indeo3.c
Go to the documentation of this file.
1 /*
2  * Indeo Video v3 compatible decoder
3  * Copyright (c) 2009 - 2011 Maxim Poliakovski
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 
32 #include "libavutil/imgutils.h"
33 #include "libavutil/intreadwrite.h"
34 #include "avcodec.h"
35 #include "dsputil.h"
36 #include "bytestream.h"
37 #include "get_bits.h"
38 
39 #include "indeo3data.h"
40 
41 /* RLE opcodes. */
42 enum {
43  RLE_ESC_F9 = 249,
44  RLE_ESC_FA = 250,
45  RLE_ESC_FB = 251,
46  RLE_ESC_FC = 252,
47  RLE_ESC_FD = 253,
48  RLE_ESC_FE = 254,
49  RLE_ESC_FF = 255
50 };
51 
52 
53 /* Some constants for parsing frame bitstream flags. */
54 #define BS_8BIT_PEL (1 << 1)
55 #define BS_KEYFRAME (1 << 2)
56 #define BS_MV_Y_HALF (1 << 4)
57 #define BS_MV_X_HALF (1 << 5)
58 #define BS_NONREF (1 << 8)
59 #define BS_BUFFER 9
60 
61 
62 typedef struct Plane {
63  uint8_t *buffers[2];
64  uint8_t *pixels[2];
65  uint32_t width;
66  uint32_t height;
67  uint32_t pitch;
68 } Plane;
69 
70 #define CELL_STACK_MAX 20
71 
72 typedef struct Cell {
73  int16_t xpos;
74  int16_t ypos;
75  int16_t width;
76  int16_t height;
77  uint8_t tree;
78  const int8_t *mv_ptr;
79 } Cell;
80 
81 typedef struct Indeo3DecodeContext {
85 
88  int skip_bits;
89  const uint8_t *next_cell_data;
90  const uint8_t *last_byte;
91  const int8_t *mc_vectors;
92  unsigned num_vectors;
93 
94  int16_t width, height;
95  uint32_t frame_num;
96  uint32_t data_size;
97  uint16_t frame_flags;
98  uint8_t cb_offset;
99  uint8_t buf_sel;
100  const uint8_t *y_data_ptr;
101  const uint8_t *v_data_ptr;
102  const uint8_t *u_data_ptr;
103  int32_t y_data_size;
104  int32_t v_data_size;
105  int32_t u_data_size;
106  const uint8_t *alt_quant;
109 
110 
111 static uint8_t requant_tab[8][128];
112 
113 /*
114  * Build the static requantization table.
115  * This table is used to remap pixel values according to a specific
116  * quant index and thus avoid overflows while adding deltas.
117  */
118 static av_cold void build_requant_tab(void)
119 {
120  static int8_t offsets[8] = { 1, 1, 2, -3, -3, 3, 4, 4 };
121  static int8_t deltas [8] = { 0, 1, 0, 4, 4, 1, 0, 1 };
122 
123  int i, j, step;
124 
125  for (i = 0; i < 8; i++) {
126  step = i + 2;
127  for (j = 0; j < 128; j++)
128  requant_tab[i][j] = (j + offsets[i]) / step * step + deltas[i];
129  }
130 
131  /* some last elements calculated above will have values >= 128 */
132  /* pixel values shall never exceed 127 so set them to non-overflowing values */
133  /* according with the quantization step of the respective section */
134  requant_tab[0][127] = 126;
135  requant_tab[1][119] = 118;
136  requant_tab[1][120] = 118;
137  requant_tab[2][126] = 124;
138  requant_tab[2][127] = 124;
139  requant_tab[6][124] = 120;
140  requant_tab[6][125] = 120;
141  requant_tab[6][126] = 120;
142  requant_tab[6][127] = 120;
143 
144  /* Patch for compatibility with the Intel's binary decoders */
145  requant_tab[1][7] = 10;
146  requant_tab[4][8] = 10;
147 }
148 
149 
151  AVCodecContext *avctx)
152 {
153  int p, luma_width, luma_height, chroma_width, chroma_height;
154  int luma_pitch, chroma_pitch, luma_size, chroma_size;
155 
156  luma_width = ctx->width;
157  luma_height = ctx->height;
158 
159  if (luma_width < 16 || luma_width > 640 ||
160  luma_height < 16 || luma_height > 480 ||
161  luma_width & 3 || luma_height & 3) {
162  av_log(avctx, AV_LOG_ERROR, "Invalid picture dimensions: %d x %d!\n",
163  luma_width, luma_height);
164  return AVERROR_INVALIDDATA;
165  }
166 
167  chroma_width = FFALIGN(luma_width >> 2, 4);
168  chroma_height = FFALIGN(luma_height >> 2, 4);
169 
170  luma_pitch = FFALIGN(luma_width, 16);
171  chroma_pitch = FFALIGN(chroma_width, 16);
172 
173  /* Calculate size of the luminance plane. */
174  /* Add one line more for INTRA prediction. */
175  luma_size = luma_pitch * (luma_height + 1);
176 
177  /* Calculate size of a chrominance planes. */
178  /* Add one line more for INTRA prediction. */
179  chroma_size = chroma_pitch * (chroma_height + 1);
180 
181  /* allocate frame buffers */
182  for (p = 0; p < 3; p++) {
183  ctx->planes[p].pitch = !p ? luma_pitch : chroma_pitch;
184  ctx->planes[p].width = !p ? luma_width : chroma_width;
185  ctx->planes[p].height = !p ? luma_height : chroma_height;
186 
187  ctx->planes[p].buffers[0] = av_malloc(!p ? luma_size : chroma_size);
188  ctx->planes[p].buffers[1] = av_malloc(!p ? luma_size : chroma_size);
189 
190  /* fill the INTRA prediction lines with the middle pixel value = 64 */
191  memset(ctx->planes[p].buffers[0], 0x40, ctx->planes[p].pitch);
192  memset(ctx->planes[p].buffers[1], 0x40, ctx->planes[p].pitch);
193 
194  /* set buffer pointers = buf_ptr + pitch and thus skip the INTRA prediction line */
195  ctx->planes[p].pixels[0] = ctx->planes[p].buffers[0] + ctx->planes[p].pitch;
196  ctx->planes[p].pixels[1] = ctx->planes[p].buffers[1] + ctx->planes[p].pitch;
197  memset(ctx->planes[p].pixels[0], 0, ctx->planes[p].pitch * ctx->planes[p].height);
198  memset(ctx->planes[p].pixels[1], 0, ctx->planes[p].pitch * ctx->planes[p].height);
199  }
200 
201  return 0;
202 }
203 
204 
206 {
207  int p;
208 
209  for (p = 0; p < 3; p++) {
210  av_freep(&ctx->planes[p].buffers[0]);
211  av_freep(&ctx->planes[p].buffers[1]);
212  ctx->planes[p].pixels[0] = ctx->planes[p].pixels[1] = 0;
213  }
214 }
215 
216 
225 static void copy_cell(Indeo3DecodeContext *ctx, Plane *plane, Cell *cell)
226 {
227  int h, w, mv_x, mv_y, offset, offset_dst;
228  uint8_t *src, *dst;
229 
230  /* setup output and reference pointers */
231  offset_dst = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
232  dst = plane->pixels[ctx->buf_sel] + offset_dst;
233  mv_y = cell->mv_ptr[0];
234  mv_x = cell->mv_ptr[1];
235  offset = offset_dst + mv_y * plane->pitch + mv_x;
236  src = plane->pixels[ctx->buf_sel ^ 1] + offset;
237 
238  h = cell->height << 2;
239 
240  for (w = cell->width; w > 0;) {
241  /* copy using 16xH blocks */
242  if (!((cell->xpos << 2) & 15) && w >= 4) {
243  for (; w >= 4; src += 16, dst += 16, w -= 4)
244  ctx->dsp.put_no_rnd_pixels_tab[0][0](dst, src, plane->pitch, h);
245  }
246 
247  /* copy using 8xH blocks */
248  if (!((cell->xpos << 2) & 7) && w >= 2) {
249  ctx->dsp.put_no_rnd_pixels_tab[1][0](dst, src, plane->pitch, h);
250  w -= 2;
251  src += 8;
252  dst += 8;
253  }
254 
255  if (w >= 1) {
256  copy_block4(dst, src, plane->pitch, plane->pitch, h);
257  w--;
258  src += 4;
259  dst += 4;
260  }
261  }
262 }
263 
264 
265 /* Average 4/8 pixels at once without rounding using SWAR */
266 #define AVG_32(dst, src, ref) \
267  AV_WN32A(dst, ((AV_RN32A(src) + AV_RN32A(ref)) >> 1) & 0x7F7F7F7FUL)
268 
269 #define AVG_64(dst, src, ref) \
270  AV_WN64A(dst, ((AV_RN64A(src) + AV_RN64A(ref)) >> 1) & 0x7F7F7F7F7F7F7F7FULL)
271 
272 
273 /*
274  * Replicate each even pixel as follows:
275  * ABCDEFGH -> AACCEEGG
276  */
277 static inline uint64_t replicate64(uint64_t a) {
278 #if HAVE_BIGENDIAN
279  a &= 0xFF00FF00FF00FF00ULL;
280  a |= a >> 8;
281 #else
282  a &= 0x00FF00FF00FF00FFULL;
283  a |= a << 8;
284 #endif
285  return a;
286 }
287 
288 static inline uint32_t replicate32(uint32_t a) {
289 #if HAVE_BIGENDIAN
290  a &= 0xFF00FF00UL;
291  a |= a >> 8;
292 #else
293  a &= 0x00FF00FFUL;
294  a |= a << 8;
295 #endif
296  return a;
297 }
298 
299 
300 /* Fill n lines with 64bit pixel value pix */
301 static inline void fill_64(uint8_t *dst, const uint64_t pix, int32_t n,
302  int32_t row_offset)
303 {
304  for (; n > 0; dst += row_offset, n--)
305  AV_WN64A(dst, pix);
306 }
307 
308 
309 /* Error codes for cell decoding. */
310 enum {
317 };
318 
319 
320 #define BUFFER_PRECHECK \
321 if (*data_ptr >= last_ptr) \
322  return IV3_OUT_OF_DATA; \
323 
324 #define RLE_BLOCK_COPY \
325  if (cell->mv_ptr || !skip_flag) \
326  copy_block4(dst, ref, row_offset, row_offset, 4 << v_zoom)
327 
328 #define RLE_BLOCK_COPY_8 \
329  pix64 = AV_RN64A(ref);\
330  if (is_first_row) {/* special prediction case: top line of a cell */\
331  pix64 = replicate64(pix64);\
332  fill_64(dst + row_offset, pix64, 7, row_offset);\
333  AVG_64(dst, ref, dst + row_offset);\
334  } else \
335  fill_64(dst, pix64, 8, row_offset)
336 
337 #define RLE_LINES_COPY \
338  copy_block4(dst, ref, row_offset, row_offset, num_lines << v_zoom)
339 
340 #define RLE_LINES_COPY_M10 \
341  pix64 = AV_RN64A(ref);\
342  if (is_top_of_cell) {\
343  pix64 = replicate64(pix64);\
344  fill_64(dst + row_offset, pix64, (num_lines << 1) - 1, row_offset);\
345  AVG_64(dst, ref, dst + row_offset);\
346  } else \
347  fill_64(dst, pix64, num_lines << 1, row_offset)
348 
349 #define APPLY_DELTA_4 \
350  AV_WN16A(dst + line_offset ,\
351  (AV_RN16A(ref ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
352  AV_WN16A(dst + line_offset + 2,\
353  (AV_RN16A(ref + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
354  if (mode >= 3) {\
355  if (is_top_of_cell && !cell->ypos) {\
356  AV_COPY32(dst, dst + row_offset);\
357  } else {\
358  AVG_32(dst, ref, dst + row_offset);\
359  }\
360  }
361 
362 #define APPLY_DELTA_8 \
363  /* apply two 32-bit VQ deltas to next even line */\
364  if (is_top_of_cell) { \
365  AV_WN32A(dst + row_offset , \
366  (replicate32(AV_RN32A(ref )) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
367  AV_WN32A(dst + row_offset + 4, \
368  (replicate32(AV_RN32A(ref + 4)) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
369  } else { \
370  AV_WN32A(dst + row_offset , \
371  (AV_RN32A(ref ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
372  AV_WN32A(dst + row_offset + 4, \
373  (AV_RN32A(ref + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
374  } \
375  /* odd lines are not coded but rather interpolated/replicated */\
376  /* first line of the cell on the top of image? - replicate */\
377  /* otherwise - interpolate */\
378  if (is_top_of_cell && !cell->ypos) {\
379  AV_COPY64(dst, dst + row_offset);\
380  } else \
381  AVG_64(dst, ref, dst + row_offset);
382 
383 
384 #define APPLY_DELTA_1011_INTER \
385  if (mode == 10) { \
386  AV_WN32A(dst , \
387  (AV_RN32A(dst ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
388  AV_WN32A(dst + 4 , \
389  (AV_RN32A(dst + 4 ) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
390  AV_WN32A(dst + row_offset , \
391  (AV_RN32A(dst + row_offset ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
392  AV_WN32A(dst + row_offset + 4, \
393  (AV_RN32A(dst + row_offset + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
394  } else { \
395  AV_WN16A(dst , \
396  (AV_RN16A(dst ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
397  AV_WN16A(dst + 2 , \
398  (AV_RN16A(dst + 2 ) + delta_tab->deltas[dyad2]) & 0x7F7F);\
399  AV_WN16A(dst + row_offset , \
400  (AV_RN16A(dst + row_offset ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
401  AV_WN16A(dst + row_offset + 2, \
402  (AV_RN16A(dst + row_offset + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
403  }
404 
405 
406 static int decode_cell_data(Cell *cell, uint8_t *block, uint8_t *ref_block,
407  int pitch, int h_zoom, int v_zoom, int mode,
408  const vqEntry *delta[2], int swap_quads[2],
409  const uint8_t **data_ptr, const uint8_t *last_ptr)
410 {
411  int x, y, line, num_lines;
412  int rle_blocks = 0;
413  uint8_t code, *dst, *ref;
414  const vqEntry *delta_tab;
415  unsigned int dyad1, dyad2;
416  uint64_t pix64;
417  int skip_flag = 0, is_top_of_cell, is_first_row = 1;
418  int row_offset, blk_row_offset, line_offset;
419 
420  row_offset = pitch;
421  blk_row_offset = (row_offset << (2 + v_zoom)) - (cell->width << 2);
422  line_offset = v_zoom ? row_offset : 0;
423 
424  if (cell->height & v_zoom || cell->width & h_zoom)
425  return IV3_BAD_DATA;
426 
427  for (y = 0; y < cell->height; is_first_row = 0, y += 1 + v_zoom) {
428  for (x = 0; x < cell->width; x += 1 + h_zoom) {
429  ref = ref_block;
430  dst = block;
431 
432  if (rle_blocks > 0) {
433  if (mode <= 4) {
435  } else if (mode == 10 && !cell->mv_ptr) {
437  }
438  rle_blocks--;
439  } else {
440  for (line = 0; line < 4;) {
441  num_lines = 1;
442  is_top_of_cell = is_first_row && !line;
443 
444  /* select primary VQ table for odd, secondary for even lines */
445  if (mode <= 4)
446  delta_tab = delta[line & 1];
447  else
448  delta_tab = delta[1];
450  code = bytestream_get_byte(data_ptr);
451  if (code < 248) {
452  if (code < delta_tab->num_dyads) {
454  dyad1 = bytestream_get_byte(data_ptr);
455  dyad2 = code;
456  if (dyad1 >= delta_tab->num_dyads || dyad1 >= 248)
457  return IV3_BAD_DATA;
458  } else {
459  /* process QUADS */
460  code -= delta_tab->num_dyads;
461  dyad1 = code / delta_tab->quad_exp;
462  dyad2 = code % delta_tab->quad_exp;
463  if (swap_quads[line & 1])
464  FFSWAP(unsigned int, dyad1, dyad2);
465  }
466  if (mode <= 4) {
468  } else if (mode == 10 && !cell->mv_ptr) {
470  } else {
472  }
473  } else {
474  /* process RLE codes */
475  switch (code) {
476  case RLE_ESC_FC:
477  skip_flag = 0;
478  rle_blocks = 1;
479  code = 253;
480  /* FALLTHROUGH */
481  case RLE_ESC_FF:
482  case RLE_ESC_FE:
483  case RLE_ESC_FD:
484  num_lines = 257 - code - line;
485  if (num_lines <= 0)
486  return IV3_BAD_RLE;
487  if (mode <= 4) {
489  } else if (mode == 10 && !cell->mv_ptr) {
491  }
492  break;
493  case RLE_ESC_FB:
495  code = bytestream_get_byte(data_ptr);
496  rle_blocks = (code & 0x1F) - 1; /* set block counter */
497  if (code >= 64 || rle_blocks < 0)
498  return IV3_BAD_COUNTER;
499  skip_flag = code & 0x20;
500  num_lines = 4 - line; /* enforce next block processing */
501  if (mode >= 10 || (cell->mv_ptr || !skip_flag)) {
502  if (mode <= 4) {
504  } else if (mode == 10 && !cell->mv_ptr) {
506  }
507  }
508  break;
509  case RLE_ESC_F9:
510  skip_flag = 1;
511  rle_blocks = 1;
512  /* FALLTHROUGH */
513  case RLE_ESC_FA:
514  if (line)
515  return IV3_BAD_RLE;
516  num_lines = 4; /* enforce next block processing */
517  if (cell->mv_ptr) {
518  if (mode <= 4) {
520  } else if (mode == 10 && !cell->mv_ptr) {
522  }
523  }
524  break;
525  default:
526  return IV3_UNSUPPORTED;
527  }
528  }
529 
530  line += num_lines;
531  ref += row_offset * (num_lines << v_zoom);
532  dst += row_offset * (num_lines << v_zoom);
533  }
534  }
535 
536  /* move to next horizontal block */
537  block += 4 << h_zoom;
538  ref_block += 4 << h_zoom;
539  }
540 
541  /* move to next line of blocks */
542  ref_block += blk_row_offset;
543  block += blk_row_offset;
544  }
545  return IV3_NOERR;
546 }
547 
548 
563  Plane *plane, Cell *cell, const uint8_t *data_ptr,
564  const uint8_t *last_ptr)
565 {
566  int x, mv_x, mv_y, mode, vq_index, prim_indx, second_indx;
567  int zoom_fac;
568  int offset, error = 0, swap_quads[2];
569  uint8_t code, *block, *ref_block = 0;
570  const vqEntry *delta[2];
571  const uint8_t *data_start = data_ptr;
572 
573  /* get coding mode and VQ table index from the VQ descriptor byte */
574  code = *data_ptr++;
575  mode = code >> 4;
576  vq_index = code & 0xF;
577 
578  /* setup output and reference pointers */
579  offset = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
580  block = plane->pixels[ctx->buf_sel] + offset;
581  if (!cell->mv_ptr) {
582  /* use previous line as reference for INTRA cells */
583  ref_block = block - plane->pitch;
584  } else if (mode >= 10) {
585  /* for mode 10 and 11 INTER first copy the predicted cell into the current one */
586  /* so we don't need to do data copying for each RLE code later */
587  copy_cell(ctx, plane, cell);
588  } else {
589  /* set the pointer to the reference pixels for modes 0-4 INTER */
590  mv_y = cell->mv_ptr[0];
591  mv_x = cell->mv_ptr[1];
592  offset += mv_y * plane->pitch + mv_x;
593  ref_block = plane->pixels[ctx->buf_sel ^ 1] + offset;
594  }
595 
596  /* select VQ tables as follows: */
597  /* modes 0 and 3 use only the primary table for all lines in a block */
598  /* while modes 1 and 4 switch between primary and secondary tables on alternate lines */
599  if (mode == 1 || mode == 4) {
600  code = ctx->alt_quant[vq_index];
601  prim_indx = (code >> 4) + ctx->cb_offset;
602  second_indx = (code & 0xF) + ctx->cb_offset;
603  } else {
604  vq_index += ctx->cb_offset;
605  prim_indx = second_indx = vq_index;
606  }
607 
608  if (prim_indx >= 24 || second_indx >= 24) {
609  av_log(avctx, AV_LOG_ERROR, "Invalid VQ table indexes! Primary: %d, secondary: %d!\n",
610  prim_indx, second_indx);
611  return AVERROR_INVALIDDATA;
612  }
613 
614  delta[0] = &vq_tab[second_indx];
615  delta[1] = &vq_tab[prim_indx];
616  swap_quads[0] = second_indx >= 16;
617  swap_quads[1] = prim_indx >= 16;
618 
619  /* requantize the prediction if VQ index of this cell differs from VQ index */
620  /* of the predicted cell in order to avoid overflows. */
621  if (vq_index >= 8 && ref_block) {
622  for (x = 0; x < cell->width << 2; x++)
623  ref_block[x] = requant_tab[vq_index & 7][ref_block[x]];
624  }
625 
626  error = IV3_NOERR;
627 
628  switch (mode) {
629  case 0: /*------------------ MODES 0 & 1 (4x4 block processing) --------------------*/
630  case 1:
631  case 3: /*------------------ MODES 3 & 4 (4x8 block processing) --------------------*/
632  case 4:
633  if (mode >= 3 && cell->mv_ptr) {
634  av_log(avctx, AV_LOG_ERROR, "Attempt to apply Mode 3/4 to an INTER cell!\n");
635  return AVERROR_INVALIDDATA;
636  }
637 
638  zoom_fac = mode >= 3;
639  error = decode_cell_data(cell, block, ref_block, plane->pitch, 0, zoom_fac,
640  mode, delta, swap_quads, &data_ptr, last_ptr);
641  break;
642  case 10: /*-------------------- MODE 10 (8x8 block processing) ---------------------*/
643  case 11: /*----------------- MODE 11 (4x8 INTER block processing) ------------------*/
644  if (mode == 10 && !cell->mv_ptr) { /* MODE 10 INTRA processing */
645  error = decode_cell_data(cell, block, ref_block, plane->pitch, 1, 1,
646  mode, delta, swap_quads, &data_ptr, last_ptr);
647  } else { /* mode 10 and 11 INTER processing */
648  if (mode == 11 && !cell->mv_ptr) {
649  av_log(avctx, AV_LOG_ERROR, "Attempt to use Mode 11 for an INTRA cell!\n");
650  return AVERROR_INVALIDDATA;
651  }
652 
653  zoom_fac = mode == 10;
654  error = decode_cell_data(cell, block, ref_block, plane->pitch,
655  zoom_fac, 1, mode, delta, swap_quads,
656  &data_ptr, last_ptr);
657  }
658  break;
659  default:
660  av_log(avctx, AV_LOG_ERROR, "Unsupported coding mode: %d\n", mode);
661  return AVERROR_INVALIDDATA;
662  }//switch mode
663 
664  switch (error) {
665  case IV3_BAD_RLE:
666  av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE code %X is not allowed at the current line\n",
667  mode, data_ptr[-1]);
668  return AVERROR_INVALIDDATA;
669  case IV3_BAD_DATA:
670  av_log(avctx, AV_LOG_ERROR, "Mode %d: invalid VQ data\n", mode);
671  return AVERROR_INVALIDDATA;
672  case IV3_BAD_COUNTER:
673  av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE-FB invalid counter: %d\n", mode, code);
674  return AVERROR_INVALIDDATA;
675  case IV3_UNSUPPORTED:
676  av_log(avctx, AV_LOG_ERROR, "Mode %d: unsupported RLE code: %X\n", mode, data_ptr[-1]);
677  return AVERROR_INVALIDDATA;
678  case IV3_OUT_OF_DATA:
679  av_log(avctx, AV_LOG_ERROR, "Mode %d: attempt to read past end of buffer\n", mode);
680  return AVERROR_INVALIDDATA;
681  }
682 
683  return data_ptr - data_start; /* report number of bytes consumed from the input buffer */
684 }
685 
686 
687 /* Binary tree codes. */
688 enum {
689  H_SPLIT = 0,
690  V_SPLIT = 1,
693 };
694 
695 
696 #define SPLIT_CELL(size, new_size) (new_size) = ((size) > 2) ? ((((size) + 2) >> 2) << 1) : 1
697 
698 #define UPDATE_BITPOS(n) \
699  ctx->skip_bits += (n); \
700  ctx->need_resync = 1
701 
702 #define RESYNC_BITSTREAM \
703  if (ctx->need_resync && !(get_bits_count(&ctx->gb) & 7)) { \
704  skip_bits_long(&ctx->gb, ctx->skip_bits); \
705  ctx->skip_bits = 0; \
706  ctx->need_resync = 0; \
707  }
708 
709 #define CHECK_CELL \
710  if (curr_cell.xpos + curr_cell.width > (plane->width >> 2) || \
711  curr_cell.ypos + curr_cell.height > (plane->height >> 2)) { \
712  av_log(avctx, AV_LOG_ERROR, "Invalid cell: x=%d, y=%d, w=%d, h=%d\n", \
713  curr_cell.xpos, curr_cell.ypos, curr_cell.width, curr_cell.height); \
714  return AVERROR_INVALIDDATA; \
715  }
716 
717 
719  Plane *plane, int code, Cell *ref_cell,
720  const int depth, const int strip_width)
721 {
722  Cell curr_cell;
723  int bytes_used;
724 
725  if (depth <= 0) {
726  av_log(avctx, AV_LOG_ERROR, "Stack overflow (corrupted binary tree)!\n");
727  return AVERROR_INVALIDDATA; // unwind recursion
728  }
729 
730  curr_cell = *ref_cell; // clone parent cell
731  if (code == H_SPLIT) {
732  SPLIT_CELL(ref_cell->height, curr_cell.height);
733  ref_cell->ypos += curr_cell.height;
734  ref_cell->height -= curr_cell.height;
735  if (ref_cell->height <= 0 || curr_cell.height <= 0)
736  return AVERROR_INVALIDDATA;
737  } else if (code == V_SPLIT) {
738  if (curr_cell.width > strip_width) {
739  /* split strip */
740  curr_cell.width = (curr_cell.width <= (strip_width << 1) ? 1 : 2) * strip_width;
741  } else
742  SPLIT_CELL(ref_cell->width, curr_cell.width);
743  ref_cell->xpos += curr_cell.width;
744  ref_cell->width -= curr_cell.width;
745  if (ref_cell->width <= 0 || curr_cell.width <= 0)
746  return AVERROR_INVALIDDATA;
747  }
748 
749  while (1) { /* loop until return */
751  switch (code = get_bits(&ctx->gb, 2)) {
752  case H_SPLIT:
753  case V_SPLIT:
754  if (parse_bintree(ctx, avctx, plane, code, &curr_cell, depth - 1, strip_width))
755  return AVERROR_INVALIDDATA;
756  break;
757  case INTRA_NULL:
758  if (!curr_cell.tree) { /* MC tree INTRA code */
759  curr_cell.mv_ptr = 0; /* mark the current strip as INTRA */
760  curr_cell.tree = 1; /* enter the VQ tree */
761  } else { /* VQ tree NULL code */
763  code = get_bits(&ctx->gb, 2);
764  if (code >= 2) {
765  av_log(avctx, AV_LOG_ERROR, "Invalid VQ_NULL code: %d\n", code);
766  return AVERROR_INVALIDDATA;
767  }
768  if (code == 1)
769  av_log(avctx, AV_LOG_ERROR, "SkipCell procedure not implemented yet!\n");
770 
771  CHECK_CELL
772  if (!curr_cell.mv_ptr)
773  return AVERROR_INVALIDDATA;
774  copy_cell(ctx, plane, &curr_cell);
775  return 0;
776  }
777  break;
778  case INTER_DATA:
779  if (!curr_cell.tree) { /* MC tree INTER code */
780  unsigned mv_idx;
781  /* get motion vector index and setup the pointer to the mv set */
782  if (!ctx->need_resync)
783  ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
784  mv_idx = *(ctx->next_cell_data++) << 1;
785  if (mv_idx >= ctx->num_vectors) {
786  av_log(avctx, AV_LOG_ERROR, "motion vector index out of range\n");
787  return AVERROR_INVALIDDATA;
788  }
789  curr_cell.mv_ptr = &ctx->mc_vectors[mv_idx];
790  curr_cell.tree = 1; /* enter the VQ tree */
791  UPDATE_BITPOS(8);
792  } else { /* VQ tree DATA code */
793  if (!ctx->need_resync)
794  ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
795 
796  CHECK_CELL
797  bytes_used = decode_cell(ctx, avctx, plane, &curr_cell,
798  ctx->next_cell_data, ctx->last_byte);
799  if (bytes_used < 0)
800  return AVERROR_INVALIDDATA;
801 
802  UPDATE_BITPOS(bytes_used << 3);
803  ctx->next_cell_data += bytes_used;
804  return 0;
805  }
806  break;
807  }
808  }//while
809 
810  return 0;
811 }
812 
813 
815  Plane *plane, const uint8_t *data, int32_t data_size,
816  int32_t strip_width)
817 {
818  Cell curr_cell;
819  unsigned num_vectors;
820 
821  /* each plane data starts with mc_vector_count field, */
822  /* an optional array of motion vectors followed by the vq data */
823  num_vectors = bytestream_get_le32(&data);
824  if (num_vectors > 256) {
825  av_log(ctx->avctx, AV_LOG_ERROR,
826  "Read invalid number of motion vectors %d\n", num_vectors);
827  return AVERROR_INVALIDDATA;
828  }
829  if (num_vectors * 2 >= data_size)
830  return AVERROR_INVALIDDATA;
831 
832  ctx->num_vectors = num_vectors;
833  ctx->mc_vectors = num_vectors ? data : 0;
834 
835  /* init the bitreader */
836  init_get_bits(&ctx->gb, &data[num_vectors * 2], (data_size - num_vectors * 2) << 3);
837  ctx->skip_bits = 0;
838  ctx->need_resync = 0;
839 
840  ctx->last_byte = data + data_size - 1;
841 
842  /* initialize the 1st cell and set its dimensions to whole plane */
843  curr_cell.xpos = curr_cell.ypos = 0;
844  curr_cell.width = plane->width >> 2;
845  curr_cell.height = plane->height >> 2;
846  curr_cell.tree = 0; // we are in the MC tree now
847  curr_cell.mv_ptr = 0; // no motion vector = INTRA cell
848 
849  return parse_bintree(ctx, avctx, plane, INTRA_NULL, &curr_cell, CELL_STACK_MAX, strip_width);
850 }
851 
852 
853 #define OS_HDR_ID MKBETAG('F', 'R', 'M', 'H')
854 
856  const uint8_t *buf, int buf_size)
857 {
858  const uint8_t *buf_ptr = buf, *bs_hdr;
859  uint32_t frame_num, word2, check_sum, data_size;
860  uint32_t y_offset, u_offset, v_offset, starts[3], ends[3];
861  uint16_t height, width;
862  int i, j;
863 
864  /* parse and check the OS header */
865  frame_num = bytestream_get_le32(&buf_ptr);
866  word2 = bytestream_get_le32(&buf_ptr);
867  check_sum = bytestream_get_le32(&buf_ptr);
868  data_size = bytestream_get_le32(&buf_ptr);
869 
870  if ((frame_num ^ word2 ^ data_size ^ OS_HDR_ID) != check_sum) {
871  av_log(avctx, AV_LOG_ERROR, "OS header checksum mismatch!\n");
872  return AVERROR_INVALIDDATA;
873  }
874 
875  /* parse the bitstream header */
876  bs_hdr = buf_ptr;
877 
878  if (bytestream_get_le16(&buf_ptr) != 32) {
879  av_log(avctx, AV_LOG_ERROR, "Unsupported codec version!\n");
880  return AVERROR_INVALIDDATA;
881  }
882 
883  ctx->frame_num = frame_num;
884  ctx->frame_flags = bytestream_get_le16(&buf_ptr);
885  ctx->data_size = (bytestream_get_le32(&buf_ptr) + 7) >> 3;
886  ctx->cb_offset = *buf_ptr++;
887 
888  if (ctx->data_size == 16)
889  return 4;
890  if (ctx->data_size > buf_size)
891  ctx->data_size = buf_size;
892 
893  buf_ptr += 3; // skip reserved byte and checksum
894 
895  /* check frame dimensions */
896  height = bytestream_get_le16(&buf_ptr);
897  width = bytestream_get_le16(&buf_ptr);
898  if (av_image_check_size(width, height, 0, avctx))
899  return AVERROR_INVALIDDATA;
900 
901  if (width != ctx->width || height != ctx->height) {
902  int res;
903 
904  av_dlog(avctx, "Frame dimensions changed!\n");
905 
906  if (width < 16 || width > 640 ||
907  height < 16 || height > 480 ||
908  width & 3 || height & 3) {
909  av_log(avctx, AV_LOG_ERROR,
910  "Invalid picture dimensions: %d x %d!\n", width, height);
911  return AVERROR_INVALIDDATA;
912  }
913 
914  ctx->width = width;
915  ctx->height = height;
916 
917  free_frame_buffers(ctx);
918  if ((res = allocate_frame_buffers(ctx, avctx)) < 0)
919  return res;
920  avcodec_set_dimensions(avctx, width, height);
921  }
922 
923  y_offset = bytestream_get_le32(&buf_ptr);
924  v_offset = bytestream_get_le32(&buf_ptr);
925  u_offset = bytestream_get_le32(&buf_ptr);
926 
927  /* unfortunately there is no common order of planes in the buffer */
928  /* so we use that sorting algo for determining planes data sizes */
929  starts[0] = y_offset;
930  starts[1] = v_offset;
931  starts[2] = u_offset;
932 
933  for (j = 0; j < 3; j++) {
934  ends[j] = ctx->data_size;
935  for (i = 2; i >= 0; i--)
936  if (starts[i] < ends[j] && starts[i] > starts[j])
937  ends[j] = starts[i];
938  }
939 
940  ctx->y_data_size = ends[0] - starts[0];
941  ctx->v_data_size = ends[1] - starts[1];
942  ctx->u_data_size = ends[2] - starts[2];
943  if (FFMAX3(y_offset, v_offset, u_offset) >= ctx->data_size - 16 ||
944  FFMIN3(ctx->y_data_size, ctx->v_data_size, ctx->u_data_size) <= 0) {
945  av_log(avctx, AV_LOG_ERROR, "One of the y/u/v offsets is invalid\n");
946  return AVERROR_INVALIDDATA;
947  }
948 
949  ctx->y_data_ptr = bs_hdr + y_offset;
950  ctx->v_data_ptr = bs_hdr + v_offset;
951  ctx->u_data_ptr = bs_hdr + u_offset;
952  ctx->alt_quant = buf_ptr + sizeof(uint32_t);
953 
954  if (ctx->data_size == 16) {
955  av_log(avctx, AV_LOG_DEBUG, "Sync frame encountered!\n");
956  return 16;
957  }
958 
959  if (ctx->frame_flags & BS_8BIT_PEL) {
960  av_log_ask_for_sample(avctx, "8-bit pixel format\n");
961  return AVERROR_PATCHWELCOME;
962  }
963 
964  if (ctx->frame_flags & BS_MV_X_HALF || ctx->frame_flags & BS_MV_Y_HALF) {
965  av_log_ask_for_sample(avctx, "halfpel motion vectors\n");
966  return AVERROR_PATCHWELCOME;
967  }
968 
969  return 0;
970 }
971 
972 
982 static void output_plane(const Plane *plane, int buf_sel, uint8_t *dst, int dst_pitch)
983 {
984  int x,y;
985  const uint8_t *src = plane->pixels[buf_sel];
986  uint32_t pitch = plane->pitch;
987 
988  for (y = 0; y < plane->height; y++) {
989  /* convert four pixels at once using SWAR */
990  for (x = 0; x < plane->width >> 2; x++) {
991  AV_WN32A(dst, (AV_RN32A(src) & 0x7F7F7F7F) << 1);
992  src += 4;
993  dst += 4;
994  }
995 
996  for (x <<= 2; x < plane->width; x++)
997  *dst++ = *src++ << 1;
998 
999  src += pitch - plane->width;
1000  dst += dst_pitch - plane->width;
1001  }
1002 }
1003 
1004 
1006 {
1007  Indeo3DecodeContext *ctx = avctx->priv_data;
1008 
1009  ctx->avctx = avctx;
1010  ctx->width = avctx->width;
1011  ctx->height = avctx->height;
1012  avctx->pix_fmt = PIX_FMT_YUV410P;
1013 
1015 
1016  dsputil_init(&ctx->dsp, avctx);
1017 
1018  allocate_frame_buffers(ctx, avctx);
1019 
1020  return 0;
1021 }
1022 
1023 
1024 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
1025  AVPacket *avpkt)
1026 {
1027  Indeo3DecodeContext *ctx = avctx->priv_data;
1028  const uint8_t *buf = avpkt->data;
1029  int buf_size = avpkt->size;
1030  int res;
1031 
1032  res = decode_frame_headers(ctx, avctx, buf, buf_size);
1033  if (res < 0)
1034  return res;
1035 
1036  /* skip sync(null) frames */
1037  if (res) {
1038  // we have processed 16 bytes but no data was decoded
1039  *data_size = 0;
1040  return buf_size;
1041  }
1042 
1043  /* skip droppable INTER frames if requested */
1044  if (ctx->frame_flags & BS_NONREF &&
1045  (avctx->skip_frame >= AVDISCARD_NONREF))
1046  return 0;
1047 
1048  /* skip INTER frames if requested */
1049  if (!(ctx->frame_flags & BS_KEYFRAME) && avctx->skip_frame >= AVDISCARD_NONKEY)
1050  return 0;
1051 
1052  /* use BS_BUFFER flag for buffer switching */
1053  ctx->buf_sel = (ctx->frame_flags >> BS_BUFFER) & 1;
1054 
1055  /* decode luma plane */
1056  if ((res = decode_plane(ctx, avctx, ctx->planes, ctx->y_data_ptr, ctx->y_data_size, 40)))
1057  return res;
1058 
1059  /* decode chroma planes */
1060  if ((res = decode_plane(ctx, avctx, &ctx->planes[1], ctx->u_data_ptr, ctx->u_data_size, 10)))
1061  return res;
1062 
1063  if ((res = decode_plane(ctx, avctx, &ctx->planes[2], ctx->v_data_ptr, ctx->v_data_size, 10)))
1064  return res;
1065 
1066  if (ctx->frame.data[0])
1067  avctx->release_buffer(avctx, &ctx->frame);
1068 
1069  ctx->frame.reference = 0;
1070  if ((res = avctx->get_buffer(avctx, &ctx->frame)) < 0) {
1071  av_log(ctx->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1072  return res;
1073  }
1074 
1075  output_plane(&ctx->planes[0], ctx->buf_sel, ctx->frame.data[0], ctx->frame.linesize[0]);
1076  output_plane(&ctx->planes[1], ctx->buf_sel, ctx->frame.data[1], ctx->frame.linesize[1]);
1077  output_plane(&ctx->planes[2], ctx->buf_sel, ctx->frame.data[2], ctx->frame.linesize[2]);
1078 
1079  *data_size = sizeof(AVFrame);
1080  *(AVFrame*)data = ctx->frame;
1081 
1082  return buf_size;
1083 }
1084 
1085 
1087 {
1088  Indeo3DecodeContext *ctx = avctx->priv_data;
1089 
1090  free_frame_buffers(avctx->priv_data);
1091 
1092  if (ctx->frame.data[0])
1093  avctx->release_buffer(avctx, &ctx->frame);
1094 
1095  return 0;
1096 }
1097 
1099  .name = "indeo3",
1100  .type = AVMEDIA_TYPE_VIDEO,
1101  .id = CODEC_ID_INDEO3,
1102  .priv_data_size = sizeof(Indeo3DecodeContext),
1103  .init = decode_init,
1104  .close = decode_close,
1105  .decode = decode_frame,
1106  .long_name = NULL_IF_CONFIG_SMALL("Intel Indeo 3"),
1107 };