2 * ALAC (Apple Lossless Audio Codec) decoder
3 * Copyright (c) 2005 David Hammerton
5 * This file is part of FFmpeg.
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.
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.
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
24 * ALAC (Apple Lossless Audio Codec) decoder
25 * @author 2005 David Hammerton
27 * For more information on the ALAC format, visit:
28 * http://crazney.net/programs/itunes/alac.html
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
41 * 32bit max sample per frame
45 * 8bit initial history
49 * 32bit max coded frame size
56 #include "bitstream.h"
57 #include "bytestream.h"
59 #define ALAC_EXTRADATA_SIZE 36
60 #define MAX_CHANNELS 2
64 AVCodecContext *avctx;
66 /* init to 0; first frame decode should initialize from extradata and
68 int context_initialized;
75 int32_t *predicterror_buffer[MAX_CHANNELS];
77 int32_t *outputsamples_buffer[MAX_CHANNELS];
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 */
95 static void allocate_buffers(ALACContext *alac)
98 for (chan = 0; chan < MAX_CHANNELS; chan++) {
99 alac->predicterror_buffer[chan] =
100 av_malloc(alac->setinfo_max_samples_per_frame * 4);
102 alac->outputsamples_buffer[chan] =
103 av_malloc(alac->setinfo_max_samples_per_frame * 4);
107 static int alac_set_info(ALACContext *alac)
109 unsigned char *ptr = alac->avctx->extradata;
115 if(AV_RB32(ptr) >= UINT_MAX/4){
116 av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
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++;
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);
133 alac->setinfo_86 = bytestream_get_be32(&ptr);
135 alac->setinfo_8a_rate = bytestream_get_be32(&ptr);
137 allocate_buffers(alac);
142 static inline int count_leading_zeros(int32_t input)
144 return 31-av_log2(input);
147 static void bastardized_rice_decompress(ALACContext *alac,
148 int32_t *output_buffer,
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 */
158 unsigned int history = rice_initialhistory;
159 int sign_modifier = 0;
161 for (output_count = 0; output_count < output_size; output_count++) {
166 /* read x - number of 1s before 0 represent the rice */
167 while (x <= 8 && get_bits1(&alac->gb)) {
172 if (x > 8) { /* RICE THRESHOLD */
173 /* use alternative encoding */
176 value = get_bits(&alac->gb, readsamplesize);
178 /* mask value to readsamplesize size */
179 if (readsamplesize != 32)
180 value &= (0xffffffff >> (32 - readsamplesize));
184 /* standard rice encoding */
186 int k; /* size of extra bits */
188 /* read k, that is bits as is */
189 k = 31 - rice_kmodifier - count_leading_zeros((history >> 9) + 3);
197 extrabits = show_bits(&alac->gb, k);
199 /* multiply x by 2^k - 1, as part of their strange algorithm */
204 get_bits(&alac->gb, k);
206 get_bits(&alac->gb, k - 1);
211 x_modified = sign_modifier + x;
212 final_val = (x_modified + 1) / 2;
213 if (x_modified & 1) final_val *= -1;
215 output_buffer[output_count] = final_val;
219 /* now update the history */
220 history += (x_modified * rice_historymult)
221 - ((history * rice_historymult) >> 9);
223 if (x_modified > 0xffff)
226 /* special case: there may be compressed blocks of 0 */
227 if ((history < 128) && (output_count+1 < output_size)) {
233 while (x <= 8 && get_bits1(&alac->gb)) {
238 block_size = get_bits(&alac->gb, 16);
239 block_size &= 0xffff;
244 k = count_leading_zeros(history) + ((history + 16) >> 6 /* / 64 */) - 24;
246 extrabits = show_bits(&alac->gb, k);
248 block_size = (((1 << k) - 1) & rice_kmodifier_mask) * x
254 get_bits(&alac->gb, k - 1);
256 get_bits(&alac->gb, k);
260 if (block_size > 0) {
261 memset(&output_buffer[output_count+1], 0, block_size * 4);
262 output_count += block_size;
266 if (block_size > 0xffff)
274 #define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits))
276 #define SIGN_ONLY(v) \
281 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
285 int16_t *predictor_coef_table,
286 int predictor_coef_num,
287 int predictor_quantitization)
291 /* first sample always copies */
292 *buffer_out = *error_buffer;
294 if (!predictor_coef_num) {
295 if (output_size <= 1) return;
296 memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
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
304 if (output_size <= 1) return;
305 for (i = 0; i < output_size - 1; i++) {
309 prev_value = buffer_out[i];
310 error_value = error_buffer[i+1];
311 buffer_out[i+1] = SIGN_EXTENDED32((prev_value + error_value), readsamplesize);
316 /* read warm-up samples */
317 if (predictor_coef_num > 0) {
319 for (i = 0; i < predictor_coef_num; i++) {
322 val = buffer_out[i] + error_buffer[i+1];
324 val = SIGN_EXTENDED32(val, readsamplesize);
326 buffer_out[i+1] = val;
331 /* 4 and 8 are very common cases (the only ones i've seen). these
332 * should be unrolled and optimised
334 if (predictor_coef_num == 4) {
335 /* FIXME: optimised general case */
339 if (predictor_coef_table == 8) {
340 /* FIXME: optimised general case */
347 if (predictor_coef_num > 0) {
348 for (i = predictor_coef_num + 1;
354 int error_val = error_buffer[i];
356 for (j = 0; j < predictor_coef_num; j++) {
357 sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
358 predictor_coef_table[j];
361 outval = (1 << (predictor_quantitization-1)) + sum;
362 outval = outval >> predictor_quantitization;
363 outval = outval + buffer_out[0] + error_val;
364 outval = SIGN_EXTENDED32(outval, readsamplesize);
366 buffer_out[predictor_coef_num+1] = outval;
369 int predictor_num = predictor_coef_num - 1;
371 while (predictor_num >= 0 && error_val > 0) {
372 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
373 int sign = SIGN_ONLY(val);
375 predictor_coef_table[predictor_num] -= sign;
377 val *= sign; /* absolute value */
379 error_val -= ((val >> predictor_quantitization) *
380 (predictor_coef_num - predictor_num));
384 } else if (error_val < 0) {
385 int predictor_num = predictor_coef_num - 1;
387 while (predictor_num >= 0 && error_val < 0) {
388 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
389 int sign = - SIGN_ONLY(val);
391 predictor_coef_table[predictor_num] -= sign;
393 val *= sign; /* neg value */
395 error_val -= ((val >> predictor_quantitization) *
396 (predictor_coef_num - predictor_num));
407 static void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b,
409 int numchannels, int numsamples,
410 uint8_t interlacing_shift,
411 uint8_t interlacing_leftweight)
414 if (numsamples <= 0) return;
416 /* weighted interlacing */
417 if (interlacing_leftweight) {
418 for (i = 0; i < numsamples; i++) {
419 int32_t difference, midright;
423 midright = buffer_a[i];
424 difference = buffer_b[i];
427 right = midright - ((difference * interlacing_leftweight) >> interlacing_shift);
428 left = (midright - ((difference * interlacing_leftweight) >> interlacing_shift))
431 buffer_out[i*numchannels] = left;
432 buffer_out[i*numchannels + 1] = right;
438 /* otherwise basic interlacing took place */
439 for (i = 0; i < numsamples; i++) {
445 buffer_out[i*numchannels] = left;
446 buffer_out[i*numchannels + 1] = right;
450 static int alac_decode_frame(AVCodecContext *avctx,
451 void *outbuffer, int *outputsize,
452 uint8_t *inbuffer, int input_buffer_size)
454 ALACContext *alac = avctx->priv_data;
457 int32_t outputsamples;
462 uint8_t interlacing_shift;
463 uint8_t interlacing_leftweight;
465 /* short-circuit null buffers */
466 if (!inbuffer || !input_buffer_size)
467 return input_buffer_size;
469 /* initialize from the extradata */
470 if (!alac->context_initialized) {
471 if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
472 av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
473 ALAC_EXTRADATA_SIZE);
474 return input_buffer_size;
476 if (alac_set_info(alac)) {
477 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
478 return input_buffer_size;
480 alac->context_initialized = 1;
483 init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
485 channels = get_bits(&alac->gb, 3) + 1;
486 if (channels > MAX_CHANNELS) {
487 av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
489 return input_buffer_size;
492 /* 2^result = something to do with output waiting.
493 * perhaps matters if we read > 1 frame in a pass?
495 get_bits(&alac->gb, 4);
497 get_bits(&alac->gb, 12); /* unknown, skip 12 bits */
499 /* the output sample size is stored soon */
500 hassize = get_bits(&alac->gb, 1);
502 wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
504 /* whether the frame is compressed */
505 isnotcompressed = get_bits(&alac->gb, 1);
508 /* now read the number of samples as a 32bit integer */
509 outputsamples = get_bits(&alac->gb, 32);
511 outputsamples = alac->setinfo_max_samples_per_frame;
513 *outputsize = outputsamples * alac->bytespersample;
514 readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
516 if (!isnotcompressed) {
517 /* so it is compressed */
518 int16_t predictor_coef_table[channels][32];
519 int predictor_coef_num[channels];
520 int prediction_type[channels];
521 int prediction_quantitization[channels];
522 int ricemodifier[channels];
525 interlacing_shift = get_bits(&alac->gb, 8);
526 interlacing_leftweight = get_bits(&alac->gb, 8);
528 for (chan = 0; chan < channels; chan++) {
529 prediction_type[chan] = get_bits(&alac->gb, 4);
530 prediction_quantitization[chan] = get_bits(&alac->gb, 4);
532 ricemodifier[chan] = get_bits(&alac->gb, 3);
533 predictor_coef_num[chan] = get_bits(&alac->gb, 5);
535 /* read the predictor table */
536 for (i = 0; i < predictor_coef_num[chan]; i++) {
537 predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
542 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
545 for (chan = 0; chan < channels; chan++) {
546 bastardized_rice_decompress(alac,
547 alac->predicterror_buffer[chan],
550 alac->setinfo_rice_initialhistory,
551 alac->setinfo_rice_kmodifier,
552 ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
553 (1 << alac->setinfo_rice_kmodifier) - 1);
555 if (prediction_type[chan] == 0) {
557 predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
558 alac->outputsamples_buffer[chan],
561 predictor_coef_table[chan],
562 predictor_coef_num[chan],
563 prediction_quantitization[chan]);
565 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
566 /* i think the only other prediction type (or perhaps this is just a
567 * boolean?) runs adaptive fir twice.. like:
568 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
569 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
575 /* not compressed, easy case */
576 if (alac->setinfo_sample_size <= 16) {
578 for (chan = 0; chan < channels; chan++) {
579 for (i = 0; i < outputsamples; i++) {
582 audiobits = get_bits(&alac->gb, alac->setinfo_sample_size);
583 audiobits = SIGN_EXTENDED32(audiobits, readsamplesize);
585 alac->outputsamples_buffer[chan][i] = audiobits;
590 for (chan = 0; chan < channels; chan++) {
591 for (i = 0; i < outputsamples; i++) {
594 audiobits = get_bits(&alac->gb, 16);
595 /* special case of sign extension..
596 * as we'll be ORing the low 16bits into this */
597 audiobits = audiobits << 16;
598 audiobits = audiobits >> (32 - alac->setinfo_sample_size);
599 audiobits |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);
601 alac->outputsamples_buffer[chan][i] = audiobits;
605 /* wasted_bytes = 0; */
606 interlacing_shift = 0;
607 interlacing_leftweight = 0;
610 switch(alac->setinfo_sample_size) {
613 deinterlace_16(alac->outputsamples_buffer[0],
614 alac->outputsamples_buffer[1],
619 interlacing_leftweight);
622 for (i = 0; i < outputsamples; i++) {
623 int16_t sample = alac->outputsamples_buffer[0][i];
624 ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
632 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
638 return input_buffer_size;
641 static int alac_decode_init(AVCodecContext * avctx)
643 ALACContext *alac = avctx->priv_data;
645 alac->context_initialized = 0;
647 alac->samplesize = alac->avctx->bits_per_sample;
648 alac->numchannels = alac->avctx->channels;
649 alac->bytespersample = (alac->samplesize / 8) * alac->numchannels;
654 static int alac_decode_close(AVCodecContext *avctx)
656 ALACContext *alac = avctx->priv_data;
659 for (chan = 0; chan < MAX_CHANNELS; chan++) {
660 av_free(alac->predicterror_buffer[chan]);
661 av_free(alac->outputsamples_buffer[chan]);
667 AVCodec alac_decoder = {