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++;
127 alac->setinfo_7f = *ptr++; // channels?
128 alac->setinfo_80 = bytestream_get_be16(&ptr);
129 /* max coded frame size */
130 alac->setinfo_82 = bytestream_get_be32(&ptr);
132 alac->setinfo_86 = bytestream_get_be32(&ptr);
134 alac->setinfo_8a_rate = bytestream_get_be32(&ptr);
136 allocate_buffers(alac);
141 /* hideously inefficient. could use a bitmask search,
142 * alternatively bsr on x86,
144 static int count_leading_zeros(int32_t input)
147 while (!(0x80000000 & input) && i < 32) {
154 static void bastardized_rice_decompress(ALACContext *alac,
155 int32_t *output_buffer,
157 int readsamplesize, /* arg_10 */
158 int rice_initialhistory, /* arg424->b */
159 int rice_kmodifier, /* arg424->d */
160 int rice_historymult, /* arg424->c */
161 int rice_kmodifier_mask /* arg424->e */
165 unsigned int history = rice_initialhistory;
166 int sign_modifier = 0;
168 for (output_count = 0; output_count < output_size; output_count++) {
173 /* read x - number of 1s before 0 represent the rice */
174 while (x <= 8 && get_bits1(&alac->gb)) {
179 if (x > 8) { /* RICE THRESHOLD */
180 /* use alternative encoding */
183 value = get_bits(&alac->gb, readsamplesize);
185 /* mask value to readsamplesize size */
186 if (readsamplesize != 32)
187 value &= (0xffffffff >> (32 - readsamplesize));
191 /* standard rice encoding */
193 int k; /* size of extra bits */
195 /* read k, that is bits as is */
196 k = 31 - rice_kmodifier - count_leading_zeros((history >> 9) + 3);
204 extrabits = show_bits(&alac->gb, k);
206 /* multiply x by 2^k - 1, as part of their strange algorithm */
211 get_bits(&alac->gb, k);
213 get_bits(&alac->gb, k - 1);
218 x_modified = sign_modifier + x;
219 final_val = (x_modified + 1) / 2;
220 if (x_modified & 1) final_val *= -1;
222 output_buffer[output_count] = final_val;
226 /* now update the history */
227 history += (x_modified * rice_historymult)
228 - ((history * rice_historymult) >> 9);
230 if (x_modified > 0xffff)
233 /* special case: there may be compressed blocks of 0 */
234 if ((history < 128) && (output_count+1 < output_size)) {
240 while (x <= 8 && get_bits1(&alac->gb)) {
245 block_size = get_bits(&alac->gb, 16);
246 block_size &= 0xffff;
251 k = count_leading_zeros(history) + ((history + 16) >> 6 /* / 64 */) - 24;
253 extrabits = show_bits(&alac->gb, k);
255 block_size = (((1 << k) - 1) & rice_kmodifier_mask) * x
261 get_bits(&alac->gb, k - 1);
263 get_bits(&alac->gb, k);
267 if (block_size > 0) {
268 memset(&output_buffer[output_count+1], 0, block_size * 4);
269 output_count += block_size;
273 if (block_size > 0xffff)
281 #define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits))
283 #define SIGN_ONLY(v) \
288 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
292 int16_t *predictor_coef_table,
293 int predictor_coef_num,
294 int predictor_quantitization)
298 /* first sample always copies */
299 *buffer_out = *error_buffer;
301 if (!predictor_coef_num) {
302 if (output_size <= 1) return;
303 memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
307 if (predictor_coef_num == 0x1f) { /* 11111 - max value of predictor_coef_num */
308 /* second-best case scenario for fir decompression,
309 * error describes a small difference from the previous sample only
311 if (output_size <= 1) return;
312 for (i = 0; i < output_size - 1; i++) {
316 prev_value = buffer_out[i];
317 error_value = error_buffer[i+1];
318 buffer_out[i+1] = SIGN_EXTENDED32((prev_value + error_value), readsamplesize);
323 /* read warm-up samples */
324 if (predictor_coef_num > 0) {
326 for (i = 0; i < predictor_coef_num; i++) {
329 val = buffer_out[i] + error_buffer[i+1];
331 val = SIGN_EXTENDED32(val, readsamplesize);
333 buffer_out[i+1] = val;
338 /* 4 and 8 are very common cases (the only ones i've seen). these
339 * should be unrolled and optimised
341 if (predictor_coef_num == 4) {
342 /* FIXME: optimised general case */
346 if (predictor_coef_table == 8) {
347 /* FIXME: optimised general case */
354 if (predictor_coef_num > 0) {
355 for (i = predictor_coef_num + 1;
361 int error_val = error_buffer[i];
363 for (j = 0; j < predictor_coef_num; j++) {
364 sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
365 predictor_coef_table[j];
368 outval = (1 << (predictor_quantitization-1)) + sum;
369 outval = outval >> predictor_quantitization;
370 outval = outval + buffer_out[0] + error_val;
371 outval = SIGN_EXTENDED32(outval, readsamplesize);
373 buffer_out[predictor_coef_num+1] = outval;
376 int predictor_num = predictor_coef_num - 1;
378 while (predictor_num >= 0 && error_val > 0) {
379 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
380 int sign = SIGN_ONLY(val);
382 predictor_coef_table[predictor_num] -= sign;
384 val *= sign; /* absolute value */
386 error_val -= ((val >> predictor_quantitization) *
387 (predictor_coef_num - predictor_num));
391 } else if (error_val < 0) {
392 int predictor_num = predictor_coef_num - 1;
394 while (predictor_num >= 0 && error_val < 0) {
395 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
396 int sign = - SIGN_ONLY(val);
398 predictor_coef_table[predictor_num] -= sign;
400 val *= sign; /* neg value */
402 error_val -= ((val >> predictor_quantitization) *
403 (predictor_coef_num - predictor_num));
414 static void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b,
416 int numchannels, int numsamples,
417 uint8_t interlacing_shift,
418 uint8_t interlacing_leftweight)
421 if (numsamples <= 0) return;
423 /* weighted interlacing */
424 if (interlacing_leftweight) {
425 for (i = 0; i < numsamples; i++) {
426 int32_t difference, midright;
430 midright = buffer_a[i];
431 difference = buffer_b[i];
434 right = midright - ((difference * interlacing_leftweight) >> interlacing_shift);
435 left = (midright - ((difference * interlacing_leftweight) >> interlacing_shift))
438 buffer_out[i*numchannels] = left;
439 buffer_out[i*numchannels + 1] = right;
445 /* otherwise basic interlacing took place */
446 for (i = 0; i < numsamples; i++) {
452 buffer_out[i*numchannels] = left;
453 buffer_out[i*numchannels + 1] = right;
457 static int alac_decode_frame(AVCodecContext *avctx,
458 void *outbuffer, int *outputsize,
459 uint8_t *inbuffer, int input_buffer_size)
461 ALACContext *alac = avctx->priv_data;
464 int32_t outputsamples;
469 uint8_t interlacing_shift;
470 uint8_t interlacing_leftweight;
472 /* short-circuit null buffers */
473 if (!inbuffer || !input_buffer_size)
474 return input_buffer_size;
476 /* initialize from the extradata */
477 if (!alac->context_initialized) {
478 if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
479 av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
480 ALAC_EXTRADATA_SIZE);
481 return input_buffer_size;
483 if (alac_set_info(alac)) {
484 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
485 return input_buffer_size;
487 alac->context_initialized = 1;
490 init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
492 channels = get_bits(&alac->gb, 3) + 1;
494 /* 2^result = something to do with output waiting.
495 * perhaps matters if we read > 1 frame in a pass?
497 get_bits(&alac->gb, 4);
499 get_bits(&alac->gb, 12); /* unknown, skip 12 bits */
501 hassize = get_bits(&alac->gb, 1); /* the output sample size is stored soon */
503 wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
505 isnotcompressed = get_bits(&alac->gb, 1); /* whether the frame is compressed */
508 /* now read the number of samples,
509 * as a 32bit integer */
510 outputsamples = get_bits(&alac->gb, 32);
512 outputsamples = alac->setinfo_max_samples_per_frame;
514 *outputsize = outputsamples * alac->bytespersample;
515 readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
518 case 1: { /* 1 channel */
521 if (!isnotcompressed) {
522 /* so it is compressed */
523 int16_t predictor_coef_table[32];
524 int predictor_coef_num;
526 int prediction_quantitization;
529 /* FIXME: skip 16 bits, not sure what they are. seem to be used in
530 * two channel case */
531 get_bits(&alac->gb, 8);
532 get_bits(&alac->gb, 8);
534 prediction_type = get_bits(&alac->gb, 4);
535 prediction_quantitization = get_bits(&alac->gb, 4);
537 ricemodifier = get_bits(&alac->gb, 3);
538 predictor_coef_num = get_bits(&alac->gb, 5);
540 /* read the predictor table */
541 for (i = 0; i < predictor_coef_num; i++) {
542 predictor_coef_table[i] = (int16_t)get_bits(&alac->gb, 16);
546 /* these bytes seem to have something to do with
549 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
552 bastardized_rice_decompress(alac,
553 alac->predicterror_buffer[0],
556 alac->setinfo_rice_initialhistory,
557 alac->setinfo_rice_kmodifier,
558 ricemodifier * alac->setinfo_rice_historymult / 4,
559 (1 << alac->setinfo_rice_kmodifier) - 1);
561 if (prediction_type == 0) {
563 predictor_decompress_fir_adapt(alac->predicterror_buffer[0],
564 alac->outputsamples_buffer[0],
567 predictor_coef_table,
569 prediction_quantitization);
571 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type);
572 /* i think the only other prediction type (or perhaps this is just a
573 * boolean?) runs adaptive fir twice.. like:
574 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
575 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
581 /* not compressed, easy case */
582 if (readsamplesize <= 16) {
584 for (i = 0; i < outputsamples; i++) {
585 int32_t audiobits = get_bits(&alac->gb, readsamplesize);
587 audiobits = SIGN_EXTENDED32(audiobits, readsamplesize);
589 alac->outputsamples_buffer[0][i] = audiobits;
593 for (i = 0; i < outputsamples; i++) {
596 audiobits = get_bits(&alac->gb, 16);
597 /* special case of sign extension..
598 * as we'll be ORing the low 16bits into this */
599 audiobits = audiobits << 16;
600 audiobits = audiobits >> (32 - readsamplesize);
602 audiobits |= get_bits(&alac->gb, readsamplesize - 16);
604 alac->outputsamples_buffer[0][i] = audiobits;
607 /* wasted_bytes = 0; // unused */
610 switch(alac->setinfo_sample_size) {
613 for (i = 0; i < outputsamples; i++) {
614 int16_t sample = alac->outputsamples_buffer[0][i];
615 ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
622 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
629 case 2: { /* 2 channels */
631 if (!isnotcompressed) {
633 int16_t predictor_coef_table[channels][32];
634 int predictor_coef_num[channels];
635 int prediction_type[channels];
636 int prediction_quantitization[channels];
637 int ricemodifier[channels];
641 interlacing_shift = get_bits(&alac->gb, 8);
642 interlacing_leftweight = get_bits(&alac->gb, 8);
644 for (chan = 0; chan < channels; chan++) {
645 prediction_type[chan] = get_bits(&alac->gb, 4);
646 prediction_quantitization[chan] = get_bits(&alac->gb, 4);
648 ricemodifier[chan] = get_bits(&alac->gb, 3);
649 predictor_coef_num[chan] = get_bits(&alac->gb, 5);
651 /* read the predictor table */
652 for (i = 0; i < predictor_coef_num[chan]; i++) {
653 predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
659 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
662 for (chan = 0; chan < channels; chan++) {
663 bastardized_rice_decompress(alac,
664 alac->predicterror_buffer[chan],
667 alac->setinfo_rice_initialhistory,
668 alac->setinfo_rice_kmodifier,
669 ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
670 (1 << alac->setinfo_rice_kmodifier) - 1);
672 if (prediction_type[chan] == 0) {
674 predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
675 alac->outputsamples_buffer[chan],
678 predictor_coef_table[chan],
679 predictor_coef_num[chan],
680 prediction_quantitization[chan]);
683 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
687 /* not compressed, easy case */
688 if (alac->setinfo_sample_size <= 16) {
690 for (chan = 0; chan < channels; chan++) {
691 for (i = 0; i < outputsamples; i++) {
694 audiobits = get_bits(&alac->gb, alac->setinfo_sample_size);
695 audiobits = SIGN_EXTENDED32(audiobits, alac->setinfo_sample_size);
696 alac->outputsamples_buffer[chan][i] = audiobits;
701 for (chan = 0; chan < channels; chan++) {
702 for (i = 0; i < outputsamples; i++) {
705 audiobits = get_bits(&alac->gb, 16);
706 audiobits = audiobits << 16;
707 audiobits = audiobits >> (32 - alac->setinfo_sample_size);
708 audiobits |= get_bits(&alac->gb, alac->setinfo_sample_size - 16);
710 alac->outputsamples_buffer[chan][i] = audiobits;
714 /* wasted_bytes = 0; */
715 interlacing_shift = 0;
716 interlacing_leftweight = 0;
719 switch(alac->setinfo_sample_size) {
721 deinterlace_16(alac->outputsamples_buffer[0],
722 alac->outputsamples_buffer[1],
727 interlacing_leftweight);
733 av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
743 return input_buffer_size;
746 static int alac_decode_init(AVCodecContext * avctx)
748 ALACContext *alac = avctx->priv_data;
750 alac->context_initialized = 0;
752 alac->samplesize = alac->avctx->bits_per_sample;
753 alac->numchannels = alac->avctx->channels;
754 alac->bytespersample = (alac->samplesize / 8) * alac->numchannels;
759 static int alac_decode_close(AVCodecContext *avctx)
761 ALACContext *alac = avctx->priv_data;
764 for (chan = 0; chan < MAX_CHANNELS; chan++) {
765 av_free(alac->predicterror_buffer[chan]);
766 av_free(alac->outputsamples_buffer[chan]);
772 AVCodec alac_decoder = {