Test for supported number of channels
[libav.git] / libavcodec / alac.c
1 /*
2 * ALAC (Apple Lossless Audio Codec) decoder
3 * Copyright (c) 2005 David Hammerton
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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 alac.c
24 * ALAC (Apple Lossless Audio Codec) decoder
25 * @author 2005 David Hammerton
26 *
27 * For more information on the ALAC format, visit:
28 * http://crazney.net/programs/itunes/alac.html
29 *
30 * Note: This decoder expects a 36- (0x24-)byte QuickTime atom to be
31 * passed through the extradata[_size] fields. This atom is tacked onto
32 * the end of an 'alac' stsd atom and has the following format:
33 * bytes 0-3 atom size (0x24), big-endian
34 * bytes 4-7 atom type ('alac', not the 'alac' tag from start of stsd)
35 * bytes 8-35 data bytes needed by decoder
36 *
37 * Extradata:
38 * 32bit size
39 * 32bit tag (=alac)
40 * 32bit zero?
41 * 32bit max sample per frame
42 * 8bit ?? (zero?)
43 * 8bit sample size
44 * 8bit history mult
45 * 8bit initial history
46 * 8bit kmodifier
47 * 8bit channels?
48 * 16bit ??
49 * 32bit max coded frame size
50 * 32bit bitrate?
51 * 32bit samplerate
52 */
53
54
55 #include "avcodec.h"
56 #include "bitstream.h"
57 #include "bytestream.h"
58
59 #define ALAC_EXTRADATA_SIZE 36
60 #define MAX_CHANNELS 2
61
62 typedef struct {
63
64 AVCodecContext *avctx;
65 GetBitContext gb;
66 /* init to 0; first frame decode should initialize from extradata and
67 * set this to 1 */
68 int context_initialized;
69
70 int samplesize;
71 int numchannels;
72 int bytespersample;
73
74 /* buffers */
75 int32_t *predicterror_buffer[MAX_CHANNELS];
76
77 int32_t *outputsamples_buffer[MAX_CHANNELS];
78
79 /* stuff from setinfo */
80 uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */ /* max samples per frame? */
81 uint8_t setinfo_7a; /* 0x00 */
82 uint8_t setinfo_sample_size; /* 0x10 */
83 uint8_t setinfo_rice_historymult; /* 0x28 */
84 uint8_t setinfo_rice_initialhistory; /* 0x0a */
85 uint8_t setinfo_rice_kmodifier; /* 0x0e */
86 uint8_t setinfo_7f; /* 0x02 */
87 uint16_t setinfo_80; /* 0x00ff */
88 uint32_t setinfo_82; /* 0x000020e7 */ /* max sample size?? */
89 uint32_t setinfo_86; /* 0x00069fe4 */ /* bit rate (average)?? */
90 uint32_t setinfo_8a_rate; /* 0x0000ac44 */
91 /* end setinfo stuff */
92
93 } ALACContext;
94
95 static void allocate_buffers(ALACContext *alac)
96 {
97 int chan;
98 for (chan = 0; chan < MAX_CHANNELS; chan++) {
99 alac->predicterror_buffer[chan] =
100 av_malloc(alac->setinfo_max_samples_per_frame * 4);
101
102 alac->outputsamples_buffer[chan] =
103 av_malloc(alac->setinfo_max_samples_per_frame * 4);
104 }
105 }
106
107 static int alac_set_info(ALACContext *alac)
108 {
109 unsigned char *ptr = alac->avctx->extradata;
110
111 ptr += 4; /* size */
112 ptr += 4; /* alac */
113 ptr += 4; /* 0 ? */
114
115 if(AV_RB32(ptr) >= UINT_MAX/4){
116 av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
117 return -1;
118 }
119
120 /* buffer size / 2 ? */
121 alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr);
122 alac->setinfo_7a = *ptr++;
123 alac->setinfo_sample_size = *ptr++;
124 alac->setinfo_rice_historymult = *ptr++;
125 alac->setinfo_rice_initialhistory = *ptr++;
126 alac->setinfo_rice_kmodifier = *ptr++;
127 /* channels? */
128 alac->setinfo_7f = *ptr++;
129 alac->setinfo_80 = bytestream_get_be16(&ptr);
130 /* max coded frame size */
131 alac->setinfo_82 = bytestream_get_be32(&ptr);
132 /* bitrate ? */
133 alac->setinfo_86 = bytestream_get_be32(&ptr);
134 /* samplerate */
135 alac->setinfo_8a_rate = bytestream_get_be32(&ptr);
136
137 allocate_buffers(alac);
138
139 return 0;
140 }
141
142 /* hideously inefficient. could use a bitmask search,
143 * alternatively bsr on x86,
144 */
145 static int count_leading_zeros(int32_t input)
146 {
147 int i = 0;
148 while (!(0x80000000 & input) && i < 32) {
149 i++;
150 input = input << 1;
151 }
152 return i;
153 }
154
155 static void bastardized_rice_decompress(ALACContext *alac,
156 int32_t *output_buffer,
157 int output_size,
158 int readsamplesize, /* arg_10 */
159 int rice_initialhistory, /* arg424->b */
160 int rice_kmodifier, /* arg424->d */
161 int rice_historymult, /* arg424->c */
162 int rice_kmodifier_mask /* arg424->e */
163 )
164 {
165 int output_count;
166 unsigned int history = rice_initialhistory;
167 int sign_modifier = 0;
168
169 for (output_count = 0; output_count < output_size; output_count++) {
170 int32_t x = 0;
171 int32_t x_modified;
172 int32_t final_val;
173
174 /* read x - number of 1s before 0 represent the rice */
175 while (x <= 8 && get_bits1(&alac->gb)) {
176 x++;
177 }
178
179
180 if (x > 8) { /* RICE THRESHOLD */
181 /* use alternative encoding */
182 int32_t value;
183
184 value = get_bits(&alac->gb, readsamplesize);
185
186 /* mask value to readsamplesize size */
187 if (readsamplesize != 32)
188 value &= (0xffffffff >> (32 - readsamplesize));
189
190 x = value;
191 } else {
192 /* standard rice encoding */
193 int extrabits;
194 int k; /* size of extra bits */
195
196 /* read k, that is bits as is */
197 k = 31 - rice_kmodifier - count_leading_zeros((history >> 9) + 3);
198
199 if (k < 0)
200 k += rice_kmodifier;
201 else
202 k = rice_kmodifier;
203
204 if (k != 1) {
205 extrabits = show_bits(&alac->gb, k);
206
207 /* multiply x by 2^k - 1, as part of their strange algorithm */
208 x = (x << k) - x;
209
210 if (extrabits > 1) {
211 x += extrabits - 1;
212 get_bits(&alac->gb, k);
213 } else {
214 get_bits(&alac->gb, k - 1);
215 }
216 }
217 }
218
219 x_modified = sign_modifier + x;
220 final_val = (x_modified + 1) / 2;
221 if (x_modified & 1) final_val *= -1;
222
223 output_buffer[output_count] = final_val;
224
225 sign_modifier = 0;
226
227 /* now update the history */
228 history += (x_modified * rice_historymult)
229 - ((history * rice_historymult) >> 9);
230
231 if (x_modified > 0xffff)
232 history = 0xffff;
233
234 /* special case: there may be compressed blocks of 0 */
235 if ((history < 128) && (output_count+1 < output_size)) {
236 int block_size;
237
238 sign_modifier = 1;
239
240 x = 0;
241 while (x <= 8 && get_bits1(&alac->gb)) {
242 x++;
243 }
244
245 if (x > 8) {
246 block_size = get_bits(&alac->gb, 16);
247 block_size &= 0xffff;
248 } else {
249 int k;
250 int extrabits;
251
252 k = count_leading_zeros(history) + ((history + 16) >> 6 /* / 64 */) - 24;
253
254 extrabits = show_bits(&alac->gb, k);
255
256 block_size = (((1 << k) - 1) & rice_kmodifier_mask) * x
257 + extrabits - 1;
258
259 if (extrabits < 2) {
260 x = 1 - extrabits;
261 block_size += x;
262 get_bits(&alac->gb, k - 1);
263 } else {
264 get_bits(&alac->gb, k);
265 }
266 }
267
268 if (block_size > 0) {
269 memset(&output_buffer[output_count+1], 0, block_size * 4);
270 output_count += block_size;
271
272 }
273
274 if (block_size > 0xffff)
275 sign_modifier = 0;
276
277 history = 0;
278 }
279 }
280 }
281
282 #define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits))
283
284 #define SIGN_ONLY(v) \
285 ((v < 0) ? (-1) : \
286 ((v > 0) ? (1) : \
287 (0)))
288
289 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
290 int32_t *buffer_out,
291 int output_size,
292 int readsamplesize,
293 int16_t *predictor_coef_table,
294 int predictor_coef_num,
295 int predictor_quantitization)
296 {
297 int i;
298
299 /* first sample always copies */
300 *buffer_out = *error_buffer;
301
302 if (!predictor_coef_num) {
303 if (output_size <= 1) return;
304 memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
305 return;
306 }
307
308 if (predictor_coef_num == 0x1f) { /* 11111 - max value of predictor_coef_num */
309 /* second-best case scenario for fir decompression,
310 * error describes a small difference from the previous sample only
311 */
312 if (output_size <= 1) return;
313 for (i = 0; i < output_size - 1; i++) {
314 int32_t prev_value;
315 int32_t error_value;
316
317 prev_value = buffer_out[i];
318 error_value = error_buffer[i+1];
319 buffer_out[i+1] = SIGN_EXTENDED32((prev_value + error_value), readsamplesize);
320 }
321 return;
322 }
323
324 /* read warm-up samples */
325 if (predictor_coef_num > 0) {
326 int i;
327 for (i = 0; i < predictor_coef_num; i++) {
328 int32_t val;
329
330 val = buffer_out[i] + error_buffer[i+1];
331
332 val = SIGN_EXTENDED32(val, readsamplesize);
333
334 buffer_out[i+1] = val;
335 }
336 }
337
338 #if 0
339 /* 4 and 8 are very common cases (the only ones i've seen). these
340 * should be unrolled and optimised
341 */
342 if (predictor_coef_num == 4) {
343 /* FIXME: optimised general case */
344 return;
345 }
346
347 if (predictor_coef_table == 8) {
348 /* FIXME: optimised general case */
349 return;
350 }
351 #endif
352
353
354 /* general case */
355 if (predictor_coef_num > 0) {
356 for (i = predictor_coef_num + 1;
357 i < output_size;
358 i++) {
359 int j;
360 int sum = 0;
361 int outval;
362 int error_val = error_buffer[i];
363
364 for (j = 0; j < predictor_coef_num; j++) {
365 sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
366 predictor_coef_table[j];
367 }
368
369 outval = (1 << (predictor_quantitization-1)) + sum;
370 outval = outval >> predictor_quantitization;
371 outval = outval + buffer_out[0] + error_val;
372 outval = SIGN_EXTENDED32(outval, readsamplesize);
373
374 buffer_out[predictor_coef_num+1] = outval;
375
376 if (error_val > 0) {
377 int predictor_num = predictor_coef_num - 1;
378
379 while (predictor_num >= 0 && error_val > 0) {
380 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
381 int sign = SIGN_ONLY(val);
382
383 predictor_coef_table[predictor_num] -= sign;
384
385 val *= sign; /* absolute value */
386
387 error_val -= ((val >> predictor_quantitization) *
388 (predictor_coef_num - predictor_num));
389
390 predictor_num--;
391 }
392 } else if (error_val < 0) {
393 int predictor_num = predictor_coef_num - 1;
394
395 while (predictor_num >= 0 && error_val < 0) {
396 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
397 int sign = - SIGN_ONLY(val);
398
399 predictor_coef_table[predictor_num] -= sign;
400
401 val *= sign; /* neg value */
402
403 error_val -= ((val >> predictor_quantitization) *
404 (predictor_coef_num - predictor_num));
405
406 predictor_num--;
407 }
408 }
409
410 buffer_out++;
411 }
412 }
413 }
414
415 static void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b,
416 int16_t *buffer_out,
417 int numchannels, int numsamples,
418 uint8_t interlacing_shift,
419 uint8_t interlacing_leftweight)
420 {
421 int i;
422 if (numsamples <= 0) return;
423
424 /* weighted interlacing */
425 if (interlacing_leftweight) {
426 for (i = 0; i < numsamples; i++) {
427 int32_t difference, midright;
428 int16_t left;
429 int16_t right;
430
431 midright = buffer_a[i];
432 difference = buffer_b[i];
433
434
435 right = midright - ((difference * interlacing_leftweight) >> interlacing_shift);
436 left = (midright - ((difference * interlacing_leftweight) >> interlacing_shift))
437 + difference;
438
439 buffer_out[i*numchannels] = left;
440 buffer_out[i*numchannels + 1] = right;
441 }
442
443 return;
444 }
445
446 /* otherwise basic interlacing took place */
447 for (i = 0; i < numsamples; i++) {
448 int16_t left, right;
449
450 left = buffer_a[i];
451 right = buffer_b[i];
452
453 buffer_out[i*numchannels] = left;
454 buffer_out[i*numchannels + 1] = right;
455 }
456 }
457
458 static int alac_decode_frame(AVCodecContext *avctx,
459 void *outbuffer, int *outputsize,
460 uint8_t *inbuffer, int input_buffer_size)
461 {
462 ALACContext *alac = avctx->priv_data;
463
464 int channels;
465 int32_t outputsamples;
466 int hassize;
467 int readsamplesize;
468 int wasted_bytes;
469 int isnotcompressed;
470 uint8_t interlacing_shift;
471 uint8_t interlacing_leftweight;
472
473 /* short-circuit null buffers */
474 if (!inbuffer || !input_buffer_size)
475 return input_buffer_size;
476
477 /* initialize from the extradata */
478 if (!alac->context_initialized) {
479 if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
480 av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
481 ALAC_EXTRADATA_SIZE);
482 return input_buffer_size;
483 }
484 if (alac_set_info(alac)) {
485 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
486 return input_buffer_size;
487 }
488 alac->context_initialized = 1;
489 }
490
491 init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
492
493 channels = get_bits(&alac->gb, 3) + 1;
494 if (channels > MAX_CHANNELS) {
495 av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
496 MAX_CHANNELS);
497 return input_buffer_size;
498 }
499
500 /* 2^result = something to do with output waiting.
501 * perhaps matters if we read > 1 frame in a pass?
502 */
503 get_bits(&alac->gb, 4);
504
505 get_bits(&alac->gb, 12); /* unknown, skip 12 bits */
506
507 /* the output sample size is stored soon */
508 hassize = get_bits(&alac->gb, 1);
509
510 wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
511
512 /* whether the frame is compressed */
513 isnotcompressed = get_bits(&alac->gb, 1);
514
515 if (hassize) {
516 /* now read the number of samples as a 32bit integer */
517 outputsamples = get_bits(&alac->gb, 32);
518 } else
519 outputsamples = alac->setinfo_max_samples_per_frame;
520
521 *outputsize = outputsamples * alac->bytespersample;
522 readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
523
524 if (!isnotcompressed) {
525 /* so it is compressed */
526 int16_t predictor_coef_table[channels][32];
527 int predictor_coef_num[channels];
528 int prediction_type[channels];
529 int prediction_quantitization[channels];
530 int ricemodifier[channels];
531 int i, chan;
532
533 interlacing_shift = get_bits(&alac->gb, 8);
534 interlacing_leftweight = get_bits(&alac->gb, 8);
535
536 for (chan = 0; chan < channels; chan++) {
537 prediction_type[chan] = get_bits(&alac->gb, 4);
538 prediction_quantitization[chan] = get_bits(&alac->gb, 4);
539
540 ricemodifier[chan] = get_bits(&alac->gb, 3);
541 predictor_coef_num[chan] = get_bits(&alac->gb, 5);
542
543 /* read the predictor table */
544 for (i = 0; i < predictor_coef_num[chan]; i++) {
545 predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
546 }
547 }
548
549 if (wasted_bytes) {
550 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
551 }
552
553 for (chan = 0; chan < channels; chan++) {
554 bastardized_rice_decompress(alac,
555 alac->predicterror_buffer[chan],
556 outputsamples,
557 readsamplesize,
558 alac->setinfo_rice_initialhistory,
559 alac->setinfo_rice_kmodifier,
560 ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
561 (1 << alac->setinfo_rice_kmodifier) - 1);
562
563 if (prediction_type[chan] == 0) {
564 /* adaptive fir */
565 predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
566 alac->outputsamples_buffer[chan],
567 outputsamples,
568 readsamplesize,
569 predictor_coef_table[chan],
570 predictor_coef_num[chan],
571 prediction_quantitization[chan]);
572 } else {
573 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
574 /* i think the only other prediction type (or perhaps this is just a
575 * boolean?) runs adaptive fir twice.. like:
576 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
577 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
578 * little strange..
579 */
580 }
581 }
582 } else {
583 /* not compressed, easy case */
584 if (alac->setinfo_sample_size <= 16) {
585 int i, chan;
586 for (chan = 0; chan < channels; chan++) {
587 for (i = 0; i < outputsamples; i++) {
588 int32_t audiobits;
589
590 audiobits = get_bits(&alac->gb, alac->setinfo_sample_size);
591 audiobits = SIGN_EXTENDED32(audiobits, readsamplesize);
592
593 alac->outputsamples_buffer[chan][i] = audiobits;
594 }
595 }
596 } else {
597 int i, chan;
598 for (chan = 0; chan < channels; chan++) {
599 for (i = 0; i < outputsamples; i++) {
600 int32_t audiobits;
601
602 audiobits = get_bits(&alac->gb, 16);
603 /* special case of sign extension..
604 * as we'll be ORing the low 16bits into this */
605 audiobits = audiobits << 16;
606 audiobits = audiobits >> (32 - alac->setinfo_sample_size);
607 audiobits |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);
608
609 alac->outputsamples_buffer[chan][i] = audiobits;
610 }
611 }
612 }
613 /* wasted_bytes = 0; */
614 interlacing_shift = 0;
615 interlacing_leftweight = 0;
616 }
617
618 switch(alac->setinfo_sample_size) {
619 case 16: {
620 if (channels == 2) {
621 deinterlace_16(alac->outputsamples_buffer[0],
622 alac->outputsamples_buffer[1],
623 (int16_t*)outbuffer,
624 alac->numchannels,
625 outputsamples,
626 interlacing_shift,
627 interlacing_leftweight);
628 } else {
629 int i;
630 for (i = 0; i < outputsamples; i++) {
631 int16_t sample = alac->outputsamples_buffer[0][i];
632 ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
633 }
634 }
635 break;
636 }
637 case 20:
638 case 24:
639 case 32:
640 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
641 break;
642 default:
643 break;
644 }
645
646 return input_buffer_size;
647 }
648
649 static int alac_decode_init(AVCodecContext * avctx)
650 {
651 ALACContext *alac = avctx->priv_data;
652 alac->avctx = avctx;
653 alac->context_initialized = 0;
654
655 alac->samplesize = alac->avctx->bits_per_sample;
656 alac->numchannels = alac->avctx->channels;
657 alac->bytespersample = (alac->samplesize / 8) * alac->numchannels;
658
659 return 0;
660 }
661
662 static int alac_decode_close(AVCodecContext *avctx)
663 {
664 ALACContext *alac = avctx->priv_data;
665
666 int chan;
667 for (chan = 0; chan < MAX_CHANNELS; chan++) {
668 av_free(alac->predicterror_buffer[chan]);
669 av_free(alac->outputsamples_buffer[chan]);
670 }
671
672 return 0;
673 }
674
675 AVCodec alac_decoder = {
676 "alac",
677 CODEC_TYPE_AUDIO,
678 CODEC_ID_ALAC,
679 sizeof(ALACContext),
680 alac_decode_init,
681 NULL,
682 alac_decode_close,
683 alac_decode_frame,
684 };