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/channel_layout.h"
51 #include "bytestream.h"
55 #include "alac_data.h"
57 #define ALAC_EXTRADATA_SIZE 36
60 AVCodecContext *avctx;
64 int32_t *predict_error_buffer[2];
65 int32_t *output_samples_buffer[2];
66 int32_t *extra_bits_buffer[2];
68 uint32_t max_samples_per_frame;
70 uint8_t rice_history_mult;
71 uint8_t rice_initial_history;
74 int extra_bits; /**< number of extra bits beyond 16-bit */
75 int nb_samples; /**< number of samples in the current frame */
78 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
80 unsigned int x = get_unary_0_9(gb);
82 if (x > 8) { /* RICE THRESHOLD */
83 /* use alternative encoding */
84 x = get_bits_long(gb, bps);
86 int extrabits = show_bits(gb, k);
88 /* multiply x by 2^k - 1, as part of their strange algorithm */
100 static void rice_decompress(ALACContext *alac, int32_t *output_buffer,
101 int nb_samples, int bps, int rice_history_mult)
104 unsigned int history = alac->rice_initial_history;
105 int sign_modifier = 0;
107 for (i = 0; i < nb_samples; i++) {
111 /* calculate rice param and decode next value */
112 k = av_log2((history >> 9) + 3);
113 k = FFMIN(k, alac->rice_limit);
114 x = decode_scalar(&alac->gb, k, bps);
117 output_buffer[i] = (x >> 1) ^ -(x & 1);
119 /* update the history */
123 history += x * rice_history_mult -
124 ((history * rice_history_mult) >> 9);
126 /* special case: there may be compressed blocks of 0 */
127 if ((history < 128) && (i + 1 < nb_samples)) {
130 /* calculate rice param and decode block size */
131 k = 7 - av_log2(history) + ((history + 16) >> 6);
132 k = FFMIN(k, alac->rice_limit);
133 block_size = decode_scalar(&alac->gb, k, 16);
135 if (block_size > 0) {
136 if (block_size >= nb_samples - i) {
137 av_log(alac->avctx, AV_LOG_ERROR,
138 "invalid zero block size of %d %d %d\n", block_size,
140 block_size = nb_samples - i - 1;
142 memset(&output_buffer[i + 1], 0,
143 block_size * sizeof(*output_buffer));
146 if (block_size <= 0xffff)
153 static inline int sign_only(int v)
155 return v ? FFSIGN(v) : 0;
158 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
159 int nb_samples, int bps, int16_t *lpc_coefs,
160 int lpc_order, int lpc_quant)
163 int32_t *pred = buffer_out;
165 /* first sample always copies */
166 *buffer_out = *error_buffer;
172 memcpy(&buffer_out[1], &error_buffer[1],
173 (nb_samples - 1) * sizeof(*buffer_out));
177 if (lpc_order == 31) {
178 /* simple 1st-order prediction */
179 for (i = 1; i < nb_samples; i++) {
180 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
186 /* read warm-up samples */
187 for (i = 1; i <= lpc_order; i++)
188 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
190 /* NOTE: 4 and 8 are very common cases that could be optimized. */
192 for (; i < nb_samples; i++) {
195 int error_val = error_buffer[i];
200 for (j = 0; j < lpc_order; j++)
201 val += (pred[j] - d) * lpc_coefs[j];
202 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
203 val += d + error_val;
204 buffer_out[i] = sign_extend(val, bps);
206 /* adapt LPC coefficients */
207 error_sign = sign_only(error_val);
209 for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {
212 sign = sign_only(val) * error_sign;
213 lpc_coefs[j] -= sign;
215 error_val -= (val >> lpc_quant) * (j + 1);
221 static void decorrelate_stereo(int32_t *buffer[2], int nb_samples,
222 int decorr_shift, int decorr_left_weight)
226 for (i = 0; i < nb_samples; i++) {
232 a -= (b * decorr_left_weight) >> decorr_shift;
240 static void append_extra_bits(int32_t *buffer[2], int32_t *extra_bits_buffer[2],
241 int extra_bits, int channels, int nb_samples)
245 for (ch = 0; ch < channels; ch++)
246 for (i = 0; i < nb_samples; i++)
247 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
250 static int decode_element(AVCodecContext *avctx, AVFrame *frame, int ch_index,
253 ALACContext *alac = avctx->priv_data;
254 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
255 uint32_t output_samples;
258 skip_bits(&alac->gb, 4); /* element instance tag */
259 skip_bits(&alac->gb, 12); /* unused header bits */
261 /* the number of output samples is stored in the frame */
262 has_size = get_bits1(&alac->gb);
264 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
265 bps = alac->sample_size - alac->extra_bits + channels - 1;
267 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
268 return AVERROR_PATCHWELCOME;
271 /* whether the frame is compressed */
272 is_compressed = !get_bits1(&alac->gb);
275 output_samples = get_bits_long(&alac->gb, 32);
277 output_samples = alac->max_samples_per_frame;
278 if (!output_samples || output_samples > alac->max_samples_per_frame) {
279 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %d\n",
281 return AVERROR_INVALIDDATA;
283 if (!alac->nb_samples) {
284 /* get output buffer */
285 frame->nb_samples = output_samples;
286 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
287 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
290 } else if (output_samples != alac->nb_samples) {
291 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %u != %d\n",
292 output_samples, alac->nb_samples);
293 return AVERROR_INVALIDDATA;
295 alac->nb_samples = output_samples;
296 if (alac->sample_size > 16) {
297 for (ch = 0; ch < channels; ch++)
298 alac->output_samples_buffer[ch] = (int32_t *)frame->extended_data[ch_index + ch];
302 int16_t lpc_coefs[2][32];
304 int prediction_type[2];
306 int rice_history_mult[2];
308 decorr_shift = get_bits(&alac->gb, 8);
309 decorr_left_weight = get_bits(&alac->gb, 8);
311 for (ch = 0; ch < channels; ch++) {
312 prediction_type[ch] = get_bits(&alac->gb, 4);
313 lpc_quant[ch] = get_bits(&alac->gb, 4);
314 rice_history_mult[ch] = get_bits(&alac->gb, 3);
315 lpc_order[ch] = get_bits(&alac->gb, 5);
317 /* read the predictor table */
318 for (i = lpc_order[ch] - 1; i >= 0; i--)
319 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
322 if (alac->extra_bits) {
323 for (i = 0; i < alac->nb_samples; i++) {
324 for (ch = 0; ch < channels; ch++)
325 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
328 for (ch = 0; ch < channels; ch++) {
329 rice_decompress(alac, alac->predict_error_buffer[ch],
330 alac->nb_samples, bps,
331 rice_history_mult[ch] * alac->rice_history_mult / 4);
333 /* adaptive FIR filter */
334 if (prediction_type[ch] == 15) {
335 /* Prediction type 15 runs the adaptive FIR twice.
336 * The first pass uses the special-case coef_num = 31, while
337 * the second pass uses the coefs from the bitstream.
339 * However, this prediction type is not currently used by the
342 lpc_prediction(alac->predict_error_buffer[ch],
343 alac->predict_error_buffer[ch],
344 alac->nb_samples, bps, NULL, 31, 0);
345 } else if (prediction_type[ch] > 0) {
346 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
347 prediction_type[ch]);
349 lpc_prediction(alac->predict_error_buffer[ch],
350 alac->output_samples_buffer[ch], alac->nb_samples,
351 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
354 /* not compressed, easy case */
355 for (i = 0; i < alac->nb_samples; i++) {
356 for (ch = 0; ch < channels; ch++) {
357 alac->output_samples_buffer[ch][i] =
358 get_sbits_long(&alac->gb, alac->sample_size);
361 alac->extra_bits = 0;
363 decorr_left_weight = 0;
366 if (channels == 2 && decorr_left_weight) {
367 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
368 decorr_shift, decorr_left_weight);
371 if (alac->extra_bits) {
372 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
373 alac->extra_bits, channels, alac->nb_samples);
376 switch(alac->sample_size) {
378 for (ch = 0; ch < channels; ch++) {
379 int16_t *outbuffer = (int16_t *)frame->extended_data[ch_index + ch];
380 for (i = 0; i < alac->nb_samples; i++)
381 *outbuffer++ = alac->output_samples_buffer[ch][i];
385 for (ch = 0; ch < channels; ch++) {
386 for (i = 0; i < alac->nb_samples; i++)
387 alac->output_samples_buffer[ch][i] <<= 8;
395 static int alac_decode_frame(AVCodecContext *avctx, void *data,
396 int *got_frame_ptr, AVPacket *avpkt)
398 ALACContext *alac = avctx->priv_data;
399 AVFrame *frame = data;
400 enum AlacRawDataBlockType element;
402 int ch, ret, got_end;
404 init_get_bits(&alac->gb, avpkt->data, avpkt->size * 8);
407 alac->nb_samples = 0;
409 while (get_bits_left(&alac->gb) >= 3) {
410 element = get_bits(&alac->gb, 3);
411 if (element == TYPE_END) {
415 if (element > TYPE_CPE && element != TYPE_LFE) {
416 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d", element);
417 return AVERROR_PATCHWELCOME;
420 channels = (element == TYPE_CPE) ? 2 : 1;
421 if (ch + channels > alac->channels) {
422 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
423 return AVERROR_INVALIDDATA;
426 ret = decode_element(avctx, frame,
427 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
429 if (ret < 0 && get_bits_left(&alac->gb))
435 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
436 return AVERROR_INVALIDDATA;
439 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
440 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
441 avpkt->size * 8 - get_bits_count(&alac->gb));
449 static av_cold int alac_decode_close(AVCodecContext *avctx)
451 ALACContext *alac = avctx->priv_data;
454 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
455 av_freep(&alac->predict_error_buffer[ch]);
456 if (alac->sample_size == 16)
457 av_freep(&alac->output_samples_buffer[ch]);
458 av_freep(&alac->extra_bits_buffer[ch]);
464 static int allocate_buffers(ALACContext *alac)
467 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
469 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
470 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
471 buf_size, buf_alloc_fail);
473 if (alac->sample_size == 16) {
474 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
475 buf_size, buf_alloc_fail);
478 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
479 buf_size, buf_alloc_fail);
483 alac_decode_close(alac->avctx);
484 return AVERROR(ENOMEM);
487 static int alac_set_info(ALACContext *alac)
491 bytestream2_init(&gb, alac->avctx->extradata,
492 alac->avctx->extradata_size);
494 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
496 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
497 if (!alac->max_samples_per_frame || alac->max_samples_per_frame > INT_MAX) {
498 av_log(alac->avctx, AV_LOG_ERROR, "max samples per frame invalid: %u\n",
499 alac->max_samples_per_frame);
500 return AVERROR_INVALIDDATA;
502 bytestream2_skipu(&gb, 1); // compatible version
503 alac->sample_size = bytestream2_get_byteu(&gb);
504 alac->rice_history_mult = bytestream2_get_byteu(&gb);
505 alac->rice_initial_history = bytestream2_get_byteu(&gb);
506 alac->rice_limit = bytestream2_get_byteu(&gb);
507 alac->channels = bytestream2_get_byteu(&gb);
508 bytestream2_get_be16u(&gb); // maxRun
509 bytestream2_get_be32u(&gb); // max coded frame size
510 bytestream2_get_be32u(&gb); // average bitrate
511 bytestream2_get_be32u(&gb); // samplerate
516 static av_cold int alac_decode_init(AVCodecContext * avctx)
519 ALACContext *alac = avctx->priv_data;
522 /* initialize from the extradata */
523 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
524 av_log(avctx, AV_LOG_ERROR, "alac: extradata is too small\n");
525 return AVERROR_INVALIDDATA;
527 if (alac_set_info(alac)) {
528 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
532 switch (alac->sample_size) {
533 case 16: avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
536 case 32: avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
538 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
539 return AVERROR_PATCHWELCOME;
541 avctx->bits_per_raw_sample = alac->sample_size;
543 if (alac->channels < 1) {
544 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
545 alac->channels = avctx->channels;
547 if (alac->channels > ALAC_MAX_CHANNELS)
548 alac->channels = avctx->channels;
550 avctx->channels = alac->channels;
552 if (avctx->channels > ALAC_MAX_CHANNELS) {
553 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
555 return AVERROR_PATCHWELCOME;
557 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
559 if ((ret = allocate_buffers(alac)) < 0) {
560 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
567 AVCodec ff_alac_decoder = {
569 .type = AVMEDIA_TYPE_AUDIO,
570 .id = AV_CODEC_ID_ALAC,
571 .priv_data_size = sizeof(ALACContext),
572 .init = alac_decode_init,
573 .close = alac_decode_close,
574 .decode = alac_decode_frame,
575 .capabilities = CODEC_CAP_DR1,
576 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),