?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
lavc decoders: work with refcounted frames.
[FFMpeg-mirror/mplayer-patches.git] / libavcodec / indeo3.c
blobe519361aa24dab68c78b1e77ca3d34b139a744d1
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 #include "avcodec.h"
35 #include "dsputil.h"
36 #include "bytestream.h"
37 #include "get_bits.h"
38 #include "internal.h"
39
40 #include "indeo3data.h"
41
42 /* RLE opcodes. */
43 enum {
44 RLE_ESC_F9 = 249, ///< same as RLE_ESC_FA + do the same with next block
45 RLE_ESC_FA = 250, ///< INTRA: skip block, INTER: copy data from reference
46 RLE_ESC_FB = 251, ///< apply null delta to N blocks / skip N blocks
47 RLE_ESC_FC = 252, ///< same as RLE_ESC_FD + do the same with next block
48 RLE_ESC_FD = 253, ///< apply null delta to all remaining lines of this block
49 RLE_ESC_FE = 254, ///< apply null delta to all lines up to the 3rd line
50 RLE_ESC_FF = 255 ///< apply null delta to all lines up to the 2nd line
51 };
52
53
54 /* Some constants for parsing frame bitstream flags. */
55 #define BS_8BIT_PEL (1 << 1) ///< 8bit pixel bitdepth indicator
56 #define BS_KEYFRAME (1 << 2) ///< intra frame indicator
57 #define BS_MV_Y_HALF (1 << 4) ///< vertical mv halfpel resolution indicator
58 #define BS_MV_X_HALF (1 << 5) ///< horizontal mv halfpel resolution indicator
59 #define BS_NONREF (1 << 8) ///< nonref (discardable) frame indicator
60 #define BS_BUFFER 9 ///< indicates which of two frame buffers should be used
61
62
63 typedef struct Plane {
64 uint8_t *buffers[2];
65 uint8_t *pixels[2]; ///< pointer to the actual pixel data of the buffers above
66 uint32_t width;
67 uint32_t height;
68 uint32_t pitch;
69 } Plane;
70
71 #define CELL_STACK_MAX 20
72
73 typedef struct Cell {
74 int16_t xpos; ///< cell coordinates in 4x4 blocks
75 int16_t ypos;
76 int16_t width; ///< cell width in 4x4 blocks
77 int16_t height; ///< cell height in 4x4 blocks
78 uint8_t tree; ///< tree id: 0- MC tree, 1 - VQ tree
79 const int8_t *mv_ptr; ///< ptr to the motion vector if any
80 } Cell;
81
82 typedef struct Indeo3DecodeContext {
83 AVCodecContext *avctx;
84 DSPContext dsp;
85
86 GetBitContext gb;
87 int need_resync;
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; ///< number of motion vectors in mc_vectors
93
94 int16_t width, height;
95 uint32_t frame_num; ///< current frame number (zero-based)
96 uint32_t data_size; ///< size of the frame data in bytes
97 uint16_t frame_flags; ///< frame properties
98 uint8_t cb_offset; ///< needed for selecting VQ tables
99 uint8_t buf_sel; ///< active frame buffer: 0 - primary, 1 -secondary
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; ///< secondary VQ table set for the modes 1 and 4
107 Plane planes[3];
108 } Indeo3DecodeContext;
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
150 static av_cold int allocate_frame_buffers(Indeo3DecodeContext *ctx,
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
205 static av_cold void free_frame_buffers(Indeo3DecodeContext *ctx)
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
217 /**
218 * Copy pixels of the cell(x + mv_x, y + mv_y) from the previous frame into
219 * the cell(x, y) in the current frame.
220 *
221 * @param ctx pointer to the decoder context
222 * @param plane pointer to the plane descriptor
223 * @param cell pointer to the cell descriptor
224 */
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 ctx->dsp.put_no_rnd_pixels_tab[2][0](dst, src, 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 {
311 IV3_NOERR = 0,
312 IV3_BAD_RLE = 1,
313 IV3_BAD_DATA = 2,
314 IV3_BAD_COUNTER = 3,
315 IV3_UNSUPPORTED = 4,
316 IV3_OUT_OF_DATA = 5
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 ctx->dsp.put_pixels_tab[2][0](dst, ref, 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 ctx->dsp.put_pixels_tab[2][0](dst, ref, 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(Indeo3DecodeContext *ctx, Cell *cell,
407 uint8_t *block, uint8_t *ref_block,
408 int pitch, int h_zoom, int v_zoom, int mode,
409 const vqEntry *delta[2], int swap_quads[2],
410 const uint8_t **data_ptr, const uint8_t *last_ptr)
411 {
412 int x, y, line, num_lines;
413 int rle_blocks = 0;
414 uint8_t code, *dst, *ref;
415 const vqEntry *delta_tab;
416 unsigned int dyad1, dyad2;
417 uint64_t pix64;
418 int skip_flag = 0, is_top_of_cell, is_first_row = 1;
419 int row_offset, blk_row_offset, line_offset;
420
421 row_offset = pitch;
422 blk_row_offset = (row_offset << (2 + v_zoom)) - (cell->width << 2);
423 line_offset = v_zoom ? row_offset : 0;
424
425 if (cell->height & v_zoom || cell->width & h_zoom)
426 return IV3_BAD_DATA;
427
428 for (y = 0; y < cell->height; is_first_row = 0, y += 1 + v_zoom) {
429 for (x = 0; x < cell->width; x += 1 + h_zoom) {
430 ref = ref_block;
431 dst = block;
432
433 if (rle_blocks > 0) {
434 if (mode <= 4) {
435 RLE_BLOCK_COPY;
436 } else if (mode == 10 && !cell->mv_ptr) {
437 RLE_BLOCK_COPY_8;
438 }
439 rle_blocks--;
440 } else {
441 for (line = 0; line < 4;) {
442 num_lines = 1;
443 is_top_of_cell = is_first_row && !line;
444
445 /* select primary VQ table for odd, secondary for even lines */
446 if (mode <= 4)
447 delta_tab = delta[line & 1];
448 else
449 delta_tab = delta[1];
450 BUFFER_PRECHECK;
451 code = bytestream_get_byte(data_ptr);
452 if (code < 248) {
453 if (code < delta_tab->num_dyads) {
454 BUFFER_PRECHECK;
455 dyad1 = bytestream_get_byte(data_ptr);
456 dyad2 = code;
457 if (dyad1 >= delta_tab->num_dyads || dyad1 >= 248)
458 return IV3_BAD_DATA;
459 } else {
460 /* process QUADS */
461 code -= delta_tab->num_dyads;
462 dyad1 = code / delta_tab->quad_exp;
463 dyad2 = code % delta_tab->quad_exp;
464 if (swap_quads[line & 1])
465 FFSWAP(unsigned int, dyad1, dyad2);
466 }
467 if (mode <= 4) {
468 APPLY_DELTA_4;
469 } else if (mode == 10 && !cell->mv_ptr) {
470 APPLY_DELTA_8;
471 } else {
472 APPLY_DELTA_1011_INTER;
473 }
474 } else {
475 /* process RLE codes */
476 switch (code) {
477 case RLE_ESC_FC:
478 skip_flag = 0;
479 rle_blocks = 1;
480 code = 253;
481 /* FALLTHROUGH */
482 case RLE_ESC_FF:
483 case RLE_ESC_FE:
484 case RLE_ESC_FD:
485 num_lines = 257 - code - line;
486 if (num_lines <= 0)
487 return IV3_BAD_RLE;
488 if (mode <= 4) {
489 RLE_LINES_COPY;
490 } else if (mode == 10 && !cell->mv_ptr) {
491 RLE_LINES_COPY_M10;
492 }
493 break;
494 case RLE_ESC_FB:
495 BUFFER_PRECHECK;
496 code = bytestream_get_byte(data_ptr);
497 rle_blocks = (code & 0x1F) - 1; /* set block counter */
498 if (code >= 64 || rle_blocks < 0)
499 return IV3_BAD_COUNTER;
500 skip_flag = code & 0x20;
501 num_lines = 4 - line; /* enforce next block processing */
502 if (mode >= 10 || (cell->mv_ptr || !skip_flag)) {
503 if (mode <= 4) {
504 RLE_LINES_COPY;
505 } else if (mode == 10 && !cell->mv_ptr) {
506 RLE_LINES_COPY_M10;
507 }
508 }
509 break;
510 case RLE_ESC_F9:
511 skip_flag = 1;
512 rle_blocks = 1;
513 /* FALLTHROUGH */
514 case RLE_ESC_FA:
515 if (line)
516 return IV3_BAD_RLE;
517 num_lines = 4; /* enforce next block processing */
518 if (cell->mv_ptr) {
519 if (mode <= 4) {
520 RLE_LINES_COPY;
521 } else if (mode == 10 && !cell->mv_ptr) {
522 RLE_LINES_COPY_M10;
523 }
524 }
525 break;
526 default:
527 return IV3_UNSUPPORTED;
528 }
529 }
530
531 line += num_lines;
532 ref += row_offset * (num_lines << v_zoom);
533 dst += row_offset * (num_lines << v_zoom);
534 }
535 }
536
537 /* move to next horizontal block */
538 block += 4 << h_zoom;
539 ref_block += 4 << h_zoom;
540 }
541
542 /* move to next line of blocks */
543 ref_block += blk_row_offset;
544 block += blk_row_offset;
545 }
546 return IV3_NOERR;
547 }
548
549
550 /**
551 * Decode a vector-quantized cell.
552 * It consists of several routines, each of which handles one or more "modes"
553 * with which a cell can be encoded.
554 *
555 * @param ctx pointer to the decoder context
556 * @param avctx ptr to the AVCodecContext
557 * @param plane pointer to the plane descriptor
558 * @param cell pointer to the cell descriptor
559 * @param data_ptr pointer to the compressed data
560 * @param last_ptr pointer to the last byte to catch reads past end of buffer
561 * @return number of consumed bytes or negative number in case of error
562 */
563 static int decode_cell(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
564 Plane *plane, Cell *cell, const uint8_t *data_ptr,
565 const uint8_t *last_ptr)
566 {
567 int x, mv_x, mv_y, mode, vq_index, prim_indx, second_indx;
568 int zoom_fac;
569 int offset, error = 0, swap_quads[2];
570 uint8_t code, *block, *ref_block = 0;
571 const vqEntry *delta[2];
572 const uint8_t *data_start = data_ptr;
573
574 /* get coding mode and VQ table index from the VQ descriptor byte */
575 code = *data_ptr++;
576 mode = code >> 4;
577 vq_index = code & 0xF;
578
579 /* setup output and reference pointers */
580 offset = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
581 block = plane->pixels[ctx->buf_sel] + offset;
582 if (!cell->mv_ptr) {
583 /* use previous line as reference for INTRA cells */
584 ref_block = block - plane->pitch;
585 } else if (mode >= 10) {
586 /* for mode 10 and 11 INTER first copy the predicted cell into the current one */
587 /* so we don't need to do data copying for each RLE code later */
588 copy_cell(ctx, plane, cell);
589 } else {
590 /* set the pointer to the reference pixels for modes 0-4 INTER */
591 mv_y = cell->mv_ptr[0];
592 mv_x = cell->mv_ptr[1];
593 offset += mv_y * plane->pitch + mv_x;
594 ref_block = plane->pixels[ctx->buf_sel ^ 1] + offset;
595 }
596
597 /* select VQ tables as follows: */
598 /* modes 0 and 3 use only the primary table for all lines in a block */
599 /* while modes 1 and 4 switch between primary and secondary tables on alternate lines */
600 if (mode == 1 || mode == 4) {
601 code = ctx->alt_quant[vq_index];
602 prim_indx = (code >> 4) + ctx->cb_offset;
603 second_indx = (code & 0xF) + ctx->cb_offset;
604 } else {
605 vq_index += ctx->cb_offset;
606 prim_indx = second_indx = vq_index;
607 }
608
609 if (prim_indx >= 24 || second_indx >= 24) {
610 av_log(avctx, AV_LOG_ERROR, "Invalid VQ table indexes! Primary: %d, secondary: %d!\n",
611 prim_indx, second_indx);
612 return AVERROR_INVALIDDATA;
613 }
614
615 delta[0] = &vq_tab[second_indx];
616 delta[1] = &vq_tab[prim_indx];
617 swap_quads[0] = second_indx >= 16;
618 swap_quads[1] = prim_indx >= 16;
619
620 /* requantize the prediction if VQ index of this cell differs from VQ index */
621 /* of the predicted cell in order to avoid overflows. */
622 if (vq_index >= 8 && ref_block) {
623 for (x = 0; x < cell->width << 2; x++)
624 ref_block[x] = requant_tab[vq_index & 7][ref_block[x]];
625 }
626
627 error = IV3_NOERR;
628
629 switch (mode) {
630 case 0: /*------------------ MODES 0 & 1 (4x4 block processing) --------------------*/
631 case 1:
632 case 3: /*------------------ MODES 3 & 4 (4x8 block processing) --------------------*/
633 case 4:
634 if (mode >= 3 && cell->mv_ptr) {
635 av_log(avctx, AV_LOG_ERROR, "Attempt to apply Mode 3/4 to an INTER cell!\n");
636 return AVERROR_INVALIDDATA;
637 }
638
639 zoom_fac = mode >= 3;
640 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
641 0, zoom_fac, mode, delta, swap_quads,
642 &data_ptr, last_ptr);
643 break;
644 case 10: /*-------------------- MODE 10 (8x8 block processing) ---------------------*/
645 case 11: /*----------------- MODE 11 (4x8 INTER block processing) ------------------*/
646 if (mode == 10 && !cell->mv_ptr) { /* MODE 10 INTRA processing */
647 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
648 1, 1, mode, delta, swap_quads,
649 &data_ptr, last_ptr);
650 } else { /* mode 10 and 11 INTER processing */
651 if (mode == 11 && !cell->mv_ptr) {
652 av_log(avctx, AV_LOG_ERROR, "Attempt to use Mode 11 for an INTRA cell!\n");
653 return AVERROR_INVALIDDATA;
654 }
655
656 zoom_fac = mode == 10;
657 error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
658 zoom_fac, 1, mode, delta, swap_quads,
659 &data_ptr, last_ptr);
660 }
661 break;
662 default:
663 av_log(avctx, AV_LOG_ERROR, "Unsupported coding mode: %d\n", mode);
664 return AVERROR_INVALIDDATA;
665 }//switch mode
666
667 switch (error) {
668 case IV3_BAD_RLE:
669 av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE code %X is not allowed at the current line\n",
670 mode, data_ptr[-1]);
671 return AVERROR_INVALIDDATA;
672 case IV3_BAD_DATA:
673 av_log(avctx, AV_LOG_ERROR, "Mode %d: invalid VQ data\n", mode);
674 return AVERROR_INVALIDDATA;
675 case IV3_BAD_COUNTER:
676 av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE-FB invalid counter: %d\n", mode, code);
677 return AVERROR_INVALIDDATA;
678 case IV3_UNSUPPORTED:
679 av_log(avctx, AV_LOG_ERROR, "Mode %d: unsupported RLE code: %X\n", mode, data_ptr[-1]);
680 return AVERROR_INVALIDDATA;
681 case IV3_OUT_OF_DATA:
682 av_log(avctx, AV_LOG_ERROR, "Mode %d: attempt to read past end of buffer\n", mode);
683 return AVERROR_INVALIDDATA;
684 }
685
686 return data_ptr - data_start; /* report number of bytes consumed from the input buffer */
687 }
688
689
690 /* Binary tree codes. */
691 enum {
692 H_SPLIT = 0,
693 V_SPLIT = 1,
694 INTRA_NULL = 2,
695 INTER_DATA = 3
696 };
697
698
699 #define SPLIT_CELL(size, new_size) (new_size) = ((size) > 2) ? ((((size) + 2) >> 2) << 1) : 1
700
701 #define UPDATE_BITPOS(n) \
702 ctx->skip_bits += (n); \
703 ctx->need_resync = 1
704
705 #define RESYNC_BITSTREAM \
706 if (ctx->need_resync && !(get_bits_count(&ctx->gb) & 7)) { \
707 skip_bits_long(&ctx->gb, ctx->skip_bits); \
708 ctx->skip_bits = 0; \
709 ctx->need_resync = 0; \
710 }
711
712 #define CHECK_CELL \
713 if (curr_cell.xpos + curr_cell.width > (plane->width >> 2) || \
714 curr_cell.ypos + curr_cell.height > (plane->height >> 2)) { \
715 av_log(avctx, AV_LOG_ERROR, "Invalid cell: x=%d, y=%d, w=%d, h=%d\n", \
716 curr_cell.xpos, curr_cell.ypos, curr_cell.width, curr_cell.height); \
717 return AVERROR_INVALIDDATA; \
718 }
719
720
721 static int parse_bintree(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
722 Plane *plane, int code, Cell *ref_cell,
723 const int depth, const int strip_width)
724 {
725 Cell curr_cell;
726 int bytes_used;
727
728 if (depth <= 0) {
729 av_log(avctx, AV_LOG_ERROR, "Stack overflow (corrupted binary tree)!\n");
730 return AVERROR_INVALIDDATA; // unwind recursion
731 }
732
733 curr_cell = *ref_cell; // clone parent cell
734 if (code == H_SPLIT) {
735 SPLIT_CELL(ref_cell->height, curr_cell.height);
736 ref_cell->ypos += curr_cell.height;
737 ref_cell->height -= curr_cell.height;
738 if (ref_cell->height <= 0 || curr_cell.height <= 0)
739 return AVERROR_INVALIDDATA;
740 } else if (code == V_SPLIT) {
741 if (curr_cell.width > strip_width) {
742 /* split strip */
743 curr_cell.width = (curr_cell.width <= (strip_width << 1) ? 1 : 2) * strip_width;
744 } else
745 SPLIT_CELL(ref_cell->width, curr_cell.width);
746 ref_cell->xpos += curr_cell.width;
747 ref_cell->width -= curr_cell.width;
748 if (ref_cell->width <= 0 || curr_cell.width <= 0)
749 return AVERROR_INVALIDDATA;
750 }
751
752 while (1) { /* loop until return */
753 RESYNC_BITSTREAM;
754 switch (code = get_bits(&ctx->gb, 2)) {
755 case H_SPLIT:
756 case V_SPLIT:
757 if (parse_bintree(ctx, avctx, plane, code, &curr_cell, depth - 1, strip_width))
758 return AVERROR_INVALIDDATA;
759 break;
760 case INTRA_NULL:
761 if (!curr_cell.tree) { /* MC tree INTRA code */
762 curr_cell.mv_ptr = 0; /* mark the current strip as INTRA */
763 curr_cell.tree = 1; /* enter the VQ tree */
764 } else { /* VQ tree NULL code */
765 RESYNC_BITSTREAM;
766 code = get_bits(&ctx->gb, 2);
767 if (code >= 2) {
768 av_log(avctx, AV_LOG_ERROR, "Invalid VQ_NULL code: %d\n", code);
769 return AVERROR_INVALIDDATA;
770 }
771 if (code == 1)
772 av_log(avctx, AV_LOG_ERROR, "SkipCell procedure not implemented yet!\n");
773
774 CHECK_CELL
775 if (!curr_cell.mv_ptr)
776 return AVERROR_INVALIDDATA;
777 copy_cell(ctx, plane, &curr_cell);
778 return 0;
779 }
780 break;
781 case INTER_DATA:
782 if (!curr_cell.tree) { /* MC tree INTER code */
783 unsigned mv_idx;
784 /* get motion vector index and setup the pointer to the mv set */
785 if (!ctx->need_resync)
786 ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
787 mv_idx = *(ctx->next_cell_data++);
788 if (mv_idx >= ctx->num_vectors) {
789 av_log(avctx, AV_LOG_ERROR, "motion vector index out of range\n");
790 return AVERROR_INVALIDDATA;
791 }
792 curr_cell.mv_ptr = &ctx->mc_vectors[mv_idx << 1];
793 curr_cell.tree = 1; /* enter the VQ tree */
794 UPDATE_BITPOS(8);
795 } else { /* VQ tree DATA code */
796 if (!ctx->need_resync)
797 ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
798
799 CHECK_CELL
800 bytes_used = decode_cell(ctx, avctx, plane, &curr_cell,
801 ctx->next_cell_data, ctx->last_byte);
802 if (bytes_used < 0)
803 return AVERROR_INVALIDDATA;
804
805 UPDATE_BITPOS(bytes_used << 3);
806 ctx->next_cell_data += bytes_used;
807 return 0;
808 }
809 break;
810 }
811 }//while
812
813 return 0;
814 }
815
816
817 static int decode_plane(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
818 Plane *plane, const uint8_t *data, int32_t data_size,
819 int32_t strip_width)
820 {
821 Cell curr_cell;
822 unsigned num_vectors;
823
824 /* each plane data starts with mc_vector_count field, */
825 /* an optional array of motion vectors followed by the vq data */
826 num_vectors = bytestream_get_le32(&data);
827 if (num_vectors > 256) {
828 av_log(ctx->avctx, AV_LOG_ERROR,
829 "Read invalid number of motion vectors %d\n", num_vectors);
830 return AVERROR_INVALIDDATA;
831 }
832 if (num_vectors * 2 >= data_size)
833 return AVERROR_INVALIDDATA;
834
835 ctx->num_vectors = num_vectors;
836 ctx->mc_vectors = num_vectors ? data : 0;
837
838 /* init the bitreader */
839 init_get_bits(&ctx->gb, &data[num_vectors * 2], (data_size - num_vectors * 2) << 3);
840 ctx->skip_bits = 0;
841 ctx->need_resync = 0;
842
843 ctx->last_byte = data + data_size - 1;
844
845 /* initialize the 1st cell and set its dimensions to whole plane */
846 curr_cell.xpos = curr_cell.ypos = 0;
847 curr_cell.width = plane->width >> 2;
848 curr_cell.height = plane->height >> 2;
849 curr_cell.tree = 0; // we are in the MC tree now
850 curr_cell.mv_ptr = 0; // no motion vector = INTRA cell
851
852 return parse_bintree(ctx, avctx, plane, INTRA_NULL, &curr_cell, CELL_STACK_MAX, strip_width);
853 }
854
855
856 #define OS_HDR_ID MKBETAG('F', 'R', 'M', 'H')
857
858 static int decode_frame_headers(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
859 const uint8_t *buf, int buf_size)
860 {
861 const uint8_t *buf_ptr = buf, *bs_hdr;
862 uint32_t frame_num, word2, check_sum, data_size;
863 uint32_t y_offset, u_offset, v_offset, starts[3], ends[3];
864 uint16_t height, width;
865 int i, j;
866
867 /* parse and check the OS header */
868 frame_num = bytestream_get_le32(&buf_ptr);
869 word2 = bytestream_get_le32(&buf_ptr);
870 check_sum = bytestream_get_le32(&buf_ptr);
871 data_size = bytestream_get_le32(&buf_ptr);
872
873 if ((frame_num ^ word2 ^ data_size ^ OS_HDR_ID) != check_sum) {
874 av_log(avctx, AV_LOG_ERROR, "OS header checksum mismatch!\n");
875 return AVERROR_INVALIDDATA;
876 }
877
878 /* parse the bitstream header */
879 bs_hdr = buf_ptr;
880
881 if (bytestream_get_le16(&buf_ptr) != 32) {
882 av_log(avctx, AV_LOG_ERROR, "Unsupported codec version!\n");
883 return AVERROR_INVALIDDATA;
884 }
885
886 ctx->frame_num = frame_num;
887 ctx->frame_flags = bytestream_get_le16(&buf_ptr);
888 ctx->data_size = (bytestream_get_le32(&buf_ptr) + 7) >> 3;
889 ctx->cb_offset = *buf_ptr++;
890
891 if (ctx->data_size == 16)
892 return 4;
893 if (ctx->data_size > buf_size)
894 ctx->data_size = buf_size;
895
896 buf_ptr += 3; // skip reserved byte and checksum
897
898 /* check frame dimensions */
899 height = bytestream_get_le16(&buf_ptr);
900 width = bytestream_get_le16(&buf_ptr);
901 if (av_image_check_size(width, height, 0, avctx))
902 return AVERROR_INVALIDDATA;
903
904 if (width != ctx->width || height != ctx->height) {
905 int res;
906
907 av_dlog(avctx, "Frame dimensions changed!\n");
908
909 if (width < 16 || width > 640 ||
910 height < 16 || height > 480 ||
911 width & 3 || height & 3) {
912 av_log(avctx, AV_LOG_ERROR,
913 "Invalid picture dimensions: %d x %d!\n", width, height);
914 return AVERROR_INVALIDDATA;
915 }
916
917 ctx->width = width;
918 ctx->height = height;
919
920 free_frame_buffers(ctx);
921 if ((res = allocate_frame_buffers(ctx, avctx)) < 0)
922 return res;
923 avcodec_set_dimensions(avctx, width, height);
924 }
925
926 y_offset = bytestream_get_le32(&buf_ptr);
927 v_offset = bytestream_get_le32(&buf_ptr);
928 u_offset = bytestream_get_le32(&buf_ptr);
929
930 /* unfortunately there is no common order of planes in the buffer */
931 /* so we use that sorting algo for determining planes data sizes */
932 starts[0] = y_offset;
933 starts[1] = v_offset;
934 starts[2] = u_offset;
935
936 for (j = 0; j < 3; j++) {
937 ends[j] = ctx->data_size;
938 for (i = 2; i >= 0; i--)
939 if (starts[i] < ends[j] && starts[i] > starts[j])
940 ends[j] = starts[i];
941 }
942
943 ctx->y_data_size = ends[0] - starts[0];
944 ctx->v_data_size = ends[1] - starts[1];
945 ctx->u_data_size = ends[2] - starts[2];
946 if (FFMAX3(y_offset, v_offset, u_offset) >= ctx->data_size - 16 ||
947 FFMIN3(ctx->y_data_size, ctx->v_data_size, ctx->u_data_size) <= 0) {
948 av_log(avctx, AV_LOG_ERROR, "One of the y/u/v offsets is invalid\n");
949 return AVERROR_INVALIDDATA;
950 }
951
952 ctx->y_data_ptr = bs_hdr + y_offset;
953 ctx->v_data_ptr = bs_hdr + v_offset;
954 ctx->u_data_ptr = bs_hdr + u_offset;
955 ctx->alt_quant = buf_ptr + sizeof(uint32_t);
956
957 if (ctx->data_size == 16) {
958 av_log(avctx, AV_LOG_DEBUG, "Sync frame encountered!\n");
959 return 16;
960 }
961
962 if (ctx->frame_flags & BS_8BIT_PEL) {
963 av_log_ask_for_sample(avctx, "8-bit pixel format\n");
964 return AVERROR_PATCHWELCOME;
965 }
966
967 if (ctx->frame_flags & BS_MV_X_HALF || ctx->frame_flags & BS_MV_Y_HALF) {
968 av_log_ask_for_sample(avctx, "halfpel motion vectors\n");
969 return AVERROR_PATCHWELCOME;
970 }
971
972 return 0;
973 }
974
975
976 /**
977 * Convert and output the current plane.
978 * All pixel values will be upsampled by shifting right by one bit.
979 *
980 * @param[in] plane pointer to the descriptor of the plane being processed
981 * @param[in] buf_sel indicates which frame buffer the input data stored in
982 * @param[out] dst pointer to the buffer receiving converted pixels
983 * @param[in] dst_pitch pitch for moving to the next y line
984 * @param[in] dst_height output plane height
985 */
986 static void output_plane(const Plane *plane, int buf_sel, uint8_t *dst,
987 int dst_pitch, int dst_height)
988 {
989 int x,y;
990 const uint8_t *src = plane->pixels[buf_sel];
991 uint32_t pitch = plane->pitch;
992
993 dst_height = FFMIN(dst_height, plane->height);
994 for (y = 0; y < dst_height; y++) {
995 /* convert four pixels at once using SWAR */
996 for (x = 0; x < plane->width >> 2; x++) {
997 AV_WN32A(dst, (AV_RN32A(src) & 0x7F7F7F7F) << 1);
998 src += 4;
999 dst += 4;
1000 }
1001
1002 for (x <<= 2; x < plane->width; x++)
1003 *dst++ = *src++ << 1;
1004
1005 src += pitch - plane->width;
1006 dst += dst_pitch - plane->width;
1007 }
1008 }
1009
1010
1011 static av_cold int decode_init(AVCodecContext *avctx)
1012 {
1013 Indeo3DecodeContext *ctx = avctx->priv_data;
1014
1015 ctx->avctx = avctx;
1016 ctx->width = avctx->width;
1017 ctx->height = avctx->height;
1018 avctx->pix_fmt = AV_PIX_FMT_YUV410P;
1019
1020 build_requant_tab();
1021
1022 ff_dsputil_init(&ctx->dsp, avctx);
1023
1024 allocate_frame_buffers(ctx, avctx);
1025
1026 return 0;
1027 }
1028
1029
1030 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
1031 AVPacket *avpkt)
1032 {
1033 Indeo3DecodeContext *ctx = avctx->priv_data;
1034 const uint8_t *buf = avpkt->data;
1035 int buf_size = avpkt->size;
1036 AVFrame *frame = data;
1037 int res;
1038
1039 res = decode_frame_headers(ctx, avctx, buf, buf_size);
1040 if (res < 0)
1041 return res;
1042
1043 /* skip sync(null) frames */
1044 if (res) {
1045 // we have processed 16 bytes but no data was decoded
1046 *got_frame = 0;
1047 return buf_size;
1048 }
1049
1050 /* skip droppable INTER frames if requested */
1051 if (ctx->frame_flags & BS_NONREF &&
1052 (avctx->skip_frame >= AVDISCARD_NONREF))
1053 return 0;
1054
1055 /* skip INTER frames if requested */
1056 if (!(ctx->frame_flags & BS_KEYFRAME) && avctx->skip_frame >= AVDISCARD_NONKEY)
1057 return 0;
1058
1059 /* use BS_BUFFER flag for buffer switching */
1060 ctx->buf_sel = (ctx->frame_flags >> BS_BUFFER) & 1;
1061
1062 /* decode luma plane */
1063 if ((res = decode_plane(ctx, avctx, ctx->planes, ctx->y_data_ptr, ctx->y_data_size, 40)))
1064 return res;
1065
1066 /* decode chroma planes */
1067 if ((res = decode_plane(ctx, avctx, &ctx->planes[1], ctx->u_data_ptr, ctx->u_data_size, 10)))
1068 return res;
1069
1070 if ((res = decode_plane(ctx, avctx, &ctx->planes[2], ctx->v_data_ptr, ctx->v_data_size, 10)))
1071 return res;
1072
1073 if ((res = ff_get_buffer(avctx, frame, 0)) < 0) {
1074 av_log(ctx->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1075 return res;
1076 }
1077
1078 output_plane(&ctx->planes[0], ctx->buf_sel,
1079 frame->data[0], frame->linesize[0],
1080 avctx->height);
1081 output_plane(&ctx->planes[1], ctx->buf_sel,
1082 frame->data[1], frame->linesize[1],
1083 (avctx->height + 3) >> 2);
1084 output_plane(&ctx->planes[2], ctx->buf_sel,
1085 frame->data[2], frame->linesize[2],
1086 (avctx->height + 3) >> 2);
1087
1088 *got_frame = 1;
1089
1090 return buf_size;
1091 }
1092
1093
1094 static av_cold int decode_close(AVCodecContext *avctx)
1095 {
1096 free_frame_buffers(avctx->priv_data);
1097
1098 return 0;
1099 }
1100
1101 AVCodec ff_indeo3_decoder = {
1102 .name = "indeo3",
1103 .type = AVMEDIA_TYPE_VIDEO,
1104 .id = AV_CODEC_ID_INDEO3,
1105 .priv_data_size = sizeof(Indeo3DecodeContext),
1106 .init = decode_init,
1107 .close = decode_close,
1108 .decode = decode_frame,
1109 .capabilities = CODEC_CAP_DR1,
1110 .long_name = NULL_IF_CONFIG_SMALL("Intel Indeo 3"),
1111 };
A couple of patches to make building with MPlayer easier