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
63 typedef struct ALACContext {
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 if (!alac->rice_limit) {
320 avpriv_request_sample(alac->avctx, "Compression with rice limit 0");
321 return AVERROR(ENOSYS);
324 decorr_shift = get_bits(&alac->gb, 8);
325 decorr_left_weight = get_bits(&alac->gb, 8);
327 for (ch = 0; ch < channels; ch++) {
328 prediction_type[ch] = get_bits(&alac->gb, 4);
329 lpc_quant[ch] = get_bits(&alac->gb, 4);
330 rice_history_mult[ch] = get_bits(&alac->gb, 3);
331 lpc_order[ch] = get_bits(&alac->gb, 5);
333 if (lpc_order[ch] >= alac->max_samples_per_frame)
334 return AVERROR_INVALIDDATA;
336 /* read the predictor table */
337 for (i = lpc_order[ch] - 1; i >= 0; i--)
338 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
341 if (alac->extra_bits) {
342 for (i = 0; i < alac->nb_samples; i++) {
343 if(get_bits_left(&alac->gb) <= 0)
345 for (ch = 0; ch < channels; ch++)
346 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
349 for (ch = 0; ch < channels; ch++) {
350 int ret=rice_decompress(alac, alac->predict_error_buffer[ch],
351 alac->nb_samples, bps,
352 rice_history_mult[ch] * alac->rice_history_mult / 4);
356 /* adaptive FIR filter */
357 if (prediction_type[ch] == 15) {
358 /* Prediction type 15 runs the adaptive FIR twice.
359 * The first pass uses the special-case coef_num = 31, while
360 * the second pass uses the coefs from the bitstream.
362 * However, this prediction type is not currently used by the
365 lpc_prediction(alac->predict_error_buffer[ch],
366 alac->predict_error_buffer[ch],
367 alac->nb_samples, bps, NULL, 31, 0);
368 } else if (prediction_type[ch] > 0) {
369 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
370 prediction_type[ch]);
372 lpc_prediction(alac->predict_error_buffer[ch],
373 alac->output_samples_buffer[ch], alac->nb_samples,
374 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
377 /* not compressed, easy case */
378 for (i = 0; i < alac->nb_samples; i++) {
379 if(get_bits_left(&alac->gb) <= 0)
381 for (ch = 0; ch < channels; ch++) {
382 alac->output_samples_buffer[ch][i] =
383 get_sbits_long(&alac->gb, alac->sample_size);
386 alac->extra_bits = 0;
388 decorr_left_weight = 0;
391 if (alac->extra_bits && alac->extra_bit_bug) {
392 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
393 alac->extra_bits, channels, alac->nb_samples);
396 if (channels == 2 && decorr_left_weight) {
397 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
398 decorr_shift, decorr_left_weight);
401 if (alac->extra_bits && !alac->extra_bit_bug) {
402 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
403 alac->extra_bits, channels, alac->nb_samples);
406 if(av_sample_fmt_is_planar(avctx->sample_fmt)) {
407 switch(alac->sample_size) {
409 for (ch = 0; ch < channels; ch++) {
410 int16_t *outbuffer = (int16_t *)frame->extended_data[ch_index + ch];
411 for (i = 0; i < alac->nb_samples; i++)
412 *outbuffer++ = alac->output_samples_buffer[ch][i];
416 for (ch = 0; ch < channels; ch++) {
417 for (i = 0; i < alac->nb_samples; i++)
418 alac->output_samples_buffer[ch][i] <<= 8;
423 switch(alac->sample_size) {
425 int16_t *outbuffer = ((int16_t *)frame->extended_data[0]) + ch_index;
426 for (i = 0; i < alac->nb_samples; i++) {
427 for (ch = 0; ch < channels; ch++)
428 *outbuffer++ = alac->output_samples_buffer[ch][i];
429 outbuffer += alac->channels - channels;
434 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
435 for (i = 0; i < alac->nb_samples; i++) {
436 for (ch = 0; ch < channels; ch++)
437 *outbuffer++ = alac->output_samples_buffer[ch][i] << 8;
438 outbuffer += alac->channels - channels;
443 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
444 for (i = 0; i < alac->nb_samples; i++) {
445 for (ch = 0; ch < channels; ch++)
446 *outbuffer++ = alac->output_samples_buffer[ch][i];
447 outbuffer += alac->channels - channels;
457 static int alac_decode_frame(AVCodecContext *avctx, void *data,
458 int *got_frame_ptr, AVPacket *avpkt)
460 ALACContext *alac = avctx->priv_data;
461 AVFrame *frame = data;
462 enum AlacRawDataBlockType element;
464 int ch, ret, got_end;
466 if ((ret = init_get_bits8(&alac->gb, avpkt->data, avpkt->size)) < 0)
470 alac->nb_samples = 0;
472 while (get_bits_left(&alac->gb) >= 3) {
473 element = get_bits(&alac->gb, 3);
474 if (element == TYPE_END) {
478 if (element > TYPE_CPE && element != TYPE_LFE) {
479 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d\n", element);
480 return AVERROR_PATCHWELCOME;
483 channels = (element == TYPE_CPE) ? 2 : 1;
484 if (ch + channels > alac->channels ||
485 ff_alac_channel_layout_offsets[alac->channels - 1][ch] + channels > alac->channels) {
486 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
487 return AVERROR_INVALIDDATA;
490 ret = decode_element(avctx, frame,
491 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
493 if (ret < 0 && get_bits_left(&alac->gb))
499 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
500 return AVERROR_INVALIDDATA;
503 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
504 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
505 avpkt->size * 8 - get_bits_count(&alac->gb));
508 if (alac->channels == ch)
511 av_log(avctx, AV_LOG_WARNING, "Failed to decode all channels\n");
516 static av_cold int alac_decode_close(AVCodecContext *avctx)
518 ALACContext *alac = avctx->priv_data;
521 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
522 av_freep(&alac->predict_error_buffer[ch]);
523 if (!alac->direct_output)
524 av_freep(&alac->output_samples_buffer[ch]);
525 av_freep(&alac->extra_bits_buffer[ch]);
531 static int allocate_buffers(ALACContext *alac)
534 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
536 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
537 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
538 buf_size, buf_alloc_fail);
540 alac->direct_output = alac->sample_size > 16 && av_sample_fmt_is_planar(alac->avctx->sample_fmt);
541 if (!alac->direct_output) {
542 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
543 buf_size, buf_alloc_fail);
546 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
547 buf_size, buf_alloc_fail);
551 alac_decode_close(alac->avctx);
552 return AVERROR(ENOMEM);
555 static int alac_set_info(ALACContext *alac)
559 bytestream2_init(&gb, alac->avctx->extradata,
560 alac->avctx->extradata_size);
562 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
564 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
565 if (!alac->max_samples_per_frame ||
566 alac->max_samples_per_frame > INT_MAX / sizeof(int32_t)) {
567 av_log(alac->avctx, AV_LOG_ERROR,
568 "max samples per frame invalid: %"PRIu32"\n",
569 alac->max_samples_per_frame);
570 return AVERROR_INVALIDDATA;
572 bytestream2_skipu(&gb, 1); // compatible version
573 alac->sample_size = bytestream2_get_byteu(&gb);
574 alac->rice_history_mult = bytestream2_get_byteu(&gb);
575 alac->rice_initial_history = bytestream2_get_byteu(&gb);
576 alac->rice_limit = bytestream2_get_byteu(&gb);
577 alac->channels = bytestream2_get_byteu(&gb);
578 bytestream2_get_be16u(&gb); // maxRun
579 bytestream2_get_be32u(&gb); // max coded frame size
580 bytestream2_get_be32u(&gb); // average bitrate
581 bytestream2_get_be32u(&gb); // samplerate
586 static av_cold int alac_decode_init(AVCodecContext * avctx)
590 ALACContext *alac = avctx->priv_data;
593 /* initialize from the extradata */
594 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
595 av_log(avctx, AV_LOG_ERROR, "extradata is too small\n");
596 return AVERROR_INVALIDDATA;
598 if (alac_set_info(alac)) {
599 av_log(avctx, AV_LOG_ERROR, "set_info failed\n");
603 req_packed = LIBAVCODEC_VERSION_MAJOR < 55 && !av_sample_fmt_is_planar(avctx->request_sample_fmt);
604 switch (alac->sample_size) {
605 case 16: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S16 : AV_SAMPLE_FMT_S16P;
608 case 32: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S32P;
610 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
611 return AVERROR_PATCHWELCOME;
613 avctx->bits_per_raw_sample = alac->sample_size;
615 if (alac->channels < 1) {
616 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
617 alac->channels = avctx->channels;
619 if (alac->channels > ALAC_MAX_CHANNELS)
620 alac->channels = avctx->channels;
622 avctx->channels = alac->channels;
624 if (avctx->channels > ALAC_MAX_CHANNELS || avctx->channels <= 0 ) {
625 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
627 return AVERROR_PATCHWELCOME;
629 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
631 if ((ret = allocate_buffers(alac)) < 0) {
632 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
639 static int init_thread_copy(AVCodecContext *avctx)
641 ALACContext *alac = avctx->priv_data;
643 return allocate_buffers(alac);
646 static const AVOption options[] = {
647 { "extra_bits_bug", "Force non-standard decoding process",
648 offsetof(ALACContext, extra_bit_bug), AV_OPT_TYPE_INT, { .i64 = 0 },
649 0, 1, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM },
653 static const AVClass alac_class = {
654 .class_name = "alac",
655 .item_name = av_default_item_name,
657 .version = LIBAVUTIL_VERSION_INT,
660 AVCodec ff_alac_decoder = {
662 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
663 .type = AVMEDIA_TYPE_AUDIO,
664 .id = AV_CODEC_ID_ALAC,
665 .priv_data_size = sizeof(ALACContext),
666 .init = alac_decode_init,
667 .close = alac_decode_close,
668 .decode = alac_decode_frame,
669 .init_thread_copy = ONLY_IF_THREADS_ENABLED(init_thread_copy),
670 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
671 .priv_class = &alac_class