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)
50 #include "libavutil/channel_layout.h"
53 #include "bytestream.h"
57 #include "alac_data.h"
59 #define ALAC_EXTRADATA_SIZE 36
62 AVCodecContext *avctx;
66 int32_t *predict_error_buffer[2];
67 int32_t *output_samples_buffer[2];
68 int32_t *extra_bits_buffer[2];
70 uint32_t max_samples_per_frame;
72 uint8_t rice_history_mult;
73 uint8_t rice_initial_history;
76 int extra_bits; /**< number of extra bits beyond 16-bit */
77 int nb_samples; /**< number of samples in the current frame */
80 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
82 unsigned int x = get_unary_0_9(gb);
84 if (x > 8) { /* RICE THRESHOLD */
85 /* use alternative encoding */
86 x = get_bits_long(gb, bps);
88 int extrabits = show_bits(gb, k);
90 /* multiply x by 2^k - 1, as part of their strange algorithm */
102 static void rice_decompress(ALACContext *alac, int32_t *output_buffer,
103 int nb_samples, int bps, int rice_history_mult)
106 unsigned int history = alac->rice_initial_history;
107 int sign_modifier = 0;
109 for (i = 0; i < nb_samples; i++) {
113 /* calculate rice param and decode next value */
114 k = av_log2((history >> 9) + 3);
115 k = FFMIN(k, alac->rice_limit);
116 x = decode_scalar(&alac->gb, k, bps);
119 output_buffer[i] = (x >> 1) ^ -(x & 1);
121 /* update the history */
125 history += x * rice_history_mult -
126 ((history * rice_history_mult) >> 9);
128 /* special case: there may be compressed blocks of 0 */
129 if ((history < 128) && (i + 1 < nb_samples)) {
132 /* calculate rice param and decode block size */
133 k = 7 - av_log2(history) + ((history + 16) >> 6);
134 k = FFMIN(k, alac->rice_limit);
135 block_size = decode_scalar(&alac->gb, k, 16);
137 if (block_size > 0) {
138 if (block_size >= nb_samples - i) {
139 av_log(alac->avctx, AV_LOG_ERROR,
140 "invalid zero block size of %d %d %d\n", block_size,
142 block_size = nb_samples - i - 1;
144 memset(&output_buffer[i + 1], 0,
145 block_size * sizeof(*output_buffer));
148 if (block_size <= 0xffff)
155 static inline int sign_only(int v)
157 return v ? FFSIGN(v) : 0;
160 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
161 int nb_samples, int bps, int16_t *lpc_coefs,
162 int lpc_order, int lpc_quant)
165 int32_t *pred = buffer_out;
167 /* first sample always copies */
168 *buffer_out = *error_buffer;
174 memcpy(&buffer_out[1], &error_buffer[1],
175 (nb_samples - 1) * sizeof(*buffer_out));
179 if (lpc_order == 31) {
180 /* simple 1st-order prediction */
181 for (i = 1; i < nb_samples; i++) {
182 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
188 /* read warm-up samples */
189 for (i = 1; i <= lpc_order; i++)
190 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
192 /* NOTE: 4 and 8 are very common cases that could be optimized. */
194 for (; i < nb_samples; i++) {
197 int error_val = error_buffer[i];
202 for (j = 0; j < lpc_order; j++)
203 val += (pred[j] - d) * lpc_coefs[j];
204 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
205 val += d + error_val;
206 buffer_out[i] = sign_extend(val, bps);
208 /* adapt LPC coefficients */
209 error_sign = sign_only(error_val);
211 for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {
214 sign = sign_only(val) * error_sign;
215 lpc_coefs[j] -= sign;
217 error_val -= (val >> lpc_quant) * (j + 1);
223 static void decorrelate_stereo(int32_t *buffer[2], int nb_samples,
224 int decorr_shift, int decorr_left_weight)
228 for (i = 0; i < nb_samples; i++) {
234 a -= (b * decorr_left_weight) >> decorr_shift;
242 static void append_extra_bits(int32_t *buffer[2], int32_t *extra_bits_buffer[2],
243 int extra_bits, int channels, int nb_samples)
247 for (ch = 0; ch < channels; ch++)
248 for (i = 0; i < nb_samples; i++)
249 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
252 static int decode_element(AVCodecContext *avctx, AVFrame *frame, int ch_index,
255 ALACContext *alac = avctx->priv_data;
256 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
257 uint32_t output_samples;
260 skip_bits(&alac->gb, 4); /* element instance tag */
261 skip_bits(&alac->gb, 12); /* unused header bits */
263 /* the number of output samples is stored in the frame */
264 has_size = get_bits1(&alac->gb);
266 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
267 bps = alac->sample_size - alac->extra_bits + channels - 1;
269 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
270 return AVERROR_PATCHWELCOME;
273 /* whether the frame is compressed */
274 is_compressed = !get_bits1(&alac->gb);
277 output_samples = get_bits_long(&alac->gb, 32);
279 output_samples = alac->max_samples_per_frame;
280 if (!output_samples || output_samples > alac->max_samples_per_frame) {
281 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %"PRIu32"\n",
283 return AVERROR_INVALIDDATA;
285 if (!alac->nb_samples) {
286 /* get output buffer */
287 frame->nb_samples = output_samples;
288 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
289 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
292 } else if (output_samples != alac->nb_samples) {
293 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %"PRIu32" != %d\n",
294 output_samples, alac->nb_samples);
295 return AVERROR_INVALIDDATA;
297 alac->nb_samples = output_samples;
298 if (alac->sample_size > 16) {
299 for (ch = 0; ch < channels; ch++)
300 alac->output_samples_buffer[ch] = (int32_t *)frame->extended_data[ch_index + ch];
304 int16_t lpc_coefs[2][32];
306 int prediction_type[2];
308 int rice_history_mult[2];
310 decorr_shift = get_bits(&alac->gb, 8);
311 decorr_left_weight = get_bits(&alac->gb, 8);
313 for (ch = 0; ch < channels; ch++) {
314 prediction_type[ch] = get_bits(&alac->gb, 4);
315 lpc_quant[ch] = get_bits(&alac->gb, 4);
316 rice_history_mult[ch] = get_bits(&alac->gb, 3);
317 lpc_order[ch] = get_bits(&alac->gb, 5);
319 if (lpc_order[ch] >= alac->max_samples_per_frame)
320 return AVERROR_INVALIDDATA;
322 /* read the predictor table */
323 for (i = lpc_order[ch] - 1; i >= 0; i--)
324 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
327 if (alac->extra_bits) {
328 for (i = 0; i < alac->nb_samples; i++) {
329 for (ch = 0; ch < channels; ch++)
330 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
333 for (ch = 0; ch < channels; ch++) {
334 rice_decompress(alac, alac->predict_error_buffer[ch],
335 alac->nb_samples, bps,
336 rice_history_mult[ch] * alac->rice_history_mult / 4);
338 /* adaptive FIR filter */
339 if (prediction_type[ch] == 15) {
340 /* Prediction type 15 runs the adaptive FIR twice.
341 * The first pass uses the special-case coef_num = 31, while
342 * the second pass uses the coefs from the bitstream.
344 * However, this prediction type is not currently used by the
347 lpc_prediction(alac->predict_error_buffer[ch],
348 alac->predict_error_buffer[ch],
349 alac->nb_samples, bps, NULL, 31, 0);
350 } else if (prediction_type[ch] > 0) {
351 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
352 prediction_type[ch]);
354 lpc_prediction(alac->predict_error_buffer[ch],
355 alac->output_samples_buffer[ch], alac->nb_samples,
356 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
359 /* not compressed, easy case */
360 for (i = 0; i < alac->nb_samples; i++) {
361 for (ch = 0; ch < channels; ch++) {
362 alac->output_samples_buffer[ch][i] =
363 get_sbits_long(&alac->gb, alac->sample_size);
366 alac->extra_bits = 0;
368 decorr_left_weight = 0;
371 if (channels == 2 && decorr_left_weight) {
372 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
373 decorr_shift, decorr_left_weight);
376 if (alac->extra_bits) {
377 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
378 alac->extra_bits, channels, alac->nb_samples);
381 switch(alac->sample_size) {
383 for (ch = 0; ch < channels; ch++) {
384 int16_t *outbuffer = (int16_t *)frame->extended_data[ch_index + ch];
385 for (i = 0; i < alac->nb_samples; i++)
386 *outbuffer++ = alac->output_samples_buffer[ch][i];
390 for (ch = 0; ch < channels; ch++) {
391 for (i = 0; i < alac->nb_samples; i++)
392 alac->output_samples_buffer[ch][i] <<= 8;
400 static int alac_decode_frame(AVCodecContext *avctx, void *data,
401 int *got_frame_ptr, AVPacket *avpkt)
403 ALACContext *alac = avctx->priv_data;
404 AVFrame *frame = data;
405 enum AlacRawDataBlockType element;
407 int ch, ret, got_end;
409 init_get_bits(&alac->gb, avpkt->data, avpkt->size * 8);
412 alac->nb_samples = 0;
414 while (get_bits_left(&alac->gb) >= 3) {
415 element = get_bits(&alac->gb, 3);
416 if (element == TYPE_END) {
420 if (element > TYPE_CPE && element != TYPE_LFE) {
421 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d", element);
422 return AVERROR_PATCHWELCOME;
425 channels = (element == TYPE_CPE) ? 2 : 1;
426 if (ch + channels > alac->channels ||
427 ff_alac_channel_layout_offsets[alac->channels - 1][ch] + channels > alac->channels) {
428 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
429 return AVERROR_INVALIDDATA;
432 ret = decode_element(avctx, frame,
433 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
435 if (ret < 0 && get_bits_left(&alac->gb))
441 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
442 return AVERROR_INVALIDDATA;
445 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
446 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
447 avpkt->size * 8 - get_bits_count(&alac->gb));
455 static av_cold int alac_decode_close(AVCodecContext *avctx)
457 ALACContext *alac = avctx->priv_data;
460 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
461 av_freep(&alac->predict_error_buffer[ch]);
462 if (alac->sample_size == 16)
463 av_freep(&alac->output_samples_buffer[ch]);
464 av_freep(&alac->extra_bits_buffer[ch]);
470 static int allocate_buffers(ALACContext *alac)
473 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
475 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
476 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
477 buf_size, buf_alloc_fail);
479 if (alac->sample_size == 16) {
480 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
481 buf_size, buf_alloc_fail);
484 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
485 buf_size, buf_alloc_fail);
489 alac_decode_close(alac->avctx);
490 return AVERROR(ENOMEM);
493 static int alac_set_info(ALACContext *alac)
497 bytestream2_init(&gb, alac->avctx->extradata,
498 alac->avctx->extradata_size);
500 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
502 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
503 if (!alac->max_samples_per_frame ||
504 alac->max_samples_per_frame > INT_MAX / sizeof(int32_t)) {
505 av_log(alac->avctx, AV_LOG_ERROR,
506 "max samples per frame invalid: %"PRIu32"\n",
507 alac->max_samples_per_frame);
508 return AVERROR_INVALIDDATA;
510 bytestream2_skipu(&gb, 1); // compatible version
511 alac->sample_size = bytestream2_get_byteu(&gb);
512 alac->rice_history_mult = bytestream2_get_byteu(&gb);
513 alac->rice_initial_history = bytestream2_get_byteu(&gb);
514 alac->rice_limit = bytestream2_get_byteu(&gb);
515 alac->channels = bytestream2_get_byteu(&gb);
516 bytestream2_get_be16u(&gb); // maxRun
517 bytestream2_get_be32u(&gb); // max coded frame size
518 bytestream2_get_be32u(&gb); // average bitrate
519 bytestream2_get_be32u(&gb); // samplerate
524 static av_cold int alac_decode_init(AVCodecContext * avctx)
527 ALACContext *alac = avctx->priv_data;
530 /* initialize from the extradata */
531 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
532 av_log(avctx, AV_LOG_ERROR, "alac: extradata is too small\n");
533 return AVERROR_INVALIDDATA;
535 if (alac_set_info(alac)) {
536 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
540 switch (alac->sample_size) {
541 case 16: avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
544 case 32: avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
546 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
547 return AVERROR_PATCHWELCOME;
549 avctx->bits_per_raw_sample = alac->sample_size;
551 if (alac->channels < 1) {
552 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
553 alac->channels = avctx->channels;
555 if (alac->channels > ALAC_MAX_CHANNELS)
556 alac->channels = avctx->channels;
558 avctx->channels = alac->channels;
560 if (avctx->channels > ALAC_MAX_CHANNELS) {
561 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
563 return AVERROR_PATCHWELCOME;
565 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
567 if ((ret = allocate_buffers(alac)) < 0) {
568 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
575 AVCodec ff_alac_decoder = {
577 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
578 .type = AVMEDIA_TYPE_AUDIO,
579 .id = AV_CODEC_ID_ALAC,
580 .priv_data_size = sizeof(ALACContext),
581 .init = alac_decode_init,
582 .close = alac_decode_close,
583 .decode = alac_decode_frame,
584 .capabilities = CODEC_CAP_DR1,