utvideo.c
Go to the documentation of this file.
1 /*
2  * Ut Video decoder
3  * Copyright (c) 2011 Konstantin Shishkov
4  *
5  * This file is part of Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
27 #include <stdlib.h>
28 
29 #include "libavutil/intreadwrite.h"
30 #include "avcodec.h"
31 #include "bytestream.h"
32 #include "get_bits.h"
33 #include "dsputil.h"
34 #include "thread.h"
35 
36 enum {
37  PRED_NONE = 0,
41 };
42 
43 typedef struct UtvideoContext {
47 
49  int planes;
50  int slices;
54 
55  uint8_t *slice_bits;
58 
59 typedef struct HuffEntry {
60  uint8_t sym;
61  uint8_t len;
62 } HuffEntry;
63 
64 static int huff_cmp(const void *a, const void *b)
65 {
66  const HuffEntry *aa = a, *bb = b;
67  return (aa->len - bb->len)*256 + aa->sym - bb->sym;
68 }
69 
70 static int build_huff(const uint8_t *src, VLC *vlc, int *fsym)
71 {
72  int i;
73  HuffEntry he[256];
74  int last;
75  uint32_t codes[256];
76  uint8_t bits[256];
77  uint8_t syms[256];
78  uint32_t code;
79 
80  *fsym = -1;
81  for (i = 0; i < 256; i++) {
82  he[i].sym = i;
83  he[i].len = *src++;
84  }
85  qsort(he, 256, sizeof(*he), huff_cmp);
86 
87  if (!he[0].len) {
88  *fsym = he[0].sym;
89  return 0;
90  }
91  if (he[0].len > 32)
92  return -1;
93 
94  last = 255;
95  while (he[last].len == 255 && last)
96  last--;
97 
98  code = 1;
99  for (i = last; i >= 0; i--) {
100  codes[i] = code >> (32 - he[i].len);
101  bits[i] = he[i].len;
102  syms[i] = he[i].sym;
103  code += 0x80000000u >> (he[i].len - 1);
104  }
105 
106  return ff_init_vlc_sparse(vlc, FFMIN(he[last].len, 9), last + 1,
107  bits, sizeof(*bits), sizeof(*bits),
108  codes, sizeof(*codes), sizeof(*codes),
109  syms, sizeof(*syms), sizeof(*syms), 0);
110 }
111 
112 static int decode_plane(UtvideoContext *c, int plane_no,
113  uint8_t *dst, int step, int stride,
114  int width, int height,
115  const uint8_t *src, int src_size, int use_pred)
116 {
117  int i, j, slice, pix;
118  int sstart, send;
119  VLC vlc;
120  GetBitContext gb;
121  int prev, fsym;
122  const int cmask = ~(!plane_no && c->avctx->pix_fmt == PIX_FMT_YUV420P);
123 
124  if (build_huff(src, &vlc, &fsym)) {
125  av_log(c->avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
126  return AVERROR_INVALIDDATA;
127  }
128  if (fsym >= 0) { // build_huff reported a symbol to fill slices with
129  send = 0;
130  for (slice = 0; slice < c->slices; slice++) {
131  uint8_t *dest;
132 
133  sstart = send;
134  send = (height * (slice + 1) / c->slices) & cmask;
135  dest = dst + sstart * stride;
136 
137  prev = 0x80;
138  for (j = sstart; j < send; j++) {
139  for (i = 0; i < width * step; i += step) {
140  pix = fsym;
141  if (use_pred) {
142  prev += pix;
143  pix = prev;
144  }
145  dest[i] = pix;
146  }
147  dest += stride;
148  }
149  }
150  return 0;
151  }
152 
153  src += 256;
154  src_size -= 256;
155 
156  send = 0;
157  for (slice = 0; slice < c->slices; slice++) {
158  uint8_t *dest;
159  int slice_data_start, slice_data_end, slice_size;
160 
161  sstart = send;
162  send = (height * (slice + 1) / c->slices) & cmask;
163  dest = dst + sstart * stride;
164 
165  // slice offset and size validation was done earlier
166  slice_data_start = slice ? AV_RL32(src + slice * 4 - 4) : 0;
167  slice_data_end = AV_RL32(src + slice * 4);
168  slice_size = slice_data_end - slice_data_start;
169 
170  if (!slice_size) {
171  for (j = sstart; j < send; j++) {
172  for (i = 0; i < width * step; i += step)
173  dest[i] = 0x80;
174  dest += stride;
175  }
176  continue;
177  }
178 
179  memcpy(c->slice_bits, src + slice_data_start + c->slices * 4, slice_size);
180  memset(c->slice_bits + slice_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
181  c->dsp.bswap_buf((uint32_t*)c->slice_bits, (uint32_t*)c->slice_bits,
182  (slice_data_end - slice_data_start + 3) >> 2);
183  init_get_bits(&gb, c->slice_bits, slice_size * 8);
184 
185  prev = 0x80;
186  for (j = sstart; j < send; j++) {
187  for (i = 0; i < width * step; i += step) {
188  if (get_bits_left(&gb) <= 0) {
189  av_log(c->avctx, AV_LOG_ERROR, "Slice decoding ran out of bits\n");
190  goto fail;
191  }
192  pix = get_vlc2(&gb, vlc.table, vlc.bits, 4);
193  if (pix < 0) {
194  av_log(c->avctx, AV_LOG_ERROR, "Decoding error\n");
195  goto fail;
196  }
197  if (use_pred) {
198  prev += pix;
199  pix = prev;
200  }
201  dest[i] = pix;
202  }
203  dest += stride;
204  }
205  if (get_bits_left(&gb) > 32)
206  av_log(c->avctx, AV_LOG_WARNING, "%d bits left after decoding slice\n",
207  get_bits_left(&gb));
208  }
209 
210  ff_free_vlc(&vlc);
211 
212  return 0;
213 fail:
214  ff_free_vlc(&vlc);
215  return AVERROR_INVALIDDATA;
216 }
217 
218 static const int rgb_order[4] = { 1, 2, 0, 3 };
219 
220 static void restore_rgb_planes(uint8_t *src, int step, int stride, int width, int height)
221 {
222  int i, j;
223  uint8_t r, g, b;
224 
225  for (j = 0; j < height; j++) {
226  for (i = 0; i < width * step; i += step) {
227  r = src[i];
228  g = src[i + 1];
229  b = src[i + 2];
230  src[i] = r + g - 0x80;
231  src[i + 2] = b + g - 0x80;
232  }
233  src += stride;
234  }
235 }
236 
237 static void restore_median(uint8_t *src, int step, int stride,
238  int width, int height, int slices, int rmode)
239 {
240  int i, j, slice;
241  int A, B, C;
242  uint8_t *bsrc;
243  int slice_start, slice_height;
244  const int cmask = ~rmode;
245 
246  for (slice = 0; slice < slices; slice++) {
247  slice_start = ((slice * height) / slices) & cmask;
248  slice_height = ((((slice + 1) * height) / slices) & cmask) - slice_start;
249 
250  bsrc = src + slice_start * stride;
251 
252  // first line - left neighbour prediction
253  bsrc[0] += 0x80;
254  A = bsrc[0];
255  for (i = step; i < width * step; i += step) {
256  bsrc[i] += A;
257  A = bsrc[i];
258  }
259  bsrc += stride;
260  if (slice_height == 1)
261  continue;
262  // second line - first element has top predition, the rest uses median
263  C = bsrc[-stride];
264  bsrc[0] += C;
265  A = bsrc[0];
266  for (i = step; i < width * step; i += step) {
267  B = bsrc[i - stride];
268  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
269  C = B;
270  A = bsrc[i];
271  }
272  bsrc += stride;
273  // the rest of lines use continuous median prediction
274  for (j = 2; j < slice_height; j++) {
275  for (i = 0; i < width * step; i += step) {
276  B = bsrc[i - stride];
277  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
278  C = B;
279  A = bsrc[i];
280  }
281  bsrc += stride;
282  }
283  }
284 }
285 
286 /* UtVideo interlaced mode treats every two lines as a single one,
287  * so restoring function should take care of possible padding between
288  * two parts of the same "line".
289  */
290 static void restore_median_il(uint8_t *src, int step, int stride,
291  int width, int height, int slices, int rmode)
292 {
293  int i, j, slice;
294  int A, B, C;
295  uint8_t *bsrc;
296  int slice_start, slice_height;
297  const int cmask = ~(rmode ? 3 : 1);
298  const int stride2 = stride << 1;
299 
300  for (slice = 0; slice < slices; slice++) {
301  slice_start = ((slice * height) / slices) & cmask;
302  slice_height = ((((slice + 1) * height) / slices) & cmask) - slice_start;
303  slice_height >>= 1;
304 
305  bsrc = src + slice_start * stride;
306 
307  // first line - left neighbour prediction
308  bsrc[0] += 0x80;
309  A = bsrc[0];
310  for (i = step; i < width * step; i += step) {
311  bsrc[i] += A;
312  A = bsrc[i];
313  }
314  for (i = 0; i < width * step; i += step) {
315  bsrc[stride + i] += A;
316  A = bsrc[stride + i];
317  }
318  bsrc += stride2;
319  if (slice_height == 1)
320  continue;
321  // second line - first element has top predition, the rest uses median
322  C = bsrc[-stride2];
323  bsrc[0] += C;
324  A = bsrc[0];
325  for (i = step; i < width * step; i += step) {
326  B = bsrc[i - stride2];
327  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
328  C = B;
329  A = bsrc[i];
330  }
331  for (i = 0; i < width * step; i += step) {
332  B = bsrc[i - stride];
333  bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
334  C = B;
335  A = bsrc[stride + i];
336  }
337  bsrc += stride2;
338  // the rest of lines use continuous median prediction
339  for (j = 2; j < slice_height; j++) {
340  for (i = 0; i < width * step; i += step) {
341  B = bsrc[i - stride2];
342  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
343  C = B;
344  A = bsrc[i];
345  }
346  for (i = 0; i < width * step; i += step) {
347  B = bsrc[i - stride];
348  bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
349  C = B;
350  A = bsrc[i + stride];
351  }
352  bsrc += stride2;
353  }
354  }
355 }
356 
357 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt)
358 {
359  const uint8_t *buf = avpkt->data;
360  int buf_size = avpkt->size;
361  UtvideoContext *c = avctx->priv_data;
362  int i, j;
363  const uint8_t *plane_start[5];
364  int plane_size, max_slice_size = 0, slice_start, slice_end, slice_size;
365  int ret;
366  GetByteContext gb;
367 
368  if (c->pic.data[0])
369  ff_thread_release_buffer(avctx, &c->pic);
370 
371  c->pic.reference = 1;
373  if ((ret = ff_thread_get_buffer(avctx, &c->pic)) < 0) {
374  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
375  return ret;
376  }
377 
378  ff_thread_finish_setup(avctx);
379 
380  /* parse plane structure to retrieve frame flags and validate slice offsets */
381  bytestream2_init(&gb, buf, buf_size);
382  for (i = 0; i < c->planes; i++) {
383  plane_start[i] = gb.buffer;
384  if (bytestream2_get_bytes_left(&gb) < 256 + 4 * c->slices) {
385  av_log(avctx, AV_LOG_ERROR, "Insufficient data for a plane\n");
386  return AVERROR_INVALIDDATA;
387  }
388  bytestream2_skipu(&gb, 256);
389  slice_start = 0;
390  slice_end = 0;
391  for (j = 0; j < c->slices; j++) {
392  slice_end = bytestream2_get_le32u(&gb);
393  slice_size = slice_end - slice_start;
394  if (slice_end <= 0 || slice_size <= 0 ||
395  bytestream2_get_bytes_left(&gb) < slice_end) {
396  av_log(avctx, AV_LOG_ERROR, "Incorrect slice size\n");
397  return AVERROR_INVALIDDATA;
398  }
399  slice_start = slice_end;
400  max_slice_size = FFMAX(max_slice_size, slice_size);
401  }
402  plane_size = slice_end;
403  bytestream2_skipu(&gb, plane_size);
404  }
405  plane_start[c->planes] = gb.buffer;
407  av_log(avctx, AV_LOG_ERROR, "Not enough data for frame information\n");
408  return AVERROR_INVALIDDATA;
409  }
410  c->frame_info = bytestream2_get_le32u(&gb);
411  av_log(avctx, AV_LOG_DEBUG, "frame information flags %X\n", c->frame_info);
412 
413  c->frame_pred = (c->frame_info >> 8) & 3;
414 
415  if (c->frame_pred == PRED_GRADIENT) {
416  av_log_ask_for_sample(avctx, "Frame uses gradient prediction\n");
417  return AVERROR_PATCHWELCOME;
418  }
419 
421  max_slice_size + FF_INPUT_BUFFER_PADDING_SIZE);
422 
423  if (!c->slice_bits) {
424  av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer\n");
425  return AVERROR(ENOMEM);
426  }
427 
428  switch (c->avctx->pix_fmt) {
429  case PIX_FMT_RGB24:
430  case PIX_FMT_RGBA:
431  for (i = 0; i < c->planes; i++) {
432  ret = decode_plane(c, i, c->pic.data[0] + rgb_order[i], c->planes,
433  c->pic.linesize[0], avctx->width, avctx->height,
434  plane_start[i], plane_start[i + 1] - plane_start[i],
435  c->frame_pred == PRED_LEFT);
436  if (ret)
437  return ret;
438  if (c->frame_pred == PRED_MEDIAN)
439  restore_median(c->pic.data[0] + rgb_order[i], c->planes,
440  c->pic.linesize[0], avctx->width, avctx->height,
441  c->slices, 0);
442  }
443  restore_rgb_planes(c->pic.data[0], c->planes, c->pic.linesize[0],
444  avctx->width, avctx->height);
445  break;
446  case PIX_FMT_YUV420P:
447  for (i = 0; i < 3; i++) {
448  ret = decode_plane(c, i, c->pic.data[i], 1,
449  c->pic.linesize[i], avctx->width >> !!i, avctx->height >> !!i,
450  plane_start[i], plane_start[i + 1] - plane_start[i],
451  c->frame_pred == PRED_LEFT);
452  if (ret)
453  return ret;
454  if (c->frame_pred == PRED_MEDIAN) {
455  if (!c->interlaced) {
456  restore_median(c->pic.data[i], 1, c->pic.linesize[i],
457  avctx->width >> !!i, avctx->height >> !!i,
458  c->slices, !i);
459  } else {
460  restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
461  avctx->width >> !!i,
462  avctx->height >> !!i,
463  c->slices, !i);
464  }
465  }
466  }
467  break;
468  case PIX_FMT_YUV422P:
469  for (i = 0; i < 3; i++) {
470  ret = decode_plane(c, i, c->pic.data[i], 1,
471  c->pic.linesize[i], avctx->width >> !!i, avctx->height,
472  plane_start[i], plane_start[i + 1] - plane_start[i],
473  c->frame_pred == PRED_LEFT);
474  if (ret)
475  return ret;
476  if (c->frame_pred == PRED_MEDIAN) {
477  if (!c->interlaced) {
478  restore_median(c->pic.data[i], 1, c->pic.linesize[i],
479  avctx->width >> !!i, avctx->height,
480  c->slices, 0);
481  } else {
482  restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
483  avctx->width >> !!i, avctx->height,
484  c->slices, 0);
485  }
486  }
487  }
488  break;
489  }
490 
491  *data_size = sizeof(AVFrame);
492  *(AVFrame*)data = c->pic;
493 
494  /* always report that the buffer was completely consumed */
495  return buf_size;
496 }
497 
499 {
500  UtvideoContext * const c = avctx->priv_data;
501 
502  c->avctx = avctx;
503 
504  dsputil_init(&c->dsp, avctx);
505 
506  if (avctx->extradata_size < 16) {
507  av_log(avctx, AV_LOG_ERROR, "Insufficient extradata size %d, should be at least 16\n",
508  avctx->extradata_size);
509  return AVERROR_INVALIDDATA;
510  }
511 
512  av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",
513  avctx->extradata[3], avctx->extradata[2],
514  avctx->extradata[1], avctx->extradata[0]);
515  av_log(avctx, AV_LOG_DEBUG, "Original format %X\n", AV_RB32(avctx->extradata + 4));
516  c->frame_info_size = AV_RL32(avctx->extradata + 8);
517  c->flags = AV_RL32(avctx->extradata + 12);
518 
519  if (c->frame_info_size != 4)
520  av_log_ask_for_sample(avctx, "Frame info is not 4 bytes\n");
521  av_log(avctx, AV_LOG_DEBUG, "Encoding parameters %08X\n", c->flags);
522  c->slices = (c->flags >> 24) + 1;
523  c->compression = c->flags & 1;
524  c->interlaced = c->flags & 0x800;
525 
526  c->slice_bits_size = 0;
527 
528  switch (avctx->codec_tag) {
529  case MKTAG('U', 'L', 'R', 'G'):
530  c->planes = 3;
531  avctx->pix_fmt = PIX_FMT_RGB24;
532  break;
533  case MKTAG('U', 'L', 'R', 'A'):
534  c->planes = 4;
535  avctx->pix_fmt = PIX_FMT_RGBA;
536  break;
537  case MKTAG('U', 'L', 'Y', '0'):
538  c->planes = 3;
539  avctx->pix_fmt = PIX_FMT_YUV420P;
540  break;
541  case MKTAG('U', 'L', 'Y', '2'):
542  c->planes = 3;
543  avctx->pix_fmt = PIX_FMT_YUV422P;
544  break;
545  default:
546  av_log(avctx, AV_LOG_ERROR, "Unknown Ut Video FOURCC provided (%08X)\n",
547  avctx->codec_tag);
548  return AVERROR_INVALIDDATA;
549  }
550 
551  return 0;
552 }
553 
555 {
556  UtvideoContext * const c = avctx->priv_data;
557 
558  if (c->pic.data[0])
559  ff_thread_release_buffer(avctx, &c->pic);
560 
561  av_freep(&c->slice_bits);
562 
563  return 0;
564 }
565 
567  .name = "utvideo",
568  .type = AVMEDIA_TYPE_VIDEO,
569  .id = CODEC_ID_UTVIDEO,
570  .priv_data_size = sizeof(UtvideoContext),
571  .init = decode_init,
572  .close = decode_end,
573  .decode = decode_frame,
574  .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
575  .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
576 };
577