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;
65 int32_t *predict_error_buffer[2];
66 int32_t *output_samples_buffer[2];
67 int32_t *extra_bits_buffer[2];
69 uint32_t max_samples_per_frame;
71 uint8_t rice_history_mult;
72 uint8_t rice_initial_history;
75 int extra_bits; /**< number of extra bits beyond 16-bit */
76 int nb_samples; /**< number of samples in the current frame */
79 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
81 unsigned int x = get_unary_0_9(gb);
83 if (x > 8) { /* RICE THRESHOLD */
84 /* use alternative encoding */
85 x = get_bits_long(gb, bps);
87 int extrabits = show_bits(gb, k);
89 /* multiply x by 2^k - 1, as part of their strange algorithm */
101 static void rice_decompress(ALACContext *alac, int32_t *output_buffer,
102 int nb_samples, int bps, int rice_history_mult)
105 unsigned int history = alac->rice_initial_history;
106 int sign_modifier = 0;
108 for (i = 0; i < nb_samples; i++) {
112 /* calculate rice param and decode next value */
113 k = av_log2((history >> 9) + 3);
114 k = FFMIN(k, alac->rice_limit);
115 x = decode_scalar(&alac->gb, k, bps);
118 output_buffer[i] = (x >> 1) ^ -(x & 1);
120 /* update the history */
124 history += x * rice_history_mult -
125 ((history * rice_history_mult) >> 9);
127 /* special case: there may be compressed blocks of 0 */
128 if ((history < 128) && (i + 1 < nb_samples)) {
131 /* calculate rice param and decode block size */
132 k = 7 - av_log2(history) + ((history + 16) >> 6);
133 k = FFMIN(k, alac->rice_limit);
134 block_size = decode_scalar(&alac->gb, k, 16);
136 if (block_size > 0) {
137 if (block_size >= nb_samples - i) {
138 av_log(alac->avctx, AV_LOG_ERROR,
139 "invalid zero block size of %d %d %d\n", block_size,
141 block_size = nb_samples - i - 1;
143 memset(&output_buffer[i + 1], 0,
144 block_size * sizeof(*output_buffer));
147 if (block_size <= 0xffff)
154 static inline int sign_only(int v)
156 return v ? FFSIGN(v) : 0;
159 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
160 int nb_samples, int bps, int16_t *lpc_coefs,
161 int lpc_order, int lpc_quant)
164 int32_t *pred = buffer_out;
166 /* first sample always copies */
167 *buffer_out = *error_buffer;
173 memcpy(&buffer_out[1], &error_buffer[1],
174 (nb_samples - 1) * sizeof(*buffer_out));
178 if (lpc_order == 31) {
179 /* simple 1st-order prediction */
180 for (i = 1; i < nb_samples; i++) {
181 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
187 /* read warm-up samples */
188 for (i = 1; i <= lpc_order; i++)
189 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
191 /* NOTE: 4 and 8 are very common cases that could be optimized. */
193 for (; i < nb_samples; i++) {
196 int error_val = error_buffer[i];
201 for (j = 0; j < lpc_order; j++)
202 val += (pred[j] - d) * lpc_coefs[j];
203 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
204 val += d + error_val;
205 buffer_out[i] = sign_extend(val, bps);
207 /* adapt LPC coefficients */
208 error_sign = sign_only(error_val);
210 for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {
213 sign = sign_only(val) * error_sign;
214 lpc_coefs[j] -= sign;
216 error_val -= (val >> lpc_quant) * (j + 1);
222 static void decorrelate_stereo(int32_t *buffer[2], int nb_samples,
223 int decorr_shift, int decorr_left_weight)
227 for (i = 0; i < nb_samples; i++) {
233 a -= (b * decorr_left_weight) >> decorr_shift;
241 static void append_extra_bits(int32_t *buffer[2], int32_t *extra_bits_buffer[2],
242 int extra_bits, int channels, int nb_samples)
246 for (ch = 0; ch < channels; ch++)
247 for (i = 0; i < nb_samples; i++)
248 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
251 static int decode_element(AVCodecContext *avctx, void *data, int ch_index,
254 ALACContext *alac = avctx->priv_data;
255 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
256 uint32_t output_samples;
259 skip_bits(&alac->gb, 4); /* element instance tag */
260 skip_bits(&alac->gb, 12); /* unused header bits */
262 /* the number of output samples is stored in the frame */
263 has_size = get_bits1(&alac->gb);
265 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
266 bps = alac->sample_size - alac->extra_bits + channels - 1;
268 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
269 return AVERROR_PATCHWELCOME;
272 /* whether the frame is compressed */
273 is_compressed = !get_bits1(&alac->gb);
276 output_samples = get_bits_long(&alac->gb, 32);
278 output_samples = alac->max_samples_per_frame;
279 if (!output_samples || output_samples > alac->max_samples_per_frame) {
280 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %d\n",
282 return AVERROR_INVALIDDATA;
284 if (!alac->nb_samples) {
285 /* get output buffer */
286 alac->frame.nb_samples = output_samples;
287 if ((ret = ff_get_buffer(avctx, &alac->frame)) < 0) {
288 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
291 } else if (output_samples != alac->nb_samples) {
292 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %u != %d\n",
293 output_samples, alac->nb_samples);
294 return AVERROR_INVALIDDATA;
296 alac->nb_samples = output_samples;
297 if (alac->sample_size > 16) {
298 for (ch = 0; ch < channels; ch++)
299 alac->output_samples_buffer[ch] = (int32_t *)alac->frame.extended_data[ch_index + ch];
303 int16_t lpc_coefs[2][32];
305 int prediction_type[2];
307 int rice_history_mult[2];
309 decorr_shift = get_bits(&alac->gb, 8);
310 decorr_left_weight = get_bits(&alac->gb, 8);
312 for (ch = 0; ch < channels; ch++) {
313 prediction_type[ch] = get_bits(&alac->gb, 4);
314 lpc_quant[ch] = get_bits(&alac->gb, 4);
315 rice_history_mult[ch] = get_bits(&alac->gb, 3);
316 lpc_order[ch] = get_bits(&alac->gb, 5);
318 /* read the predictor table */
319 for (i = lpc_order[ch] - 1; i >= 0; i--)
320 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
323 if (alac->extra_bits) {
324 for (i = 0; i < alac->nb_samples; i++) {
325 for (ch = 0; ch < channels; ch++)
326 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
329 for (ch = 0; ch < channels; ch++) {
330 rice_decompress(alac, alac->predict_error_buffer[ch],
331 alac->nb_samples, bps,
332 rice_history_mult[ch] * alac->rice_history_mult / 4);
334 /* adaptive FIR filter */
335 if (prediction_type[ch] == 15) {
336 /* Prediction type 15 runs the adaptive FIR twice.
337 * The first pass uses the special-case coef_num = 31, while
338 * the second pass uses the coefs from the bitstream.
340 * However, this prediction type is not currently used by the
343 lpc_prediction(alac->predict_error_buffer[ch],
344 alac->predict_error_buffer[ch],
345 alac->nb_samples, bps, NULL, 31, 0);
346 } else if (prediction_type[ch] > 0) {
347 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
348 prediction_type[ch]);
350 lpc_prediction(alac->predict_error_buffer[ch],
351 alac->output_samples_buffer[ch], alac->nb_samples,
352 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
355 /* not compressed, easy case */
356 for (i = 0; i < alac->nb_samples; i++) {
357 for (ch = 0; ch < channels; ch++) {
358 alac->output_samples_buffer[ch][i] =
359 get_sbits_long(&alac->gb, alac->sample_size);
362 alac->extra_bits = 0;
364 decorr_left_weight = 0;
367 if (channels == 2 && decorr_left_weight) {
368 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
369 decorr_shift, decorr_left_weight);
372 if (alac->extra_bits) {
373 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
374 alac->extra_bits, channels, alac->nb_samples);
377 switch(alac->sample_size) {
379 for (ch = 0; ch < channels; ch++) {
380 int16_t *outbuffer = (int16_t *)alac->frame.extended_data[ch_index + ch];
381 for (i = 0; i < alac->nb_samples; i++)
382 *outbuffer++ = alac->output_samples_buffer[ch][i];
386 for (ch = 0; ch < channels; ch++) {
387 for (i = 0; i < alac->nb_samples; i++)
388 alac->output_samples_buffer[ch][i] <<= 8;
396 static int alac_decode_frame(AVCodecContext *avctx, void *data,
397 int *got_frame_ptr, AVPacket *avpkt)
399 ALACContext *alac = avctx->priv_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, data,
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));
445 *(AVFrame *)data = alac->frame;
450 static av_cold int alac_decode_close(AVCodecContext *avctx)
452 ALACContext *alac = avctx->priv_data;
455 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
456 av_freep(&alac->predict_error_buffer[ch]);
457 if (alac->sample_size == 16)
458 av_freep(&alac->output_samples_buffer[ch]);
459 av_freep(&alac->extra_bits_buffer[ch]);
465 static int allocate_buffers(ALACContext *alac)
468 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
470 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
471 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
472 buf_size, buf_alloc_fail);
474 if (alac->sample_size == 16) {
475 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
476 buf_size, buf_alloc_fail);
479 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
480 buf_size, buf_alloc_fail);
484 alac_decode_close(alac->avctx);
485 return AVERROR(ENOMEM);
488 static int alac_set_info(ALACContext *alac)
492 bytestream2_init(&gb, alac->avctx->extradata,
493 alac->avctx->extradata_size);
495 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
497 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
498 if (!alac->max_samples_per_frame || alac->max_samples_per_frame > INT_MAX) {
499 av_log(alac->avctx, AV_LOG_ERROR, "max samples per frame invalid: %u\n",
500 alac->max_samples_per_frame);
501 return AVERROR_INVALIDDATA;
503 bytestream2_skipu(&gb, 1); // compatible version
504 alac->sample_size = bytestream2_get_byteu(&gb);
505 alac->rice_history_mult = bytestream2_get_byteu(&gb);
506 alac->rice_initial_history = bytestream2_get_byteu(&gb);
507 alac->rice_limit = bytestream2_get_byteu(&gb);
508 alac->channels = bytestream2_get_byteu(&gb);
509 bytestream2_get_be16u(&gb); // maxRun
510 bytestream2_get_be32u(&gb); // max coded frame size
511 bytestream2_get_be32u(&gb); // average bitrate
512 bytestream2_get_be32u(&gb); // samplerate
517 static av_cold int alac_decode_init(AVCodecContext * avctx)
520 ALACContext *alac = avctx->priv_data;
523 /* initialize from the extradata */
524 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
525 av_log(avctx, AV_LOG_ERROR, "alac: extradata is too small\n");
526 return AVERROR_INVALIDDATA;
528 if (alac_set_info(alac)) {
529 av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
533 switch (alac->sample_size) {
534 case 16: avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
537 case 32: avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
539 default: av_log_ask_for_sample(avctx, "Sample depth %d is not supported.\n",
541 return AVERROR_PATCHWELCOME;
543 avctx->bits_per_raw_sample = alac->sample_size;
545 if (alac->channels < 1) {
546 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
547 alac->channels = avctx->channels;
549 if (alac->channels > ALAC_MAX_CHANNELS)
550 alac->channels = avctx->channels;
552 avctx->channels = alac->channels;
554 if (avctx->channels > ALAC_MAX_CHANNELS) {
555 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
557 return AVERROR_PATCHWELCOME;
559 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
561 if ((ret = allocate_buffers(alac)) < 0) {
562 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
566 avcodec_get_frame_defaults(&alac->frame);
567 avctx->coded_frame = &alac->frame;
572 AVCodec ff_alac_decoder = {
574 .type = AVMEDIA_TYPE_AUDIO,
575 .id = AV_CODEC_ID_ALAC,
576 .priv_data_size = sizeof(ALACContext),
577 .init = alac_decode_init,
578 .close = alac_decode_close,
579 .decode = alac_decode_frame,
580 .capabilities = CODEC_CAP_DR1,
581 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),