2 * FLAC (Free Lossless Audio Codec) decoder
3 * Copyright (c) 2003 Alex Beregszaszi
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
23 * @file libavcodec/flacdec.c
24 * FLAC (Free Lossless Audio Codec) decoder
25 * @author Alex Beregszaszi
27 * For more information on the FLAC format, visit:
28 * http://flac.sourceforge.net/
30 * This decoder can be used in 1 of 2 ways: Either raw FLAC data can be fed
31 * through, starting from the initial 'fLaC' signature; or by passing the
32 * 34-byte streaminfo structure through avctx->extradata[_size] followed
33 * by data starting with the 0xFFF8 marker.
38 #include "libavutil/crc.h"
41 #include "bitstream.h"
48 #define MAX_CHANNELS 8
49 #define MAX_BLOCKSIZE 65535
51 enum decorrelation_type {
58 typedef struct FLACContext {
61 AVCodecContext *avctx; ///< parent AVCodecContext
62 GetBitContext gb; ///< GetBitContext initialized to start at the current frame
64 int blocksize; ///< number of samples in the current frame
65 int curr_bps; ///< bps for current subframe, adjusted for channel correlation and wasted bits
66 int sample_shift; ///< shift required to make output samples 16-bit or 32-bit
67 int is32; ///< flag to indicate if output should be 32-bit instead of 16-bit
68 enum decorrelation_type decorrelation; ///< channel decorrelation type in the current frame
70 int32_t *decoded[MAX_CHANNELS]; ///< decoded samples
72 unsigned int bitstream_size;
73 unsigned int bitstream_index;
74 unsigned int allocated_bitstream_size;
77 static const int sample_rate_table[] =
79 88200, 176400, 192000,
80 8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000,
83 static const int sample_size_table[] =
84 { 0, 8, 12, 0, 16, 20, 24, 0 };
86 static const int blocksize_table[] = {
87 0, 192, 576<<0, 576<<1, 576<<2, 576<<3, 0, 0,
88 256<<0, 256<<1, 256<<2, 256<<3, 256<<4, 256<<5, 256<<6, 256<<7
91 static int64_t get_utf8(GetBitContext *gb)
94 GET_UTF8(val, get_bits(gb, 8), return -1;)
98 static void allocate_buffers(FLACContext *s);
100 int ff_flac_is_extradata_valid(AVCodecContext *avctx,
101 enum FLACExtradataFormat *format,
102 uint8_t **streaminfo_start)
104 if (!avctx->extradata || avctx->extradata_size < FLAC_STREAMINFO_SIZE) {
105 av_log(avctx, AV_LOG_ERROR, "extradata NULL or too small.\n");
108 if (AV_RL32(avctx->extradata) != MKTAG('f','L','a','C')) {
109 /* extradata contains STREAMINFO only */
110 if (avctx->extradata_size != FLAC_STREAMINFO_SIZE) {
111 av_log(avctx, AV_LOG_WARNING, "extradata contains %d bytes too many.\n",
112 FLAC_STREAMINFO_SIZE-avctx->extradata_size);
114 *format = FLAC_EXTRADATA_FORMAT_STREAMINFO;
115 *streaminfo_start = avctx->extradata;
117 if (avctx->extradata_size < 8+FLAC_STREAMINFO_SIZE) {
118 av_log(avctx, AV_LOG_ERROR, "extradata too small.\n");
121 *format = FLAC_EXTRADATA_FORMAT_FULL_HEADER;
122 *streaminfo_start = &avctx->extradata[8];
127 static av_cold int flac_decode_init(AVCodecContext *avctx)
129 enum FLACExtradataFormat format;
131 FLACContext *s = avctx->priv_data;
134 avctx->sample_fmt = SAMPLE_FMT_S16;
136 /* for now, the raw FLAC header is allowed to be passed to the decoder as
137 frame data instead of extradata. */
138 if (!avctx->extradata)
141 if (!ff_flac_is_extradata_valid(avctx, &format, &streaminfo))
144 /* initialize based on the demuxer-supplied streamdata header */
145 ff_flac_parse_streaminfo(avctx, (FLACStreaminfo *)s, streaminfo);
151 static void dump_headers(AVCodecContext *avctx, FLACStreaminfo *s)
153 av_log(avctx, AV_LOG_DEBUG, " Blocksize: %d .. %d\n", s->min_blocksize,
155 av_log(avctx, AV_LOG_DEBUG, " Max Framesize: %d\n", s->max_framesize);
156 av_log(avctx, AV_LOG_DEBUG, " Samplerate: %d\n", s->samplerate);
157 av_log(avctx, AV_LOG_DEBUG, " Channels: %d\n", s->channels);
158 av_log(avctx, AV_LOG_DEBUG, " Bits: %d\n", s->bps);
161 static void allocate_buffers(FLACContext *s)
165 assert(s->max_blocksize);
167 if (s->max_framesize == 0 && s->max_blocksize) {
168 // FIXME header overhead
169 s->max_framesize= (s->channels * s->bps * s->max_blocksize + 7)/ 8;
172 for (i = 0; i < s->channels; i++) {
173 s->decoded[i] = av_realloc(s->decoded[i],
174 sizeof(int32_t)*s->max_blocksize);
177 if (s->allocated_bitstream_size < s->max_framesize)
178 s->bitstream= av_fast_realloc(s->bitstream,
179 &s->allocated_bitstream_size,
183 void ff_flac_parse_streaminfo(AVCodecContext *avctx, struct FLACStreaminfo *s,
184 const uint8_t *buffer)
187 init_get_bits(&gb, buffer, FLAC_STREAMINFO_SIZE*8);
189 /* mandatory streaminfo */
190 s->min_blocksize = get_bits(&gb, 16);
191 s->max_blocksize = get_bits(&gb, 16);
193 skip_bits(&gb, 24); /* skip min frame size */
194 s->max_framesize = get_bits_long(&gb, 24);
196 s->samplerate = get_bits_long(&gb, 20);
197 s->channels = get_bits(&gb, 3) + 1;
198 s->bps = get_bits(&gb, 5) + 1;
200 avctx->channels = s->channels;
201 avctx->sample_rate = s->samplerate;
202 avctx->bits_per_raw_sample = s->bps;
204 avctx->sample_fmt = SAMPLE_FMT_S32;
206 avctx->sample_fmt = SAMPLE_FMT_S16;
208 s->samples = get_bits_long(&gb, 32) << 4;
209 s->samples |= get_bits(&gb, 4);
211 skip_bits_long(&gb, 64); /* md5 sum */
212 skip_bits_long(&gb, 64); /* md5 sum */
214 dump_headers(avctx, s);
218 * Parse a list of metadata blocks. This list of blocks must begin with
220 * @param s the flac decoding context containing the gb bit reader used to
222 * @return 1 if some metadata was read, 0 if no fLaC marker was found
224 static int metadata_parse(FLACContext *s)
226 int i, metadata_last, metadata_type, metadata_size, streaminfo_updated=0;
227 int initial_pos= get_bits_count(&s->gb);
229 if (show_bits_long(&s->gb, 32) == MKBETAG('f','L','a','C')) {
230 skip_bits_long(&s->gb, 32);
233 metadata_last = get_bits1(&s->gb);
234 metadata_type = get_bits(&s->gb, 7);
235 metadata_size = get_bits_long(&s->gb, 24);
237 if (get_bits_count(&s->gb) + 8*metadata_size > s->gb.size_in_bits) {
238 skip_bits_long(&s->gb, initial_pos - get_bits_count(&s->gb));
243 switch (metadata_type) {
244 case FLAC_METADATA_TYPE_STREAMINFO:
245 ff_flac_parse_streaminfo(s->avctx, (FLACStreaminfo *)s,
246 s->gb.buffer+get_bits_count(&s->gb)/8);
247 streaminfo_updated = 1;
250 for (i = 0; i < metadata_size; i++)
251 skip_bits(&s->gb, 8);
254 } while (!metadata_last);
256 if (streaminfo_updated)
263 static int decode_residuals(FLACContext *s, int channel, int pred_order)
265 int i, tmp, partition, method_type, rice_order;
266 int sample = 0, samples;
268 method_type = get_bits(&s->gb, 2);
269 if (method_type > 1) {
270 av_log(s->avctx, AV_LOG_ERROR, "illegal residual coding method %d\n",
275 rice_order = get_bits(&s->gb, 4);
277 samples= s->blocksize >> rice_order;
278 if (pred_order > samples) {
279 av_log(s->avctx, AV_LOG_ERROR, "invalid predictor order: %i > %i\n",
280 pred_order, samples);
286 for (partition = 0; partition < (1 << rice_order); partition++) {
287 tmp = get_bits(&s->gb, method_type == 0 ? 4 : 5);
288 if (tmp == (method_type == 0 ? 15 : 31)) {
289 tmp = get_bits(&s->gb, 5);
290 for (; i < samples; i++, sample++)
291 s->decoded[channel][sample] = get_sbits_long(&s->gb, tmp);
293 for (; i < samples; i++, sample++) {
294 s->decoded[channel][sample] = get_sr_golomb_flac(&s->gb, tmp, INT_MAX, 0);
303 static int decode_subframe_fixed(FLACContext *s, int channel, int pred_order)
305 const int blocksize = s->blocksize;
306 int32_t *decoded = s->decoded[channel];
307 int av_uninit(a), av_uninit(b), av_uninit(c), av_uninit(d), i;
309 /* warm up samples */
310 for (i = 0; i < pred_order; i++) {
311 decoded[i] = get_sbits_long(&s->gb, s->curr_bps);
314 if (decode_residuals(s, channel, pred_order) < 0)
318 a = decoded[pred_order-1];
320 b = a - decoded[pred_order-2];
322 c = b - decoded[pred_order-2] + decoded[pred_order-3];
324 d = c - decoded[pred_order-2] + 2*decoded[pred_order-3] - decoded[pred_order-4];
326 switch (pred_order) {
330 for (i = pred_order; i < blocksize; i++)
331 decoded[i] = a += decoded[i];
334 for (i = pred_order; i < blocksize; i++)
335 decoded[i] = a += b += decoded[i];
338 for (i = pred_order; i < blocksize; i++)
339 decoded[i] = a += b += c += decoded[i];
342 for (i = pred_order; i < blocksize; i++)
343 decoded[i] = a += b += c += d += decoded[i];
346 av_log(s->avctx, AV_LOG_ERROR, "illegal pred order %d\n", pred_order);
353 static int decode_subframe_lpc(FLACContext *s, int channel, int pred_order)
356 int coeff_prec, qlevel;
357 int coeffs[pred_order];
358 int32_t *decoded = s->decoded[channel];
360 /* warm up samples */
361 for (i = 0; i < pred_order; i++) {
362 decoded[i] = get_sbits_long(&s->gb, s->curr_bps);
365 coeff_prec = get_bits(&s->gb, 4) + 1;
366 if (coeff_prec == 16) {
367 av_log(s->avctx, AV_LOG_ERROR, "invalid coeff precision\n");
370 qlevel = get_sbits(&s->gb, 5);
372 av_log(s->avctx, AV_LOG_ERROR, "qlevel %d not supported, maybe buggy stream\n",
377 for (i = 0; i < pred_order; i++) {
378 coeffs[i] = get_sbits(&s->gb, coeff_prec);
381 if (decode_residuals(s, channel, pred_order) < 0)
386 for (i = pred_order; i < s->blocksize; i++) {
388 for (j = 0; j < pred_order; j++)
389 sum += (int64_t)coeffs[j] * decoded[i-j-1];
390 decoded[i] += sum >> qlevel;
393 for (i = pred_order; i < s->blocksize-1; i += 2) {
395 int d = decoded[i-pred_order];
397 for (j = pred_order-1; j > 0; j--) {
405 d = decoded[i] += s0 >> qlevel;
407 decoded[i+1] += s1 >> qlevel;
409 if (i < s->blocksize) {
411 for (j = 0; j < pred_order; j++)
412 sum += coeffs[j] * decoded[i-j-1];
413 decoded[i] += sum >> qlevel;
420 static inline int decode_subframe(FLACContext *s, int channel)
422 int type, wasted = 0;
425 s->curr_bps = s->bps;
427 if (s->decorrelation == RIGHT_SIDE)
430 if (s->decorrelation == LEFT_SIDE || s->decorrelation == MID_SIDE)
433 if (s->curr_bps > 32) {
434 ff_log_missing_feature(s->avctx, "decorrelated bit depth > 32", 0);
438 if (get_bits1(&s->gb)) {
439 av_log(s->avctx, AV_LOG_ERROR, "invalid subframe padding\n");
442 type = get_bits(&s->gb, 6);
444 if (get_bits1(&s->gb)) {
446 while (!get_bits1(&s->gb))
448 s->curr_bps -= wasted;
451 //FIXME use av_log2 for types
453 tmp = get_sbits_long(&s->gb, s->curr_bps);
454 for (i = 0; i < s->blocksize; i++)
455 s->decoded[channel][i] = tmp;
456 } else if (type == 1) {
457 for (i = 0; i < s->blocksize; i++)
458 s->decoded[channel][i] = get_sbits_long(&s->gb, s->curr_bps);
459 } else if ((type >= 8) && (type <= 12)) {
460 if (decode_subframe_fixed(s, channel, type & ~0x8) < 0)
462 } else if (type >= 32) {
463 if (decode_subframe_lpc(s, channel, (type & ~0x20)+1) < 0)
466 av_log(s->avctx, AV_LOG_ERROR, "invalid coding type\n");
472 for (i = 0; i < s->blocksize; i++)
473 s->decoded[channel][i] <<= wasted;
479 static int decode_frame(FLACContext *s, int alloc_data_size)
481 int blocksize_code, sample_rate_code, sample_size_code, assignment, i, crc8;
482 int decorrelation, bps, blocksize, samplerate;
484 blocksize_code = get_bits(&s->gb, 4);
486 sample_rate_code = get_bits(&s->gb, 4);
488 assignment = get_bits(&s->gb, 4); /* channel assignment */
489 if (assignment < 8 && s->channels == assignment+1)
490 decorrelation = INDEPENDENT;
491 else if (assignment >=8 && assignment < 11 && s->channels == 2)
492 decorrelation = LEFT_SIDE + assignment - 8;
494 av_log(s->avctx, AV_LOG_ERROR, "unsupported channel assignment %d (channels=%d)\n",
495 assignment, s->channels);
499 sample_size_code = get_bits(&s->gb, 3);
500 if (sample_size_code == 0)
502 else if ((sample_size_code != 3) && (sample_size_code != 7))
503 bps = sample_size_table[sample_size_code];
505 av_log(s->avctx, AV_LOG_ERROR, "invalid sample size code (%d)\n",
510 s->avctx->sample_fmt = SAMPLE_FMT_S32;
511 s->sample_shift = 32 - bps;
514 s->avctx->sample_fmt = SAMPLE_FMT_S16;
515 s->sample_shift = 16 - bps;
518 s->bps = s->avctx->bits_per_raw_sample = bps;
520 if (get_bits1(&s->gb)) {
521 av_log(s->avctx, AV_LOG_ERROR, "broken stream, invalid padding\n");
525 if (get_utf8(&s->gb) < 0) {
526 av_log(s->avctx, AV_LOG_ERROR, "utf8 fscked\n");
530 if (blocksize_code == 0)
531 blocksize = s->min_blocksize;
532 else if (blocksize_code == 6)
533 blocksize = get_bits(&s->gb, 8)+1;
534 else if (blocksize_code == 7)
535 blocksize = get_bits(&s->gb, 16)+1;
537 blocksize = blocksize_table[blocksize_code];
539 if (blocksize > s->max_blocksize) {
540 av_log(s->avctx, AV_LOG_ERROR, "blocksize %d > %d\n", blocksize,
545 if (blocksize * s->channels * sizeof(int16_t) > alloc_data_size)
548 if (sample_rate_code == 0)
549 samplerate= s->samplerate;
550 else if (sample_rate_code < 12)
551 samplerate = sample_rate_table[sample_rate_code];
552 else if (sample_rate_code == 12)
553 samplerate = get_bits(&s->gb, 8) * 1000;
554 else if (sample_rate_code == 13)
555 samplerate = get_bits(&s->gb, 16);
556 else if (sample_rate_code == 14)
557 samplerate = get_bits(&s->gb, 16) * 10;
559 av_log(s->avctx, AV_LOG_ERROR, "illegal sample rate code %d\n",
564 skip_bits(&s->gb, 8);
565 crc8 = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0,
566 s->gb.buffer, get_bits_count(&s->gb)/8);
568 av_log(s->avctx, AV_LOG_ERROR, "header crc mismatch crc=%2X\n", crc8);
572 s->blocksize = blocksize;
573 s->samplerate = samplerate;
575 s->decorrelation= decorrelation;
577 // dump_headers(s->avctx, (FLACStreaminfo *)s);
580 for (i = 0; i < s->channels; i++) {
581 if (decode_subframe(s, i) < 0)
585 align_get_bits(&s->gb);
588 skip_bits(&s->gb, 16); /* data crc */
593 static int flac_decode_frame(AVCodecContext *avctx,
594 void *data, int *data_size,
595 const uint8_t *buf, int buf_size)
597 FLACContext *s = avctx->priv_data;
598 int tmp = 0, i, j = 0, input_buf_size = 0;
599 int16_t *samples_16 = data;
600 int32_t *samples_32 = data;
601 int alloc_data_size= *data_size;
605 if (s->max_framesize == 0) {
606 s->max_framesize= FFMAX(4, buf_size); // should hopefully be enough for the first header
607 s->bitstream= av_fast_realloc(s->bitstream, &s->allocated_bitstream_size, s->max_framesize);
610 if (1 && s->max_framesize) { //FIXME truncated
611 if (s->bitstream_size < 4 || AV_RL32(s->bitstream) != MKTAG('f','L','a','C'))
612 buf_size= FFMIN(buf_size, s->max_framesize - FFMIN(s->bitstream_size, s->max_framesize));
613 input_buf_size= buf_size;
615 if (s->bitstream_size + buf_size < buf_size || s->bitstream_index + s->bitstream_size + buf_size < s->bitstream_index)
618 if (s->allocated_bitstream_size < s->bitstream_size + buf_size)
619 s->bitstream= av_fast_realloc(s->bitstream, &s->allocated_bitstream_size, s->bitstream_size + buf_size);
621 if (s->bitstream_index + s->bitstream_size + buf_size > s->allocated_bitstream_size) {
622 memmove(s->bitstream, &s->bitstream[s->bitstream_index],
624 s->bitstream_index=0;
626 memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size],
628 buf= &s->bitstream[s->bitstream_index];
629 buf_size += s->bitstream_size;
630 s->bitstream_size= buf_size;
632 if (buf_size < s->max_framesize && input_buf_size) {
633 return input_buf_size;
637 init_get_bits(&s->gb, buf, buf_size*8);
639 if (metadata_parse(s))
642 tmp = show_bits(&s->gb, 16);
643 if ((tmp & 0xFFFE) != 0xFFF8) {
644 av_log(s->avctx, AV_LOG_ERROR, "FRAME HEADER not here\n");
645 while (get_bits_count(&s->gb)/8+2 < buf_size && (show_bits(&s->gb, 16) & 0xFFFE) != 0xFFF8)
646 skip_bits(&s->gb, 8);
647 goto end; // we may not have enough bits left to decode a frame, so try next time
649 skip_bits(&s->gb, 16);
650 if (decode_frame(s, alloc_data_size) < 0) {
651 av_log(s->avctx, AV_LOG_ERROR, "decode_frame() failed\n");
653 s->bitstream_index=0;
657 #define DECORRELATE(left, right)\
658 assert(s->channels == 2);\
659 for (i = 0; i < s->blocksize; i++) {\
660 int a= s->decoded[0][i];\
661 int b= s->decoded[1][i];\
663 *samples_32++ = (left) << s->sample_shift;\
664 *samples_32++ = (right) << s->sample_shift;\
666 *samples_16++ = (left) << s->sample_shift;\
667 *samples_16++ = (right) << s->sample_shift;\
672 switch (s->decorrelation) {
674 for (j = 0; j < s->blocksize; j++) {
675 for (i = 0; i < s->channels; i++) {
677 *samples_32++ = s->decoded[i][j] << s->sample_shift;
679 *samples_16++ = s->decoded[i][j] << s->sample_shift;
688 DECORRELATE( (a-=b>>1) + b, a)
691 *data_size = s->blocksize * s->channels * (s->is32 ? 4 : 2);
694 i= (get_bits_count(&s->gb)+7)/8;
696 av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
698 s->bitstream_index=0;
702 if (s->bitstream_size) {
703 s->bitstream_index += i;
704 s->bitstream_size -= i;
705 return input_buf_size;
710 static av_cold int flac_decode_close(AVCodecContext *avctx)
712 FLACContext *s = avctx->priv_data;
715 for (i = 0; i < s->channels; i++) {
716 av_freep(&s->decoded[i]);
718 av_freep(&s->bitstream);
723 static void flac_flush(AVCodecContext *avctx)
725 FLACContext *s = avctx->priv_data;
728 s->bitstream_index= 0;
731 AVCodec flac_decoder = {
742 .long_name= NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"),