useless variable
[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 #include "unary.h"
59
60 #define ALAC_EXTRADATA_SIZE 36
61 #define MAX_CHANNELS 2
62
63 typedef struct {
64
65 AVCodecContext *avctx;
66 GetBitContext gb;
67 /* init to 0; first frame decode should initialize from extradata and
68 * set this to 1 */
69 int context_initialized;
70
71 int samplesize;
72 int numchannels;
73 int bytespersample;
74
75 /* buffers */
76 int32_t *predicterror_buffer[MAX_CHANNELS];
77
78 int32_t *outputsamples_buffer[MAX_CHANNELS];
79
80 /* stuff from setinfo */
81 uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */ /* max samples per frame? */
82 uint8_t setinfo_7a; /* 0x00 */
83 uint8_t setinfo_sample_size; /* 0x10 */
84 uint8_t setinfo_rice_historymult; /* 0x28 */
85 uint8_t setinfo_rice_initialhistory; /* 0x0a */
86 uint8_t setinfo_rice_kmodifier; /* 0x0e */
87 uint8_t setinfo_7f; /* 0x02 */
88 uint16_t setinfo_80; /* 0x00ff */
89 uint32_t setinfo_82; /* 0x000020e7 */ /* max sample size?? */
90 uint32_t setinfo_86; /* 0x00069fe4 */ /* bit rate (average)?? */
91 uint32_t setinfo_8a_rate; /* 0x0000ac44 */
92 /* end setinfo stuff */
93
94 } ALACContext;
95
96 static void allocate_buffers(ALACContext *alac)
97 {
98 int chan;
99 for (chan = 0; chan < MAX_CHANNELS; chan++) {
100 alac->predicterror_buffer[chan] =
101 av_malloc(alac->setinfo_max_samples_per_frame * 4);
102
103 alac->outputsamples_buffer[chan] =
104 av_malloc(alac->setinfo_max_samples_per_frame * 4);
105 }
106 }
107
108 static int alac_set_info(ALACContext *alac)
109 {
110 const unsigned char *ptr = alac->avctx->extradata;
111
112 ptr += 4; /* size */
113 ptr += 4; /* alac */
114 ptr += 4; /* 0 ? */
115
116 if(AV_RB32(ptr) >= UINT_MAX/4){
117 av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
118 return -1;
119 }
120
121 /* buffer size / 2 ? */
122 alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr);
123 alac->setinfo_7a = *ptr++;
124 alac->setinfo_sample_size = *ptr++;
125 alac->setinfo_rice_historymult = *ptr++;
126 alac->setinfo_rice_initialhistory = *ptr++;
127 alac->setinfo_rice_kmodifier = *ptr++;
128 /* channels? */
129 alac->setinfo_7f = *ptr++;
130 alac->setinfo_80 = bytestream_get_be16(&ptr);
131 /* max coded frame size */
132 alac->setinfo_82 = bytestream_get_be32(&ptr);
133 /* bitrate ? */
134 alac->setinfo_86 = bytestream_get_be32(&ptr);
135 /* samplerate */
136 alac->setinfo_8a_rate = bytestream_get_be32(&ptr);
137
138 allocate_buffers(alac);
139
140 return 0;
141 }
142
143 static inline int count_leading_zeros(int32_t input)
144 {
145 return 31-av_log2(input);
146 }
147
148 static void bastardized_rice_decompress(ALACContext *alac,
149 int32_t *output_buffer,
150 int output_size,
151 int readsamplesize, /* arg_10 */
152 int rice_initialhistory, /* arg424->b */
153 int rice_kmodifier, /* arg424->d */
154 int rice_historymult, /* arg424->c */
155 int rice_kmodifier_mask /* arg424->e */
156 )
157 {
158 int output_count;
159 unsigned int history = rice_initialhistory;
160 int sign_modifier = 0;
161
162 for (output_count = 0; output_count < output_size; output_count++) {
163 int32_t x;
164 int32_t x_modified;
165 int32_t final_val;
166
167 /* read x - number of 1s before 0 represent the rice */
168 x = get_unary_0_9(&alac->gb);
169
170 if (x > 8) { /* RICE THRESHOLD */
171 /* use alternative encoding */
172 x = get_bits(&alac->gb, readsamplesize);
173
174 /* mask value to readsamplesize size */
175 if (readsamplesize != 32)
176 x &= (0xffffffff >> (32 - readsamplesize));
177 } else {
178 /* standard rice encoding */
179 int extrabits;
180 int k; /* size of extra bits */
181
182 /* read k, that is bits as is */
183 k = 31 - count_leading_zeros((history >> 9) + 3);
184
185 if (k >= rice_kmodifier)
186 k = rice_kmodifier;
187
188 if (k != 1) {
189 extrabits = show_bits(&alac->gb, k);
190
191 /* multiply x by 2^k - 1, as part of their strange algorithm */
192 x = (x << k) - x;
193
194 if (extrabits > 1) {
195 x += extrabits - 1;
196 skip_bits(&alac->gb, k);
197 } else
198 skip_bits(&alac->gb, k - 1);
199 }
200 }
201
202 x_modified = sign_modifier + x;
203 final_val = (x_modified + 1) / 2;
204 if (x_modified & 1) final_val *= -1;
205
206 output_buffer[output_count] = final_val;
207
208 sign_modifier = 0;
209
210 /* now update the history */
211 history += x_modified * rice_historymult
212 - ((history * rice_historymult) >> 9);
213
214 if (x_modified > 0xffff)
215 history = 0xffff;
216
217 /* special case: there may be compressed blocks of 0 */
218 if ((history < 128) && (output_count+1 < output_size)) {
219 int block_size;
220
221 sign_modifier = 1;
222
223 x = get_unary_0_9(&alac->gb);
224
225 if (x > 8) {
226 block_size = get_bits(&alac->gb, 16);
227 block_size &= 0xffff;
228 } else {
229 int k;
230 int extrabits;
231
232 k = count_leading_zeros(history) + ((history + 16) >> 6 /* / 64 */) - 24;
233
234 extrabits = show_bits(&alac->gb, k);
235
236 block_size = (((1 << k) - 1) & rice_kmodifier_mask) * x
237 + extrabits - 1;
238
239 if (extrabits < 2) {
240 x = 1 - extrabits;
241 block_size += x;
242 skip_bits(&alac->gb, k - 1);
243 } else {
244 skip_bits(&alac->gb, k);
245 }
246 }
247
248 if (block_size > 0) {
249 memset(&output_buffer[output_count+1], 0, block_size * 4);
250 output_count += block_size;
251 }
252
253 if (block_size > 0xffff)
254 sign_modifier = 0;
255
256 history = 0;
257 }
258 }
259 }
260
261 static inline int32_t extend_sign32(int32_t val, int bits)
262 {
263 return (val << (32 - bits)) >> (32 - bits);
264 }
265
266 static inline int sign_only(int v)
267 {
268 return v ? FFSIGN(v) : 0;
269 }
270
271 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
272 int32_t *buffer_out,
273 int output_size,
274 int readsamplesize,
275 int16_t *predictor_coef_table,
276 int predictor_coef_num,
277 int predictor_quantitization)
278 {
279 int i;
280
281 /* first sample always copies */
282 *buffer_out = *error_buffer;
283
284 if (!predictor_coef_num) {
285 if (output_size <= 1)
286 return;
287
288 memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
289 return;
290 }
291
292 if (predictor_coef_num == 0x1f) { /* 11111 - max value of predictor_coef_num */
293 /* second-best case scenario for fir decompression,
294 * error describes a small difference from the previous sample only
295 */
296 if (output_size <= 1)
297 return;
298 for (i = 0; i < output_size - 1; i++) {
299 int32_t prev_value;
300 int32_t error_value;
301
302 prev_value = buffer_out[i];
303 error_value = error_buffer[i+1];
304 buffer_out[i+1] =
305 extend_sign32((prev_value + error_value), readsamplesize);
306 }
307 return;
308 }
309
310 /* read warm-up samples */
311 if (predictor_coef_num > 0)
312 for (i = 0; i < predictor_coef_num; i++) {
313 int32_t val;
314
315 val = buffer_out[i] + error_buffer[i+1];
316 val = extend_sign32(val, readsamplesize);
317 buffer_out[i+1] = val;
318 }
319
320 #if 0
321 /* 4 and 8 are very common cases (the only ones i've seen). these
322 * should be unrolled and optimized
323 */
324 if (predictor_coef_num == 4) {
325 /* FIXME: optimized general case */
326 return;
327 }
328
329 if (predictor_coef_table == 8) {
330 /* FIXME: optimized general case */
331 return;
332 }
333 #endif
334
335 /* general case */
336 if (predictor_coef_num > 0) {
337 for (i = predictor_coef_num + 1; i < output_size; i++) {
338 int j;
339 int sum = 0;
340 int outval;
341 int error_val = error_buffer[i];
342
343 for (j = 0; j < predictor_coef_num; j++) {
344 sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
345 predictor_coef_table[j];
346 }
347
348 outval = (1 << (predictor_quantitization-1)) + sum;
349 outval = outval >> predictor_quantitization;
350 outval = outval + buffer_out[0] + error_val;
351 outval = extend_sign32(outval, readsamplesize);
352
353 buffer_out[predictor_coef_num+1] = outval;
354
355 if (error_val > 0) {
356 int predictor_num = predictor_coef_num - 1;
357
358 while (predictor_num >= 0 && error_val > 0) {
359 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
360 int sign = sign_only(val);
361
362 predictor_coef_table[predictor_num] -= sign;
363
364 val *= sign; /* absolute value */
365
366 error_val -= ((val >> predictor_quantitization) *
367 (predictor_coef_num - predictor_num));
368
369 predictor_num--;
370 }
371 } else if (error_val < 0) {
372 int predictor_num = predictor_coef_num - 1;
373
374 while (predictor_num >= 0 && error_val < 0) {
375 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
376 int sign = - sign_only(val);
377
378 predictor_coef_table[predictor_num] -= sign;
379
380 val *= sign; /* neg value */
381
382 error_val -= ((val >> predictor_quantitization) *
383 (predictor_coef_num - predictor_num));
384
385 predictor_num--;
386 }
387 }
388
389 buffer_out++;
390 }
391 }
392 }
393
394 static void reconstruct_stereo_16(int32_t *buffer[MAX_CHANNELS],
395 int16_t *buffer_out,
396 int numchannels, int numsamples,
397 uint8_t interlacing_shift,
398 uint8_t interlacing_leftweight)
399 {
400 int i;
401 if (numsamples <= 0)
402 return;
403
404 /* weighted interlacing */
405 if (interlacing_leftweight) {
406 for (i = 0; i < numsamples; i++) {
407 int32_t a, b;
408
409 a = buffer[0][i];
410 b = buffer[1][i];
411
412 a -= (b * interlacing_leftweight) >> interlacing_shift;
413 b += a;
414
415 buffer_out[i*numchannels] = b;
416 buffer_out[i*numchannels + 1] = a;
417 }
418
419 return;
420 }
421
422 /* otherwise basic interlacing took place */
423 for (i = 0; i < numsamples; i++) {
424 int16_t left, right;
425
426 left = buffer[0][i];
427 right = buffer[1][i];
428
429 buffer_out[i*numchannels] = left;
430 buffer_out[i*numchannels + 1] = right;
431 }
432 }
433
434 static int alac_decode_frame(AVCodecContext *avctx,
435 void *outbuffer, int *outputsize,
436 const uint8_t *inbuffer, int input_buffer_size)
437 {
438 ALACContext *alac = avctx->priv_data;
439
440 int channels;
441 int32_t outputsamples;
442 int hassize;
443 int readsamplesize;
444 int wasted_bytes;
445 int isnotcompressed;
446 uint8_t interlacing_shift;
447 uint8_t interlacing_leftweight;
448
449 /* short-circuit null buffers */
450 if (!inbuffer || !input_buffer_size)
451 return input_buffer_size;
452
453 /* initialize from the extradata */
454 if (!alac->context_initialized) {
455 if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
456 av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
457 ALAC_EXTRADATA_SIZE);
458 return input_buffer_size;
459 }
460 if (alac_set_info(alac)) {
461 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
462 return input_buffer_size;
463 }
464 alac->context_initialized = 1;
465 }
466
467 init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
468
469 channels = get_bits(&alac->gb, 3) + 1;
470 if (channels > MAX_CHANNELS) {
471 av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
472 MAX_CHANNELS);
473 return input_buffer_size;
474 }
475
476 /* 2^result = something to do with output waiting.
477 * perhaps matters if we read > 1 frame in a pass?
478 */
479 skip_bits(&alac->gb, 4);
480
481 skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */
482
483 /* the output sample size is stored soon */
484 hassize = get_bits1(&alac->gb);
485
486 wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
487
488 /* whether the frame is compressed */
489 isnotcompressed = get_bits1(&alac->gb);
490
491 if (hassize) {
492 /* now read the number of samples as a 32bit integer */
493 outputsamples = get_bits(&alac->gb, 32);
494 } else
495 outputsamples = alac->setinfo_max_samples_per_frame;
496
497 *outputsize = outputsamples * alac->bytespersample;
498 readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
499
500 if (!isnotcompressed) {
501 /* so it is compressed */
502 int16_t predictor_coef_table[channels][32];
503 int predictor_coef_num[channels];
504 int prediction_type[channels];
505 int prediction_quantitization[channels];
506 int ricemodifier[channels];
507 int i, chan;
508
509 interlacing_shift = get_bits(&alac->gb, 8);
510 interlacing_leftweight = get_bits(&alac->gb, 8);
511
512 for (chan = 0; chan < channels; chan++) {
513 prediction_type[chan] = get_bits(&alac->gb, 4);
514 prediction_quantitization[chan] = get_bits(&alac->gb, 4);
515
516 ricemodifier[chan] = get_bits(&alac->gb, 3);
517 predictor_coef_num[chan] = get_bits(&alac->gb, 5);
518
519 /* read the predictor table */
520 for (i = 0; i < predictor_coef_num[chan]; i++)
521 predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
522 }
523
524 if (wasted_bytes)
525 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
526
527 for (chan = 0; chan < channels; chan++) {
528 bastardized_rice_decompress(alac,
529 alac->predicterror_buffer[chan],
530 outputsamples,
531 readsamplesize,
532 alac->setinfo_rice_initialhistory,
533 alac->setinfo_rice_kmodifier,
534 ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
535 (1 << alac->setinfo_rice_kmodifier) - 1);
536
537 if (prediction_type[chan] == 0) {
538 /* adaptive fir */
539 predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
540 alac->outputsamples_buffer[chan],
541 outputsamples,
542 readsamplesize,
543 predictor_coef_table[chan],
544 predictor_coef_num[chan],
545 prediction_quantitization[chan]);
546 } else {
547 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
548 /* I think the only other prediction type (or perhaps this is
549 * just a boolean?) runs adaptive fir twice.. like:
550 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
551 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
552 * little strange..
553 */
554 }
555 }
556 } else {
557 /* not compressed, easy case */
558 if (alac->setinfo_sample_size <= 16) {
559 int i, chan;
560 for (chan = 0; chan < channels; chan++)
561 for (i = 0; i < outputsamples; i++) {
562 int32_t audiobits;
563
564 audiobits = get_bits(&alac->gb, alac->setinfo_sample_size);
565 audiobits = extend_sign32(audiobits, readsamplesize);
566
567 alac->outputsamples_buffer[chan][i] = audiobits;
568 }
569 } else {
570 int i, chan;
571 for (chan = 0; chan < channels; chan++)
572 for (i = 0; i < outputsamples; i++) {
573 int32_t audiobits;
574
575 audiobits = get_bits(&alac->gb, 16);
576 /* special case of sign extension..
577 * as we'll be ORing the low 16bits into this */
578 audiobits = audiobits << 16;
579 audiobits = audiobits >> (32 - alac->setinfo_sample_size);
580 audiobits |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);
581
582 alac->outputsamples_buffer[chan][i] = audiobits;
583 }
584 }
585 /* wasted_bytes = 0; */
586 interlacing_shift = 0;
587 interlacing_leftweight = 0;
588 }
589
590 switch(alac->setinfo_sample_size) {
591 case 16:
592 if (channels == 2) {
593 reconstruct_stereo_16(alac->outputsamples_buffer,
594 (int16_t*)outbuffer,
595 alac->numchannels,
596 outputsamples,
597 interlacing_shift,
598 interlacing_leftweight);
599 } else {
600 int i;
601 for (i = 0; i < outputsamples; i++) {
602 int16_t sample = alac->outputsamples_buffer[0][i];
603 ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
604 }
605 }
606 break;
607 case 20:
608 case 24:
609 // It is not clear if there exist any encoder that creates 24 bit ALAC
610 // files. iTunes convert 24 bit raw files to 16 bit before encoding.
611 case 32:
612 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
613 break;
614 default:
615 break;
616 }
617
618 return input_buffer_size;
619 }
620
621 static av_cold int alac_decode_init(AVCodecContext * avctx)
622 {
623 ALACContext *alac = avctx->priv_data;
624 alac->avctx = avctx;
625 alac->context_initialized = 0;
626
627 alac->samplesize = alac->avctx->bits_per_sample;
628 alac->numchannels = alac->avctx->channels;
629 alac->bytespersample = (alac->samplesize / 8) * alac->numchannels;
630
631 return 0;
632 }
633
634 static av_cold int alac_decode_close(AVCodecContext *avctx)
635 {
636 ALACContext *alac = avctx->priv_data;
637
638 int chan;
639 for (chan = 0; chan < MAX_CHANNELS; chan++) {
640 av_free(alac->predicterror_buffer[chan]);
641 av_free(alac->outputsamples_buffer[chan]);
642 }
643
644 return 0;
645 }
646
647 AVCodec alac_decoder = {
648 "alac",
649 CODEC_TYPE_AUDIO,
650 CODEC_ID_ALAC,
651 sizeof(ALACContext),
652 alac_decode_init,
653 NULL,
654 alac_decode_close,
655 alac_decode_frame,
656 };