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)
48 #include "libavutil/channel_layout.h"
51 #include "bytestream.h"
56 #include "alac_data.h"
58 #define ALAC_EXTRADATA_SIZE 36
61 AVCodecContext *avctx;
65 int32_t *predict_error_buffer[2];
66 int32_t *output_samples_buffer[2];
67 int32_t *extra_bits_buffer[2];
69 uint32_t max_samples_per_frame;
71 uint8_t rice_history_mult;
72 uint8_t rice_initial_history;
75 int extra_bits; /**< number of extra bits beyond 16-bit */
76 int nb_samples; /**< number of samples in the current frame */
81 static inline unsigned int decode_scalar(GetBitContext *gb, int k, int bps)
83 unsigned int x = get_unary_0_9(gb);
85 if (x > 8) { /* RICE THRESHOLD */
86 /* use alternative encoding */
87 x = get_bits_long(gb, bps);
89 int extrabits = show_bits(gb, k);
91 /* multiply x by 2^k - 1, as part of their strange algorithm */
103 static int rice_decompress(ALACContext *alac, int32_t *output_buffer,
104 int nb_samples, int bps, int rice_history_mult)
107 unsigned int history = alac->rice_initial_history;
108 int sign_modifier = 0;
110 for (i = 0; i < nb_samples; i++) {
114 if(get_bits_left(&alac->gb) <= 0)
117 /* calculate rice param and decode next value */
118 k = av_log2((history >> 9) + 3);
119 k = FFMIN(k, alac->rice_limit);
120 x = decode_scalar(&alac->gb, k, bps);
123 output_buffer[i] = (x >> 1) ^ -(x & 1);
125 /* update the history */
129 history += x * rice_history_mult -
130 ((history * rice_history_mult) >> 9);
132 /* special case: there may be compressed blocks of 0 */
133 if ((history < 128) && (i + 1 < nb_samples)) {
136 /* calculate rice param and decode block size */
137 k = 7 - av_log2(history) + ((history + 16) >> 6);
138 k = FFMIN(k, alac->rice_limit);
139 block_size = decode_scalar(&alac->gb, k, 16);
141 if (block_size > 0) {
142 if (block_size >= nb_samples - i) {
143 av_log(alac->avctx, AV_LOG_ERROR,
144 "invalid zero block size of %d %d %d\n", block_size,
146 block_size = nb_samples - i - 1;
148 memset(&output_buffer[i + 1], 0,
149 block_size * sizeof(*output_buffer));
152 if (block_size <= 0xffff)
160 static inline int sign_only(int v)
162 return v ? FFSIGN(v) : 0;
165 static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,
166 int nb_samples, int bps, int16_t *lpc_coefs,
167 int lpc_order, int lpc_quant)
170 int32_t *pred = buffer_out;
172 /* first sample always copies */
173 *buffer_out = *error_buffer;
179 memcpy(&buffer_out[1], &error_buffer[1],
180 (nb_samples - 1) * sizeof(*buffer_out));
184 if (lpc_order == 31) {
185 /* simple 1st-order prediction */
186 for (i = 1; i < nb_samples; i++) {
187 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i],
193 /* read warm-up samples */
194 for (i = 1; i <= lpc_order && i < nb_samples; i++)
195 buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);
197 /* NOTE: 4 and 8 are very common cases that could be optimized. */
199 for (; i < nb_samples; i++) {
202 int error_val = error_buffer[i];
207 for (j = 0; j < lpc_order; j++)
208 val += (pred[j] - d) * lpc_coefs[j];
209 val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;
210 val += d + error_val;
211 buffer_out[i] = sign_extend(val, bps);
213 /* adapt LPC coefficients */
214 error_sign = sign_only(error_val);
216 for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {
219 sign = sign_only(val) * error_sign;
220 lpc_coefs[j] -= sign;
222 error_val -= (val >> lpc_quant) * (j + 1);
228 static void decorrelate_stereo(int32_t *buffer[2], int nb_samples,
229 int decorr_shift, int decorr_left_weight)
233 for (i = 0; i < nb_samples; i++) {
239 a -= (b * decorr_left_weight) >> decorr_shift;
247 static void append_extra_bits(int32_t *buffer[2], int32_t *extra_bits_buffer[2],
248 int extra_bits, int channels, int nb_samples)
252 for (ch = 0; ch < channels; ch++)
253 for (i = 0; i < nb_samples; i++)
254 buffer[ch][i] = (buffer[ch][i] << extra_bits) | extra_bits_buffer[ch][i];
257 static int decode_element(AVCodecContext *avctx, AVFrame *frame, int ch_index,
260 ALACContext *alac = avctx->priv_data;
261 int has_size, bps, is_compressed, decorr_shift, decorr_left_weight, ret;
262 uint32_t output_samples;
265 skip_bits(&alac->gb, 4); /* element instance tag */
266 skip_bits(&alac->gb, 12); /* unused header bits */
268 /* the number of output samples is stored in the frame */
269 has_size = get_bits1(&alac->gb);
271 alac->extra_bits = get_bits(&alac->gb, 2) << 3;
272 bps = alac->sample_size - alac->extra_bits + channels - 1;
274 av_log(avctx, AV_LOG_ERROR, "bps is unsupported: %d\n", bps);
275 return AVERROR_PATCHWELCOME;
278 /* whether the frame is compressed */
279 is_compressed = !get_bits1(&alac->gb);
282 output_samples = get_bits_long(&alac->gb, 32);
284 output_samples = alac->max_samples_per_frame;
285 if (!output_samples || output_samples > alac->max_samples_per_frame) {
286 av_log(avctx, AV_LOG_ERROR, "invalid samples per frame: %d\n",
288 return AVERROR_INVALIDDATA;
290 if (!alac->nb_samples) {
291 ThreadFrame tframe = { .f = frame };
292 /* get output buffer */
293 frame->nb_samples = output_samples;
294 if ((ret = ff_thread_get_buffer(avctx, &tframe, 0)) < 0)
296 } else if (output_samples != alac->nb_samples) {
297 av_log(avctx, AV_LOG_ERROR, "sample count mismatch: %u != %d\n",
298 output_samples, alac->nb_samples);
299 return AVERROR_INVALIDDATA;
301 alac->nb_samples = output_samples;
302 if (alac->direct_output) {
303 for (ch = 0; ch < channels; ch++)
304 alac->output_samples_buffer[ch] = (int32_t *)frame->extended_data[ch_index + ch];
308 int16_t lpc_coefs[2][32];
310 int prediction_type[2];
312 int rice_history_mult[2];
314 decorr_shift = get_bits(&alac->gb, 8);
315 decorr_left_weight = get_bits(&alac->gb, 8);
317 for (ch = 0; ch < channels; ch++) {
318 prediction_type[ch] = get_bits(&alac->gb, 4);
319 lpc_quant[ch] = get_bits(&alac->gb, 4);
320 rice_history_mult[ch] = get_bits(&alac->gb, 3);
321 lpc_order[ch] = get_bits(&alac->gb, 5);
323 if (lpc_order[ch] >= alac->max_samples_per_frame)
324 return AVERROR_INVALIDDATA;
326 /* read the predictor table */
327 for (i = lpc_order[ch] - 1; i >= 0; i--)
328 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);
331 if (alac->extra_bits) {
332 for (i = 0; i < alac->nb_samples; i++) {
333 if(get_bits_left(&alac->gb) <= 0)
335 for (ch = 0; ch < channels; ch++)
336 alac->extra_bits_buffer[ch][i] = get_bits(&alac->gb, alac->extra_bits);
339 for (ch = 0; ch < channels; ch++) {
340 int ret=rice_decompress(alac, alac->predict_error_buffer[ch],
341 alac->nb_samples, bps,
342 rice_history_mult[ch] * alac->rice_history_mult / 4);
346 /* adaptive FIR filter */
347 if (prediction_type[ch] == 15) {
348 /* Prediction type 15 runs the adaptive FIR twice.
349 * The first pass uses the special-case coef_num = 31, while
350 * the second pass uses the coefs from the bitstream.
352 * However, this prediction type is not currently used by the
355 lpc_prediction(alac->predict_error_buffer[ch],
356 alac->predict_error_buffer[ch],
357 alac->nb_samples, bps, NULL, 31, 0);
358 } else if (prediction_type[ch] > 0) {
359 av_log(avctx, AV_LOG_WARNING, "unknown prediction type: %i\n",
360 prediction_type[ch]);
362 lpc_prediction(alac->predict_error_buffer[ch],
363 alac->output_samples_buffer[ch], alac->nb_samples,
364 bps, lpc_coefs[ch], lpc_order[ch], lpc_quant[ch]);
367 /* not compressed, easy case */
368 for (i = 0; i < alac->nb_samples; i++) {
369 if(get_bits_left(&alac->gb) <= 0)
371 for (ch = 0; ch < channels; ch++) {
372 alac->output_samples_buffer[ch][i] =
373 get_sbits_long(&alac->gb, alac->sample_size);
376 alac->extra_bits = 0;
378 decorr_left_weight = 0;
381 if (channels == 2 && decorr_left_weight) {
382 decorrelate_stereo(alac->output_samples_buffer, alac->nb_samples,
383 decorr_shift, decorr_left_weight);
386 if (alac->extra_bits) {
387 append_extra_bits(alac->output_samples_buffer, alac->extra_bits_buffer,
388 alac->extra_bits, channels, alac->nb_samples);
391 if(av_sample_fmt_is_planar(avctx->sample_fmt)) {
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] <<= 8;
408 switch(alac->sample_size) {
410 int16_t *outbuffer = ((int16_t *)frame->extended_data[0]) + ch_index;
411 for (i = 0; i < alac->nb_samples; i++) {
412 for (ch = 0; ch < channels; ch++)
413 *outbuffer++ = alac->output_samples_buffer[ch][i];
414 outbuffer += alac->channels - channels;
419 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
420 for (i = 0; i < alac->nb_samples; i++) {
421 for (ch = 0; ch < channels; ch++)
422 *outbuffer++ = alac->output_samples_buffer[ch][i] << 8;
423 outbuffer += alac->channels - channels;
428 int32_t *outbuffer = ((int32_t *)frame->extended_data[0]) + ch_index;
429 for (i = 0; i < alac->nb_samples; i++) {
430 for (ch = 0; ch < channels; ch++)
431 *outbuffer++ = alac->output_samples_buffer[ch][i];
432 outbuffer += alac->channels - channels;
442 static int alac_decode_frame(AVCodecContext *avctx, void *data,
443 int *got_frame_ptr, AVPacket *avpkt)
445 ALACContext *alac = avctx->priv_data;
446 AVFrame *frame = data;
447 enum AlacRawDataBlockType element;
449 int ch, ret, got_end;
451 if ((ret = init_get_bits8(&alac->gb, avpkt->data, avpkt->size)) < 0)
455 alac->nb_samples = 0;
457 while (get_bits_left(&alac->gb) >= 3) {
458 element = get_bits(&alac->gb, 3);
459 if (element == TYPE_END) {
463 if (element > TYPE_CPE && element != TYPE_LFE) {
464 av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d\n", element);
465 return AVERROR_PATCHWELCOME;
468 channels = (element == TYPE_CPE) ? 2 : 1;
469 if (ch + channels > alac->channels ||
470 ff_alac_channel_layout_offsets[alac->channels - 1][ch] + channels > alac->channels) {
471 av_log(avctx, AV_LOG_ERROR, "invalid element channel count\n");
472 return AVERROR_INVALIDDATA;
475 ret = decode_element(avctx, frame,
476 ff_alac_channel_layout_offsets[alac->channels - 1][ch],
478 if (ret < 0 && get_bits_left(&alac->gb))
484 av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");
485 return AVERROR_INVALIDDATA;
488 if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {
489 av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",
490 avpkt->size * 8 - get_bits_count(&alac->gb));
498 static av_cold int alac_decode_close(AVCodecContext *avctx)
500 ALACContext *alac = avctx->priv_data;
503 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
504 av_freep(&alac->predict_error_buffer[ch]);
505 if (!alac->direct_output)
506 av_freep(&alac->output_samples_buffer[ch]);
507 av_freep(&alac->extra_bits_buffer[ch]);
513 static int allocate_buffers(ALACContext *alac)
516 int buf_size = alac->max_samples_per_frame * sizeof(int32_t);
518 for (ch = 0; ch < FFMIN(alac->channels, 2); ch++) {
519 FF_ALLOC_OR_GOTO(alac->avctx, alac->predict_error_buffer[ch],
520 buf_size, buf_alloc_fail);
522 alac->direct_output = alac->sample_size > 16 && av_sample_fmt_is_planar(alac->avctx->sample_fmt);
523 if (!alac->direct_output) {
524 FF_ALLOC_OR_GOTO(alac->avctx, alac->output_samples_buffer[ch],
525 buf_size, buf_alloc_fail);
528 FF_ALLOC_OR_GOTO(alac->avctx, alac->extra_bits_buffer[ch],
529 buf_size, buf_alloc_fail);
533 alac_decode_close(alac->avctx);
534 return AVERROR(ENOMEM);
537 static int alac_set_info(ALACContext *alac)
541 bytestream2_init(&gb, alac->avctx->extradata,
542 alac->avctx->extradata_size);
544 bytestream2_skipu(&gb, 12); // size:4, alac:4, version:4
546 alac->max_samples_per_frame = bytestream2_get_be32u(&gb);
547 if (!alac->max_samples_per_frame ||
548 alac->max_samples_per_frame > INT_MAX / sizeof(int32_t)) {
549 av_log(alac->avctx, AV_LOG_ERROR, "max samples per frame invalid: %u\n",
550 alac->max_samples_per_frame);
551 return AVERROR_INVALIDDATA;
553 bytestream2_skipu(&gb, 1); // compatible version
554 alac->sample_size = bytestream2_get_byteu(&gb);
555 alac->rice_history_mult = bytestream2_get_byteu(&gb);
556 alac->rice_initial_history = bytestream2_get_byteu(&gb);
557 alac->rice_limit = bytestream2_get_byteu(&gb);
558 alac->channels = bytestream2_get_byteu(&gb);
559 bytestream2_get_be16u(&gb); // maxRun
560 bytestream2_get_be32u(&gb); // max coded frame size
561 bytestream2_get_be32u(&gb); // average bitrate
562 bytestream2_get_be32u(&gb); // samplerate
567 static av_cold int alac_decode_init(AVCodecContext * avctx)
571 ALACContext *alac = avctx->priv_data;
574 /* initialize from the extradata */
575 if (alac->avctx->extradata_size < ALAC_EXTRADATA_SIZE) {
576 av_log(avctx, AV_LOG_ERROR, "extradata is too small\n");
577 return AVERROR_INVALIDDATA;
579 if (alac_set_info(alac)) {
580 av_log(avctx, AV_LOG_ERROR, "set_info failed\n");
584 req_packed = LIBAVCODEC_VERSION_MAJOR < 55 && !av_sample_fmt_is_planar(avctx->request_sample_fmt);
585 switch (alac->sample_size) {
586 case 16: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S16 : AV_SAMPLE_FMT_S16P;
589 case 32: avctx->sample_fmt = req_packed ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S32P;
591 default: avpriv_request_sample(avctx, "Sample depth %d", alac->sample_size);
592 return AVERROR_PATCHWELCOME;
594 avctx->bits_per_raw_sample = alac->sample_size;
596 if (alac->channels < 1) {
597 av_log(avctx, AV_LOG_WARNING, "Invalid channel count\n");
598 alac->channels = avctx->channels;
600 if (alac->channels > ALAC_MAX_CHANNELS)
601 alac->channels = avctx->channels;
603 avctx->channels = alac->channels;
605 if (avctx->channels > ALAC_MAX_CHANNELS || avctx->channels <= 0 ) {
606 av_log(avctx, AV_LOG_ERROR, "Unsupported channel count: %d\n",
608 return AVERROR_PATCHWELCOME;
610 avctx->channel_layout = ff_alac_channel_layouts[alac->channels - 1];
612 if ((ret = allocate_buffers(alac)) < 0) {
613 av_log(avctx, AV_LOG_ERROR, "Error allocating buffers\n");
620 static int init_thread_copy(AVCodecContext *avctx)
622 ALACContext *alac = avctx->priv_data;
624 return allocate_buffers(alac);
627 AVCodec ff_alac_decoder = {
629 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
630 .type = AVMEDIA_TYPE_AUDIO,
631 .id = AV_CODEC_ID_ALAC,
632 .priv_data_size = sizeof(ALACContext),
633 .init = alac_decode_init,
634 .close = alac_decode_close,
635 .decode = alac_decode_frame,
636 .init_thread_copy = ONLY_IF_THREADS_ENABLED(init_thread_copy),
637 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,