indeo: Convert to the new bitstream reader
[libav.git] / libavcodec / indeo3.c
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
22 /**
23 * @file
24 * This is a decoder for Intel Indeo Video v3.
25 * It is based on vector quantization, run-length coding and motion compensation.
26 * Known container formats: .avi and .mov
27 * Known FOURCCs: 'IV31', 'IV32'
28 *
29 * @see http://wiki.multimedia.cx/index.php?title=Indeo_3
30 */
31
32 #include "libavutil/imgutils.h"
33 #include "libavutil/intreadwrite.h"
34
35 #include "avcodec.h"
36 #include "bitstream.h"
37 #include "bytestream.h"
38 #include "hpeldsp.h"
39 #include "internal.h"
40
41 #include "indeo3data.h"
42
43 /* RLE opcodes. */
44 enum {
45 RLE_ESC_F9 = 249, ///< same as RLE_ESC_FA + do the same with next block
46 RLE_ESC_FA = 250, ///< INTRA: skip block, INTER: copy data from reference
47 RLE_ESC_FB = 251, ///< apply null delta to N blocks / skip N blocks
48 RLE_ESC_FC = 252, ///< same as RLE_ESC_FD + do the same with next block
49 RLE_ESC_FD = 253, ///< apply null delta to all remaining lines of this block
50 RLE_ESC_FE = 254, ///< apply null delta to all lines up to the 3rd line
51 RLE_ESC_FF = 255 ///< apply null delta to all lines up to the 2nd line
52 };
53
54
55 /* Some constants for parsing frame bitstream flags. */
56 #define BS_8BIT_PEL (1 << 1) ///< 8-bit pixel bitdepth indicator
57 #define BS_KEYFRAME (1 << 2) ///< intra frame indicator
58 #define BS_MV_Y_HALF (1 << 4) ///< vertical mv halfpel resolution indicator
59 #define BS_MV_X_HALF (1 << 5) ///< horizontal mv halfpel resolution indicator
60 #define BS_NONREF (1 << 8) ///< nonref (discardable) frame indicator
61 #define BS_BUFFER 9 ///< indicates which of two frame buffers should be used
62
63
64 typedef struct Plane {
65 uint8_t *buffers[2];
66 uint8_t *pixels[2]; ///< pointer to the actual pixel data of the buffers above
67 uint32_t width;
68 uint32_t height;
69 ptrdiff_t pitch;
70 } Plane;
71
72 #define CELL_STACK_MAX 20
73
74 typedef struct Cell {
75 int16_t xpos; ///< cell coordinates in 4x4 blocks
76 int16_t ypos;
77 int16_t width; ///< cell width in 4x4 blocks
78 int16_t height; ///< cell height in 4x4 blocks
79 uint8_t tree; ///< tree id: 0- MC tree, 1 - VQ tree
80 const int8_t *mv_ptr; ///< ptr to the motion vector if any
81 } Cell;
82
83 typedef struct Indeo3DecodeContext {
84 AVCodecContext *avctx;
85 HpelDSPContext hdsp;
86
87 BitstreamContext bc;
88 int need_resync;
89 int skip_bits;
90 const uint8_t *next_cell_data;
91 const uint8_t *last_byte;
92 const int8_t *mc_vectors;
93 unsigned num_vectors; ///< number of motion vectors in mc_vectors
94
95 int16_t width, height;
96 uint32_t frame_num; ///< current frame number (zero-based)
97 uint32_t data_size; ///< size of the frame data in bytes
98 uint16_t frame_flags; ///< frame properties
99 uint8_t cb_offset; ///< needed for selecting VQ tables
100 uint8_t buf_sel; ///< active frame buffer: 0 - primary, 1 -secondary
101 const uint8_t *y_data_ptr;
102 const uint8_t *v_data_ptr;
103 const uint8_t *u_data_ptr;
104 int32_t y_data_size;
105 int32_t v_data_size;
106 int32_t u_data_size;
107 const uint8_t *alt_quant; ///< secondary VQ table set for the modes 1 and 4
108 Plane planes[3];
109 } Indeo3DecodeContext;
110
111
112 static uint8_t requant_tab[8][128];
113
114 /*
115 * Build the static requantization table.
116 * This table is used to remap pixel values according to a specific
117 * quant index and thus avoid overflows while adding deltas.
118 */
119 static av_cold void build_requant_tab(void)
120 {
121 static int8_t offsets[8] = { 1, 1, 2, -3, -3, 3, 4, 4 };
122 static int8_t deltas [8] = { 0, 1, 0, 4, 4, 1, 0, 1 };
123
124 int i, j, step;
125
126 for (i = 0; i < 8; i++) {
127 step = i + 2;
128 for (j = 0; j < 128; j++)
129 requant_tab[i][j] = (j + offsets[i]) / step * step + deltas[i];
130 }
131
132 /* some last elements calculated above will have values >= 128 */
133 /* pixel values shall never exceed 127 so set them to non-overflowing values */
134 /* according with the quantization step of the respective section */
135 requant_tab[0][127] = 126;
136 requant_tab[1][119] = 118;
137 requant_tab[1][120] = 118;
138 requant_tab[2][126] = 124;
139 requant_tab[2][127] = 124;
140 requant_tab[6][124] = 120;
141 requant_tab[6][125] = 120;
142 requant_tab[6][126] = 120;
143 requant_tab[6][127] = 120;
144
145 /* Patch for compatibility with the Intel's binary decoders */
146 requant_tab[1][7] = 10;
147 requant_tab[4][8] = 10;
148 }
149
150
151 static av_cold int allocate_frame_buffers(Indeo3DecodeContext *ctx,
152 AVCodecContext *avctx)
153 {
154 int p, luma_width, luma_height, chroma_width, chroma_height;
155 int luma_size, chroma_size;
156 ptrdiff_t luma_pitch, chroma_pitch;
157
158 luma_width = ctx->width;
159 luma_height = ctx->height;
160
161 if (luma_width < 16 || luma_width > 640 ||
162 luma_height < 16 || luma_height > 480 ||
163 luma_width & 3 || luma_height & 3) {
164 av_log(avctx, AV_LOG_ERROR, "Invalid picture dimensions: %d x %d!\n",
165 luma_width, luma_height);
166 return AVERROR_INVALIDDATA;
167 }
168
169 chroma_width = FFALIGN(luma_width >> 2, 4);
170 chroma_height = FFALIGN(luma_height >> 2, 4);
171
172 luma_pitch = FFALIGN(luma_width, 16);
173 chroma_pitch = FFALIGN(chroma_width, 16);
174
175 /* Calculate size of the luminance plane. */
176 /* Add one line more for INTRA prediction. */
177 luma_size = luma_pitch * (luma_height + 1);
178
179 /* Calculate size of a chrominance planes. */
180 /* Add one line more for INTRA prediction. */
181 chroma_size = chroma_pitch * (chroma_height + 1);
182
183 /* allocate frame buffers */
184 for (p = 0; p < 3; p++) {
185 ctx->planes[p].pitch = !p ? luma_pitch : chroma_pitch;
186 ctx->planes[p].width = !p ? luma_width : chroma_width;
187 ctx->planes[p].height = !p ? luma_height : chroma_height;
188
189 ctx->planes[p].buffers[0] = av_malloc(!p ? luma_size : chroma_size);
190 ctx->planes[p].buffers[1] = av_malloc(!p ? luma_size : chroma_size);
191
192 /* fill the INTRA prediction lines with the middle pixel value = 64 */
193 memset(ctx->planes[p].buffers[0], 0x40, ctx->planes[p].pitch);
194 memset(ctx->planes[p].buffers[1], 0x40, ctx->planes[p].pitch);
195
196 /* set buffer pointers = buf_ptr + pitch and thus skip the INTRA prediction line */
197 ctx->planes[p].pixels[0] = ctx->planes[p].buffers[0] + ctx->planes[p].pitch;
198 ctx->planes[p].pixels[1] = ctx->planes[p].buffers[1] + ctx->planes[p].pitch;
199 memset(ctx->planes[p].pixels[0], 0, ctx->planes[p].pitch * ctx->planes[p].height);
200 memset(ctx->planes[p].pixels[1], 0, ctx->planes[p].pitch * ctx->planes[p].height);
201 }
202
203 return 0;
204 }
205
206
207 static av_cold void free_frame_buffers(Indeo3DecodeContext *ctx)
208 {
209 int p;
210
211 for (p = 0; p < 3; p++) {
212 av_freep(&ctx->planes[p].buffers[0]);
213 av_freep(&ctx->planes[p].buffers[1]);
214 ctx->planes[p].pixels[0] = ctx->planes[p].pixels[1] = 0;
215 }
216 }
217
218
219 /**
220 * Copy pixels of the cell(x + mv_x, y + mv_y) from the previous frame into
221 * the cell(x, y) in the current frame.
222 *
223 * @param ctx pointer to the decoder context
224 * @param plane pointer to the plane descriptor
225 * @param cell pointer to the cell descriptor
226 */
227 static int copy_cell(Indeo3DecodeContext *ctx, Plane *plane, Cell *cell)
228 {
229 int h, w, mv_x, mv_y, offset, offset_dst;
230 uint8_t *src, *dst;
231
232 /* setup output and reference pointers */
233 offset_dst = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
234 dst = plane->pixels[ctx->buf_sel] + offset_dst;
235 mv_y = cell->mv_ptr[0];
236 mv_x = cell->mv_ptr[1];
237
238 /* -1 because there is an extra line on top for prediction */
239 if ((cell->ypos << 2) + mv_y < -1 || (cell->xpos << 2) + mv_x < 0 ||
240 ((cell->ypos + cell->height) << 2) + mv_y > plane->height ||
241 ((cell->xpos + cell->width) << 2) + mv_x > plane->width) {
242 av_log(ctx->avctx, AV_LOG_ERROR,
243 "Motion vectors point out of the frame.\n");
244 return AVERROR_INVALIDDATA;
245 }
246
247 offset = offset_dst + mv_y * plane->pitch + mv_x;
248 src = plane->pixels[ctx->buf_sel ^ 1] + offset;
249
250 h = cell->height << 2;
251
252 for (w = cell->width; w > 0;) {
253 /* copy using 16xH blocks */
254 if (!((cell->xpos << 2) & 15) && w >= 4) {
255 for (; w >= 4; src += 16, dst += 16, w -= 4)
256 ctx->hdsp.put_pixels_tab[0][0](dst, src, plane->pitch, h);
257 }
258
259 /* copy using 8xH blocks */
260 if (!((cell->xpos << 2) & 7) && w >= 2) {
261 ctx->hdsp.put_pixels_tab[1][0](dst, src, plane->pitch, h);
262 w -= 2;
263 src += 8;
264 dst += 8;
265 } else if (w >= 1) {
266 ctx->hdsp.put_pixels_tab[2][0](dst, src, plane->pitch, h);
267 w--;
268 src += 4;
269 dst += 4;
270 }
271 }
272
273 return 0;
274 }
275
276
277 /* Average 4/8 pixels at once without rounding using SWAR */
278 #define AVG_32(dst, src, ref) \
279 AV_WN32A(dst, ((AV_RN32(src) + AV_RN32(ref)) >> 1) & 0x7F7F7F7FUL)
280
281 #define AVG_64(dst, src, ref) \
282 AV_WN64A(dst, ((AV_RN64(src) + AV_RN64(ref)) >> 1) & 0x7F7F7F7F7F7F7F7FULL)
283
284
285 /*
286 * Replicate each even pixel as follows:
287 * ABCDEFGH -> AACCEEGG
288 */
289 static inline uint64_t replicate64(uint64_t a) {
290 #if HAVE_BIGENDIAN
291 a &= 0xFF00FF00FF00FF00ULL;
292 a |= a >> 8;
293 #else
294 a &= 0x00FF00FF00FF00FFULL;
295 a |= a << 8;
296 #endif
297 return a;
298 }
299
300 static inline uint32_t replicate32(uint32_t a) {
301 #if HAVE_BIGENDIAN
302 a &= 0xFF00FF00UL;
303 a |= a >> 8;
304 #else
305 a &= 0x00FF00FFUL;
306 a |= a << 8;
307 #endif
308 return a;
309 }
310
311
312 /* Fill n lines with 64-bit pixel value pix */
313 static inline void fill_64(uint8_t *dst, const uint64_t pix, int32_t n,
314 int32_t row_offset)
315 {
316 for (; n > 0; dst += row_offset, n--)
317 AV_WN64A(dst, pix);
318 }
319
320
321 /* Error codes for cell decoding. */
322 enum {
323 IV3_NOERR = 0,
324 IV3_BAD_RLE = 1,
325 IV3_BAD_DATA = 2,
326 IV3_BAD_COUNTER = 3,
327 IV3_UNSUPPORTED = 4,
328 IV3_OUT_OF_DATA = 5
329 };
330
331
332 #define BUFFER_PRECHECK \
333 if (*data_ptr >= last_ptr) \
334 return IV3_OUT_OF_DATA; \
335
336 #define RLE_BLOCK_COPY \
337 if (cell->mv_ptr || !skip_flag) \
338 ctx->hdsp.put_pixels_tab[2][0](dst, ref, row_offset, 4 << v_zoom)
339
340 #define RLE_BLOCK_COPY_8 \
341 pix64 = AV_RN64(ref);\
342 if (is_first_row) {/* special prediction case: top line of a cell */\
343 pix64 = replicate64(pix64);\
344 fill_64(dst + row_offset, pix64, 7, row_offset);\
345 AVG_64(dst, ref, dst + row_offset);\
346 } else \
347 fill_64(dst, pix64, 8, row_offset)
348
349 #define RLE_LINES_COPY \
350 ctx->hdsp.put_pixels_tab[2][0](dst, ref, row_offset, num_lines << v_zoom)
351
352 #define RLE_LINES_COPY_M10 \
353 pix64 = AV_RN64(ref);\
354 if (is_top_of_cell) {\
355 pix64 = replicate64(pix64);\
356 fill_64(dst + row_offset, pix64, (num_lines << 1) - 1, row_offset);\
357 AVG_64(dst, ref, dst + row_offset);\
358 } else \
359 fill_64(dst, pix64, num_lines << 1, row_offset)
360
361 #define APPLY_DELTA_4 \
362 AV_WN16A(dst + line_offset ,\
363 (AV_RN16(ref ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
364 AV_WN16A(dst + line_offset + 2,\
365 (AV_RN16(ref + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
366 if (mode >= 3) {\
367 if (is_top_of_cell && !cell->ypos) {\
368 AV_COPY32U(dst, dst + row_offset);\
369 } else {\
370 AVG_32(dst, ref, dst + row_offset);\
371 }\
372 }
373
374 #define APPLY_DELTA_8 \
375 /* apply two 32-bit VQ deltas to next even line */\
376 if (is_top_of_cell) { \
377 AV_WN32A(dst + row_offset , \
378 (replicate32(AV_RN32(ref )) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
379 AV_WN32A(dst + row_offset + 4, \
380 (replicate32(AV_RN32(ref + 4)) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
381 } else { \
382 AV_WN32A(dst + row_offset , \
383 (AV_RN32(ref ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
384 AV_WN32A(dst + row_offset + 4, \
385 (AV_RN32(ref + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
386 } \
387 /* odd lines are not coded but rather interpolated/replicated */\
388 /* first line of the cell on the top of image? - replicate */\
389 /* otherwise - interpolate */\
390 if (is_top_of_cell && !cell->ypos) {\
391 AV_COPY64U(dst, dst + row_offset);\
392 } else \
393 AVG_64(dst, ref, dst + row_offset);
394
395
396 #define APPLY_DELTA_1011_INTER \
397 if (mode == 10) { \
398 AV_WN32A(dst , \
399 (AV_RN32(dst ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
400 AV_WN32A(dst + 4 , \
401 (AV_RN32(dst + 4 ) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
402 AV_WN32A(dst + row_offset , \
403 (AV_RN32(dst + row_offset ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
404 AV_WN32A(dst + row_offset + 4, \
405 (AV_RN32(dst + row_offset + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
406 } else { \
407 AV_WN16A(dst , \
408 (AV_RN16(dst ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
409 AV_WN16A(dst + 2 , \
410 (AV_RN16(dst + 2 ) + delta_tab->deltas[dyad2]) & 0x7F7F);\
411 AV_WN16A(dst + row_offset , \
412 (AV_RN16(dst + row_offset ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
413 AV_WN16A(dst + row_offset + 2, \
414 (AV_RN16(dst + row_offset + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
415 }
416
417
418 static int decode_cell_data(Indeo3DecodeContext *ctx, Cell *cell,
419 uint8_t *block, uint8_t *ref_block,
420 ptrdiff_t row_offset, int h_zoom, int v_zoom, int mode,
421 const vqEntry *delta[2], int swap_quads[2],
422 const uint8_t **data_ptr, const uint8_t *last_ptr)
423 {
424 int x, y, line, num_lines;
425 int rle_blocks = 0;
426 uint8_t code, *dst, *ref;
427 const vqEntry *delta_tab;
428 unsigned int dyad1, dyad2;
429 uint64_t pix64;
430 int skip_flag = 0, is_top_of_cell, is_first_row = 1;
431 int blk_row_offset, line_offset;
432
433 blk_row_offset = (row_offset << (2 + v_zoom)) - (cell->width << 2);
434 line_offset = v_zoom ? row_offset : 0;
435
436 if (cell->height & v_zoom || cell->width & h_zoom)
437 return IV3_BAD_DATA;
438
439 for (y = 0; y < cell->height; is_first_row = 0, y += 1 + v_zoom) {
440 for (x = 0; x < cell->width; x += 1 + h_zoom) {
441 ref = ref_block;
442 dst = block;
443
444 if (rle_blocks > 0) {
445 if (mode <= 4) {
446 RLE_BLOCK_COPY;
447 } else if (mode == 10 && !cell->mv_ptr) {
448 RLE_BLOCK_COPY_8;
449 }
450 rle_blocks--;
451 } else {
452 for (line = 0; line < 4;) {
453 num_lines = 1;
454 is_top_of_cell = is_first_row && !line;
455
456 /* select primary VQ table for odd, secondary for even lines */
457 if (mode <= 4)
458 delta_tab = delta[line & 1];
459 else
460 delta_tab = delta[1];
461 BUFFER_PRECHECK;
462 code = bytestream_get_byte(data_ptr);
463 if (code < 248) {
464 if (code < delta_tab->num_dyads) {
465 BUFFER_PRECHECK;
466 dyad1 = bytestream_get_byte(data_ptr);
467 dyad2 = code;
468 if (dyad1 >= delta_tab->num_dyads || dyad1 >= 248)
469 return IV3_BAD_DATA;
470 } else {
471 /* process QUADS */
472 code -= delta_tab->num_dyads;
473 dyad1 = code / delta_tab->quad_exp;
474 dyad2 = code % delta_tab->quad_exp;
475 if (swap_quads[line & 1])
476 FFSWAP(unsigned int, dyad1, dyad2);
477 }
478 if (mode <= 4) {
479 APPLY_DELTA_4;
480 } else if (mode == 10 && !cell->mv_ptr) {
481 APPLY_DELTA_8;
482 } else {
483 APPLY_DELTA_1011_INTER;
484 }
485 } else {
486 /* process RLE codes */
487 switch (code) {
488 case RLE_ESC_FC:
489 skip_flag = 0;
490 rle_blocks = 1;
491 code = 253;
492 /* FALLTHROUGH */
493 case RLE_ESC_FF:
494 case RLE_ESC_FE:
495 case RLE_ESC_FD:
496 num_lines = 257 - code - line;
497 if (num_lines <= 0)
498 return IV3_BAD_RLE;
499 if (mode <= 4) {
500 RLE_LINES_COPY;
501 } else if (mode == 10 && !cell->mv_ptr) {
502 RLE_LINES_COPY_M10;
503 }
504 break;
505 case RLE_ESC_FB:
506 BUFFER_PRECHECK;
507 code = bytestream_get_byte(data_ptr);
508 rle_blocks = (code & 0x1F) - 1; /* set block counter */
509 if (code >= 64 || rle_blocks < 0)
510 return IV3_BAD_COUNTER;
511 skip_flag = code & 0x20;
512 num_lines = 4 - line; /* enforce next block processing */
513 if (mode >= 10 || (cell->mv_ptr || !skip_flag)) {
514 if (mode <= 4) {
515 RLE_LINES_COPY;
516 } else if (mode == 10 && !cell->mv_ptr) {
517 RLE_LINES_COPY_M10;
518 }
519 }
520 break;
521 case RLE_ESC_F9:
522 skip_flag = 1;
523 rle_blocks = 1;
524 /* FALLTHROUGH */
525 case RLE_ESC_FA:
526 if (line)
527 return IV3_BAD_RLE;
528 num_lines = 4; /* enforce next block processing */
529 if (cell->mv_ptr) {
530 if (mode <= 4) {
531 RLE_LINES_COPY;
532 } else if (mode == 10 && !cell->mv_ptr) {
533 RLE_LINES_COPY_M10;
534 }
535 }
536 break;
537 default:
538 return IV3_UNSUPPORTED;
539 }
540 }
541
542 line += num_lines;
543 ref += row_offset * (num_lines << v_zoom);
544 dst += row_offset * (num_lines << v_zoom);
545 }
546 }
547
548 /* move to next horizontal block */
549 block += 4 << h_zoom;
550 ref_block += 4 << h_zoom;
551 }
552
553 /* move to next line of blocks */
554 ref_block += blk_row_offset;
555 block += blk_row_offset;
556 }
557 return IV3_NOERR;
558 }
559
560
561 /**
562 * Decode a vector-quantized cell.
563 * It consists of several routines, each of which handles one or more "modes"
564 * with which a cell can be encoded.
565 *
566 * @param ctx pointer to the decoder context
567 * @param avctx ptr to the AVCodecContext
568 * @param plane pointer to the plane descriptor
569 * @param cell pointer to the cell descriptor
570 * @param data_ptr pointer to the compressed data
571 * @param last_ptr pointer to the last byte to catch reads past end of buffer
572 * @return number of consumed bytes or negative number in case of error
573 */
574 static int decode_cell(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
575 Plane *plane, Cell *cell, const uint8_t *data_ptr,
576 const uint8_t *last_ptr)
577 {
578 int x, mv_x, mv_y, mode, vq_index, prim_indx, second_indx;
579 int zoom_fac;
580 int offset, error = 0, swap_quads[2];
581 uint8_t code, *block, *ref_block = 0;
582 const vqEntry *delta[2];
583 const uint8_t *data_start = data_ptr;
584
585 /* get coding mode and VQ table index from the VQ descriptor byte */
586 code = *data_ptr++;
587 mode = code >> 4;
588 vq_index = code & 0xF;
589
590 /* setup output and reference pointers */
591 offset = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
592 block = plane->pixels[ctx->buf_sel] + offset;
593 if (!cell->mv_ptr) {
594 /* use previous line as reference for INTRA cells */
595 ref_block = block - plane->pitch;
596 } else if (mode >= 10) {
597 /* for mode 10 and 11 INTER first copy the predicted cell into the current one */
598 /* so we don't need to do data copying for each RLE code later */
599 int ret = copy_cell(ctx, plane, cell);
600 if (ret < 0)
601 return ret;
602 } else {
603 /* set the pointer to the reference pixels for modes 0-4 INTER */
604 mv_y = cell->mv_ptr[0];
605 mv_x = cell->mv_ptr[1];
606
607 /* -1 because there is an extra line on top for prediction */
608 if ((cell->ypos << 2) + mv_y < -1 || (cell->xpos << 2) + mv_x < 0 ||
609 ((cell->ypos + cell->height) << 2) + mv_y > plane->height ||
610 ((cell->xpos + cell->width) << 2) + mv_x > plane->width) {
611 av_log(ctx->avctx, AV_LOG_ERROR,
612 "Motion vectors point out of the frame.\n");
613 return AVERROR_INVALIDDATA;
614 }
615
616 offset += mv_y * plane->pitch + mv_x;
617 ref_block = plane->pixels[ctx->buf_sel ^ 1] + offset;
618 }
619
620 /* select VQ tables as follows: */
621 /* modes 0 and 3 use only the primary table for all lines in a block */
622 /* while modes 1 and 4 switch between primary and secondary tables on alternate lines */
623 if (mode == 1 || mode == 4) {
624 code = ctx->alt_quant[vq_index];
625 prim_indx = (code >> 4) + ctx->cb_offset;
626 second_indx = (code & 0xF) + ctx->cb_offset;
627 } else {
628 vq_index += ctx->cb_offset;
629 prim_indx = second_indx = vq_index;
630 }
631
632 if (prim_indx >= 24 || second_indx >= 24) {
633 av_log(avctx, AV_LOG_ERROR, "Invalid VQ table indexes! Primary: %d, secondary: %d!\n",
634 prim_indx, second_indx);
635 return AVERROR_INVALIDDATA;
636 }
637
638 delta[0] = &vq_tab[second_indx];
639 delta[1] = &vq_tab[prim_indx];
640 swap_quads[0] = second_indx >= 16;
641 swap_quads[1] = prim_indx >= 16;
642
643 /* requantize the prediction if VQ index of this cell differs from VQ index */
644 /* of the predicted cell in order to avoid overflows. */
645 if (vq_index >= 8 && ref_block) {
646 for (x = 0; x < cell->width << 2; x++)
647 ref_block[x] = requant_tab[vq_index & 7][ref_block[x]];
648 }
649
650 error = IV3_NOERR;
651
652 switch (mode) {
653 case 0: /*------------------ MODES 0 & 1 (4x4 block processing) --------------------*/
654 case 1:
655 case 3: /*------------------ MODES 3 & 4 (4x8 block processing) --------------------*/
656 case 4:
657 if (mode >= 3 && cell->mv_ptr) {
658 av_log(avctx, AV_LOG_ERROR, "Attempt to apply Mode 3/4 to an INTER cell!\n");
659 return AVERROR_INVALIDDATA;
660 }
661
662 zoom_fac = mode >= 3;
663 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
664 0, zoom_fac, mode, delta, swap_quads,
665 &data_ptr, last_ptr);
666 break;
667 case 10: /*-------------------- MODE 10 (8x8 block processing) ---------------------*/
668 case 11: /*----------------- MODE 11 (4x8 INTER block processing) ------------------*/
669 if (mode == 10 && !cell->mv_ptr) { /* MODE 10 INTRA processing */
670 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
671 1, 1, mode, delta, swap_quads,
672 &data_ptr, last_ptr);
673 } else { /* mode 10 and 11 INTER processing */
674 if (mode == 11 && !cell->mv_ptr) {
675 av_log(avctx, AV_LOG_ERROR, "Attempt to use Mode 11 for an INTRA cell!\n");
676 return AVERROR_INVALIDDATA;
677 }
678
679 zoom_fac = mode == 10;
680 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
681 zoom_fac, 1, mode, delta, swap_quads,
682 &data_ptr, last_ptr);
683 }
684 break;
685 default:
686 av_log(avctx, AV_LOG_ERROR, "Unsupported coding mode: %d\n", mode);
687 return AVERROR_INVALIDDATA;
688 }//switch mode
689
690 switch (error) {
691 case IV3_BAD_RLE:
692 av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE code %X is not allowed at the current line\n",
693 mode, data_ptr[-1]);
694 return AVERROR_INVALIDDATA;
695 case IV3_BAD_DATA:
696 av_log(avctx, AV_LOG_ERROR, "Mode %d: invalid VQ data\n", mode);
697 return AVERROR_INVALIDDATA;
698 case IV3_BAD_COUNTER:
699 av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE-FB invalid counter: %d\n", mode, code);
700 return AVERROR_INVALIDDATA;
701 case IV3_UNSUPPORTED:
702 av_log(avctx, AV_LOG_ERROR, "Mode %d: unsupported RLE code: %X\n", mode, data_ptr[-1]);
703 return AVERROR_INVALIDDATA;
704 case IV3_OUT_OF_DATA:
705 av_log(avctx, AV_LOG_ERROR, "Mode %d: attempt to read past end of buffer\n", mode);
706 return AVERROR_INVALIDDATA;
707 }
708
709 return data_ptr - data_start; /* report number of bytes consumed from the input buffer */
710 }
711
712
713 /* Binary tree codes. */
714 enum {
715 H_SPLIT = 0,
716 V_SPLIT = 1,
717 INTRA_NULL = 2,
718 INTER_DATA = 3
719 };
720
721
722 #define SPLIT_CELL(size, new_size) (new_size) = ((size) > 2) ? ((((size) + 2) >> 2) << 1) : 1
723
724 #define UPDATE_BITPOS(n) \
725 ctx->skip_bits += (n); \
726 ctx->need_resync = 1
727
728 #define RESYNC_BITSTREAM \
729 if (ctx->need_resync && !(bitstream_tell(&ctx->bc) & 7)) { \
730 bitstream_skip(&ctx->bc, ctx->skip_bits); \
731 ctx->skip_bits = 0; \
732 ctx->need_resync = 0; \
733 }
734
735 #define CHECK_CELL \
736 if (curr_cell.xpos + curr_cell.width > (plane->width >> 2) || \
737 curr_cell.ypos + curr_cell.height > (plane->height >> 2)) { \
738 av_log(avctx, AV_LOG_ERROR, "Invalid cell: x=%d, y=%d, w=%d, h=%d\n", \
739 curr_cell.xpos, curr_cell.ypos, curr_cell.width, curr_cell.height); \
740 return AVERROR_INVALIDDATA; \
741 }
742
743
744 static int parse_bintree(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
745 Plane *plane, int code, Cell *ref_cell,
746 const int depth, const int strip_width)
747 {
748 Cell curr_cell;
749 int bytes_used, ret;
750
751 if (depth <= 0) {
752 av_log(avctx, AV_LOG_ERROR, "Stack overflow (corrupted binary tree)!\n");
753 return AVERROR_INVALIDDATA; // unwind recursion
754 }
755
756 curr_cell = *ref_cell; // clone parent cell
757 if (code == H_SPLIT) {
758 SPLIT_CELL(ref_cell->height, curr_cell.height);
759 ref_cell->ypos += curr_cell.height;
760 ref_cell->height -= curr_cell.height;
761 if (ref_cell->height <= 0 || curr_cell.height <= 0)
762 return AVERROR_INVALIDDATA;
763 } else if (code == V_SPLIT) {
764 if (curr_cell.width > strip_width) {
765 /* split strip */
766 curr_cell.width = (curr_cell.width <= (strip_width << 1) ? 1 : 2) * strip_width;
767 } else
768 SPLIT_CELL(ref_cell->width, curr_cell.width);
769 ref_cell->xpos += curr_cell.width;
770 ref_cell->width -= curr_cell.width;
771 if (ref_cell->width <= 0 || curr_cell.width <= 0)
772 return AVERROR_INVALIDDATA;
773 }
774
775 while (1) { /* loop until return */
776 RESYNC_BITSTREAM;
777 switch (code = bitstream_read(&ctx->bc, 2)) {
778 case H_SPLIT:
779 case V_SPLIT:
780 if (parse_bintree(ctx, avctx, plane, code, &curr_cell, depth - 1, strip_width))
781 return AVERROR_INVALIDDATA;
782 break;
783 case INTRA_NULL:
784 if (!curr_cell.tree) { /* MC tree INTRA code */
785 curr_cell.mv_ptr = 0; /* mark the current strip as INTRA */
786 curr_cell.tree = 1; /* enter the VQ tree */
787 } else { /* VQ tree NULL code */
788 RESYNC_BITSTREAM;
789 code = bitstream_read(&ctx->bc, 2);
790 if (code >= 2) {
791 av_log(avctx, AV_LOG_ERROR, "Invalid VQ_NULL code: %d\n", code);
792 return AVERROR_INVALIDDATA;
793 }
794 if (code == 1)
795 av_log(avctx, AV_LOG_ERROR, "SkipCell procedure not implemented yet!\n");
796
797 CHECK_CELL
798 if (!curr_cell.mv_ptr)
799 return AVERROR_INVALIDDATA;
800 ret = copy_cell(ctx, plane, &curr_cell);
801 return ret;
802 }
803 break;
804 case INTER_DATA:
805 if (!curr_cell.tree) { /* MC tree INTER code */
806 unsigned mv_idx;
807 /* get motion vector index and setup the pointer to the mv set */
808 if (!ctx->need_resync)
809 ctx->next_cell_data = &ctx->bc.buffer[(bitstream_tell(&ctx->bc) + 7) >> 3];
810 mv_idx = *(ctx->next_cell_data++);
811 if (mv_idx >= ctx->num_vectors) {
812 av_log(avctx, AV_LOG_ERROR, "motion vector index out of range\n");
813 return AVERROR_INVALIDDATA;
814 }
815 curr_cell.mv_ptr = &ctx->mc_vectors[mv_idx << 1];
816 curr_cell.tree = 1; /* enter the VQ tree */
817 UPDATE_BITPOS(8);
818 } else { /* VQ tree DATA code */
819 if (!ctx->need_resync)
820 ctx->next_cell_data = &ctx->bc.buffer[(bitstream_tell(&ctx->bc) + 7) >> 3];
821
822 CHECK_CELL
823 bytes_used = decode_cell(ctx, avctx, plane, &curr_cell,
824 ctx->next_cell_data, ctx->last_byte);
825 if (bytes_used < 0)
826 return AVERROR_INVALIDDATA;
827
828 UPDATE_BITPOS(bytes_used << 3);
829 ctx->next_cell_data += bytes_used;
830 return 0;
831 }
832 break;
833 }
834 }//while
835 }
836
837
838 static int decode_plane(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
839 Plane *plane, const uint8_t *data, int32_t data_size,
840 int32_t strip_width)
841 {
842 Cell curr_cell;
843 unsigned num_vectors;
844
845 /* each plane data starts with mc_vector_count field, */
846 /* an optional array of motion vectors followed by the vq data */
847 num_vectors = bytestream_get_le32(&data);
848 if (num_vectors > 256) {
849 av_log(ctx->avctx, AV_LOG_ERROR,
850 "Read invalid number of motion vectors %d\n", num_vectors);
851 return AVERROR_INVALIDDATA;
852 }
853 if (num_vectors * 2 >= data_size)
854 return AVERROR_INVALIDDATA;
855
856 ctx->num_vectors = num_vectors;
857 ctx->mc_vectors = num_vectors ? data : 0;
858
859 /* init the bitreader */
860 bitstream_init(&ctx->bc, &data[num_vectors * 2], (data_size - num_vectors * 2) << 3);
861 ctx->skip_bits = 0;
862 ctx->need_resync = 0;
863
864 ctx->last_byte = data + data_size - 1;
865
866 /* initialize the 1st cell and set its dimensions to whole plane */
867 curr_cell.xpos = curr_cell.ypos = 0;
868 curr_cell.width = plane->width >> 2;
869 curr_cell.height = plane->height >> 2;
870 curr_cell.tree = 0; // we are in the MC tree now
871 curr_cell.mv_ptr = 0; // no motion vector = INTRA cell
872
873 return parse_bintree(ctx, avctx, plane, INTRA_NULL, &curr_cell, CELL_STACK_MAX, strip_width);
874 }
875
876
877 #define OS_HDR_ID MKBETAG('F', 'R', 'M', 'H')
878
879 static int decode_frame_headers(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
880 const uint8_t *buf, int buf_size)
881 {
882 GetByteContext gb;
883 const uint8_t *bs_hdr;
884 uint32_t frame_num, word2, check_sum, data_size;
885 uint32_t y_offset, u_offset, v_offset, starts[3], ends[3];
886 uint16_t height, width;
887 int i, j;
888
889 bytestream2_init(&gb, buf, buf_size);
890
891 /* parse and check the OS header */
892 frame_num = bytestream2_get_le32(&gb);
893 word2 = bytestream2_get_le32(&gb);
894 check_sum = bytestream2_get_le32(&gb);
895 data_size = bytestream2_get_le32(&gb);
896
897 if ((frame_num ^ word2 ^ data_size ^ OS_HDR_ID) != check_sum) {
898 av_log(avctx, AV_LOG_ERROR, "OS header checksum mismatch!\n");
899 return AVERROR_INVALIDDATA;
900 }
901
902 /* parse the bitstream header */
903 bs_hdr = gb.buffer;
904
905 if (bytestream2_get_le16(&gb) != 32) {
906 av_log(avctx, AV_LOG_ERROR, "Unsupported codec version!\n");
907 return AVERROR_INVALIDDATA;
908 }
909
910 ctx->frame_num = frame_num;
911 ctx->frame_flags = bytestream2_get_le16(&gb);
912 ctx->data_size = (bytestream2_get_le32(&gb) + 7) >> 3;
913 ctx->cb_offset = bytestream2_get_byte(&gb);
914
915 if (ctx->data_size == 16)
916 return 4;
917 ctx->data_size = FFMIN(ctx->data_size, buf_size - 16);
918
919 bytestream2_skip(&gb, 3); // skip reserved byte and checksum
920
921 /* check frame dimensions */
922 height = bytestream2_get_le16(&gb);
923 width = bytestream2_get_le16(&gb);
924 if (av_image_check_size(width, height, 0, avctx))
925 return AVERROR_INVALIDDATA;
926
927 if (width != ctx->width || height != ctx->height) {
928 int res;
929
930 ff_dlog(avctx, "Frame dimensions changed!\n");
931
932 if (width < 16 || width > 640 ||
933 height < 16 || height > 480 ||
934 width & 3 || height & 3) {
935 av_log(avctx, AV_LOG_ERROR,
936 "Invalid picture dimensions: %d x %d!\n", width, height);
937 return AVERROR_INVALIDDATA;
938 }
939
940 ctx->width = width;
941 ctx->height = height;
942
943 free_frame_buffers(ctx);
944 if ((res = allocate_frame_buffers(ctx, avctx)) < 0)
945 return res;
946 if ((res = ff_set_dimensions(avctx, width, height)) < 0)
947 return res;
948 }
949
950 y_offset = bytestream2_get_le32(&gb);
951 v_offset = bytestream2_get_le32(&gb);
952 u_offset = bytestream2_get_le32(&gb);
953 bytestream2_skip(&gb, 4);
954
955 /* unfortunately there is no common order of planes in the buffer */
956 /* so we use that sorting algo for determining planes data sizes */
957 starts[0] = y_offset;
958 starts[1] = v_offset;
959 starts[2] = u_offset;
960
961 for (j = 0; j < 3; j++) {
962 ends[j] = ctx->data_size;
963 for (i = 2; i >= 0; i--)
964 if (starts[i] < ends[j] && starts[i] > starts[j])
965 ends[j] = starts[i];
966 }
967
968 ctx->y_data_size = ends[0] - starts[0];
969 ctx->v_data_size = ends[1] - starts[1];
970 ctx->u_data_size = ends[2] - starts[2];
971 if (FFMAX3(y_offset, v_offset, u_offset) >= ctx->data_size - 16 ||
972 FFMIN3(y_offset, v_offset, u_offset) < gb.buffer - bs_hdr + 16 ||
973 FFMIN3(ctx->y_data_size, ctx->v_data_size, ctx->u_data_size) <= 0) {
974 av_log(avctx, AV_LOG_ERROR, "One of the y/u/v offsets is invalid\n");
975 return AVERROR_INVALIDDATA;
976 }
977
978 ctx->y_data_ptr = bs_hdr + y_offset;
979 ctx->v_data_ptr = bs_hdr + v_offset;
980 ctx->u_data_ptr = bs_hdr + u_offset;
981 ctx->alt_quant = gb.buffer;
982
983 if (ctx->data_size == 16) {
984 av_log(avctx, AV_LOG_DEBUG, "Sync frame encountered!\n");
985 return 16;
986 }
987
988 if (ctx->frame_flags & BS_8BIT_PEL) {
989 avpriv_request_sample(avctx, "8-bit pixel format");
990 return AVERROR_PATCHWELCOME;
991 }
992
993 if (ctx->frame_flags & BS_MV_X_HALF || ctx->frame_flags & BS_MV_Y_HALF) {
994 avpriv_request_sample(avctx, "Halfpel motion vectors");
995 return AVERROR_PATCHWELCOME;
996 }
997
998 return 0;
999 }
1000
1001
1002 /**
1003 * Convert and output the current plane.
1004 * All pixel values will be upsampled by shifting right by one bit.
1005 *
1006 * @param[in] plane pointer to the descriptor of the plane being processed
1007 * @param[in] buf_sel indicates which frame buffer the input data stored in
1008 * @param[out] dst pointer to the buffer receiving converted pixels
1009 * @param[in] dst_pitch pitch for moving to the next y line
1010 * @param[in] dst_height output plane height
1011 */
1012 static void output_plane(const Plane *plane, int buf_sel, uint8_t *dst,
1013 ptrdiff_t dst_pitch, int dst_height)
1014 {
1015 int x,y;
1016 const uint8_t *src = plane->pixels[buf_sel];
1017 ptrdiff_t pitch = plane->pitch;
1018
1019 dst_height = FFMIN(dst_height, plane->height);
1020 for (y = 0; y < dst_height; y++) {
1021 /* convert four pixels at once using SWAR */
1022 for (x = 0; x < plane->width >> 2; x++) {
1023 AV_WN32A(dst, (AV_RN32A(src) & 0x7F7F7F7F) << 1);
1024 src += 4;
1025 dst += 4;
1026 }
1027
1028 for (x <<= 2; x < plane->width; x++)
1029 *dst++ = *src++ << 1;
1030
1031 src += pitch - plane->width;
1032 dst += dst_pitch - plane->width;
1033 }
1034 }
1035
1036
1037 static av_cold int decode_init(AVCodecContext *avctx)
1038 {
1039 Indeo3DecodeContext *ctx = avctx->priv_data;
1040
1041 ctx->avctx = avctx;
1042 ctx->width = avctx->width;
1043 ctx->height = avctx->height;
1044 avctx->pix_fmt = AV_PIX_FMT_YUV410P;
1045
1046 build_requant_tab();
1047
1048 ff_hpeldsp_init(&ctx->hdsp, avctx->flags);
1049
1050 allocate_frame_buffers(ctx, avctx);
1051
1052 return 0;
1053 }
1054
1055
1056 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
1057 AVPacket *avpkt)
1058 {
1059 Indeo3DecodeContext *ctx = avctx->priv_data;
1060 const uint8_t *buf = avpkt->data;
1061 int buf_size = avpkt->size;
1062 AVFrame *frame = data;
1063 int res;
1064
1065 res = decode_frame_headers(ctx, avctx, buf, buf_size);
1066 if (res < 0)
1067 return res;
1068
1069 /* skip sync(null) frames */
1070 if (res) {
1071 // we have processed 16 bytes but no data was decoded
1072 *got_frame = 0;
1073 return buf_size;
1074 }
1075
1076 /* skip droppable INTER frames if requested */
1077 if (ctx->frame_flags & BS_NONREF &&
1078 (avctx->skip_frame >= AVDISCARD_NONREF))
1079 return 0;
1080
1081 /* skip INTER frames if requested */
1082 if (!(ctx->frame_flags & BS_KEYFRAME) && avctx->skip_frame >= AVDISCARD_NONKEY)
1083 return 0;
1084
1085 /* use BS_BUFFER flag for buffer switching */
1086 ctx->buf_sel = (ctx->frame_flags >> BS_BUFFER) & 1;
1087
1088 /* decode luma plane */
1089 if ((res = decode_plane(ctx, avctx, ctx->planes, ctx->y_data_ptr, ctx->y_data_size, 40)))
1090 return res;
1091
1092 /* decode chroma planes */
1093 if ((res = decode_plane(ctx, avctx, &ctx->planes[1], ctx->u_data_ptr, ctx->u_data_size, 10)))
1094 return res;
1095
1096 if ((res = decode_plane(ctx, avctx, &ctx->planes[2], ctx->v_data_ptr, ctx->v_data_size, 10)))
1097 return res;
1098
1099 if ((res = ff_get_buffer(avctx, frame, 0)) < 0) {
1100 av_log(ctx->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1101 return res;
1102 }
1103
1104 output_plane(&ctx->planes[0], ctx->buf_sel,
1105 frame->data[0], frame->linesize[0],
1106 avctx->height);
1107 output_plane(&ctx->planes[1], ctx->buf_sel,
1108 frame->data[1], frame->linesize[1],
1109 (avctx->height + 3) >> 2);
1110 output_plane(&ctx->planes[2], ctx->buf_sel,
1111 frame->data[2], frame->linesize[2],
1112 (avctx->height + 3) >> 2);
1113
1114 *got_frame = 1;
1115
1116 return buf_size;
1117 }
1118
1119
1120 static av_cold int decode_close(AVCodecContext *avctx)
1121 {
1122 free_frame_buffers(avctx->priv_data);
1123
1124 return 0;
1125 }
1126
1127 AVCodec ff_indeo3_decoder = {
1128 .name = "indeo3",
1129 .long_name = NULL_IF_CONFIG_SMALL("Intel Indeo 3"),
1130 .type = AVMEDIA_TYPE_VIDEO,
1131 .id = AV_CODEC_ID_INDEO3,
1132 .priv_data_size = sizeof(Indeo3DecodeContext),
1133 .init = decode_init,
1134 .close = decode_close,
1135 .decode = decode_frame,
1136 .capabilities = AV_CODEC_CAP_DR1,
1137 };