Libav
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 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 
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,
47 };
48 
49 typedef struct LagarithContext {
52  int zeros;
53  int zeros_rem;
58 
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 
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 > 257 - i)
162  prob = 257 - 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 
222 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
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  int i, align_width = (width - 1) & ~31;
253  /* Left prediction only for first line */
254  L = l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1,
255  align_width, buf[0]);
256  for (i = align_width + 1; i < width; i++)
257  buf[i] += buf[i - 1];
258  } else {
259  /* Left pixel is actually prev_row[width] */
260  L = buf[width - stride - 1];
261 
262  if (line == 1) {
263  /* Second line, left predict first pixel, the rest of the line is median predicted
264  * NOTE: In the case of RGB this pixel is top predicted */
265  TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
266  } else {
267  /* Top left is 2 rows back, last pixel */
268  TL = buf[width - (2 * stride) - 1];
269  }
270 
271  add_lag_median_prediction(buf, buf - stride, buf,
272  width, &L, &TL);
273  }
274 }
275 
277  int width, int stride, int line,
278  int is_luma)
279 {
280  int L, TL;
281 
282  if (!line) {
283  int i, align_width;
284  if (is_luma) {
285  buf++;
286  width--;
287  }
288 
289  align_width = (width - 1) & ~31;
290  l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1, align_width, buf[0]);
291 
292  for (i = align_width + 1; i < width; i++)
293  buf[i] += buf[i - 1];
294 
295  return;
296  }
297  if (line == 1) {
298  const int HEAD = is_luma ? 4 : 2;
299  int i;
300 
301  L = buf[width - stride - 1];
302  TL = buf[HEAD - stride - 1];
303  for (i = 0; i < HEAD; i++) {
304  L += buf[i];
305  buf[i] = L;
306  }
307  for (; i < width; i++) {
308  L = mid_pred(L & 0xFF, buf[i - stride], (L + buf[i - stride] - TL) & 0xFF) + buf[i];
309  TL = buf[i - stride];
310  buf[i] = L;
311  }
312  } else {
313  TL = buf[width - (2 * stride) - 1];
314  L = buf[width - stride - 1];
315  l->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
316  &L, &TL);
317  }
318 }
319 
321  uint8_t *dst, int width, int stride,
322  int esc_count)
323 {
324  int i = 0;
325  int ret = 0;
326 
327  if (!esc_count)
328  esc_count = -1;
329 
330  /* Output any zeros remaining from the previous run */
331 handle_zeros:
332  if (l->zeros_rem) {
333  int count = FFMIN(l->zeros_rem, width - i);
334  memset(dst + i, 0, count);
335  i += count;
336  l->zeros_rem -= count;
337  }
338 
339  while (i < width) {
340  dst[i] = lag_get_rac(rac);
341  ret++;
342 
343  if (dst[i])
344  l->zeros = 0;
345  else
346  l->zeros++;
347 
348  i++;
349  if (l->zeros == esc_count) {
350  int index = lag_get_rac(rac);
351  ret++;
352 
353  l->zeros = 0;
354 
355  l->zeros_rem = lag_calc_zero_run(index);
356  goto handle_zeros;
357  }
358  }
359  return ret;
360 }
361 
363  const uint8_t *src, const uint8_t *src_end,
364  int width, int esc_count)
365 {
366  int i = 0;
367  int count;
368  uint8_t zero_run = 0;
369  const uint8_t *src_start = src;
370  uint8_t mask1 = -(esc_count < 2);
371  uint8_t mask2 = -(esc_count < 3);
372  uint8_t *end = dst + (width - 2);
373 
374 output_zeros:
375  if (l->zeros_rem) {
376  count = FFMIN(l->zeros_rem, width - i);
377  if (end - dst < count) {
378  av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
379  return AVERROR_INVALIDDATA;
380  }
381 
382  memset(dst, 0, count);
383  l->zeros_rem -= count;
384  dst += count;
385  }
386 
387  while (dst < end) {
388  i = 0;
389  while (!zero_run && dst + i < end) {
390  i++;
391  if (src + i >= src_end)
392  return AVERROR_INVALIDDATA;
393  zero_run =
394  !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
395  }
396  if (zero_run) {
397  zero_run = 0;
398  i += esc_count;
399  memcpy(dst, src, i);
400  dst += i;
401  l->zeros_rem = lag_calc_zero_run(src[i]);
402 
403  src += i + 1;
404  goto output_zeros;
405  } else {
406  memcpy(dst, src, i);
407  src += i;
408  dst += i;
409  }
410  }
411  return src_start - src;
412 }
413 
414 
415 
417  int width, int height, int stride,
418  const uint8_t *src, int src_size)
419 {
420  int i = 0;
421  int read = 0;
422  uint32_t length;
423  uint32_t offset = 1;
424  int esc_count = src[0];
425  GetBitContext gb;
426  lag_rac rac;
427  const uint8_t *src_end = src + src_size;
428 
429  rac.avctx = l->avctx;
430  l->zeros = 0;
431 
432  if (esc_count < 4) {
433  length = width * height;
434  if (esc_count && AV_RL32(src + 1) < length) {
435  length = AV_RL32(src + 1);
436  offset += 4;
437  }
438 
439  init_get_bits(&gb, src + offset, src_size * 8);
440 
441  if (lag_read_prob_header(&rac, &gb) < 0)
442  return -1;
443 
444  ff_lag_rac_init(&rac, &gb, length - stride);
445 
446  for (i = 0; i < height; i++)
447  read += lag_decode_line(l, &rac, dst + (i * stride), width,
448  stride, esc_count);
449 
450  if (read > length)
452  "Output more bytes than length (%d of %d)\n", read,
453  length);
454  } else if (esc_count < 8) {
455  esc_count -= 4;
456  if (esc_count > 0) {
457  /* Zero run coding only, no range coding. */
458  for (i = 0; i < height; i++) {
459  int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
460  src_end, width, esc_count);
461  if (res < 0)
462  return res;
463  src += res;
464  }
465  } else {
466  if (src_size < width * height)
467  return AVERROR_INVALIDDATA; // buffer not big enough
468  /* Plane is stored uncompressed */
469  for (i = 0; i < height; i++) {
470  memcpy(dst + (i * stride), src, width);
471  src += width;
472  }
473  }
474  } else if (esc_count == 0xff) {
475  /* Plane is a solid run of given value */
476  for (i = 0; i < height; i++)
477  memset(dst + i * stride, src[1], width);
478  /* Do not apply prediction.
479  Note: memset to 0 above, setting first value to src[1]
480  and applying prediction gives the same result. */
481  return 0;
482  } else {
484  "Invalid zero run escape code! (%#x)\n", esc_count);
485  return -1;
486  }
487 
488  if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
489  for (i = 0; i < height; i++) {
490  lag_pred_line(l, dst, width, stride, i);
491  dst += stride;
492  }
493  } else {
494  for (i = 0; i < height; i++) {
495  lag_pred_line_yuy2(l, dst, width, stride, i,
496  width == l->avctx->width);
497  dst += stride;
498  }
499  }
500 
501  return 0;
502 }
503 
513  void *data, int *got_frame, AVPacket *avpkt)
514 {
515  const uint8_t *buf = avpkt->data;
516  int buf_size = avpkt->size;
517  LagarithContext *l = avctx->priv_data;
518  ThreadFrame frame = { .f = data };
519  AVFrame *const p = data;
520  uint8_t frametype = 0;
521  uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
522  uint32_t offs[4];
523  uint8_t *srcs[4], *dst;
524  int i, j, planes = 3;
525 
526  p->key_frame = 1;
527 
528  frametype = buf[0];
529 
530  offset_gu = AV_RL32(buf + 1);
531  offset_bv = AV_RL32(buf + 5);
532 
533  switch (frametype) {
534  case FRAME_SOLID_RGBA:
535  avctx->pix_fmt = AV_PIX_FMT_RGB32;
536 
537  if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
538  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
539  return -1;
540  }
541 
542  dst = p->data[0];
543  for (j = 0; j < avctx->height; j++) {
544  for (i = 0; i < avctx->width; i++)
545  AV_WN32(dst + i * 4, offset_gu);
546  dst += p->linesize[0];
547  }
548  break;
549  case FRAME_ARITH_RGBA:
550  avctx->pix_fmt = AV_PIX_FMT_RGB32;
551  planes = 4;
552  offset_ry += 4;
553  offs[3] = AV_RL32(buf + 9);
554  case FRAME_ARITH_RGB24:
555  case FRAME_U_RGB24:
556  if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
557  avctx->pix_fmt = AV_PIX_FMT_RGB24;
558 
559  if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
560  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
561  return -1;
562  }
563 
564  offs[0] = offset_bv;
565  offs[1] = offset_gu;
566  offs[2] = offset_ry;
567 
568  l->rgb_stride = FFALIGN(avctx->width, 16);
570  l->rgb_stride * avctx->height * planes + 1);
571  if (!l->rgb_planes) {
572  av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
573  return AVERROR(ENOMEM);
574  }
575  for (i = 0; i < planes; i++)
576  srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
577  if (offset_ry >= buf_size ||
578  offset_gu >= buf_size ||
579  offset_bv >= buf_size ||
580  (planes == 4 && offs[3] >= buf_size)) {
581  av_log(avctx, AV_LOG_ERROR,
582  "Invalid frame offsets\n");
583  return AVERROR_INVALIDDATA;
584  }
585  for (i = 0; i < planes; i++)
586  lag_decode_arith_plane(l, srcs[i],
587  avctx->width, avctx->height,
588  -l->rgb_stride, buf + offs[i],
589  buf_size - offs[i]);
590  dst = p->data[0];
591  for (i = 0; i < planes; i++)
592  srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
593  for (j = 0; j < avctx->height; j++) {
594  for (i = 0; i < avctx->width; i++) {
595  uint8_t r, g, b, a;
596  r = srcs[0][i];
597  g = srcs[1][i];
598  b = srcs[2][i];
599  r += g;
600  b += g;
601  if (frametype == FRAME_ARITH_RGBA) {
602  a = srcs[3][i];
603  AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
604  } else {
605  dst[i * 3 + 0] = r;
606  dst[i * 3 + 1] = g;
607  dst[i * 3 + 2] = b;
608  }
609  }
610  dst += p->linesize[0];
611  for (i = 0; i < planes; i++)
612  srcs[i] += l->rgb_stride;
613  }
614  break;
615  case FRAME_ARITH_YUY2:
616  avctx->pix_fmt = AV_PIX_FMT_YUV422P;
617 
618  if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
619  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
620  return -1;
621  }
622 
623  if (offset_ry >= buf_size ||
624  offset_gu >= buf_size ||
625  offset_bv >= buf_size) {
626  av_log(avctx, AV_LOG_ERROR,
627  "Invalid frame offsets\n");
628  return AVERROR_INVALIDDATA;
629  }
630 
631  lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
632  p->linesize[0], buf + offset_ry,
633  buf_size - offset_ry);
634  lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
635  avctx->height, p->linesize[1],
636  buf + offset_gu, buf_size - offset_gu);
637  lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
638  avctx->height, p->linesize[2],
639  buf + offset_bv, buf_size - offset_bv);
640  break;
641  case FRAME_ARITH_YV12:
642  avctx->pix_fmt = AV_PIX_FMT_YUV420P;
643 
644  if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
645  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
646  return -1;
647  }
648 
649  if (offset_ry >= buf_size ||
650  offset_gu >= buf_size ||
651  offset_bv >= buf_size) {
652  av_log(avctx, AV_LOG_ERROR,
653  "Invalid frame offsets\n");
654  return AVERROR_INVALIDDATA;
655  }
656 
657  lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
658  p->linesize[0], buf + offset_ry,
659  buf_size - offset_ry);
660  lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
661  avctx->height / 2, p->linesize[2],
662  buf + offset_gu, buf_size - offset_gu);
663  lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
664  avctx->height / 2, p->linesize[1],
665  buf + offset_bv, buf_size - offset_bv);
666  break;
667  default:
668  av_log(avctx, AV_LOG_ERROR,
669  "Unsupported Lagarith frame type: %#x\n", frametype);
670  return -1;
671  }
672 
673  *got_frame = 1;
674 
675  return buf_size;
676 }
677 
679 {
680  LagarithContext *l = avctx->priv_data;
681  l->avctx = avctx;
682 
683  ff_dsputil_init(&l->dsp, avctx);
684 
685  return 0;
686 }
687 
689 {
690  LagarithContext *l = avctx->priv_data;
691 
692  av_freep(&l->rgb_planes);
693 
694  return 0;
695 }
696 
698  .name = "lagarith",
699  .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
700  .type = AVMEDIA_TYPE_VIDEO,
701  .id = AV_CODEC_ID_LAGARITH,
702  .priv_data_size = sizeof(LagarithContext),
706  .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
707 };