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"
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;
80 int extra_bits; /**< number of extra bits beyond 16-bit */
81 int nb_samples; /**< number of samples in the current frame */
89 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
91 unsigned int x = get_unary_0_9(gb);
93 if (x > 8) { /* RICE THRESHOLD */
94 /* use alternative encoding */
95 x = get_bits_long(gb, bps);
97 int extrabits = show_bits(gb, k);
99 /* multiply x by 2^k - 1, as part of their strange algorithm */
106 skip_bits(gb, k - 1);
111 static int rice_decompress(ALACContext *alac, int32_t *output_buffer,
112 int nb_samples, int bps, int rice_history_mult)
115 unsigned int history = alac->rice_initial_history;
116 int sign_modifier = 0;
118 for (i = 0; i < nb_samples; i++) {
122 if(get_bits_left(&alac->gb) <= 0)
125 /* calculate rice param and decode next value */
126 k = av_log2((history >> 9) + 3);
127 k = FFMIN(k, alac->rice_limit);
128 x = decode_scalar(&alac->gb, k, bps);
131 output_buffer[i] = (x >> 1) ^ -(x & 1);
133 /* update the history */
137 history += x * rice_history_mult -
138 ((history * rice_history_mult) >> 9);
140 /* special case: there may be compressed blocks of 0 */
141 if ((history < 128) && (i + 1 < nb_samples)) {
144 /* calculate rice param and decode block size */
145 k = 7 - av_log2(history) + ((history + 16) >> 6);
146 k = FFMIN(k, alac->rice_limit);
147 block_size = decode_scalar(&alac->gb, k, 16);
149 if (block_size > 0) {
150 if (block_size >= nb_samples - i) {
151 av_log(alac->avctx, AV_LOG_ERROR,
152 "invalid zero block size of %d %d %d\n", block_size,
154 block_size = nb_samples - i - 1;
156 memset(&output_buffer[i + 1], 0,
157 block_size * sizeof(*output_buffer));
160 if (block_size <= 0xffff)
168 static inline int sign_only(int v)
170 return v ? FFSIGN(v) : 0;
173 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
174 int nb_samples, int bps, int16_t *lpc_coefs,
175 int lpc_order, int lpc_quant)
178 int32_t *pred = buffer_out;
180 /* first sample always copies */
181 *buffer_out = *error_buffer;
187 memcpy(&buffer_out[1], &error_buffer[1],
188 (nb_samples - 1) * sizeof(*buffer_out));
192 if (lpc_order == 31) {
193 /* simple 1st-order prediction */
194 for (i = 1; i < nb_samples; i++) {
195 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
201 /* read warm-up samples */
202 for (i = 1; i <= lpc_order && i < nb_samples; i++)
203 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
205 /* NOTE: 4 and 8 are very common cases that could be optimized. */
207 for (; i < nb_samples; i++) {
210 int error_val = error_buffer[i];
215 for (j = 0; j < lpc_order; j++)
216 val += (pred[j] - d) * lpc_coefs[j];
217 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
218 val += d + error_val;
219 buffer_out[i] = sign_extend(val, bps);
221 /* adapt LPC coefficients */
222 error_sign = sign_only(error_val);
224 for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {
227 sign = sign_only(val) * error_sign;
228 lpc_coefs[j] -= sign;
230 error_val -= (val >> lpc_quant) * (j + 1);
236 static int decode_element(AVCodecContext *avctx, AVFrame *frame, int ch_index,
239 ALACContext *alac = avctx->priv_data;
240 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
241 uint32_t output_samples;
244 skip_bits(&alac->gb, 4); /* element instance tag */
245 skip_bits(&alac->gb, 12); /* unused header bits */
247 /* the number of output samples is stored in the frame */
248 has_size = get_bits1(&alac->gb);
250 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
251 bps = alac->sample_size - alac->extra_bits + channels - 1;
253 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
254 return AVERROR_PATCHWELCOME;
257 /* whether the frame is compressed */
258 is_compressed = !get_bits1(&alac->gb);
261 output_samples = get_bits_long(&alac->gb, 32);
263 output_samples = alac->max_samples_per_frame;
264 if (!output_samples || output_samples > alac->max_samples_per_frame) {
265 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %"PRIu32"\n",
267 return AVERROR_INVALIDDATA;
269 if (!alac->nb_samples) {
270 ThreadFrame tframe = { .f = frame };
271 /* get output buffer */
272 frame->nb_samples = output_samples;
273 if ((ret = ff_thread_get_buffer(avctx, &tframe, 0)) < 0)
275 } else if (output_samples != alac->nb_samples) {
276 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %"PRIu32" != %d\n",
277 output_samples, alac->nb_samples);
278 return AVERROR_INVALIDDATA;
280 alac->nb_samples = output_samples;
281 if (alac->direct_output) {
282 for (ch = 0; ch < channels; ch++)
283 alac->output_samples_buffer[ch] = (int32_t *)frame->extended_data[ch_index + ch];
287 int16_t lpc_coefs[2][32];
289 int prediction_type[2];
291 int rice_history_mult[2];
293 if (!alac->rice_limit) {
294 avpriv_request_sample(alac->avctx,
295 "Compression with rice limit 0");
296 return AVERROR(ENOSYS);
299 decorr_shift = get_bits(&alac->gb, 8);
300 decorr_left_weight = get_bits(&alac->gb, 8);
302 for (ch = 0; ch < channels; ch++) {
303 prediction_type[ch] = get_bits(&alac->gb, 4);
304 lpc_quant[ch] = get_bits(&alac->gb, 4);
305 rice_history_mult[ch] = get_bits(&alac->gb, 3);
306 lpc_order[ch] = get_bits(&alac->gb, 5);
308 if (lpc_order[ch] >= alac->max_samples_per_frame)
309 return AVERROR_INVALIDDATA;
311 /* read the predictor table */
312 for (i = lpc_order[ch] - 1; i >= 0; i--)
313 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
316 if (alac->extra_bits) {
317 for (i = 0; i < alac->nb_samples; i++) {
318 if(get_bits_left(&alac->gb) <= 0)
320 for (ch = 0; ch < channels; ch++)
321 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
324 for (ch = 0; ch < channels; ch++) {
325 int ret=rice_decompress(alac, alac->predict_error_buffer[ch],
326 alac->nb_samples, bps,
327 rice_history_mult[ch] * alac->rice_history_mult / 4);
331 /* adaptive FIR filter */
332 if (prediction_type[ch] == 15) {
333 /* Prediction type 15 runs the adaptive FIR twice.
334 * The first pass uses the special-case coef_num = 31, while
335 * the second pass uses the coefs from the bitstream.
337 * However, this prediction type is not currently used by the
340 lpc_prediction(alac->predict_error_buffer[ch],
341 alac->predict_error_buffer[ch],
342 alac->nb_samples, bps, NULL, 31, 0);
343 } else if (prediction_type[ch] > 0) {
344 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
345 prediction_type[ch]);
347 lpc_prediction(alac->predict_error_buffer[ch],
348 alac->output_samples_buffer[ch], alac->nb_samples,
349 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
352 /* not compressed, easy case */
353 for (i = 0; i < alac->nb_samples; i++) {
354 if(get_bits_left(&alac->gb) <= 0)
356 for (ch = 0; ch < channels; ch++) {
357 alac->output_samples_buffer[ch][i] =
358 get_sbits_long(&alac->gb, alac->sample_size);
361 alac->extra_bits = 0;
363 decorr_left_weight = 0;
367 if (alac->extra_bits && alac->extra_bit_bug) {
368 alac->dsp.append_extra_bits[1](alac->output_samples_buffer, alac->extra_bits_buffer,
369 alac->extra_bits, channels, alac->nb_samples);
372 if (decorr_left_weight) {
373 alac->dsp.decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
374 decorr_shift, decorr_left_weight);
377 if (alac->extra_bits && !alac->extra_bit_bug) {
378 alac->dsp.append_extra_bits[1](alac->output_samples_buffer, alac->extra_bits_buffer,
379 alac->extra_bits, channels, alac->nb_samples);
381 } else if (alac->extra_bits) {
382 alac->dsp.append_extra_bits[0](alac->output_samples_buffer, alac->extra_bits_buffer,
383 alac->extra_bits, channels, alac->nb_samples);
386 switch(alac->sample_size) {
388 for (ch = 0; ch < channels; ch++) {
389 int16_t *outbuffer = (int16_t *)frame->extended_data[ch_index + ch];
390 for (i = 0; i < alac->nb_samples; i++)
391 *outbuffer++ = alac->output_samples_buffer[ch][i];
395 for (ch = 0; ch < channels; ch++) {
396 for (i = 0; i < alac->nb_samples; i++)
397 alac->output_samples_buffer[ch][i] <<= 8;
405 static int alac_decode_frame(AVCodecContext *avctx, void *data,
406 int *got_frame_ptr, AVPacket *avpkt)
408 ALACContext *alac = avctx->priv_data;
409 AVFrame *frame = data;
410 enum AlacRawDataBlockType element;
412 int ch, ret, got_end;
414 if ((ret = init_get_bits8(&alac->gb, avpkt->data, avpkt->size)) < 0)
418 alac->nb_samples = 0;
420 while (get_bits_left(&alac->gb) >= 3) {
421 element = get_bits(&alac->gb, 3);
422 if (element == TYPE_END) {
426 if (element > TYPE_CPE && element != TYPE_LFE) {
427 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d\n", element);
428 return AVERROR_PATCHWELCOME;
431 channels = (element == TYPE_CPE) ? 2 : 1;
432 if (ch + channels > alac->channels ||
433 ff_alac_channel_layout_offsets[alac->channels - 1][ch] + channels > alac->channels) {
434 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
435 return AVERROR_INVALIDDATA;
438 ret = decode_element(avctx, frame,
439 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
441 if (ret < 0 && get_bits_left(&alac->gb))
447 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
448 return AVERROR_INVALIDDATA;
451 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
452 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
453 avpkt->size * 8 - get_bits_count(&alac->gb));
456 if (alac->channels == ch)
459 av_log(avctx, AV_LOG_WARNING, "Failed to decode all channels\n");
464 static av_cold int alac_decode_close(AVCodecContext *avctx)
466 ALACContext *alac = avctx->priv_data;
469 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
470 av_freep(&alac->predict_error_buffer[ch]);
471 if (!alac->direct_output)
472 av_freep(&alac->output_samples_buffer[ch]);
473 av_freep(&alac->extra_bits_buffer[ch]);
479 static int allocate_buffers(ALACContext *alac)
482 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
484 for (ch = 0; ch < 2; ch++) {
485 alac->predict_error_buffer[ch] = NULL;
486 alac->output_samples_buffer[ch] = NULL;
487 alac->extra_bits_buffer[ch] = NULL;
490 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
491 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
492 buf_size, buf_alloc_fail);
494 alac->direct_output = alac->sample_size > 16;
495 if (!alac->direct_output) {
496 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
497 buf_size + AV_INPUT_BUFFER_PADDING_SIZE, buf_alloc_fail);
500 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
501 buf_size + AV_INPUT_BUFFER_PADDING_SIZE, buf_alloc_fail);
505 alac_decode_close(alac->avctx);
506 return AVERROR(ENOMEM);
509 static int alac_set_info(ALACContext *alac)
513 bytestream2_init(&gb, alac->avctx->extradata,
514 alac->avctx->extradata_size);
516 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
518 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
519 if (!alac->max_samples_per_frame ||
520 alac->max_samples_per_frame > INT_MAX / sizeof(int32_t)) {
521 av_log(alac->avctx, AV_LOG_ERROR,
522 "max samples per frame invalid: %"PRIu32"\n",
523 alac->max_samples_per_frame);
524 return AVERROR_INVALIDDATA;
526 bytestream2_skipu(&gb, 1); // compatible version
527 alac->sample_size = bytestream2_get_byteu(&gb);
528 alac->rice_history_mult = bytestream2_get_byteu(&gb);
529 alac->rice_initial_history = bytestream2_get_byteu(&gb);
530 alac->rice_limit = bytestream2_get_byteu(&gb);
531 alac->channels = bytestream2_get_byteu(&gb);
532 bytestream2_get_be16u(&gb); // maxRun
533 bytestream2_get_be32u(&gb); // max coded frame size
534 bytestream2_get_be32u(&gb); // average bitrate
535 bytestream2_get_be32u(&gb); // samplerate
540 static av_cold int alac_decode_init(AVCodecContext * avctx)
543 ALACContext *alac = avctx->priv_data;
546 /* initialize from the extradata */
547 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
548 av_log(avctx, AV_LOG_ERROR, "extradata is too small\n");
549 return AVERROR_INVALIDDATA;
551 if (alac_set_info(alac)) {
552 av_log(avctx, AV_LOG_ERROR, "set_info failed\n");
556 switch (alac->sample_size) {
557 case 16: avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
560 case 32: avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
562 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
563 return AVERROR_PATCHWELCOME;
565 avctx->bits_per_raw_sample = alac->sample_size;
567 if (alac->channels < 1) {
568 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
569 alac->channels = avctx->channels;
571 if (alac->channels > ALAC_MAX_CHANNELS)
572 alac->channels = avctx->channels;
574 avctx->channels = alac->channels;
576 if (avctx->channels > ALAC_MAX_CHANNELS || avctx->channels <= 0 ) {
577 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
579 return AVERROR_PATCHWELCOME;
581 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
583 if ((ret = allocate_buffers(alac)) < 0) {
584 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
588 ff_alacdsp_init(&alac->dsp);
594 static int init_thread_copy(AVCodecContext *avctx)
596 ALACContext *alac = avctx->priv_data;
598 return allocate_buffers(alac);
602 static const AVOption options[] = {
603 { "extra_bits_bug", "Force non-standard decoding process",
604 offsetof(ALACContext, extra_bit_bug), AV_OPT_TYPE_INT, { .i64 = 0 },
605 0, 1, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM },
609 static const AVClass alac_class = {
610 .class_name = "alac",
611 .item_name = av_default_item_name,
613 .version = LIBAVUTIL_VERSION_INT,
616 AVCodec ff_alac_decoder = {
618 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
619 .type = AVMEDIA_TYPE_AUDIO,
620 .id = AV_CODEC_ID_ALAC,
621 .priv_data_size = sizeof(ALACContext),
622 .init = alac_decode_init,
623 .close = alac_decode_close,
624 .decode = alac_decode_frame,
625 .init_thread_copy = ONLY_IF_THREADS_ENABLED(init_thread_copy),
626 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,
627 .priv_class = &alac_class