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