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  bsrc = src + slice_start * stride;
219 
220  // first line - left neighbour prediction
221  bsrc[0] += 0x80;
222  A = bsrc[0];
223  for (i = step; i < width * step; i += step) {
224  bsrc[i] += A;
225  A = bsrc[i];
226  }
227  bsrc += stride;
228  if (slice_height <= 1)
229  continue;
230  // second line - first element has top prediction, the rest uses median
231  C = bsrc[-stride];
232  bsrc[0] += C;
233  A = bsrc[0];
234  for (i = step; i < width * step; i += step) {
235  B = bsrc[i - stride];
236  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
237  C = B;
238  A = bsrc[i];
239  }
240  bsrc += stride;
241  // the rest of lines use continuous median prediction
242  for (j = 2; j < slice_height; j++) {
243  for (i = 0; i < width * step; i += step) {
244  B = bsrc[i - stride];
245  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
246  C = B;
247  A = bsrc[i];
248  }
249  bsrc += stride;
250  }
251  }
252 }
253 
254 /* UtVideo interlaced mode treats every two lines as a single one,
255  * so restoring function should take care of possible padding between
256  * two parts of the same "line".
257  */
258 static void restore_median_il(uint8_t *src, int step, int stride,
259  int width, int height, int slices, int rmode)
260 {
261  int i, j, slice;
262  int A, B, C;
263  uint8_t *bsrc;
264  int slice_start, slice_height;
265  const int cmask = ~(rmode ? 3 : 1);
266  const int stride2 = stride << 1;
267 
268  for (slice = 0; slice < slices; slice++) {
269  slice_start = ((slice * height) / slices) & cmask;
270  slice_height = ((((slice + 1) * height) / slices) & cmask) -
271  slice_start;
272  slice_height >>= 1;
273 
274  bsrc = src + slice_start * stride;
275 
276  // first line - left neighbour prediction
277  bsrc[0] += 0x80;
278  A = bsrc[0];
279  for (i = step; i < width * step; i += step) {
280  bsrc[i] += A;
281  A = bsrc[i];
282  }
283  for (i = 0; i < width * step; i += step) {
284  bsrc[stride + i] += A;
285  A = bsrc[stride + i];
286  }
287  bsrc += stride2;
288  if (slice_height <= 1)
289  continue;
290  // second line - first element has top prediction, the rest uses median
291  C = bsrc[-stride2];
292  bsrc[0] += C;
293  A = bsrc[0];
294  for (i = step; i < width * step; i += step) {
295  B = bsrc[i - stride2];
296  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
297  C = B;
298  A = bsrc[i];
299  }
300  for (i = 0; i < width * step; i += step) {
301  B = bsrc[i - stride];
302  bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
303  C = B;
304  A = bsrc[stride + i];
305  }
306  bsrc += stride2;
307  // the rest of lines use continuous median prediction
308  for (j = 2; j < slice_height; j++) {
309  for (i = 0; i < width * step; i += step) {
310  B = bsrc[i - stride2];
311  bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
312  C = B;
313  A = bsrc[i];
314  }
315  for (i = 0; i < width * step; i += step) {
316  B = bsrc[i - stride];
317  bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
318  C = B;
319  A = bsrc[i + stride];
320  }
321  bsrc += stride2;
322  }
323  }
324 }
325 
326 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
327  AVPacket *avpkt)
328 {
329  const uint8_t *buf = avpkt->data;
330  int buf_size = avpkt->size;
331  UtvideoContext *c = avctx->priv_data;
332  int i, j;
333  const uint8_t *plane_start[5];
334  int plane_size, max_slice_size = 0, slice_start, slice_end, slice_size;
335  int ret;
336  GetByteContext gb;
337  ThreadFrame frame = { .f = data };
338 
339  if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
340  return ret;
341 
342  /* parse plane structure to get frame flags and validate slice offsets */
343  bytestream2_init(&gb, buf, buf_size);
344  for (i = 0; i < c->planes; i++) {
345  plane_start[i] = gb.buffer;
346  if (bytestream2_get_bytes_left(&gb) < 256 + 4 * c->slices) {
347  av_log(avctx, AV_LOG_ERROR, "Insufficient data for a plane\n");
348  return AVERROR_INVALIDDATA;
349  }
350  bytestream2_skipu(&gb, 256);
351  slice_start = 0;
352  slice_end = 0;
353  for (j = 0; j < c->slices; j++) {
354  slice_end = bytestream2_get_le32u(&gb);
355  slice_size = slice_end - slice_start;
356  if (slice_end < 0 || slice_size < 0 ||
357  bytestream2_get_bytes_left(&gb) < slice_end) {
358  av_log(avctx, AV_LOG_ERROR, "Incorrect slice size\n");
359  return AVERROR_INVALIDDATA;
360  }
361  slice_start = slice_end;
362  max_slice_size = FFMAX(max_slice_size, slice_size);
363  }
364  plane_size = slice_end;
365  bytestream2_skipu(&gb, plane_size);
366  }
367  plane_start[c->planes] = gb.buffer;
369  av_log(avctx, AV_LOG_ERROR, "Not enough data for frame information\n");
370  return AVERROR_INVALIDDATA;
371  }
372  c->frame_info = bytestream2_get_le32u(&gb);
373  av_log(avctx, AV_LOG_DEBUG, "frame information flags %"PRIX32"\n",
374  c->frame_info);
375 
376  c->frame_pred = (c->frame_info >> 8) & 3;
377 
378  if (c->frame_pred == PRED_GRADIENT) {
379  avpriv_request_sample(avctx, "Frame with gradient prediction");
380  return AVERROR_PATCHWELCOME;
381  }
382 
384  max_slice_size + FF_INPUT_BUFFER_PADDING_SIZE);
385 
386  if (!c->slice_bits) {
387  av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer\n");
388  return AVERROR(ENOMEM);
389  }
390 
391  switch (c->avctx->pix_fmt) {
392  case AV_PIX_FMT_RGB24:
393  case AV_PIX_FMT_RGBA:
394  for (i = 0; i < c->planes; i++) {
395  ret = decode_plane(c, i, frame.f->data[0] + ff_ut_rgb_order[i],
396  c->planes, frame.f->linesize[0], avctx->width,
397  avctx->height, plane_start[i],
398  c->frame_pred == PRED_LEFT);
399  if (ret)
400  return ret;
401  if (c->frame_pred == PRED_MEDIAN) {
402  if (!c->interlaced) {
403  restore_median(frame.f->data[0] + ff_ut_rgb_order[i],
404  c->planes, frame.f->linesize[0], avctx->width,
405  avctx->height, c->slices, 0);
406  } else {
407  restore_median_il(frame.f->data[0] + ff_ut_rgb_order[i],
408  c->planes, frame.f->linesize[0],
409  avctx->width, avctx->height, c->slices,
410  0);
411  }
412  }
413  }
414  restore_rgb_planes(frame.f->data[0], c->planes, frame.f->linesize[0],
415  avctx->width, avctx->height);
416  break;
417  case AV_PIX_FMT_YUV420P:
418  for (i = 0; i < 3; i++) {
419  ret = decode_plane(c, i, frame.f->data[i], 1, frame.f->linesize[i],
420  avctx->width >> !!i, avctx->height >> !!i,
421  plane_start[i], c->frame_pred == PRED_LEFT);
422  if (ret)
423  return ret;
424  if (c->frame_pred == PRED_MEDIAN) {
425  if (!c->interlaced) {
426  restore_median(frame.f->data[i], 1, frame.f->linesize[i],
427  avctx->width >> !!i, avctx->height >> !!i,
428  c->slices, !i);
429  } else {
430  restore_median_il(frame.f->data[i], 1, frame.f->linesize[i],
431  avctx->width >> !!i,
432  avctx->height >> !!i,
433  c->slices, !i);
434  }
435  }
436  }
437  break;
438  case AV_PIX_FMT_YUV422P:
439  for (i = 0; i < 3; i++) {
440  ret = decode_plane(c, i, frame.f->data[i], 1, frame.f->linesize[i],
441  avctx->width >> !!i, avctx->height,
442  plane_start[i], c->frame_pred == PRED_LEFT);
443  if (ret)
444  return ret;
445  if (c->frame_pred == PRED_MEDIAN) {
446  if (!c->interlaced) {
447  restore_median(frame.f->data[i], 1, frame.f->linesize[i],
448  avctx->width >> !!i, avctx->height,
449  c->slices, 0);
450  } else {
451  restore_median_il(frame.f->data[i], 1, frame.f->linesize[i],
452  avctx->width >> !!i, avctx->height,
453  c->slices, 0);
454  }
455  }
456  }
457  break;
458  }
459 
460  frame.f->key_frame = 1;
461  frame.f->pict_type = AV_PICTURE_TYPE_I;
462  frame.f->interlaced_frame = !!c->interlaced;
463 
464  *got_frame = 1;
465 
466  /* always report that the buffer was completely consumed */
467  return buf_size;
468 }
469 
471 {
472  UtvideoContext * const c = avctx->priv_data;
473 
474  c->avctx = avctx;
475 
476  ff_bswapdsp_init(&c->bdsp);
477 
478  if (avctx->extradata_size < 16) {
479  av_log(avctx, AV_LOG_ERROR,
480  "Insufficient extradata size %d, should be at least 16\n",
481  avctx->extradata_size);
482  return AVERROR_INVALIDDATA;
483  }
484 
485  av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",
486  avctx->extradata[3], avctx->extradata[2],
487  avctx->extradata[1], avctx->extradata[0]);
488  av_log(avctx, AV_LOG_DEBUG, "Original format %"PRIX32"\n",
489  AV_RB32(avctx->extradata + 4));
490  c->frame_info_size = AV_RL32(avctx->extradata + 8);
491  c->flags = AV_RL32(avctx->extradata + 12);
492 
493  if (c->frame_info_size != 4)
494  avpriv_request_sample(avctx, "Frame info not 4 bytes");
495  av_log(avctx, AV_LOG_DEBUG, "Encoding parameters %08"PRIX32"\n", c->flags);
496  c->slices = (c->flags >> 24) + 1;
497  c->compression = c->flags & 1;
498  c->interlaced = c->flags & 0x800;
499 
500  c->slice_bits_size = 0;
501 
502  switch (avctx->codec_tag) {
503  case MKTAG('U', 'L', 'R', 'G'):
504  c->planes = 3;
505  avctx->pix_fmt = AV_PIX_FMT_RGB24;
506  break;
507  case MKTAG('U', 'L', 'R', 'A'):
508  c->planes = 4;
509  avctx->pix_fmt = AV_PIX_FMT_RGBA;
510  break;
511  case MKTAG('U', 'L', 'Y', '0'):
512  c->planes = 3;
513  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
514  avctx->colorspace = AVCOL_SPC_BT470BG;
515  break;
516  case MKTAG('U', 'L', 'Y', '2'):
517  c->planes = 3;
518  avctx->pix_fmt = AV_PIX_FMT_YUV422P;
519  avctx->colorspace = AVCOL_SPC_BT470BG;
520  break;
521  case MKTAG('U', 'L', 'H', '0'):
522  c->planes = 3;
523  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
524  avctx->colorspace = AVCOL_SPC_BT709;
525  break;
526  case MKTAG('U', 'L', 'H', '2'):
527  c->planes = 3;
528  avctx->pix_fmt = AV_PIX_FMT_YUV422P;
529  avctx->colorspace = AVCOL_SPC_BT709;
530  break;
531  default:
532  av_log(avctx, AV_LOG_ERROR, "Unknown Ut Video FOURCC provided (%08X)\n",
533  avctx->codec_tag);
534  return AVERROR_INVALIDDATA;
535  }
536 
537  return 0;
538 }
539 
541 {
542  UtvideoContext * const c = avctx->priv_data;
543 
544  av_freep(&c->slice_bits);
545 
546  return 0;
547 }
548 
550  .name = "utvideo",
551  .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
552  .type = AVMEDIA_TYPE_VIDEO,
553  .id = AV_CODEC_ID_UTVIDEO,
554  .priv_data_size = sizeof(UtvideoContext),
555  .init = decode_init,
556  .close = decode_end,
557  .decode = decode_frame,
558  .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
559 };