2 * ALAC (Apple Lossless Audio Codec) decoder
3 * Copyright (c) 2005 David Hammerton
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; 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-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:
34 * 32bit tag version (0)
35 * 32bit samples per frame (used when not set explicitly in the frames)
36 * 8bit compatible version (0)
38 * 8bit history mult (40)
39 * 8bit initial history (14)
40 * 8bit rice param limit (10)
43 * 32bit max coded frame size (0 means unknown)
44 * 32bit average bitrate (0 means unknown)
48 #include "libavutil/audioconvert.h"
51 #include "bytestream.h"
55 #define ALAC_EXTRADATA_SIZE 36
56 #define MAX_CHANNELS 8
60 AVCodecContext *avctx;
67 int32_t *predict_error_buffer[2];
68 int32_t *output_samples_buffer[2];
69 int32_t *extra_bits_buffer[2];
71 uint32_t max_samples_per_frame;
73 uint8_t rice_history_mult;
74 uint8_t rice_initial_history;
77 int extra_bits; /**< number of extra bits beyond 16-bit */
78 int nb_samples; /**< number of samples in the current frame */
81 enum RawDataBlockType {
82 /* At the moment, only SCE, CPE, LFE, and END are recognized. */
93 static const uint8_t alac_channel_layout_offsets[8][8] = {
100 { 2, 0, 1, 4, 5, 6, 3 },
101 { 2, 6, 7, 0, 1, 4, 5, 3 }
104 static const uint16_t alac_channel_layouts[8] = {
107 AV_CH_LAYOUT_SURROUND,
108 AV_CH_LAYOUT_4POINT0,
109 AV_CH_LAYOUT_5POINT0_BACK,
110 AV_CH_LAYOUT_5POINT1_BACK,
111 AV_CH_LAYOUT_6POINT1_BACK,
112 AV_CH_LAYOUT_7POINT1_WIDE_BACK
115 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
117 unsigned int x = get_unary_0_9(gb);
119 if (x > 8) { /* RICE THRESHOLD */
120 /* use alternative encoding */
121 x = get_bits_long(gb, bps);
123 int extrabits = show_bits(gb, k);
125 /* multiply x by 2^k - 1, as part of their strange algorithm */
132 skip_bits(gb, k - 1);
137 static void rice_decompress(ALACContext *alac, int32_t *output_buffer,
138 int nb_samples, int bps, int rice_history_mult)
141 unsigned int history = alac->rice_initial_history;
142 int sign_modifier = 0;
144 for (i = 0; i < nb_samples; i++) {
148 /* read k, that is bits as is */
149 k = av_log2((history >> 9) + 3);
150 k = FFMIN(k, alac->rice_limit);
151 x = decode_scalar(&alac->gb, k, bps);
155 output_buffer[i] = (x >> 1) ^ -(x & 1);
157 /* now update the history */
161 history += x * rice_history_mult -
162 ((history * rice_history_mult) >> 9);
164 /* special case: there may be compressed blocks of 0 */
165 if ((history < 128) && (i + 1 < nb_samples)) {
168 k = 7 - av_log2(history) + ((history + 16) >> 6 /* / 64 */);
169 k = FFMIN(k, alac->rice_limit);
171 block_size = decode_scalar(&alac->gb, k, 16);
173 if (block_size > 0) {
174 if (block_size >= nb_samples - i) {
175 av_log(alac->avctx, AV_LOG_ERROR,
176 "invalid zero block size of %d %d %d\n", block_size,
178 block_size = nb_samples - i - 1;
180 memset(&output_buffer[i + 1], 0,
181 block_size * sizeof(*output_buffer));
185 if (block_size <= 0xffff)
193 static inline int sign_only(int v)
195 return v ? FFSIGN(v) : 0;
198 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
199 int nb_samples, int bps, int16_t *lpc_coefs,
200 int lpc_order, int lpc_quant)
204 /* first sample always copies */
205 *buffer_out = *error_buffer;
211 memcpy(&buffer_out[1], &error_buffer[1],
212 (nb_samples - 1) * sizeof(*buffer_out));
216 if (lpc_order == 31) {
217 /* simple 1st-order prediction */
218 for (i = 1; i < nb_samples; i++) {
219 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
225 /* read warm-up samples */
226 for (i = 0; i < lpc_order; i++) {
227 buffer_out[i + 1] = sign_extend(buffer_out[i] + error_buffer[i + 1],
231 /* NOTE: 4 and 8 are very common cases that could be optimized. */
234 for (i = lpc_order; i < nb_samples - 1; i++) {
237 int error_val = error_buffer[i + 1];
239 int d = buffer_out[i - lpc_order];
241 for (j = 0; j < lpc_order; j++) {
242 val += (buffer_out[i - j] - d) * lpc_coefs[j];
245 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
246 val += d + error_val;
248 buffer_out[i + 1] = sign_extend(val, bps);
250 /* adapt LPC coefficients */
251 error_sign = sign_only(error_val);
253 for (j = lpc_order - 1; j >= 0 && error_val * error_sign > 0; j--) {
255 val = d - buffer_out[i - j];
256 sign = sign_only(val) * error_sign;
257 lpc_coefs[j] -= sign;
259 error_val -= (val >> lpc_quant) * (lpc_order - j);
265 static void decorrelate_stereo(int32_t *buffer[2],
266 int nb_samples, uint8_t decorr_shift,
267 uint8_t decorr_left_weight)
271 for (i = 0; i < nb_samples; i++) {
277 a -= (b * decorr_left_weight) >> decorr_shift;
285 static void append_extra_bits(int32_t *buffer[2],
286 int32_t *extra_bits_buffer[2],
287 int extra_bits, int channels, int nb_samples)
291 for (ch = 0; ch < channels; ch++)
292 for (i = 0; i < nb_samples; i++)
293 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
296 static int decode_element(AVCodecContext *avctx, void *data, int ch_index,
299 ALACContext *alac = avctx->priv_data;
303 uint8_t decorr_shift;
304 uint8_t decorr_left_weight;
305 uint32_t output_samples;
308 skip_bits(&alac->gb, 4); /* element instance tag */
309 skip_bits(&alac->gb, 12); /* unused header bits */
311 /* the number of output samples is stored in the frame */
312 has_size = get_bits1(&alac->gb);
314 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
315 bps = alac->sample_size - alac->extra_bits + channels - 1;
317 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
318 return AVERROR_PATCHWELCOME;
321 /* whether the frame is compressed */
322 is_compressed = !get_bits1(&alac->gb);
325 output_samples = get_bits_long(&alac->gb, 32);
327 output_samples = alac->max_samples_per_frame;
328 if (!output_samples || output_samples > alac->max_samples_per_frame) {
329 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %d\n",
331 return AVERROR_INVALIDDATA;
333 if (!alac->nb_samples) {
334 /* get output buffer */
335 alac->frame.nb_samples = output_samples;
336 if ((ret = avctx->get_buffer(avctx, &alac->frame)) < 0) {
337 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
340 if (alac->sample_size > 16) {
341 for (ch = 0; ch < channels; ch++)
342 alac->output_samples_buffer[ch] = (int32_t *)alac->frame.extended_data[ch_index + ch];
344 } else if (output_samples != alac->nb_samples) {
345 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %u != %d\n",
346 output_samples, alac->nb_samples);
347 return AVERROR_INVALIDDATA;
349 alac->nb_samples = output_samples;
352 int16_t lpc_coefs[2][32];
354 int prediction_type[2];
356 int rice_history_mult[2];
358 decorr_shift = get_bits(&alac->gb, 8);
359 decorr_left_weight = get_bits(&alac->gb, 8);
361 for (ch = 0; ch < channels; ch++) {
362 prediction_type[ch] = get_bits(&alac->gb, 4);
363 lpc_quant[ch] = get_bits(&alac->gb, 4);
364 rice_history_mult[ch] = get_bits(&alac->gb, 3);
365 lpc_order[ch] = get_bits(&alac->gb, 5);
367 /* read the predictor table */
368 for (i = 0; i < lpc_order[ch]; i++)
369 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
372 if (alac->extra_bits) {
373 for (i = 0; i < alac->nb_samples; i++) {
374 for (ch = 0; ch < channels; ch++)
375 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
378 for (ch = 0; ch < channels; ch++) {
379 rice_decompress(alac, alac->predict_error_buffer[ch],
380 alac->nb_samples, bps,
381 rice_history_mult[ch] * alac->rice_history_mult / 4);
383 /* adaptive FIR filter */
384 if (prediction_type[ch] == 15) {
385 /* Prediction type 15 runs the adaptive FIR twice.
386 * The first pass uses the special-case coef_num = 31, while
387 * the second pass uses the coefs from the bitstream.
389 * However, this prediction type is not currently used by the
392 lpc_prediction(alac->predict_error_buffer[ch],
393 alac->predict_error_buffer[ch],
394 alac->nb_samples, bps, NULL, 31, 0);
395 } else if (prediction_type[ch] > 0) {
396 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
397 prediction_type[ch]);
399 lpc_prediction(alac->predict_error_buffer[ch],
400 alac->output_samples_buffer[ch], alac->nb_samples,
401 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
404 /* not compressed, easy case */
405 for (i = 0; i < alac->nb_samples; i++) {
406 for (ch = 0; ch < channels; ch++) {
407 alac->output_samples_buffer[ch][i] = get_sbits_long(&alac->gb, alac->sample_size);
410 alac->extra_bits = 0;
412 decorr_left_weight = 0;
415 if (channels == 2 && decorr_left_weight) {
416 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
417 decorr_shift, decorr_left_weight);
420 if (alac->extra_bits) {
421 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
422 alac->extra_bits, channels, alac->nb_samples);
425 switch(alac->sample_size) {
427 for (ch = 0; ch < channels; ch++) {
428 int16_t *outbuffer = (int16_t *)alac->frame.extended_data[ch_index + ch];
429 for (i = 0; i < alac->nb_samples; i++)
430 *outbuffer++ = alac->output_samples_buffer[ch][i];
434 for (ch = 0; ch < channels; ch++) {
435 for (i = 0; i < alac->nb_samples; i++)
436 alac->output_samples_buffer[ch][i] <<= 8;
444 static int alac_decode_frame(AVCodecContext *avctx, void *data,
445 int *got_frame_ptr, AVPacket *avpkt)
447 ALACContext *alac = avctx->priv_data;
448 enum RawDataBlockType element;
452 init_get_bits(&alac->gb, avpkt->data, avpkt->size * 8);
454 alac->nb_samples = 0;
456 while (get_bits_left(&alac->gb)) {
457 element = get_bits(&alac->gb, 3);
458 if (element == TYPE_END)
460 if (element > TYPE_CPE && element != TYPE_LFE) {
461 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d", element);
462 return AVERROR_PATCHWELCOME;
465 channels = (element == TYPE_CPE) ? 2 : 1;
466 if (ch + channels > alac->channels) {
467 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
468 return AVERROR_INVALIDDATA;
471 ret = decode_element(avctx, data,
472 alac_channel_layout_offsets[alac->channels - 1][ch],
480 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8)
481 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
482 avpkt->size * 8 - get_bits_count(&alac->gb));
485 *(AVFrame *)data = alac->frame;
490 static av_cold int alac_decode_close(AVCodecContext *avctx)
492 ALACContext *alac = avctx->priv_data;
495 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
496 av_freep(&alac->predict_error_buffer[ch]);
497 if (alac->sample_size == 16)
498 av_freep(&alac->output_samples_buffer[ch]);
499 av_freep(&alac->extra_bits_buffer[ch]);
505 static int allocate_buffers(ALACContext *alac)
508 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
509 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
511 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
512 buf_size, buf_alloc_fail);
514 if (alac->sample_size == 16) {
515 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
516 buf_size, buf_alloc_fail);
519 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
520 buf_size, buf_alloc_fail);
524 alac_decode_close(alac->avctx);
525 return AVERROR(ENOMEM);
528 static int alac_set_info(ALACContext *alac)
532 bytestream2_init(&gb, alac->avctx->extradata,
533 alac->avctx->extradata_size);
535 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
537 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
538 if (!alac->max_samples_per_frame || alac->max_samples_per_frame > INT_MAX) {
539 av_log(alac->avctx, AV_LOG_ERROR, "max samples per frame invalid: %u\n",
540 alac->max_samples_per_frame);
541 return AVERROR_INVALIDDATA;
543 bytestream2_skipu(&gb, 1); // compatible version
544 alac->sample_size = bytestream2_get_byteu(&gb);
545 alac->rice_history_mult = bytestream2_get_byteu(&gb);
546 alac->rice_initial_history = bytestream2_get_byteu(&gb);
547 alac->rice_limit = bytestream2_get_byteu(&gb);
548 alac->channels = bytestream2_get_byteu(&gb);
549 bytestream2_get_be16u(&gb); // maxRun
550 bytestream2_get_be32u(&gb); // max coded frame size
551 bytestream2_get_be32u(&gb); // average bitrate
552 bytestream2_get_be32u(&gb); // samplerate
557 static av_cold int alac_decode_init(AVCodecContext * avctx)
560 ALACContext *alac = avctx->priv_data;
563 /* initialize from the extradata */
564 if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
565 av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
566 ALAC_EXTRADATA_SIZE);
569 if (alac_set_info(alac)) {
570 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
574 switch (alac->sample_size) {
575 case 16: avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
578 case 32: avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
580 default: av_log_ask_for_sample(avctx, "Sample depth %d is not supported.\n",
582 return AVERROR_PATCHWELCOME;
585 if (alac->channels < 1) {
586 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
587 alac->channels = avctx->channels;
589 if (alac->channels > MAX_CHANNELS)
590 alac->channels = avctx->channels;
592 avctx->channels = alac->channels;
594 if (avctx->channels > MAX_CHANNELS) {
595 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
597 return AVERROR_PATCHWELCOME;
599 avctx->channel_layout = alac_channel_layouts[alac->channels - 1];
601 if ((ret = allocate_buffers(alac)) < 0) {
602 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
606 avcodec_get_frame_defaults(&alac->frame);
607 avctx->coded_frame = &alac->frame;
612 AVCodec ff_alac_decoder = {
614 .type = AVMEDIA_TYPE_AUDIO,
616 .priv_data_size = sizeof(ALACContext),
617 .init = alac_decode_init,
618 .close = alac_decode_close,
619 .decode = alac_decode_frame,
620 .capabilities = CODEC_CAP_DR1,
621 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),