_prob[6])) {
1393  if (!vpx_rac_get_prob_branchy(&c, token_prob[7])) { // DCT_CAT1
1395  } else { // DCT_CAT2
1396  coeff = 7;
1397  coeff += vpx_rac_get_prob(&c, vp8_dct_cat2_prob[0]) << 1;
1399  }
1400  } else { // DCT_CAT3 and up
1401  int a = vpx_rac_get_prob(&c, token_prob[8]);
1402  int b = vpx_rac_get_prob(&c, token_prob[9 + a]);
1403  int cat = (a << 1) + b;
1404  coeff = 3 + (8 << cat);
1406  }
1407  }
1408  token_prob = probs[i + 1][2];
1409  }
1410  block[scan[i]] = (vp89_rac_get(&c) ? -coeff : coeff) * qmul[!!i];
1411  } while (++i < 16);
1412 
1413  *r = c;
1414  return i;
1415 }
1416 
1417 static av_always_inline
1418 int inter_predict_dc(int16_t block[16], int16_t pred[2])
1419 {
1420  int16_t dc = block[0];
1421  int ret = 0;
1422 
1423  if (pred[1] > 3) {
1424  dc += pred[0];
1425  ret = 1;
1426  }
1427 
1428  if (!pred[0] | !dc | ((int32_t)pred[0] ^ (int32_t)dc) >> 31) {
1429  block[0] = pred[0] = dc;
1430  pred[1] = 0;
1431  } else {
1432  if (pred[0] == dc)
1433  pred[1]++;
1434  block[0] = pred[0] = dc;
1435  }
1436 
1437  return ret;
1438 }
1439 
1441  int16_t block[16],
1442  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
1443  int i, const uint8_t *token_prob,
1444  const int16_t qmul[2],
1445  const uint8_t scan[16])
1446 {
1447  return decode_block_coeffs_internal(r, block, probs, i,
1448  token_prob, qmul, scan, IS_VP7);
1449 }
1450 
1451 #ifndef vp8_decode_block_coeffs_internal
1453  int16_t block[16],
1454  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
1455  int i, const uint8_t *token_prob,
1456  const int16_t qmul[2])
1457 {
1458  return decode_block_coeffs_internal(r, block, probs, i,
1459  token_prob, qmul, ff_zigzag_scan, IS_VP8);
1460 }
1461 #endif
1462 
1463 /**
1464  * @param c arithmetic bitstream reader context
1465  * @param block destination for block coefficients
1466  * @param probs probabilities to use when reading trees from the bitstream
1467  * @param i initial coeff index, 0 unless a separate DC block is coded
1468  * @param zero_nhood the initial prediction context for number of surrounding
1469  * all-zero blocks (only left/top, so 0-2)
1470  * @param qmul array holding the dc/ac dequant factor at position 0/1
1471  * @param scan scan pattern (VP7 only)
1472  *
1473  * @return 0 if no coeffs were decoded
1474  * otherwise, the index of the last coeff decoded plus one
1475  */
1476 static av_always_inline
1478  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
1479  int i, int zero_nhood, const int16_t qmul[2],
1480  const uint8_t scan[16], int vp7)
1481 {
1482  const uint8_t *token_prob = probs[i][zero_nhood];
1483  if (!vpx_rac_get_prob_branchy(c, token_prob[0])) // DCT_EOB
1484  return 0;
1485  return vp7 ? vp7_decode_block_coeffs_internal(c, block, probs, i,
1486  token_prob, qmul, scan)
1488  token_prob, qmul);
1489 }
1490 
1491 static av_always_inline
1493  VP8Macroblock *mb, uint8_t t_nnz[9], uint8_t l_nnz[9],
1494  int is_vp7)
1495 {
1496  int i, x, y, luma_start = 0, luma_ctx = 3;
1497  int nnz_pred, nnz, nnz_total = 0;
1498  int segment = mb->segment;
1499  int block_dc = 0;
1500 
1501  if (mb->mode != MODE_I4x4 && (is_vp7 || mb->mode != VP8_MVMODE_SPLIT)) {
1502  nnz_pred = t_nnz[8] + l_nnz[8];
1503 
1504  // decode DC values and do hadamard
1505  nnz = decode_block_coeffs(c, td->block_dc, s->prob->token[1], 0,
1506  nnz_pred, s->qmat[segment].luma_dc_qmul,
1507  ff_zigzag_scan, is_vp7);
1508  l_nnz[8] = t_nnz[8] = !!nnz;
1509 
1510  if (is_vp7 && mb->mode > MODE_I4x4) {
1511  nnz |= inter_predict_dc(td->block_dc,
1512  s->inter_dc_pred[mb->ref_frame - 1]);
1513  }
1514 
1515  if (nnz) {
1516  nnz_total += nnz;
1517  block_dc = 1;
1518  if (nnz == 1)
1519  s->vp8dsp.vp8_luma_dc_wht_dc(td->block, td->block_dc);
1520  else
1521  s->vp8dsp.vp8_luma_dc_wht(td->block, td->block_dc);
1522  }
1523  luma_start = 1;
1524  luma_ctx = 0;
1525  }
1526 
1527  // luma blocks
1528  for (y = 0; y < 4; y++)
1529  for (x = 0; x < 4; x++) {
1530  nnz_pred = l_nnz[y] + t_nnz[x];
1531  nnz = decode_block_coeffs(c, td->block[y][x],
1532  s->prob->token[luma_ctx],
1533  luma_start, nnz_pred,
1534  s->qmat[segment].luma_qmul,
1535  s->prob[0].scan, is_vp7);
1536  /* nnz+block_dc may be one more than the actual last index,
1537  * but we don't care */
1538  td->non_zero_count_cache[y][x] = nnz + block_dc;
1539  t_nnz[x] = l_nnz[y] = !!nnz;
1540  nnz_total += nnz;
1541  }
1542 
1543  // chroma blocks
1544  // TODO: what to do about dimensions? 2nd dim for luma is x,
1545  // but for chroma it's (y<<1)|x
1546  for (i = 4; i < 6; i++)
1547  for (y = 0; y < 2; y++)
1548  for (x = 0; x < 2; x++) {
1549  nnz_pred = l_nnz[i + 2 * y] + t_nnz[i + 2 * x];
1550  nnz = decode_block_coeffs(c, td->block[i][(y << 1) + x],
1551  s->prob->token[2], 0, nnz_pred,
1552  s->qmat[segment].chroma_qmul,
1553  s->prob[0].scan, is_vp7);
1554  td->non_zero_count_cache[i][(y << 1) + x] = nnz;
1555  t_nnz[i + 2 * x] = l_nnz[i + 2 * y] = !!nnz;
1556  nnz_total += nnz;
1557  }
1558 
1559  // if there were no coded coeffs despite the macroblock not being marked skip,
1560  // we MUST not do the inner loop filter and should not do IDCT
1561  // Since skip isn't used for bitstream prediction, just manually set it.
1562  if (!nnz_total)
1563  mb->skip = 1;
1564 }
1565 
1566 static av_always_inline
1567 void backup_mb_border(uint8_t *top_border, const uint8_t *src_y,
1568  const uint8_t *src_cb, const uint8_t *src_cr,
1569  ptrdiff_t linesize, ptrdiff_t uvlinesize, int simple)
1570 {
1571  AV_COPY128(top_border, src_y + 15 * linesize);
1572  if (!simple) {
1573  AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1574  AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
1575  }
1576 }
1577 
1578 static av_always_inline
1579 void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb,
1580  uint8_t *src_cr, ptrdiff_t linesize, ptrdiff_t uvlinesize, int mb_x,
1581  int mb_y, int mb_width, int simple, int xchg)
1582 {
1583  uint8_t *top_border_m1 = top_border - 32; // for TL prediction
1584  src_y -= linesize;
1585  src_cb -= uvlinesize;
1586  src_cr -= uvlinesize;
1587 
1588 #define XCHG(a, b, xchg) \
1589  do { \
1590  if (xchg) \
1591  AV_SWAP64(b, a); \
1592  else \
1593  AV_COPY64(b, a); \
1594  } while (0)
1595 
1596  XCHG(top_border_m1 + 8, src_y - 8, xchg);
1597  XCHG(top_border, src_y, xchg);
1598  XCHG(top_border + 8, src_y + 8, 1);
1599  if (mb_x < mb_width - 1)
1600  XCHG(top_border + 32, src_y + 16, 1);
1601 
1602  // only copy chroma for normal loop filter
1603  // or to initialize the top row to 127
1604  if (!simple || !mb_y) {
1605  XCHG(top_border_m1 + 16, src_cb - 8, xchg);
1606  XCHG(top_border_m1 + 24, src_cr - 8, xchg);
1607  XCHG(top_border + 16, src_cb, 1);
1608  XCHG(top_border + 24, src_cr, 1);
1609  }
1610 }
1611 
1612 static av_always_inline
1613 int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y)
1614 {
1615  if (!mb_x)
1616  return mb_y ? TOP_DC_PRED8x8 : DC_128_PRED8x8;
1617  else
1618  return mb_y ? mode : LEFT_DC_PRED8x8;
1619 }
1620 
1621 static av_always_inline
1622 int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y, int vp7)
1623 {
1624  if (!mb_x)
1625  return mb_y ? VERT_PRED8x8 : (vp7 ? DC_128_PRED8x8 : DC_129_PRED8x8);
1626  else
1627  return mb_y ? mode : HOR_PRED8x8;
1628 }
1629 
1630 static av_always_inline
1631 int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y, int vp7)
1632 {
1633  switch (mode) {
1634  case DC_PRED8x8:
1635  return check_dc_pred8x8_mode(mode, mb_x, mb_y);
1636  case VERT_PRED8x8:
1637  return !mb_y ? (vp7 ? DC_128_PRED8x8 : DC_127_PRED8x8) : mode;
1638  case HOR_PRED8x8:
1639  return !mb_x ? (vp7 ? DC_128_PRED8x8 : DC_129_PRED8x8) : mode;
1640  case PLANE_PRED8x8: /* TM */
1641  return check_tm_pred8x8_mode(mode, mb_x, mb_y, vp7);
1642  }
1643  return mode;
1644 }
1645 
1646 static av_always_inline
1647 int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y, int vp7)
1648 {
1649  if (!mb_x) {
1650  return mb_y ? VERT_VP8_PRED : (vp7 ? DC_128_PRED : DC_129_PRED);
1651  } else {
1652  return mb_y ? mode : HOR_VP8_PRED;
1653  }
1654 }
1655 
1656 static av_always_inline
1657 int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y,
1658  int *copy_buf, int vp7)
1659 {
1660  switch (mode) {
1661  case VERT_PRED:
1662  if (!mb_x && mb_y) {
1663  *copy_buf = 1;
1664  return mode;
1665  }
1666  /* fall-through */
1667  case DIAG_DOWN_LEFT_PRED:
1668  case VERT_LEFT_PRED:
1669  return !mb_y ? (vp7 ? DC_128_PRED : DC_127_PRED) : mode;
1670  case HOR_PRED:
1671  if (!mb_y) {
1672  *copy_buf = 1;
1673  return mode;
1674  }
1675  /* fall-through */
1676  case HOR_UP_PRED:
1677  return !mb_x ? (vp7 ? DC_128_PRED : DC_129_PRED) : mode;
1678  case TM_VP8_PRED:
1679  return check_tm_pred4x4_mode(mode, mb_x, mb_y, vp7);
1680  case DC_PRED: /* 4x4 DC doesn't use the same "H.264-style" exceptions
1681  * as 16x16/8x8 DC */
1682  case DIAG_DOWN_RIGHT_PRED:
1683  case VERT_RIGHT_PRED:
1684  case HOR_DOWN_PRED:
1685  if (!mb_y || !mb_x)
1686  *copy_buf = 1;
1687  return mode;
1688  }
1689  return mode;
1690 }
1691 
1692 static av_always_inline
1693 void intra_predict(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3],
1694  VP8Macroblock *mb, int mb_x, int mb_y, int is_vp7)
1695 {
1696  int x, y, mode, nnz;
1697  uint32_t tr;
1698 
1699  /* for the first row, we need to run xchg_mb_border to init the top edge
1700  * to 127 otherwise, skip it if we aren't going to deblock */
1701  if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0)
1702  xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2],
1703  s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
1704  s->filter.simple, 1);
1705 
1706  if (mb->mode < MODE_I4x4) {
1707  mode = check_intra_pred8x8_mode_emuedge(mb->mode, mb_x, mb_y, is_vp7);
1708  s->hpc.pred16x16[mode](dst[0], s->linesize);
1709  } else {
1710  uint8_t *ptr = dst[0];
1711  const uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
1712  const uint8_t lo = is_vp7 ? 128 : 127;
1713  const uint8_t hi = is_vp7 ? 128 : 129;
1714  const uint8_t tr_top[4] = { lo, lo, lo, lo };
1715 
1716  // all blocks on the right edge of the macroblock use bottom edge
1717  // the top macroblock for their topright edge
1718  const uint8_t *tr_right = ptr - s->linesize + 16;
1719 
1720  // if we're on the right edge of the frame, said edge is extended
1721  // from the top macroblock
1722  if (mb_y && mb_x == s->mb_width - 1) {
1723  tr = tr_right[-1] * 0x01010101u;
1724  tr_right = (uint8_t *) &tr;
1725  }
1726 
1727  if (mb->skip)
1728  AV_ZERO128(td->non_zero_count_cache);
1729 
1730  for (y = 0; y < 4; y++) {
1731  const uint8_t *topright = ptr + 4 - s->linesize;
1732  for (x = 0; x < 4; x++) {
1733  int copy = 0;
1734  ptrdiff_t linesize = s->linesize;
1735  uint8_t *dst = ptr + 4 * x;
1736  LOCAL_ALIGNED(4, uint8_t, copy_dst, [5 * 8]);
1737 
1738  if ((y == 0 || x == 3) && mb_y == 0) {
1739  topright = tr_top;
1740  } else if (x == 3)
1741  topright = tr_right;
1742 
1743  mode = check_intra_pred4x4_mode_emuedge(intra4x4[x], mb_x + x,
1744  mb_y + y, &copy, is_vp7);
1745  if (copy) {
1746  dst = copy_dst + 12;
1747  linesize = 8;
1748  if (!(mb_y + y)) {
1749  copy_dst[3] = lo;
1750  AV_WN32A(copy_dst + 4, lo * 0x01010101U);
1751  } else {
1752  AV_COPY32(copy_dst + 4, ptr + 4 * x - s->linesize);
1753  if (!(mb_x + x)) {
1754  copy_dst[3] = hi;
1755  } else {
1756  copy_dst[3] = ptr[4 * x - s->linesize - 1];
1757  }
1758  }
1759  if (!(mb_x + x)) {
1760  copy_dst[11] =
1761  copy_dst[19] =
1762  copy_dst[27] =
1763  copy_dst[35] = hi;
1764  } else {
1765  copy_dst[11] = ptr[4 * x - 1];
1766  copy_dst[19] = ptr[4 * x + s->linesize - 1];
1767  copy_dst[27] = ptr[4 * x + s->linesize * 2 - 1];
1768  copy_dst[35] = ptr[4 * x + s->linesize * 3 - 1];
1769  }
1770  }
1771  s->hpc.pred4x4[mode](dst, topright, linesize);
1772  if (copy) {
1773  AV_COPY32(ptr + 4 * x, copy_dst + 12);
1774  AV_COPY32(ptr + 4 * x + s->linesize, copy_dst + 20);
1775  AV_COPY32(ptr + 4 * x + s->linesize * 2, copy_dst + 28);
1776  AV_COPY32(ptr + 4 * x + s->linesize * 3, copy_dst + 36);
1777  }
1778 
1779  nnz = td->non_zero_count_cache[y][x];
1780  if (nnz) {
1781  if (nnz == 1)
1782  s->vp8dsp.vp8_idct_dc_add(ptr + 4 * x,
1783  td->block[y][x], s->linesize);
1784  else
1785  s->vp8dsp.vp8_idct_add(ptr + 4 * x,
1786  td->block[y][x], s->linesize);
1787  }
1788  topright += 4;
1789  }
1790 
1791  ptr += 4 * s->linesize;
1792  intra4x4 += 4;
1793  }
1794  }
1795 
1796  mode = check_intra_pred8x8_mode_emuedge(mb->chroma_pred_mode,
1797  mb_x, mb_y, is_vp7);
1798  s->hpc.pred8x8[mode](dst[1], s->uvlinesize);
1799  s->hpc.pred8x8[mode](dst[2], s->uvlinesize);
1800 
1801  if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0)
1802  xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2],
1803  s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
1804  s->filter.simple, 0);
1805 }
1806 
1807 static const uint8_t subpel_idx[3][8] = {
1808  { 0, 1, 2, 1, 2, 1, 2, 1 }, // nr. of left extra pixels,
1809  // also function pointer index
1810  { 0, 3, 5, 3, 5, 3, 5, 3 }, // nr. of extra pixels required
1811  { 0, 2, 3, 2, 3, 2, 3, 2 }, // nr. of right extra pixels
1812 };
1813 
1814 /**
1815  * luma MC function
1816  *
1817  * @param s VP8 decoding context
1818  * @param dst target buffer for block data at block position
1819  * @param ref reference picture buffer at origin (0, 0)
1820  * @param mv motion vector (relative to block position) to get pixel data from
1821  * @param x_off horizontal position of block from origin (0, 0)
1822  * @param y_off vertical position of block from origin (0, 0)
1823  * @param block_w width of block (16, 8 or 4)
1824  * @param block_h height of block (always same as block_w)
1825  * @param width width of src/dst plane data
1826  * @param height height of src/dst plane data
1827  * @param linesize size of a single line of plane data, including padding
1828  * @param mc_func motion compensation function pointers (bilinear or sixtap MC)
1829  */
1830 static av_always_inline
1831 void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst,
1832  const ThreadFrame *ref, const VP8mv *mv,
1833  int x_off, int y_off, int block_w, int block_h,
1834  int width, int height, ptrdiff_t linesize,
1835  vp8_mc_func mc_func[3][3])
1836 {
1837  const uint8_t *src = ref->f->data[0];
1838 
1839  if (AV_RN32A(mv)) {
1840  ptrdiff_t src_linesize = linesize;
1841 
1842  int mx = (mv->x * 2) & 7, mx_idx = subpel_idx[0][mx];
1843  int my = (mv->y * 2) & 7, my_idx = subpel_idx[0][my];
1844 
1845  x_off += mv->x >> 2;
1846  y_off += mv->y >> 2;
1847 
1848  // edge emulation
1849  ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 4, 0);
1850  src += y_off * linesize + x_off;
1851  if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] ||
1852  y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
1853  s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
1854  src - my_idx * linesize - mx_idx,
1855  EDGE_EMU_LINESIZE, linesize,
1856  block_w + subpel_idx[1][mx],
1857  block_h + subpel_idx[1][my],
1858  x_off - mx_idx, y_off - my_idx,
1859  width, height);
1860  src = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
1861  src_linesize = EDGE_EMU_LINESIZE;
1862  }
1863  mc_func[my_idx][mx_idx](dst, linesize, src, src_linesize, block_h, mx, my);
1864  } else {
1865  ff_thread_await_progress(ref, (3 + y_off + block_h) >> 4, 0);
1866  mc_func[0][0](dst, linesize, src + y_off * linesize + x_off,
1867  linesize, block_h, 0, 0);
1868  }
1869 }
1870 
1871 /**
1872  * chroma MC function
1873  *
1874  * @param s VP8 decoding context
1875  * @param dst1 target buffer for block data at block position (U plane)
1876  * @param dst2 target buffer for block data at block position (V plane)
1877  * @param ref reference picture buffer at origin (0, 0)
1878  * @param mv motion vector (relative to block position) to get pixel data from
1879  * @param x_off horizontal position of block from origin (0, 0)
1880  * @param y_off vertical position of block from origin (0, 0)
1881  * @param block_w width of block (16, 8 or 4)
1882  * @param block_h height of block (always same as block_w)
1883  * @param width width of src/dst plane data
1884  * @param height height of src/dst plane data
1885  * @param linesize size of a single line of plane data, including padding
1886  * @param mc_func motion compensation function pointers (bilinear or sixtap MC)
1887  */
1888 static av_always_inline
1889 void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1,
1890  uint8_t *dst2, const ThreadFrame *ref, const VP8mv *mv,
1891  int x_off, int y_off, int block_w, int block_h,
1892  int width, int height, ptrdiff_t linesize,
1893  vp8_mc_func mc_func[3][3])
1894 {
1895  const uint8_t *src1 = ref->f->data[1], *src2 = ref->f->data[2];
1896 
1897  if (AV_RN32A(mv)) {
1898  int mx = mv->x & 7, mx_idx = subpel_idx[0][mx];
1899  int my = mv->y & 7, my_idx = subpel_idx[0][my];
1900 
1901  x_off += mv->x >> 3;
1902  y_off += mv->y >> 3;
1903 
1904  // edge emulation
1905  src1 += y_off * linesize + x_off;
1906  src2 += y_off * linesize + x_off;
1907  ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3, 0);
1908  if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] ||
1909  y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
1910  s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
1911  src1 - my_idx * linesize - mx_idx,
1912  EDGE_EMU_LINESIZE, linesize,
1913  block_w + subpel_idx[1][mx],
1914  block_h + subpel_idx[1][my],
1915  x_off - mx_idx, y_off - my_idx, width, height);
1916  src1 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
1917  mc_func[my_idx][mx_idx](dst1, linesize, src1, EDGE_EMU_LINESIZE, block_h, mx, my);
1918 
1919  s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
1920  src2 - my_idx * linesize - mx_idx,
1921  EDGE_EMU_LINESIZE, linesize,
1922  block_w + subpel_idx[1][mx],
1923  block_h + subpel_idx[1][my],
1924  x_off - mx_idx, y_off - my_idx, width, height);
1925  src2 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
1926  mc_func[my_idx][mx_idx](dst2, linesize, src2, EDGE_EMU_LINESIZE, block_h, mx, my);
1927  } else {
1928  mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my);
1929  mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my);
1930  }
1931  } else {
1932  ff_thread_await_progress(ref, (3 + y_off + block_h) >> 3, 0);
1933  mc_func[0][0](dst1, linesize, src1 + y_off * linesize + x_off, linesize, block_h, 0, 0);
1934  mc_func[0][0](dst2, linesize, src2 + y_off * linesize + x_off, linesize, block_h, 0, 0);
1935  }
1936 }
1937 
1938 static av_always_inline
1939 void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3],
1940  const ThreadFrame *ref_frame, int x_off, int y_off,
1941  int bx_off, int by_off, int block_w, int block_h,
1942  int width, int height, const VP8mv *mv)
1943 {
1944  VP8mv uvmv = *mv;
1945 
1946  /* Y */
1947  vp8_mc_luma(s, td, dst[0] + by_off * s->linesize + bx_off,
1948  ref_frame, mv, x_off + bx_off, y_off + by_off,
1949  block_w, block_h, width, height, s->linesize,
1950  s->put_pixels_tab[block_w == 8]);
1951 
1952  /* U/V */
1953  if (s->profile == 3) {
1954  /* this block only applies VP8; it is safe to check
1955  * only the profile, as VP7 profile <= 1 */
1956  uvmv.x &= ~7;
1957  uvmv.y &= ~7;
1958  }
1959  x_off >>= 1;
1960  y_off >>= 1;
1961  bx_off >>= 1;
1962  by_off >>= 1;
1963  width >>= 1;
1964  height >>= 1;
1965  block_w >>= 1;
1966  block_h >>= 1;
1967  vp8_mc_chroma(s, td, dst[1] + by_off * s->uvlinesize + bx_off,
1968  dst[2] + by_off * s->uvlinesize + bx_off, ref_frame,
1969  &uvmv, x_off + bx_off, y_off + by_off,
1970  block_w, block_h, width, height, s->uvlinesize,
1971  s->put_pixels_tab[1 + (block_w == 4)]);
1972 }
1973 
1974 /* Fetch pixels for estimated mv 4 macroblocks ahead.
1975  * Optimized for 64-byte cache lines. Inspired by ffh264 prefetch_motion. */
1976 static av_always_inline
1978  int mb_x, int mb_y, int mb_xy, int ref)
1979 {
1980  /* Don't prefetch refs that haven't been used very often this frame. */
1981  if (s->ref_count[ref - 1] > (mb_xy >> 5)) {
1982  int x_off = mb_x << 4, y_off = mb_y << 4;
1983  int mx = (mb->mv.x >> 2) + x_off + 8;
1984  int my = (mb->mv.y >> 2) + y_off;
1985  uint8_t **src = s->framep[ref]->tf.f->data;
1986  int off = mx + (my + (mb_x & 3) * 4) * s->linesize + 64;
1987  /* For threading, a ff_thread_await_progress here might be useful, but
1988  * it actually slows down the decoder. Since a bad prefetch doesn't
1989  * generate bad decoder output, we don't run it here. */
1990  s->vdsp.prefetch(src[0] + off, s->linesize, 4);
1991  off = (mx >> 1) + ((my >> 1) + (mb_x & 7)) * s->uvlinesize + 64;
1992  s->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
1993  }
1994 }
1995 
1996 /**
1997  * Apply motion vectors to prediction buffer, chapter 18.
1998  */
1999 static av_always_inline
2000 void inter_predict(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3],
2001  VP8Macroblock *mb, int mb_x, int mb_y)
2002 {
2003  int x_off = mb_x << 4, y_off = mb_y << 4;
2004  int width = 16 * s->mb_width, height = 16 * s->mb_height;
2005  const ThreadFrame *ref = &s->framep[mb->ref_frame]->tf;
2006  const VP8mv *bmv = mb->bmv;
2007 
2008  switch (mb->partitioning) {
2009  case VP8_SPLITMVMODE_NONE:
2010  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2011  0, 0, 16, 16, width, height, &mb->mv);
2012  break;
2013  case VP8_SPLITMVMODE_4x4: {
2014  int x, y;
2015  VP8mv uvmv;
2016 
2017  /* Y */
2018  for (y = 0; y < 4; y++) {
2019  for (x = 0; x < 4; x++) {
2020  vp8_mc_luma(s, td, dst[0] + 4 * y * s->linesize + x * 4,
2021  ref, &bmv[4 * y + x],
2022  4 * x + x_off, 4 * y + y_off, 4, 4,
2023  width, height, s->linesize,
2024  s->put_pixels_tab[2]);
2025  }
2026  }
2027 
2028  /* U/V */
2029  x_off >>= 1;
2030  y_off >>= 1;
2031  width >>= 1;
2032  height >>= 1;
2033  for (y = 0; y < 2; y++) {
2034  for (x = 0; x < 2; x++) {
2035  uvmv.x = mb->bmv[2 * y * 4 + 2 * x ].x +
2036  mb->bmv[2 * y * 4 + 2 * x + 1].x +
2037  mb->bmv[(2 * y + 1) * 4 + 2 * x ].x +
2038  mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].x;
2039  uvmv.y = mb->bmv[2 * y * 4 + 2 * x ].y +
2040  mb->bmv[2 * y * 4 + 2 * x + 1].y +
2041  mb->bmv[(2 * y + 1) * 4 + 2 * x ].y +
2042  mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].y;
2043  uvmv.x = (uvmv.x + 2 + FF_SIGNBIT(uvmv.x)) >> 2;
2044  uvmv.y = (uvmv.y + 2 + FF_SIGNBIT(uvmv.y)) >> 2;
2045  if (s->profile == 3) {
2046  uvmv.x &= ~7;
2047  uvmv.y &= ~7;
2048  }
2049  vp8_mc_chroma(s, td, dst[1] + 4 * y * s->uvlinesize + x * 4,
2050  dst[2] + 4 * y * s->uvlinesize + x * 4, ref,
2051  &uvmv, 4 * x + x_off, 4 * y + y_off, 4, 4,
2052  width, height, s->uvlinesize,
2053  s->put_pixels_tab[2]);
2054  }
2055  }
2056  break;
2057  }
2058  case VP8_SPLITMVMODE_16x8:
2059  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2060  0, 0, 16, 8, width, height, &bmv[0]);
2061  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2062  0, 8, 16, 8, width, height, &bmv[1]);
2063  break;
2064  case VP8_SPLITMVMODE_8x16:
2065  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2066  0, 0, 8, 16, width, height, &bmv[0]);
2067  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2068  8, 0, 8, 16, width, height, &bmv[1]);
2069  break;
2070  case VP8_SPLITMVMODE_8x8:
2071  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2072  0, 0, 8, 8, width, height, &bmv[0]);
2073  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2074  8, 0, 8, 8, width, height, &bmv[1]);
2075  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2076  0, 8, 8, 8, width, height, &bmv[2]);
2077  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2078  8, 8, 8, 8, width, height, &bmv[3]);
2079  break;
2080  }
2081 }
2082 
2083 static av_always_inline
2084 void idct_mb(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3],
2085  const VP8Macroblock *mb)
2086 {
2087  int x, y, ch;
2088 
2089  if (mb->mode != MODE_I4x4) {
2090  uint8_t *y_dst = dst[0];
2091  for (y = 0; y < 4; y++) {
2092  uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[y]);
2093  if (nnz4) {
2094  if (nnz4 & ~0x01010101) {
2095  for (x = 0; x < 4; x++) {
2096  if ((uint8_t) nnz4 == 1)
2097  s->vp8dsp.vp8_idct_dc_add(y_dst + 4 * x,
2098  td->block[y][x],
2099  s->linesize);
2100  else if ((uint8_t) nnz4 > 1)
2101  s->vp8dsp.vp8_idct_add(y_dst + 4 * x,
2102  td->block[y][x],
2103  s->linesize);
2104  nnz4 >>= 8;
2105  if (!nnz4)
2106  break;
2107  }
2108  } else {
2109  s->vp8dsp.vp8_idct_dc_add4y(y_dst, td->block[y], s->linesize);
2110  }
2111  }
2112  y_dst += 4 * s->linesize;
2113  }
2114  }
2115 
2116  for (ch = 0; ch < 2; ch++) {
2117  uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[4 + ch]);
2118  if (nnz4) {
2119  uint8_t *ch_dst = dst[1 + ch];
2120  if (nnz4 & ~0x01010101) {
2121  for (y = 0; y < 2; y++) {
2122  for (x = 0; x < 2; x++) {
2123  if ((uint8_t) nnz4 == 1)
2124  s->vp8dsp.vp8_idct_dc_add(ch_dst + 4 * x,
2125  td->block[4 + ch][(y << 1) + x],
2126  s->uvlinesize);
2127  else if ((uint8_t) nnz4 > 1)
2128  s->vp8dsp.vp8_idct_add(ch_dst + 4 * x,
2129  td->block[4 + ch][(y << 1) + x],
2130  s->uvlinesize);
2131  nnz4 >>= 8;
2132  if (!nnz4)
2133  goto chroma_idct_end;
2134  }
2135  ch_dst += 4 * s->uvlinesize;
2136  }
2137  } else {
2138  s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, td->block[4 + ch], s->uvlinesize);
2139  }
2140  }
2141 chroma_idct_end:
2142  ;
2143  }
2144 }
2145 
2146 static av_always_inline
2148  VP8FilterStrength *f, int is_vp7)
2149 {
2150  int interior_limit, filter_level;
2151 
2152  if (s->segmentation.enabled) {
2153  filter_level = s->segmentation.filter_level[mb->segment];
2154  if (!s->segmentation.absolute_vals)
2155  filter_level += s->filter.level;
2156  } else
2157  filter_level = s->filter.level;
2158 
2159  if (s->lf_delta.enabled) {
2160  filter_level += s->lf_delta.ref[mb->ref_frame];
2161  filter_level += s->lf_delta.mode[mb->mode];
2162  }
2163 
2164  filter_level = av_clip_uintp2(filter_level, 6);
2165 
2166  interior_limit = filter_level;
2167  if (s->filter.sharpness) {
2168  interior_limit >>= (s->filter.sharpness + 3) >> 2;
2169  interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness);
2170  }
2171  interior_limit = FFMAX(interior_limit, 1);
2172 
2173  f->filter_level = filter_level;
2174  f->inner_limit = interior_limit;
2175  f->inner_filter = is_vp7 || !mb->skip || mb->mode == MODE_I4x4 ||
2176  mb->mode == VP8_MVMODE_SPLIT;
2177 }
2178 
2179 static av_always_inline
2180 void filter_mb(const VP8Context *s, uint8_t *const dst[3], const VP8FilterStrength *f,
2181  int mb_x, int mb_y, int is_vp7)
2182 {
2183  int mbedge_lim, bedge_lim_y, bedge_lim_uv, hev_thresh;
2184  int filter_level = f->filter_level;
2185  int inner_limit = f->inner_limit;
2186  int inner_filter = f->inner_filter;
2187  ptrdiff_t linesize = s->linesize;
2188  ptrdiff_t uvlinesize = s->uvlinesize;
2189  static const uint8_t hev_thresh_lut[2][64] = {
2190  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
2191  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2192  3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
2193  3, 3, 3, 3 },
2194  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
2195  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
2196  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2197  2, 2, 2, 2 }
2198  };
2199 
2200  if (!filter_level)
2201  return;
2202 
2203  if (is_vp7) {
2204  bedge_lim_y = filter_level;
2205  bedge_lim_uv = filter_level * 2;
2206  mbedge_lim = filter_level + 2;
2207  } else {
2208  bedge_lim_y =
2209  bedge_lim_uv = filter_level * 2 + inner_limit;
2210  mbedge_lim = bedge_lim_y + 4;
2211  }
2212 
2213  hev_thresh = hev_thresh_lut[s->keyframe][filter_level];
2214 
2215  if (mb_x) {
2216  s->vp8dsp.vp8_h_loop_filter16y(dst[0], linesize,
2217  mbedge_lim, inner_limit, hev_thresh);
2218  s->vp8dsp.vp8_h_loop_filter8uv(dst[1], dst[2], uvlinesize,
2219  mbedge_lim, inner_limit, hev_thresh);
2220  }
2221 
2222 #define H_LOOP_FILTER_16Y_INNER(cond) \
2223  if (cond && inner_filter) { \
2224  s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 4, linesize, \
2225  bedge_lim_y, inner_limit, \
2226  hev_thresh); \
2227  s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 8, linesize, \
2228  bedge_lim_y, inner_limit, \
2229  hev_thresh); \
2230  s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 12, linesize, \
2231  bedge_lim_y, inner_limit, \
2232  hev_thresh); \
2233  s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4, \
2234  uvlinesize, bedge_lim_uv, \
2235  inner_limit, hev_thresh); \
2236  }
2237 
2238  H_LOOP_FILTER_16Y_INNER(!is_vp7)
2239 
2240  if (mb_y) {
2241  s->vp8dsp.vp8_v_loop_filter16y(dst[0], linesize,
2242  mbedge_lim, inner_limit, hev_thresh);
2243  s->vp8dsp.vp8_v_loop_filter8uv(dst[1], dst[2], uvlinesize,
2244  mbedge_lim, inner_limit, hev_thresh);
2245  }
2246 
2247  if (inner_filter) {
2248  s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 4 * linesize,
2249  linesize, bedge_lim_y,
2250  inner_limit, hev_thresh);
2251  s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 8 * linesize,
2252  linesize, bedge_lim_y,
2253  inner_limit, hev_thresh);
2254  s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 12 * linesize,
2255  linesize, bedge_lim_y,
2256  inner_limit, hev_thresh);
2257  s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] + 4 * uvlinesize,
2258  dst[2] + 4 * uvlinesize,
2259  uvlinesize, bedge_lim_uv,
2260  inner_limit, hev_thresh);
2261  }
2262 
2263  H_LOOP_FILTER_16Y_INNER(is_vp7)
2264 }
2265 
2266 static av_always_inline
2267 void filter_mb_simple(const VP8Context *s, uint8_t *dst, const VP8FilterStrength *f,
2268  int mb_x, int mb_y)
2269 {
2270  int mbedge_lim, bedge_lim;
2271  int filter_level = f->filter_level;
2272  int inner_limit = f->inner_limit;
2273  int inner_filter = f->inner_filter;
2274  ptrdiff_t linesize = s->linesize;
2275 
2276  if (!filter_level)
2277  return;
2278 
2279  bedge_lim = 2 * filter_level + inner_limit;
2280  mbedge_lim = bedge_lim + 4;
2281 
2282  if (mb_x)
2283  s->vp8dsp.vp8_h_loop_filter_simple(dst, linesize, mbedge_lim);
2284  if (inner_filter) {
2285  s->vp8dsp.vp8_h_loop_filter_simple(dst + 4, linesize, bedge_lim);
2286  s->vp8dsp.vp8_h_loop_filter_simple(dst + 8, linesize, bedge_lim);
2287  s->vp8dsp.vp8_h_loop_filter_simple(dst + 12, linesize, bedge_lim);
2288  }
2289 
2290  if (mb_y)
2291  s->vp8dsp.vp8_v_loop_filter_simple(dst, linesize, mbedge_lim);
2292  if (inner_filter) {
2293  s->vp8dsp.vp8_v_loop_filter_simple(dst + 4 * linesize, linesize, bedge_lim);
2294  s->vp8dsp.vp8_v_loop_filter_simple(dst + 8 * linesize, linesize, bedge_lim);
2295  s->vp8dsp.vp8_v_loop_filter_simple(dst + 12 * linesize, linesize, bedge_lim);
2296  }
2297 }
2298 
2299 #define MARGIN (16 << 2)
2300 static av_always_inline
2302  const VP8Frame *prev_frame, int is_vp7)
2303 {
2304  VP8Context *s = avctx->priv_data;
2305  int mb_x, mb_y;
2306 
2307  s->mv_bounds.mv_min.y = -MARGIN;
2308  s->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
2309  for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
2310  VP8Macroblock *mb = s->macroblocks_base +
2311  ((s->mb_width + 1) * (mb_y + 1) + 1);
2312  int mb_xy = mb_y * s->mb_width;
2313 
2314  AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101);
2315 
2316  s->mv_bounds.mv_min.x = -MARGIN;
2317  s->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
2318 
2319  for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
2320  if (vpx_rac_is_end(&s->c)) {
2321  return AVERROR_INVALIDDATA;
2322  }
2323  if (mb_y == 0)
2324  AV_WN32A((mb - s->mb_width - 1)->intra4x4_pred_mode_top,
2325  DC_PRED * 0x01010101);
2326  decode_mb_mode(s, &s->mv_bounds, mb, mb_x, mb_y, curframe->seg_map + mb_xy,
2327  prev_frame && prev_frame->seg_map ?
2328  prev_frame->seg_map + mb_xy : NULL, 1, is_vp7);
2329  s->mv_bounds.mv_min.x -= 64;
2330  s->mv_bounds.mv_max.x -= 64;
2331  }
2332  s->mv_bounds.mv_min.y -= 64;
2333  s->mv_bounds.mv_max.y -= 64;
2334  }
2335  return 0;
2336 }
2337 
2338 static int vp7_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame,
2339  const VP8Frame *prev_frame)
2340 {
2341  return vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP7);
2342 }
2343 
2344 static int vp8_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame,
2345  const VP8Frame *prev_frame)
2346 {
2347  return vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP8);
2348 }
2349 
2350 #if HAVE_THREADS
2351 #define check_thread_pos(td, otd, mb_x_check, mb_y_check) \
2352  do { \
2353  int tmp = (mb_y_check << 16) | (mb_x_check & 0xFFFF); \
2354  if (atomic_load(&otd->thread_mb_pos) < tmp) { \
2355  pthread_mutex_lock(&otd->lock); \
2356  atomic_store(&td->wait_mb_pos, tmp); \
2357  do { \
2358  if (atomic_load(&otd->thread_mb_pos) >= tmp) \
2359  break; \
2360  pthread_cond_wait(&otd->cond, &otd->lock); \
2361  } while (1); \
2362  atomic_store(&td->wait_mb_pos, INT_MAX); \
2363  pthread_mutex_unlock(&otd->lock); \
2364  } \
2365  } while (0)
2366 
2367 #define update_pos(td, mb_y, mb_x) \
2368  do { \
2369  int pos = (mb_y << 16) | (mb_x & 0xFFFF); \
2370  int sliced_threading = (avctx->active_thread_type == FF_THREAD_SLICE) && \
2371  (num_jobs > 1); \
2372  int is_null = !next_td || !prev_td; \
2373  int pos_check = (is_null) ? 1 : \
2374  (next_td != td && pos >= atomic_load(&next_td->wait_mb_pos)) || \
2375  (prev_td != td && pos >= atomic_load(&prev_td->wait_mb_pos)); \
2376  atomic_store(&td->thread_mb_pos, pos); \
2377  if (sliced_threading && pos_check) { \
2378  pthread_mutex_lock(&td->lock); \
2379  pthread_cond_broadcast(&td->cond); \
2380  pthread_mutex_unlock(&td->lock); \
2381  } \
2382  } while (0)
2383 #else
2384 #define check_thread_pos(td, otd, mb_x_check, mb_y_check) while(0)
2385 #define update_pos(td, mb_y, mb_x) while(0)
2386 #endif
2387 
2389  int jobnr, int threadnr, int is_vp7)
2390 {
2391  VP8Context *s = avctx->priv_data;
2392  VP8ThreadData *prev_td, *next_td, *td = &s->thread_data[threadnr];
2393  int mb_y = atomic_load(&td->thread_mb_pos) >> 16;
2394  int mb_x, mb_xy = mb_y * s->mb_width;
2395  int num_jobs = s->num_jobs;
2396  const VP8Frame *prev_frame = s->prev_frame;
2397  VP8Frame *curframe = s->curframe;
2398  VPXRangeCoder *coeff_c = &s->coeff_partition[mb_y & (s->num_coeff_partitions - 1)];
2399 
2400  VP8Macroblock *mb;
2401  uint8_t *dst[3] = {
2402  curframe->tf.f->data[0] + 16 * mb_y * s->linesize,
2403  curframe->tf.f->data[1] + 8 * mb_y * s->uvlinesize,
2404  curframe->tf.f->data[2] + 8 * mb_y * s->uvlinesize
2405  };
2406 
2407  if (vpx_rac_is_end(&s->c))
2408  return AVERROR_INVALIDDATA;
2409 
2410  if (mb_y == 0)
2411  prev_td = td;
2412  else
2413  prev_td = &s->thread_data[(jobnr + num_jobs - 1) % num_jobs];
2414  if (mb_y == s->mb_height - 1)
2415  next_td = td;
2416  else
2417  next_td = &s->thread_data[(jobnr + 1) % num_jobs];
2418  if (s->mb_layout == 1)
2419  mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1);
2420  else {
2421  // Make sure the previous frame has read its segmentation map,
2422  // if we re-use the same map.
2423  if (prev_frame && s->segmentation.enabled &&
2424  !s->segmentation.update_map)
2425  ff_thread_await_progress(&prev_frame->tf, mb_y, 0);
2426  mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2;
2427  memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock
2428  AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101);
2429  }
2430 
2431  if (!is_vp7 || mb_y == 0)
2432  memset(td->left_nnz, 0, sizeof(td->left_nnz));
2433 
2434  td->mv_bounds.mv_min.x = -MARGIN;
2435  td->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
2436 
2437  for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
2438  if (vpx_rac_is_end(&s->c))
2439  return AVERROR_INVALIDDATA;
2440  // Wait for previous thread to read mb_x+2, and reach mb_y-1.
2441  if (prev_td != td) {
2442  if (threadnr != 0) {
2443  check_thread_pos(td, prev_td,
2444  mb_x + (is_vp7 ? 2 : 1),
2445  mb_y - (is_vp7 ? 2 : 1));
2446  } else {
2447  check_thread_pos(td, prev_td,
2448  mb_x + (is_vp7 ? 2 : 1) + s->mb_width + 3,
2449  mb_y - (is_vp7 ? 2 : 1));
2450  }
2451  }
2452 
2453  s->vdsp.prefetch(dst[0] + (mb_x & 3) * 4 * s->linesize + 64,
2454  s->linesize, 4);
2455  s->vdsp.prefetch(dst[1] + (mb_x & 7) * s->uvlinesize + 64,
2456  dst[2] - dst[1], 2);
2457 
2458  if (!s->mb_layout)
2459  decode_mb_mode(s, &td->mv_bounds, mb, mb_x, mb_y, curframe->seg_map + mb_xy,
2460  prev_frame && prev_frame->seg_map ?
2461  prev_frame->seg_map + mb_xy : NULL, 0, is_vp7);
2462 
2463  prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_PREVIOUS);
2464 
2465  if (!mb->skip) {
2466  if (vpx_rac_is_end(coeff_c))
2467  return AVERROR_INVALIDDATA;
2468  decode_mb_coeffs(s, td, coeff_c, mb, s->top_nnz[mb_x], td->left_nnz, is_vp7);
2469  }
2470 
2471  if (mb->mode <= MODE_I4x4)
2472  intra_predict(s, td, dst, mb, mb_x, mb_y, is_vp7);
2473  else
2474  inter_predict(s, td, dst, mb, mb_x, mb_y);
2475 
2476  prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_GOLDEN);
2477 
2478  if (!mb->skip) {
2479  idct_mb(s, td, dst, mb);
2480  } else {
2481  AV_ZERO64(td->left_nnz);
2482  AV_WN64(s->top_nnz[mb_x], 0); // array of 9, so unaligned
2483 
2484  /* Reset DC block predictors if they would exist
2485  * if the mb had coefficients */
2486  if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
2487  td->left_nnz[8] = 0;
2488  s->top_nnz[mb_x][8] = 0;
2489  }
2490  }
2491 
2492  if (s->deblock_filter)
2493  filter_level_for_mb(s, mb, &td->filter_strength[mb_x], is_vp7);
2494 
2495  if (s->deblock_filter && num_jobs != 1 && threadnr == num_jobs - 1) {
2496  if (s->filter.simple)
2497  backup_mb_border(s->top_border[mb_x + 1], dst[0],
2498  NULL, NULL, s->linesize, 0, 1);
2499  else
2500  backup_mb_border(s->top_border[mb_x + 1], dst[0],
2501  dst[1], dst[2], s->linesize, s->uvlinesize, 0);
2502  }
2503 
2504  prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_ALTREF);
2505 
2506  dst[0] += 16;
2507  dst[1] += 8;
2508  dst[2] += 8;
2509  td->mv_bounds.mv_min.x -= 64;
2510  td->mv_bounds.mv_max.x -= 64;
2511 
2512  if (mb_x == s->mb_width + 1) {
2513  update_pos(td, mb_y, s->mb_width + 3);
2514  } else {
2515  update_pos(td, mb_y, mb_x);
2516  }
2517  }
2518  return 0;
2519 }
2520 
2521 static int vp7_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
2522  int jobnr, int threadnr)
2523 {
2524  return decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 1);
2525 }
2526 
2527 static int vp8_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
2528  int jobnr, int threadnr)
2529 {
2530  return decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 0);
2531 }
2532 
2533 static av_always_inline void filter_mb_row(AVCodecContext *avctx, void *tdata,
2534  int jobnr, int threadnr, int is_vp7)
2535 {
2536  VP8Context *s = avctx->priv_data;
2537  VP8ThreadData *td = &s->thread_data[threadnr];
2538  int mb_x, mb_y = atomic_load(&td->thread_mb_pos) >> 16, num_jobs = s->num_jobs;
2539  AVFrame *curframe = s->curframe->tf.f;
2540  VP8Macroblock *mb;
2541  VP8ThreadData *prev_td, *next_td;
2542  uint8_t *dst[3] = {
2543  curframe->data[0] + 16 * mb_y * s->linesize,
2544  curframe->data[1] + 8 * mb_y * s->uvlinesize,
2545  curframe->data[2] + 8 * mb_y * s->uvlinesize
2546  };
2547 
2548  if (s->mb_layout == 1)
2549  mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1);
2550  else
2551  mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2;
2552 
2553  if (mb_y == 0)
2554  prev_td = td;
2555  else
2556  prev_td = &s->thread_data[(jobnr + num_jobs - 1) % num_jobs];
2557  if (mb_y == s->mb_height - 1)
2558  next_td = td;
2559  else
2560  next_td = &s->thread_data[(jobnr + 1) % num_jobs];
2561 
2562  for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb++) {
2563  const VP8FilterStrength *f = &td->filter_strength[mb_x];
2564  if (prev_td != td)
2565  check_thread_pos(td, prev_td,
2566  (mb_x + 1) + (s->mb_width + 3), mb_y - 1);
2567  if (next_td != td)
2568  if (next_td != &s->thread_data[0])
2569  check_thread_pos(td, next_td, mb_x + 1, mb_y + 1);
2570 
2571  if (num_jobs == 1) {
2572  if (s->filter.simple)
2573  backup_mb_border(s->top_border[mb_x + 1], dst[0],
2574  NULL, NULL, s->linesize, 0, 1);
2575  else
2576  backup_mb_border(s->top_border[mb_x + 1], dst[0],
2577  dst[1], dst[2], s->linesize, s->uvlinesize, 0);
2578  }
2579 
2580  if (s->filter.simple)
2581  filter_mb_simple(s, dst[0], f, mb_x, mb_y);
2582  else
2583  filter_mb(s, dst, f, mb_x, mb_y, is_vp7);
2584  dst[0] += 16;
2585  dst[1] += 8;
2586  dst[2] += 8;
2587 
2588  update_pos(td, mb_y, (s->mb_width + 3) + mb_x);
2589  }
2590 }
2591 
2592 static void vp7_filter_mb_row(AVCodecContext *avctx, void *tdata,
2593  int jobnr, int threadnr)
2594 {
2595  filter_mb_row(avctx, tdata, jobnr, threadnr, 1);
2596 }
2597 
2598 static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata,
2599  int jobnr, int threadnr)
2600 {
2601  filter_mb_row(avctx, tdata, jobnr, threadnr, 0);
2602 }
2603 
2604 static av_always_inline
2605 int vp78_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr,
2606  int threadnr, int is_vp7)
2607 {
2608  const VP8Context *s = avctx->priv_data;
2609  VP8ThreadData *td = &s->thread_data[jobnr];
2610  VP8ThreadData *next_td = NULL, *prev_td = NULL;
2611  VP8Frame *curframe = s->curframe;
2612  int mb_y, num_jobs = s->num_jobs;
2613  int ret;
2614 
2615  td->thread_nr = threadnr;
2616  td->mv_bounds.mv_min.y = -MARGIN - 64 * threadnr;
2617  td->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN - 64 * threadnr;
2618  for (mb_y = jobnr; mb_y < s->mb_height; mb_y += num_jobs) {
2619  atomic_store(&td->thread_mb_pos, mb_y << 16);
2620  ret = s->decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr);
2621  if (ret < 0) {
2622  update_pos(td, s->mb_height, INT_MAX & 0xFFFF);
2623  return ret;
2624  }
2625  if (s->deblock_filter)
2626  s->filter_mb_row(avctx, tdata, jobnr, threadnr);
2627  update_pos(td, mb_y, INT_MAX & 0xFFFF);
2628 
2629  td->mv_bounds.mv_min.y -= 64 * num_jobs;
2630  td->mv_bounds.mv_max.y -= 64 * num_jobs;
2631 
2632  if (avctx->active_thread_type == FF_THREAD_FRAME)
2633  ff_thread_report_progress(&curframe->tf, mb_y, 0);
2634  }
2635 
2636  return 0;
2637 }
2638 
2639 static int vp7_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata,
2640  int jobnr, int threadnr)
2641 {
2642  return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP7);
2643 }
2644 
2645 static int vp8_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata,
2646  int jobnr, int threadnr)
2647 {
2648  return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP8);
2649 }
2650 
2651 static av_always_inline
2652 int vp78_decode_frame(AVCodecContext *avctx, AVFrame *rframe, int *got_frame,
2653  const AVPacket *avpkt, int is_vp7)
2654 {
2655  VP8Context *s = avctx->priv_data;
2656  int ret, i, referenced, num_jobs;
2657  enum AVDiscard skip_thresh;
2658  VP8Frame *av_uninit(curframe), *prev_frame;
2659 
2660  if (is_vp7)
2661  ret = vp7_decode_frame_header(s, avpkt->data, avpkt->size);
2662  else
2663  ret = vp8_decode_frame_header(s, avpkt->data, avpkt->size);
2664 
2665  if (ret < 0)
2666  goto err;
2667 
2668  if (s->actually_webp) {
2669  // avctx->pix_fmt already set in caller.
2670  } else if (!is_vp7 && s->pix_fmt == AV_PIX_FMT_NONE) {
2671  s->pix_fmt = get_pixel_format(s);
2672  if (s->pix_fmt < 0) {
2673  ret = AVERROR(EINVAL);
2674  goto err;
2675  }
2676  avctx->pix_fmt = s->pix_fmt;
2677  }
2678 
2679  prev_frame = s->framep[VP8_FRAME_CURRENT];
2680 
2681  referenced = s->update_last || s->update_golden == VP8_FRAME_CURRENT ||
2682  s->update_altref == VP8_FRAME_CURRENT;
2683 
2684  skip_thresh = !referenced ? AVDISCARD_NONREF
2685  : !s->keyframe ? AVDISCARD_NONKEY
2686  : AVDISCARD_ALL;
2687 
2688  if (avctx->skip_frame >= skip_thresh) {
2689  s->invisible = 1;
2690  memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4);
2691  goto skip_decode;
2692  }
2693  s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh;
2694 
2695  // release no longer referenced frames
2696  for (i = 0; i < 5; i++)
2697  if (s->frames[i].tf.f->buf[0] &&
2698  &s->frames[i] != prev_frame &&
2699  &s->frames[i] != s->framep[VP8_FRAME_PREVIOUS] &&
2700  &s->frames[i] != s->framep[VP8_FRAME_GOLDEN] &&
2701  &s->frames[i] != s->framep[VP8_FRAME_ALTREF])
2702  vp8_release_frame(&s->frames[i]);
2703 
2704  curframe = s->framep[VP8_FRAME_CURRENT] = vp8_find_free_buffer(s);
2705 
2706  if (!s->colorspace)
2707  avctx->colorspace = AVCOL_SPC_BT470BG;
2708  if (s->fullrange)
2709  avctx->color_range = AVCOL_RANGE_JPEG;
2710  else
2711  avctx->color_range = AVCOL_RANGE_MPEG;
2712 
2713  /* Given that arithmetic probabilities are updated every frame, it's quite
2714  * likely that the values we have on a random interframe are complete
2715  * junk if we didn't start decode on a keyframe. So just don't display
2716  * anything rather than junk. */
2717  if (!s->keyframe && (!s->framep[VP8_FRAME_PREVIOUS] ||
2718  !s->framep[VP8_FRAME_GOLDEN] ||
2719  !s->framep[VP8_FRAME_ALTREF])) {
2720  av_log(avctx, AV_LOG_WARNING,
2721  "Discarding interframe without a prior keyframe!\n");
2723  goto err;
2724  }
2725 
2726  if (s->keyframe)
2727  curframe->tf.f->flags |= AV_FRAME_FLAG_KEY;
2728  else
2729  curframe->tf.f->flags &= ~AV_FRAME_FLAG_KEY;
2730  curframe->tf.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2732  if ((ret = vp8_alloc_frame(s, curframe, referenced)) < 0)
2733  goto err;
2734 
2735  // check if golden and altref are swapped
2736  if (s->update_altref != VP8_FRAME_NONE)
2737  s->next_framep[VP8_FRAME_ALTREF] = s->framep[s->update_altref];
2738  else
2739  s->next_framep[VP8_FRAME_ALTREF] = s->framep[VP8_FRAME_ALTREF];
2740 
2741  if (s->update_golden != VP8_FRAME_NONE)
2742  s->next_framep[VP8_FRAME_GOLDEN] = s->framep[s->update_golden];
2743  else
2744  s->next_framep[VP8_FRAME_GOLDEN] = s->framep[VP8_FRAME_GOLDEN];
2745 
2746  if (s->update_last)
2747  s->next_framep[VP8_FRAME_PREVIOUS] = curframe;
2748  else
2749  s->next_framep[VP8_FRAME_PREVIOUS] = s->framep[VP8_FRAME_PREVIOUS];
2750 
2751  s->next_framep[VP8_FRAME_CURRENT] = curframe;
2752 
2753  if (ffcodec(avctx->codec)->update_thread_context)
2754  ff_thread_finish_setup(avctx);
2755 
2756  if (avctx->hwaccel) {
2757  const FFHWAccel *hwaccel = ffhwaccel(avctx->hwaccel);
2758  ret = hwaccel->start_frame(avctx, avpkt->data, avpkt->size);
2759  if (ret < 0)
2760  goto err;
2761 
2762  ret = hwaccel->decode_slice(avctx, avpkt->data, avpkt->size);
2763  if (ret < 0)
2764  goto err;
2765 
2766  ret = hwaccel->end_frame(avctx);
2767  if (ret < 0)
2768  goto err;
2769 
2770  } else {
2771  s->linesize = curframe->tf.f->linesize[0];
2772  s->uvlinesize = curframe->tf.f->linesize[1];
2773 
2774  memset(s->top_nnz, 0, s->mb_width * sizeof(*s->top_nnz));
2775  /* Zero macroblock structures for top/top-left prediction
2776  * from outside the frame. */
2777  if (!s->mb_layout)
2778  memset(s->macroblocks + s->mb_height * 2 - 1, 0,
2779  (s->mb_width + 1) * sizeof(*s->macroblocks));
2780  if (!s->mb_layout && s->keyframe)
2781  memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width * 4);
2782 
2783  memset(s->ref_count, 0, sizeof(s->ref_count));
2784 
2785  if (s->mb_layout == 1) {
2786  // Make sure the previous frame has read its segmentation map,
2787  // if we re-use the same map.
2788  if (prev_frame && s->segmentation.enabled &&
2789  !s->segmentation.update_map)
2790  ff_thread_await_progress(&prev_frame->tf, 1, 0);
2791  if (is_vp7)
2792  ret = vp7_decode_mv_mb_modes(avctx, curframe, prev_frame);
2793  else
2794  ret = vp8_decode_mv_mb_modes(avctx, curframe, prev_frame);
2795  if (ret < 0)
2796  goto err;
2797  }
2798 
2799  if (avctx->active_thread_type == FF_THREAD_FRAME)
2800  num_jobs = 1;
2801  else
2802  num_jobs = FFMIN(s->num_coeff_partitions, avctx->thread_count);
2803  s->num_jobs = num_jobs;
2804  s->curframe = curframe;
2805  s->prev_frame = prev_frame;
2806  s->mv_bounds.mv_min.y = -MARGIN;
2807  s->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
2808  for (i = 0; i < MAX_THREADS; i++) {
2809  VP8ThreadData *td = &s->thread_data[i];
2810  atomic_init(&td->thread_mb_pos, 0);
2811  atomic_init(&td->wait_mb_pos, INT_MAX);
2812  }
2813  if (is_vp7)
2814  avctx->execute2(avctx, vp7_decode_mb_row_sliced, s->thread_data, NULL,
2815  num_jobs);
2816  else
2817  avctx->execute2(avctx, vp8_decode_mb_row_sliced, s->thread_data, NULL,
2818  num_jobs);
2819  }
2820 
2821  ff_thread_report_progress(&curframe->tf, INT_MAX, 0);
2822  memcpy(&s->framep[0], &s->next_framep[0], sizeof(s->framep[0]) * 4);
2823 
2824 skip_decode:
2825  // if future frames don't use the updated probabilities,
2826  // reset them to the values we saved
2827  if (!s->update_probabilities)
2828  s->prob[0] = s->prob[1];
2829 
2830  if (!s->invisible) {
2831  if ((ret = av_frame_ref(rframe, curframe->tf.f)) < 0)
2832  return ret;
2833  *got_frame = 1;
2834  }
2835 
2836  return avpkt->size;
2837 err:
2838  memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4);
2839  return ret;
2840 }
2841 
2843  int *got_frame, AVPacket *avpkt)
2844 {
2845  return vp78_decode_frame(avctx, frame, got_frame, avpkt, IS_VP8);
2846 }
2847 
2848 #if CONFIG_VP7_DECODER
2849 static int vp7_decode_frame(AVCodecContext *avctx, AVFrame *frame,
2850  int *got_frame, AVPacket *avpkt)
2851 {
2852  return vp78_decode_frame(avctx, frame, got_frame, avpkt, IS_VP7);
2853 }
2854 #endif /* CONFIG_VP7_DECODER */
2855 
2857 {
2858  VP8Context *s = avctx->priv_data;
2859  int i;
2860 
2861  vp8_decode_flush_impl(avctx, 1);
2862  for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++)
2863  av_frame_free(&s->frames[i].tf.f);
2864 
2865  return 0;
2866 }
2867 
2869 {
2870  int i;
2871  for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++) {
2872  s->frames[i].tf.f = av_frame_alloc();
2873  if (!s->frames[i].tf.f)
2874  return AVERROR(ENOMEM);
2875  }
2876  return 0;
2877 }
2878 
2879 static av_always_inline
2880 int vp78_decode_init(AVCodecContext *avctx, int is_vp7)
2881 {
2882  VP8Context *s = avctx->priv_data;
2883  int ret;
2884 
2885  s->avctx = avctx;
2886  s->vp7 = avctx->codec->id == AV_CODEC_ID_VP7;
2887  s->pix_fmt = AV_PIX_FMT_NONE;
2888  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2889 
2890  ff_videodsp_init(&s->vdsp, 8);
2891 
2892  ff_vp78dsp_init(&s->vp8dsp);
2893  if (CONFIG_VP7_DECODER && is_vp7) {
2894  ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP7, 8, 1);
2895  ff_vp7dsp_init(&s->vp8dsp);
2896  s->decode_mb_row_no_filter = vp7_decode_mb_row_no_filter;
2897  s->filter_mb_row = vp7_filter_mb_row;
2898  } else if (CONFIG_VP8_DECODER && !is_vp7) {
2899  ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP8, 8, 1);
2900  ff_vp8dsp_init(&s->vp8dsp);
2901  s->decode_mb_row_no_filter = vp8_decode_mb_row_no_filter;
2902  s->filter_mb_row = vp8_filter_mb_row;
2903  }
2904 
2905  /* does not change for VP8 */
2906  memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan));
2907 
2908  if ((ret = vp8_init_frames(s)) < 0) {
2909  ff_vp8_decode_free(avctx);
2910  return ret;
2911  }
2912 
2913  return 0;
2914 }
2915 
2916 #if CONFIG_VP7_DECODER
2917 static int vp7_decode_init(AVCodecContext *avctx)
2918 {
2919  return vp78_decode_init(avctx, IS_VP7);
2920 }
2921 #endif /* CONFIG_VP7_DECODER */
2922 
2924 {
2925  return vp78_decode_init(avctx, IS_VP8);
2926 }
2927 
2928 #if CONFIG_VP8_DECODER
2929 #if HAVE_THREADS
2930 #define REBASE(pic) ((pic) ? (pic) - &s_src->frames[0] + &s->frames[0] : NULL)
2931 
2932 static int vp8_decode_update_thread_context(AVCodecContext *dst,
2933  const AVCodecContext *src)
2934 {
2935  VP8Context *s = dst->priv_data, *s_src = src->priv_data;
2936  int i;
2937 
2938  if (s->macroblocks_base &&
2939  (s_src->mb_width != s->mb_width || s_src->mb_height != s->mb_height)) {
2940  free_buffers(s);
2941  s->mb_width = s_src->mb_width;
2942  s->mb_height = s_src->mb_height;
2943  }
2944 
2945  s->pix_fmt = s_src->pix_fmt;
2946  s->prob[0] = s_src->prob[!s_src->update_probabilities];
2947  s->segmentation = s_src->segmentation;
2948  s->lf_delta = s_src->lf_delta;
2949  memcpy(s->sign_bias, s_src->sign_bias, sizeof(s->sign_bias));
2950 
2951  for (i = 0; i < FF_ARRAY_ELEMS(s_src->frames); i++) {
2952  if (s_src->frames[i].tf.f->buf[0]) {
2953  int ret = vp8_ref_frame(&s->frames[i], &s_src->frames[i]);
2954  if (ret < 0)
2955  return ret;
2956  }
2957  }
2958 
2959  s->framep[0] = REBASE(s_src->next_framep[0]);
2960  s->framep[1] = REBASE(s_src->next_framep[1]);
2961  s->framep[2] = REBASE(s_src->next_framep[2]);
2962  s->framep[3] = REBASE(s_src->next_framep[3]);
2963 
2964  return 0;
2965 }
2966 #endif /* HAVE_THREADS */
2967 #endif /* CONFIG_VP8_DECODER */
2968 
2969 #if CONFIG_VP7_DECODER
2970 const FFCodec ff_vp7_decoder = {
2971  .p.name = "vp7",
2972  CODEC_LONG_NAME("On2 VP7"),
2973  .p.type = AVMEDIA_TYPE_VIDEO,
2974  .p.id = AV_CODEC_ID_VP7,
2975  .priv_data_size = sizeof(VP8Context),
2976  .init = vp7_decode_init,
2977  .close = ff_vp8_decode_free,
2978  FF_CODEC_DECODE_CB(vp7_decode_frame),
2979  .p.capabilities = AV_CODEC_CAP_DR1,
2980  .flush = vp8_decode_flush,
2981 };
2982 #endif /* CONFIG_VP7_DECODER */
2983 
2984 #if CONFIG_VP8_DECODER
2985 const FFCodec ff_vp8_decoder = {
2986  .p.name = "vp8",
2987  CODEC_LONG_NAME("On2 VP8"),
2988  .p.type = AVMEDIA_TYPE_VIDEO,
2989  .p.id = AV_CODEC_ID_VP8,
2990  .priv_data_size = sizeof(VP8Context),
2992  .close = ff_vp8_decode_free,
2994  .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS |
2996  .caps_internal = FF_CODEC_CAP_ALLOCATE_PROGRESS,
2997  .flush = vp8_decode_flush,
2998  UPDATE_THREAD_CONTEXT(vp8_decode_update_thread_context),
2999  .hw_configs = (const AVCodecHWConfigInternal *const []) {
3000 #if CONFIG_VP8_VAAPI_HWACCEL
3001  HWACCEL_VAAPI(vp8),
3002 #endif
3003 #if CONFIG_VP8_NVDEC_HWACCEL
3004  HWACCEL_NVDEC(vp8),
3005 #endif
3006  NULL
3007  },
3008 };
3009 #endif /* CONFIG_VP7_DECODER */
vp8_mode_contexts
static const int vp8_mode_contexts[6][4]
Definition: vp8data.h:118
hwconfig.h
vp8_dct_cat1_prob
static const uint8_t vp8_dct_cat1_prob[]
Definition: vp8data.h:336
FFCodec::update_thread_context
int(* update_thread_context)(struct AVCodecContext *dst, const struct AVCodecContext *src)
Copy necessary context variables from a previous thread context to the current one.
Definition: codec_internal.h:157
decode_mb_mode
static av_always_inline void decode_mb_mode(VP8Context *s, const VP8mvbounds *mv_bounds, VP8Macroblock *mb, int mb_x, int mb_y, uint8_t *segment, const uint8_t *ref, int layout, int is_vp7)
Definition: vp8.c:1265
VP7_MV_PRED_COUNT
#define VP7_MV_PRED_COUNT
Definition: vp8data.h:68
AVCodecContext::hwaccel
const struct AVHWAccel * hwaccel
Hardware accelerator in use.
Definition: avcodec.h:1435
ff_vp8_decode_free
av_cold int ff_vp8_decode_free(AVCodecContext *avctx)
Definition: vp8.c:2856
vp7_pred4x4_mode
static const uint8_t vp7_pred4x4_mode[]
Definition: vp8data.h:33
HOR_PRED8x8
#define HOR_PRED8x8
Definition: h264pred.h:69
decode_block_coeffs_internal
static av_always_inline int decode_block_coeffs_internal(VPXRangeCoder *r, int16_t block[16], uint8_t probs[16][3][NUM_DCT_TOKENS - 1], int i, const uint8_t *token_prob, const int16_t qmul[2], const uint8_t scan[16], int vp7)
Definition: vp8.c:1358
vp8_release_frame
static void vp8_release_frame(VP8Frame *f)
Definition: vp8.c:123
vp7_mv_pred
static const VP7MVPred vp7_mv_pred[VP7_MV_PRED_COUNT]
Definition: vp8data.h:69
AV_LOG_WARNING
#define AV_LOG_WARNING
Something somehow does not look correct.
Definition: log.h:186
AV_PIX_FMT_CUDA
@ AV_PIX_FMT_CUDA
HW acceleration through CUDA.
Definition: pixfmt.h:253
td
#define td
Definition: regdef.h:70
AVPixelFormat
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
vp8_decode_block_coeffs_internal
static int vp8_decode_block_coeffs_internal(VPXRangeCoder *r, int16_t block[16], uint8_t probs[16][3][NUM_DCT_TOKENS - 1], int i, const uint8_t *token_prob, const int16_t qmul[2])
Definition: vp8.c:1452
vp7_read_mv_component
static int vp7_read_mv_component(VPXRangeCoder *c, const uint8_t *p)
Definition: vp8.c:911
vp7_calculate_mb_offset
static int vp7_calculate_mb_offset(int mb_x, int mb_y, int mb_width, int xoffset, int yoffset, int boundary, int *edge_x, int *edge_y)
The vp7 reference decoder uses a padding macroblock column (added to right edge of the frame) to guar...
Definition: vp8.c:1021
av_clip
#define av_clip
Definition: common.h:96
atomic_store
#define atomic_store(object, desired)
Definition: stdatomic.h:85
backup_mb_border
static av_always_inline void backup_mb_border(uint8_t *top_border, const uint8_t *src_y, const uint8_t *src_cb, const uint8_t *src_cr, ptrdiff_t linesize, ptrdiff_t uvlinesize, int simple)
Definition: vp8.c:1567
VP8Macroblock::partitioning
uint8_t partitioning
Definition: vp8.h:101
VP8_FRAME_CURRENT
@ VP8_FRAME_CURRENT
Definition: vp8.h:44
r
const char * r
Definition: vf_curves.c:126
vp7_filter_mb_row
static void vp7_filter_mb_row(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2592
AVERROR
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later. That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another. Frame references ownership and permissions
DC_PRED8x8
#define DC_PRED8x8
Definition: h264pred.h:68
IS_VP7
#define IS_VP7
Definition: vp8dsp.h:99
AVCodecContext::colorspace
enum AVColorSpace colorspace
YUV colorspace type.
Definition: avcodec.h:1029
vp78_decode_init
static av_always_inline int vp78_decode_init(AVCodecContext *avctx, int is_vp7)
Definition: vp8.c:2880
mem_internal.h
DC_128_PRED
@ DC_128_PRED
Definition: vp9.h:58
ff_get_format
int ff_get_format(AVCodecContext *avctx, const enum AVPixelFormat *fmt)
Select the (possibly hardware accelerated) pixel format.
Definition: decode.c:1264
u
#define u(width, name, range_min, range_max)
Definition: cbs_h2645.c:250
check_tm_pred8x8_mode
static av_always_inline int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y, int vp7)
Definition: vp8.c:1622
vp8_submv_prob
static const uint8_t vp8_submv_prob[5][3]
Definition: vp8data.h:153
av_clip_uintp2
#define av_clip_uintp2
Definition: common.h:120
pthread_mutex_init
static av_always_inline int pthread_mutex_init(pthread_mutex_t *mutex, const pthread_mutexattr_t *attr)
Definition: os2threads.h:104
vp7_ydc_qlookup
static const uint16_t vp7_ydc_qlookup[]
Definition: vp8data.h:754
src1
const pixel * src1
Definition: h264pred_template.c:421
HOR_VP8_PRED
#define HOR_VP8_PRED
unaveraged version of HOR_PRED, see
Definition: h264pred.h:63
mv
static const int8_t mv[256][2]
Definition: 4xm.c:80
vp7_decode_mvs
static av_always_inline void vp7_decode_mvs(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, int layout)
Definition: vp8.c:1040
vp7_mv_default_prob
static const uint8_t vp7_mv_default_prob[2][17]
Definition: vp8data.h:720
check_intra_pred4x4_mode_emuedge
static av_always_inline int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y, int *copy_buf, int vp7)
Definition: vp8.c:1657
av_frame_free
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:100
check_tm_pred4x4_mode
static av_always_inline int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y, int vp7)
Definition: vp8.c:1647
vp7_y2dc_qlookup
static const uint16_t vp7_y2dc_qlookup[]
Definition: vp8data.h:779
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:340
TM_VP8_PRED
@ TM_VP8_PRED
Definition: vp9.h:55
AVCOL_RANGE_JPEG
@ AVCOL_RANGE_JPEG
Full range content.
Definition: pixfmt.h:673
vp8_mc_chroma
static av_always_inline void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1, uint8_t *dst2, const ThreadFrame *ref, const VP8mv *mv, int x_off, int y_off, int block_w, int block_h, int width, int height, ptrdiff_t linesize, vp8_mc_func mc_func[3][3])
chroma MC function
Definition: vp8.c:1889
AVPacket::data
uint8_t * data
Definition: packet.h:491
inter_predict_dc
static av_always_inline int inter_predict_dc(int16_t block[16], int16_t pred[2])
Definition: vp8.c:1418
DC_PRED
@ DC_PRED
Definition: vp9.h:48
b
#define b
Definition: input.c:41
VP7_MVC_SIZE
#define VP7_MVC_SIZE
Definition: vp8.c:468
VERT_LEFT_PRED
@ VERT_LEFT_PRED
Definition: vp9.h:53
vp8_get_quants
static void vp8_get_quants(VP8Context *s)
Definition: vp8.c:381
FFCodec
Definition: codec_internal.h:127
AV_WN32A
#define AV_WN32A(p, v)
Definition: intreadwrite.h:536
FF_HW_SIMPLE_CALL
#define FF_HW_SIMPLE_CALL(avctx, function)
Definition: hwaccel_internal.h:174
cat
#define cat(a, bpp, b)
Definition: vp9dsp_init.h:32
FFMAX
#define FFMAX(a, b)
Definition: macros.h:47
VP8mvbounds
Definition: vp8.h:115
vp89_rac.h
inter_predict
static av_always_inline void inter_predict(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], VP8Macroblock *mb, int mb_x, int mb_y)
Apply motion vectors to prediction buffer, chapter 18.
Definition: vp8.c:2000
VP8_SPLITMVMODE_4x4
@ VP8_SPLITMVMODE_4x4
4x4 blocks of 4x4px each
Definition: vp8.h:80
AVFrame::buf
AVBufferRef * buf[AV_NUM_DATA_POINTERS]
AVBuffer references backing the data for this frame.
Definition: frame.h:590
ff_set_dimensions
int ff_set_dimensions(AVCodecContext *s, int width, int height)
Check that the provided frame dimensions are valid and set them on the codec context.
Definition: utils.c:94
VP8_FRAME_ALTREF
@ VP8_FRAME_ALTREF
Definition: vp8.h:47
VERT_VP8_PRED
#define VERT_VP8_PRED
for VP8, VERT_PRED is the average of
Definition: h264pred.h:60
VPXRangeCoder
Definition: vpx_rac.h:35
thread.h
ff_thread_await_progress
the pkt_dts and pkt_pts fields in AVFrame will work as usual Restrictions on codec whose streams don t reset across will not work because their bitstreams cannot be decoded in parallel *The contents of buffers must not be read before ff_thread_await_progress() has been called on them. reget_buffer() and buffer age optimizations no longer work. *The contents of buffers must not be written to after ff_thread_report_progress() has been called on them. This includes draw_edges(). Porting codecs to frame threading
ThreadFrame::f
AVFrame * f
Definition: threadframe.h:28
AVFrame::data
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:361
VP8_MVC_SIZE
#define VP8_MVC_SIZE
Definition: vp8.c:469
vp8_decode_mb_row_no_filter
static int vp8_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2527
vp8_rac_get_sint
static int vp8_rac_get_sint(VPXRangeCoder *c, int bits)
Definition: vp8.c:50
vp8_pred8x8c_tree
static const int8_t vp8_pred8x8c_tree[3][2]
Definition: vp8data.h:180
XCHG
#define XCHG(a, b, xchg)
bit
#define bit(string, value)
Definition: cbs_mpeg2.c:56
update_pos
#define update_pos(td, mb_y, mb_x)
Definition: vp8.c:2385
vp8.h
get_bmv_ptr
static const VP8mv * get_bmv_ptr(const VP8Macroblock *mb, int subblock)
Definition: vp8.c:1034
update_dimensions
static av_always_inline int update_dimensions(VP8Context *s, int width, int height, int is_vp7)
Definition: vp8.c:209
AVCOL_SPC_BT470BG
@ AVCOL_SPC_BT470BG
also ITU-R BT601-6 625 / ITU-R BT1358 625 / ITU-R BT1700 625 PAL & SECAM / IEC 61966-2-4 xvYCC601
Definition: pixfmt.h:605
VP8_SPLITMVMODE_8x8
@ VP8_SPLITMVMODE_8x8
2x2 blocks of 8x8px each
Definition: vp8.h:79
FFCodec::p
AVCodec p
The public AVCodec.
Definition: codec_internal.h:131
FFHWAccel
Definition: hwaccel_internal.h:34
AVCodecContext::codec
const struct AVCodec * codec
Definition: avcodec.h:450
DC_127_PRED
@ DC_127_PRED
Definition: vp9.h:59
vp8_mv_update_prob
static const uint8_t vp8_mv_update_prob[2][19]
Definition: vp8data.h:709
AVCodecContext::skip_frame
enum AVDiscard skip_frame
Skip decoding for selected frames.
Definition: avcodec.h:1766
fail
#define fail()
Definition: checkasm.h:138
VERT_PRED
@ VERT_PRED
Definition: vp9.h:46
AVCodecContext::thread_count
int thread_count
thread count is used to decide how many independent tasks should be passed to execute()
Definition: avcodec.h:1532
ff_vp8_decoder
const FFCodec ff_vp8_decoder
VP8mv::y
int16_t y
Definition: vp8.h:86
DIAG_DOWN_RIGHT_PRED
@ DIAG_DOWN_RIGHT_PRED
Definition: vp9.h:50
FFHWAccel::end_frame
int(* end_frame)(AVCodecContext *avctx)
Called at the end of each frame or field picture.
Definition: hwaccel_internal.h:97
MAX_THREADS
#define MAX_THREADS
Definition: frame_thread_encoder.c:34
ff_videodsp_init
av_cold void ff_videodsp_init(VideoDSPContext *ctx, int bpc)
Definition: videodsp.c:39
update
static av_always_inline void update(SilenceDetectContext *s, AVFrame *insamples, int is_silence, int current_sample, int64_t nb_samples_notify, AVRational time_base)
Definition: af_silencedetect.c:77
VP8Macroblock::bmv
VP8mv bmv[16]
Definition: vp8.h:107
idct_mb
static av_always_inline void idct_mb(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], const VP8Macroblock *mb)
Definition: vp8.c:2084
check_intra_pred8x8_mode_emuedge
static av_always_inline int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y, int vp7)
Definition: vp8.c:1631
filter_level_for_mb
static av_always_inline void filter_level_for_mb(const VP8Context *s, const VP8Macroblock *mb, VP8FilterStrength *f, int is_vp7)
Definition: vp8.c:2147
read_mv_component
static av_always_inline int read_mv_component(VPXRangeCoder *c, const uint8_t *p, int vp7)
Motion vector coding, 17.1.
Definition: vp8.c:883
vp8_filter_mb_row
static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2598
vp7_get_quants
static void vp7_get_quants(VP8Context *s)
Definition: vp8.c:362
refstruct.h
VP8_SPLITMVMODE_16x8
@ VP8_SPLITMVMODE_16x8
2 16x8 blocks (vertical)
Definition: vp8.h:77
vp8_init_frames
static av_cold int vp8_init_frames(VP8Context *s)
Definition: vp8.c:2868
av_frame_alloc
AVFrame * av_frame_alloc(void)
Allocate an AVFrame and set its fields to default values.
Definition: frame.c:88
vp8_mc_part
static av_always_inline void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], const ThreadFrame *ref_frame, int x_off, int y_off, int bx_off, int by_off, int block_w, int block_h, int width, int height, const VP8mv *mv)
Definition: vp8.c:1939
FFHWAccel::start_frame
int(* start_frame)(AVCodecContext *avctx, const uint8_t *buf, uint32_t buf_size)
Called at the beginning of each frame or field picture.
Definition: hwaccel_internal.h:59
vp8_mc_luma
static av_always_inline void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst, const ThreadFrame *ref, const VP8mv *mv, int x_off, int y_off, int block_w, int block_h, int width, int height, ptrdiff_t linesize, vp8_mc_func mc_func[3][3])
luma MC function
Definition: vp8.c:1831
ff_vp7dsp_init
void ff_vp7dsp_init(VP8DSPContext *c)
AV_LOG_ERROR
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:180
ff_vp8dsp_init
void ff_vp8dsp_init(VP8DSPContext *c)
FF_ARRAY_ELEMS
#define FF_ARRAY_ELEMS(a)
Definition: sinewin_tablegen.c:29
HOR_PRED
@ HOR_PRED
Definition: vp9.h:47
av_cold
#define av_cold
Definition: attributes.h:90
vp8_dct_cat2_prob
static const uint8_t vp8_dct_cat2_prob[]
Definition: vp8data.h:339
AV_FRAME_FLAG_KEY
#define AV_FRAME_FLAG_KEY
A flag to mark frames that are keyframes.
Definition: frame.h:628
ff_thread_report_progress
void ff_thread_report_progress(ThreadFrame *f, int n, int field)
Notify later decoding threads when part of their reference picture is ready.
Definition: pthread_frame.c:589
LOCAL_ALIGNED
#define LOCAL_ALIGNED(a, t, v,...)
Definition: mem_internal.h:112
width
#define width
ref_frame
static const AVFrame * ref_frame(const struct pl_frame_mix *mix)
Definition: vf_libplacebo.c:783
vp8_pred4x4_mode
static const uint8_t vp8_pred4x4_mode[]
Definition: vp8data.h:40
FF_CODEC_DECODE_CB
#define FF_CODEC_DECODE_CB(func)
Definition: codec_internal.h:306
ff_hwaccel_frame_priv_alloc
int ff_hwaccel_frame_priv_alloc(AVCodecContext *avctx, void **hwaccel_picture_private)
Allocate a hwaccel frame private data if the provided avctx uses a hwaccel method that needs it.
Definition: decode.c:1838
s
#define s(width, name)
Definition: cbs_vp9.c:198
filter_mb_simple
static av_always_inline void filter_mb_simple(const VP8Context *s, uint8_t *dst, const VP8FilterStrength *f, int mb_x, int mb_y)
Definition: vp8.c:2267
vpx_rac_renorm
static av_always_inline unsigned int vpx_rac_renorm(VPXRangeCoder *c)
Definition: vpx_rac.h:58
AV_ZERO64
#define AV_ZERO64(d)
Definition: intreadwrite.h:631
vp8_pred8x8c_prob_inter
static const uint8_t vp8_pred8x8c_prob_inter[3]
Definition: vp8data.h:189
DC_129_PRED8x8
#define DC_129_PRED8x8
Definition: h264pred.h:86
AV_ZERO32
#define AV_ZERO32(d)
Definition: intreadwrite.h:627
AV_GET_BUFFER_FLAG_REF
#define AV_GET_BUFFER_FLAG_REF
The decoder will keep a reference to the frame and may reuse it later.
Definition: avcodec.h:421
VP8Frame::tf
ThreadFrame tf
Definition: vp8.h:153
init
int(* init)(AVBSFContext *ctx)
Definition: dts2pts_bsf.c:365
vp8_pred16x16_tree_intra
static const int8_t vp8_pred16x16_tree_intra[4][2]
Definition: vp8data.h:47
bits
uint8_t bits
Definition: vp3data.h:128
parse_segment_info
static void parse_segment_info(VP8Context *s)
Definition: vp8.c:284
vp8_pred4x4_prob_inter
static const uint8_t vp8_pred4x4_prob_inter[9]
Definition: vp8data.h:192
pix_fmts
static enum AVPixelFormat pix_fmts[]
Definition: libkvazaar.c:304
vp8_mbsplits
static const uint8_t vp8_mbsplits[5][16]
Definition: vp8data.h:127
ctx
AVFormatContext * ctx
Definition: movenc.c:48
vp78_decode_frame
static av_always_inline int vp78_decode_frame(AVCodecContext *avctx, AVFrame *rframe, int *got_frame, const AVPacket *avpkt, int is_vp7)
Definition: vp8.c:2652
decode.h
AV_RL16
uint64_t_TMPL AV_WL64 unsigned int_TMPL AV_WL32 unsigned int_TMPL AV_WL24 unsigned int_TMPL AV_RL16
Definition: bytestream.h:94
vp7_mode_contexts
static const int vp7_mode_contexts[31][4]
Definition: vp8data.h:84
vp78_decode_mv_mb_modes
static av_always_inline int vp78_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *curframe, const VP8Frame *prev_frame, int is_vp7)
Definition: vp8.c:2301
AV_PIX_FMT_YUV420P
@ AV_PIX_FMT_YUV420P
planar YUV 4:2:0, 12bpp, (1 Cr & Cb sample per 2x2 Y samples)
Definition: pixfmt.h:66
atomic_load
#define atomic_load(object)
Definition: stdatomic.h:93
VP8_SPLITMVMODE_8x16
@ VP8_SPLITMVMODE_8x16
2 8x16 blocks (horizontal)
Definition: vp8.h:78
CODEC_LONG_NAME
#define CODEC_LONG_NAME(str)
Definition: codec_internal.h:272
VP8Frame::seg_map
uint8_t * seg_map
RefStruct reference.
Definition: vp8.h:154
frame
static AVFrame * frame
Definition: demux_decode.c:54
ff_thread_ref_frame
int ff_thread_ref_frame(ThreadFrame *dst, const ThreadFrame *src)
Definition: utils.c:871
vp8_mv_default_prob
static const uint8_t vp8_mv_default_prob[2][19]
Definition: vp8data.h:731
vp8_coeff_band_indexes
static const int8_t vp8_coeff_band_indexes[8][10]
Definition: vp8data.h:325
TOP_DC_PRED8x8
#define TOP_DC_PRED8x8
Definition: h264pred.h:75
if
if(ret)
Definition: filter_design.txt:179
vp8_pred16x16_prob_inter
static const uint8_t vp8_pred16x16_prob_inter[4]
Definition: vp8data.h:164
vp8_decode_flush_impl
static void vp8_decode_flush_impl(AVCodecContext *avctx, int free_mem)
Definition: vp8.c:147
AV_CODEC_CAP_FRAME_THREADS
#define AV_CODEC_CAP_FRAME_THREADS
Codec supports frame-level multithreading.
Definition: codec.h:110
AVDISCARD_ALL
@ AVDISCARD_ALL
discard all
Definition: defs.h:219
threadframe.h
vp8_rac_get_nn
static int vp8_rac_get_nn(VPXRangeCoder *c)
Definition: vp8.c:65
clamp_mv
static av_always_inline void clamp_mv(const VP8mvbounds *s, VP8mv *dst, const VP8mv *src)
Definition: vp8.c:872
NULL
#define NULL
Definition: coverity.c:32
sizes
static const int sizes[][2]
Definition: img2dec.c:58
AV_COPY128
#define AV_COPY128(d, s)
Definition: intreadwrite.h:607
AVCodecContext::color_range
enum AVColorRange color_range
MPEG vs JPEG YUV range.
Definition: avcodec.h:1039
vp7_decode_mb_row_sliced
static int vp7_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2639
AV_COPY64
#define AV_COPY64(d, s)
Definition: intreadwrite.h:603
hwaccel_internal.h
vp8_update_dimensions
static int vp8_update_dimensions(VP8Context *s, int width, int height)
Definition: vp8.c:278
VP8FilterStrength
Definition: vp8.h:89
NUM_DCT_TOKENS
@ NUM_DCT_TOKENS
Definition: vp8.h:64
AV_PICTURE_TYPE_I
@ AV_PICTURE_TYPE_I
Intra.
Definition: avutil.h:279
vp89_rac_get_uint
static av_unused int vp89_rac_get_uint(VPXRangeCoder *c, int bits)
Definition: vp89_rac.h:41
check_thread_pos
#define check_thread_pos(td, otd, mb_x_check, mb_y_check)
Definition: vp8.c:2384
VP7MVPred
Definition: vp8data.h:61
ff_thread_release_ext_buffer
void ff_thread_release_ext_buffer(ThreadFrame *f)
Unref a ThreadFrame.
Definition: pthread_frame.c:1012
mathops.h
vp8_mc_func
void(* vp8_mc_func)(uint8_t *dst, ptrdiff_t dstStride, const uint8_t *src, ptrdiff_t srcStride, int h, int x, int y)
Definition: vp8dsp.h:33
vp7_yac_qlookup
static const uint16_t vp7_yac_qlookup[]
Definition: vp8data.h:766
vp8_token_default_probs
static const uint8_t vp8_token_default_probs[4][8][3][NUM_DCT_TOKENS - 1]
Definition: vp8data.h:345
ff_refstruct_allocz
static void * ff_refstruct_allocz(size_t size)
Equivalent to ff_refstruct_alloc_ext(size, 0, NULL, NULL)
Definition: refstruct.h:105
VERT_PRED8x8
#define VERT_PRED8x8
Definition: h264pred.h:70
vp8_token_update_probs
static const uint8_t vp8_token_update_probs[4][8][3][NUM_DCT_TOKENS - 1]
Definition: vp8data.h:516
vp8_mbsplit_count
static const uint8_t vp8_mbsplit_count[4]
Definition: vp8data.h:142
UPDATE_THREAD_CONTEXT
#define UPDATE_THREAD_CONTEXT(func)
Definition: codec_internal.h:281
vp8_decode_mvs
static av_always_inline void vp8_decode_mvs(VP8Context *s, const VP8mvbounds *mv_bounds, VP8Macroblock *mb, int mb_x, int mb_y, int layout)
Definition: vp8.c:1130
AV_ZERO128
#define AV_ZERO128(d)
Definition: intreadwrite.h:635
FF_HW_HAS_CB
#define FF_HW_HAS_CB(avctx, function)
Definition: hwaccel_internal.h:177
VP8FrameType
VP8FrameType
Definition: vp8.h:42
decode_mb_row_no_filter
static av_always_inline int decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr, int is_vp7)
Definition: vp8.c:2388
VP8mv
Definition: vp8.h:84
vp7_feature_value_size
static const uint8_t vp7_feature_value_size[2][4]
Definition: vp8data.h:742
index
int index
Definition: gxfenc.c:89
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
VP8Frame
Definition: vp8.h:152
vp8.h
ff_vp8_decode_init
av_cold int ff_vp8_decode_init(AVCodecContext *avctx)
Definition: vp8.c:2923
VP8_FRAME_GOLDEN
@ VP8_FRAME_GOLDEN
Definition: vp8.h:46
FF_SIGNBIT
#define FF_SIGNBIT(x)
Definition: mathops.h:130
vp8_mbfirstidx
static const uint8_t vp8_mbfirstidx[4][16]
Definition: vp8data.h:135
DC_127_PRED8x8
#define DC_127_PRED8x8
Definition: h264pred.h:85
AVDISCARD_NONKEY
@ AVDISCARD_NONKEY
discard all frames except keyframes
Definition: defs.h:218
f
f
Definition: af_crystalizer.c:121
xchg_mb_border
static av_always_inline void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, ptrdiff_t linesize, ptrdiff_t uvlinesize, int mb_x, int mb_y, int mb_width, int simple, int xchg)
Definition: vp8.c:1579
ff_zigzag_scan
const uint8_t ff_zigzag_scan[16+1]
Definition: mathtables.c:109
AV_CODEC_CAP_DR1
#define AV_CODEC_CAP_DR1
Codec uses get_buffer() or get_encode_buffer() for allocating buffers and supports custom allocators.
Definition: codec.h:52
vp8_pred4x4_tree
static const int8_t vp8_pred4x4_tree[9][2]
Definition: vp8data.h:168
MV_EDGE_CHECK
#define MV_EDGE_CHECK(n)
AVPacket::size
int size
Definition: packet.h:492
dc
Tag MUST be and< 10hcoeff half pel interpolation filter coefficients, hcoeff[0] are the 2 middle coefficients[1] are the next outer ones and so on, resulting in a filter like:...eff[2], hcoeff[1], hcoeff[0], hcoeff[0], hcoeff[1], hcoeff[2] ... the sign of the coefficients is not explicitly stored but alternates after each coeff and coeff[0] is positive, so ...,+,-,+,-,+,+,-,+,-,+,... hcoeff[0] is not explicitly stored but found by subtracting the sum of all stored coefficients with signs from 32 hcoeff[0]=32 - hcoeff[1] - hcoeff[2] - ... a good choice for hcoeff and htaps is htaps=6 hcoeff={40,-10, 2} an alternative which requires more computations at both encoder and decoder side and may or may not be better is htaps=8 hcoeff={42,-14, 6,-2}ref_frames minimum of the number of available reference frames and max_ref_frames for example the first frame after a key frame always has ref_frames=1spatial_decomposition_type wavelet type 0 is a 9/7 symmetric compact integer wavelet 1 is a 5/3 symmetric compact integer wavelet others are reserved stored as delta from last, last is reset to 0 if always_reset||keyframeqlog quality(logarithmic quantizer scale) stored as delta from last, last is reset to 0 if always_reset||keyframemv_scale stored as delta from last, last is reset to 0 if always_reset||keyframe FIXME check that everything works fine if this changes between framesqbias dequantization bias stored as delta from last, last is reset to 0 if always_reset||keyframeblock_max_depth maximum depth of the block tree stored as delta from last, last is reset to 0 if always_reset||keyframequant_table quantization tableHighlevel bitstream structure:==============================--------------------------------------------|Header|--------------------------------------------|------------------------------------|||Block0||||split?||||yes no||||......... intra?||||:Block01 :yes no||||:Block02 :....... ..........||||:Block03 ::y DC ::ref index:||||:Block04 ::cb DC ::motion x :||||......... :cr DC ::motion y :||||....... ..........|||------------------------------------||------------------------------------|||Block1|||...|--------------------------------------------|------------ ------------ ------------|||Y subbands||Cb subbands||Cr subbands||||--- ---||--- ---||--- ---|||||LL0||HL0||||LL0||HL0||||LL0||HL0|||||--- ---||--- ---||--- ---||||--- ---||--- ---||--- ---|||||LH0||HH0||||LH0||HH0||||LH0||HH0|||||--- ---||--- ---||--- ---||||--- ---||--- ---||--- ---|||||HL1||LH1||||HL1||LH1||||HL1||LH1|||||--- ---||--- ---||--- ---||||--- ---||--- ---||--- ---|||||HH1||HL2||||HH1||HL2||||HH1||HL2|||||...||...||...|||------------ ------------ ------------|--------------------------------------------Decoding process:=================------------|||Subbands|------------||||------------|Intra DC||||LL0 subband prediction ------------|\ Dequantization ------------------- \||Reference frames|\ IDWT|------- -------|Motion \|||Frame 0||Frame 1||Compensation . OBMC v -------|------- -------|--------------. \------> Frame n output Frame Frame<----------------------------------/|...|------------------- Range Coder:============Binary Range Coder:------------------- The implemented range coder is an adapted version based upon "Range encoding: an algorithm for removing redundancy from a digitised message." by G. N. N. Martin. The symbols encoded by the Snow range coder are bits(0|1). The associated probabilities are not fix but change depending on the symbol mix seen so far. bit seen|new state ---------+----------------------------------------------- 0|256 - state_transition_table[256 - old_state];1|state_transition_table[old_state];state_transition_table={ 0, 0, 0, 0, 0, 0, 0, 0, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 190, 191, 192, 194, 194, 195, 196, 197, 198, 199, 200, 201, 202, 202, 204, 205, 206, 207, 208, 209, 209, 210, 211, 212, 213, 215, 215, 216, 217, 218, 219, 220, 220, 222, 223, 224, 225, 226, 227, 227, 229, 229, 230, 231, 232, 234, 234, 235, 236, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 248, 0, 0, 0, 0, 0, 0, 0};FIXME Range Coding of integers:------------------------- FIXME Neighboring Blocks:===================left and top are set to the respective blocks unless they are outside of the image in which case they are set to the Null block top-left is set to the top left block unless it is outside of the image in which case it is set to the left block if this block has no larger parent block or it is at the left side of its parent block and the top right block is not outside of the image then the top right block is used for top-right else the top-left block is used Null block y, cb, cr are 128 level, ref, mx and my are 0 Motion Vector Prediction:=========================1. the motion vectors of all the neighboring blocks are scaled to compensate for the difference of reference frames scaled_mv=(mv *(256 *(current_reference+1)/(mv.reference+1))+128)> the median of the scaled top and top right vectors is used as motion vector prediction the used motion vector is the sum of the predictor and(mvx_diff, mvy_diff) *mv_scale Intra DC Prediction block[y][x] dc[1]
Definition: snow.txt:400
copy
static void copy(const float *p1, float *p2, const int length)
Definition: vf_vaguedenoiser.c:185
av_frame_ref
int av_frame_ref(AVFrame *dst, const AVFrame *src)
Set up a new reference to the data described by the source frame.
Definition: frame.c:361
codec_internal.h
vp8_coeff_band
static const uint8_t vp8_coeff_band[16]
Definition: vp8data.h:319
subpel_idx
static const uint8_t subpel_idx[3][8]
Definition: vp8.c:1807
vp7_update_dimensions
static int vp7_update_dimensions(VP8Context *s, int width, int height)
Definition: vp8.c:273
EDGE_EMU_LINESIZE
#define EDGE_EMU_LINESIZE
Definition: vp8.h:146
size
int size
Definition: twinvq_data.h:10344
VERT_RIGHT_PRED
@ VERT_RIGHT_PRED
Definition: vp9.h:51
FF_CODEC_CAP_ALLOCATE_PROGRESS
#define FF_CODEC_CAP_ALLOCATE_PROGRESS
Definition: codec_internal.h:69
free_buffers
static void free_buffers(VP8Context *s)
Definition: vp8.c:83
DC_128_PRED8x8
#define DC_128_PRED8x8
Definition: h264pred.h:76
ffcodec
static const av_always_inline FFCodec * ffcodec(const AVCodec *codec)
Definition: codec_internal.h:325
decode_block_coeffs
static av_always_inline int decode_block_coeffs(VPXRangeCoder *c, int16_t block[16], uint8_t probs[16][3][NUM_DCT_TOKENS - 1], int i, int zero_nhood, const int16_t qmul[2], const uint8_t scan[16], int vp7)
Definition: vp8.c:1477
FF_THREAD_SLICE
#define FF_THREAD_SLICE
Decode more than one part of a single frame at once.
Definition: avcodec.h:1544
ff_vp8_dct_cat_prob
const uint8_t *const ff_vp8_dct_cat_prob[]
Definition: vp8data.c:36
vp8_pred8x8c_prob_intra
static const uint8_t vp8_pred8x8c_prob_intra[3]
Definition: vp8data.h:186
vp8_decode_mv_mb_modes
static int vp8_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame, const VP8Frame *prev_frame)
Definition: vp8.c:2344
AV_RL24
uint64_t_TMPL AV_WL64 unsigned int_TMPL AV_WL32 unsigned int_TMPL AV_RL24
Definition: bytestream.h:93
AVCodecHWConfigInternal
Definition: hwconfig.h:25
vp8_pred4x4_prob_intra
static const uint8_t vp8_pred4x4_prob_intra[10][10][9]
Definition: vp8data.h:196
VP8ThreadData
Definition: vp8.h:120
height
#define height
a
The reader does not expect b to be semantically here and if the code is changed by maybe adding a a division or other the signedness will almost certainly be mistaken To avoid this confusion a new type was SUINT is the C unsigned type but it holds a signed int to use the same example SUINT a
Definition: undefined.txt:41
AV_CODEC_CAP_SLICE_THREADS
#define AV_CODEC_CAP_SLICE_THREADS
Codec supports slice-based (or partition-based) multithreading.
Definition: codec.h:114
vp8_decode_frame_header
static int vp8_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
Definition: vp8.c:731
vp8_mbsplit_prob
static const uint8_t vp8_mbsplit_prob[3]
Definition: vp8data.h:145
setup_partitions
static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size)
Definition: vp8.c:330
vp8_pred16x16_tree_inter
static const int8_t vp8_pred16x16_tree_inter[4][2]
Definition: vp8data.h:54
vp7_feature_index_tree
static const int8_t vp7_feature_index_tree[4][2]
Definition: vp8data.h:747
AVCodecContext::skip_loop_filter
enum AVDiscard skip_loop_filter
Skip loop filtering for selected frames.
Definition: avcodec.h:1752
HWACCEL_NVDEC
#define HWACCEL_NVDEC(codec)
Definition: hwconfig.h:68
PLANE_PRED8x8
#define PLANE_PRED8x8
Definition: h264pred.h:71
vpx_rac_is_end
static av_always_inline int vpx_rac_is_end(VPXRangeCoder *c)
returns 1 if the end of the stream has been reached, 0 otherwise.
Definition: vpx_rac.h:51
AV_PIX_FMT_VAAPI
@ AV_PIX_FMT_VAAPI
Hardware acceleration through VA-API, data[3] contains a VASurfaceID.
Definition: pixfmt.h:119
mb
#define mb
Definition: vf_colormatrix.c:99
pthread_cond_destroy
static av_always_inline int pthread_cond_destroy(pthread_cond_t *cond)
Definition: os2threads.h:144
MODE_I4x4
#define MODE_I4x4
Definition: vp8.h:68
FF_THREAD_FRAME
#define FF_THREAD_FRAME
Decode more than one frame at once.
Definition: avcodec.h:1543
H_LOOP_FILTER_16Y_INNER
#define H_LOOP_FILTER_16Y_INNER(cond)
vp78_decode_mb_row_sliced
static av_always_inline int vp78_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr, int is_vp7)
Definition: vp8.c:2605
AVCodec::id
enum AVCodecID id
Definition: codec.h:201
pthread_mutex_destroy
static av_always_inline int pthread_mutex_destroy(pthread_mutex_t *mutex)
Definition: os2threads.h:112
layout
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel layout
Definition: filter_design.txt:18
AV_CODEC_ID_VP7
@ AV_CODEC_ID_VP7
Definition: codec_id.h:233
ref_to_update
static VP8FrameType ref_to_update(VP8Context *s, int update, VP8FrameType ref)
Determine which buffers golden and altref should be updated with after this frame.
Definition: vp8.c:427
vp8_read_mv_component
static int vp8_read_mv_component(VPXRangeCoder *c, const uint8_t *p)
Definition: vp8.c:916
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:255
DC_129_PRED
@ DC_129_PRED
Definition: vp9.h:60
modes
static const SiprModeParam modes[MODE_COUNT]
Definition: sipr.c:70
vpx_rac.h
src2
const pixel * src2
Definition: h264pred_template.c:422
vp7_fade_frame
static int vp7_fade_frame(VP8Context *s, int alpha, int beta)
Definition: vp8.c:529
av_always_inline
#define av_always_inline
Definition: attributes.h:49
FFMIN
#define FFMIN(a, b)
Definition: macros.h:49
vpx_rac_get_prob_branchy
static av_always_inline int vpx_rac_get_prob_branchy(VPXRangeCoder *c, int prob)
Definition: vpx_rac.h:99
intra_predict
static av_always_inline void intra_predict(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], VP8Macroblock *mb, int mb_x, int mb_y, int is_vp7)
Definition: vp8.c:1693
AV_COPY32
#define AV_COPY32(d, s)
Definition: intreadwrite.h:599
av_mallocz
void * av_mallocz(size_t size)
Allocate a memory block with alignment suitable for all memory accesses (including vectors if availab...
Definition: mem.c:254
AVCodec::name
const char * name
Name of the codec implementation.
Definition: codec.h:194
VP8mv::x
int16_t x
Definition: vp8.h:85
vp78_reset_probability_tables
static void vp78_reset_probability_tables(VP8Context *s)
Definition: vp8.c:443
vp7_decode_frame_header
static int vp7_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
Definition: vp8.c:567
VP8_FRAME_NONE
@ VP8_FRAME_NONE
Definition: vp8.h:43
profile
int profile
Definition: mxfenc.c:2115
fade
static void fade(uint8_t *dst, ptrdiff_t dst_linesize, const uint8_t *src, ptrdiff_t src_linesize, int width, int height, int alpha, int beta)
Definition: vp8.c:513
decode_intra4x4_modes
static av_always_inline void decode_intra4x4_modes(VP8Context *s, VPXRangeCoder *c, VP8Macroblock *mb, int mb_x, int keyframe, int layout)
Definition: vp8.c:1230
VP8Macroblock
Definition: vp8.h:95
AVCodecContext::pix_fmt
enum AVPixelFormat pix_fmt
Pixel format, see AV_PIX_FMT_xxx.
Definition: avcodec.h:658
ff_thread_get_ext_buffer
int ff_thread_get_ext_buffer(AVCodecContext *avctx, ThreadFrame *f, int flags)
Wrapper around ff_get_buffer() for frame-multithreaded codecs.
Definition: pthread_frame.c:984
AVCOL_RANGE_MPEG
@ AVCOL_RANGE_MPEG
Narrow or limited range content.
Definition: pixfmt.h:656
vp8_decode_mb_row_sliced
static int vp8_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2645
ff_vp7_decoder
const FFCodec ff_vp7_decoder
VP8_SPLITMVMODE_NONE
@ VP8_SPLITMVMODE_NONE
(only used in prediction) no split MVs
Definition: vp8.h:81
LEFT_DC_PRED8x8
#define LEFT_DC_PRED8x8
Definition: h264pred.h:74
prefetch_motion
static av_always_inline void prefetch_motion(const VP8Context *s, const VP8Macroblock *mb, int mb_x, int mb_y, int mb_xy, int ref)
Definition: vp8.c:1977
avcodec.h
AV_RN32A
#define AV_RN32A(p)
Definition: intreadwrite.h:524
vp89_rac_get_tree
static av_always_inline int vp89_rac_get_tree(VPXRangeCoder *c, const int8_t(*tree)[2], const uint8_t *probs)
Definition: vp89_rac.h:54
decode_mb_coeffs
static av_always_inline void decode_mb_coeffs(VP8Context *s, VP8ThreadData *td, VPXRangeCoder *c, VP8Macroblock *mb, uint8_t t_nnz[9], uint8_t l_nnz[9], int is_vp7)
Definition: vp8.c:1492
av_uninit
#define av_uninit(x)
Definition: attributes.h:154
ret
ret
Definition: filter_design.txt:187
AV_LOG_FATAL
#define AV_LOG_FATAL
Something went wrong and recovery is not possible.
Definition: log.h:174
check_dc_pred8x8_mode
static av_always_inline int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y)
Definition: vp8.c:1613
pred
static const float pred[4]
Definition: siprdata.h:259
vp7_decode_mb_row_no_filter
static int vp7_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2521
FFSWAP
#define FFSWAP(type, a, b)
Definition: macros.h:52
vp8_pred16x16_prob_intra
static const uint8_t vp8_pred16x16_prob_intra[4]
Definition: vp8data.h:161
vp8_ac_qlookup
static const uint16_t vp8_ac_qlookup[VP8_MAX_QUANT+1]
Definition: vp8data.h:698
prob
#define prob(name, subs,...)
Definition: cbs_vp9.c:325
ff_vpx_init_range_decoder
int ff_vpx_init_range_decoder(VPXRangeCoder *c, const uint8_t *buf, int buf_size)
Definition: vpx_rac.c:42
ff_refstruct_replace
void ff_refstruct_replace(void *dstp, const void *src)
Ensure *dstp refers to the same object as src.
Definition: refstruct.c:156
ff_thread_finish_setup
the pkt_dts and pkt_pts fields in AVFrame will work as usual Restrictions on codec whose streams don t reset across will not work because their bitstreams cannot be decoded in parallel *The contents of buffers must not be read before as well as code calling up to before the decode process starts Call ff_thread_finish_setup() afterwards. If some code can 't be moved
left
Tag MUST be and< 10hcoeff half pel interpolation filter coefficients, hcoeff[0] are the 2 middle coefficients[1] are the next outer ones and so on, resulting in a filter like:...eff[2], hcoeff[1], hcoeff[0], hcoeff[0], hcoeff[1], hcoeff[2] ... the sign of the coefficients is not explicitly stored but alternates after each coeff and coeff[0] is positive, so ...,+,-,+,-,+,+,-,+,-,+,... hcoeff[0] is not explicitly stored but found by subtracting the sum of all stored coefficients with signs from 32 hcoeff[0]=32 - hcoeff[1] - hcoeff[2] - ... a good choice for hcoeff and htaps is htaps=6 hcoeff={40,-10, 2} an alternative which requires more computations at both encoder and decoder side and may or may not be better is htaps=8 hcoeff={42,-14, 6,-2}ref_frames minimum of the number of available reference frames and max_ref_frames for example the first frame after a key frame always has ref_frames=1spatial_decomposition_type wavelet type 0 is a 9/7 symmetric compact integer wavelet 1 is a 5/3 symmetric compact integer wavelet others are reserved stored as delta from last, last is reset to 0 if always_reset||keyframeqlog quality(logarithmic quantizer scale) stored as delta from last, last is reset to 0 if always_reset||keyframemv_scale stored as delta from last, last is reset to 0 if always_reset||keyframe FIXME check that everything works fine if this changes between framesqbias dequantization bias stored as delta from last, last is reset to 0 if always_reset||keyframeblock_max_depth maximum depth of the block tree stored as delta from last, last is reset to 0 if always_reset||keyframequant_table quantization tableHighlevel bitstream structure:==============================--------------------------------------------|Header|--------------------------------------------|------------------------------------|||Block0||||split?||||yes no||||......... intra?||||:Block01 :yes no||||:Block02 :....... ..........||||:Block03 ::y DC ::ref index:||||:Block04 ::cb DC ::motion x :||||......... :cr DC ::motion y :||||....... ..........|||------------------------------------||------------------------------------|||Block1|||...|--------------------------------------------|------------ ------------ ------------|||Y subbands||Cb subbands||Cr subbands||||--- ---||--- ---||--- ---|||||LL0||HL0||||LL0||HL0||||LL0||HL0|||||--- ---||--- ---||--- ---||||--- ---||--- ---||--- ---|||||LH0||HH0||||LH0||HH0||||LH0||HH0|||||--- ---||--- ---||--- ---||||--- ---||--- ---||--- ---|||||HL1||LH1||||HL1||LH1||||HL1||LH1|||||--- ---||--- ---||--- ---||||--- ---||--- ---||--- ---|||||HH1||HL2||||HH1||HL2||||HH1||HL2|||||...||...||...|||------------ ------------ ------------|--------------------------------------------Decoding process:=================------------|||Subbands|------------||||------------|Intra DC||||LL0 subband prediction ------------|\ Dequantization ------------------- \||Reference frames|\ IDWT|------- -------|Motion \|||Frame 0||Frame 1||Compensation . OBMC v -------|------- -------|--------------. \------> Frame n output Frame Frame<----------------------------------/|...|------------------- Range Coder:============Binary Range Coder:------------------- The implemented range coder is an adapted version based upon "Range encoding: an algorithm for removing redundancy from a digitised message." by G. N. N. Martin. The symbols encoded by the Snow range coder are bits(0|1). The associated probabilities are not fix but change depending on the symbol mix seen so far. bit seen|new state ---------+----------------------------------------------- 0|256 - state_transition_table[256 - old_state];1|state_transition_table[old_state];state_transition_table={ 0, 0, 0, 0, 0, 0, 0, 0, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 190, 191, 192, 194, 194, 195, 196, 197, 198, 199, 200, 201, 202, 202, 204, 205, 206, 207, 208, 209, 209, 210, 211, 212, 213, 215, 215, 216, 217, 218, 219, 220, 220, 222, 223, 224, 225, 226, 227, 227, 229, 229, 230, 231, 232, 234, 234, 235, 236, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 248, 0, 0, 0, 0, 0, 0, 0};FIXME Range Coding of integers:------------------------- FIXME Neighboring Blocks:===================left and top are set to the respective blocks unless they are outside of the image in which case they are set to the Null block top-left is set to the top left block unless it is outside of the image in which case it is set to the left block if this block has no larger parent block or it is at the left side of its parent block and the top right block is not outside of the image then the top right block is used for top-right else the top-left block is used Null block y, cb, cr are 128 level, ref, mx and my are 0 Motion Vector Prediction:=========================1. the motion vectors of all the neighboring blocks are scaled to compensate for the difference of reference frames scaled_mv=(mv *(256 *(current_reference+1)/(mv.reference+1))+128)> the median of the scaled left
Definition: snow.txt:386
AV_RL32
uint64_t_TMPL AV_WL64 unsigned int_TMPL AV_RL32
Definition: bytestream.h:92
vp89_rac_get
static av_always_inline int vp89_rac_get(VPXRangeCoder *c)
Definition: vp89_rac.h:36
AVCodecContext
main external API structure.
Definition: avcodec.h:441
AVCodecContext::active_thread_type
int active_thread_type
Which multithreading methods are in use by the codec.
Definition: avcodec.h:1551
VP8Macroblock::intra4x4_pred_mode_top
uint8_t intra4x4_pred_mode_top[4]
Definition: vp8.h:105
decode_splitmvs
static av_always_inline int decode_splitmvs(const VP8Context *s, VPXRangeCoder *c, VP8Macroblock *mb, int layout, int is_vp7)
Split motion vector prediction, 16.4.
Definition: vp8.c:939
ThreadFrame
Definition: threadframe.h:27
ff_h264_pred_init
av_cold void ff_h264_pred_init(H264PredContext *h, int codec_id, const int bit_depth, int chroma_format_idc)
Set the intra prediction function pointers.
Definition: h264pred.c:437
HOR_UP_PRED
@ HOR_UP_PRED
Definition: vp9.h:54
vp8data.h
mode
mode
Definition: ebur128.h:83
AV_PIX_FMT_NONE
@ AV_PIX_FMT_NONE
Definition: pixfmt.h:65
ffhwaccel
static const FFHWAccel * ffhwaccel(const AVHWAccel *codec)
Definition: hwaccel_internal.h:166
FFHWAccel::decode_slice
int(* decode_slice)(AVCodecContext *avctx, const uint8_t *buf, uint32_t buf_size)
Callback for each slice.
Definition: hwaccel_internal.h:86
VP8Macroblock::mode
uint8_t mode
Definition: vp8.h:99
IS_VP8
#define IS_VP8
Definition: vp8dsp.h:100
VP8_MVMODE_SPLIT
@ VP8_MVMODE_SPLIT
Definition: vp8.h:73
ref
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:112
HOR_DOWN_PRED
@ HOR_DOWN_PRED
Definition: vp9.h:52
vp7_decode_block_coeffs_internal
static int vp7_decode_block_coeffs_internal(VPXRangeCoder *r, int16_t block[16], uint8_t probs[16][3][NUM_DCT_TOKENS - 1], int i, const uint8_t *token_prob, const int16_t qmul[2], const uint8_t scan[16])
Definition: vp8.c:1440
filter_mb
static av_always_inline void filter_mb(const VP8Context *s, uint8_t *const dst[3], const VP8FilterStrength *f, int mb_x, int mb_y, int is_vp7)
Definition: vp8.c:2180
segment
Definition: hls.c:76
av_clip_uint8
#define av_clip_uint8
Definition: common.h:102
vp78_update_probability_tables
static void vp78_update_probability_tables(VP8Context *s)
Definition: vp8.c:452
AV_PICTURE_TYPE_P
@ AV_PICTURE_TYPE_P
Predicted.
Definition: avutil.h:280
AVMEDIA_TYPE_VIDEO
@ AVMEDIA_TYPE_VIDEO
Definition: avutil.h:201
flush
void(* flush)(AVBSFContext *ctx)
Definition: dts2pts_bsf.c:367
vp78_update_pred16x16_pred8x8_mvc_probabilities
static void vp78_update_pred16x16_pred8x8_mvc_probabilities(VP8Context *s, int mvc_size)
Definition: vp8.c:471
avpriv_request_sample
#define avpriv_request_sample(...)
Definition: tableprint_vlc.h:36
update_refs
static void update_refs(VP8Context *s)
Definition: vp8.c:491
vp7_decode_mv_mb_modes
static int vp7_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame, const VP8Frame *prev_frame)
Definition: vp8.c:2338
update_lf_deltas
static void update_lf_deltas(VP8Context *s)
Definition: vp8.c:306
filter_mb_row
static av_always_inline void filter_mb_row(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr, int is_vp7)
Definition: vp8.c:2533
alpha
static const int16_t alpha[]
Definition: ilbcdata.h:55
AVPacket
This structure stores compressed data.
Definition: packet.h:468
AVCodecContext::priv_data
void * priv_data
Definition: avcodec.h:468
get_pixel_format
static enum AVPixelFormat get_pixel_format(VP8Context *s)
Definition: vp8.c:192
av_freep
#define av_freep(p)
Definition: tableprint_vlc.h:34
ff_vp8_decode_frame
int ff_vp8_decode_frame(AVCodecContext *avctx, AVFrame *frame, int *got_frame, AVPacket *avpkt)
Definition: vp8.c:2842
src
INIT_CLIP pixel * src
Definition: h264pred_template.c:418
VP7MVPred::score
uint8_t score
Definition: vp8data.h:65
VP8Context
Definition: vp8.h:160
HWACCEL_VAAPI
#define HWACCEL_VAAPI(codec)
Definition: hwconfig.h:70
DIAG_DOWN_LEFT_PRED
@ DIAG_DOWN_LEFT_PRED
Definition: vp9.h:49
vp8_find_free_buffer
static VP8Frame * vp8_find_free_buffer(VP8Context *s)
Definition: vp8.c:168
int32_t
int32_t
Definition: audioconvert.c:56
AV_CODEC_ID_VP8
@ AV_CODEC_ID_VP8
Definition: codec_id.h:192
AVFrame::linesize
int linesize[AV_NUM_DATA_POINTERS]
For video, a positive or negative value, which is typically indicating the size in bytes of each pict...
Definition: frame.h:385
coeff
static const double coeff[2][5]
Definition: vf_owdenoise.c:79
block
The exact code depends on how similar the blocks are and how related they are to the block
Definition: filter_design.txt:207
av_log
#define av_log(a,...)
Definition: tableprint_vlc.h:27
AV_WN64
#define AV_WN64(p, v)
Definition: intreadwrite.h:378
VP8_MVMODE_ZERO
@ VP8_MVMODE_ZERO
Definition: vp8.h:71
AVERROR_INVALIDDATA
#define AVERROR_INVALIDDATA
Invalid data found when processing input.
Definition: error.h:61
pthread_cond_init
static av_always_inline int pthread_cond_init(pthread_cond_t *cond, const pthread_condattr_t *attr)
Definition: os2threads.h:133
vp7_y2ac_qlookup
static const uint16_t vp7_y2ac_qlookup[]
Definition: vp8data.h:792
atomic_init
#define atomic_init(obj, value)
Definition: stdatomic.h:33
vp7_submv_prob
static const uint8_t vp7_submv_prob[3]
Definition: vp8data.h:149
AVDiscard
AVDiscard
Definition: defs.h:210
AVDISCARD_NONREF
@ AVDISCARD_NONREF
discard all non reference
Definition: defs.h:215
vp8_rac_get_coeff
static int vp8_rac_get_coeff(VPXRangeCoder *c, const uint8_t *prob)
Definition: vp8.c:72
vp8_dc_qlookup
static const uint8_t vp8_dc_qlookup[VP8_MAX_QUANT+1]
Definition: vp8data.h:687
copy_chroma
static void copy_chroma(AVFrame *dst, const AVFrame *src, int width, int height)
Definition: vp8.c:502
VP8_FRAME_PREVIOUS
@ VP8_FRAME_PREVIOUS
Definition: vp8.h:45
vpx_rac_get_prob
#define vpx_rac_get_prob
Definition: vpx_rac.h:82
vp8_decode_flush
static void vp8_decode_flush(AVCodecContext *avctx)
Definition: vp8.c:163
AVCodecContext::execute2
int(* execute2)(struct AVCodecContext *c, int(*func)(struct AVCodecContext *c2, void *arg, int jobnr, int threadnr), void *arg2, int *ret, int count)
The codec may call this to execute several independent things.
Definition: avcodec.h:1581
VP8_MVMODE_MV
@ VP8_MVMODE_MV
Definition: vp8.h:72
MARGIN
#define MARGIN
Definition: vp8.c:2299
ff_refstruct_unref
void ff_refstruct_unref(void *objp)
Decrement the reference count of the underlying object and automatically free the object if there are...
Definition: refstruct.c:116
vp8_alloc_frame
static int vp8_alloc_frame(VP8Context *s, VP8Frame *f, int ref)
Definition: vp8.c:103
get_submv_prob
static const av_always_inline uint8_t * get_submv_prob(uint32_t left, uint32_t top, int is_vp7)
Definition: vp8.c:922
ff_vp78dsp_init
av_cold void ff_vp78dsp_init(VP8DSPContext *dsp)
Definition: vp8dsp.c:668
VP8Frame::hwaccel_picture_private
void * hwaccel_picture_private
RefStruct reference.
Definition: vp8.h:156