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)
50 #include "libavutil/channel_layout.h"
51 #include "libavutil/opt.h"
54 #include "bytestream.h"
59 #include "alac_data.h"
61 #define ALAC_EXTRADATA_SIZE 36
65 AVCodecContext *avctx;
69 int32_t *predict_error_buffer[2];
70 int32_t *output_samples_buffer[2];
71 int32_t *extra_bits_buffer[2];
73 uint32_t max_samples_per_frame;
75 uint8_t rice_history_mult;
76 uint8_t rice_initial_history;
79 int extra_bits; /**< number of extra bits beyond 16-bit */
80 int nb_samples; /**< number of samples in the current frame */
86 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
88 unsigned int x = get_unary_0_9(gb);
90 if (x > 8) { /* RICE THRESHOLD */
91 /* use alternative encoding */
92 x = get_bits_long(gb, bps);
94 int extrabits = show_bits(gb, k);
96 /* multiply x by 2^k - 1, as part of their strange algorithm */
103 skip_bits(gb, k - 1);
108 static int rice_decompress(ALACContext *alac, int32_t *output_buffer,
109 int nb_samples, int bps, int rice_history_mult)
112 unsigned int history = alac->rice_initial_history;
113 int sign_modifier = 0;
115 for (i = 0; i < nb_samples; i++) {
119 if(get_bits_left(&alac->gb) <= 0)
122 /* calculate rice param and decode next value */
123 k = av_log2((history >> 9) + 3);
124 k = FFMIN(k, alac->rice_limit);
125 x = decode_scalar(&alac->gb, k, bps);
128 output_buffer[i] = (x >> 1) ^ -(x & 1);
130 /* update the history */
134 history += x * rice_history_mult -
135 ((history * rice_history_mult) >> 9);
137 /* special case: there may be compressed blocks of 0 */
138 if ((history < 128) && (i + 1 < nb_samples)) {
141 /* calculate rice param and decode block size */
142 k = 7 - av_log2(history) + ((history + 16) >> 6);
143 k = FFMIN(k, alac->rice_limit);
144 block_size = decode_scalar(&alac->gb, k, 16);
146 if (block_size > 0) {
147 if (block_size >= nb_samples - i) {
148 av_log(alac->avctx, AV_LOG_ERROR,
149 "invalid zero block size of %d %d %d\n", block_size,
151 block_size = nb_samples - i - 1;
153 memset(&output_buffer[i + 1], 0,
154 block_size * sizeof(*output_buffer));
157 if (block_size <= 0xffff)
165 static inline int sign_only(int v)
167 return v ? FFSIGN(v) : 0;
170 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
171 int nb_samples, int bps, int16_t *lpc_coefs,
172 int lpc_order, int lpc_quant)
175 int32_t *pred = buffer_out;
177 /* first sample always copies */
178 *buffer_out = *error_buffer;
184 memcpy(&buffer_out[1], &error_buffer[1],
185 (nb_samples - 1) * sizeof(*buffer_out));
189 if (lpc_order == 31) {
190 /* simple 1st-order prediction */
191 for (i = 1; i < nb_samples; i++) {
192 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
198 /* read warm-up samples */
199 for (i = 1; i <= lpc_order && i < nb_samples; i++)
200 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
202 /* NOTE: 4 and 8 are very common cases that could be optimized. */
204 for (; i < nb_samples; i++) {
207 int error_val = error_buffer[i];
212 for (j = 0; j < lpc_order; j++)
213 val += (pred[j] - d) * lpc_coefs[j];
214 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
215 val += d + error_val;
216 buffer_out[i] = sign_extend(val, bps);
218 /* adapt LPC coefficients */
219 error_sign = sign_only(error_val);
221 for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {
224 sign = sign_only(val) * error_sign;
225 lpc_coefs[j] -= sign;
227 error_val -= (val >> lpc_quant) * (j + 1);
233 static void decorrelate_stereo(int32_t *buffer[2], int nb_samples,
234 int decorr_shift, int decorr_left_weight)
238 for (i = 0; i < nb_samples; i++) {
244 a -= (b * decorr_left_weight) >> decorr_shift;
252 static void append_extra_bits(int32_t *buffer[2], int32_t *extra_bits_buffer[2],
253 int extra_bits, int channels, int nb_samples)
257 for (ch = 0; ch < channels; ch++)
258 for (i = 0; i < nb_samples; i++)
259 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
262 static int decode_element(AVCodecContext *avctx, AVFrame *frame, int ch_index,
265 ALACContext *alac = avctx->priv_data;
266 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
267 uint32_t output_samples;
270 skip_bits(&alac->gb, 4); /* element instance tag */
271 skip_bits(&alac->gb, 12); /* unused header bits */
273 /* the number of output samples is stored in the frame */
274 has_size = get_bits1(&alac->gb);
276 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
277 bps = alac->sample_size - alac->extra_bits + channels - 1;
279 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
280 return AVERROR_PATCHWELCOME;
283 /* whether the frame is compressed */
284 is_compressed = !get_bits1(&alac->gb);
287 output_samples = get_bits_long(&alac->gb, 32);
289 output_samples = alac->max_samples_per_frame;
290 if (!output_samples || output_samples > alac->max_samples_per_frame) {
291 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %"PRIu32"\n",
293 return AVERROR_INVALIDDATA;
295 if (!alac->nb_samples) {
296 ThreadFrame tframe = { .f = frame };
297 /* get output buffer */
298 frame->nb_samples = output_samples;
299 if ((ret = ff_thread_get_buffer(avctx, &tframe, 0)) < 0)
301 } else if (output_samples != alac->nb_samples) {
302 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %"PRIu32" != %d\n",
303 output_samples, alac->nb_samples);
304 return AVERROR_INVALIDDATA;
306 alac->nb_samples = output_samples;
307 if (alac->direct_output) {
308 for (ch = 0; ch < channels; ch++)
309 alac->output_samples_buffer[ch] = (int32_t *)frame->extended_data[ch_index + ch];
313 int16_t lpc_coefs[2][32];
315 int prediction_type[2];
317 int rice_history_mult[2];
319 decorr_shift = get_bits(&alac->gb, 8);
320 decorr_left_weight = get_bits(&alac->gb, 8);
322 for (ch = 0; ch < channels; ch++) {
323 prediction_type[ch] = get_bits(&alac->gb, 4);
324 lpc_quant[ch] = get_bits(&alac->gb, 4);
325 rice_history_mult[ch] = get_bits(&alac->gb, 3);
326 lpc_order[ch] = get_bits(&alac->gb, 5);
328 if (lpc_order[ch] >= alac->max_samples_per_frame)
329 return AVERROR_INVALIDDATA;
331 /* read the predictor table */
332 for (i = lpc_order[ch] - 1; i >= 0; i--)
333 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
336 if (alac->extra_bits) {
337 for (i = 0; i < alac->nb_samples; i++) {
338 if(get_bits_left(&alac->gb) <= 0)
340 for (ch = 0; ch < channels; ch++)
341 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
344 for (ch = 0; ch < channels; ch++) {
345 int ret=rice_decompress(alac, alac->predict_error_buffer[ch],
346 alac->nb_samples, bps,
347 rice_history_mult[ch] * alac->rice_history_mult / 4);
351 /* adaptive FIR filter */
352 if (prediction_type[ch] == 15) {
353 /* Prediction type 15 runs the adaptive FIR twice.
354 * The first pass uses the special-case coef_num = 31, while
355 * the second pass uses the coefs from the bitstream.
357 * However, this prediction type is not currently used by the
360 lpc_prediction(alac->predict_error_buffer[ch],
361 alac->predict_error_buffer[ch],
362 alac->nb_samples, bps, NULL, 31, 0);
363 } else if (prediction_type[ch] > 0) {
364 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
365 prediction_type[ch]);
367 lpc_prediction(alac->predict_error_buffer[ch],
368 alac->output_samples_buffer[ch], alac->nb_samples,
369 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
372 /* not compressed, easy case */
373 for (i = 0; i < alac->nb_samples; i++) {
374 if(get_bits_left(&alac->gb) <= 0)
376 for (ch = 0; ch < channels; ch++) {
377 alac->output_samples_buffer[ch][i] =
378 get_sbits_long(&alac->gb, alac->sample_size);
381 alac->extra_bits = 0;
383 decorr_left_weight = 0;
386 if (alac->extra_bits && alac->extra_bit_bug) {
387 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
388 alac->extra_bits, channels, alac->nb_samples);
391 if (channels == 2 && decorr_left_weight) {
392 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
393 decorr_shift, decorr_left_weight);
396 if (alac->extra_bits && !alac->extra_bit_bug) {
397 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
398 alac->extra_bits, channels, alac->nb_samples);
401 if(av_sample_fmt_is_planar(avctx->sample_fmt)) {
402 switch(alac->sample_size) {
404 for (ch = 0; ch < channels; ch++) {
405 int16_t *outbuffer = (int16_t *)frame->extended_data[ch_index + ch];
406 for (i = 0; i < alac->nb_samples; i++)
407 *outbuffer++ = alac->output_samples_buffer[ch][i];
411 for (ch = 0; ch < channels; ch++) {
412 for (i = 0; i < alac->nb_samples; i++)
413 alac->output_samples_buffer[ch][i] <<= 8;
418 switch(alac->sample_size) {
420 int16_t *outbuffer = ((int16_t *)frame->extended_data[0]) + ch_index;
421 for (i = 0; i < alac->nb_samples; i++) {
422 for (ch = 0; ch < channels; ch++)
423 *outbuffer++ = alac->output_samples_buffer[ch][i];
424 outbuffer += alac->channels - channels;
429 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
430 for (i = 0; i < alac->nb_samples; i++) {
431 for (ch = 0; ch < channels; ch++)
432 *outbuffer++ = alac->output_samples_buffer[ch][i] << 8;
433 outbuffer += alac->channels - channels;
438 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
439 for (i = 0; i < alac->nb_samples; i++) {
440 for (ch = 0; ch < channels; ch++)
441 *outbuffer++ = alac->output_samples_buffer[ch][i];
442 outbuffer += alac->channels - channels;
452 static int alac_decode_frame(AVCodecContext *avctx, void *data,
453 int *got_frame_ptr, AVPacket *avpkt)
455 ALACContext *alac = avctx->priv_data;
456 AVFrame *frame = data;
457 enum AlacRawDataBlockType element;
459 int ch, ret, got_end;
461 if ((ret = init_get_bits8(&alac->gb, avpkt->data, avpkt->size)) < 0)
465 alac->nb_samples = 0;
467 while (get_bits_left(&alac->gb) >= 3) {
468 element = get_bits(&alac->gb, 3);
469 if (element == TYPE_END) {
473 if (element > TYPE_CPE && element != TYPE_LFE) {
474 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d\n", element);
475 return AVERROR_PATCHWELCOME;
478 channels = (element == TYPE_CPE) ? 2 : 1;
479 if (ch + channels > alac->channels ||
480 ff_alac_channel_layout_offsets[alac->channels - 1][ch] + channels > alac->channels) {
481 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
482 return AVERROR_INVALIDDATA;
485 ret = decode_element(avctx, frame,
486 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
488 if (ret < 0 && get_bits_left(&alac->gb))
494 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
495 return AVERROR_INVALIDDATA;
498 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
499 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
500 avpkt->size * 8 - get_bits_count(&alac->gb));
503 if (alac->channels == ch)
506 av_log(avctx, AV_LOG_WARNING, "Failed to decode all channels\n");
511 static av_cold int alac_decode_close(AVCodecContext *avctx)
513 ALACContext *alac = avctx->priv_data;
516 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
517 av_freep(&alac->predict_error_buffer[ch]);
518 if (!alac->direct_output)
519 av_freep(&alac->output_samples_buffer[ch]);
520 av_freep(&alac->extra_bits_buffer[ch]);
526 static int allocate_buffers(ALACContext *alac)
529 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
531 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
532 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
533 buf_size, buf_alloc_fail);
535 alac->direct_output = alac->sample_size > 16 && av_sample_fmt_is_planar(alac->avctx->sample_fmt);
536 if (!alac->direct_output) {
537 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
538 buf_size, buf_alloc_fail);
541 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
542 buf_size, buf_alloc_fail);
546 alac_decode_close(alac->avctx);
547 return AVERROR(ENOMEM);
550 static int alac_set_info(ALACContext *alac)
554 bytestream2_init(&gb, alac->avctx->extradata,
555 alac->avctx->extradata_size);
557 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
559 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
560 if (!alac->max_samples_per_frame ||
561 alac->max_samples_per_frame > INT_MAX / sizeof(int32_t)) {
562 av_log(alac->avctx, AV_LOG_ERROR,
563 "max samples per frame invalid: %"PRIu32"\n",
564 alac->max_samples_per_frame);
565 return AVERROR_INVALIDDATA;
567 bytestream2_skipu(&gb, 1); // compatible version
568 alac->sample_size = bytestream2_get_byteu(&gb);
569 alac->rice_history_mult = bytestream2_get_byteu(&gb);
570 alac->rice_initial_history = bytestream2_get_byteu(&gb);
571 alac->rice_limit = bytestream2_get_byteu(&gb);
572 alac->channels = bytestream2_get_byteu(&gb);
573 bytestream2_get_be16u(&gb); // maxRun
574 bytestream2_get_be32u(&gb); // max coded frame size
575 bytestream2_get_be32u(&gb); // average bitrate
576 bytestream2_get_be32u(&gb); // samplerate
581 static av_cold int alac_decode_init(AVCodecContext * avctx)
585 ALACContext *alac = avctx->priv_data;
588 /* initialize from the extradata */
589 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
590 av_log(avctx, AV_LOG_ERROR, "extradata is too small\n");
591 return AVERROR_INVALIDDATA;
593 if (alac_set_info(alac)) {
594 av_log(avctx, AV_LOG_ERROR, "set_info failed\n");
598 req_packed = LIBAVCODEC_VERSION_MAJOR < 55 && !av_sample_fmt_is_planar(avctx->request_sample_fmt);
599 switch (alac->sample_size) {
600 case 16: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S16 : AV_SAMPLE_FMT_S16P;
603 case 32: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S32P;
605 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
606 return AVERROR_PATCHWELCOME;
608 avctx->bits_per_raw_sample = alac->sample_size;
610 if (alac->channels < 1) {
611 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
612 alac->channels = avctx->channels;
614 if (alac->channels > ALAC_MAX_CHANNELS)
615 alac->channels = avctx->channels;
617 avctx->channels = alac->channels;
619 if (avctx->channels > ALAC_MAX_CHANNELS || avctx->channels <= 0 ) {
620 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
622 return AVERROR_PATCHWELCOME;
624 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
626 if ((ret = allocate_buffers(alac)) < 0) {
627 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
634 static int init_thread_copy(AVCodecContext *avctx)
636 ALACContext *alac = avctx->priv_data;
638 return allocate_buffers(alac);
641 static const AVOption options[] = {
642 { "extra_bits_bug", "Force non-standard decoding process",
643 offsetof(ALACContext, extra_bit_bug), AV_OPT_TYPE_INT, { .i64 = 0 },
644 0, 1, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM },
648 static const AVClass alac_class = {
649 .class_name = "alac",
650 .item_name = av_default_item_name,
652 .version = LIBAVUTIL_VERSION_INT,
655 AVCodec ff_alac_decoder = {
657 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
658 .type = AVMEDIA_TYPE_AUDIO,
659 .id = AV_CODEC_ID_ALAC,
660 .priv_data_size = sizeof(ALACContext),
661 .init = alac_decode_init,
662 .close = alac_decode_close,
663 .decode = alac_decode_frame,
664 .init_thread_copy = ONLY_IF_THREADS_ENABLED(init_thread_copy),
665 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
666 .priv_class = &alac_class