FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
lagarith.c
Go to the documentation of this file.
1 /*
2  * Lagarith lossless decoder
3  * Copyright (c) 2009 Nathan Caldwell <saintdev (at) gmail.com>
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  * Lagarith lossless decoder
25  * @author Nathan Caldwell
26  */
27 
28 #include "avcodec.h"
29 #include "get_bits.h"
30 #include "mathops.h"
31 #include "dsputil.h"
32 #include "lagarithrac.h"
33 #include "thread.h"
34 
36  FRAME_RAW = 1, /**< uncompressed */
37  FRAME_U_RGB24 = 2, /**< unaligned RGB24 */
38  FRAME_ARITH_YUY2 = 3, /**< arithmetic coded YUY2 */
39  FRAME_ARITH_RGB24 = 4, /**< arithmetic coded RGB24 */
40  FRAME_SOLID_GRAY = 5, /**< solid grayscale color frame */
41  FRAME_SOLID_COLOR = 6, /**< solid non-grayscale color frame */
42  FRAME_OLD_ARITH_RGB = 7, /**< obsolete arithmetic coded RGB (no longer encoded by upstream since version 1.1.0) */
43  FRAME_ARITH_RGBA = 8, /**< arithmetic coded RGBA */
44  FRAME_SOLID_RGBA = 9, /**< solid RGBA color frame */
45  FRAME_ARITH_YV12 = 10, /**< arithmetic coded YV12 */
46  FRAME_REDUCED_RES = 11, /**< reduced resolution YV12 frame */
47 };
48 
49 typedef struct LagarithContext {
53  int zeros; /**< number of consecutive zero bytes encountered */
54  int zeros_rem; /**< number of zero bytes remaining to output */
58 
59 /**
60  * Compute the 52bit mantissa of 1/(double)denom.
61  * This crazy format uses floats in an entropy coder and we have to match x86
62  * rounding exactly, thus ordinary floats aren't portable enough.
63  * @param denom denominator
64  * @return 52bit mantissa
65  * @see softfloat_mul
66  */
67 static uint64_t softfloat_reciprocal(uint32_t denom)
68 {
69  int shift = av_log2(denom - 1) + 1;
70  uint64_t ret = (1ULL << 52) / denom;
71  uint64_t err = (1ULL << 52) - ret * denom;
72  ret <<= shift;
73  err <<= shift;
74  err += denom / 2;
75  return ret + err / denom;
76 }
77 
78 /**
79  * (uint32_t)(x*f), where f has the given mantissa, and exponent 0
80  * Used in combination with softfloat_reciprocal computes x/(double)denom.
81  * @param x 32bit integer factor
82  * @param mantissa mantissa of f with exponent 0
83  * @return 32bit integer value (x*f)
84  * @see softfloat_reciprocal
85  */
86 static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
87 {
88  uint64_t l = x * (mantissa & 0xffffffff);
89  uint64_t h = x * (mantissa >> 32);
90  h += l >> 32;
91  l &= 0xffffffff;
92  l += 1 << av_log2(h >> 21);
93  h += l >> 32;
94  return h >> 20;
95 }
96 
97 static uint8_t lag_calc_zero_run(int8_t x)
98 {
99  return (x << 1) ^ (x >> 7);
100 }
101 
102 static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
103 {
104  static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
105  int i;
106  int bit = 0;
107  int bits = 0;
108  int prevbit = 0;
109  unsigned val;
110 
111  for (i = 0; i < 7; i++) {
112  if (prevbit && bit)
113  break;
114  prevbit = bit;
115  bit = get_bits1(gb);
116  if (bit && !prevbit)
117  bits += series[i];
118  }
119  bits--;
120  if (bits < 0 || bits > 31) {
121  *value = 0;
122  return -1;
123  } else if (bits == 0) {
124  *value = 0;
125  return 0;
126  }
127 
128  val = get_bits_long(gb, bits);
129  val |= 1 << bits;
130 
131  *value = val - 1;
132 
133  return 0;
134 }
135 
137 {
138  int i, j, scale_factor;
139  unsigned prob, cumulative_target;
140  unsigned cumul_prob = 0;
141  unsigned scaled_cumul_prob = 0;
142 
143  rac->prob[0] = 0;
144  rac->prob[257] = UINT_MAX;
145  /* Read probabilities from bitstream */
146  for (i = 1; i < 257; i++) {
147  if (lag_decode_prob(gb, &rac->prob[i]) < 0) {
148  av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability encountered.\n");
149  return -1;
150  }
151  if ((uint64_t)cumul_prob + rac->prob[i] > UINT_MAX) {
152  av_log(rac->avctx, AV_LOG_ERROR, "Integer overflow encountered in cumulative probability calculation.\n");
153  return -1;
154  }
155  cumul_prob += rac->prob[i];
156  if (!rac->prob[i]) {
157  if (lag_decode_prob(gb, &prob)) {
158  av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
159  return -1;
160  }
161  if (prob > 256 - i)
162  prob = 256 - i;
163  for (j = 0; j < prob; j++)
164  rac->prob[++i] = 0;
165  }
166  }
167 
168  if (!cumul_prob) {
169  av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
170  return -1;
171  }
172 
173  /* Scale probabilities so cumulative probability is an even power of 2. */
174  scale_factor = av_log2(cumul_prob);
175 
176  if (cumul_prob & (cumul_prob - 1)) {
177  uint64_t mul = softfloat_reciprocal(cumul_prob);
178  for (i = 1; i < 257; i++) {
179  rac->prob[i] = softfloat_mul(rac->prob[i], mul);
180  scaled_cumul_prob += rac->prob[i];
181  }
182 
183  scale_factor++;
184  cumulative_target = 1 << scale_factor;
185 
186  if (scaled_cumul_prob > cumulative_target) {
187  av_log(rac->avctx, AV_LOG_ERROR,
188  "Scaled probabilities are larger than target!\n");
189  return -1;
190  }
191 
192  scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
193 
194  for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
195  if (rac->prob[i]) {
196  rac->prob[i]++;
197  scaled_cumul_prob--;
198  }
199  /* Comment from reference source:
200  * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
201  * // since the compression change is negligible and fixing it
202  * // breaks backwards compatibility
203  * b =- (signed int)b;
204  * b &= 0xFF;
205  * } else {
206  * b++;
207  * b &= 0x7f;
208  * }
209  */
210  }
211  }
212 
213  rac->scale = scale_factor;
214 
215  /* Fill probability array with cumulative probability for each symbol. */
216  for (i = 1; i < 257; i++)
217  rac->prob[i] += rac->prob[i - 1];
218 
219  return 0;
220 }
221 
223  uint8_t *diff, int w, int *left,
224  int *left_top)
225 {
226  /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
227  * However the &0xFF on the gradient predictor yealds incorrect output
228  * for lagarith.
229  */
230  int i;
231  uint8_t l, lt;
232 
233  l = *left;
234  lt = *left_top;
235 
236  for (i = 0; i < w; i++) {
237  l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
238  lt = src1[i];
239  dst[i] = l;
240  }
241 
242  *left = l;
243  *left_top = lt;
244 }
245 
246 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
247  int width, int stride, int line)
248 {
249  int L, TL;
250 
251  if (!line) {
252  /* Left prediction only for first line */
253  L = l->dsp.add_hfyu_left_prediction(buf, buf,
254  width, 0);
255  } else {
256  /* Left pixel is actually prev_row[width] */
257  L = buf[width - stride - 1];
258 
259  if (line == 1) {
260  /* Second line, left predict first pixel, the rest of the line is median predicted
261  * NOTE: In the case of RGB this pixel is top predicted */
262  TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
263  } else {
264  /* Top left is 2 rows back, last pixel */
265  TL = buf[width - (2 * stride) - 1];
266  }
267 
268  add_lag_median_prediction(buf, buf - stride, buf,
269  width, &L, &TL);
270  }
271 }
272 
274  int width, int stride, int line,
275  int is_luma)
276 {
277  int L, TL;
278 
279  if (!line) {
280  L= buf[0];
281  if (is_luma)
282  buf[0] = 0;
283  l->dsp.add_hfyu_left_prediction(buf, buf, width, 0);
284  if (is_luma)
285  buf[0] = L;
286  return;
287  }
288  if (line == 1) {
289  const int HEAD = is_luma ? 4 : 2;
290  int i;
291 
292  L = buf[width - stride - 1];
293  TL = buf[HEAD - stride - 1];
294  for (i = 0; i < HEAD; i++) {
295  L += buf[i];
296  buf[i] = L;
297  }
298  for (; i<width; i++) {
299  L = mid_pred(L&0xFF, buf[i-stride], (L + buf[i-stride] - TL)&0xFF) + buf[i];
300  TL = buf[i-stride];
301  buf[i]= L;
302  }
303  } else {
304  TL = buf[width - (2 * stride) - 1];
305  L = buf[width - stride - 1];
306  l->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
307  &L, &TL);
308  }
309 }
310 
312  uint8_t *dst, int width, int stride,
313  int esc_count)
314 {
315  int i = 0;
316  int ret = 0;
317 
318  if (!esc_count)
319  esc_count = -1;
320 
321  /* Output any zeros remaining from the previous run */
322 handle_zeros:
323  if (l->zeros_rem) {
324  int count = FFMIN(l->zeros_rem, width - i);
325  memset(dst + i, 0, count);
326  i += count;
327  l->zeros_rem -= count;
328  }
329 
330  while (i < width) {
331  dst[i] = lag_get_rac(rac);
332  ret++;
333 
334  if (dst[i])
335  l->zeros = 0;
336  else
337  l->zeros++;
338 
339  i++;
340  if (l->zeros == esc_count) {
341  int index = lag_get_rac(rac);
342  ret++;
343 
344  l->zeros = 0;
345 
346  l->zeros_rem = lag_calc_zero_run(index);
347  goto handle_zeros;
348  }
349  }
350  return ret;
351 }
352 
354  const uint8_t *src, const uint8_t *src_end,
355  int width, int esc_count)
356 {
357  int i = 0;
358  int count;
359  uint8_t zero_run = 0;
360  const uint8_t *src_start = src;
361  uint8_t mask1 = -(esc_count < 2);
362  uint8_t mask2 = -(esc_count < 3);
363  uint8_t *end = dst + (width - 2);
364 
365 output_zeros:
366  if (l->zeros_rem) {
367  count = FFMIN(l->zeros_rem, width - i);
368  if (end - dst < count) {
369  av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
370  return AVERROR_INVALIDDATA;
371  }
372 
373  memset(dst, 0, count);
374  l->zeros_rem -= count;
375  dst += count;
376  }
377 
378  while (dst < end) {
379  i = 0;
380  while (!zero_run && dst + i < end) {
381  i++;
382  if (i+2 >= src_end - src)
383  return AVERROR_INVALIDDATA;
384  zero_run =
385  !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
386  }
387  if (zero_run) {
388  zero_run = 0;
389  i += esc_count;
390  memcpy(dst, src, i);
391  dst += i;
392  l->zeros_rem = lag_calc_zero_run(src[i]);
393 
394  src += i + 1;
395  goto output_zeros;
396  } else {
397  memcpy(dst, src, i);
398  src += i;
399  dst += i;
400  }
401  }
402  return src - src_start;
403 }
404 
405 
406 
408  int width, int height, int stride,
409  const uint8_t *src, int src_size)
410 {
411  int i = 0;
412  int read = 0;
413  uint32_t length;
414  uint32_t offset = 1;
415  int esc_count;
416  GetBitContext gb;
417  lag_rac rac;
418  const uint8_t *src_end = src + src_size;
419 
420  rac.avctx = l->avctx;
421  l->zeros = 0;
422 
423  if(src_size < 2)
424  return AVERROR_INVALIDDATA;
425 
426  esc_count = src[0];
427  if (esc_count < 4) {
428  length = width * height;
429  if(src_size < 5)
430  return AVERROR_INVALIDDATA;
431  if (esc_count && AV_RL32(src + 1) < length) {
432  length = AV_RL32(src + 1);
433  offset += 4;
434  }
435 
436  init_get_bits(&gb, src + offset, src_size * 8);
437 
438  if (lag_read_prob_header(&rac, &gb) < 0)
439  return -1;
440 
441  ff_lag_rac_init(&rac, &gb, length - stride);
442 
443  for (i = 0; i < height; i++)
444  read += lag_decode_line(l, &rac, dst + (i * stride), width,
445  stride, esc_count);
446 
447  if (read > length)
449  "Output more bytes than length (%d of %d)\n", read,
450  length);
451  } else if (esc_count < 8) {
452  esc_count -= 4;
453  if (esc_count > 0) {
454  /* Zero run coding only, no range coding. */
455  for (i = 0; i < height; i++) {
456  int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
457  src_end, width, esc_count);
458  if (res < 0)
459  return res;
460  src += res;
461  }
462  } else {
463  if (src_size < width * height)
464  return AVERROR_INVALIDDATA; // buffer not big enough
465  /* Plane is stored uncompressed */
466  for (i = 0; i < height; i++) {
467  memcpy(dst + (i * stride), src, width);
468  src += width;
469  }
470  }
471  } else if (esc_count == 0xff) {
472  /* Plane is a solid run of given value */
473  for (i = 0; i < height; i++)
474  memset(dst + i * stride, src[1], width);
475  /* Do not apply prediction.
476  Note: memset to 0 above, setting first value to src[1]
477  and applying prediction gives the same result. */
478  return 0;
479  } else {
481  "Invalid zero run escape code! (%#x)\n", esc_count);
482  return -1;
483  }
484 
485  if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
486  for (i = 0; i < height; i++) {
487  lag_pred_line(l, dst, width, stride, i);
488  dst += stride;
489  }
490  } else {
491  for (i = 0; i < height; i++) {
492  lag_pred_line_yuy2(l, dst, width, stride, i,
493  width == l->avctx->width);
494  dst += stride;
495  }
496  }
497 
498  return 0;
499 }
500 
501 /**
502  * Decode a frame.
503  * @param avctx codec context
504  * @param data output AVFrame
505  * @param data_size size of output data or 0 if no picture is returned
506  * @param avpkt input packet
507  * @return number of consumed bytes on success or negative if decode fails
508  */
510  void *data, int *got_frame, AVPacket *avpkt)
511 {
512  const uint8_t *buf = avpkt->data;
513  unsigned int buf_size = avpkt->size;
514  LagarithContext *l = avctx->priv_data;
515  AVFrame *const p = &l->picture;
516  uint8_t frametype = 0;
517  uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
518  uint32_t offs[4];
519  uint8_t *srcs[4], *dst;
520  int i, j, planes = 3;
521  int ret;
522 
523  AVFrame *picture = data;
524 
525  if (p->data[0])
526  ff_thread_release_buffer(avctx, p);
527 
528  p->reference = 0;
529  p->key_frame = 1;
530 
531  frametype = buf[0];
532 
533  offset_gu = AV_RL32(buf + 1);
534  offset_bv = AV_RL32(buf + 5);
535 
536  switch (frametype) {
537  case FRAME_SOLID_RGBA:
538  avctx->pix_fmt = AV_PIX_FMT_RGB32;
539  case FRAME_SOLID_GRAY:
540  if (frametype == FRAME_SOLID_GRAY)
541  if (avctx->bits_per_coded_sample == 24) {
542  avctx->pix_fmt = AV_PIX_FMT_RGB24;
543  } else {
544  avctx->pix_fmt = AV_PIX_FMT_0RGB32;
545  planes = 4;
546  }
547 
548  if ((ret = ff_thread_get_buffer(avctx, p)) < 0) {
549  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
550  return ret;
551  }
552 
553  dst = p->data[0];
554  if (frametype == FRAME_SOLID_RGBA) {
555  for (j = 0; j < avctx->height; j++) {
556  for (i = 0; i < avctx->width; i++)
557  AV_WN32(dst + i * 4, offset_gu);
558  dst += p->linesize[0];
559  }
560  } else {
561  for (j = 0; j < avctx->height; j++) {
562  memset(dst, buf[1], avctx->width * planes);
563  dst += p->linesize[0];
564  }
565  }
566  break;
567  case FRAME_ARITH_RGBA:
568  avctx->pix_fmt = AV_PIX_FMT_RGB32;
569  planes = 4;
570  offset_ry += 4;
571  offs[3] = AV_RL32(buf + 9);
572  case FRAME_ARITH_RGB24:
573  case FRAME_U_RGB24:
574  if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
575  avctx->pix_fmt = AV_PIX_FMT_RGB24;
576 
577  if ((ret = ff_thread_get_buffer(avctx, p)) < 0) {
578  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
579  return ret;
580  }
581 
582  offs[0] = offset_bv;
583  offs[1] = offset_gu;
584  offs[2] = offset_ry;
585 
586  if (!l->rgb_planes) {
587  l->rgb_stride = FFALIGN(avctx->width, 16);
588  l->rgb_planes = av_malloc(l->rgb_stride * avctx->height * 4 + 16);
589  if (!l->rgb_planes) {
590  av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
591  return AVERROR(ENOMEM);
592  }
593  }
594  for (i = 0; i < planes; i++)
595  srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
596  for (i = 0; i < planes; i++)
597  if (buf_size <= offs[i]) {
598  av_log(avctx, AV_LOG_ERROR,
599  "Invalid frame offsets\n");
600  return AVERROR_INVALIDDATA;
601  }
602 
603  for (i = 0; i < planes; i++)
604  lag_decode_arith_plane(l, srcs[i],
605  avctx->width, avctx->height,
606  -l->rgb_stride, buf + offs[i],
607  buf_size - offs[i]);
608  dst = p->data[0];
609  for (i = 0; i < planes; i++)
610  srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
611  for (j = 0; j < avctx->height; j++) {
612  for (i = 0; i < avctx->width; i++) {
613  uint8_t r, g, b, a;
614  r = srcs[0][i];
615  g = srcs[1][i];
616  b = srcs[2][i];
617  r += g;
618  b += g;
619  if (frametype == FRAME_ARITH_RGBA) {
620  a = srcs[3][i];
621  AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
622  } else {
623  dst[i * 3 + 0] = r;
624  dst[i * 3 + 1] = g;
625  dst[i * 3 + 2] = b;
626  }
627  }
628  dst += p->linesize[0];
629  for (i = 0; i < planes; i++)
630  srcs[i] += l->rgb_stride;
631  }
632  break;
633  case FRAME_ARITH_YUY2:
634  avctx->pix_fmt = AV_PIX_FMT_YUV422P;
635 
636  if ((ret = ff_thread_get_buffer(avctx, p)) < 0) {
637  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
638  return ret;
639  }
640 
641  if (offset_ry >= buf_size ||
642  offset_gu >= buf_size ||
643  offset_bv >= buf_size) {
644  av_log(avctx, AV_LOG_ERROR,
645  "Invalid frame offsets\n");
646  return AVERROR_INVALIDDATA;
647  }
648 
649  lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
650  p->linesize[0], buf + offset_ry,
651  buf_size - offset_ry);
652  lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
653  avctx->height, p->linesize[1],
654  buf + offset_gu, buf_size - offset_gu);
655  lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
656  avctx->height, p->linesize[2],
657  buf + offset_bv, buf_size - offset_bv);
658  break;
659  case FRAME_ARITH_YV12:
660  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
661 
662  if ((ret = ff_thread_get_buffer(avctx, p)) < 0) {
663  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
664  return ret;
665  }
666  if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
667  return AVERROR_INVALIDDATA;
668  }
669 
670  if (offset_ry >= buf_size ||
671  offset_gu >= buf_size ||
672  offset_bv >= buf_size) {
673  av_log(avctx, AV_LOG_ERROR,
674  "Invalid frame offsets\n");
675  return AVERROR_INVALIDDATA;
676  }
677 
678  lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
679  p->linesize[0], buf + offset_ry,
680  buf_size - offset_ry);
681  lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
682  avctx->height / 2, p->linesize[2],
683  buf + offset_gu, buf_size - offset_gu);
684  lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
685  avctx->height / 2, p->linesize[1],
686  buf + offset_bv, buf_size - offset_bv);
687  break;
688  default:
689  av_log(avctx, AV_LOG_ERROR,
690  "Unsupported Lagarith frame type: %#x\n", frametype);
691  return AVERROR_PATCHWELCOME;
692  }
693 
694  *picture = *p;
695  *got_frame = 1;
696 
697  return buf_size;
698 }
699 
701 {
702  LagarithContext *l = avctx->priv_data;
703  l->avctx = avctx;
704 
705  ff_dsputil_init(&l->dsp, avctx);
706 
707  return 0;
708 }
709 
711 {
712  LagarithContext *l = avctx->priv_data;
713 
714  if (l->picture.data[0])
715  ff_thread_release_buffer(avctx, &l->picture);
716  av_freep(&l->rgb_planes);
717 
718  return 0;
719 }
720 
722  .name = "lagarith",
723  .type = AVMEDIA_TYPE_VIDEO,
724  .id = AV_CODEC_ID_LAGARITH,
725  .priv_data_size = sizeof(LagarithContext),
729  .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
730  .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
731 };