FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
utvideodec.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 FFmpeg.
6  *
7  * FFmpeg 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  * FFmpeg 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 FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * Ut Video decoder
25  */
26 
27 #include <inttypes.h>
28 #include <stdlib.h>
29 
30 #include "libavutil/intreadwrite.h"
31 #include "avcodec.h"
32 #include "bswapdsp.h"
33 #include "bytestream.h"
34 #include "get_bits.h"
35 #include "thread.h"
36 #include "utvideo.h"
37 
38 static int build_huff(const uint8_t *src, VLC *vlc, int *fsym)
39 {
40  int i;
41  HuffEntry he[256];
42  int last;
43  uint32_t codes[256];
44  uint8_t bits[256];
45  uint8_t syms[256];
46  uint32_t code;
47 
48  *fsym = -1;
49  for (i = 0; i < 256; i++) {
50  he[i].sym = i;
51  he[i].len = *src++;
52  }
53  qsort(he, 256, sizeof(*he), ff_ut_huff_cmp_len);
54 
55  if (!he[0].len) {
56  *fsym = he[0].sym;
57  return 0;
58  }
59  if (he[0].len > 32)
60  return -1;
61 
62  last = 255;
63  while (he[last].len == 255 && last)
64  last--;
65 
66  code = 1;
67  for (i = last; i >= 0; i--) {
68  codes[i] = code >> (32 - he[i].len);
69  bits[i] = he[i].len;
70  syms[i] = he[i].sym;
71  code += 0x80000000u >> (he[i].len - 1);
72  }
73 
74  return ff_init_vlc_sparse(vlc, FFMIN(he[last].len, 11), last + 1,
75  bits, sizeof(*bits), sizeof(*bits),
76  codes, sizeof(*codes), sizeof(*codes),
77  syms, sizeof(*syms), sizeof(*syms), 0);
78 }
79 
80 static int decode_plane(UtvideoContext *c, int plane_no,
81  uint8_t *dst, int step, int stride,
82  int width, int height,
83  const uint8_t *src, int use_pred)
84 {
85  int i, j, slice, pix;
86  int sstart, send;
87  VLC vlc;
88  GetBitContext gb;
89  int prev, fsym;
90  const int cmask = ~(!plane_no && c->avctx->pix_fmt == AV_PIX_FMT_YUV420P);
91 
92  if (build_huff(src, &vlc, &fsym)) {
93  av_log(c->avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
94  return AVERROR_INVALIDDATA;
95  }
96  if (fsym >= 0) { // build_huff reported a symbol to fill slices with
97  send = 0;
98  for (slice = 0; slice < c->slices; slice++) {
99  uint8_t *dest;
100 
101  sstart = send;
102  send = (height * (slice + 1) / c->slices) & cmask;
103  dest = dst + sstart * stride;
104 
105  prev = 0x80;
106  for (j = sstart; j < send; j++) {
107  for (i = 0; i < width * step; i += step) {
108  pix = fsym;
109  if (use_pred) {
110  prev += pix;
111  pix = prev;
112  }
113  dest[i] = pix;
114  }
115  dest += stride;
116  }
117  }
118  return 0;
119  }
120 
121  src += 256;
122 
123  send = 0;
124  for (slice = 0; slice < c->slices; slice++) {
125  uint8_t *dest;
126  int slice_data_start, slice_data_end, slice_size;
127 
128  sstart = send;
129  send = (height * (slice + 1) / c->slices) & cmask;
130  dest = dst + sstart * stride;
131 
132  // slice offset and size validation was done earlier
133  slice_data_start = slice ? AV_RL32(src + slice * 4 - 4) : 0;
134  slice_data_end = AV_RL32(src + slice * 4);
135  slice_size = slice_data_end - slice_data_start;
136 
137  if (!slice_size) {
138  av_log(c->avctx, AV_LOG_ERROR, "Plane has more than one symbol "
139  "yet a slice has a length of zero.\n");
140  goto fail;
141  }
142 
143  memcpy(c->slice_bits, src + slice_data_start + c->slices * 4,
144  slice_size);
145  memset(c->slice_bits + slice_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
146  c->bdsp.bswap_buf((uint32_t *) c->slice_bits,
147  (uint32_t *) c->slice_bits,
148  (slice_data_end - slice_data_start + 3) >> 2);
149  init_get_bits(&gb, c->slice_bits, slice_size * 8);
150 
151  prev = 0x80;
152  for (j = sstart; j < send; j++) {
153  for (i = 0; i < width * step; i += step) {
154  if (get_bits_left(&gb) <= 0) {
156  "Slice decoding ran out of bits\n");
157  goto fail;
158  }
159  pix = get_vlc2(&gb, vlc.table, vlc.bits, 3);
160  if (pix < 0) {
161  av_log(c->avctx, AV_LOG_ERROR, "Decoding error\n");
162  goto fail;
163  }
164  if (use_pred) {
165  prev += pix;
166  pix = prev;
167  }
168  dest[i] = pix;
169  }
170  dest += stride;
171  }
172  if (get_bits_left(&gb) > 32)
174  "%d bits left after decoding slice\n", get_bits_left(&gb));
175  }
176 
177  ff_free_vlc(&vlc);
178 
179  return 0;
180 fail:
181  ff_free_vlc(&vlc);
182  return AVERROR_INVALIDDATA;
183 }
184 
185 static void restore_rgb_planes(uint8_t *src, int step, int stride, int width,
186  int height)
187 {
188  int i, j;
189  uint8_t r, g, b;
190 
191  for (j = 0; j < height; j++) {
192  for (i = 0; i < width * step; i += step) {
193  r = src[i];
194  g = src[i + 1];
195  b = src[i + 2];
196  src[i] = r + g - 0x80;
197  src[i + 2] = b + g - 0x80;
198  }
199  src += stride;
200  }
201 }
202 
203 static void restore_median(uint8_t *src, int step, int stride,
204  int width, int height, int slices, int rmode)
205 {
206  int i, j, slice;
207  int A, B, C;
208  uint8_t *bsrc;
209  int slice_start, slice_height;
210  const int cmask = ~rmode;
211 
212  for (slice = 0; slice < slices; slice++) {
213  slice_start = ((slice * height) / slices) & cmask;
214  slice_height = ((((slice + 1) * height) / slices) & cmask) -
215  slice_start;
216 
217  bsrc = src + slice_start * stride;
218 
219  // first line - left neighbour prediction
220  bsrc[0] += 0x80;
221  A = bsrc[0];
222  for (i = step; i < width * step; i += step) {
223  bsrc[i] += A;
224  A = bsrc[i];
225  }
226  bsrc += stride;
227  if (slice_height == 1)
228  continue;
229  // second line - first element has top prediction, the rest uses median
230  C = bsrc[-stride];
231  bsrc[0] += C;
232  A = bsrc[0];
233  for (i = step; i < width * step; i += step) {
234  B = bsrc[i - stride];
235  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
236  C = B;
237  A = bsrc[i];
238  }
239  bsrc += stride;
240  // the rest of lines use continuous median prediction
241  for (j = 2; j < slice_height; j++) {
242  for (i = 0; i < width * step; i += step) {
243  B = bsrc[i - stride];
244  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
245  C = B;
246  A = bsrc[i];
247  }
248  bsrc += stride;
249  }
250  }
251 }
252 
253 /* UtVideo interlaced mode treats every two lines as a single one,
254  * so restoring function should take care of possible padding between
255  * two parts of the same "line".
256  */
257 static void restore_median_il(uint8_t *src, int step, int stride,
258  int width, int height, int slices, int rmode)
259 {
260  int i, j, slice;
261  int A, B, C;
262  uint8_t *bsrc;
263  int slice_start, slice_height;
264  const int cmask = ~(rmode ? 3 : 1);
265  const int stride2 = stride << 1;
266 
267  for (slice = 0; slice < slices; slice++) {
268  slice_start = ((slice * height) / slices) & cmask;
269  slice_height = ((((slice + 1) * height) / slices) & cmask) -
270  slice_start;
271  slice_height >>= 1;
272 
273  bsrc = src + slice_start * stride;
274 
275  // first line - left neighbour prediction
276  bsrc[0] += 0x80;
277  A = bsrc[0];
278  for (i = step; i < width * step; i += step) {
279  bsrc[i] += A;
280  A = bsrc[i];
281  }
282  for (i = 0; i < width * step; i += step) {
283  bsrc[stride + i] += A;
284  A = bsrc[stride + i];
285  }
286  bsrc += stride2;
287  if (slice_height == 1)
288  continue;
289  // second line - first element has top prediction, the rest uses median
290  C = bsrc[-stride2];
291  bsrc[0] += C;
292  A = bsrc[0];
293  for (i = step; i < width * step; i += step) {
294  B = bsrc[i - stride2];
295  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
296  C = B;
297  A = bsrc[i];
298  }
299  for (i = 0; i < width * step; i += step) {
300  B = bsrc[i - stride];
301  bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
302  C = B;
303  A = bsrc[stride + i];
304  }
305  bsrc += stride2;
306  // the rest of lines use continuous median prediction
307  for (j = 2; j < slice_height; j++) {
308  for (i = 0; i < width * step; i += step) {
309  B = bsrc[i - stride2];
310  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
311  C = B;
312  A = bsrc[i];
313  }
314  for (i = 0; i < width * step; i += step) {
315  B = bsrc[i - stride];
316  bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
317  C = B;
318  A = bsrc[i + stride];
319  }
320  bsrc += stride2;
321  }
322  }
323 }
324 
325 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
326  AVPacket *avpkt)
327 {
328  const uint8_t *buf = avpkt->data;
329  int buf_size = avpkt->size;
330  UtvideoContext *c = avctx->priv_data;
331  int i, j;
332  const uint8_t *plane_start[5];
333  int plane_size, max_slice_size = 0, slice_start, slice_end, slice_size;
334  int ret;
335  GetByteContext gb;
336  ThreadFrame frame = { .f = data };
337 
338  if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
339  return ret;
340 
341  /* parse plane structure to get frame flags and validate slice offsets */
342  bytestream2_init(&gb, buf, buf_size);
343  for (i = 0; i < c->planes; i++) {
344  plane_start[i] = gb.buffer;
345  if (bytestream2_get_bytes_left(&gb) < 256 + 4 * c->slices) {
346  av_log(avctx, AV_LOG_ERROR, "Insufficient data for a plane\n");
347  return AVERROR_INVALIDDATA;
348  }
349  bytestream2_skipu(&gb, 256);
350  slice_start = 0;
351  slice_end = 0;
352  for (j = 0; j < c->slices; j++) {
353  slice_end = bytestream2_get_le32u(&gb);
354  slice_size = slice_end - slice_start;
355  if (slice_end < 0 || slice_size < 0 ||
356  bytestream2_get_bytes_left(&gb) < slice_end) {
357  av_log(avctx, AV_LOG_ERROR, "Incorrect slice size\n");
358  return AVERROR_INVALIDDATA;
359  }
360  slice_start = slice_end;
361  max_slice_size = FFMAX(max_slice_size, slice_size);
362  }
363  plane_size = slice_end;
364  bytestream2_skipu(&gb, plane_size);
365  }
366  plane_start[c->planes] = gb.buffer;
368  av_log(avctx, AV_LOG_ERROR, "Not enough data for frame information\n");
369  return AVERROR_INVALIDDATA;
370  }
371  c->frame_info = bytestream2_get_le32u(&gb);
372  av_log(avctx, AV_LOG_DEBUG, "frame information flags %"PRIX32"\n",
373  c->frame_info);
374 
375  c->frame_pred = (c->frame_info >> 8) & 3;
376 
377  if (c->frame_pred == PRED_GRADIENT) {
378  avpriv_request_sample(avctx, "Frame with gradient prediction");
379  return AVERROR_PATCHWELCOME;
380  }
381 
383  max_slice_size + FF_INPUT_BUFFER_PADDING_SIZE);
384 
385  if (!c->slice_bits) {
386  av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer\n");
387  return AVERROR(ENOMEM);
388  }
389 
390  switch (c->avctx->pix_fmt) {
391  case AV_PIX_FMT_RGB24:
392  case AV_PIX_FMT_RGBA:
393  for (i = 0; i < c->planes; i++) {
394  ret = decode_plane(c, i, frame.f->data[0] + ff_ut_rgb_order[i],
395  c->planes, frame.f->linesize[0], avctx->width,
396  avctx->height, plane_start[i],
397  c->frame_pred == PRED_LEFT);
398  if (ret)
399  return ret;
400  if (c->frame_pred == PRED_MEDIAN) {
401  if (!c->interlaced) {
402  restore_median(frame.f->data[0] + ff_ut_rgb_order[i],
403  c->planes, frame.f->linesize[0], avctx->width,
404  avctx->height, c->slices, 0);
405  } else {
406  restore_median_il(frame.f->data[0] + ff_ut_rgb_order[i],
407  c->planes, frame.f->linesize[0],
408  avctx->width, avctx->height, c->slices,
409  0);
410  }
411  }
412  }
413  restore_rgb_planes(frame.f->data[0], c->planes, frame.f->linesize[0],
414  avctx->width, avctx->height);
415  break;
416  case AV_PIX_FMT_YUV420P:
417  for (i = 0; i < 3; i++) {
418  ret = decode_plane(c, i, frame.f->data[i], 1, frame.f->linesize[i],
419  avctx->width >> !!i, avctx->height >> !!i,
420  plane_start[i], c->frame_pred == PRED_LEFT);
421  if (ret)
422  return ret;
423  if (c->frame_pred == PRED_MEDIAN) {
424  if (!c->interlaced) {
425  restore_median(frame.f->data[i], 1, frame.f->linesize[i],
426  avctx->width >> !!i, avctx->height >> !!i,
427  c->slices, !i);
428  } else {
429  restore_median_il(frame.f->data[i], 1, frame.f->linesize[i],
430  avctx->width >> !!i,
431  avctx->height >> !!i,
432  c->slices, !i);
433  }
434  }
435  }
436  break;
437  case AV_PIX_FMT_YUV422P:
438  for (i = 0; i < 3; i++) {
439  ret = decode_plane(c, i, frame.f->data[i], 1, frame.f->linesize[i],
440  avctx->width >> !!i, avctx->height,
441  plane_start[i], c->frame_pred == PRED_LEFT);
442  if (ret)
443  return ret;
444  if (c->frame_pred == PRED_MEDIAN) {
445  if (!c->interlaced) {
446  restore_median(frame.f->data[i], 1, frame.f->linesize[i],
447  avctx->width >> !!i, avctx->height,
448  c->slices, 0);
449  } else {
450  restore_median_il(frame.f->data[i], 1, frame.f->linesize[i],
451  avctx->width >> !!i, avctx->height,
452  c->slices, 0);
453  }
454  }
455  }
456  break;
457  }
458 
459  frame.f->key_frame = 1;
460  frame.f->pict_type = AV_PICTURE_TYPE_I;
461  frame.f->interlaced_frame = !!c->interlaced;
462 
463  *got_frame = 1;
464 
465  /* always report that the buffer was completely consumed */
466  return buf_size;
467 }
468 
470 {
471  UtvideoContext * const c = avctx->priv_data;
472 
473  c->avctx = avctx;
474 
475  ff_bswapdsp_init(&c->bdsp);
476 
477  if (avctx->extradata_size < 16) {
478  av_log(avctx, AV_LOG_ERROR,
479  "Insufficient extradata size %d, should be at least 16\n",
480  avctx->extradata_size);
481  return AVERROR_INVALIDDATA;
482  }
483 
484  av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",
485  avctx->extradata[3], avctx->extradata[2],
486  avctx->extradata[1], avctx->extradata[0]);
487  av_log(avctx, AV_LOG_DEBUG, "Original format %"PRIX32"\n",
488  AV_RB32(avctx->extradata + 4));
489  c->frame_info_size = AV_RL32(avctx->extradata + 8);
490  c->flags = AV_RL32(avctx->extradata + 12);
491 
492  if (c->frame_info_size != 4)
493  avpriv_request_sample(avctx, "Frame info not 4 bytes");
494  av_log(avctx, AV_LOG_DEBUG, "Encoding parameters %08"PRIX32"\n", c->flags);
495  c->slices = (c->flags >> 24) + 1;
496  c->compression = c->flags & 1;
497  c->interlaced = c->flags & 0x800;
498 
499  c->slice_bits_size = 0;
500 
501  switch (avctx->codec_tag) {
502  case MKTAG('U', 'L', 'R', 'G'):
503  c->planes = 3;
504  avctx->pix_fmt = AV_PIX_FMT_RGB24;
505  break;
506  case MKTAG('U', 'L', 'R', 'A'):
507  c->planes = 4;
508  avctx->pix_fmt = AV_PIX_FMT_RGBA;
509  break;
510  case MKTAG('U', 'L', 'Y', '0'):
511  c->planes = 3;
512  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
513  avctx->colorspace = AVCOL_SPC_BT470BG;
514  break;
515  case MKTAG('U', 'L', 'Y', '2'):
516  c->planes = 3;
517  avctx->pix_fmt = AV_PIX_FMT_YUV422P;
518  avctx->colorspace = AVCOL_SPC_BT470BG;
519  break;
520  case MKTAG('U', 'L', 'H', '0'):
521  c->planes = 3;
522  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
523  avctx->colorspace = AVCOL_SPC_BT709;
524  break;
525  case MKTAG('U', 'L', 'H', '2'):
526  c->planes = 3;
527  avctx->pix_fmt = AV_PIX_FMT_YUV422P;
528  avctx->colorspace = AVCOL_SPC_BT709;
529  break;
530  default:
531  av_log(avctx, AV_LOG_ERROR, "Unknown Ut Video FOURCC provided (%08X)\n",
532  avctx->codec_tag);
533  return AVERROR_INVALIDDATA;
534  }
535 
536  return 0;
537 }
538 
540 {
541  UtvideoContext * const c = avctx->priv_data;
542 
543  av_freep(&c->slice_bits);
544 
545  return 0;
546 }
547 
549  .name = "utvideo",
550  .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
551  .type = AVMEDIA_TYPE_VIDEO,
552  .id = AV_CODEC_ID_UTVIDEO,
553  .priv_data_size = sizeof(UtvideoContext),
554  .init = decode_init,
555  .close = decode_end,
556  .decode = decode_frame,
557  .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
558 };