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:
33 * 32 bits tag ("alac")
34 * 32 bits tag version (0)
35 * 32 bits samples per frame (used when not set explicitly in the frames)
36 * 8 bits compatible version (0)
38 * 8 bits history mult (40)
39 * 8 bits initial history (10)
40 * 8 bits rice param limit (14)
42 * 16 bits maxRun (255)
43 * 32 bits max coded frame size (0 means unknown)
44 * 32 bits average bitrate (0 means unknown)
50 #include "libavutil/channel_layout.h"
51 #include "libavutil/opt.h"
54 #include "bytestream.h"
59 #include "alac_data.h"
62 #define ALAC_EXTRADATA_SIZE 36
64 typedef struct ALACContext {
66 AVCodecContext *avctx;
70 int32_t *predict_error_buffer[2];
71 int32_t *output_samples_buffer[2];
72 int32_t *extra_bits_buffer[2];
74 uint32_t max_samples_per_frame;
76 uint8_t rice_history_mult;
77 uint8_t rice_initial_history;
81 int extra_bits; /**< number of extra bits beyond 16-bit */
82 int nb_samples; /**< number of samples in the current frame */
90 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
92 unsigned int x = get_unary_0_9(gb);
94 if (x > 8) { /* RICE THRESHOLD */
95 /* use alternative encoding */
96 x = get_bits_long(gb, bps);
98 int extrabits = show_bits(gb, k);
100 /* multiply x by 2^k - 1, as part of their strange algorithm */
107 skip_bits(gb, k - 1);
112 static int rice_decompress(ALACContext *alac, int32_t *output_buffer,
113 int nb_samples, int bps, int rice_history_mult)
116 unsigned int history = alac->rice_initial_history;
117 int sign_modifier = 0;
119 for (i = 0; i < nb_samples; i++) {
123 if(get_bits_left(&alac->gb) <= 0)
124 return AVERROR_INVALIDDATA;
126 /* calculate rice param and decode next value */
127 k = av_log2((history >> 9) + 3);
128 k = FFMIN(k, alac->rice_limit);
129 x = decode_scalar(&alac->gb, k, bps);
132 output_buffer[i] = (x >> 1) ^ -(x & 1);
134 /* update the history */
138 history += x * rice_history_mult -
139 ((history * rice_history_mult) >> 9);
141 /* special case: there may be compressed blocks of 0 */
142 if ((history < 128) && (i + 1 < nb_samples)) {
145 /* calculate rice param and decode block size */
146 k = 7 - av_log2(history) + ((history + 16) >> 6);
147 k = FFMIN(k, alac->rice_limit);
148 block_size = decode_scalar(&alac->gb, k, 16);
150 if (block_size > 0) {
151 if (block_size >= nb_samples - i) {
152 av_log(alac->avctx, AV_LOG_ERROR,
153 "invalid zero block size of %d %d %d\n", block_size,
155 block_size = nb_samples - i - 1;
157 memset(&output_buffer[i + 1], 0,
158 block_size * sizeof(*output_buffer));
161 if (block_size <= 0xffff)
169 static inline int sign_only(int v)
171 return v ? FFSIGN(v) : 0;
174 static void lpc_prediction(int32_t *error_buffer, uint32_t *buffer_out,
175 int nb_samples, int bps, int16_t *lpc_coefs,
176 int lpc_order, int lpc_quant)
179 uint32_t *pred = buffer_out;
181 /* first sample always copies */
182 *buffer_out = *error_buffer;
188 memcpy(&buffer_out[1], &error_buffer[1],
189 (nb_samples - 1) * sizeof(*buffer_out));
193 if (lpc_order == 31) {
194 /* simple 1st-order prediction */
195 for (i = 1; i < nb_samples; i++) {
196 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
202 /* read warm-up samples */
203 for (i = 1; i <= lpc_order && i < nb_samples; i++)
204 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
206 /* NOTE: 4 and 8 are very common cases that could be optimized. */
208 for (; i < nb_samples; i++) {
211 unsigned error_val = error_buffer[i];
216 for (j = 0; j < lpc_order; j++)
217 val += (pred[j] - d) * lpc_coefs[j];
218 val = (val + (1LL << (lpc_quant - 1))) >> lpc_quant;
219 val += d + error_val;
220 buffer_out[i] = sign_extend(val, bps);
222 /* adapt LPC coefficients */
223 error_sign = sign_only(error_val);
225 for (j = 0; j < lpc_order && (int)(error_val * error_sign) > 0; j++) {
228 sign = sign_only(val) * error_sign;
229 lpc_coefs[j] -= sign;
230 val *= (unsigned)sign;
231 error_val -= (val >> lpc_quant) * (j + 1U);
237 static int decode_element(AVCodecContext *avctx, AVFrame *frame, int ch_index,
240 ALACContext *alac = avctx->priv_data;
241 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
242 uint32_t output_samples;
245 skip_bits(&alac->gb, 4); /* element instance tag */
246 skip_bits(&alac->gb, 12); /* unused header bits */
248 /* the number of output samples is stored in the frame */
249 has_size = get_bits1(&alac->gb);
251 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
252 bps = alac->sample_size - alac->extra_bits + channels - 1;
254 avpriv_report_missing_feature(avctx, "bps %d", bps);
255 return AVERROR_PATCHWELCOME;
258 return AVERROR_INVALIDDATA;
260 /* whether the frame is compressed */
261 is_compressed = !get_bits1(&alac->gb);
264 output_samples = get_bits_long(&alac->gb, 32);
266 output_samples = alac->max_samples_per_frame;
267 if (!output_samples || output_samples > alac->max_samples_per_frame) {
268 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %"PRIu32"\n",
270 return AVERROR_INVALIDDATA;
272 if (!alac->nb_samples) {
273 ThreadFrame tframe = { .f = frame };
274 /* get output buffer */
275 frame->nb_samples = output_samples;
276 if ((ret = ff_thread_get_buffer(avctx, &tframe, 0)) < 0)
278 } else if (output_samples != alac->nb_samples) {
279 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %"PRIu32" != %d\n",
280 output_samples, alac->nb_samples);
281 return AVERROR_INVALIDDATA;
283 alac->nb_samples = output_samples;
284 if (alac->direct_output) {
285 for (ch = 0; ch < channels; ch++)
286 alac->output_samples_buffer[ch] = (int32_t *)frame->extended_data[ch_index + ch];
290 int16_t lpc_coefs[2][32];
292 int prediction_type[2];
294 int rice_history_mult[2];
296 if (!alac->rice_limit) {
297 avpriv_request_sample(alac->avctx,
298 "Compression with rice limit 0");
299 return AVERROR(ENOSYS);
302 decorr_shift = get_bits(&alac->gb, 8);
303 decorr_left_weight = get_bits(&alac->gb, 8);
305 if (channels == 2 && decorr_left_weight && decorr_shift > 31)
306 return AVERROR_INVALIDDATA;
308 for (ch = 0; ch < channels; ch++) {
309 prediction_type[ch] = get_bits(&alac->gb, 4);
310 lpc_quant[ch] = get_bits(&alac->gb, 4);
311 rice_history_mult[ch] = get_bits(&alac->gb, 3);
312 lpc_order[ch] = get_bits(&alac->gb, 5);
314 if (lpc_order[ch] >= alac->max_samples_per_frame || !lpc_quant[ch])
315 return AVERROR_INVALIDDATA;
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 if(get_bits_left(&alac->gb) <= 0)
325 return AVERROR_INVALIDDATA;
326 for (ch = 0; ch < channels; ch++)
327 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
330 for (ch = 0; ch < channels; ch++) {
331 int ret=rice_decompress(alac, alac->predict_error_buffer[ch],
332 alac->nb_samples, bps,
333 rice_history_mult[ch] * alac->rice_history_mult / 4);
337 /* adaptive FIR filter */
338 if (prediction_type[ch] == 15) {
339 /* Prediction type 15 runs the adaptive FIR twice.
340 * The first pass uses the special-case coef_num = 31, while
341 * the second pass uses the coefs from the bitstream.
343 * However, this prediction type is not currently used by the
346 lpc_prediction(alac->predict_error_buffer[ch],
347 alac->predict_error_buffer[ch],
348 alac->nb_samples, bps, NULL, 31, 0);
349 } else if (prediction_type[ch] > 0) {
350 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
351 prediction_type[ch]);
353 lpc_prediction(alac->predict_error_buffer[ch],
354 alac->output_samples_buffer[ch], alac->nb_samples,
355 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
358 /* not compressed, easy case */
359 for (i = 0; i < alac->nb_samples; i++) {
360 if(get_bits_left(&alac->gb) <= 0)
361 return AVERROR_INVALIDDATA;
362 for (ch = 0; ch < channels; ch++) {
363 alac->output_samples_buffer[ch][i] =
364 get_sbits_long(&alac->gb, alac->sample_size);
367 alac->extra_bits = 0;
369 decorr_left_weight = 0;
373 if (alac->extra_bits && alac->extra_bit_bug) {
374 alac->dsp.append_extra_bits[1](alac->output_samples_buffer, alac->extra_bits_buffer,
375 alac->extra_bits, channels, alac->nb_samples);
378 if (decorr_left_weight) {
379 alac->dsp.decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
380 decorr_shift, decorr_left_weight);
383 if (alac->extra_bits && !alac->extra_bit_bug) {
384 alac->dsp.append_extra_bits[1](alac->output_samples_buffer, alac->extra_bits_buffer,
385 alac->extra_bits, channels, alac->nb_samples);
387 } else if (alac->extra_bits) {
388 alac->dsp.append_extra_bits[0](alac->output_samples_buffer, alac->extra_bits_buffer,
389 alac->extra_bits, channels, alac->nb_samples);
392 switch(alac->sample_size) {
394 for (ch = 0; ch < channels; ch++) {
395 int16_t *outbuffer = (int16_t *)frame->extended_data[ch_index + ch];
396 for (i = 0; i < alac->nb_samples; i++)
397 *outbuffer++ = alac->output_samples_buffer[ch][i];
401 for (ch = 0; ch < channels; ch++) {
402 for (i = 0; i < alac->nb_samples; i++)
403 alac->output_samples_buffer[ch][i] *= 1U << 12;
407 for (ch = 0; ch < channels; ch++) {
408 for (i = 0; i < alac->nb_samples; i++)
409 alac->output_samples_buffer[ch][i] *= 1U << 8;
417 static int alac_decode_frame(AVCodecContext *avctx, void *data,
418 int *got_frame_ptr, AVPacket *avpkt)
420 ALACContext *alac = avctx->priv_data;
421 AVFrame *frame = data;
422 enum AlacRawDataBlockType element;
424 int ch, ret, got_end;
426 if ((ret = init_get_bits8(&alac->gb, avpkt->data, avpkt->size)) < 0)
430 alac->nb_samples = 0;
432 while (get_bits_left(&alac->gb) >= 3) {
433 element = get_bits(&alac->gb, 3);
434 if (element == TYPE_END) {
438 if (element > TYPE_CPE && element != TYPE_LFE) {
439 avpriv_report_missing_feature(avctx, "Syntax element %d", element);
440 return AVERROR_PATCHWELCOME;
443 channels = (element == TYPE_CPE) ? 2 : 1;
444 if (ch + channels > alac->channels ||
445 ff_alac_channel_layout_offsets[alac->channels - 1][ch] + channels > alac->channels) {
446 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
447 return AVERROR_INVALIDDATA;
450 ret = decode_element(avctx, frame,
451 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
453 if (ret < 0 && get_bits_left(&alac->gb))
459 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
460 return AVERROR_INVALIDDATA;
463 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
464 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
465 avpkt->size * 8 - get_bits_count(&alac->gb));
468 if (alac->channels == ch && alac->nb_samples)
471 av_log(avctx, AV_LOG_WARNING, "Failed to decode all channels\n");
476 static av_cold int alac_decode_close(AVCodecContext *avctx)
478 ALACContext *alac = avctx->priv_data;
481 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
482 av_freep(&alac->predict_error_buffer[ch]);
483 if (!alac->direct_output)
484 av_freep(&alac->output_samples_buffer[ch]);
485 av_freep(&alac->extra_bits_buffer[ch]);
491 static int allocate_buffers(ALACContext *alac)
494 unsigned buf_size = alac->max_samples_per_frame * sizeof(int32_t);
495 unsigned extra_buf_size = buf_size + AV_INPUT_BUFFER_PADDING_SIZE;
497 for (ch = 0; ch < 2; ch++) {
498 alac->predict_error_buffer[ch] = NULL;
499 alac->output_samples_buffer[ch] = NULL;
500 alac->extra_bits_buffer[ch] = NULL;
503 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
504 if (!(alac->predict_error_buffer[ch] = av_malloc(buf_size)))
505 return AVERROR(ENOMEM);
507 alac->direct_output = alac->sample_size > 16;
508 if (!alac->direct_output) {
509 if (!(alac->output_samples_buffer[ch] = av_malloc(extra_buf_size)))
510 return AVERROR(ENOMEM);
513 if (!(alac->extra_bits_buffer[ch] = av_malloc(extra_buf_size)))
514 return AVERROR(ENOMEM);
519 static int alac_set_info(ALACContext *alac)
523 bytestream2_init(&gb, alac->avctx->extradata,
524 alac->avctx->extradata_size);
526 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
528 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
529 if (!alac->max_samples_per_frame ||
530 alac->max_samples_per_frame > 4096 * 4096) {
531 av_log(alac->avctx, AV_LOG_ERROR,
532 "max samples per frame invalid: %"PRIu32"\n",
533 alac->max_samples_per_frame);
534 return AVERROR_INVALIDDATA;
536 bytestream2_skipu(&gb, 1); // compatible version
537 alac->sample_size = bytestream2_get_byteu(&gb);
538 alac->rice_history_mult = bytestream2_get_byteu(&gb);
539 alac->rice_initial_history = bytestream2_get_byteu(&gb);
540 alac->rice_limit = bytestream2_get_byteu(&gb);
541 alac->channels = bytestream2_get_byteu(&gb);
542 bytestream2_get_be16u(&gb); // maxRun
543 bytestream2_get_be32u(&gb); // max coded frame size
544 bytestream2_get_be32u(&gb); // average bitrate
545 alac->sample_rate = bytestream2_get_be32u(&gb);
550 static av_cold int alac_decode_init(AVCodecContext * avctx)
553 ALACContext *alac = avctx->priv_data;
556 /* initialize from the extradata */
557 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
558 av_log(avctx, AV_LOG_ERROR, "extradata is too small\n");
559 return AVERROR_INVALIDDATA;
561 if ((ret = alac_set_info(alac)) < 0) {
562 av_log(avctx, AV_LOG_ERROR, "set_info failed\n");
566 switch (alac->sample_size) {
567 case 16: avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
571 case 32: avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
573 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
574 return AVERROR_PATCHWELCOME;
576 avctx->bits_per_raw_sample = alac->sample_size;
577 avctx->sample_rate = alac->sample_rate;
579 if (alac->channels < 1) {
580 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
581 alac->channels = avctx->channels;
583 if (alac->channels > ALAC_MAX_CHANNELS)
584 alac->channels = avctx->channels;
586 avctx->channels = alac->channels;
588 if (avctx->channels > ALAC_MAX_CHANNELS || avctx->channels <= 0 ) {
589 avpriv_report_missing_feature(avctx, "Channel count %d",
591 return AVERROR_PATCHWELCOME;
593 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
595 if ((ret = allocate_buffers(alac)) < 0) {
596 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
600 ff_alacdsp_init(&alac->dsp);
605 static const AVOption options[] = {
606 { "extra_bits_bug", "Force non-standard decoding process",
607 offsetof(ALACContext, extra_bit_bug), AV_OPT_TYPE_BOOL, { .i64 = 0 },
608 0, 1, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM },
612 static const AVClass alac_class = {
613 .class_name = "alac",
614 .item_name = av_default_item_name,
616 .version = LIBAVUTIL_VERSION_INT,
619 AVCodec ff_alac_decoder = {
621 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
622 .type = AVMEDIA_TYPE_AUDIO,
623 .id = AV_CODEC_ID_ALAC,
624 .priv_data_size = sizeof(ALACContext),
625 .init = alac_decode_init,
626 .close = alac_decode_close,
627 .decode = alac_decode_frame,
628 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,
629 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
630 .priv_class = &alac_class