da0ef08eeea3eb200af607801c4efdb04c978d0b
[libav.git] / libavcodec / shorten.c
1 /*
2 * Shorten decoder
3 * Copyright (c) 2005 Jeff Muizelaar
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 * Shorten decoder
25 * @author Jeff Muizelaar
26 *
27 */
28
29 #include <limits.h>
30 #include "avcodec.h"
31 #include "bytestream.h"
32 #include "get_bits.h"
33 #include "golomb.h"
34
35 #define MAX_CHANNELS 8
36 #define MAX_BLOCKSIZE 65535
37
38 #define OUT_BUFFER_SIZE 16384
39
40 #define ULONGSIZE 2
41
42 #define WAVE_FORMAT_PCM 0x0001
43
44 #define DEFAULT_BLOCK_SIZE 256
45
46 #define TYPESIZE 4
47 #define CHANSIZE 0
48 #define LPCQSIZE 2
49 #define ENERGYSIZE 3
50 #define BITSHIFTSIZE 2
51
52 #define TYPE_S16HL 3
53 #define TYPE_S16LH 5
54
55 #define NWRAP 3
56 #define NSKIPSIZE 1
57
58 #define LPCQUANT 5
59 #define V2LPCQOFFSET (1 << LPCQUANT)
60
61 #define FNSIZE 2
62 #define FN_DIFF0 0
63 #define FN_DIFF1 1
64 #define FN_DIFF2 2
65 #define FN_DIFF3 3
66 #define FN_QUIT 4
67 #define FN_BLOCKSIZE 5
68 #define FN_BITSHIFT 6
69 #define FN_QLPC 7
70 #define FN_ZERO 8
71 #define FN_VERBATIM 9
72
73 /** indicates if the FN_* command is audio or non-audio */
74 static const uint8_t is_audio_command[10] = { 1, 1, 1, 1, 0, 0, 0, 1, 1, 0 };
75
76 #define VERBATIM_CKSIZE_SIZE 5
77 #define VERBATIM_BYTE_SIZE 8
78 #define CANONICAL_HEADER_SIZE 44
79
80 typedef struct ShortenContext {
81 AVCodecContext *avctx;
82 AVFrame frame;
83 GetBitContext gb;
84
85 int min_framesize, max_framesize;
86 int channels;
87
88 int32_t *decoded[MAX_CHANNELS];
89 int32_t *offset[MAX_CHANNELS];
90 int *coeffs;
91 uint8_t *bitstream;
92 int bitstream_size;
93 int bitstream_index;
94 unsigned int allocated_bitstream_size;
95 int header_size;
96 uint8_t header[OUT_BUFFER_SIZE];
97 int version;
98 int cur_chan;
99 int bitshift;
100 int nmean;
101 int internal_ftype;
102 int nwrap;
103 int blocksize;
104 int bitindex;
105 int32_t lpcqoffset;
106 int got_header;
107 int got_quit_command;
108 } ShortenContext;
109
110 static av_cold int shorten_decode_init(AVCodecContext * avctx)
111 {
112 ShortenContext *s = avctx->priv_data;
113 s->avctx = avctx;
114 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
115
116 avcodec_get_frame_defaults(&s->frame);
117 avctx->coded_frame = &s->frame;
118
119 return 0;
120 }
121
122 static int allocate_buffers(ShortenContext *s)
123 {
124 int i, chan;
125 int *coeffs;
126 void *tmp_ptr;
127
128 for (chan=0; chan<s->channels; chan++) {
129 if(FFMAX(1, s->nmean) >= UINT_MAX/sizeof(int32_t)){
130 av_log(s->avctx, AV_LOG_ERROR, "nmean too large\n");
131 return -1;
132 }
133 if(s->blocksize + s->nwrap >= UINT_MAX/sizeof(int32_t) || s->blocksize + s->nwrap <= (unsigned)s->nwrap){
134 av_log(s->avctx, AV_LOG_ERROR, "s->blocksize + s->nwrap too large\n");
135 return -1;
136 }
137
138 tmp_ptr = av_realloc(s->offset[chan], sizeof(int32_t)*FFMAX(1, s->nmean));
139 if (!tmp_ptr)
140 return AVERROR(ENOMEM);
141 s->offset[chan] = tmp_ptr;
142
143 tmp_ptr = av_realloc(s->decoded[chan], sizeof(int32_t)*(s->blocksize + s->nwrap));
144 if (!tmp_ptr)
145 return AVERROR(ENOMEM);
146 s->decoded[chan] = tmp_ptr;
147 for (i=0; i<s->nwrap; i++)
148 s->decoded[chan][i] = 0;
149 s->decoded[chan] += s->nwrap;
150 }
151
152 coeffs = av_realloc(s->coeffs, s->nwrap * sizeof(*s->coeffs));
153 if (!coeffs)
154 return AVERROR(ENOMEM);
155 s->coeffs = coeffs;
156
157 return 0;
158 }
159
160
161 static inline unsigned int get_uint(ShortenContext *s, int k)
162 {
163 if (s->version != 0)
164 k = get_ur_golomb_shorten(&s->gb, ULONGSIZE);
165 return get_ur_golomb_shorten(&s->gb, k);
166 }
167
168
169 static void fix_bitshift(ShortenContext *s, int32_t *buffer)
170 {
171 int i;
172
173 if (s->bitshift != 0)
174 for (i = 0; i < s->blocksize; i++)
175 buffer[i] <<= s->bitshift;
176 }
177
178
179 static void init_offset(ShortenContext *s)
180 {
181 int32_t mean = 0;
182 int chan, i;
183 int nblock = FFMAX(1, s->nmean);
184 /* initialise offset */
185 switch (s->internal_ftype)
186 {
187 case TYPE_S16HL:
188 case TYPE_S16LH:
189 mean = 0;
190 break;
191 default:
192 av_log(s->avctx, AV_LOG_ERROR, "unknown audio type");
193 abort();
194 }
195
196 for (chan = 0; chan < s->channels; chan++)
197 for (i = 0; i < nblock; i++)
198 s->offset[chan][i] = mean;
199 }
200
201 static int decode_wave_header(AVCodecContext *avctx, const uint8_t *header,
202 int header_size)
203 {
204 int len;
205 short wave_format;
206
207
208 if (bytestream_get_le32(&header) != MKTAG('R','I','F','F')) {
209 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
210 return -1;
211 }
212
213 header += 4; /* chunk size */;
214
215 if (bytestream_get_le32(&header) != MKTAG('W','A','V','E')) {
216 av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
217 return -1;
218 }
219
220 while (bytestream_get_le32(&header) != MKTAG('f','m','t',' ')) {
221 len = bytestream_get_le32(&header);
222 header += len;
223 }
224 len = bytestream_get_le32(&header);
225
226 if (len < 16) {
227 av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
228 return -1;
229 }
230
231 wave_format = bytestream_get_le16(&header);
232
233 switch (wave_format) {
234 case WAVE_FORMAT_PCM:
235 break;
236 default:
237 av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
238 return -1;
239 }
240
241 header += 2; // skip channels (already got from shorten header)
242 avctx->sample_rate = bytestream_get_le32(&header);
243 header += 4; // skip bit rate (represents original uncompressed bit rate)
244 header += 2; // skip block align (not needed)
245 avctx->bits_per_coded_sample = bytestream_get_le16(&header);
246
247 if (avctx->bits_per_coded_sample != 16) {
248 av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample\n");
249 return -1;
250 }
251
252 len -= 16;
253 if (len > 0)
254 av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
255
256 return 0;
257 }
258
259 static void interleave_buffer(int16_t *samples, int nchan, int blocksize,
260 int32_t **buffer)
261 {
262 int i, chan;
263 for (i=0; i<blocksize; i++)
264 for (chan=0; chan < nchan; chan++)
265 *samples++ = av_clip_int16(buffer[chan][i]);
266 }
267
268 static const int fixed_coeffs[3][3] = {
269 { 1, 0, 0 },
270 { 2, -1, 0 },
271 { 3, -3, 1 }
272 };
273
274 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
275 int residual_size, int32_t coffset)
276 {
277 int pred_order, sum, qshift, init_sum, i, j;
278 const int *coeffs;
279
280 if (command == FN_QLPC) {
281 /* read/validate prediction order */
282 pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
283 if (pred_order > s->nwrap) {
284 av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n", pred_order);
285 return AVERROR(EINVAL);
286 }
287 /* read LPC coefficients */
288 for (i=0; i<pred_order; i++)
289 s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
290 coeffs = s->coeffs;
291
292 qshift = LPCQUANT;
293 } else {
294 /* fixed LPC coeffs */
295 pred_order = command;
296 coeffs = fixed_coeffs[pred_order-1];
297 qshift = 0;
298 }
299
300 /* subtract offset from previous samples to use in prediction */
301 if (command == FN_QLPC && coffset)
302 for (i = -pred_order; i < 0; i++)
303 s->decoded[channel][i] -= coffset;
304
305 /* decode residual and do LPC prediction */
306 init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
307 for (i=0; i < s->blocksize; i++) {
308 sum = init_sum;
309 for (j=0; j<pred_order; j++)
310 sum += coeffs[j] * s->decoded[channel][i-j-1];
311 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + (sum >> qshift);
312 }
313
314 /* add offset to current samples */
315 if (command == FN_QLPC && coffset)
316 for (i = 0; i < s->blocksize; i++)
317 s->decoded[channel][i] += coffset;
318
319 return 0;
320 }
321
322 static int read_header(ShortenContext *s)
323 {
324 int i, ret;
325 int maxnlpc = 0;
326 /* shorten signature */
327 if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
328 av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
329 return -1;
330 }
331
332 s->lpcqoffset = 0;
333 s->blocksize = DEFAULT_BLOCK_SIZE;
334 s->channels = 1;
335 s->nmean = -1;
336 s->version = get_bits(&s->gb, 8);
337 s->internal_ftype = get_uint(s, TYPESIZE);
338
339 s->channels = get_uint(s, CHANSIZE);
340 if (s->channels > MAX_CHANNELS) {
341 av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
342 return -1;
343 }
344 s->avctx->channels = s->channels;
345
346 /* get blocksize if version > 0 */
347 if (s->version > 0) {
348 int skip_bytes, blocksize;
349
350 blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
351 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
352 av_log(s->avctx, AV_LOG_ERROR, "invalid or unsupported block size: %d\n",
353 blocksize);
354 return AVERROR(EINVAL);
355 }
356 s->blocksize = blocksize;
357
358 maxnlpc = get_uint(s, LPCQSIZE);
359 s->nmean = get_uint(s, 0);
360
361 skip_bytes = get_uint(s, NSKIPSIZE);
362 for (i=0; i<skip_bytes; i++) {
363 skip_bits(&s->gb, 8);
364 }
365 }
366 s->nwrap = FFMAX(NWRAP, maxnlpc);
367
368 if ((ret = allocate_buffers(s)) < 0)
369 return ret;
370
371 init_offset(s);
372
373 if (s->version > 1)
374 s->lpcqoffset = V2LPCQOFFSET;
375
376 if (get_ur_golomb_shorten(&s->gb, FNSIZE) != FN_VERBATIM) {
377 av_log(s->avctx, AV_LOG_ERROR, "missing verbatim section at beginning of stream\n");
378 return -1;
379 }
380
381 s->header_size = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
382 if (s->header_size >= OUT_BUFFER_SIZE || s->header_size < CANONICAL_HEADER_SIZE) {
383 av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n", s->header_size);
384 return -1;
385 }
386
387 for (i=0; i<s->header_size; i++)
388 s->header[i] = (char)get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
389
390 if (decode_wave_header(s->avctx, s->header, s->header_size) < 0)
391 return -1;
392
393 s->cur_chan = 0;
394 s->bitshift = 0;
395
396 s->got_header = 1;
397
398 return 0;
399 }
400
401 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
402 int *got_frame_ptr, AVPacket *avpkt)
403 {
404 const uint8_t *buf = avpkt->data;
405 int buf_size = avpkt->size;
406 ShortenContext *s = avctx->priv_data;
407 int i, input_buf_size = 0;
408 int ret;
409
410 /* allocate internal bitstream buffer */
411 if(s->max_framesize == 0){
412 void *tmp_ptr;
413 s->max_framesize= 1024; // should hopefully be enough for the first header
414 tmp_ptr = av_fast_realloc(s->bitstream, &s->allocated_bitstream_size,
415 s->max_framesize);
416 if (!tmp_ptr) {
417 av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
418 return AVERROR(ENOMEM);
419 }
420 s->bitstream = tmp_ptr;
421 }
422
423 /* append current packet data to bitstream buffer */
424 if(1 && s->max_framesize){//FIXME truncated
425 buf_size= FFMIN(buf_size, s->max_framesize - s->bitstream_size);
426 input_buf_size= buf_size;
427
428 if(s->bitstream_index + s->bitstream_size + buf_size > s->allocated_bitstream_size){
429 memmove(s->bitstream, &s->bitstream[s->bitstream_index], s->bitstream_size);
430 s->bitstream_index=0;
431 }
432 if (buf)
433 memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf, buf_size);
434 buf= &s->bitstream[s->bitstream_index];
435 buf_size += s->bitstream_size;
436 s->bitstream_size= buf_size;
437
438 /* do not decode until buffer has at least max_framesize bytes or
439 the end of the file has been reached */
440 if (buf_size < s->max_framesize && avpkt->data) {
441 *got_frame_ptr = 0;
442 return input_buf_size;
443 }
444 }
445 /* init and position bitstream reader */
446 init_get_bits(&s->gb, buf, buf_size*8);
447 skip_bits(&s->gb, s->bitindex);
448
449 /* process header or next subblock */
450 if (!s->got_header) {
451 if ((ret = read_header(s)) < 0)
452 return ret;
453 *got_frame_ptr = 0;
454 goto finish_frame;
455 }
456
457 /* if quit command was read previously, don't decode anything */
458 if (s->got_quit_command) {
459 *got_frame_ptr = 0;
460 return avpkt->size;
461 }
462
463 s->cur_chan = 0;
464 while (s->cur_chan < s->channels) {
465 int cmd;
466 int len;
467
468 if (get_bits_left(&s->gb) < 3+FNSIZE) {
469 *got_frame_ptr = 0;
470 break;
471 }
472
473 cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
474
475 if (cmd > FN_VERBATIM) {
476 av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
477 *got_frame_ptr = 0;
478 break;
479 }
480
481 if (!is_audio_command[cmd]) {
482 /* process non-audio command */
483 switch (cmd) {
484 case FN_VERBATIM:
485 len = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
486 while (len--) {
487 get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
488 }
489 break;
490 case FN_BITSHIFT:
491 s->bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
492 break;
493 case FN_BLOCKSIZE: {
494 int blocksize = get_uint(s, av_log2(s->blocksize));
495 if (blocksize > s->blocksize) {
496 av_log(avctx, AV_LOG_ERROR, "Increasing block size is not supported\n");
497 return AVERROR_PATCHWELCOME;
498 }
499 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
500 av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
501 "block size: %d\n", blocksize);
502 return AVERROR(EINVAL);
503 }
504 s->blocksize = blocksize;
505 break;
506 }
507 case FN_QUIT:
508 s->got_quit_command = 1;
509 break;
510 }
511 if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
512 *got_frame_ptr = 0;
513 break;
514 }
515 } else {
516 /* process audio command */
517 int residual_size = 0;
518 int channel = s->cur_chan;
519 int32_t coffset;
520
521 /* get Rice code for residual decoding */
522 if (cmd != FN_ZERO) {
523 residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
524 /* this is a hack as version 0 differed in defintion of get_sr_golomb_shorten */
525 if (s->version == 0)
526 residual_size--;
527 }
528
529 /* calculate sample offset using means from previous blocks */
530 if (s->nmean == 0)
531 coffset = s->offset[channel][0];
532 else {
533 int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
534 for (i=0; i<s->nmean; i++)
535 sum += s->offset[channel][i];
536 coffset = sum / s->nmean;
537 if (s->version >= 2)
538 coffset >>= FFMIN(1, s->bitshift);
539 }
540
541 /* decode samples for this channel */
542 if (cmd == FN_ZERO) {
543 for (i=0; i<s->blocksize; i++)
544 s->decoded[channel][i] = 0;
545 } else {
546 if ((ret = decode_subframe_lpc(s, cmd, channel, residual_size, coffset)) < 0)
547 return ret;
548 }
549
550 /* update means with info from the current block */
551 if (s->nmean > 0) {
552 int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
553 for (i=0; i<s->blocksize; i++)
554 sum += s->decoded[channel][i];
555
556 for (i=1; i<s->nmean; i++)
557 s->offset[channel][i-1] = s->offset[channel][i];
558
559 if (s->version < 2)
560 s->offset[channel][s->nmean - 1] = sum / s->blocksize;
561 else
562 s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
563 }
564
565 /* copy wrap samples for use with next block */
566 for (i=-s->nwrap; i<0; i++)
567 s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
568
569 /* shift samples to add in unused zero bits which were removed
570 during encoding */
571 fix_bitshift(s, s->decoded[channel]);
572
573 /* if this is the last channel in the block, output the samples */
574 s->cur_chan++;
575 if (s->cur_chan == s->channels) {
576 /* get output buffer */
577 s->frame.nb_samples = s->blocksize;
578 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
579 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
580 return ret;
581 }
582 /* interleave output */
583 interleave_buffer((int16_t *)s->frame.data[0], s->channels,
584 s->blocksize, s->decoded);
585
586 *got_frame_ptr = 1;
587 *(AVFrame *)data = s->frame;
588 }
589 }
590 }
591 if (s->cur_chan < s->channels)
592 *got_frame_ptr = 0;
593
594 finish_frame:
595 s->bitindex = get_bits_count(&s->gb) - 8*((get_bits_count(&s->gb))/8);
596 i= (get_bits_count(&s->gb))/8;
597 if (i > buf_size) {
598 av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
599 s->bitstream_size=0;
600 s->bitstream_index=0;
601 return -1;
602 }
603 if (s->bitstream_size) {
604 s->bitstream_index += i;
605 s->bitstream_size -= i;
606 return input_buf_size;
607 } else
608 return i;
609 }
610
611 static av_cold int shorten_decode_close(AVCodecContext *avctx)
612 {
613 ShortenContext *s = avctx->priv_data;
614 int i;
615
616 for (i = 0; i < s->channels; i++) {
617 s->decoded[i] -= s->nwrap;
618 av_freep(&s->decoded[i]);
619 av_freep(&s->offset[i]);
620 }
621 av_freep(&s->bitstream);
622 av_freep(&s->coeffs);
623
624 return 0;
625 }
626
627 AVCodec ff_shorten_decoder = {
628 .name = "shorten",
629 .type = AVMEDIA_TYPE_AUDIO,
630 .id = CODEC_ID_SHORTEN,
631 .priv_data_size = sizeof(ShortenContext),
632 .init = shorten_decode_init,
633 .close = shorten_decode_close,
634 .decode = shorten_decode_frame,
635 .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
636 .long_name= NULL_IF_CONFIG_SMALL("Shorten"),
637 };