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.h"
30 #include "libavutil/mem_internal.h"
31 
32 #include "avcodec.h"
33 #include "codec_internal.h"
34 #include "decode.h"
35 #include "hwaccel_internal.h"
36 #include "hwconfig.h"
37 #include "mathops.h"
38 #include "progressframe.h"
39 #include "refstruct.h"
40 #include "thread.h"
41 #include "vp8.h"
42 #include "vp89_rac.h"
43 #include "vp8data.h"
44 #include "vpx_rac.h"
45 
46 #if ARCH_ARM
47 # include "arm/vp8.h"
48 #endif
49 
50 // fixme: add 1 bit to all the calls to this?
52 {
53  int v;
54 
55  if (!vp89_rac_get(c))
56  return 0;
57 
58  v = vp89_rac_get_uint(c, bits);
59 
60  if (vp89_rac_get(c))
61  v = -v;
62 
63  return v;
64 }
65 
67 {
68  int v = vp89_rac_get_uint(c, 7) << 1;
69  return v + !v;
70 }
71 
72 // DCTextra
73 static int vp8_rac_get_coeff(VPXRangeCoder *c, const uint8_t *prob)
74 {
75  int v = 0;
76 
77  do {
78  v = (v<<1) + vpx_rac_get_prob(c, *prob++);
79  } while (*prob);
80 
81  return v;
82 }
83 
84 static void free_buffers(VP8Context *s)
85 {
86  int i;
87  if (s->thread_data)
88  for (i = 0; i < MAX_THREADS; i++) {
89 #if HAVE_THREADS
90  pthread_cond_destroy(&s->thread_data[i].cond);
91  pthread_mutex_destroy(&s->thread_data[i].lock);
92 #endif
93  av_freep(&s->thread_data[i].filter_strength);
94  }
95  av_freep(&s->thread_data);
96  av_freep(&s->macroblocks_base);
97  av_freep(&s->intra4x4_pred_mode_top);
98  av_freep(&s->top_nnz);
99  av_freep(&s->top_border);
100 
101  s->macroblocks = NULL;
102 }
103 
105 {
106  int ret = ff_progress_frame_get_buffer(s->avctx, &f->tf,
108  if (ret < 0)
109  return ret;
110  f->seg_map = ff_refstruct_allocz(s->mb_width * s->mb_height);
111  if (!f->seg_map) {
112  ret = AVERROR(ENOMEM);
113  goto fail;
114  }
115  ret = ff_hwaccel_frame_priv_alloc(s->avctx, &f->hwaccel_picture_private);
116  if (ret < 0)
117  goto fail;
118 
119  return 0;
120 
121 fail:
122  ff_refstruct_unref(&f->seg_map);
124  return ret;
125 }
126 
128 {
129  ff_refstruct_unref(&f->seg_map);
130  ff_refstruct_unref(&f->hwaccel_picture_private);
132 }
133 
134 static av_cold void vp8_decode_flush_impl(AVCodecContext *avctx, int free_mem)
135 {
136  VP8Context *s = avctx->priv_data;
137  int i;
138 
139  for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++)
140  vp8_release_frame(&s->frames[i]);
141  memset(s->framep, 0, sizeof(s->framep));
142 
143  if (free_mem)
144  free_buffers(s);
145 
146  if (FF_HW_HAS_CB(avctx, flush))
147  FF_HW_SIMPLE_CALL(avctx, flush);
148 }
149 
151 {
152  vp8_decode_flush_impl(avctx, 0);
153 }
154 
156 {
157  VP8Frame *frame = NULL;
158  int i;
159 
160  // find a free buffer
161  for (i = 0; i < 5; i++)
162  if (&s->frames[i] != s->framep[VP8_FRAME_CURRENT] &&
163  &s->frames[i] != s->framep[VP8_FRAME_PREVIOUS] &&
164  &s->frames[i] != s->framep[VP8_FRAME_GOLDEN] &&
165  &s->frames[i] != s->framep[VP8_FRAME_ALTREF]) {
166  frame = &s->frames[i];
167  break;
168  }
169  if (i == 5) {
170  av_log(s->avctx, AV_LOG_FATAL, "Ran out of free frames!\n");
171  abort();
172  }
173  if (frame->tf.f)
175 
176  return frame;
177 }
178 
180 {
181  enum AVPixelFormat pix_fmts[] = {
182 #if CONFIG_VP8_VAAPI_HWACCEL
184 #endif
185 #if CONFIG_VP8_NVDEC_HWACCEL
187 #endif
190  };
191 
192  return ff_get_format(s->avctx, pix_fmts);
193 }
194 
195 static av_always_inline
196 int update_dimensions(VP8Context *s, int width, int height, int is_vp7)
197 {
198  AVCodecContext *avctx = s->avctx;
199  int i, ret, dim_reset = 0;
200 
201  if (width != s->avctx->width || ((width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) && s->macroblocks_base ||
202  height != s->avctx->height) {
203  vp8_decode_flush_impl(s->avctx, 1);
204 
205  ret = ff_set_dimensions(s->avctx, width, height);
206  if (ret < 0)
207  return ret;
208 
209  dim_reset = (s->macroblocks_base != NULL);
210  }
211 
212  if ((s->pix_fmt == AV_PIX_FMT_NONE || dim_reset) &&
213  !s->actually_webp && !is_vp7) {
214  s->pix_fmt = get_pixel_format(s);
215  if (s->pix_fmt < 0)
216  return AVERROR(EINVAL);
217  avctx->pix_fmt = s->pix_fmt;
218  }
219 
220  s->mb_width = (s->avctx->coded_width + 15) / 16;
221  s->mb_height = (s->avctx->coded_height + 15) / 16;
222 
223  s->mb_layout = is_vp7 || avctx->active_thread_type == FF_THREAD_SLICE &&
224  avctx->thread_count > 1;
225  if (!s->mb_layout) { // Frame threading and one thread
226  s->macroblocks_base = av_mallocz((s->mb_width + s->mb_height * 2 + 1) *
227  sizeof(*s->macroblocks));
228  s->intra4x4_pred_mode_top = av_mallocz(s->mb_width * 4);
229  } else // Sliced threading
230  s->macroblocks_base = av_mallocz((s->mb_width + 2) * (s->mb_height + 2) *
231  sizeof(*s->macroblocks));
232  s->top_nnz = av_mallocz(s->mb_width * sizeof(*s->top_nnz));
233  s->top_border = av_mallocz((s->mb_width + 1) * sizeof(*s->top_border));
234  s->thread_data = av_mallocz(MAX_THREADS * sizeof(VP8ThreadData));
235 
236  if (!s->macroblocks_base || !s->top_nnz || !s->top_border ||
237  !s->thread_data || (!s->intra4x4_pred_mode_top && !s->mb_layout)) {
238  free_buffers(s);
239  return AVERROR(ENOMEM);
240  }
241 
242  for (i = 0; i < MAX_THREADS; i++) {
243  s->thread_data[i].filter_strength =
244  av_mallocz(s->mb_width * sizeof(*s->thread_data[0].filter_strength));
245  if (!s->thread_data[i].filter_strength) {
246  free_buffers(s);
247  return AVERROR(ENOMEM);
248  }
249 #if HAVE_THREADS
250  ret = pthread_mutex_init(&s->thread_data[i].lock, NULL);
251  if (ret) {
252  free_buffers(s);
253  return AVERROR(ret);
254  }
255  ret = pthread_cond_init(&s->thread_data[i].cond, NULL);
256  if (ret) {
257  free_buffers(s);
258  return AVERROR(ret);
259  }
260 #endif
261  }
262 
263  s->macroblocks = s->macroblocks_base + 1;
264 
265  return 0;
266 }
267 
269 {
271 }
272 
274 {
276 }
277 
278 
280 {
281  VPXRangeCoder *c = &s->c;
282  int i;
283 
284  s->segmentation.update_map = vp89_rac_get(c);
285  s->segmentation.update_feature_data = vp89_rac_get(c);
286 
287  if (s->segmentation.update_feature_data) {
288  s->segmentation.absolute_vals = vp89_rac_get(c);
289 
290  for (i = 0; i < 4; i++)
291  s->segmentation.base_quant[i] = vp8_rac_get_sint(c, 7);
292 
293  for (i = 0; i < 4; i++)
294  s->segmentation.filter_level[i] = vp8_rac_get_sint(c, 6);
295  }
296  if (s->segmentation.update_map)
297  for (i = 0; i < 3; i++)
298  s->prob->segmentid[i] = vp89_rac_get(c) ? vp89_rac_get_uint(c, 8) : 255;
299 }
300 
302 {
303  VPXRangeCoder *c = &s->c;
304  int i;
305 
306  for (i = 0; i < 4; i++) {
307  if (vp89_rac_get(c)) {
308  s->lf_delta.ref[i] = vp89_rac_get_uint(c, 6);
309 
310  if (vp89_rac_get(c))
311  s->lf_delta.ref[i] = -s->lf_delta.ref[i];
312  }
313  }
314 
315  for (i = MODE_I4x4; i <= VP8_MVMODE_SPLIT; i++) {
316  if (vp89_rac_get(c)) {
317  s->lf_delta.mode[i] = vp89_rac_get_uint(c, 6);
318 
319  if (vp89_rac_get(c))
320  s->lf_delta.mode[i] = -s->lf_delta.mode[i];
321  }
322  }
323 }
324 
325 static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size)
326 {
327  const uint8_t *sizes = buf;
328  int i;
329  int ret;
330 
331  s->num_coeff_partitions = 1 << vp89_rac_get_uint(&s->c, 2);
332 
333  buf += 3 * (s->num_coeff_partitions - 1);
334  buf_size -= 3 * (s->num_coeff_partitions - 1);
335  if (buf_size < 0)
336  return -1;
337 
338  for (i = 0; i < s->num_coeff_partitions - 1; i++) {
339  int size = AV_RL24(sizes + 3 * i);
340  if (buf_size - size < 0)
341  return -1;
342  s->coeff_partition_size[i] = size;
343 
344  ret = ff_vpx_init_range_decoder(&s->coeff_partition[i], buf, size);
345  if (ret < 0)
346  return ret;
347  buf += size;
348  buf_size -= size;
349  }
350 
351  s->coeff_partition_size[i] = buf_size;
352 
353  return ff_vpx_init_range_decoder(&s->coeff_partition[i], buf, buf_size);
354 }
355 
357 {
358  VPXRangeCoder *c = &s->c;
359 
360  int yac_qi = vp89_rac_get_uint(c, 7);
361  int ydc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
362  int y2dc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
363  int y2ac_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
364  int uvdc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
365  int uvac_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi;
366 
367  s->qmat[0].luma_qmul[0] = vp7_ydc_qlookup[ydc_qi];
368  s->qmat[0].luma_qmul[1] = vp7_yac_qlookup[yac_qi];
369  s->qmat[0].luma_dc_qmul[0] = vp7_y2dc_qlookup[y2dc_qi];
370  s->qmat[0].luma_dc_qmul[1] = vp7_y2ac_qlookup[y2ac_qi];
371  s->qmat[0].chroma_qmul[0] = FFMIN(vp7_ydc_qlookup[uvdc_qi], 132);
372  s->qmat[0].chroma_qmul[1] = vp7_yac_qlookup[uvac_qi];
373 }
374 
376 {
377  VPXRangeCoder *c = &s->c;
378  int i, base_qi;
379 
380  s->quant.yac_qi = vp89_rac_get_uint(c, 7);
381  s->quant.ydc_delta = vp8_rac_get_sint(c, 4);
382  s->quant.y2dc_delta = vp8_rac_get_sint(c, 4);
383  s->quant.y2ac_delta = vp8_rac_get_sint(c, 4);
384  s->quant.uvdc_delta = vp8_rac_get_sint(c, 4);
385  s->quant.uvac_delta = vp8_rac_get_sint(c, 4);
386 
387  for (i = 0; i < 4; i++) {
388  if (s->segmentation.enabled) {
389  base_qi = s->segmentation.base_quant[i];
390  if (!s->segmentation.absolute_vals)
391  base_qi += s->quant.yac_qi;
392  } else
393  base_qi = s->quant.yac_qi;
394 
395  s->qmat[i].luma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.ydc_delta, 7)];
396  s->qmat[i].luma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi, 7)];
397  s->qmat[i].luma_dc_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.y2dc_delta, 7)] * 2;
398  /* 101581>>16 is equivalent to 155/100 */
399  s->qmat[i].luma_dc_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi + s->quant.y2ac_delta, 7)] * 101581 >> 16;
400  s->qmat[i].chroma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.uvdc_delta, 7)];
401  s->qmat[i].chroma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi + s->quant.uvac_delta, 7)];
402 
403  s->qmat[i].luma_dc_qmul[1] = FFMAX(s->qmat[i].luma_dc_qmul[1], 8);
404  s->qmat[i].chroma_qmul[0] = FFMIN(s->qmat[i].chroma_qmul[0], 132);
405  }
406 }
407 
408 /**
409  * Determine which buffers golden and altref should be updated with after this frame.
410  * The spec isn't clear here, so I'm going by my understanding of what libvpx does
411  *
412  * Intra frames update all 3 references
413  * Inter frames update VP8_FRAME_PREVIOUS if the update_last flag is set
414  * If the update (golden|altref) flag is set, it's updated with the current frame
415  * if update_last is set, and VP8_FRAME_PREVIOUS otherwise.
416  * If the flag is not set, the number read means:
417  * 0: no update
418  * 1: VP8_FRAME_PREVIOUS
419  * 2: update golden with altref, or update altref with golden
420  */
422 {
423  VPXRangeCoder *c = &s->c;
424 
425  if (update)
426  return VP8_FRAME_CURRENT;
427 
428  switch (vp89_rac_get_uint(c, 2)) {
429  case 1:
430  return VP8_FRAME_PREVIOUS;
431  case 2:
433  }
434  return VP8_FRAME_NONE;
435 }
436 
438 {
439  int i, j;
440  for (i = 0; i < 4; i++)
441  for (j = 0; j < 16; j++)
442  memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]],
443  sizeof(s->prob->token[i][j]));
444 }
445 
447 {
448  VPXRangeCoder *c = &s->c;
449  int i, j, k, l, m;
450 
451  for (i = 0; i < 4; i++)
452  for (j = 0; j < 8; j++)
453  for (k = 0; k < 3; k++)
454  for (l = 0; l < NUM_DCT_TOKENS-1; l++)
456  int prob = vp89_rac_get_uint(c, 8);
457  for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++)
458  s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob;
459  }
460 }
461 
462 #define VP7_MVC_SIZE 17
463 #define VP8_MVC_SIZE 19
464 
466  int mvc_size)
467 {
468  VPXRangeCoder *c = &s->c;
469  int i, j;
470 
471  if (vp89_rac_get(c))
472  for (i = 0; i < 4; i++)
473  s->prob->pred16x16[i] = vp89_rac_get_uint(c, 8);
474  if (vp89_rac_get(c))
475  for (i = 0; i < 3; i++)
476  s->prob->pred8x8c[i] = vp89_rac_get_uint(c, 8);
477 
478  // 17.2 MV probability update
479  for (i = 0; i < 2; i++)
480  for (j = 0; j < mvc_size; j++)
482  s->prob->mvc[i][j] = vp8_rac_get_nn(c);
483 }
484 
485 static void update_refs(VP8Context *s)
486 {
487  VPXRangeCoder *c = &s->c;
488 
489  int update_golden = vp89_rac_get(c);
490  int update_altref = vp89_rac_get(c);
491 
492  s->update_golden = ref_to_update(s, update_golden, VP8_FRAME_GOLDEN);
493  s->update_altref = ref_to_update(s, update_altref, VP8_FRAME_ALTREF);
494 }
495 
496 static void copy_chroma(AVFrame *dst, const AVFrame *src, int width, int height)
497 {
498  int i, j;
499 
500  for (j = 1; j < 3; j++) {
501  for (i = 0; i < height / 2; i++)
502  memcpy(dst->data[j] + i * dst->linesize[j],
503  src->data[j] + i * src->linesize[j], width / 2);
504  }
505 }
506 
507 static void fade(uint8_t *dst, ptrdiff_t dst_linesize,
508  const uint8_t *src, ptrdiff_t src_linesize,
509  int width, int height,
510  int alpha, int beta)
511 {
512  int i, j;
513  for (j = 0; j < height; j++) {
514  const uint8_t *src2 = src + j * src_linesize;
515  uint8_t *dst2 = dst + j * dst_linesize;
516  for (i = 0; i < width; i++) {
517  uint8_t y = src2[i];
518  dst2[i] = av_clip_uint8(y + ((y * beta) >> 8) + alpha);
519  }
520  }
521 }
522 
523 static int vp7_fade_frame(VP8Context *s, int alpha, int beta)
524 {
525  int ret;
526 
527  if (!s->keyframe && (alpha || beta)) {
528  int width = s->mb_width * 16;
529  int height = s->mb_height * 16;
530  const AVFrame *src;
531  AVFrame *dst;
532 
533  if (!s->framep[VP8_FRAME_PREVIOUS] ||
534  !s->framep[VP8_FRAME_GOLDEN]) {
535  av_log(s->avctx, AV_LOG_WARNING, "Discarding interframe without a prior keyframe!\n");
536  return AVERROR_INVALIDDATA;
537  }
538 
539  src =
540  dst = s->framep[VP8_FRAME_PREVIOUS]->tf.f;
541 
542  /* preserve the golden frame, write a new previous frame */
543  if (s->framep[VP8_FRAME_GOLDEN] == s->framep[VP8_FRAME_PREVIOUS]) {
545  if ((ret = vp8_alloc_frame(s, s->framep[VP8_FRAME_PREVIOUS], 1)) < 0)
546  return ret;
547 
548  dst = s->framep[VP8_FRAME_PREVIOUS]->tf.f;
549 
551  }
552 
553  fade(dst->data[0], dst->linesize[0],
554  src->data[0], src->linesize[0],
555  width, height, alpha, beta);
556  }
557 
558  return 0;
559 }
560 
561 static int vp7_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
562 {
563  VPXRangeCoder *c = &s->c;
564  int part1_size, hscale, vscale, i, j, ret;
565  int width = s->avctx->width;
566  int height = s->avctx->height;
567  int alpha = 0;
568  int beta = 0;
569  int fade_present = 1;
570 
571  if (buf_size < 4) {
572  return AVERROR_INVALIDDATA;
573  }
574 
575  s->profile = (buf[0] >> 1) & 7;
576  if (s->profile > 1) {
577  avpriv_request_sample(s->avctx, "Unknown profile %d", s->profile);
578  return AVERROR_INVALIDDATA;
579  }
580 
581  s->keyframe = !(buf[0] & 1);
582  s->invisible = 0;
583  part1_size = AV_RL24(buf) >> 4;
584 
585  if (buf_size < 4 - s->profile + part1_size) {
586  av_log(s->avctx, AV_LOG_ERROR, "Buffer size %d is too small, needed : %d\n", buf_size, 4 - s->profile + part1_size);
587  return AVERROR_INVALIDDATA;
588  }
589 
590  buf += 4 - s->profile;
591  buf_size -= 4 - s->profile;
592 
593  memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, sizeof(s->put_pixels_tab));
594 
595  ret = ff_vpx_init_range_decoder(c, buf, part1_size);
596  if (ret < 0)
597  return ret;
598  buf += part1_size;
599  buf_size -= part1_size;
600 
601  /* A. Dimension information (keyframes only) */
602  if (s->keyframe) {
603  width = vp89_rac_get_uint(c, 12);
604  height = vp89_rac_get_uint(c, 12);
605  hscale = vp89_rac_get_uint(c, 2);
606  vscale = vp89_rac_get_uint(c, 2);
607  if (hscale || vscale)
608  avpriv_request_sample(s->avctx, "Upscaling");
609 
610  s->update_golden = s->update_altref = VP8_FRAME_CURRENT;
612  memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter,
613  sizeof(s->prob->pred16x16));
614  memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter,
615  sizeof(s->prob->pred8x8c));
616  for (i = 0; i < 2; i++)
617  memcpy(s->prob->mvc[i], vp7_mv_default_prob[i],
618  sizeof(vp7_mv_default_prob[i]));
619  memset(&s->segmentation, 0, sizeof(s->segmentation));
620  memset(&s->lf_delta, 0, sizeof(s->lf_delta));
621  memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan));
622  }
623 
624  if (s->keyframe || s->profile > 0)
625  memset(s->inter_dc_pred, 0 , sizeof(s->inter_dc_pred));
626 
627  /* B. Decoding information for all four macroblock-level features */
628  for (i = 0; i < 4; i++) {
629  s->feature_enabled[i] = vp89_rac_get(c);
630  if (s->feature_enabled[i]) {
631  s->feature_present_prob[i] = vp89_rac_get_uint(c, 8);
632 
633  for (j = 0; j < 3; j++)
634  s->feature_index_prob[i][j] =
635  vp89_rac_get(c) ? vp89_rac_get_uint(c, 8) : 255;
636 
637  if (vp7_feature_value_size[s->profile][i])
638  for (j = 0; j < 4; j++)
639  s->feature_value[i][j] =
641  }
642  }
643 
644  s->segmentation.enabled = 0;
645  s->segmentation.update_map = 0;
646  s->lf_delta.enabled = 0;
647 
648  s->num_coeff_partitions = 1;
649  ret = ff_vpx_init_range_decoder(&s->coeff_partition[0], buf, buf_size);
650  if (ret < 0)
651  return ret;
652 
653  if (!s->macroblocks_base || /* first frame */
654  width != s->avctx->width || height != s->avctx->height ||
655  (width + 15) / 16 != s->mb_width || (height + 15) / 16 != s->mb_height) {
656  if ((ret = vp7_update_dimensions(s, width, height)) < 0)
657  return ret;
658  }
659 
660  /* C. Dequantization indices */
661  vp7_get_quants(s);
662 
663  /* D. Golden frame update flag (a Flag) for interframes only */
664  if (!s->keyframe) {
665  s->update_golden = vp89_rac_get(c) ? VP8_FRAME_CURRENT : VP8_FRAME_NONE;
666  s->sign_bias[VP8_FRAME_GOLDEN] = 0;
667  }
668 
669  s->update_last = 1;
670  s->update_probabilities = 1;
671 
672  if (s->profile > 0) {
673  s->update_probabilities = vp89_rac_get(c);
674  if (!s->update_probabilities)
675  s->prob[1] = s->prob[0];
676 
677  if (!s->keyframe)
678  fade_present = vp89_rac_get(c);
679  }
680 
681  if (vpx_rac_is_end(c))
682  return AVERROR_INVALIDDATA;
683  /* E. Fading information for previous frame */
684  if (fade_present && vp89_rac_get(c)) {
685  alpha = (int8_t) vp89_rac_get_uint(c, 8);
686  beta = (int8_t) vp89_rac_get_uint(c, 8);
687  }
688 
689  /* F. Loop filter type */
690  if (!s->profile)
691  s->filter.simple = vp89_rac_get(c);
692 
693  /* G. DCT coefficient ordering specification */
694  if (vp89_rac_get(c))
695  for (i = 1; i < 16; i++)
696  s->prob[0].scan[i] = ff_zigzag_scan[vp89_rac_get_uint(c, 4)];
697 
698  /* H. Loop filter levels */
699  if (s->profile > 0)
700  s->filter.simple = vp89_rac_get(c);
701  s->filter.level = vp89_rac_get_uint(c, 6);
702  s->filter.sharpness = vp89_rac_get_uint(c, 3);
703 
704  /* I. DCT coefficient probability update; 13.3 Token Probability Updates */
706 
707  s->mbskip_enabled = 0;
708 
709  /* J. The remaining frame header data occurs ONLY FOR INTERFRAMES */
710  if (!s->keyframe) {
711  s->prob->intra = vp89_rac_get_uint(c, 8);
712  s->prob->last = vp89_rac_get_uint(c, 8);
714  }
715 
716  if (vpx_rac_is_end(c))
717  return AVERROR_INVALIDDATA;
718 
719  if ((ret = vp7_fade_frame(s, alpha, beta)) < 0)
720  return ret;
721 
722  return 0;
723 }
724 
725 static int vp8_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
726 {
727  VPXRangeCoder *c = &s->c;
728  int header_size, hscale, vscale, ret;
729  int width = s->avctx->width;
730  int height = s->avctx->height;
731 
732  if (buf_size < 3) {
733  av_log(s->avctx, AV_LOG_ERROR, "Insufficent data (%d) for header\n", buf_size);
734  return AVERROR_INVALIDDATA;
735  }
736 
737  s->keyframe = !(buf[0] & 1);
738  s->profile = (buf[0]>>1) & 7;
739  s->invisible = !(buf[0] & 0x10);
740  header_size = AV_RL24(buf) >> 5;
741  buf += 3;
742  buf_size -= 3;
743 
744  s->header_partition_size = header_size;
745 
746  if (s->profile > 3)
747  av_log(s->avctx, AV_LOG_WARNING, "Unknown profile %d\n", s->profile);
748 
749  if (!s->profile)
750  memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab,
751  sizeof(s->put_pixels_tab));
752  else // profile 1-3 use bilinear, 4+ aren't defined so whatever
753  memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_bilinear_pixels_tab,
754  sizeof(s->put_pixels_tab));
755 
756  if (header_size > buf_size - 7 * s->keyframe) {
757  av_log(s->avctx, AV_LOG_ERROR, "Header size larger than data provided\n");
758  return AVERROR_INVALIDDATA;
759  }
760 
761  if (s->keyframe) {
762  if (AV_RL24(buf) != 0x2a019d) {
763  av_log(s->avctx, AV_LOG_ERROR,
764  "Invalid start code 0x%x\n", AV_RL24(buf));
765  return AVERROR_INVALIDDATA;
766  }
767  width = AV_RL16(buf + 3) & 0x3fff;
768  height = AV_RL16(buf + 5) & 0x3fff;
769  hscale = buf[4] >> 6;
770  vscale = buf[6] >> 6;
771  buf += 7;
772  buf_size -= 7;
773 
774  if (hscale || vscale)
775  avpriv_request_sample(s->avctx, "Upscaling");
776 
777  s->update_golden = s->update_altref = VP8_FRAME_CURRENT;
779  memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter,
780  sizeof(s->prob->pred16x16));
781  memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter,
782  sizeof(s->prob->pred8x8c));
783  memcpy(s->prob->mvc, vp8_mv_default_prob,
784  sizeof(s->prob->mvc));
785  memset(&s->segmentation, 0, sizeof(s->segmentation));
786  memset(&s->lf_delta, 0, sizeof(s->lf_delta));
787  }
788 
789  ret = ff_vpx_init_range_decoder(c, buf, header_size);
790  if (ret < 0)
791  return ret;
792  buf += header_size;
793  buf_size -= header_size;
794 
795  if (s->keyframe) {
796  s->colorspace = vp89_rac_get(c);
797  if (s->colorspace)
798  av_log(s->avctx, AV_LOG_WARNING, "Unspecified colorspace\n");
799  s->fullrange = vp89_rac_get(c);
800  }
801 
802  if ((s->segmentation.enabled = vp89_rac_get(c)))
804  else
805  s->segmentation.update_map = 0; // FIXME: move this to some init function?
806 
807  s->filter.simple = vp89_rac_get(c);
808  s->filter.level = vp89_rac_get_uint(c, 6);
809  s->filter.sharpness = vp89_rac_get_uint(c, 3);
810 
811  if ((s->lf_delta.enabled = vp89_rac_get(c))) {
812  s->lf_delta.update = vp89_rac_get(c);
813  if (s->lf_delta.update)
815  }
816 
817  if (setup_partitions(s, buf, buf_size)) {
818  av_log(s->avctx, AV_LOG_ERROR, "Invalid partitions\n");
819  return AVERROR_INVALIDDATA;
820  }
821 
822  if (!s->macroblocks_base || /* first frame */
823  width != s->avctx->width || height != s->avctx->height ||
824  (width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height)
825  if ((ret = vp8_update_dimensions(s, width, height)) < 0)
826  return ret;
827 
828  vp8_get_quants(s);
829 
830  if (!s->keyframe) {
831  update_refs(s);
832  s->sign_bias[VP8_FRAME_GOLDEN] = vp89_rac_get(c);
833  s->sign_bias[VP8_FRAME_ALTREF] = vp89_rac_get(c);
834  }
835 
836  // if we aren't saving this frame's probabilities for future frames,
837  // make a copy of the current probabilities
838  if (!(s->update_probabilities = vp89_rac_get(c)))
839  s->prob[1] = s->prob[0];
840 
841  s->update_last = s->keyframe || vp89_rac_get(c);
842 
844 
845  if ((s->mbskip_enabled = vp89_rac_get(c)))
846  s->prob->mbskip = vp89_rac_get_uint(c, 8);
847 
848  if (!s->keyframe) {
849  s->prob->intra = vp89_rac_get_uint(c, 8);
850  s->prob->last = vp89_rac_get_uint(c, 8);
851  s->prob->golden = vp89_rac_get_uint(c, 8);
853  }
854 
855  // Record the entropy coder state here so that hwaccels can use it.
856  s->c.code_word = vpx_rac_renorm(&s->c);
857  s->coder_state_at_header_end.input = s->c.buffer - (-s->c.bits / 8);
858  s->coder_state_at_header_end.range = s->c.high;
859  s->coder_state_at_header_end.value = s->c.code_word >> 16;
860  s->coder_state_at_header_end.bit_count = -s->c.bits % 8;
861 
862  return 0;
863 }
864 
865 static av_always_inline
866 void clamp_mv(const VP8mvbounds *s, VP8mv *dst, const VP8mv *src)
867 {
868  dst->x = av_clip(src->x, av_clip(s->mv_min.x, INT16_MIN, INT16_MAX),
869  av_clip(s->mv_max.x, INT16_MIN, INT16_MAX));
870  dst->y = av_clip(src->y, av_clip(s->mv_min.y, INT16_MIN, INT16_MAX),
871  av_clip(s->mv_max.y, INT16_MIN, INT16_MAX));
872 }
873 
874 /**
875  * Motion vector coding, 17.1.
876  */
877 static av_always_inline int read_mv_component(VPXRangeCoder *c, const uint8_t *p, int vp7)
878 {
879  int bit, x = 0;
880 
881  if (vpx_rac_get_prob_branchy(c, p[0])) {
882  int i;
883 
884  for (i = 0; i < 3; i++)
885  x += vpx_rac_get_prob(c, p[9 + i]) << i;
886  for (i = (vp7 ? 7 : 9); i > 3; i--)
887  x += vpx_rac_get_prob(c, p[9 + i]) << i;
888  if (!(x & (vp7 ? 0xF0 : 0xFFF0)) || vpx_rac_get_prob(c, p[12]))
889  x += 8;
890  } else {
891  // small_mvtree
892  const uint8_t *ps = p + 2;
893  bit = vpx_rac_get_prob(c, *ps);
894  ps += 1 + 3 * bit;
895  x += 4 * bit;
896  bit = vpx_rac_get_prob(c, *ps);
897  ps += 1 + bit;
898  x += 2 * bit;
899  x += vpx_rac_get_prob(c, *ps);
900  }
901 
902  return (x && vpx_rac_get_prob(c, p[1])) ? -x : x;
903 }
904 
905 static int vp7_read_mv_component(VPXRangeCoder *c, const uint8_t *p)
906 {
907  return read_mv_component(c, p, 1);
908 }
909 
910 static int vp8_read_mv_component(VPXRangeCoder *c, const uint8_t *p)
911 {
912  return read_mv_component(c, p, 0);
913 }
914 
915 static av_always_inline
916 const uint8_t *get_submv_prob(uint32_t left, uint32_t top, int is_vp7)
917 {
918  if (is_vp7)
919  return vp7_submv_prob;
920 
921  if (left == top)
922  return vp8_submv_prob[4 - !!left];
923  if (!top)
924  return vp8_submv_prob[2];
925  return vp8_submv_prob[1 - !!left];
926 }
927 
928 /**
929  * Split motion vector prediction, 16.4.
930  * @returns the number of motion vectors parsed (2, 4 or 16)
931  */
932 static av_always_inline
934  int layout, int is_vp7)
935 {
936  int part_idx;
937  int n, num;
938  const VP8Macroblock *top_mb;
939  const VP8Macroblock *left_mb = &mb[-1];
940  const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning];
941  const uint8_t *mbsplits_top, *mbsplits_cur, *firstidx;
942  const VP8mv *top_mv;
943  const VP8mv *left_mv = left_mb->bmv;
944  const VP8mv *cur_mv = mb->bmv;
945 
946  if (!layout) // layout is inlined, s->mb_layout is not
947  top_mb = &mb[2];
948  else
949  top_mb = &mb[-s->mb_width - 1];
950  mbsplits_top = vp8_mbsplits[top_mb->partitioning];
951  top_mv = top_mb->bmv;
952 
956  else
957  part_idx = VP8_SPLITMVMODE_8x8;
958  } else {
959  part_idx = VP8_SPLITMVMODE_4x4;
960  }
961 
962  num = vp8_mbsplit_count[part_idx];
963  mbsplits_cur = vp8_mbsplits[part_idx],
964  firstidx = vp8_mbfirstidx[part_idx];
965  mb->partitioning = part_idx;
966 
967  for (n = 0; n < num; n++) {
968  int k = firstidx[n];
969  uint32_t left, above;
970  const uint8_t *submv_prob;
971 
972  if (!(k & 3))
973  left = AV_RN32A(&left_mv[mbsplits_left[k + 3]]);
974  else
975  left = AV_RN32A(&cur_mv[mbsplits_cur[k - 1]]);
976  if (k <= 3)
977  above = AV_RN32A(&top_mv[mbsplits_top[k + 12]]);
978  else
979  above = AV_RN32A(&cur_mv[mbsplits_cur[k - 4]]);
980 
981  submv_prob = get_submv_prob(left, above, is_vp7);
982 
983  if (vpx_rac_get_prob_branchy(c, submv_prob[0])) {
984  if (vpx_rac_get_prob_branchy(c, submv_prob[1])) {
985  if (vpx_rac_get_prob_branchy(c, submv_prob[2])) {
986  mb->bmv[n].y = mb->mv.y +
987  read_mv_component(c, s->prob->mvc[0], is_vp7);
988  mb->bmv[n].x = mb->mv.x +
989  read_mv_component(c, s->prob->mvc[1], is_vp7);
990  } else {
991  AV_ZERO32(&mb->bmv[n]);
992  }
993  } else {
994  AV_WN32A(&mb->bmv[n], above);
995  }
996  } else {
997  AV_WN32A(&mb->bmv[n], left);
998  }
999  }
1000 
1001  return num;
1002 }
1003 
1004 /**
1005  * The vp7 reference decoder uses a padding macroblock column (added to right
1006  * edge of the frame) to guard against illegal macroblock offsets. The
1007  * algorithm has bugs that permit offsets to straddle the padding column.
1008  * This function replicates those bugs.
1009  *
1010  * @param[out] edge_x macroblock x address
1011  * @param[out] edge_y macroblock y address
1012  *
1013  * @return macroblock offset legal (boolean)
1014  */
1015 static int vp7_calculate_mb_offset(int mb_x, int mb_y, int mb_width,
1016  int xoffset, int yoffset, int boundary,
1017  int *edge_x, int *edge_y)
1018 {
1019  int vwidth = mb_width + 1;
1020  int new = (mb_y + yoffset) * vwidth + mb_x + xoffset;
1021  if (new < boundary || new % vwidth == vwidth - 1)
1022  return 0;
1023  *edge_y = new / vwidth;
1024  *edge_x = new % vwidth;
1025  return 1;
1026 }
1027 
1028 static const VP8mv *get_bmv_ptr(const VP8Macroblock *mb, int subblock)
1029 {
1030  return &mb->bmv[mb->mode == VP8_MVMODE_SPLIT ? vp8_mbsplits[mb->partitioning][subblock] : 0];
1031 }
1032 
1033 static av_always_inline
1035  int mb_x, int mb_y, int layout)
1036 {
1037  enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR };
1038  enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
1039  int idx = CNT_ZERO;
1040  VP8mv near_mv[3];
1041  uint8_t cnt[3] = { 0 };
1042  VPXRangeCoder *c = &s->c;
1043  int i;
1044 
1045  AV_ZERO32(&near_mv[0]);
1046  AV_ZERO32(&near_mv[1]);
1047  AV_ZERO32(&near_mv[2]);
1048 
1049  for (i = 0; i < VP7_MV_PRED_COUNT; i++) {
1050  const VP7MVPred * pred = &vp7_mv_pred[i];
1051  int edge_x, edge_y;
1052 
1053  if (vp7_calculate_mb_offset(mb_x, mb_y, s->mb_width, pred->xoffset,
1054  pred->yoffset, !s->profile, &edge_x, &edge_y)) {
1055  const VP8Macroblock *edge = (s->mb_layout == 1)
1056  ? s->macroblocks_base + 1 + edge_x +
1057  (s->mb_width + 1) * (edge_y + 1)
1058  : s->macroblocks + edge_x +
1059  (s->mb_height - edge_y - 1) * 2;
1060  uint32_t mv = AV_RN32A(get_bmv_ptr(edge, vp7_mv_pred[i].subblock));
1061  if (mv) {
1062  if (AV_RN32A(&near_mv[CNT_NEAREST])) {
1063  if (mv == AV_RN32A(&near_mv[CNT_NEAREST])) {
1064  idx = CNT_NEAREST;
1065  } else if (AV_RN32A(&near_mv[CNT_NEAR])) {
1066  if (mv != AV_RN32A(&near_mv[CNT_NEAR]))
1067  continue;
1068  idx = CNT_NEAR;
1069  } else {
1070  AV_WN32A(&near_mv[CNT_NEAR], mv);
1071  idx = CNT_NEAR;
1072  }
1073  } else {
1074  AV_WN32A(&near_mv[CNT_NEAREST], mv);
1075  idx = CNT_NEAREST;
1076  }
1077  } else {
1078  idx = CNT_ZERO;
1079  }
1080  } else {
1081  idx = CNT_ZERO;
1082  }
1083  cnt[idx] += vp7_mv_pred[i].score;
1084  }
1085 
1086  mb->partitioning = VP8_SPLITMVMODE_NONE;
1087 
1088  if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_ZERO]][0])) {
1089  mb->mode = VP8_MVMODE_MV;
1090 
1091  if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAREST]][1])) {
1092 
1093  if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][2])) {
1094 
1095  if (cnt[CNT_NEAREST] > cnt[CNT_NEAR])
1096  AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAREST] ? 0 : AV_RN32A(&near_mv[CNT_NEAREST]));
1097  else
1098  AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAR] ? 0 : AV_RN32A(&near_mv[CNT_NEAR]));
1099 
1100  if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][3])) {
1101  mb->mode = VP8_MVMODE_SPLIT;
1102  mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP7) - 1];
1103  } else {
1104  mb->mv.y += vp7_read_mv_component(c, s->prob->mvc[0]);
1105  mb->mv.x += vp7_read_mv_component(c, s->prob->mvc[1]);
1106  mb->bmv[0] = mb->mv;
1107  }
1108  } else {
1109  mb->mv = near_mv[CNT_NEAR];
1110  mb->bmv[0] = mb->mv;
1111  }
1112  } else {
1113  mb->mv = near_mv[CNT_NEAREST];
1114  mb->bmv[0] = mb->mv;
1115  }
1116  } else {
1117  mb->mode = VP8_MVMODE_ZERO;
1118  AV_ZERO32(&mb->mv);
1119  mb->bmv[0] = mb->mv;
1120  }
1121 }
1122 
1123 static av_always_inline
1125  int mb_x, int mb_y, int layout)
1126 {
1127  VP8Macroblock *mb_edge[3] = { 0 /* top */,
1128  mb - 1 /* left */,
1129  0 /* top-left */ };
1130  enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV };
1131  enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
1132  int idx = CNT_ZERO;
1133  int cur_sign_bias = s->sign_bias[mb->ref_frame];
1134  const int8_t *sign_bias = s->sign_bias;
1135  VP8mv near_mv[4];
1136  uint8_t cnt[4] = { 0 };
1137  VPXRangeCoder *c = &s->c;
1138 
1139  if (!layout) { // layout is inlined (s->mb_layout is not)
1140  mb_edge[0] = mb + 2;
1141  mb_edge[2] = mb + 1;
1142  } else {
1143  mb_edge[0] = mb - s->mb_width - 1;
1144  mb_edge[2] = mb - s->mb_width - 2;
1145  }
1146 
1147  AV_ZERO32(&near_mv[0]);
1148  AV_ZERO32(&near_mv[1]);
1149  AV_ZERO32(&near_mv[2]);
1150 
1151  /* Process MB on top, left and top-left */
1152 #define MV_EDGE_CHECK(n) \
1153  { \
1154  const VP8Macroblock *edge = mb_edge[n]; \
1155  int edge_ref = edge->ref_frame; \
1156  if (edge_ref != VP8_FRAME_CURRENT) { \
1157  uint32_t mv = AV_RN32A(&edge->mv); \
1158  if (mv) { \
1159  if (cur_sign_bias != sign_bias[edge_ref]) { \
1160  /* SWAR negate of the values in mv. */ \
1161  mv = ~mv; \
1162  mv = ((mv & 0x7fff7fff) + \
1163  0x00010001) ^ (mv & 0x80008000); \
1164  } \
1165  if (!n || mv != AV_RN32A(&near_mv[idx])) \
1166  AV_WN32A(&near_mv[++idx], mv); \
1167  cnt[idx] += 1 + (n != 2); \
1168  } else \
1169  cnt[CNT_ZERO] += 1 + (n != 2); \
1170  } \
1171  }
1172 
1173  MV_EDGE_CHECK(0)
1174  MV_EDGE_CHECK(1)
1175  MV_EDGE_CHECK(2)
1176 
1177  mb->partitioning = VP8_SPLITMVMODE_NONE;
1178  if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_ZERO]][0])) {
1179  mb->mode = VP8_MVMODE_MV;
1180 
1181  /* If we have three distinct MVs, merge first and last if they're the same */
1182  if (cnt[CNT_SPLITMV] &&
1183  AV_RN32A(&near_mv[1 + VP8_EDGE_TOP]) == AV_RN32A(&near_mv[1 + VP8_EDGE_TOPLEFT]))
1184  cnt[CNT_NEAREST] += 1;
1185 
1186  /* Swap near and nearest if necessary */
1187  if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) {
1188  FFSWAP(uint8_t, cnt[CNT_NEAREST], cnt[CNT_NEAR]);
1189  FFSWAP(VP8mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR]);
1190  }
1191 
1192  if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAREST]][1])) {
1193  if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAR]][2])) {
1194  /* Choose the best mv out of 0,0 and the nearest mv */
1195  clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]);
1196  cnt[CNT_SPLITMV] = ((mb_edge[VP8_EDGE_LEFT]->mode == VP8_MVMODE_SPLIT) +
1197  (mb_edge[VP8_EDGE_TOP]->mode == VP8_MVMODE_SPLIT)) * 2 +
1198  (mb_edge[VP8_EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT);
1199 
1200  if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_SPLITMV]][3])) {
1201  mb->mode = VP8_MVMODE_SPLIT;
1202  mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP8) - 1];
1203  } else {
1204  mb->mv.y += vp8_read_mv_component(c, s->prob->mvc[0]);
1205  mb->mv.x += vp8_read_mv_component(c, s->prob->mvc[1]);
1206  mb->bmv[0] = mb->mv;
1207  }
1208  } else {
1209  clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAR]);
1210  mb->bmv[0] = mb->mv;
1211  }
1212  } else {
1213  clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAREST]);
1214  mb->bmv[0] = mb->mv;
1215  }
1216  } else {
1217  mb->mode = VP8_MVMODE_ZERO;
1218  AV_ZERO32(&mb->mv);
1219  mb->bmv[0] = mb->mv;
1220  }
1221 }
1222 
1223 static av_always_inline
1225  int mb_x, int keyframe, int layout)
1226 {
1227  uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
1228 
1229  if (layout) {
1230  VP8Macroblock *mb_top = mb - s->mb_width - 1;
1231  memcpy(mb->intra4x4_pred_mode_top, mb_top->intra4x4_pred_mode_top, 4);
1232  }
1233  if (keyframe) {
1234  int x, y;
1235  uint8_t *top;
1236  uint8_t *const left = s->intra4x4_pred_mode_left;
1237  if (layout)
1238  top = mb->intra4x4_pred_mode_top;
1239  else
1240  top = s->intra4x4_pred_mode_top + 4 * mb_x;
1241  for (y = 0; y < 4; y++) {
1242  for (x = 0; x < 4; x++) {
1243  const uint8_t *ctx;
1244  ctx = vp8_pred4x4_prob_intra[top[x]][left[y]];
1245  *intra4x4 = vp89_rac_get_tree(c, vp8_pred4x4_tree, ctx);
1246  left[y] = top[x] = *intra4x4;
1247  intra4x4++;
1248  }
1249  }
1250  } else {
1251  int i;
1252  for (i = 0; i < 16; i++)
1253  intra4x4[i] = vp89_rac_get_tree(c, vp8_pred4x4_tree,
1255  }
1256 }
1257 
1258 static av_always_inline
1259 void decode_mb_mode(VP8Context *s, const VP8mvbounds *mv_bounds,
1260  VP8Macroblock *mb, int mb_x, int mb_y,
1261  uint8_t *segment, const uint8_t *ref, int layout, int is_vp7)
1262 {
1263  VPXRangeCoder *c = &s->c;
1264  static const char * const vp7_feature_name[] = { "q-index",
1265  "lf-delta",
1266  "partial-golden-update",
1267  "blit-pitch" };
1268  if (is_vp7) {
1269  int i;
1270  *segment = 0;
1271  for (i = 0; i < 4; i++) {
1272  if (s->feature_enabled[i]) {
1273  if (vpx_rac_get_prob_branchy(c, s->feature_present_prob[i])) {
1275  s->feature_index_prob[i]);
1276  av_log(s->avctx, AV_LOG_WARNING,
1277  "Feature %s present in macroblock (value 0x%x)\n",
1278  vp7_feature_name[i], s->feature_value[i][index]);
1279  }
1280  }
1281  }
1282  } else if (s->segmentation.update_map) {
1283  int bit = vpx_rac_get_prob(c, s->prob->segmentid[0]);
1284  *segment = vpx_rac_get_prob(c, s->prob->segmentid[1+bit]) + 2*bit;
1285  } else if (s->segmentation.enabled)
1286  *segment = ref ? *ref : *segment;
1287  mb->segment = *segment;
1288 
1289  mb->skip = s->mbskip_enabled ? vpx_rac_get_prob(c, s->prob->mbskip) : 0;
1290 
1291  if (s->keyframe) {
1294 
1295  if (mb->mode == MODE_I4x4) {
1296  decode_intra4x4_modes(s, c, mb, mb_x, 1, layout);
1297  } else {
1298  const uint32_t modes = (is_vp7 ? vp7_pred4x4_mode
1299  : vp8_pred4x4_mode)[mb->mode] * 0x01010101u;
1300  if (s->mb_layout)
1301  AV_WN32A(mb->intra4x4_pred_mode_top, modes);
1302  else
1303  AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes);
1304  AV_WN32A(s->intra4x4_pred_mode_left, modes);
1305  }
1306 
1307  mb->chroma_pred_mode = vp89_rac_get_tree(c, vp8_pred8x8c_tree,
1309  mb->ref_frame = VP8_FRAME_CURRENT;
1310  } else if (vpx_rac_get_prob_branchy(c, s->prob->intra)) {
1311  // inter MB, 16.2
1312  if (vpx_rac_get_prob_branchy(c, s->prob->last))
1313  mb->ref_frame =
1314  (!is_vp7 && vpx_rac_get_prob(c, s->prob->golden)) ? VP8_FRAME_ALTREF
1315  : VP8_FRAME_GOLDEN;
1316  else
1317  mb->ref_frame = VP8_FRAME_PREVIOUS;
1318  s->ref_count[mb->ref_frame - 1]++;
1319 
1320  // motion vectors, 16.3
1321  if (is_vp7)
1322  vp7_decode_mvs(s, mb, mb_x, mb_y, layout);
1323  else
1324  vp8_decode_mvs(s, mv_bounds, mb, mb_x, mb_y, layout);
1325  } else {
1326  // intra MB, 16.1
1328  s->prob->pred16x16);
1329 
1330  if (mb->mode == MODE_I4x4)
1331  decode_intra4x4_modes(s, c, mb, mb_x, 0, layout);
1332 
1333  mb->chroma_pred_mode = vp89_rac_get_tree(c, vp8_pred8x8c_tree,
1334  s->prob->pred8x8c);
1335  mb->ref_frame = VP8_FRAME_CURRENT;
1336  mb->partitioning = VP8_SPLITMVMODE_NONE;
1337  AV_ZERO32(&mb->bmv[0]);
1338  }
1339 }
1340 
1341 /**
1342  * @param r arithmetic bitstream reader context
1343  * @param block destination for block coefficients
1344  * @param probs probabilities to use when reading trees from the bitstream
1345  * @param i initial coeff index, 0 unless a separate DC block is coded
1346  * @param qmul array holding the dc/ac dequant factor at position 0/1
1347  *
1348  * @return 0 if no coeffs were decoded
1349  * otherwise, the index of the last coeff decoded plus one
1350  */
1351 static av_always_inline
1353  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
1354  int i, const uint8_t *token_prob, const int16_t qmul[2],
1355  const uint8_t scan[16], int vp7)
1356 {
1357  VPXRangeCoder c = *r;
1358  goto skip_eob;
1359  do {
1360  int coeff;
1361 restart:
1362  if (!vpx_rac_get_prob_branchy(&c, token_prob[0])) // DCT_EOB
1363  break;
1364 
1365 skip_eob:
1366  if (!vpx_rac_get_prob_branchy(&c, token_prob[1])) { // DCT_0
1367  if (++i == 16)
1368  break; // invalid input; blocks should end with EOB
1369  token_prob = probs[i][0];
1370  if (vp7)
1371  goto restart;
1372  goto skip_eob;
1373  }
1374 
1375  if (!vpx_rac_get_prob_branchy(&c, token_prob[2])) { // DCT_1
1376  coeff = 1;
1377  token_prob = probs[i + 1][1];
1378  } else {
1379  if (!vpx_rac_get_prob_branchy(&c, token_prob[3])) { // DCT 2,3,4
1380  coeff = vpx_rac_get_prob_branchy(&c, token_prob[4]);
1381  if (coeff)
1382  coeff += vpx_rac_get_prob(&c, token_prob[5]);
1383  coeff += 2;
1384  } else {
1385  // DCT_CAT*
1386  if (!vpx_rac_get_prob_branchy(&c, token_prob[6])) {
1387  if (!vpx_rac_get_prob_branchy(&c, token_prob[7])) { // DCT_CAT1
1389  } else { // DCT_CAT2
1390  coeff = 7;
1391  coeff += vpx_rac_get_prob(&c, vp8_dct_cat2_prob[0]) << 1;
1393  }
1394  } else { // DCT_CAT3 and up
1395  int a = vpx_rac_get_prob(&c, token_prob[8]);
1396  int b = vpx_rac_get_prob(&c, token_prob[9 + a]);
1397  int cat = (a << 1) + b;
1398  coeff = 3 + (8 << cat);
1400  }
1401  }
1402  token_prob = probs[i + 1][2];
1403  }
1404  block[scan[i]] = (vp89_rac_get(&c) ? -coeff : coeff) * qmul[!!i];
1405  } while (++i < 16);
1406 
1407  *r = c;
1408  return i;
1409 }
1410 
1411 static av_always_inline
1412 int inter_predict_dc(int16_t block[16], int16_t pred[2])
1413 {
1414  int16_t dc = block[0];
1415  int ret = 0;
1416 
1417  if (pred[1] > 3) {
1418  dc += pred[0];
1419  ret = 1;
1420  }
1421 
1422  if (!pred[0] | !dc | ((int32_t)pred[0] ^ (int32_t)dc) >> 31) {
1423  block[0] = pred[0] = dc;
1424  pred[1] = 0;
1425  } else {
1426  if (pred[0] == dc)
1427  pred[1]++;
1428  block[0] = pred[0] = dc;
1429  }
1430 
1431  return ret;
1432 }
1433 
1435  int16_t block[16],
1436  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
1437  int i, const uint8_t *token_prob,
1438  const int16_t qmul[2],
1439  const uint8_t scan[16])
1440 {
1441  return decode_block_coeffs_internal(r, block, probs, i,
1442  token_prob, qmul, scan, IS_VP7);
1443 }
1444 
1445 #ifndef vp8_decode_block_coeffs_internal
1447  int16_t block[16],
1448  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
1449  int i, const uint8_t *token_prob,
1450  const int16_t qmul[2])
1451 {
1452  return decode_block_coeffs_internal(r, block, probs, i,
1453  token_prob, qmul, ff_zigzag_scan, IS_VP8);
1454 }
1455 #endif
1456 
1457 /**
1458  * @param c arithmetic bitstream reader context
1459  * @param block destination for block coefficients
1460  * @param probs probabilities to use when reading trees from the bitstream
1461  * @param i initial coeff index, 0 unless a separate DC block is coded
1462  * @param zero_nhood the initial prediction context for number of surrounding
1463  * all-zero blocks (only left/top, so 0-2)
1464  * @param qmul array holding the dc/ac dequant factor at position 0/1
1465  * @param scan scan pattern (VP7 only)
1466  *
1467  * @return 0 if no coeffs were decoded
1468  * otherwise, the index of the last coeff decoded plus one
1469  */
1470 static av_always_inline
1472  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
1473  int i, int zero_nhood, const int16_t qmul[2],
1474  const uint8_t scan[16], int vp7)
1475 {
1476  const uint8_t *token_prob = probs[i][zero_nhood];
1477  if (!vpx_rac_get_prob_branchy(c, token_prob[0])) // DCT_EOB
1478  return 0;
1479  return vp7 ? vp7_decode_block_coeffs_internal(c, block, probs, i,
1480  token_prob, qmul, scan)
1482  token_prob, qmul);
1483 }
1484 
1485 static av_always_inline
1487  VP8Macroblock *mb, uint8_t t_nnz[9], uint8_t l_nnz[9],
1488  int is_vp7)
1489 {
1490  int i, x, y, luma_start = 0, luma_ctx = 3;
1491  int nnz_pred, nnz, nnz_total = 0;
1492  int segment = mb->segment;
1493  int block_dc = 0;
1494 
1495  if (mb->mode != MODE_I4x4 && (is_vp7 || mb->mode != VP8_MVMODE_SPLIT)) {
1496  nnz_pred = t_nnz[8] + l_nnz[8];
1497 
1498  // decode DC values and do hadamard
1499  nnz = decode_block_coeffs(c, td->block_dc, s->prob->token[1], 0,
1500  nnz_pred, s->qmat[segment].luma_dc_qmul,
1501  ff_zigzag_scan, is_vp7);
1502  l_nnz[8] = t_nnz[8] = !!nnz;
1503 
1504  if (is_vp7 && mb->mode > MODE_I4x4) {
1505  nnz |= inter_predict_dc(td->block_dc,
1506  s->inter_dc_pred[mb->ref_frame - 1]);
1507  }
1508 
1509  if (nnz) {
1510  nnz_total += nnz;
1511  block_dc = 1;
1512  if (nnz == 1)
1513  s->vp8dsp.vp8_luma_dc_wht_dc(td->block, td->block_dc);
1514  else
1515  s->vp8dsp.vp8_luma_dc_wht(td->block, td->block_dc);
1516  }
1517  luma_start = 1;
1518  luma_ctx = 0;
1519  }
1520 
1521  // luma blocks
1522  for (y = 0; y < 4; y++)
1523  for (x = 0; x < 4; x++) {
1524  nnz_pred = l_nnz[y] + t_nnz[x];
1525  nnz = decode_block_coeffs(c, td->block[y][x],
1526  s->prob->token[luma_ctx],
1527  luma_start, nnz_pred,
1528  s->qmat[segment].luma_qmul,
1529  s->prob[0].scan, is_vp7);
1530  /* nnz+block_dc may be one more than the actual last index,
1531  * but we don't care */
1532  td->non_zero_count_cache[y][x] = nnz + block_dc;
1533  t_nnz[x] = l_nnz[y] = !!nnz;
1534  nnz_total += nnz;
1535  }
1536 
1537  // chroma blocks
1538  // TODO: what to do about dimensions? 2nd dim for luma is x,
1539  // but for chroma it's (y<<1)|x
1540  for (i = 4; i < 6; i++)
1541  for (y = 0; y < 2; y++)
1542  for (x = 0; x < 2; x++) {
1543  nnz_pred = l_nnz[i + 2 * y] + t_nnz[i + 2 * x];
1544  nnz = decode_block_coeffs(c, td->block[i][(y << 1) + x],
1545  s->prob->token[2], 0, nnz_pred,
1546  s->qmat[segment].chroma_qmul,
1547  s->prob[0].scan, is_vp7);
1548  td->non_zero_count_cache[i][(y << 1) + x] = nnz;
1549  t_nnz[i + 2 * x] = l_nnz[i + 2 * y] = !!nnz;
1550  nnz_total += nnz;
1551  }
1552 
1553  // if there were no coded coeffs despite the macroblock not being marked skip,
1554  // we MUST not do the inner loop filter and should not do IDCT
1555  // Since skip isn't used for bitstream prediction, just manually set it.
1556  if (!nnz_total)
1557  mb->skip = 1;
1558 }
1559 
1560 static av_always_inline
1561 void backup_mb_border(uint8_t *top_border, const uint8_t *src_y,
1562  const uint8_t *src_cb, const uint8_t *src_cr,
1563  ptrdiff_t linesize, ptrdiff_t uvlinesize, int simple)
1564 {
1565  AV_COPY128(top_border, src_y + 15 * linesize);
1566  if (!simple) {
1567  AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1568  AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
1569  }
1570 }
1571 
1572 static av_always_inline
1573 void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb,
1574  uint8_t *src_cr, ptrdiff_t linesize, ptrdiff_t uvlinesize, int mb_x,
1575  int mb_y, int mb_width, int simple, int xchg)
1576 {
1577  uint8_t *top_border_m1 = top_border - 32; // for TL prediction
1578  src_y -= linesize;
1579  src_cb -= uvlinesize;
1580  src_cr -= uvlinesize;
1581 
1582 #define XCHG(a, b, xchg) \
1583  do { \
1584  if (xchg) \
1585  AV_SWAP64(b, a); \
1586  else \
1587  AV_COPY64(b, a); \
1588  } while (0)
1589 
1590  XCHG(top_border_m1 + 8, src_y - 8, xchg);
1591  XCHG(top_border, src_y, xchg);
1592  XCHG(top_border + 8, src_y + 8, 1);
1593  if (mb_x < mb_width - 1)
1594  XCHG(top_border + 32, src_y + 16, 1);
1595 
1596  // only copy chroma for normal loop filter
1597  // or to initialize the top row to 127
1598  if (!simple || !mb_y) {
1599  XCHG(top_border_m1 + 16, src_cb - 8, xchg);
1600  XCHG(top_border_m1 + 24, src_cr - 8, xchg);
1601  XCHG(top_border + 16, src_cb, 1);
1602  XCHG(top_border + 24, src_cr, 1);
1603  }
1604 }
1605 
1606 static av_always_inline
1607 int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y)
1608 {
1609  if (!mb_x)
1610  return mb_y ? TOP_DC_PRED8x8 : DC_128_PRED8x8;
1611  else
1612  return mb_y ? mode : LEFT_DC_PRED8x8;
1613 }
1614 
1615 static av_always_inline
1616 int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y, int vp7)
1617 {
1618  if (!mb_x)
1619  return mb_y ? VERT_PRED8x8 : (vp7 ? DC_128_PRED8x8 : DC_129_PRED8x8);
1620  else
1621  return mb_y ? mode : HOR_PRED8x8;
1622 }
1623 
1624 static av_always_inline
1625 int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y, int vp7)
1626 {
1627  switch (mode) {
1628  case DC_PRED8x8:
1629  return check_dc_pred8x8_mode(mode, mb_x, mb_y);
1630  case VERT_PRED8x8:
1631  return !mb_y ? (vp7 ? DC_128_PRED8x8 : DC_127_PRED8x8) : mode;
1632  case HOR_PRED8x8:
1633  return !mb_x ? (vp7 ? DC_128_PRED8x8 : DC_129_PRED8x8) : mode;
1634  case PLANE_PRED8x8: /* TM */
1635  return check_tm_pred8x8_mode(mode, mb_x, mb_y, vp7);
1636  }
1637  return mode;
1638 }
1639 
1640 static av_always_inline
1641 int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y, int vp7)
1642 {
1643  if (!mb_x) {
1644  return mb_y ? VERT_VP8_PRED : (vp7 ? DC_128_PRED : DC_129_PRED);
1645  } else {
1646  return mb_y ? mode : HOR_VP8_PRED;
1647  }
1648 }
1649 
1650 static av_always_inline
1651 int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y,
1652  int *copy_buf, int vp7)
1653 {
1654  switch (mode) {
1655  case VERT_PRED:
1656  if (!mb_x && mb_y) {
1657  *copy_buf = 1;
1658  return mode;
1659  }
1660  /* fall-through */
1661  case DIAG_DOWN_LEFT_PRED:
1662  case VERT_LEFT_PRED:
1663  return !mb_y ? (vp7 ? DC_128_PRED : DC_127_PRED) : mode;
1664  case HOR_PRED:
1665  if (!mb_y) {
1666  *copy_buf = 1;
1667  return mode;
1668  }
1669  /* fall-through */
1670  case HOR_UP_PRED:
1671  return !mb_x ? (vp7 ? DC_128_PRED : DC_129_PRED) : mode;
1672  case TM_VP8_PRED:
1673  return check_tm_pred4x4_mode(mode, mb_x, mb_y, vp7);
1674  case DC_PRED: /* 4x4 DC doesn't use the same "H.264-style" exceptions
1675  * as 16x16/8x8 DC */
1676  case DIAG_DOWN_RIGHT_PRED:
1677  case VERT_RIGHT_PRED:
1678  case HOR_DOWN_PRED:
1679  if (!mb_y || !mb_x)
1680  *copy_buf = 1;
1681  return mode;
1682  }
1683  return mode;
1684 }
1685 
1686 static av_always_inline
1687 void intra_predict(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3],
1688  VP8Macroblock *mb, int mb_x, int mb_y, int is_vp7)
1689 {
1690  int x, y, mode, nnz;
1691  uint32_t tr;
1692 
1693  /* for the first row, we need to run xchg_mb_border to init the top edge
1694  * to 127 otherwise, skip it if we aren't going to deblock */
1695  if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0)
1696  xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2],
1697  s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
1698  s->filter.simple, 1);
1699 
1700  if (mb->mode < MODE_I4x4) {
1701  mode = check_intra_pred8x8_mode_emuedge(mb->mode, mb_x, mb_y, is_vp7);
1702  s->hpc.pred16x16[mode](dst[0], s->linesize);
1703  } else {
1704  uint8_t *ptr = dst[0];
1705  const uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
1706  const uint8_t lo = is_vp7 ? 128 : 127;
1707  const uint8_t hi = is_vp7 ? 128 : 129;
1708  const uint8_t tr_top[4] = { lo, lo, lo, lo };
1709 
1710  // all blocks on the right edge of the macroblock use bottom edge
1711  // the top macroblock for their topright edge
1712  const uint8_t *tr_right = ptr - s->linesize + 16;
1713 
1714  // if we're on the right edge of the frame, said edge is extended
1715  // from the top macroblock
1716  if (mb_y && mb_x == s->mb_width - 1) {
1717  tr = tr_right[-1] * 0x01010101u;
1718  tr_right = (uint8_t *) &tr;
1719  }
1720 
1721  if (mb->skip)
1723 
1724  for (y = 0; y < 4; y++) {
1725  const uint8_t *topright = ptr + 4 - s->linesize;
1726  for (x = 0; x < 4; x++) {
1727  int copy = 0;
1728  ptrdiff_t linesize = s->linesize;
1729  uint8_t *dst = ptr + 4 * x;
1730  LOCAL_ALIGNED(4, uint8_t, copy_dst, [5 * 8]);
1731 
1732  if ((y == 0 || x == 3) && mb_y == 0) {
1733  topright = tr_top;
1734  } else if (x == 3)
1735  topright = tr_right;
1736 
1737  mode = check_intra_pred4x4_mode_emuedge(intra4x4[x], mb_x + x,
1738  mb_y + y, &copy, is_vp7);
1739  if (copy) {
1740  dst = copy_dst + 12;
1741  linesize = 8;
1742  if (!(mb_y + y)) {
1743  copy_dst[3] = lo;
1744  AV_WN32A(copy_dst + 4, lo * 0x01010101U);
1745  } else {
1746  AV_COPY32(copy_dst + 4, ptr + 4 * x - s->linesize);
1747  if (!(mb_x + x)) {
1748  copy_dst[3] = hi;
1749  } else {
1750  copy_dst[3] = ptr[4 * x - s->linesize - 1];
1751  }
1752  }
1753  if (!(mb_x + x)) {
1754  copy_dst[11] =
1755  copy_dst[19] =
1756  copy_dst[27] =
1757  copy_dst[35] = hi;
1758  } else {
1759  copy_dst[11] = ptr[4 * x - 1];
1760  copy_dst[19] = ptr[4 * x + s->linesize - 1];
1761  copy_dst[27] = ptr[4 * x + s->linesize * 2 - 1];
1762  copy_dst[35] = ptr[4 * x + s->linesize * 3 - 1];
1763  }
1764  }
1765  s->hpc.pred4x4[mode](dst, topright, linesize);
1766  if (copy) {
1767  AV_COPY32(ptr + 4 * x, copy_dst + 12);
1768  AV_COPY32(ptr + 4 * x + s->linesize, copy_dst + 20);
1769  AV_COPY32(ptr + 4 * x + s->linesize * 2, copy_dst + 28);
1770  AV_COPY32(ptr + 4 * x + s->linesize * 3, copy_dst + 36);
1771  }
1772 
1773  nnz = td->non_zero_count_cache[y][x];
1774  if (nnz) {
1775  if (nnz == 1)
1776  s->vp8dsp.vp8_idct_dc_add(ptr + 4 * x,
1777  td->block[y][x], s->linesize);
1778  else
1779  s->vp8dsp.vp8_idct_add(ptr + 4 * x,
1780  td->block[y][x], s->linesize);
1781  }
1782  topright += 4;
1783  }
1784 
1785  ptr += 4 * s->linesize;
1786  intra4x4 += 4;
1787  }
1788  }
1789 
1790  mode = check_intra_pred8x8_mode_emuedge(mb->chroma_pred_mode,
1791  mb_x, mb_y, is_vp7);
1792  s->hpc.pred8x8[mode](dst[1], s->uvlinesize);
1793  s->hpc.pred8x8[mode](dst[2], s->uvlinesize);
1794 
1795  if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0)
1796  xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2],
1797  s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
1798  s->filter.simple, 0);
1799 }
1800 
1801 static const uint8_t subpel_idx[3][8] = {
1802  { 0, 1, 2, 1, 2, 1, 2, 1 }, // nr. of left extra pixels,
1803  // also function pointer index
1804  { 0, 3, 5, 3, 5, 3, 5, 3 }, // nr. of extra pixels required
1805  { 0, 2, 3, 2, 3, 2, 3, 2 }, // nr. of right extra pixels
1806 };
1807 
1808 /**
1809  * luma MC function
1810  *
1811  * @param s VP8 decoding context
1812  * @param dst target buffer for block data at block position
1813  * @param ref reference picture buffer at origin (0, 0)
1814  * @param mv motion vector (relative to block position) to get pixel data from
1815  * @param x_off horizontal position of block from origin (0, 0)
1816  * @param y_off vertical position of block from origin (0, 0)
1817  * @param block_w width of block (16, 8 or 4)
1818  * @param block_h height of block (always same as block_w)
1819  * @param width width of src/dst plane data
1820  * @param height height of src/dst plane data
1821  * @param linesize size of a single line of plane data, including padding
1822  * @param mc_func motion compensation function pointers (bilinear or sixtap MC)
1823  */
1824 static av_always_inline
1826  const ProgressFrame *ref, const VP8mv *mv,
1827  int x_off, int y_off, int block_w, int block_h,
1828  int width, int height, ptrdiff_t linesize,
1829  vp8_mc_func mc_func[3][3])
1830 {
1831  const uint8_t *src = ref->f->data[0];
1832 
1833  if (AV_RN32A(mv)) {
1834  ptrdiff_t src_linesize = linesize;
1835 
1836  int mx = (mv->x * 2) & 7, mx_idx = subpel_idx[0][mx];
1837  int my = (mv->y * 2) & 7, my_idx = subpel_idx[0][my];
1838 
1839  x_off += mv->x >> 2;
1840  y_off += mv->y >> 2;
1841 
1842  // edge emulation
1843  ff_progress_frame_await(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 4);
1844  src += y_off * linesize + x_off;
1845  if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] ||
1846  y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
1847  s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
1848  src - my_idx * linesize - mx_idx,
1849  EDGE_EMU_LINESIZE, linesize,
1850  block_w + subpel_idx[1][mx],
1851  block_h + subpel_idx[1][my],
1852  x_off - mx_idx, y_off - my_idx,
1853  width, height);
1854  src = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
1855  src_linesize = EDGE_EMU_LINESIZE;
1856  }
1857  mc_func[my_idx][mx_idx](dst, linesize, src, src_linesize, block_h, mx, my);
1858  } else {
1859  ff_progress_frame_await(ref, (3 + y_off + block_h) >> 4);
1860  mc_func[0][0](dst, linesize, src + y_off * linesize + x_off,
1861  linesize, block_h, 0, 0);
1862  }
1863 }
1864 
1865 /**
1866  * chroma MC function
1867  *
1868  * @param s VP8 decoding context
1869  * @param dst1 target buffer for block data at block position (U plane)
1870  * @param dst2 target buffer for block data at block position (V plane)
1871  * @param ref reference picture buffer at origin (0, 0)
1872  * @param mv motion vector (relative to block position) to get pixel data from
1873  * @param x_off horizontal position of block from origin (0, 0)
1874  * @param y_off vertical position of block from origin (0, 0)
1875  * @param block_w width of block (16, 8 or 4)
1876  * @param block_h height of block (always same as block_w)
1877  * @param width width of src/dst plane data
1878  * @param height height of src/dst plane data
1879  * @param linesize size of a single line of plane data, including padding
1880  * @param mc_func motion compensation function pointers (bilinear or sixtap MC)
1881  */
1882 static av_always_inline
1883 void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1,
1884  uint8_t *dst2, const ProgressFrame *ref, const VP8mv *mv,
1885  int x_off, int y_off, int block_w, int block_h,
1886  int width, int height, ptrdiff_t linesize,
1887  vp8_mc_func mc_func[3][3])
1888 {
1889  const uint8_t *src1 = ref->f->data[1], *src2 = ref->f->data[2];
1890 
1891  if (AV_RN32A(mv)) {
1892  int mx = mv->x & 7, mx_idx = subpel_idx[0][mx];
1893  int my = mv->y & 7, my_idx = subpel_idx[0][my];
1894 
1895  x_off += mv->x >> 3;
1896  y_off += mv->y >> 3;
1897 
1898  // edge emulation
1899  src1 += y_off * linesize + x_off;
1900  src2 += y_off * linesize + x_off;
1901  ff_progress_frame_await(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3);
1902  if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] ||
1903  y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
1904  s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
1905  src1 - my_idx * linesize - mx_idx,
1906  EDGE_EMU_LINESIZE, linesize,
1907  block_w + subpel_idx[1][mx],
1908  block_h + subpel_idx[1][my],
1909  x_off - mx_idx, y_off - my_idx, width, height);
1910  src1 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
1911  mc_func[my_idx][mx_idx](dst1, linesize, src1, EDGE_EMU_LINESIZE, block_h, mx, my);
1912 
1913  s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
1914  src2 - my_idx * linesize - mx_idx,
1915  EDGE_EMU_LINESIZE, linesize,
1916  block_w + subpel_idx[1][mx],
1917  block_h + subpel_idx[1][my],
1918  x_off - mx_idx, y_off - my_idx, width, height);
1919  src2 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
1920  mc_func[my_idx][mx_idx](dst2, linesize, src2, EDGE_EMU_LINESIZE, block_h, mx, my);
1921  } else {
1922  mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my);
1923  mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my);
1924  }
1925  } else {
1926  ff_progress_frame_await(ref, (3 + y_off + block_h) >> 3);
1927  mc_func[0][0](dst1, linesize, src1 + y_off * linesize + x_off, linesize, block_h, 0, 0);
1928  mc_func[0][0](dst2, linesize, src2 + y_off * linesize + x_off, linesize, block_h, 0, 0);
1929  }
1930 }
1931 
1932 static av_always_inline
1933 void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3],
1934  const ProgressFrame *ref_frame, int x_off, int y_off,
1935  int bx_off, int by_off, int block_w, int block_h,
1936  int width, int height, const VP8mv *mv)
1937 {
1938  VP8mv uvmv = *mv;
1939 
1940  /* Y */
1941  vp8_mc_luma(s, td, dst[0] + by_off * s->linesize + bx_off,
1942  ref_frame, mv, x_off + bx_off, y_off + by_off,
1943  block_w, block_h, width, height, s->linesize,
1944  s->put_pixels_tab[block_w == 8]);
1945 
1946  /* U/V */
1947  if (s->profile == 3) {
1948  /* this block only applies VP8; it is safe to check
1949  * only the profile, as VP7 profile <= 1 */
1950  uvmv.x &= ~7;
1951  uvmv.y &= ~7;
1952  }
1953  x_off >>= 1;
1954  y_off >>= 1;
1955  bx_off >>= 1;
1956  by_off >>= 1;
1957  width >>= 1;
1958  height >>= 1;
1959  block_w >>= 1;
1960  block_h >>= 1;
1961  vp8_mc_chroma(s, td, dst[1] + by_off * s->uvlinesize + bx_off,
1962  dst[2] + by_off * s->uvlinesize + bx_off, ref_frame,
1963  &uvmv, x_off + bx_off, y_off + by_off,
1964  block_w, block_h, width, height, s->uvlinesize,
1965  s->put_pixels_tab[1 + (block_w == 4)]);
1966 }
1967 
1968 /* Fetch pixels for estimated mv 4 macroblocks ahead.
1969  * Optimized for 64-byte cache lines. Inspired by ffh264 prefetch_motion. */
1970 static av_always_inline
1972  int mb_x, int mb_y, int mb_xy, int ref)
1973 {
1974  /* Don't prefetch refs that haven't been used very often this frame. */
1975  if (s->ref_count[ref - 1] > (mb_xy >> 5)) {
1976  int x_off = mb_x << 4, y_off = mb_y << 4;
1977  int mx = (mb->mv.x >> 2) + x_off + 8;
1978  int my = (mb->mv.y >> 2) + y_off;
1979  uint8_t **src = s->framep[ref]->tf.f->data;
1980  int off = mx + (my + (mb_x & 3) * 4) * s->linesize + 64;
1981  /* For threading, a ff_thread_await_progress here might be useful, but
1982  * it actually slows down the decoder. Since a bad prefetch doesn't
1983  * generate bad decoder output, we don't run it here. */
1984  s->vdsp.prefetch(src[0] + off, s->linesize, 4);
1985  off = (mx >> 1) + ((my >> 1) + (mb_x & 7)) * s->uvlinesize + 64;
1986  s->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
1987  }
1988 }
1989 
1990 /**
1991  * Apply motion vectors to prediction buffer, chapter 18.
1992  */
1993 static av_always_inline
1994 void inter_predict(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3],
1995  VP8Macroblock *mb, int mb_x, int mb_y)
1996 {
1997  int x_off = mb_x << 4, y_off = mb_y << 4;
1998  int width = 16 * s->mb_width, height = 16 * s->mb_height;
1999  const ProgressFrame *ref = &s->framep[mb->ref_frame]->tf;
2000  const VP8mv *bmv = mb->bmv;
2001 
2002  switch (mb->partitioning) {
2003  case VP8_SPLITMVMODE_NONE:
2004  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2005  0, 0, 16, 16, width, height, &mb->mv);
2006  break;
2007  case VP8_SPLITMVMODE_4x4: {
2008  int x, y;
2009  VP8mv uvmv;
2010 
2011  /* Y */
2012  for (y = 0; y < 4; y++) {
2013  for (x = 0; x < 4; x++) {
2014  vp8_mc_luma(s, td, dst[0] + 4 * y * s->linesize + x * 4,
2015  ref, &bmv[4 * y + x],
2016  4 * x + x_off, 4 * y + y_off, 4, 4,
2017  width, height, s->linesize,
2018  s->put_pixels_tab[2]);
2019  }
2020  }
2021 
2022  /* U/V */
2023  x_off >>= 1;
2024  y_off >>= 1;
2025  width >>= 1;
2026  height >>= 1;
2027  for (y = 0; y < 2; y++) {
2028  for (x = 0; x < 2; x++) {
2029  uvmv.x = mb->bmv[2 * y * 4 + 2 * x ].x +
2030  mb->bmv[2 * y * 4 + 2 * x + 1].x +
2031  mb->bmv[(2 * y + 1) * 4 + 2 * x ].x +
2032  mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].x;
2033  uvmv.y = mb->bmv[2 * y * 4 + 2 * x ].y +
2034  mb->bmv[2 * y * 4 + 2 * x + 1].y +
2035  mb->bmv[(2 * y + 1) * 4 + 2 * x ].y +
2036  mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].y;
2037  uvmv.x = (uvmv.x + 2 + FF_SIGNBIT(uvmv.x)) >> 2;
2038  uvmv.y = (uvmv.y + 2 + FF_SIGNBIT(uvmv.y)) >> 2;
2039  if (s->profile == 3) {
2040  uvmv.x &= ~7;
2041  uvmv.y &= ~7;
2042  }
2043  vp8_mc_chroma(s, td, dst[1] + 4 * y * s->uvlinesize + x * 4,
2044  dst[2] + 4 * y * s->uvlinesize + x * 4, ref,
2045  &uvmv, 4 * x + x_off, 4 * y + y_off, 4, 4,
2046  width, height, s->uvlinesize,
2047  s->put_pixels_tab[2]);
2048  }
2049  }
2050  break;
2051  }
2052  case VP8_SPLITMVMODE_16x8:
2053  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2054  0, 0, 16, 8, width, height, &bmv[0]);
2055  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2056  0, 8, 16, 8, width, height, &bmv[1]);
2057  break;
2058  case VP8_SPLITMVMODE_8x16:
2059  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2060  0, 0, 8, 16, width, height, &bmv[0]);
2061  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2062  8, 0, 8, 16, width, height, &bmv[1]);
2063  break;
2064  case VP8_SPLITMVMODE_8x8:
2065  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2066  0, 0, 8, 8, width, height, &bmv[0]);
2067  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2068  8, 0, 8, 8, width, height, &bmv[1]);
2069  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2070  0, 8, 8, 8, width, height, &bmv[2]);
2071  vp8_mc_part(s, td, dst, ref, x_off, y_off,
2072  8, 8, 8, 8, width, height, &bmv[3]);
2073  break;
2074  }
2075 }
2076 
2077 static av_always_inline
2078 void idct_mb(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3],
2079  const VP8Macroblock *mb)
2080 {
2081  int x, y, ch;
2082 
2083  if (mb->mode != MODE_I4x4) {
2084  uint8_t *y_dst = dst[0];
2085  for (y = 0; y < 4; y++) {
2086  uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[y]);
2087  if (nnz4) {
2088  if (nnz4 & ~0x01010101) {
2089  for (x = 0; x < 4; x++) {
2090  if ((uint8_t) nnz4 == 1)
2091  s->vp8dsp.vp8_idct_dc_add(y_dst + 4 * x,
2092  td->block[y][x],
2093  s->linesize);
2094  else if ((uint8_t) nnz4 > 1)
2095  s->vp8dsp.vp8_idct_add(y_dst + 4 * x,
2096  td->block[y][x],
2097  s->linesize);
2098  nnz4 >>= 8;
2099  if (!nnz4)
2100  break;
2101  }
2102  } else {
2103  s->vp8dsp.vp8_idct_dc_add4y(y_dst, td->block[y], s->linesize);
2104  }
2105  }
2106  y_dst += 4 * s->linesize;
2107  }
2108  }
2109 
2110  for (ch = 0; ch < 2; ch++) {
2111  uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[4 + ch]);
2112  if (nnz4) {
2113  uint8_t *ch_dst = dst[1 + ch];
2114  if (nnz4 & ~0x01010101) {
2115  for (y = 0; y < 2; y++) {
2116  for (x = 0; x < 2; x++) {
2117  if ((uint8_t) nnz4 == 1)
2118  s->vp8dsp.vp8_idct_dc_add(ch_dst + 4 * x,
2119  td->block[4 + ch][(y << 1) + x],
2120  s->uvlinesize);
2121  else if ((uint8_t) nnz4 > 1)
2122  s->vp8dsp.vp8_idct_add(ch_dst + 4 * x,
2123  td->block[4 + ch][(y << 1) + x],
2124  s->uvlinesize);
2125  nnz4 >>= 8;
2126  if (!nnz4)
2127  goto chroma_idct_end;
2128  }
2129  ch_dst += 4 * s->uvlinesize;
2130  }
2131  } else {
2132  s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, td->block[4 + ch], s->uvlinesize);
2133  }
2134  }
2135 chroma_idct_end:
2136  ;
2137  }
2138 }
2139 
2140 static av_always_inline
2142  VP8FilterStrength *f, int is_vp7)
2143 {
2144  int interior_limit, filter_level;
2145 
2146  if (s->segmentation.enabled) {
2147  filter_level = s->segmentation.filter_level[mb->segment];
2148  if (!s->segmentation.absolute_vals)
2149  filter_level += s->filter.level;
2150  } else
2151  filter_level = s->filter.level;
2152 
2153  if (s->lf_delta.enabled) {
2154  filter_level += s->lf_delta.ref[mb->ref_frame];
2155  filter_level += s->lf_delta.mode[mb->mode];
2156  }
2157 
2158  filter_level = av_clip_uintp2(filter_level, 6);
2159 
2160  interior_limit = filter_level;
2161  if (s->filter.sharpness) {
2162  interior_limit >>= (s->filter.sharpness + 3) >> 2;
2163  interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness);
2164  }
2165  interior_limit = FFMAX(interior_limit, 1);
2166 
2167  f->filter_level = filter_level;
2168  f->inner_limit = interior_limit;
2169  f->inner_filter = is_vp7 || !mb->skip || mb->mode == MODE_I4x4 ||
2170  mb->mode == VP8_MVMODE_SPLIT;
2171 }
2172 
2173 static av_always_inline
2174 void filter_mb(const VP8Context *s, uint8_t *const dst[3], const VP8FilterStrength *f,
2175  int mb_x, int mb_y, int is_vp7)
2176 {
2177  int mbedge_lim, bedge_lim_y, bedge_lim_uv, hev_thresh;
2178  int filter_level = f->filter_level;
2179  int inner_limit = f->inner_limit;
2180  int inner_filter = f->inner_filter;
2181  ptrdiff_t linesize = s->linesize;
2182  ptrdiff_t uvlinesize = s->uvlinesize;
2183  static const uint8_t hev_thresh_lut[2][64] = {
2184  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
2185  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2186  3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
2187  3, 3, 3, 3 },
2188  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
2189  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
2190  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2191  2, 2, 2, 2 }
2192  };
2193 
2194  if (!filter_level)
2195  return;
2196 
2197  if (is_vp7) {
2198  bedge_lim_y = filter_level;
2199  bedge_lim_uv = filter_level * 2;
2200  mbedge_lim = filter_level + 2;
2201  } else {
2202  bedge_lim_y =
2203  bedge_lim_uv = filter_level * 2 + inner_limit;
2204  mbedge_lim = bedge_lim_y + 4;
2205  }
2206 
2207  hev_thresh = hev_thresh_lut[s->keyframe][filter_level];
2208 
2209  if (mb_x) {
2210  s->vp8dsp.vp8_h_loop_filter16y(dst[0], linesize,
2211  mbedge_lim, inner_limit, hev_thresh);
2212  s->vp8dsp.vp8_h_loop_filter8uv(dst[1], dst[2], uvlinesize,
2213  mbedge_lim, inner_limit, hev_thresh);
2214  }
2215 
2216 #define H_LOOP_FILTER_16Y_INNER(cond) \
2217  if (cond && inner_filter) { \
2218  s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 4, linesize, \
2219  bedge_lim_y, inner_limit, \
2220  hev_thresh); \
2221  s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 8, linesize, \
2222  bedge_lim_y, inner_limit, \
2223  hev_thresh); \
2224  s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 12, linesize, \
2225  bedge_lim_y, inner_limit, \
2226  hev_thresh); \
2227  s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4, \
2228  uvlinesize, bedge_lim_uv, \
2229  inner_limit, hev_thresh); \
2230  }
2231 
2232  H_LOOP_FILTER_16Y_INNER(!is_vp7)
2233 
2234  if (mb_y) {
2235  s->vp8dsp.vp8_v_loop_filter16y(dst[0], linesize,
2236  mbedge_lim, inner_limit, hev_thresh);
2237  s->vp8dsp.vp8_v_loop_filter8uv(dst[1], dst[2], uvlinesize,
2238  mbedge_lim, inner_limit, hev_thresh);
2239  }
2240 
2241  if (inner_filter) {
2242  s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 4 * linesize,
2243  linesize, bedge_lim_y,
2244  inner_limit, hev_thresh);
2245  s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 8 * linesize,
2246  linesize, bedge_lim_y,
2247  inner_limit, hev_thresh);
2248  s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 12 * linesize,
2249  linesize, bedge_lim_y,
2250  inner_limit, hev_thresh);
2251  s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] + 4 * uvlinesize,
2252  dst[2] + 4 * uvlinesize,
2253  uvlinesize, bedge_lim_uv,
2254  inner_limit, hev_thresh);
2255  }
2256 
2257  H_LOOP_FILTER_16Y_INNER(is_vp7)
2258 }
2259 
2260 static av_always_inline
2261 void filter_mb_simple(const VP8Context *s, uint8_t *dst, const VP8FilterStrength *f,
2262  int mb_x, int mb_y)
2263 {
2264  int mbedge_lim, bedge_lim;
2265  int filter_level = f->filter_level;
2266  int inner_limit = f->inner_limit;
2267  int inner_filter = f->inner_filter;
2268  ptrdiff_t linesize = s->linesize;
2269 
2270  if (!filter_level)
2271  return;
2272 
2273  bedge_lim = 2 * filter_level + inner_limit;
2274  mbedge_lim = bedge_lim + 4;
2275 
2276  if (mb_x)
2277  s->vp8dsp.vp8_h_loop_filter_simple(dst, linesize, mbedge_lim);
2278  if (inner_filter) {
2279  s->vp8dsp.vp8_h_loop_filter_simple(dst + 4, linesize, bedge_lim);
2280  s->vp8dsp.vp8_h_loop_filter_simple(dst + 8, linesize, bedge_lim);
2281  s->vp8dsp.vp8_h_loop_filter_simple(dst + 12, linesize, bedge_lim);
2282  }
2283 
2284  if (mb_y)
2285  s->vp8dsp.vp8_v_loop_filter_simple(dst, linesize, mbedge_lim);
2286  if (inner_filter) {
2287  s->vp8dsp.vp8_v_loop_filter_simple(dst + 4 * linesize, linesize, bedge_lim);
2288  s->vp8dsp.vp8_v_loop_filter_simple(dst + 8 * linesize, linesize, bedge_lim);
2289  s->vp8dsp.vp8_v_loop_filter_simple(dst + 12 * linesize, linesize, bedge_lim);
2290  }
2291 }
2292 
2293 #define MARGIN (16 << 2)
2294 static av_always_inline
2296  const VP8Frame *prev_frame, int is_vp7)
2297 {
2298  VP8Context *s = avctx->priv_data;
2299  int mb_x, mb_y;
2300 
2301  s->mv_bounds.mv_min.y = -MARGIN;
2302  s->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
2303  for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
2304  VP8Macroblock *mb = s->macroblocks_base +
2305  ((s->mb_width + 1) * (mb_y + 1) + 1);
2306  int mb_xy = mb_y * s->mb_width;
2307 
2308  AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101);
2309 
2310  s->mv_bounds.mv_min.x = -MARGIN;
2311  s->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
2312 
2313  for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
2314  if (vpx_rac_is_end(&s->c)) {
2315  return AVERROR_INVALIDDATA;
2316  }
2317  if (mb_y == 0)
2318  AV_WN32A((mb - s->mb_width - 1)->intra4x4_pred_mode_top,
2319  DC_PRED * 0x01010101);
2320  decode_mb_mode(s, &s->mv_bounds, mb, mb_x, mb_y, curframe->seg_map + mb_xy,
2321  prev_frame && prev_frame->seg_map ?
2322  prev_frame->seg_map + mb_xy : NULL, 1, is_vp7);
2323  s->mv_bounds.mv_min.x -= 64;
2324  s->mv_bounds.mv_max.x -= 64;
2325  }
2326  s->mv_bounds.mv_min.y -= 64;
2327  s->mv_bounds.mv_max.y -= 64;
2328  }
2329  return 0;
2330 }
2331 
2332 static int vp7_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame,
2333  const VP8Frame *prev_frame)
2334 {
2335  return vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP7);
2336 }
2337 
2338 static int vp8_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_VP8);
2342 }
2343 
2344 #if HAVE_THREADS
2345 #define check_thread_pos(td, otd, mb_x_check, mb_y_check) \
2346  do { \
2347  int tmp = (mb_y_check << 16) | (mb_x_check & 0xFFFF); \
2348  if (atomic_load(&otd->thread_mb_pos) < tmp) { \
2349  pthread_mutex_lock(&otd->lock); \
2350  atomic_store(&td->wait_mb_pos, tmp); \
2351  do { \
2352  if (atomic_load(&otd->thread_mb_pos) >= tmp) \
2353  break; \
2354  pthread_cond_wait(&otd->cond, &otd->lock); \
2355  } while (1); \
2356  atomic_store(&td->wait_mb_pos, INT_MAX); \
2357  pthread_mutex_unlock(&otd->lock); \
2358  } \
2359  } while (0)
2360 
2361 #define update_pos(td, mb_y, mb_x) \
2362  do { \
2363  int pos = (mb_y << 16) | (mb_x & 0xFFFF); \
2364  int sliced_threading = (avctx->active_thread_type == FF_THREAD_SLICE) && \
2365  (num_jobs > 1); \
2366  int is_null = !next_td || !prev_td; \
2367  int pos_check = (is_null) ? 1 : \
2368  (next_td != td && pos >= atomic_load(&next_td->wait_mb_pos)) || \
2369  (prev_td != td && pos >= atomic_load(&prev_td->wait_mb_pos)); \
2370  atomic_store(&td->thread_mb_pos, pos); \
2371  if (sliced_threading && pos_check) { \
2372  pthread_mutex_lock(&td->lock); \
2373  pthread_cond_broadcast(&td->cond); \
2374  pthread_mutex_unlock(&td->lock); \
2375  } \
2376  } while (0)
2377 #else
2378 #define check_thread_pos(td, otd, mb_x_check, mb_y_check) while(0)
2379 #define update_pos(td, mb_y, mb_x) while(0)
2380 #endif
2381 
2383  int jobnr, int threadnr, int is_vp7)
2384 {
2385  VP8Context *s = avctx->priv_data;
2386  VP8ThreadData *prev_td, *next_td, *td = &s->thread_data[threadnr];
2387  int mb_y = atomic_load(&td->thread_mb_pos) >> 16;
2388  int mb_x, mb_xy = mb_y * s->mb_width;
2389  int num_jobs = s->num_jobs;
2390  const VP8Frame *prev_frame = s->prev_frame;
2391  VP8Frame *curframe = s->curframe;
2392  VPXRangeCoder *coeff_c = &s->coeff_partition[mb_y & (s->num_coeff_partitions - 1)];
2393 
2394  VP8Macroblock *mb;
2395  uint8_t *dst[3] = {
2396  curframe->tf.f->data[0] + 16 * mb_y * s->linesize,
2397  curframe->tf.f->data[1] + 8 * mb_y * s->uvlinesize,
2398  curframe->tf.f->data[2] + 8 * mb_y * s->uvlinesize
2399  };
2400 
2401  if (vpx_rac_is_end(&s->c))
2402  return AVERROR_INVALIDDATA;
2403 
2404  if (mb_y == 0)
2405  prev_td = td;
2406  else
2407  prev_td = &s->thread_data[(jobnr + num_jobs - 1) % num_jobs];
2408  if (mb_y == s->mb_height - 1)
2409  next_td = td;
2410  else
2411  next_td = &s->thread_data[(jobnr + 1) % num_jobs];
2412  if (s->mb_layout == 1)
2413  mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1);
2414  else {
2415  // Make sure the previous frame has read its segmentation map,
2416  // if we re-use the same map.
2417  if (prev_frame && s->segmentation.enabled &&
2418  !s->segmentation.update_map)
2419  ff_progress_frame_await(&prev_frame->tf, mb_y);
2420  mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2;
2421  memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock
2422  AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101);
2423  }
2424 
2425  if (!is_vp7 || mb_y == 0)
2426  memset(td->left_nnz, 0, sizeof(td->left_nnz));
2427 
2428  td->mv_bounds.mv_min.x = -MARGIN;
2429  td->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
2430 
2431  for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
2432  if (vpx_rac_is_end(&s->c))
2433  return AVERROR_INVALIDDATA;
2434  // Wait for previous thread to read mb_x+2, and reach mb_y-1.
2435  if (prev_td != td) {
2436  if (threadnr != 0) {
2437  check_thread_pos(td, prev_td,
2438  mb_x + (is_vp7 ? 2 : 1),
2439  mb_y - (is_vp7 ? 2 : 1));
2440  } else {
2441  check_thread_pos(td, prev_td,
2442  mb_x + (is_vp7 ? 2 : 1) + s->mb_width + 3,
2443  mb_y - (is_vp7 ? 2 : 1));
2444  }
2445  }
2446 
2447  s->vdsp.prefetch(dst[0] + (mb_x & 3) * 4 * s->linesize + 64,
2448  s->linesize, 4);
2449  s->vdsp.prefetch(dst[1] + (mb_x & 7) * s->uvlinesize + 64,
2450  dst[2] - dst[1], 2);
2451 
2452  if (!s->mb_layout)
2453  decode_mb_mode(s, &td->mv_bounds, mb, mb_x, mb_y, curframe->seg_map + mb_xy,
2454  prev_frame && prev_frame->seg_map ?
2455  prev_frame->seg_map + mb_xy : NULL, 0, is_vp7);
2456 
2457  prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_PREVIOUS);
2458 
2459  if (!mb->skip) {
2460  if (vpx_rac_is_end(coeff_c))
2461  return AVERROR_INVALIDDATA;
2462  decode_mb_coeffs(s, td, coeff_c, mb, s->top_nnz[mb_x], td->left_nnz, is_vp7);
2463  }
2464 
2465  if (mb->mode <= MODE_I4x4)
2466  intra_predict(s, td, dst, mb, mb_x, mb_y, is_vp7);
2467  else
2468  inter_predict(s, td, dst, mb, mb_x, mb_y);
2469 
2470  prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_GOLDEN);
2471 
2472  if (!mb->skip) {
2473  idct_mb(s, td, dst, mb);
2474  } else {
2475  AV_ZERO64(td->left_nnz);
2476  AV_WN64(s->top_nnz[mb_x], 0); // array of 9, so unaligned
2477 
2478  /* Reset DC block predictors if they would exist
2479  * if the mb had coefficients */
2480  if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
2481  td->left_nnz[8] = 0;
2482  s->top_nnz[mb_x][8] = 0;
2483  }
2484  }
2485 
2486  if (s->deblock_filter)
2487  filter_level_for_mb(s, mb, &td->filter_strength[mb_x], is_vp7);
2488 
2489  if (s->deblock_filter && num_jobs != 1 && threadnr == num_jobs - 1) {
2490  if (s->filter.simple)
2491  backup_mb_border(s->top_border[mb_x + 1], dst[0],
2492  NULL, NULL, s->linesize, 0, 1);
2493  else
2494  backup_mb_border(s->top_border[mb_x + 1], dst[0],
2495  dst[1], dst[2], s->linesize, s->uvlinesize, 0);
2496  }
2497 
2498  prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_ALTREF);
2499 
2500  dst[0] += 16;
2501  dst[1] += 8;
2502  dst[2] += 8;
2503  td->mv_bounds.mv_min.x -= 64;
2504  td->mv_bounds.mv_max.x -= 64;
2505 
2506  if (mb_x == s->mb_width + 1) {
2507  update_pos(td, mb_y, s->mb_width + 3);
2508  } else {
2509  update_pos(td, mb_y, mb_x);
2510  }
2511  }
2512  return 0;
2513 }
2514 
2515 static int vp7_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
2516  int jobnr, int threadnr)
2517 {
2518  return decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 1);
2519 }
2520 
2521 static int vp8_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, 0);
2525 }
2526 
2527 static av_always_inline void filter_mb_row(AVCodecContext *avctx, void *tdata,
2528  int jobnr, int threadnr, int is_vp7)
2529 {
2530  VP8Context *s = avctx->priv_data;
2531  VP8ThreadData *td = &s->thread_data[threadnr];
2532  int mb_x, mb_y = atomic_load(&td->thread_mb_pos) >> 16, num_jobs = s->num_jobs;
2533  AVFrame *curframe = s->curframe->tf.f;
2534  VP8Macroblock *mb;
2535  VP8ThreadData *prev_td, *next_td;
2536  uint8_t *dst[3] = {
2537  curframe->data[0] + 16 * mb_y * s->linesize,
2538  curframe->data[1] + 8 * mb_y * s->uvlinesize,
2539  curframe->data[2] + 8 * mb_y * s->uvlinesize
2540  };
2541 
2542  if (s->mb_layout == 1)
2543  mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1);
2544  else
2545  mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2;
2546 
2547  if (mb_y == 0)
2548  prev_td = td;
2549  else
2550  prev_td = &s->thread_data[(jobnr + num_jobs - 1) % num_jobs];
2551  if (mb_y == s->mb_height - 1)
2552  next_td = td;
2553  else
2554  next_td = &s->thread_data[(jobnr + 1) % num_jobs];
2555 
2556  for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb++) {
2557  const VP8FilterStrength *f = &td->filter_strength[mb_x];
2558  if (prev_td != td)
2559  check_thread_pos(td, prev_td,
2560  (mb_x + 1) + (s->mb_width + 3), mb_y - 1);
2561  if (next_td != td)
2562  if (next_td != &s->thread_data[0])
2563  check_thread_pos(td, next_td, mb_x + 1, mb_y + 1);
2564 
2565  if (num_jobs == 1) {
2566  if (s->filter.simple)
2567  backup_mb_border(s->top_border[mb_x + 1], dst[0],
2568  NULL, NULL, s->linesize, 0, 1);
2569  else
2570  backup_mb_border(s->top_border[mb_x + 1], dst[0],
2571  dst[1], dst[2], s->linesize, s->uvlinesize, 0);
2572  }
2573 
2574  if (s->filter.simple)
2575  filter_mb_simple(s, dst[0], f, mb_x, mb_y);
2576  else
2577  filter_mb(s, dst, f, mb_x, mb_y, is_vp7);
2578  dst[0] += 16;
2579  dst[1] += 8;
2580  dst[2] += 8;
2581 
2582  update_pos(td, mb_y, (s->mb_width + 3) + mb_x);
2583  }
2584 }
2585 
2586 static void vp7_filter_mb_row(AVCodecContext *avctx, void *tdata,
2587  int jobnr, int threadnr)
2588 {
2589  filter_mb_row(avctx, tdata, jobnr, threadnr, 1);
2590 }
2591 
2592 static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata,
2593  int jobnr, int threadnr)
2594 {
2595  filter_mb_row(avctx, tdata, jobnr, threadnr, 0);
2596 }
2597 
2598 static av_always_inline
2599 int vp78_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr,
2600  int threadnr, int is_vp7)
2601 {
2602  const VP8Context *s = avctx->priv_data;
2603  VP8ThreadData *td = &s->thread_data[jobnr];
2604  VP8ThreadData *next_td = NULL, *prev_td = NULL;
2605  VP8Frame *curframe = s->curframe;
2606  int mb_y, num_jobs = s->num_jobs;
2607  int ret;
2608 
2609  td->thread_nr = threadnr;
2610  td->mv_bounds.mv_min.y = -MARGIN - 64 * threadnr;
2611  td->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN - 64 * threadnr;
2612  for (mb_y = jobnr; mb_y < s->mb_height; mb_y += num_jobs) {
2613  atomic_store(&td->thread_mb_pos, mb_y << 16);
2614  ret = s->decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr);
2615  if (ret < 0) {
2616  update_pos(td, s->mb_height, INT_MAX & 0xFFFF);
2617  return ret;
2618  }
2619  if (s->deblock_filter)
2620  s->filter_mb_row(avctx, tdata, jobnr, threadnr);
2621  update_pos(td, mb_y, INT_MAX & 0xFFFF);
2622 
2623  td->mv_bounds.mv_min.y -= 64 * num_jobs;
2624  td->mv_bounds.mv_max.y -= 64 * num_jobs;
2625 
2626  if (avctx->active_thread_type == FF_THREAD_FRAME)
2627  ff_progress_frame_report(&curframe->tf, mb_y);
2628  }
2629 
2630  return 0;
2631 }
2632 
2633 static int vp7_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata,
2634  int jobnr, int threadnr)
2635 {
2636  return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP7);
2637 }
2638 
2639 static int vp8_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_VP8);
2643 }
2644 
2645 static av_always_inline
2646 int vp78_decode_frame(AVCodecContext *avctx, AVFrame *rframe, int *got_frame,
2647  const AVPacket *avpkt, int is_vp7)
2648 {
2649  VP8Context *s = avctx->priv_data;
2650  int ret, i, referenced, num_jobs;
2651  enum AVDiscard skip_thresh;
2652  VP8Frame *av_uninit(curframe), *prev_frame;
2653 
2654  if (is_vp7)
2655  ret = vp7_decode_frame_header(s, avpkt->data, avpkt->size);
2656  else
2657  ret = vp8_decode_frame_header(s, avpkt->data, avpkt->size);
2658 
2659  if (ret < 0)
2660  goto err;
2661 
2662  if (!is_vp7 && s->actually_webp) {
2663  // VP8 in WebP is supposed to be intra-only. Enforce this here
2664  // to ensure that output is reproducible with frame-threading.
2665  if (!s->keyframe)
2666  return AVERROR_INVALIDDATA;
2667  // avctx->pix_fmt already set in caller.
2668  } else if (!is_vp7 && s->pix_fmt == AV_PIX_FMT_NONE) {
2669  s->pix_fmt = get_pixel_format(s);
2670  if (s->pix_fmt < 0) {
2671  ret = AVERROR(EINVAL);
2672  goto err;
2673  }
2674  avctx->pix_fmt = s->pix_fmt;
2675  }
2676 
2677  prev_frame = s->framep[VP8_FRAME_CURRENT];
2678 
2679  referenced = s->update_last || s->update_golden == VP8_FRAME_CURRENT ||
2680  s->update_altref == VP8_FRAME_CURRENT;
2681 
2682  skip_thresh = !referenced ? AVDISCARD_NONREF
2683  : !s->keyframe ? AVDISCARD_NONKEY
2684  : AVDISCARD_ALL;
2685 
2686  if (avctx->skip_frame >= skip_thresh) {
2687  s->invisible = 1;
2688  memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4);
2689  goto skip_decode;
2690  }
2691  s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh;
2692 
2693  // release no longer referenced frames
2694  for (i = 0; i < 5; i++)
2695  if (s->frames[i].tf.f &&
2696  &s->frames[i] != prev_frame &&
2697  &s->frames[i] != s->framep[VP8_FRAME_PREVIOUS] &&
2698  &s->frames[i] != s->framep[VP8_FRAME_GOLDEN] &&
2699  &s->frames[i] != s->framep[VP8_FRAME_ALTREF])
2700  vp8_release_frame(&s->frames[i]);
2701 
2702  curframe = s->framep[VP8_FRAME_CURRENT] = vp8_find_free_buffer(s);
2703 
2704  if (!s->colorspace)
2705  avctx->colorspace = AVCOL_SPC_BT470BG;
2706  if (s->fullrange)
2707  avctx->color_range = AVCOL_RANGE_JPEG;
2708  else
2709  avctx->color_range = AVCOL_RANGE_MPEG;
2710 
2711  /* Given that arithmetic probabilities are updated every frame, it's quite
2712  * likely that the values we have on a random interframe are complete
2713  * junk if we didn't start decode on a keyframe. So just don't display
2714  * anything rather than junk. */
2715  if (!s->keyframe && (!s->framep[VP8_FRAME_PREVIOUS] ||
2716  !s->framep[VP8_FRAME_GOLDEN] ||
2717  !s->framep[VP8_FRAME_ALTREF])) {
2718  av_log(avctx, AV_LOG_WARNING,
2719  "Discarding interframe without a prior keyframe!\n");
2721  goto err;
2722  }
2723 
2724  if ((ret = vp8_alloc_frame(s, curframe, referenced)) < 0)
2725  goto err;
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 
2733  // check if golden and altref are swapped
2734  if (s->update_altref != VP8_FRAME_NONE)
2735  s->next_framep[VP8_FRAME_ALTREF] = s->framep[s->update_altref];
2736  else
2737  s->next_framep[VP8_FRAME_ALTREF] = s->framep[VP8_FRAME_ALTREF];
2738 
2739  if (s->update_golden != VP8_FRAME_NONE)
2740  s->next_framep[VP8_FRAME_GOLDEN] = s->framep[s->update_golden];
2741  else
2742  s->next_framep[VP8_FRAME_GOLDEN] = s->framep[VP8_FRAME_GOLDEN];
2743 
2744  if (s->update_last)
2745  s->next_framep[VP8_FRAME_PREVIOUS] = curframe;
2746  else
2747  s->next_framep[VP8_FRAME_PREVIOUS] = s->framep[VP8_FRAME_PREVIOUS];
2748 
2749  s->next_framep[VP8_FRAME_CURRENT] = curframe;
2750 
2751  if (!is_vp7 && !s->actually_webp)
2752  ff_thread_finish_setup(avctx);
2753 
2754  if (avctx->hwaccel) {
2755  const FFHWAccel *hwaccel = ffhwaccel(avctx->hwaccel);
2756  ret = hwaccel->start_frame(avctx, avpkt->data, avpkt->size);
2757  if (ret < 0)
2758  goto err;
2759 
2760  ret = hwaccel->decode_slice(avctx, avpkt->data, avpkt->size);
2761  if (ret < 0)
2762  goto err;
2763 
2764  ret = hwaccel->end_frame(avctx);
2765  if (ret < 0)
2766  goto err;
2767 
2768  } else {
2769  s->linesize = curframe->tf.f->linesize[0];
2770  s->uvlinesize = curframe->tf.f->linesize[1];
2771 
2772  memset(s->top_nnz, 0, s->mb_width * sizeof(*s->top_nnz));
2773  /* Zero macroblock structures for top/top-left prediction
2774  * from outside the frame. */
2775  if (!s->mb_layout)
2776  memset(s->macroblocks + s->mb_height * 2 - 1, 0,
2777  (s->mb_width + 1) * sizeof(*s->macroblocks));
2778  if (!s->mb_layout && s->keyframe)
2779  memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width * 4);
2780 
2781  memset(s->ref_count, 0, sizeof(s->ref_count));
2782 
2783  if (s->mb_layout == 1) {
2784  // Make sure the previous frame has read its segmentation map,
2785  // if we re-use the same map.
2786  if (prev_frame && s->segmentation.enabled &&
2787  !s->segmentation.update_map)
2788  ff_progress_frame_await(&prev_frame->tf, 1);
2789  if (is_vp7)
2790  ret = vp7_decode_mv_mb_modes(avctx, curframe, prev_frame);
2791  else
2792  ret = vp8_decode_mv_mb_modes(avctx, curframe, prev_frame);
2793  if (ret < 0)
2794  goto err;
2795  }
2796 
2797  if (avctx->active_thread_type == FF_THREAD_FRAME)
2798  num_jobs = 1;
2799  else
2800  num_jobs = FFMIN(s->num_coeff_partitions, avctx->thread_count);
2801  s->num_jobs = num_jobs;
2802  s->curframe = curframe;
2803  s->prev_frame = prev_frame;
2804  s->mv_bounds.mv_min.y = -MARGIN;
2805  s->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
2806  for (i = 0; i < MAX_THREADS; i++) {
2807  VP8ThreadData *td = &s->thread_data[i];
2808  atomic_init(&td->thread_mb_pos, 0);
2809  atomic_init(&td->wait_mb_pos, INT_MAX);
2810  }
2811  if (is_vp7)
2812  avctx->execute2(avctx, vp7_decode_mb_row_sliced, s->thread_data, NULL,
2813  num_jobs);
2814  else
2815  avctx->execute2(avctx, vp8_decode_mb_row_sliced, s->thread_data, NULL,
2816  num_jobs);
2817  }
2818 
2819  ff_progress_frame_report(&curframe->tf, INT_MAX);
2820  memcpy(&s->framep[0], &s->next_framep[0], sizeof(s->framep[0]) * 4);
2821 
2822 skip_decode:
2823  // if future frames don't use the updated probabilities,
2824  // reset them to the values we saved
2825  if (!s->update_probabilities)
2826  s->prob[0] = s->prob[1];
2827 
2828  if (!s->invisible) {
2829  if ((ret = av_frame_ref(rframe, curframe->tf.f)) < 0)
2830  return ret;
2831  *got_frame = 1;
2832  }
2833 
2834  return avpkt->size;
2835 err:
2836  memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4);
2837  return ret;
2838 }
2839 
2841  int *got_frame, AVPacket *avpkt)
2842 {
2843  return vp78_decode_frame(avctx, frame, got_frame, avpkt, IS_VP8);
2844 }
2845 
2846 #if CONFIG_VP7_DECODER
2847 static int vp7_decode_frame(AVCodecContext *avctx, AVFrame *frame,
2848  int *got_frame, AVPacket *avpkt)
2849 {
2850  return vp78_decode_frame(avctx, frame, got_frame, avpkt, IS_VP7);
2851 }
2852 #endif /* CONFIG_VP7_DECODER */
2853 
2855 {
2856  vp8_decode_flush_impl(avctx, 1);
2857 
2858  return 0;
2859 }
2860 
2861 static av_always_inline
2862 int vp78_decode_init(AVCodecContext *avctx, int is_vp7)
2863 {
2864  VP8Context *s = avctx->priv_data;
2865 
2866  s->avctx = avctx;
2867  s->pix_fmt = AV_PIX_FMT_NONE;
2868  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2869 
2870  ff_videodsp_init(&s->vdsp, 8);
2871 
2872  ff_vp78dsp_init(&s->vp8dsp);
2873  if (CONFIG_VP7_DECODER && is_vp7) {
2874  ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP7, 8, 1);
2875  ff_vp7dsp_init(&s->vp8dsp);
2876  s->decode_mb_row_no_filter = vp7_decode_mb_row_no_filter;
2877  s->filter_mb_row = vp7_filter_mb_row;
2878  } else if (CONFIG_VP8_DECODER && !is_vp7) {
2879  ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP8, 8, 1);
2880  ff_vp8dsp_init(&s->vp8dsp);
2881  s->decode_mb_row_no_filter = vp8_decode_mb_row_no_filter;
2882  s->filter_mb_row = vp8_filter_mb_row;
2883  }
2884 
2885  /* does not change for VP8 */
2886  memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan));
2887 
2888  return 0;
2889 }
2890 
2891 #if CONFIG_VP7_DECODER
2892 static int vp7_decode_init(AVCodecContext *avctx)
2893 {
2894  return vp78_decode_init(avctx, IS_VP7);
2895 }
2896 #endif /* CONFIG_VP7_DECODER */
2897 
2899 {
2900  return vp78_decode_init(avctx, IS_VP8);
2901 }
2902 
2903 #if CONFIG_VP8_DECODER
2904 #if HAVE_THREADS
2905 static void vp8_replace_frame(VP8Frame *dst, const VP8Frame *src)
2906 {
2907  ff_progress_frame_replace(&dst->tf, &src->tf);
2908  ff_refstruct_replace(&dst->seg_map, src->seg_map);
2909  ff_refstruct_replace(&dst->hwaccel_picture_private,
2910  src->hwaccel_picture_private);
2911 }
2912 
2913 #define REBASE(pic) ((pic) ? (pic) - &s_src->frames[0] + &s->frames[0] : NULL)
2914 
2915 static int vp8_decode_update_thread_context(AVCodecContext *dst,
2916  const AVCodecContext *src)
2917 {
2918  VP8Context *s = dst->priv_data, *s_src = src->priv_data;
2919 
2920  if (s->macroblocks_base &&
2921  (s_src->mb_width != s->mb_width || s_src->mb_height != s->mb_height)) {
2922  free_buffers(s);
2923  s->mb_width = s_src->mb_width;
2924  s->mb_height = s_src->mb_height;
2925  }
2926 
2927  s->pix_fmt = s_src->pix_fmt;
2928  s->prob[0] = s_src->prob[!s_src->update_probabilities];
2929  s->segmentation = s_src->segmentation;
2930  s->lf_delta = s_src->lf_delta;
2931  memcpy(s->sign_bias, s_src->sign_bias, sizeof(s->sign_bias));
2932 
2933  for (int i = 0; i < FF_ARRAY_ELEMS(s_src->frames); i++)
2934  vp8_replace_frame(&s->frames[i], &s_src->frames[i]);
2935 
2936  s->framep[0] = REBASE(s_src->next_framep[0]);
2937  s->framep[1] = REBASE(s_src->next_framep[1]);
2938  s->framep[2] = REBASE(s_src->next_framep[2]);
2939  s->framep[3] = REBASE(s_src->next_framep[3]);
2940 
2941  return 0;
2942 }
2943 #endif /* HAVE_THREADS */
2944 #endif /* CONFIG_VP8_DECODER */
2945 
2946 #if CONFIG_VP7_DECODER
2947 const FFCodec ff_vp7_decoder = {
2948  .p.name = "vp7",
2949  CODEC_LONG_NAME("On2 VP7"),
2950  .p.type = AVMEDIA_TYPE_VIDEO,
2951  .p.id = AV_CODEC_ID_VP7,
2952  .priv_data_size = sizeof(VP8Context),
2953  .init = vp7_decode_init,
2954  .close = ff_vp8_decode_free,
2955  FF_CODEC_DECODE_CB(vp7_decode_frame),
2956  .p.capabilities = AV_CODEC_CAP_DR1,
2957  .flush = vp8_decode_flush,
2958  .caps_internal = FF_CODEC_CAP_USES_PROGRESSFRAMES,
2959 };
2960 #endif /* CONFIG_VP7_DECODER */
2961 
2962 #if CONFIG_VP8_DECODER
2963 const FFCodec ff_vp8_decoder = {
2964  .p.name = "vp8",
2965  CODEC_LONG_NAME("On2 VP8"),
2966  .p.type = AVMEDIA_TYPE_VIDEO,
2967  .p.id = AV_CODEC_ID_VP8,
2968  .priv_data_size = sizeof(VP8Context),
2970  .close = ff_vp8_decode_free,
2972  .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS |
2974  .caps_internal = FF_CODEC_CAP_USES_PROGRESSFRAMES,
2975  .flush = vp8_decode_flush,
2976  UPDATE_THREAD_CONTEXT(vp8_decode_update_thread_context),
2977  .hw_configs = (const AVCodecHWConfigInternal *const []) {
2978 #if CONFIG_VP8_VAAPI_HWACCEL
2979  HWACCEL_VAAPI(vp8),
2980 #endif
2981 #if CONFIG_VP8_NVDEC_HWACCEL
2982  HWACCEL_NVDEC(vp8),
2983 #endif
2984  NULL
2985  },
2986 };
2987 #endif /* CONFIG_VP7_DECODER */
vp8_mode_contexts
static const int vp8_mode_contexts[6][4]
Definition: vp8data.h:118
VP8ThreadData::thread_mb_pos
atomic_int thread_mb_pos
Definition: vp8.h:144
hwconfig.h
vp8_dct_cat1_prob
static const uint8_t vp8_dct_cat1_prob[]
Definition: vp8data.h:336
ff_progress_frame_report
void ff_progress_frame_report(ProgressFrame *f, int n)
Notify later decoding threads when part of their reference frame is ready.
Definition: decode.c:1893
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:1259
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:1437
ff_vp8_decode_free
av_cold int ff_vp8_decode_free(AVCodecContext *avctx)
Definition: vp8.c:2854
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:1352
vp8_release_frame
static void vp8_release_frame(VP8Frame *f)
Definition: vp8.c:127
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:215
AV_PIX_FMT_CUDA
@ AV_PIX_FMT_CUDA
HW acceleration through CUDA.
Definition: pixfmt.h:260
AVPixelFormat
AVPixelFormat
Pixel format.
Definition: pixfmt.h:71
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:1446
vp7_read_mv_component
static int vp7_read_mv_component(VPXRangeCoder *c, const uint8_t *p)
Definition: vp8.c:905
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:1015
av_clip
#define av_clip
Definition: common.h:100
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:1561
VP8Macroblock::partitioning
uint8_t partitioning
Definition: vp8.h:102
VP8_FRAME_CURRENT
@ VP8_FRAME_CURRENT
Definition: vp8.h:45
r
const char * r
Definition: vf_curves.c:127
vp7_filter_mb_row
static void vp7_filter_mb_row(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2586
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:103
AVCodecContext::colorspace
enum AVColorSpace colorspace
YUV colorspace type.
Definition: avcodec.h:691
vp78_decode_init
static av_always_inline int vp78_decode_init(AVCodecContext *avctx, int is_vp7)
Definition: vp8.c:2862
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:1277
u
#define u(width, name, range_min, range_max)
Definition: cbs_h2645.c:251
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:1616
vp8_submv_prob
static const uint8_t vp8_submv_prob[5][3]
Definition: vp8data.h:153
VP8Frame::tf
ProgressFrame tf
Definition: vp8.h:154
av_clip_uintp2
#define av_clip_uintp2
Definition: common.h:124
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:585
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:81
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:1034
vp7_mv_default_prob
static const uint8_t vp7_mv_default_prob[2][17]
Definition: vp8data.h:551
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:1651
ff_vp8_token_update_probs
const uint8_t ff_vp8_token_update_probs[4][8][3][11]
Definition: vp8data.c:43
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:1641
vp7_y2dc_qlookup
static const uint16_t vp7_y2dc_qlookup[]
Definition: vp8data.h:610
vp8_mc_chroma
static av_always_inline void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1, uint8_t *dst2, const ProgressFrame *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:1883
mode
Definition: swscale.c:52
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:389
TM_VP8_PRED
@ TM_VP8_PRED
Definition: vp9.h:55
AVCOL_RANGE_JPEG
@ AVCOL_RANGE_JPEG
Full range content.
Definition: pixfmt.h:717
AVPacket::data
uint8_t * data
Definition: packet.h:539
inter_predict_dc
static av_always_inline int inter_predict_dc(int16_t block[16], int16_t pred[2])
Definition: vp8.c:1412
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:462
ff_progress_frame_get_buffer
int ff_progress_frame_get_buffer(AVCodecContext *avctx, ProgressFrame *f, int flags)
This function sets up the ProgressFrame, i.e.
Definition: decode.c:1848
VERT_LEFT_PRED
@ VERT_LEFT_PRED
Definition: vp9.h:53
vp8_get_quants
static void vp8_get_quants(VP8Context *s)
Definition: vp8.c:375
VP8intmv::y
int y
Definition: vp8.h:113
FFCodec
Definition: codec_internal.h:127
AV_WN32A
#define AV_WN32A(p, v)
Definition: intreadwrite.h:534
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:116
vp8_decode_flush
static av_cold void vp8_decode_flush(AVCodecContext *avctx)
Definition: vp8.c:150
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:1994
VP8_SPLITMVMODE_4x4
@ VP8_SPLITMVMODE_4x4
4x4 blocks of 4x4px each
Definition: vp8.h:81
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:48
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
AVFrame::data
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:410
VP8_MVC_SIZE
#define VP8_MVC_SIZE
Definition: vp8.c:463
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:2521
vp8_rac_get_sint
static int vp8_rac_get_sint(VPXRangeCoder *c, int bits)
Definition: vp8.c:51
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:2379
vp8.h
get_bmv_ptr
static const VP8mv * get_bmv_ptr(const VP8Macroblock *mb, int subblock)
Definition: vp8.c:1028
update_dimensions
static av_always_inline int update_dimensions(VP8Context *s, int width, int height, int is_vp7)
Definition: vp8.c:196
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:646
VP8_SPLITMVMODE_8x8
@ VP8_SPLITMVMODE_8x8
2x2 blocks of 8x8px each
Definition: vp8.h:80
mx
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t mx
Definition: dsp.h:53
vp8_decode_flush_impl
static av_cold void vp8_decode_flush_impl(AVCodecContext *avctx, int free_mem)
Definition: vp8.c:134
IS_VP8
#define IS_VP8(avctx)
Definition: libvpxenc.c:53
FFCodec::p
AVCodec p
The public AVCodec.
Definition: codec_internal.h:131
FFHWAccel
Definition: hwaccel_internal.h:34
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:540
AVCodecContext::skip_frame
enum AVDiscard skip_frame
Skip decoding for selected frames.
Definition: avcodec.h:1830
fail
#define fail()
Definition: checkasm.h:193
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:1593
VP8ThreadData::non_zero_count_cache
uint8_t non_zero_count_cache[6][4]
This is the index plus one of the last non-zero coeff for each of the blocks in the current macrobloc...
Definition: vp8.h:131
ff_vp8_decoder
const FFCodec ff_vp8_decoder
VP8mv::y
int16_t y
Definition: vp8.h:87
DIAG_DOWN_RIGHT_PRED
@ DIAG_DOWN_RIGHT_PRED
Definition: vp9.h:50
MAX_THREADS
#define MAX_THREADS
Definition: frame_thread_encoder.c:37
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:78
VP8Macroblock::bmv
VP8mv bmv[16]
Definition: vp8.h:108
idct_mb
static av_always_inline void idct_mb(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], const VP8Macroblock *mb)
Definition: vp8.c:2078
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:1625
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:2141
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:877
progressframe.h
vp8_filter_mb_row
static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2592
vp7_get_quants
static void vp7_get_quants(VP8Context *s)
Definition: vp8.c:356
refstruct.h
VP8_SPLITMVMODE_16x8
@ VP8_SPLITMVMODE_16x8
2 16x8 blocks (vertical)
Definition: vp8.h:78
FF_CODEC_CAP_USES_PROGRESSFRAMES
#define FF_CODEC_CAP_USES_PROGRESSFRAMES
The decoder might make use of the ProgressFrame API.
Definition: codec_internal.h:69
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:209
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:640
LOCAL_ALIGNED
#define LOCAL_ALIGNED(a, t, v,...)
Definition: mem_internal.h:122
VP8ThreadData::left_nnz
uint8_t left_nnz[9]
For coeff decode, we need to know whether the above block had non-zero coefficients.
Definition: vp8.h:138
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:311
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:2251
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:2261
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:666
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:662
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:431
vp8_mc_luma
static av_always_inline void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst, const ProgressFrame *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:1825
VP8ThreadData::filter_strength
VP8FilterStrength * filter_strength
Definition: vp8.h:149
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:279
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:49
ff_progress_frame_unref
void ff_progress_frame_unref(ProgressFrame *f)
Give up a reference to the underlying frame contained in a ProgressFrame and reset the ProgressFrame,...
Definition: decode.c:1876
ff_progress_frame_await
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_progress_frame_await() 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_progress_frame_report() has been called on them. This includes draw_edges(). Porting codecs to frame threading
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:2646
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:2295
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:73
atomic_load
#define atomic_load(object)
Definition: stdatomic.h:93
VP8_SPLITMVMODE_8x16
@ VP8_SPLITMVMODE_8x16
2 8x16 blocks (horizontal)
Definition: vp8.h:79
CODEC_LONG_NAME
#define CODEC_LONG_NAME(str)
Definition: codec_internal.h:296
VP8Frame::seg_map
uint8_t * seg_map
RefStruct reference.
Definition: vp8.h:155
my
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t intptr_t my
Definition: dsp.h:53
vp8_mc_part
static av_always_inline void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], const ProgressFrame *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:1933
vp8_mv_default_prob
static const uint8_t vp8_mv_default_prob[2][19]
Definition: vp8data.h:562
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
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:221
vp8_rac_get_nn
static int vp8_rac_get_nn(VPXRangeCoder *c)
Definition: vp8.c:66
clamp_mv
static av_always_inline void clamp_mv(const VP8mvbounds *s, VP8mv *dst, const VP8mv *src)
Definition: vp8.c:866
NULL
#define NULL
Definition: coverity.c:32
sizes
static const int sizes[][2]
Definition: img2dec.c:60
AV_COPY128
#define AV_COPY128(d, s)
Definition: intreadwrite.h:642
AVCodecContext::color_range
enum AVColorRange color_range
MPEG vs JPEG YUV range.
Definition: avcodec.h:701
vp7_decode_mb_row_sliced
static int vp7_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2633
AV_COPY64
#define AV_COPY64(d, s)
Definition: intreadwrite.h:638
hwaccel_internal.h
vp8_update_dimensions
static int vp8_update_dimensions(VP8Context *s, int width, int height)
Definition: vp8.c:273
VP8FilterStrength
Definition: vp8.h:90
NUM_DCT_TOKENS
@ NUM_DCT_TOKENS
Definition: vp8.h:65
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:2378
VP7MVPred
Definition: vp8data.h:61
mathops.h
flush
void(* flush)(AVBSFContext *ctx)
Definition: dts2pts.c:370
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:597
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_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:305
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:1124
AV_ZERO128
#define AV_ZERO128(d)
Definition: intreadwrite.h:670
FF_HW_HAS_CB
#define FF_HW_HAS_CB(avctx, function)
Definition: hwaccel_internal.h:177
VP8FrameType
VP8FrameType
Definition: vp8.h:43
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:2382
VP8mv
Definition: vp8.h:85
vp7_feature_value_size
static const uint8_t vp7_feature_value_size[2][4]
Definition: vp8data.h:573
index
int index
Definition: gxfenc.c:90
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:153
vp8.h
ff_vp8_decode_init
av_cold int ff_vp8_decode_init(AVCodecContext *avctx)
Definition: vp8.c:2898
VP8_FRAME_GOLDEN
@ VP8_FRAME_GOLDEN
Definition: vp8.h:47
FF_SIGNBIT
#define FF_SIGNBIT(x)
Definition: mathops.h:128
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:220
f
f
Definition: af_crystalizer.c:122
init
int(* init)(AVBSFContext *ctx)
Definition: dts2pts.c:368
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:1573
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:540
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:186
height
#define height
Definition: dsp.h:85
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:400
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:1801
dst
uint8_t ptrdiff_t const uint8_t ptrdiff_t int intptr_t intptr_t int int16_t * dst
Definition: dsp.h:83
VP8ThreadData::thread_nr
int thread_nr
Definition: vp8.h:139
vp7_update_dimensions
static int vp7_update_dimensions(VP8Context *s, int width, int height)
Definition: vp8.c:268
EDGE_EMU_LINESIZE
#define EDGE_EMU_LINESIZE
Definition: vp8.h:147
size
int size
Definition: twinvq_data.h:10344
VERT_RIGHT_PRED
@ VERT_RIGHT_PRED
Definition: vp9.h:51
free_buffers
static void free_buffers(VP8Context *s)
Definition: vp8.c:84
ref_frame
static int ref_frame(VVCFrame *dst, const VVCFrame *src)
Definition: dec.c:593
DC_128_PRED8x8
#define DC_128_PRED8x8
Definition: h264pred.h:76
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:1471
FF_THREAD_SLICE
#define FF_THREAD_SLICE
Decode more than one part of a single frame at once.
Definition: avcodec.h:1605
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:2338
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
VP8ThreadData::mv_bounds
VP8mvbounds mv_bounds
Definition: vp8.h:150
vp8_pred4x4_prob_intra
static const uint8_t vp8_pred4x4_prob_intra[10][10][9]
Definition: vp8data.h:196
VP8ThreadData
Definition: vp8.h:121
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:725
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:325
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:578
AVCodecContext::skip_loop_filter
enum AVDiscard skip_loop_filter
Skip loop filtering for selected frames.
Definition: avcodec.h:1816
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:126
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:69
VP8ThreadData::edge_emu_buffer
uint8_t edge_emu_buffer[21 *EDGE_EMU_LINESIZE]
Definition: vp8.h:148
FF_THREAD_FRAME
#define FF_THREAD_FRAME
Decode more than one frame at once.
Definition: avcodec.h:1604
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:2599
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:235
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:421
vp8_read_mv_component
static int vp8_read_mv_component(VPXRangeCoder *c, const uint8_t *p)
Definition: vp8.c:910
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:256
DC_129_PRED
@ DC_129_PRED
Definition: vp9.h:60
modes
const struct mode modes[]
Definition: swscale.c:57
VP8ThreadData::wait_mb_pos
atomic_int wait_mb_pos
Definition: vp8.h:145
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:523
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:1687
AV_COPY32
#define AV_COPY32(d, s)
Definition: intreadwrite.h:634
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:256
AVCodec::name
const char * name
Name of the codec implementation.
Definition: codec.h:194
VP8mv::x
int16_t x
Definition: vp8.h:86
vp78_reset_probability_tables
static void vp78_reset_probability_tables(VP8Context *s)
Definition: vp8.c:437
vp7_decode_frame_header
static int vp7_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
Definition: vp8.c:561
VP8mvbounds::mv_max
VP8intmv mv_max
Definition: vp8.h:118
VP8_FRAME_NONE
@ VP8_FRAME_NONE
Definition: vp8.h:44
profile
int profile
Definition: mxfenc.c:2233
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:507
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:1224
VP8Macroblock
Definition: vp8.h:96
AVCodecContext::pix_fmt
enum AVPixelFormat pix_fmt
Pixel format, see AV_PIX_FMT_xxx.
Definition: avcodec.h:663
AVCOL_RANGE_MPEG
@ AVCOL_RANGE_MPEG
Narrow or limited range content.
Definition: pixfmt.h:700
vp8_decode_mb_row_sliced
static int vp8_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr)
Definition: vp8.c:2639
ff_vp7_decoder
const FFCodec ff_vp7_decoder
VP8_SPLITMVMODE_NONE
@ VP8_SPLITMVMODE_NONE
(only used in prediction) no split MVs
Definition: vp8.h:82
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:1971
avcodec.h
AV_RN32A
#define AV_RN32A(p)
Definition: intreadwrite.h:522
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:1486
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:203
check_dc_pred8x8_mode
static av_always_inline int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y)
Definition: vp8.c:1607
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:2515
FFSWAP
#define FFSWAP(type, a, b)
Definition: macros.h:52
frame
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several the filter must be ready for frames arriving randomly on any input any filter with several inputs will most likely require some kind of queuing mechanism It is perfectly acceptable to have a limited queue and to drop frames when the inputs are too unbalanced request_frame For filters that do not use the this method is called when a frame is wanted on an output For a it should directly call filter_frame on the corresponding output For a if there are queued frames already one of these frames should be pushed If the filter should request a frame on one of its repeatedly until at least one frame has been pushed Return or at least make progress towards producing a frame
Definition: filter_design.txt:264
vp8_pred16x16_prob_intra
static const uint8_t vp8_pred16x16_prob_intra[4]
Definition: vp8data.h:161
ProgressFrame::f
struct AVFrame * f
Definition: progressframe.h:74
vp8_ac_qlookup
static const uint16_t vp8_ac_qlookup[VP8_MAX_QUANT+1]
Definition: vp8data.h:529
prob
#define prob(name, subs,...)
Definition: cbs_vp9.c:325
hwaccel
static const char * hwaccel
Definition: ffplay.c:353
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:160
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
ff_progress_frame_replace
void ff_progress_frame_replace(ProgressFrame *dst, const ProgressFrame *src)
Do nothing if dst and src already refer to the same AVFrame; otherwise unreference dst and if src is ...
Definition: decode.c:1883
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:451
AVCodecContext::active_thread_type
int active_thread_type
Which multithreading methods are in use by the codec.
Definition: avcodec.h:1612
VP8Macroblock::intra4x4_pred_mode_top
uint8_t intra4x4_pred_mode_top[4]
Definition: vp8.h:106
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:933
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:72
ffhwaccel
static const FFHWAccel * ffhwaccel(const AVHWAccel *codec)
Definition: hwaccel_internal.h:166
VP8Macroblock::mode
uint8_t mode
Definition: vp8.h:100
VP8intmv::x
int x
Definition: vp8.h:112
VP8_MVMODE_SPLIT
@ VP8_MVMODE_SPLIT
Definition: vp8.h:74
ref
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:117
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:1434
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:2174
segment
Definition: hls.c:77
av_clip_uint8
#define av_clip_uint8
Definition: common.h:106
VP8mvbounds::mv_min
VP8intmv mv_min
Definition: vp8.h:117
vp78_update_probability_tables
static void vp78_update_probability_tables(VP8Context *s)
Definition: vp8.c:446
AV_PICTURE_TYPE_P
@ AV_PICTURE_TYPE_P
Predicted.
Definition: avutil.h:280
AVMEDIA_TYPE_VIDEO
@ AVMEDIA_TYPE_VIDEO
Definition: avutil.h:201
mem.h
vp78_update_pred16x16_pred8x8_mvc_probabilities
static void vp78_update_pred16x16_pred8x8_mvc_probabilities(VP8Context *s, int mvc_size)
Definition: vp8.c:465
avpriv_request_sample
#define avpriv_request_sample(...)
Definition: tableprint_vlc.h:36
update_refs
static void update_refs(VP8Context *s)
Definition: vp8.c:485
vp7_decode_mv_mb_modes
static int vp7_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame, const VP8Frame *prev_frame)
Definition: vp8.c:2332
update_lf_deltas
static void update_lf_deltas(VP8Context *s)
Definition: vp8.c:301
VP8ThreadData::block
int16_t block[6][4][16]
Definition: vp8.h:122
ProgressFrame
The ProgressFrame structure.
Definition: progressframe.h:73
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:2527
alpha
static const int16_t alpha[]
Definition: ilbcdata.h:55
AVPacket
This structure stores compressed data.
Definition: packet.h:516
AVCodecContext::priv_data
void * priv_data
Definition: avcodec.h:478
get_pixel_format
static enum AVPixelFormat get_pixel_format(VP8Context *s)
Definition: vp8.c:179
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:2840
VP7MVPred::score
uint8_t score
Definition: vp8data.h:65
VP8Context
Definition: vp8.h:161
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:155
int32_t
int32_t
Definition: audioconvert.c:56
AV_CODEC_ID_VP8
@ AV_CODEC_ID_VP8
Definition: codec_id.h:192
coeff
static const double coeff[2][5]
Definition: vf_owdenoise.c:80
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:376
VP8_MVMODE_ZERO
@ VP8_MVMODE_ZERO
Definition: vp8.h:72
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:623
atomic_init
#define atomic_init(obj, value)
Definition: stdatomic.h:33
width
#define width
Definition: dsp.h:85
vp7_submv_prob
static const uint8_t vp7_submv_prob[3]
Definition: vp8data.h:149
AVDiscard
AVDiscard
Definition: defs.h:212
AVDISCARD_NONREF
@ AVDISCARD_NONREF
discard all non reference
Definition: defs.h:217
VP8ThreadData::block_dc
int16_t block_dc[16]
Definition: vp8.h:123
vp8_rac_get_coeff
static int vp8_rac_get_coeff(VPXRangeCoder *c, const uint8_t *prob)
Definition: vp8.c:73
vp8_dc_qlookup
static const uint8_t vp8_dc_qlookup[VP8_MAX_QUANT+1]
Definition: vp8data.h:518
copy_chroma
static void copy_chroma(AVFrame *dst, const AVFrame *src, int width, int height)
Definition: vp8.c:496
VP8_FRAME_PREVIOUS
@ VP8_FRAME_PREVIOUS
Definition: vp8.h:46
vpx_rac_get_prob
#define vpx_rac_get_prob
Definition: vpx_rac.h:82
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:1642
VP8_MVMODE_MV
@ VP8_MVMODE_MV
Definition: vp8.h:73
MARGIN
#define MARGIN
Definition: vp8.c:2293
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:120
vp8_alloc_frame
static int vp8_alloc_frame(VP8Context *s, VP8Frame *f, int ref)
Definition: vp8.c:104
src
#define src
Definition: vp8dsp.c:248
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:916
ff_vp78dsp_init
av_cold void ff_vp78dsp_init(VP8DSPContext *dsp)
Definition: vp8dsp.c:668