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