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 
60  last = 255;
61  while (he[last].len == 255 && last)
62  last--;
63 
64  if (he[last].len > 32)
65  return -1;
66 
67  code = 1;
68  for (i = last; i >= 0; i--) {
69  codes[i] = code >> (32 - he[i].len);
70  bits[i] = he[i].len;
71  syms[i] = he[i].sym;
72  code += 0x80000000u >> (he[i].len - 1);
73  }
74 
75  return ff_init_vlc_sparse(vlc, FFMIN(he[last].len, 11), last + 1,
76  bits, sizeof(*bits), sizeof(*bits),
77  codes, sizeof(*codes), sizeof(*codes),
78  syms, sizeof(*syms), sizeof(*syms), 0);
79 }
80 
81 static int decode_plane(UtvideoContext *c, int plane_no,
82  uint8_t *dst, int step, int stride,
83  int width, int height,
84  const uint8_t *src, int use_pred)
85 {
86  int i, j, slice, pix;
87  int sstart, send;
88  VLC vlc;
89  GetBitContext gb;
90  int prev, fsym;
91  const int cmask = ~(!plane_no && c->avctx->pix_fmt == AV_PIX_FMT_YUV420P);
92 
93  if (build_huff(src, &vlc, &fsym)) {
94  av_log(c->avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
95  return AVERROR_INVALIDDATA;
96  }
97  if (fsym >= 0) { // build_huff reported a symbol to fill slices with
98  send = 0;
99  for (slice = 0; slice < c->slices; slice++) {
100  uint8_t *dest;
101 
102  sstart = send;
103  send = (height * (slice + 1) / c->slices) & cmask;
104  dest = dst + sstart * stride;
105 
106  prev = 0x80;
107  for (j = sstart; j < send; j++) {
108  for (i = 0; i < width * step; i += step) {
109  pix = fsym;
110  if (use_pred) {
111  prev += pix;
112  pix = prev;
113  }
114  dest[i] = pix;
115  }
116  dest += stride;
117  }
118  }
119  return 0;
120  }
121 
122  src += 256;
123 
124  send = 0;
125  for (slice = 0; slice < c->slices; slice++) {
126  uint8_t *dest;
127  int slice_data_start, slice_data_end, slice_size;
128 
129  sstart = send;
130  send = (height * (slice + 1) / c->slices) & cmask;
131  dest = dst + sstart * stride;
132 
133  // slice offset and size validation was done earlier
134  slice_data_start = slice ? AV_RL32(src + slice * 4 - 4) : 0;
135  slice_data_end = AV_RL32(src + slice * 4);
136  slice_size = slice_data_end - slice_data_start;
137 
138  if (!slice_size) {
139  av_log(c->avctx, AV_LOG_ERROR, "Plane has more than one symbol "
140  "yet a slice has a length of zero.\n");
141  goto fail;
142  }
143 
144  memcpy(c->slice_bits, src + slice_data_start + c->slices * 4,
145  slice_size);
146  memset(c->slice_bits + slice_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
147  c->bdsp.bswap_buf((uint32_t *) c->slice_bits,
148  (uint32_t *) c->slice_bits,
149  (slice_data_end - slice_data_start + 3) >> 2);
150  init_get_bits(&gb, c->slice_bits, slice_size * 8);
151 
152  prev = 0x80;
153  for (j = sstart; j < send; j++) {
154  for (i = 0; i < width * step; i += step) {
155  if (get_bits_left(&gb) <= 0) {
157  "Slice decoding ran out of bits\n");
158  goto fail;
159  }
160  pix = get_vlc2(&gb, vlc.table, vlc.bits, 3);
161  if (pix < 0) {
162  av_log(c->avctx, AV_LOG_ERROR, "Decoding error\n");
163  goto fail;
164  }
165  if (use_pred) {
166  prev += pix;
167  pix = prev;
168  }
169  dest[i] = pix;
170  }
171  dest += stride;
172  }
173  if (get_bits_left(&gb) > 32)
175  "%d bits left after decoding slice\n", get_bits_left(&gb));
176  }
177 
178  ff_free_vlc(&vlc);
179 
180  return 0;
181 fail:
182  ff_free_vlc(&vlc);
183  return AVERROR_INVALIDDATA;
184 }
185 
186 static void restore_rgb_planes(uint8_t *src, int step, int stride, int width,
187  int height)
188 {
189  int i, j;
190  uint8_t r, g, b;
191 
192  for (j = 0; j < height; j++) {
193  for (i = 0; i < width * step; i += step) {
194  r = src[i];
195  g = src[i + 1];
196  b = src[i + 2];
197  src[i] = r + g - 0x80;
198  src[i + 2] = b + g - 0x80;
199  }
200  src += stride;
201  }
202 }
203 
204 static void restore_median(uint8_t *src, int step, int stride,
205  int width, int height, int slices, int rmode)
206 {
207  int i, j, slice;
208  int A, B, C;
209  uint8_t *bsrc;
210  int slice_start, slice_height;
211  const int cmask = ~rmode;
212 
213  for (slice = 0; slice < slices; slice++) {
214  slice_start = ((slice * height) / slices) & cmask;
215  slice_height = ((((slice + 1) * height) / slices) & cmask) -
216  slice_start;
217 
218  if (!slice_height)
219  continue;
220  bsrc = src + slice_start * stride;
221 
222  // first line - left neighbour prediction
223  bsrc[0] += 0x80;
224  A = bsrc[0];
225  for (i = step; i < width * step; i += step) {
226  bsrc[i] += A;
227  A = bsrc[i];
228  }
229  bsrc += stride;
230  if (slice_height <= 1)
231  continue;
232  // second line - first element has top prediction, the rest uses median
233  C = bsrc[-stride];
234  bsrc[0] += C;
235  A = bsrc[0];
236  for (i = step; i < width * step; i += step) {
237  B = bsrc[i - stride];
238  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
239  C = B;
240  A = bsrc[i];
241  }
242  bsrc += stride;
243  // the rest of lines use continuous median prediction
244  for (j = 2; j < slice_height; j++) {
245  for (i = 0; i < width * step; i += step) {
246  B = bsrc[i - stride];
247  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
248  C = B;
249  A = bsrc[i];
250  }
251  bsrc += stride;
252  }
253  }
254 }
255 
256 /* UtVideo interlaced mode treats every two lines as a single one,
257  * so restoring function should take care of possible padding between
258  * two parts of the same "line".
259  */
260 static void restore_median_il(uint8_t *src, int step, int stride,
261  int width, int height, int slices, int rmode)
262 {
263  int i, j, slice;
264  int A, B, C;
265  uint8_t *bsrc;
266  int slice_start, slice_height;
267  const int cmask = ~(rmode ? 3 : 1);
268  const int stride2 = stride << 1;
269 
270  for (slice = 0; slice < slices; slice++) {
271  slice_start = ((slice * height) / slices) & cmask;
272  slice_height = ((((slice + 1) * height) / slices) & cmask) -
273  slice_start;
274  slice_height >>= 1;
275  if (!slice_height)
276  continue;
277 
278  bsrc = src + slice_start * stride;
279 
280  // first line - left neighbour prediction
281  bsrc[0] += 0x80;
282  A = bsrc[0];
283  for (i = step; i < width * step; i += step) {
284  bsrc[i] += A;
285  A = bsrc[i];
286  }
287  for (i = 0; i < width * step; i += step) {
288  bsrc[stride + i] += A;
289  A = bsrc[stride + i];
290  }
291  bsrc += stride2;
292  if (slice_height <= 1)
293  continue;
294  // second line - first element has top prediction, the rest uses median
295  C = bsrc[-stride2];
296  bsrc[0] += C;
297  A = bsrc[0];
298  for (i = step; i < width * step; i += step) {
299  B = bsrc[i - stride2];
300  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
301  C = B;
302  A = bsrc[i];
303  }
304  for (i = 0; i < width * step; i += step) {
305  B = bsrc[i - stride];
306  bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
307  C = B;
308  A = bsrc[stride + i];
309  }
310  bsrc += stride2;
311  // the rest of lines use continuous median prediction
312  for (j = 2; j < slice_height; j++) {
313  for (i = 0; i < width * step; i += step) {
314  B = bsrc[i - stride2];
315  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
316  C = B;
317  A = bsrc[i];
318  }
319  for (i = 0; i < width * step; i += step) {
320  B = bsrc[i - stride];
321  bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
322  C = B;
323  A = bsrc[i + stride];
324  }
325  bsrc += stride2;
326  }
327  }
328 }
329 
330 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
331  AVPacket *avpkt)
332 {
333  const uint8_t *buf = avpkt->data;
334  int buf_size = avpkt->size;
335  UtvideoContext *c = avctx->priv_data;
336  int i, j;
337  const uint8_t *plane_start[5];
338  int plane_size, max_slice_size = 0, slice_start, slice_end, slice_size;
339  int ret;
340  GetByteContext gb;
341  ThreadFrame frame = { .f = data };
342 
343  if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
344  return ret;
345 
346  /* parse plane structure to get frame flags and validate slice offsets */
347  bytestream2_init(&gb, buf, buf_size);
348  for (i = 0; i < c->planes; i++) {
349  plane_start[i] = gb.buffer;
350  if (bytestream2_get_bytes_left(&gb) < 256 + 4 * c->slices) {
351  av_log(avctx, AV_LOG_ERROR, "Insufficient data for a plane\n");
352  return AVERROR_INVALIDDATA;
353  }
354  bytestream2_skipu(&gb, 256);
355  slice_start = 0;
356  slice_end = 0;
357  for (j = 0; j < c->slices; j++) {
358  slice_end = bytestream2_get_le32u(&gb);
359  slice_size = slice_end - slice_start;
360  if (slice_end < 0 || slice_size < 0 ||
361  bytestream2_get_bytes_left(&gb) < slice_end) {
362  av_log(avctx, AV_LOG_ERROR, "Incorrect slice size\n");
363  return AVERROR_INVALIDDATA;
364  }
365  slice_start = slice_end;
366  max_slice_size = FFMAX(max_slice_size, slice_size);
367  }
368  plane_size = slice_end;
369  bytestream2_skipu(&gb, plane_size);
370  }
371  plane_start[c->planes] = gb.buffer;
373  av_log(avctx, AV_LOG_ERROR, "Not enough data for frame information\n");
374  return AVERROR_INVALIDDATA;
375  }
376  c->frame_info = bytestream2_get_le32u(&gb);
377  av_log(avctx, AV_LOG_DEBUG, "frame information flags %"PRIX32"\n",
378  c->frame_info);
379 
380  c->frame_pred = (c->frame_info >> 8) & 3;
381 
382  if (c->frame_pred == PRED_GRADIENT) {
383  avpriv_request_sample(avctx, "Frame with gradient prediction");
384  return AVERROR_PATCHWELCOME;
385  }
386 
388  max_slice_size + FF_INPUT_BUFFER_PADDING_SIZE);
389 
390  if (!c->slice_bits) {
391  av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer\n");
392  return AVERROR(ENOMEM);
393  }
394 
395  switch (c->avctx->pix_fmt) {
396  case AV_PIX_FMT_RGB24:
397  case AV_PIX_FMT_RGBA:
398  for (i = 0; i < c->planes; i++) {
399  ret = decode_plane(c, i, frame.f->data[0] + ff_ut_rgb_order[i],
400  c->planes, frame.f->linesize[0], avctx->width,
401  avctx->height, plane_start[i],
402  c->frame_pred == PRED_LEFT);
403  if (ret)
404  return ret;
405  if (c->frame_pred == PRED_MEDIAN) {
406  if (!c->interlaced) {
407  restore_median(frame.f->data[0] + ff_ut_rgb_order[i],
408  c->planes, frame.f->linesize[0], avctx->width,
409  avctx->height, c->slices, 0);
410  } else {
411  restore_median_il(frame.f->data[0] + ff_ut_rgb_order[i],
412  c->planes, frame.f->linesize[0],
413  avctx->width, avctx->height, c->slices,
414  0);
415  }
416  }
417  }
418  restore_rgb_planes(frame.f->data[0], c->planes, frame.f->linesize[0],
419  avctx->width, avctx->height);
420  break;
421  case AV_PIX_FMT_YUV420P:
422  for (i = 0; i < 3; i++) {
423  ret = decode_plane(c, i, frame.f->data[i], 1, frame.f->linesize[i],
424  avctx->width >> !!i, avctx->height >> !!i,
425  plane_start[i], c->frame_pred == PRED_LEFT);
426  if (ret)
427  return ret;
428  if (c->frame_pred == PRED_MEDIAN) {
429  if (!c->interlaced) {
430  restore_median(frame.f->data[i], 1, frame.f->linesize[i],
431  avctx->width >> !!i, avctx->height >> !!i,
432  c->slices, !i);
433  } else {
434  restore_median_il(frame.f->data[i], 1, frame.f->linesize[i],
435  avctx->width >> !!i,
436  avctx->height >> !!i,
437  c->slices, !i);
438  }
439  }
440  }
441  break;
442  case AV_PIX_FMT_YUV422P:
443  for (i = 0; i < 3; i++) {
444  ret = decode_plane(c, i, frame.f->data[i], 1, frame.f->linesize[i],
445  avctx->width >> !!i, avctx->height,
446  plane_start[i], c->frame_pred == PRED_LEFT);
447  if (ret)
448  return ret;
449  if (c->frame_pred == PRED_MEDIAN) {
450  if (!c->interlaced) {
451  restore_median(frame.f->data[i], 1, frame.f->linesize[i],
452  avctx->width >> !!i, avctx->height,
453  c->slices, 0);
454  } else {
455  restore_median_il(frame.f->data[i], 1, frame.f->linesize[i],
456  avctx->width >> !!i, avctx->height,
457  c->slices, 0);
458  }
459  }
460  }
461  break;
462  }
463 
464  frame.f->key_frame = 1;
465  frame.f->pict_type = AV_PICTURE_TYPE_I;
466  frame.f->interlaced_frame = !!c->interlaced;
467 
468  *got_frame = 1;
469 
470  /* always report that the buffer was completely consumed */
471  return buf_size;
472 }
473 
475 {
476  UtvideoContext * const c = avctx->priv_data;
477 
478  c->avctx = avctx;
479 
480  ff_bswapdsp_init(&c->bdsp);
481 
482  if (avctx->extradata_size < 16) {
483  av_log(avctx, AV_LOG_ERROR,
484  "Insufficient extradata size %d, should be at least 16\n",
485  avctx->extradata_size);
486  return AVERROR_INVALIDDATA;
487  }
488 
489  av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",
490  avctx->extradata[3], avctx->extradata[2],
491  avctx->extradata[1], avctx->extradata[0]);
492  av_log(avctx, AV_LOG_DEBUG, "Original format %"PRIX32"\n",
493  AV_RB32(avctx->extradata + 4));
494  c->frame_info_size = AV_RL32(avctx->extradata + 8);
495  c->flags = AV_RL32(avctx->extradata + 12);
496 
497  if (c->frame_info_size != 4)
498  avpriv_request_sample(avctx, "Frame info not 4 bytes");
499  av_log(avctx, AV_LOG_DEBUG, "Encoding parameters %08"PRIX32"\n", c->flags);
500  c->slices = (c->flags >> 24) + 1;
501  c->compression = c->flags & 1;
502  c->interlaced = c->flags & 0x800;
503 
504  c->slice_bits_size = 0;
505 
506  switch (avctx->codec_tag) {
507  case MKTAG('U', 'L', 'R', 'G'):
508  c->planes = 3;
509  avctx->pix_fmt = AV_PIX_FMT_RGB24;
510  break;
511  case MKTAG('U', 'L', 'R', 'A'):
512  c->planes = 4;
513  avctx->pix_fmt = AV_PIX_FMT_RGBA;
514  break;
515  case MKTAG('U', 'L', 'Y', '0'):
516  c->planes = 3;
517  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
518  avctx->colorspace = AVCOL_SPC_BT470BG;
519  break;
520  case MKTAG('U', 'L', 'Y', '2'):
521  c->planes = 3;
522  avctx->pix_fmt = AV_PIX_FMT_YUV422P;
523  avctx->colorspace = AVCOL_SPC_BT470BG;
524  break;
525  case MKTAG('U', 'L', 'H', '0'):
526  c->planes = 3;
527  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
528  avctx->colorspace = AVCOL_SPC_BT709;
529  break;
530  case MKTAG('U', 'L', 'H', '2'):
531  c->planes = 3;
532  avctx->pix_fmt = AV_PIX_FMT_YUV422P;
533  avctx->colorspace = AVCOL_SPC_BT709;
534  break;
535  default:
536  av_log(avctx, AV_LOG_ERROR, "Unknown Ut Video FOURCC provided (%08X)\n",
537  avctx->codec_tag);
538  return AVERROR_INVALIDDATA;
539  }
540 
541  return 0;
542 }
543 
545 {
546  UtvideoContext * const c = avctx->priv_data;
547 
548  av_freep(&c->slice_bits);
549 
550  return 0;
551 }
552 
554  .name = "utvideo",
555  .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
556  .type = AVMEDIA_TYPE_VIDEO,
557  .id = AV_CODEC_ID_UTVIDEO,
558  .priv_data_size = sizeof(UtvideoContext),
559  .init = decode_init,
560  .close = decode_end,
561  .decode = decode_frame,
562  .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
563 };