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"
53 #include "bytestream.h"
58 #include "alac_data.h"
60 #define ALAC_EXTRADATA_SIZE 36
63 AVCodecContext *avctx;
67 int32_t *predict_error_buffer[2];
68 int32_t *output_samples_buffer[2];
69 int32_t *extra_bits_buffer[2];
71 uint32_t max_samples_per_frame;
73 uint8_t rice_history_mult;
74 uint8_t rice_initial_history;
77 int extra_bits; /**< number of extra bits beyond 16-bit */
78 int nb_samples; /**< number of samples in the current frame */
83 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
85 unsigned int x = get_unary_0_9(gb);
87 if (x > 8) { /* RICE THRESHOLD */
88 /* use alternative encoding */
89 x = get_bits_long(gb, bps);
91 int extrabits = show_bits(gb, k);
93 /* multiply x by 2^k - 1, as part of their strange algorithm */
100 skip_bits(gb, k - 1);
105 static int rice_decompress(ALACContext *alac, int32_t *output_buffer,
106 int nb_samples, int bps, int rice_history_mult)
109 unsigned int history = alac->rice_initial_history;
110 int sign_modifier = 0;
112 for (i = 0; i < nb_samples; i++) {
116 if(get_bits_left(&alac->gb) <= 0)
119 /* calculate rice param and decode next value */
120 k = av_log2((history >> 9) + 3);
121 k = FFMIN(k, alac->rice_limit);
122 x = decode_scalar(&alac->gb, k, bps);
125 output_buffer[i] = (x >> 1) ^ -(x & 1);
127 /* update the history */
131 history += x * rice_history_mult -
132 ((history * rice_history_mult) >> 9);
134 /* special case: there may be compressed blocks of 0 */
135 if ((history < 128) && (i + 1 < nb_samples)) {
138 /* calculate rice param and decode block size */
139 k = 7 - av_log2(history) + ((history + 16) >> 6);
140 k = FFMIN(k, alac->rice_limit);
141 block_size = decode_scalar(&alac->gb, k, 16);
143 if (block_size > 0) {
144 if (block_size >= nb_samples - i) {
145 av_log(alac->avctx, AV_LOG_ERROR,
146 "invalid zero block size of %d %d %d\n", block_size,
148 block_size = nb_samples - i - 1;
150 memset(&output_buffer[i + 1], 0,
151 block_size * sizeof(*output_buffer));
154 if (block_size <= 0xffff)
162 static inline int sign_only(int v)
164 return v ? FFSIGN(v) : 0;
167 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
168 int nb_samples, int bps, int16_t *lpc_coefs,
169 int lpc_order, int lpc_quant)
172 int32_t *pred = buffer_out;
174 /* first sample always copies */
175 *buffer_out = *error_buffer;
181 memcpy(&buffer_out[1], &error_buffer[1],
182 (nb_samples - 1) * sizeof(*buffer_out));
186 if (lpc_order == 31) {
187 /* simple 1st-order prediction */
188 for (i = 1; i < nb_samples; i++) {
189 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
195 /* read warm-up samples */
196 for (i = 1; i <= lpc_order && i < nb_samples; i++)
197 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
199 /* NOTE: 4 and 8 are very common cases that could be optimized. */
201 for (; i < nb_samples; i++) {
204 int error_val = error_buffer[i];
209 for (j = 0; j < lpc_order; j++)
210 val += (pred[j] - d) * lpc_coefs[j];
211 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
212 val += d + error_val;
213 buffer_out[i] = sign_extend(val, bps);
215 /* adapt LPC coefficients */
216 error_sign = sign_only(error_val);
218 for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {
221 sign = sign_only(val) * error_sign;
222 lpc_coefs[j] -= sign;
224 error_val -= (val >> lpc_quant) * (j + 1);
230 static void decorrelate_stereo(int32_t *buffer[2], int nb_samples,
231 int decorr_shift, int decorr_left_weight)
235 for (i = 0; i < nb_samples; i++) {
241 a -= (b * decorr_left_weight) >> decorr_shift;
249 static void append_extra_bits(int32_t *buffer[2], int32_t *extra_bits_buffer[2],
250 int extra_bits, int channels, int nb_samples)
254 for (ch = 0; ch < channels; ch++)
255 for (i = 0; i < nb_samples; i++)
256 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
259 static int decode_element(AVCodecContext *avctx, AVFrame *frame, int ch_index,
262 ALACContext *alac = avctx->priv_data;
263 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
264 uint32_t output_samples;
267 skip_bits(&alac->gb, 4); /* element instance tag */
268 skip_bits(&alac->gb, 12); /* unused header bits */
270 /* the number of output samples is stored in the frame */
271 has_size = get_bits1(&alac->gb);
273 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
274 bps = alac->sample_size - alac->extra_bits + channels - 1;
276 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
277 return AVERROR_PATCHWELCOME;
280 /* whether the frame is compressed */
281 is_compressed = !get_bits1(&alac->gb);
284 output_samples = get_bits_long(&alac->gb, 32);
286 output_samples = alac->max_samples_per_frame;
287 if (!output_samples || output_samples > alac->max_samples_per_frame) {
288 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %"PRIu32"\n",
290 return AVERROR_INVALIDDATA;
292 if (!alac->nb_samples) {
293 ThreadFrame tframe = { .f = frame };
294 /* get output buffer */
295 frame->nb_samples = output_samples;
296 if ((ret = ff_thread_get_buffer(avctx, &tframe, 0)) < 0)
298 } else if (output_samples != alac->nb_samples) {
299 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %"PRIu32" != %d\n",
300 output_samples, alac->nb_samples);
301 return AVERROR_INVALIDDATA;
303 alac->nb_samples = output_samples;
304 if (alac->direct_output) {
305 for (ch = 0; ch < channels; ch++)
306 alac->output_samples_buffer[ch] = (int32_t *)frame->extended_data[ch_index + ch];
310 int16_t lpc_coefs[2][32];
312 int prediction_type[2];
314 int rice_history_mult[2];
316 decorr_shift = get_bits(&alac->gb, 8);
317 decorr_left_weight = get_bits(&alac->gb, 8);
319 for (ch = 0; ch < channels; ch++) {
320 prediction_type[ch] = get_bits(&alac->gb, 4);
321 lpc_quant[ch] = get_bits(&alac->gb, 4);
322 rice_history_mult[ch] = get_bits(&alac->gb, 3);
323 lpc_order[ch] = get_bits(&alac->gb, 5);
325 if (lpc_order[ch] >= alac->max_samples_per_frame)
326 return AVERROR_INVALIDDATA;
328 /* read the predictor table */
329 for (i = lpc_order[ch] - 1; i >= 0; i--)
330 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
333 if (alac->extra_bits) {
334 for (i = 0; i < alac->nb_samples; i++) {
335 if(get_bits_left(&alac->gb) <= 0)
337 for (ch = 0; ch < channels; ch++)
338 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
341 for (ch = 0; ch < channels; ch++) {
342 int ret=rice_decompress(alac, alac->predict_error_buffer[ch],
343 alac->nb_samples, bps,
344 rice_history_mult[ch] * alac->rice_history_mult / 4);
348 /* adaptive FIR filter */
349 if (prediction_type[ch] == 15) {
350 /* Prediction type 15 runs the adaptive FIR twice.
351 * The first pass uses the special-case coef_num = 31, while
352 * the second pass uses the coefs from the bitstream.
354 * However, this prediction type is not currently used by the
357 lpc_prediction(alac->predict_error_buffer[ch],
358 alac->predict_error_buffer[ch],
359 alac->nb_samples, bps, NULL, 31, 0);
360 } else if (prediction_type[ch] > 0) {
361 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
362 prediction_type[ch]);
364 lpc_prediction(alac->predict_error_buffer[ch],
365 alac->output_samples_buffer[ch], alac->nb_samples,
366 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
369 /* not compressed, easy case */
370 for (i = 0; i < alac->nb_samples; i++) {
371 if(get_bits_left(&alac->gb) <= 0)
373 for (ch = 0; ch < channels; ch++) {
374 alac->output_samples_buffer[ch][i] =
375 get_sbits_long(&alac->gb, alac->sample_size);
378 alac->extra_bits = 0;
380 decorr_left_weight = 0;
383 if (channels == 2 && decorr_left_weight) {
384 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
385 decorr_shift, decorr_left_weight);
388 if (alac->extra_bits) {
389 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
390 alac->extra_bits, channels, alac->nb_samples);
393 if(av_sample_fmt_is_planar(avctx->sample_fmt)) {
394 switch(alac->sample_size) {
396 for (ch = 0; ch < channels; ch++) {
397 int16_t *outbuffer = (int16_t *)frame->extended_data[ch_index + ch];
398 for (i = 0; i < alac->nb_samples; i++)
399 *outbuffer++ = alac->output_samples_buffer[ch][i];
403 for (ch = 0; ch < channels; ch++) {
404 for (i = 0; i < alac->nb_samples; i++)
405 alac->output_samples_buffer[ch][i] <<= 8;
410 switch(alac->sample_size) {
412 int16_t *outbuffer = ((int16_t *)frame->extended_data[0]) + ch_index;
413 for (i = 0; i < alac->nb_samples; i++) {
414 for (ch = 0; ch < channels; ch++)
415 *outbuffer++ = alac->output_samples_buffer[ch][i];
416 outbuffer += alac->channels - channels;
421 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
422 for (i = 0; i < alac->nb_samples; i++) {
423 for (ch = 0; ch < channels; ch++)
424 *outbuffer++ = alac->output_samples_buffer[ch][i] << 8;
425 outbuffer += alac->channels - channels;
430 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
431 for (i = 0; i < alac->nb_samples; i++) {
432 for (ch = 0; ch < channels; ch++)
433 *outbuffer++ = alac->output_samples_buffer[ch][i];
434 outbuffer += alac->channels - channels;
444 static int alac_decode_frame(AVCodecContext *avctx, void *data,
445 int *got_frame_ptr, AVPacket *avpkt)
447 ALACContext *alac = avctx->priv_data;
448 AVFrame *frame = data;
449 enum AlacRawDataBlockType element;
451 int ch, ret, got_end;
453 if ((ret = init_get_bits8(&alac->gb, avpkt->data, avpkt->size)) < 0)
457 alac->nb_samples = 0;
459 while (get_bits_left(&alac->gb) >= 3) {
460 element = get_bits(&alac->gb, 3);
461 if (element == TYPE_END) {
465 if (element > TYPE_CPE && element != TYPE_LFE) {
466 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d\n", element);
467 return AVERROR_PATCHWELCOME;
470 channels = (element == TYPE_CPE) ? 2 : 1;
471 if (ch + channels > alac->channels ||
472 ff_alac_channel_layout_offsets[alac->channels - 1][ch] + channels > alac->channels) {
473 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
474 return AVERROR_INVALIDDATA;
477 ret = decode_element(avctx, frame,
478 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
480 if (ret < 0 && get_bits_left(&alac->gb))
486 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
487 return AVERROR_INVALIDDATA;
490 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
491 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
492 avpkt->size * 8 - get_bits_count(&alac->gb));
495 if (alac->channels == ch)
498 av_log(avctx, AV_LOG_WARNING, "Failed to decode all channels\n");
503 static av_cold int alac_decode_close(AVCodecContext *avctx)
505 ALACContext *alac = avctx->priv_data;
508 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
509 av_freep(&alac->predict_error_buffer[ch]);
510 if (!alac->direct_output)
511 av_freep(&alac->output_samples_buffer[ch]);
512 av_freep(&alac->extra_bits_buffer[ch]);
518 static int allocate_buffers(ALACContext *alac)
521 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
523 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
524 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
525 buf_size, buf_alloc_fail);
527 alac->direct_output = alac->sample_size > 16 && av_sample_fmt_is_planar(alac->avctx->sample_fmt);
528 if (!alac->direct_output) {
529 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
530 buf_size, buf_alloc_fail);
533 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
534 buf_size, buf_alloc_fail);
538 alac_decode_close(alac->avctx);
539 return AVERROR(ENOMEM);
542 static int alac_set_info(ALACContext *alac)
546 bytestream2_init(&gb, alac->avctx->extradata,
547 alac->avctx->extradata_size);
549 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
551 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
552 if (!alac->max_samples_per_frame ||
553 alac->max_samples_per_frame > INT_MAX / sizeof(int32_t)) {
554 av_log(alac->avctx, AV_LOG_ERROR,
555 "max samples per frame invalid: %"PRIu32"\n",
556 alac->max_samples_per_frame);
557 return AVERROR_INVALIDDATA;
559 bytestream2_skipu(&gb, 1); // compatible version
560 alac->sample_size = bytestream2_get_byteu(&gb);
561 alac->rice_history_mult = bytestream2_get_byteu(&gb);
562 alac->rice_initial_history = bytestream2_get_byteu(&gb);
563 alac->rice_limit = bytestream2_get_byteu(&gb);
564 alac->channels = bytestream2_get_byteu(&gb);
565 bytestream2_get_be16u(&gb); // maxRun
566 bytestream2_get_be32u(&gb); // max coded frame size
567 bytestream2_get_be32u(&gb); // average bitrate
568 bytestream2_get_be32u(&gb); // samplerate
573 static av_cold int alac_decode_init(AVCodecContext * avctx)
577 ALACContext *alac = avctx->priv_data;
580 /* initialize from the extradata */
581 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
582 av_log(avctx, AV_LOG_ERROR, "extradata is too small\n");
583 return AVERROR_INVALIDDATA;
585 if (alac_set_info(alac)) {
586 av_log(avctx, AV_LOG_ERROR, "set_info failed\n");
590 req_packed = LIBAVCODEC_VERSION_MAJOR < 55 && !av_sample_fmt_is_planar(avctx->request_sample_fmt);
591 switch (alac->sample_size) {
592 case 16: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S16 : AV_SAMPLE_FMT_S16P;
595 case 32: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S32P;
597 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
598 return AVERROR_PATCHWELCOME;
600 avctx->bits_per_raw_sample = alac->sample_size;
602 if (alac->channels < 1) {
603 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
604 alac->channels = avctx->channels;
606 if (alac->channels > ALAC_MAX_CHANNELS)
607 alac->channels = avctx->channels;
609 avctx->channels = alac->channels;
611 if (avctx->channels > ALAC_MAX_CHANNELS || avctx->channels <= 0 ) {
612 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
614 return AVERROR_PATCHWELCOME;
616 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
618 if ((ret = allocate_buffers(alac)) < 0) {
619 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
626 static int init_thread_copy(AVCodecContext *avctx)
628 ALACContext *alac = avctx->priv_data;
630 return allocate_buffers(alac);
633 AVCodec ff_alac_decoder = {
635 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
636 .type = AVMEDIA_TYPE_AUDIO,
637 .id = AV_CODEC_ID_ALAC,
638 .priv_data_size = sizeof(ALACContext),
639 .init = alac_decode_init,
640 .close = alac_decode_close,
641 .decode = alac_decode_frame,
642 .init_thread_copy = ONLY_IF_THREADS_ENABLED(init_thread_copy),
643 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,