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
26 * @see http://crazney.net/programs/itunes/alac.html
28 * Note: This decoder expects a 36- (0x24-)byte QuickTime atom to be
29 * passed through the extradata[_size] fields. This atom is tacked onto
30 * the end of an 'alac' stsd atom and has the following format:
31 * bytes 0-3 atom size (0x24), big-endian
32 * bytes 4-7 atom type ('alac', not the 'alac' tag from start of stsd)
33 * bytes 8-35 data bytes needed by decoder
39 * 32bit max sample per frame
43 * 8bit initial history
47 * 32bit max coded frame size
55 #include "bytestream.h"
59 #define ALAC_EXTRADATA_SIZE 36
60 #define MAX_CHANNELS 2
64 AVCodecContext *avctx;
71 int32_t *predicterror_buffer[MAX_CHANNELS];
73 int32_t *outputsamples_buffer[MAX_CHANNELS];
75 int32_t *extra_bits_buffer[MAX_CHANNELS];
77 /* stuff from setinfo */
78 uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */ /* max samples per frame? */
79 uint8_t setinfo_sample_size; /* 0x10 */
80 uint8_t setinfo_rice_historymult; /* 0x28 */
81 uint8_t setinfo_rice_initialhistory; /* 0x0a */
82 uint8_t setinfo_rice_kmodifier; /* 0x0e */
83 /* end setinfo stuff */
85 int extra_bits; /**< number of extra bits beyond 16-bit */
88 static inline int decode_scalar(GetBitContext *gb, int k, int limit, int readsamplesize){
89 /* read x - number of 1s before 0 represent the rice */
90 int x = get_unary_0_9(gb);
92 if (x > 8) { /* RICE THRESHOLD */
93 /* use alternative encoding */
94 x = get_bits(gb, readsamplesize);
100 int extrabits = show_bits(gb, k);
102 /* multiply x by 2^k - 1, as part of their strange algorithm */
109 skip_bits(gb, k - 1);
115 static int bastardized_rice_decompress(ALACContext *alac,
116 int32_t *output_buffer,
118 int readsamplesize, /* arg_10 */
119 int rice_initialhistory, /* arg424->b */
120 int rice_kmodifier, /* arg424->d */
121 int rice_historymult, /* arg424->c */
122 int rice_kmodifier_mask /* arg424->e */
126 unsigned int history = rice_initialhistory;
127 int sign_modifier = 0;
129 for (output_count = 0; output_count < output_size; output_count++) {
134 /* standard rice encoding */
135 int k; /* size of extra bits */
137 if(get_bits_left(&alac->gb) <= 0)
140 /* read k, that is bits as is */
141 k = av_log2((history >> 9) + 3);
142 x= decode_scalar(&alac->gb, k, rice_kmodifier, readsamplesize);
144 x_modified = sign_modifier + x;
145 final_val = (x_modified + 1) / 2;
146 if (x_modified & 1) final_val *= -1;
148 output_buffer[output_count] = final_val;
152 /* now update the history */
153 history += x_modified * rice_historymult
154 - ((history * rice_historymult) >> 9);
156 if (x_modified > 0xffff)
159 /* special case: there may be compressed blocks of 0 */
160 if ((history < 128) && (output_count+1 < output_size)) {
162 unsigned int block_size;
166 k = 7 - av_log2(history) + ((history + 16) >> 6 /* / 64 */);
168 block_size= decode_scalar(&alac->gb, k, rice_kmodifier, 16);
170 if (block_size > 0) {
171 if(block_size >= output_size - output_count){
172 av_log(alac->avctx, AV_LOG_ERROR, "invalid zero block size of %d %d %d\n", block_size, output_size, output_count);
173 block_size= output_size - output_count - 1;
175 memset(&output_buffer[output_count+1], 0, block_size * 4);
176 output_count += block_size;
179 if (block_size > 0xffff)
188 static inline int sign_only(int v)
190 return v ? FFSIGN(v) : 0;
193 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
197 int16_t *predictor_coef_table,
198 int predictor_coef_num,
199 int predictor_quantitization)
203 /* first sample always copies */
204 *buffer_out = *error_buffer;
206 if (!predictor_coef_num) {
207 if (output_size <= 1)
210 memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
214 if (predictor_coef_num == 0x1f) { /* 11111 - max value of predictor_coef_num */
215 /* second-best case scenario for fir decompression,
216 * error describes a small difference from the previous sample only
218 if (output_size <= 1)
220 for (i = 0; i < output_size - 1; i++) {
224 prev_value = buffer_out[i];
225 error_value = error_buffer[i+1];
227 sign_extend((prev_value + error_value), readsamplesize);
232 /* read warm-up samples */
233 if (predictor_coef_num > 0)
234 for (i = 0; i < predictor_coef_num; i++) {
237 val = buffer_out[i] + error_buffer[i+1];
238 val = sign_extend(val, readsamplesize);
239 buffer_out[i+1] = val;
242 /* 4 and 8 are very common cases (the only ones i've seen). these
243 * should be unrolled and optimized
247 if (predictor_coef_num > 0) {
248 for (i = predictor_coef_num + 1; i < output_size; i++) {
252 int error_val = error_buffer[i];
254 for (j = 0; j < predictor_coef_num; j++) {
255 sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
256 predictor_coef_table[j];
259 outval = (1 << (predictor_quantitization-1)) + sum;
260 outval = outval >> predictor_quantitization;
261 outval = outval + buffer_out[0] + error_val;
262 outval = sign_extend(outval, readsamplesize);
264 buffer_out[predictor_coef_num+1] = outval;
267 int predictor_num = predictor_coef_num - 1;
269 while (predictor_num >= 0 && error_val > 0) {
270 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
271 int sign = sign_only(val);
273 predictor_coef_table[predictor_num] -= sign;
275 val *= sign; /* absolute value */
277 error_val -= ((val >> predictor_quantitization) *
278 (predictor_coef_num - predictor_num));
282 } else if (error_val < 0) {
283 int predictor_num = predictor_coef_num - 1;
285 while (predictor_num >= 0 && error_val < 0) {
286 int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
287 int sign = - sign_only(val);
289 predictor_coef_table[predictor_num] -= sign;
291 val *= sign; /* neg value */
293 error_val -= ((val >> predictor_quantitization) *
294 (predictor_coef_num - predictor_num));
305 static void decorrelate_stereo(int32_t *buffer[MAX_CHANNELS],
306 int numsamples, uint8_t interlacing_shift,
307 uint8_t interlacing_leftweight)
311 for (i = 0; i < numsamples; i++) {
317 a -= (b * interlacing_leftweight) >> interlacing_shift;
325 static void append_extra_bits(int32_t *buffer[MAX_CHANNELS],
326 int32_t *extra_bits_buffer[MAX_CHANNELS],
327 int extra_bits, int numchannels, int numsamples)
331 for (ch = 0; ch < numchannels; ch++)
332 for (i = 0; i < numsamples; i++)
333 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
336 static void interleave_stereo_16(int32_t *buffer[MAX_CHANNELS],
337 int16_t *buffer_out, int numsamples)
341 for (i = 0; i < numsamples; i++) {
342 *buffer_out++ = buffer[0][i];
343 *buffer_out++ = buffer[1][i];
347 static void interleave_stereo_24(int32_t *buffer[MAX_CHANNELS],
348 int32_t *buffer_out, int numsamples)
352 for (i = 0; i < numsamples; i++) {
353 *buffer_out++ = buffer[0][i] << 8;
354 *buffer_out++ = buffer[1][i] << 8;
358 static int alac_decode_frame(AVCodecContext *avctx, void *data,
359 int *got_frame_ptr, AVPacket *avpkt)
361 const uint8_t *inbuffer = avpkt->data;
362 int input_buffer_size = avpkt->size;
363 ALACContext *alac = avctx->priv_data;
366 unsigned int outputsamples;
368 unsigned int readsamplesize;
370 uint8_t interlacing_shift;
371 uint8_t interlacing_leftweight;
374 init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
376 channels = get_bits(&alac->gb, 3) + 1;
377 if (channels != avctx->channels) {
378 av_log(avctx, AV_LOG_ERROR, "frame header channel count mismatch\n");
379 return AVERROR_INVALIDDATA;
382 /* 2^result = something to do with output waiting.
383 * perhaps matters if we read > 1 frame in a pass?
385 skip_bits(&alac->gb, 4);
387 skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */
389 /* the output sample size is stored soon */
390 hassize = get_bits1(&alac->gb);
392 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
394 /* whether the frame is compressed */
395 isnotcompressed = get_bits1(&alac->gb);
398 /* now read the number of samples as a 32bit integer */
399 outputsamples = get_bits_long(&alac->gb, 32);
400 if(outputsamples > alac->setinfo_max_samples_per_frame){
401 av_log(avctx, AV_LOG_ERROR, "outputsamples %d > %d\n", outputsamples, alac->setinfo_max_samples_per_frame);
405 outputsamples = alac->setinfo_max_samples_per_frame;
407 /* get output buffer */
408 if (outputsamples > INT32_MAX) {
409 av_log(avctx, AV_LOG_ERROR, "unsupported block size: %u\n", outputsamples);
410 return AVERROR_INVALIDDATA;
412 alac->frame.nb_samples = outputsamples;
413 if ((ret = avctx->get_buffer(avctx, &alac->frame)) < 0) {
414 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
418 readsamplesize = alac->setinfo_sample_size - alac->extra_bits + channels - 1;
419 if (readsamplesize > MIN_CACHE_BITS) {
420 av_log(avctx, AV_LOG_ERROR, "readsamplesize too big (%d)\n", readsamplesize);
424 if (!isnotcompressed) {
425 /* so it is compressed */
426 int16_t predictor_coef_table[MAX_CHANNELS][32];
427 int predictor_coef_num[MAX_CHANNELS];
428 int prediction_type[MAX_CHANNELS];
429 int prediction_quantitization[MAX_CHANNELS];
430 int ricemodifier[MAX_CHANNELS];
432 interlacing_shift = get_bits(&alac->gb, 8);
433 interlacing_leftweight = get_bits(&alac->gb, 8);
435 for (ch = 0; ch < channels; ch++) {
436 prediction_type[ch] = get_bits(&alac->gb, 4);
437 prediction_quantitization[ch] = get_bits(&alac->gb, 4);
439 ricemodifier[ch] = get_bits(&alac->gb, 3);
440 predictor_coef_num[ch] = get_bits(&alac->gb, 5);
442 /* read the predictor table */
443 for (i = 0; i < predictor_coef_num[ch]; i++)
444 predictor_coef_table[ch][i] = (int16_t)get_bits(&alac->gb, 16);
447 if (alac->extra_bits) {
448 for (i = 0; i < outputsamples; i++) {
449 if(get_bits_left(&alac->gb) <= 0)
451 for (ch = 0; ch < channels; ch++)
452 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
455 for (ch = 0; ch < channels; ch++) {
456 int ret = bastardized_rice_decompress(alac,
457 alac->predicterror_buffer[ch],
460 alac->setinfo_rice_initialhistory,
461 alac->setinfo_rice_kmodifier,
462 ricemodifier[ch] * alac->setinfo_rice_historymult / 4,
463 (1 << alac->setinfo_rice_kmodifier) - 1);
467 if (prediction_type[ch] == 0) {
469 predictor_decompress_fir_adapt(alac->predicterror_buffer[ch],
470 alac->outputsamples_buffer[ch],
473 predictor_coef_table[ch],
474 predictor_coef_num[ch],
475 prediction_quantitization[ch]);
477 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[ch]);
478 /* I think the only other prediction type (or perhaps this is
479 * just a boolean?) runs adaptive fir twice.. like:
480 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
481 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
487 /* not compressed, easy case */
488 for (i = 0; i < outputsamples; i++) {
489 if(get_bits_left(&alac->gb) <= 0)
491 for (ch = 0; ch < channels; ch++) {
492 alac->outputsamples_buffer[ch][i] = get_sbits_long(&alac->gb,
493 alac->setinfo_sample_size);
496 alac->extra_bits = 0;
497 interlacing_shift = 0;
498 interlacing_leftweight = 0;
500 if (get_bits(&alac->gb, 3) != 7)
501 av_log(avctx, AV_LOG_ERROR, "Error : Wrong End Of Frame\n");
503 if (channels == 2 && interlacing_leftweight) {
504 decorrelate_stereo(alac->outputsamples_buffer, outputsamples,
505 interlacing_shift, interlacing_leftweight);
508 if (alac->extra_bits) {
509 append_extra_bits(alac->outputsamples_buffer, alac->extra_bits_buffer,
510 alac->extra_bits, alac->numchannels, outputsamples);
513 switch(alac->setinfo_sample_size) {
516 interleave_stereo_16(alac->outputsamples_buffer,
517 (int16_t *)alac->frame.data[0], outputsamples);
519 int16_t *outbuffer = (int16_t *)alac->frame.data[0];
520 for (i = 0; i < outputsamples; i++) {
521 outbuffer[i] = alac->outputsamples_buffer[0][i];
527 interleave_stereo_24(alac->outputsamples_buffer,
528 (int32_t *)alac->frame.data[0], outputsamples);
530 int32_t *outbuffer = (int32_t *)alac->frame.data[0];
531 for (i = 0; i < outputsamples; i++)
532 outbuffer[i] = alac->outputsamples_buffer[0][i] << 8;
537 if (input_buffer_size * 8 - get_bits_count(&alac->gb) > 8)
538 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n", input_buffer_size * 8 - get_bits_count(&alac->gb));
541 *(AVFrame *)data = alac->frame;
543 return input_buffer_size;
546 static av_cold int alac_decode_close(AVCodecContext *avctx)
548 ALACContext *alac = avctx->priv_data;
551 for (ch = 0; ch < alac->numchannels; ch++) {
552 av_freep(&alac->predicterror_buffer[ch]);
553 av_freep(&alac->outputsamples_buffer[ch]);
554 av_freep(&alac->extra_bits_buffer[ch]);
560 static int allocate_buffers(ALACContext *alac)
563 for (ch = 0; ch < alac->numchannels; ch++) {
564 int buf_size = alac->setinfo_max_samples_per_frame * sizeof(int32_t);
566 FF_ALLOC_OR_GOTO(alac->avctx, alac->predicterror_buffer[ch],
567 buf_size, buf_alloc_fail);
569 FF_ALLOC_OR_GOTO(alac->avctx, alac->outputsamples_buffer[ch],
570 buf_size, buf_alloc_fail);
572 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
573 buf_size, buf_alloc_fail);
577 alac_decode_close(alac->avctx);
578 return AVERROR(ENOMEM);
581 static int alac_set_info(ALACContext *alac)
583 const unsigned char *ptr = alac->avctx->extradata;
589 if(AV_RB32(ptr) >= UINT_MAX/4){
590 av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
594 /* buffer size / 2 ? */
595 alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr);
597 alac->setinfo_sample_size = *ptr++;
598 alac->setinfo_rice_historymult = *ptr++;
599 alac->setinfo_rice_initialhistory = *ptr++;
600 alac->setinfo_rice_kmodifier = *ptr++;
601 alac->numchannels = *ptr++;
602 bytestream_get_be16(&ptr); /* ??? */
603 bytestream_get_be32(&ptr); /* max coded frame size */
604 bytestream_get_be32(&ptr); /* bitrate ? */
605 bytestream_get_be32(&ptr); /* samplerate */
610 static av_cold int alac_decode_init(AVCodecContext * avctx)
613 ALACContext *alac = avctx->priv_data;
616 /* initialize from the extradata */
617 if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
618 av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
619 ALAC_EXTRADATA_SIZE);
622 if (alac_set_info(alac)) {
623 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
627 switch (alac->setinfo_sample_size) {
628 case 16: avctx->sample_fmt = AV_SAMPLE_FMT_S16;
630 case 24: avctx->sample_fmt = AV_SAMPLE_FMT_S32;
632 default: av_log_ask_for_sample(avctx, "Sample depth %d is not supported.\n",
633 alac->setinfo_sample_size);
634 return AVERROR_PATCHWELCOME;
637 if (alac->numchannels < 1) {
638 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
639 alac->numchannels = avctx->channels;
641 if (alac->numchannels > MAX_CHANNELS)
642 alac->numchannels = avctx->channels;
644 avctx->channels = alac->numchannels;
646 if (avctx->channels > MAX_CHANNELS) {
647 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
649 return AVERROR_PATCHWELCOME;
652 if ((ret = allocate_buffers(alac)) < 0) {
653 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
657 avcodec_get_frame_defaults(&alac->frame);
658 avctx->coded_frame = &alac->frame;
663 AVCodec ff_alac_decoder = {
665 .type = AVMEDIA_TYPE_AUDIO,
667 .priv_data_size = sizeof(ALACContext),
668 .init = alac_decode_init,
669 .close = alac_decode_close,
670 .decode = alac_decode_frame,
671 .capabilities = CODEC_CAP_DR1,
672 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),