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-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 (10)
40 * 8bit rice param limit (14)
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 */
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 int 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 if(get_bits_left(&alac->gb) <= 0)
116 /* calculate rice param and decode next value */
117 k = av_log2((history >> 9) + 3);
118 k = FFMIN(k, alac->rice_limit);
119 x = decode_scalar(&alac->gb, k, bps);
122 output_buffer[i] = (x >> 1) ^ -(x & 1);
124 /* update the history */
128 history += x * rice_history_mult -
129 ((history * rice_history_mult) >> 9);
131 /* special case: there may be compressed blocks of 0 */
132 if ((history < 128) && (i + 1 < nb_samples)) {
135 /* calculate rice param and decode block size */
136 k = 7 - av_log2(history) + ((history + 16) >> 6);
137 k = FFMIN(k, alac->rice_limit);
138 block_size = decode_scalar(&alac->gb, k, 16);
140 if (block_size > 0) {
141 if (block_size >= nb_samples - i) {
142 av_log(alac->avctx, AV_LOG_ERROR,
143 "invalid zero block size of %d %d %d\n", block_size,
145 block_size = nb_samples - i - 1;
147 memset(&output_buffer[i + 1], 0,
148 block_size * sizeof(*output_buffer));
151 if (block_size <= 0xffff)
159 static inline int sign_only(int v)
161 return v ? FFSIGN(v) : 0;
164 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
165 int nb_samples, int bps, int16_t *lpc_coefs,
166 int lpc_order, int lpc_quant)
169 int32_t *pred = buffer_out;
171 /* first sample always copies */
172 *buffer_out = *error_buffer;
178 memcpy(&buffer_out[1], &error_buffer[1],
179 (nb_samples - 1) * sizeof(*buffer_out));
183 if (lpc_order == 31) {
184 /* simple 1st-order prediction */
185 for (i = 1; i < nb_samples; i++) {
186 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
192 /* read warm-up samples */
193 for (i = 1; i <= lpc_order && i < nb_samples; i++)
194 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
196 /* NOTE: 4 and 8 are very common cases that could be optimized. */
198 for (; i < nb_samples; i++) {
201 int error_val = error_buffer[i];
206 for (j = 0; j < lpc_order; j++)
207 val += (pred[j] - d) * lpc_coefs[j];
208 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
209 val += d + error_val;
210 buffer_out[i] = sign_extend(val, bps);
212 /* adapt LPC coefficients */
213 error_sign = sign_only(error_val);
215 for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {
218 sign = sign_only(val) * error_sign;
219 lpc_coefs[j] -= sign;
221 error_val -= (val >> lpc_quant) * (j + 1);
227 static void decorrelate_stereo(int32_t *buffer[2], int nb_samples,
228 int decorr_shift, int decorr_left_weight)
232 for (i = 0; i < nb_samples; i++) {
238 a -= (b * decorr_left_weight) >> decorr_shift;
246 static void append_extra_bits(int32_t *buffer[2], int32_t *extra_bits_buffer[2],
247 int extra_bits, int channels, int nb_samples)
251 for (ch = 0; ch < channels; ch++)
252 for (i = 0; i < nb_samples; i++)
253 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
256 static int decode_element(AVCodecContext *avctx, AVFrame *frame, int ch_index,
259 ALACContext *alac = avctx->priv_data;
260 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
261 uint32_t output_samples;
264 skip_bits(&alac->gb, 4); /* element instance tag */
265 skip_bits(&alac->gb, 12); /* unused header bits */
267 /* the number of output samples is stored in the frame */
268 has_size = get_bits1(&alac->gb);
270 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
271 bps = alac->sample_size - alac->extra_bits + channels - 1;
273 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
274 return AVERROR_PATCHWELCOME;
277 /* whether the frame is compressed */
278 is_compressed = !get_bits1(&alac->gb);
281 output_samples = get_bits_long(&alac->gb, 32);
283 output_samples = alac->max_samples_per_frame;
284 if (!output_samples || output_samples > alac->max_samples_per_frame) {
285 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %d\n",
287 return AVERROR_INVALIDDATA;
289 if (!alac->nb_samples) {
290 /* get output buffer */
291 frame->nb_samples = output_samples;
292 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
294 } else if (output_samples != alac->nb_samples) {
295 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %u != %d\n",
296 output_samples, alac->nb_samples);
297 return AVERROR_INVALIDDATA;
299 alac->nb_samples = output_samples;
300 if (alac->direct_output) {
301 for (ch = 0; ch < channels; ch++)
302 alac->output_samples_buffer[ch] = (int32_t *)frame->extended_data[ch_index + ch];
306 int16_t lpc_coefs[2][32];
308 int prediction_type[2];
310 int rice_history_mult[2];
312 decorr_shift = get_bits(&alac->gb, 8);
313 decorr_left_weight = get_bits(&alac->gb, 8);
315 for (ch = 0; ch < channels; ch++) {
316 prediction_type[ch] = get_bits(&alac->gb, 4);
317 lpc_quant[ch] = get_bits(&alac->gb, 4);
318 rice_history_mult[ch] = get_bits(&alac->gb, 3);
319 lpc_order[ch] = get_bits(&alac->gb, 5);
321 /* read the predictor table */
322 for (i = lpc_order[ch] - 1; i >= 0; i--)
323 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
326 if (alac->extra_bits) {
327 for (i = 0; i < alac->nb_samples; i++) {
328 if(get_bits_left(&alac->gb) <= 0)
330 for (ch = 0; ch < channels; ch++)
331 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
334 for (ch = 0; ch < channels; ch++) {
335 int ret=rice_decompress(alac, alac->predict_error_buffer[ch],
336 alac->nb_samples, bps,
337 rice_history_mult[ch] * alac->rice_history_mult / 4);
341 /* adaptive FIR filter */
342 if (prediction_type[ch] == 15) {
343 /* Prediction type 15 runs the adaptive FIR twice.
344 * The first pass uses the special-case coef_num = 31, while
345 * the second pass uses the coefs from the bitstream.
347 * However, this prediction type is not currently used by the
350 lpc_prediction(alac->predict_error_buffer[ch],
351 alac->predict_error_buffer[ch],
352 alac->nb_samples, bps, NULL, 31, 0);
353 } else if (prediction_type[ch] > 0) {
354 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
355 prediction_type[ch]);
357 lpc_prediction(alac->predict_error_buffer[ch],
358 alac->output_samples_buffer[ch], alac->nb_samples,
359 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
362 /* not compressed, easy case */
363 for (i = 0; i < alac->nb_samples; i++) {
364 if(get_bits_left(&alac->gb) <= 0)
366 for (ch = 0; ch < channels; ch++) {
367 alac->output_samples_buffer[ch][i] =
368 get_sbits_long(&alac->gb, alac->sample_size);
371 alac->extra_bits = 0;
373 decorr_left_weight = 0;
376 if (channels == 2 && decorr_left_weight) {
377 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
378 decorr_shift, decorr_left_weight);
381 if (alac->extra_bits) {
382 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
383 alac->extra_bits, channels, alac->nb_samples);
386 if(av_sample_fmt_is_planar(avctx->sample_fmt)) {
387 switch(alac->sample_size) {
389 for (ch = 0; ch < channels; ch++) {
390 int16_t *outbuffer = (int16_t *)frame->extended_data[ch_index + ch];
391 for (i = 0; i < alac->nb_samples; i++)
392 *outbuffer++ = alac->output_samples_buffer[ch][i];
396 for (ch = 0; ch < channels; ch++) {
397 for (i = 0; i < alac->nb_samples; i++)
398 alac->output_samples_buffer[ch][i] <<= 8;
403 switch(alac->sample_size) {
405 int16_t *outbuffer = ((int16_t *)frame->extended_data[0]) + ch_index;
406 for (i = 0; i < alac->nb_samples; i++) {
407 for (ch = 0; ch < channels; ch++)
408 *outbuffer++ = alac->output_samples_buffer[ch][i];
409 outbuffer += alac->channels - channels;
414 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
415 for (i = 0; i < alac->nb_samples; i++) {
416 for (ch = 0; ch < channels; ch++)
417 *outbuffer++ = alac->output_samples_buffer[ch][i] << 8;
418 outbuffer += alac->channels - channels;
423 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
424 for (i = 0; i < alac->nb_samples; i++) {
425 for (ch = 0; ch < channels; ch++)
426 *outbuffer++ = alac->output_samples_buffer[ch][i];
427 outbuffer += alac->channels - channels;
437 static int alac_decode_frame(AVCodecContext *avctx, void *data,
438 int *got_frame_ptr, AVPacket *avpkt)
440 ALACContext *alac = avctx->priv_data;
441 AVFrame *frame = data;
442 enum AlacRawDataBlockType element;
444 int ch, ret, got_end;
446 init_get_bits(&alac->gb, avpkt->data, avpkt->size * 8);
449 alac->nb_samples = 0;
451 while (get_bits_left(&alac->gb) >= 3) {
452 element = get_bits(&alac->gb, 3);
453 if (element == TYPE_END) {
457 if (element > TYPE_CPE && element != TYPE_LFE) {
458 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d\n", element);
459 return AVERROR_PATCHWELCOME;
462 channels = (element == TYPE_CPE) ? 2 : 1;
463 if ( ch + channels > alac->channels
464 || ff_alac_channel_layout_offsets[alac->channels - 1][ch] + channels > alac->channels
466 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
467 return AVERROR_INVALIDDATA;
470 ret = decode_element(avctx, frame,
471 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
473 if (ret < 0 && get_bits_left(&alac->gb))
479 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
480 return AVERROR_INVALIDDATA;
483 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
484 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
485 avpkt->size * 8 - get_bits_count(&alac->gb));
493 static av_cold int alac_decode_close(AVCodecContext *avctx)
495 ALACContext *alac = avctx->priv_data;
498 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
499 av_freep(&alac->predict_error_buffer[ch]);
500 if (!alac->direct_output)
501 av_freep(&alac->output_samples_buffer[ch]);
502 av_freep(&alac->extra_bits_buffer[ch]);
508 static int allocate_buffers(ALACContext *alac)
513 if (alac->max_samples_per_frame > INT_MAX / sizeof(int32_t))
515 buf_size = alac->max_samples_per_frame * sizeof(int32_t);
517 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
518 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
519 buf_size, buf_alloc_fail);
521 alac->direct_output = alac->sample_size > 16 && av_sample_fmt_is_planar(alac->avctx->sample_fmt);
522 if (!alac->direct_output) {
523 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
524 buf_size, buf_alloc_fail);
527 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
528 buf_size, buf_alloc_fail);
532 alac_decode_close(alac->avctx);
533 return AVERROR(ENOMEM);
536 static int alac_set_info(ALACContext *alac)
540 bytestream2_init(&gb, alac->avctx->extradata,
541 alac->avctx->extradata_size);
543 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
545 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
546 if (!alac->max_samples_per_frame || alac->max_samples_per_frame > INT_MAX) {
547 av_log(alac->avctx, AV_LOG_ERROR, "max samples per frame invalid: %u\n",
548 alac->max_samples_per_frame);
549 return AVERROR_INVALIDDATA;
551 bytestream2_skipu(&gb, 1); // compatible version
552 alac->sample_size = bytestream2_get_byteu(&gb);
553 alac->rice_history_mult = bytestream2_get_byteu(&gb);
554 alac->rice_initial_history = bytestream2_get_byteu(&gb);
555 alac->rice_limit = bytestream2_get_byteu(&gb);
556 alac->channels = bytestream2_get_byteu(&gb);
557 bytestream2_get_be16u(&gb); // maxRun
558 bytestream2_get_be32u(&gb); // max coded frame size
559 bytestream2_get_be32u(&gb); // average bitrate
560 bytestream2_get_be32u(&gb); // samplerate
565 static av_cold int alac_decode_init(AVCodecContext * avctx)
569 ALACContext *alac = avctx->priv_data;
572 /* initialize from the extradata */
573 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
574 av_log(avctx, AV_LOG_ERROR, "extradata is too small\n");
575 return AVERROR_INVALIDDATA;
577 if (alac_set_info(alac)) {
578 av_log(avctx, AV_LOG_ERROR, "set_info failed\n");
582 req_packed = LIBAVCODEC_VERSION_MAJOR < 55 && !av_sample_fmt_is_planar(avctx->request_sample_fmt);
583 switch (alac->sample_size) {
584 case 16: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S16 : AV_SAMPLE_FMT_S16P;
587 case 32: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S32P;
589 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
590 return AVERROR_PATCHWELCOME;
592 avctx->bits_per_raw_sample = alac->sample_size;
594 if (alac->channels < 1) {
595 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
596 alac->channels = avctx->channels;
598 if (alac->channels > ALAC_MAX_CHANNELS)
599 alac->channels = avctx->channels;
601 avctx->channels = alac->channels;
603 if (avctx->channels > ALAC_MAX_CHANNELS || avctx->channels <= 0 ) {
604 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
606 return AVERROR_PATCHWELCOME;
608 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
610 if ((ret = allocate_buffers(alac)) < 0) {
611 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
618 AVCodec ff_alac_decoder = {
620 .type = AVMEDIA_TYPE_AUDIO,
621 .id = AV_CODEC_ID_ALAC,
622 .priv_data_size = sizeof(ALACContext),
623 .init = alac_decode_init,
624 .close = alac_decode_close,
625 .decode = alac_decode_frame,
626 .capabilities = CODEC_CAP_DR1,
627 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),