Cosmetics: indentation
[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 static inline int count_leading_zeros(int32_t input)
143 {
144 return 31-av_log2(input);
145 }
146
147 static void bastardized_rice_decompress(ALACContext *alac,
148 int32_t *output_buffer,
149 int output_size,
150 int readsamplesize, /* arg_10 */
151 int rice_initialhistory, /* arg424->b */
152 int rice_kmodifier, /* arg424->d */
153 int rice_historymult, /* arg424->c */
154 int rice_kmodifier_mask /* arg424->e */
155 )
156 {
157 int output_count;
158 unsigned int history = rice_initialhistory;
159 int sign_modifier = 0;
160
161 for (output_count = 0; output_count < output_size; output_count++) {
162 int32_t x = 0;
163 int32_t x_modified;
164 int32_t final_val;
165
166 /* read x - number of 1s before 0 represent the rice */
167 while (x <= 8 && get_bits1(&alac->gb)) {
168 x++;
169 }
170
171
172 if (x > 8) { /* RICE THRESHOLD */
173 /* use alternative encoding */
174 int32_t value;
175
176 value = get_bits(&alac->gb, readsamplesize);
177
178 /* mask value to readsamplesize size */
179 if (readsamplesize != 32)
180 value &= (0xffffffff >> (32 - readsamplesize));
181
182 x = value;
183 } else {
184 /* standard rice encoding */
185 int extrabits;
186 int k; /* size of extra bits */
187
188 /* read k, that is bits as is */
189 k = 31 - rice_kmodifier - count_leading_zeros((history >> 9) + 3);
190
191 if (k < 0)
192 k += rice_kmodifier;
193 else
194 k = rice_kmodifier;
195
196 if (k != 1) {
197 extrabits = show_bits(&alac->gb, k);
198
199 /* multiply x by 2^k - 1, as part of their strange algorithm */
200 x = (x << k) - x;
201
202 if (extrabits > 1) {
203 x += extrabits - 1;
204 get_bits(&alac->gb, k);
205 } else
206 get_bits(&alac->gb, k - 1);
207 }
208 }
209
210 x_modified = sign_modifier + x;
211 final_val = (x_modified + 1) / 2;
212 if (x_modified & 1) final_val *= -1;
213
214 output_buffer[output_count] = final_val;
215
216 sign_modifier = 0;
217
218 /* now update the history */
219 history += x_modified * rice_historymult
220 - ((history * rice_historymult) >> 9);
221
222 if (x_modified > 0xffff)
223 history = 0xffff;
224
225 /* special case: there may be compressed blocks of 0 */
226 if ((history < 128) && (output_count+1 < output_size)) {
227 int block_size;
228
229 sign_modifier = 1;
230
231 x = 0;
232 while (x <= 8 && get_bits1(&alac->gb)) {
233 x++;
234 }
235
236 if (x > 8) {
237 block_size = get_bits(&alac->gb, 16);
238 block_size &= 0xffff;
239 } else {
240 int k;
241 int extrabits;
242
243 k = count_leading_zeros(history) + ((history + 16) >> 6 /* / 64 */) - 24;
244
245 extrabits = show_bits(&alac->gb, k);
246
247 block_size = (((1 << k) - 1) & rice_kmodifier_mask) * x
248 + extrabits - 1;
249
250 if (extrabits < 2) {
251 x = 1 - extrabits;
252 block_size += x;
253 get_bits(&alac->gb, k - 1);
254 } else {
255 get_bits(&alac->gb, k);
256 }
257 }
258
259 if (block_size > 0) {
260 memset(&output_buffer[output_count+1], 0, block_size * 4);
261 output_count += block_size;
262 }
263
264 if (block_size > 0xffff)
265 sign_modifier = 0;
266
267 history = 0;
268 }
269 }
270 }
271
272 #define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits))
273
274 #define SIGN_ONLY(v) \
275 ((v < 0) ? (-1) : \
276 ((v > 0) ? (1) : \
277 (0)))
278
279 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
280 int32_t *buffer_out,
281 int output_size,
282 int readsamplesize,
283 int16_t *predictor_coef_table,
284 int predictor_coef_num,
285 int predictor_quantitization)
286 {
287 int i;
288
289 /* first sample always copies */
290 *buffer_out = *error_buffer;
291
292 if (!predictor_coef_num) {
293 if (output_size <= 1)
294 return;
295
296 memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
297 return;
298 }
299
300 if (predictor_coef_num == 0x1f) { /* 11111 - max value of predictor_coef_num */
301 /* second-best case scenario for fir decompression,
302 * error describes a small difference from the previous sample only
303 */
304 if (output_size <= 1)
305 return;
306 for (i = 0; i < output_size - 1; i++) {
307 int32_t prev_value;
308 int32_t error_value;
309
310 prev_value = buffer_out[i];
311 error_value = error_buffer[i+1];
312 buffer_out[i+1] =
313 SIGN_EXTENDED32((prev_value + error_value), readsamplesize);
314 }
315 return;
316 }
317
318 /* read warm-up samples */
319 if (predictor_coef_num > 0)
320 for (i = 0; i < predictor_coef_num; i++) {
321 int32_t val;
322
323 val = buffer_out[i] + error_buffer[i+1];
324 val = SIGN_EXTENDED32(val, readsamplesize);
325 buffer_out[i+1] = val;
326 }
327
328 #if 0
329 /* 4 and 8 are very common cases (the only ones i've seen). these
330 * should be unrolled and optimised
331 */
332 if (predictor_coef_num == 4) {
333 /* FIXME: optimised general case */
334 return;
335 }
336
337 if (predictor_coef_table == 8) {
338 /* FIXME: optimised general case */
339 return;
340 }
341 #endif
342
343 /* general case */
344 if (predictor_coef_num > 0) {
345 for (i = predictor_coef_num + 1; i < output_size; i++) {
346 int j;
347 int sum = 0;
348 int outval;
349 int error_val = error_buffer[i];
350
351 for (j = 0; j < predictor_coef_num; j++) {
352 sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
353 predictor_coef_table[j];
354 }
355
356 outval = (1 << (predictor_quantitization-1)) + sum;
357 outval = outval >> predictor_quantitization;
358 outval = outval + buffer_out[0] + error_val;
359 outval = SIGN_EXTENDED32(outval, readsamplesize);
360
361 buffer_out[predictor_coef_num+1] = outval;
362
363 if (error_val > 0) {
364 int predictor_num = predictor_coef_num - 1;
365
366 while (predictor_num >= 0 && error_val > 0) {
367 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
368 int sign = SIGN_ONLY(val);
369
370 predictor_coef_table[predictor_num] -= sign;
371
372 val *= sign; /* absolute value */
373
374 error_val -= ((val >> predictor_quantitization) *
375 (predictor_coef_num - predictor_num));
376
377 predictor_num--;
378 }
379 } else if (error_val < 0) {
380 int predictor_num = predictor_coef_num - 1;
381
382 while (predictor_num >= 0 && error_val < 0) {
383 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
384 int sign = - SIGN_ONLY(val);
385
386 predictor_coef_table[predictor_num] -= sign;
387
388 val *= sign; /* neg value */
389
390 error_val -= ((val >> predictor_quantitization) *
391 (predictor_coef_num - predictor_num));
392
393 predictor_num--;
394 }
395 }
396
397 buffer_out++;
398 }
399 }
400 }
401
402 static void reconstruct_stereo_16(int32_t *buffer[MAX_CHANNELS],
403 int16_t *buffer_out,
404 int numchannels, int numsamples,
405 uint8_t interlacing_shift,
406 uint8_t interlacing_leftweight)
407 {
408 int i;
409 if (numsamples <= 0)
410 return;
411
412 /* weighted interlacing */
413 if (interlacing_leftweight) {
414 for (i = 0; i < numsamples; i++) {
415 int32_t a, b;
416
417 a = buffer[0][i];
418 b = buffer[1][i];
419
420 a -= (b * interlacing_leftweight) >> interlacing_shift;
421 b += a;
422
423 buffer_out[i*numchannels] = b;
424 buffer_out[i*numchannels + 1] = a;
425 }
426
427 return;
428 }
429
430 /* otherwise basic interlacing took place */
431 for (i = 0; i < numsamples; i++) {
432 int16_t left, right;
433
434 left = buffer[0][i];
435 right = buffer[1][i];
436
437 buffer_out[i*numchannels] = left;
438 buffer_out[i*numchannels + 1] = right;
439 }
440 }
441
442 static int alac_decode_frame(AVCodecContext *avctx,
443 void *outbuffer, int *outputsize,
444 uint8_t *inbuffer, int input_buffer_size)
445 {
446 ALACContext *alac = avctx->priv_data;
447
448 int channels;
449 int32_t outputsamples;
450 int hassize;
451 int readsamplesize;
452 int wasted_bytes;
453 int isnotcompressed;
454 uint8_t interlacing_shift;
455 uint8_t interlacing_leftweight;
456
457 /* short-circuit null buffers */
458 if (!inbuffer || !input_buffer_size)
459 return input_buffer_size;
460
461 /* initialize from the extradata */
462 if (!alac->context_initialized) {
463 if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
464 av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
465 ALAC_EXTRADATA_SIZE);
466 return input_buffer_size;
467 }
468 if (alac_set_info(alac)) {
469 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
470 return input_buffer_size;
471 }
472 alac->context_initialized = 1;
473 }
474
475 init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
476
477 channels = get_bits(&alac->gb, 3) + 1;
478 if (channels > MAX_CHANNELS) {
479 av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
480 MAX_CHANNELS);
481 return input_buffer_size;
482 }
483
484 /* 2^result = something to do with output waiting.
485 * perhaps matters if we read > 1 frame in a pass?
486 */
487 get_bits(&alac->gb, 4);
488
489 get_bits(&alac->gb, 12); /* unknown, skip 12 bits */
490
491 /* the output sample size is stored soon */
492 hassize = get_bits(&alac->gb, 1);
493
494 wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
495
496 /* whether the frame is compressed */
497 isnotcompressed = get_bits(&alac->gb, 1);
498
499 if (hassize) {
500 /* now read the number of samples as a 32bit integer */
501 outputsamples = get_bits(&alac->gb, 32);
502 } else
503 outputsamples = alac->setinfo_max_samples_per_frame;
504
505 *outputsize = outputsamples * alac->bytespersample;
506 readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
507
508 if (!isnotcompressed) {
509 /* so it is compressed */
510 int16_t predictor_coef_table[channels][32];
511 int predictor_coef_num[channels];
512 int prediction_type[channels];
513 int prediction_quantitization[channels];
514 int ricemodifier[channels];
515 int i, chan;
516
517 interlacing_shift = get_bits(&alac->gb, 8);
518 interlacing_leftweight = get_bits(&alac->gb, 8);
519
520 for (chan = 0; chan < channels; chan++) {
521 prediction_type[chan] = get_bits(&alac->gb, 4);
522 prediction_quantitization[chan] = get_bits(&alac->gb, 4);
523
524 ricemodifier[chan] = get_bits(&alac->gb, 3);
525 predictor_coef_num[chan] = get_bits(&alac->gb, 5);
526
527 /* read the predictor table */
528 for (i = 0; i < predictor_coef_num[chan]; i++)
529 predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
530 }
531
532 if (wasted_bytes)
533 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
534
535 for (chan = 0; chan < channels; chan++) {
536 bastardized_rice_decompress(alac,
537 alac->predicterror_buffer[chan],
538 outputsamples,
539 readsamplesize,
540 alac->setinfo_rice_initialhistory,
541 alac->setinfo_rice_kmodifier,
542 ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
543 (1 << alac->setinfo_rice_kmodifier) - 1);
544
545 if (prediction_type[chan] == 0) {
546 /* adaptive fir */
547 predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
548 alac->outputsamples_buffer[chan],
549 outputsamples,
550 readsamplesize,
551 predictor_coef_table[chan],
552 predictor_coef_num[chan],
553 prediction_quantitization[chan]);
554 } else {
555 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
556 /* I think the only other prediction type (or perhaps this is
557 * just a boolean?) runs adaptive fir twice.. like:
558 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
559 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
560 * little strange..
561 */
562 }
563 }
564 } else {
565 /* not compressed, easy case */
566 if (alac->setinfo_sample_size <= 16) {
567 int i, chan;
568 for (chan = 0; chan < channels; chan++)
569 for (i = 0; i < outputsamples; i++) {
570 int32_t audiobits;
571
572 audiobits = get_bits(&alac->gb, alac->setinfo_sample_size);
573 audiobits = SIGN_EXTENDED32(audiobits, readsamplesize);
574
575 alac->outputsamples_buffer[chan][i] = audiobits;
576 }
577 } else {
578 int i, chan;
579 for (chan = 0; chan < channels; chan++)
580 for (i = 0; i < outputsamples; i++) {
581 int32_t audiobits;
582
583 audiobits = get_bits(&alac->gb, 16);
584 /* special case of sign extension..
585 * as we'll be ORing the low 16bits into this */
586 audiobits = audiobits << 16;
587 audiobits = audiobits >> (32 - alac->setinfo_sample_size);
588 audiobits |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);
589
590 alac->outputsamples_buffer[chan][i] = audiobits;
591 }
592 }
593 /* wasted_bytes = 0; */
594 interlacing_shift = 0;
595 interlacing_leftweight = 0;
596 }
597
598 switch(alac->setinfo_sample_size) {
599 case 16:
600 if (channels == 2) {
601 reconstruct_stereo_16(alac->outputsamples_buffer,
602 (int16_t*)outbuffer,
603 alac->numchannels,
604 outputsamples,
605 interlacing_shift,
606 interlacing_leftweight);
607 } else {
608 int i;
609 for (i = 0; i < outputsamples; i++) {
610 int16_t sample = alac->outputsamples_buffer[0][i];
611 ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
612 }
613 }
614 break;
615 case 20:
616 case 24:
617 case 32:
618 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
619 break;
620 default:
621 break;
622 }
623
624 return input_buffer_size;
625 }
626
627 static int alac_decode_init(AVCodecContext * avctx)
628 {
629 ALACContext *alac = avctx->priv_data;
630 alac->avctx = avctx;
631 alac->context_initialized = 0;
632
633 alac->samplesize = alac->avctx->bits_per_sample;
634 alac->numchannels = alac->avctx->channels;
635 alac->bytespersample = (alac->samplesize / 8) * alac->numchannels;
636
637 return 0;
638 }
639
640 static int alac_decode_close(AVCodecContext *avctx)
641 {
642 ALACContext *alac = avctx->priv_data;
643
644 int chan;
645 for (chan = 0; chan < MAX_CHANNELS; chan++) {
646 av_free(alac->predicterror_buffer[chan]);
647 av_free(alac->outputsamples_buffer[chan]);
648 }
649
650 return 0;
651 }
652
653 AVCodec alac_decoder = {
654 "alac",
655 CODEC_TYPE_AUDIO,
656 CODEC_ID_ALAC,
657 sizeof(ALACContext),
658 alac_decode_init,
659 NULL,
660 alac_decode_close,
661 alac_decode_frame,
662 };