3 * Copyright (c) 2001, 2002 Fabrice Bellard
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
27 #include "libavutil/attributes.h"
28 #include "libavutil/avassert.h"
29 #include "libavutil/channel_layout.h"
30 #include "libavutil/crc.h"
31 #include "libavutil/float_dsp.h"
32 #include "libavutil/libm.h"
33 #include "libavutil/thread.h"
38 #include "mpegaudiodsp.h"
42 * - test lsf / mpeg25 extensively.
45 #include "mpegaudio.h"
46 #include "mpegaudiodecheader.h"
48 #define BACKSTEP_SIZE 512
50 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES
52 /* layer 3 "granule" */
53 typedef struct GranuleDef {
58 int scalefac_compress;
63 uint8_t scalefac_scale;
64 uint8_t count1table_select;
65 int region_size[3]; /* number of huffman codes in each region */
67 int short_start, long_end; /* long/short band indexes */
68 uint8_t scale_factors[40];
69 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
72 typedef struct MPADecodeContext {
74 uint8_t last_buf[LAST_BUF_SIZE];
77 /* next header (used in free format parsing) */
78 uint32_t free_format_next_header;
81 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
82 int synth_buf_offset[MPA_MAX_CHANNELS];
83 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
84 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
85 GranuleDef granules[2][2]; /* Used in Layer 3 */
86 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
89 AVCodecContext* avctx;
91 void (*butterflies_float)(float *av_restrict v1, float *av_restrict v2, int len);
98 #include "mpegaudiodata.h"
100 #include "mpegaudio_tablegen.h"
101 /* intensity stereo coef table */
102 static INTFLOAT is_table_lsf[2][2][16];
104 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
105 static int32_t scale_factor_mult[15][3];
106 /* mult table for layer 2 group quantization */
108 #define SCALE_GEN(v) \
109 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
111 static const int32_t scale_factor_mult2[3][3] = {
112 SCALE_GEN(4.0 / 3.0), /* 3 steps */
113 SCALE_GEN(4.0 / 5.0), /* 5 steps */
114 SCALE_GEN(4.0 / 9.0), /* 9 steps */
118 * Convert region offsets to region sizes and truncate
119 * size to big_values.
121 static void region_offset2size(GranuleDef *g)
124 g->region_size[2] = 576 / 2;
125 for (i = 0; i < 3; i++) {
126 k = FFMIN(g->region_size[i], g->big_values);
127 g->region_size[i] = k - j;
132 static void init_short_region(MPADecodeContext *s, GranuleDef *g)
134 if (g->block_type == 2) {
135 if (s->sample_rate_index != 8)
136 g->region_size[0] = (36 / 2);
138 g->region_size[0] = (72 / 2);
140 if (s->sample_rate_index <= 2)
141 g->region_size[0] = (36 / 2);
142 else if (s->sample_rate_index != 8)
143 g->region_size[0] = (54 / 2);
145 g->region_size[0] = (108 / 2);
147 g->region_size[1] = (576 / 2);
150 static void init_long_region(MPADecodeContext *s, GranuleDef *g,
154 g->region_size[0] = ff_band_index_long[s->sample_rate_index][ra1 + 1];
155 /* should not overflow */
156 l = FFMIN(ra1 + ra2 + 2, 22);
157 g->region_size[1] = ff_band_index_long[s->sample_rate_index][ l];
160 static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
162 if (g->block_type == 2) {
163 if (g->switch_point) {
164 if(s->sample_rate_index == 8)
165 avpriv_request_sample(s->avctx, "switch point in 8khz");
166 /* if switched mode, we handle the 36 first samples as
167 long blocks. For 8000Hz, we handle the 72 first
168 exponents as long blocks */
169 if (s->sample_rate_index <= 2)
185 /* layer 1 unscaling */
186 /* n = number of bits of the mantissa minus 1 */
187 static inline int l1_unscale(int n, int mant, int scale_factor)
192 shift = ff_scale_factor_modshift[scale_factor];
195 val = MUL64((int)(mant + (-1U << n) + 1), scale_factor_mult[n-1][mod]);
197 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
198 return (int)((val + (1LL << (shift - 1))) >> shift);
201 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
205 shift = ff_scale_factor_modshift[scale_factor];
209 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
210 /* NOTE: at this point, 0 <= shift <= 21 */
212 val = (val + (1 << (shift - 1))) >> shift;
216 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
217 static inline int l3_unscale(int value, int exponent)
222 e = ff_table_4_3_exp [4 * value + (exponent & 3)];
223 m = ff_table_4_3_value[4 * value + (exponent & 3)];
227 av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e);
231 m = (m + ((1U << e) >> 1)) >> e;
236 static av_cold void decode_init_static(void)
240 /* scale factor multiply for layer 1 */
241 for (i = 0; i < 15; i++) {
244 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
245 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
246 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
247 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
248 ff_dlog(NULL, "%d: norm=%x s=%"PRIx32" %"PRIx32" %"PRIx32"\n", i,
250 scale_factor_mult[i][0],
251 scale_factor_mult[i][1],
252 scale_factor_mult[i][2]);
255 /* compute n ^ (4/3) and store it in mantissa/exp format */
257 mpegaudio_tableinit();
259 for (i = 0; i < 16; i++) {
263 for (j = 0; j < 2; j++) {
264 e = -(j + 1) * ((i + 1) >> 1);
267 is_table_lsf[j][k ^ 1][i] = FIXR(f);
268 is_table_lsf[j][k ][i] = FIXR(1.0);
269 ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
270 i, j, (float) is_table_lsf[j][0][i],
271 (float) is_table_lsf[j][1][i]);
274 RENAME(ff_mpa_synth_init)();
275 ff_mpegaudiodec_common_init_static();
278 static av_cold int decode_init(AVCodecContext * avctx)
280 static AVOnce init_static_once = AV_ONCE_INIT;
281 MPADecodeContext *s = avctx->priv_data;
287 AVFloatDSPContext *fdsp;
288 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
290 return AVERROR(ENOMEM);
291 s->butterflies_float = fdsp->butterflies_float;
296 ff_mpadsp_init(&s->mpadsp);
298 if (avctx->request_sample_fmt == OUT_FMT &&
299 avctx->codec_id != AV_CODEC_ID_MP3ON4)
300 avctx->sample_fmt = OUT_FMT;
302 avctx->sample_fmt = OUT_FMT_P;
303 s->err_recognition = avctx->err_recognition;
305 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
308 ff_thread_once(&init_static_once, decode_init_static);
313 #define C3 FIXHR(0.86602540378443864676/2)
314 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
315 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
316 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
318 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
320 static void imdct12(INTFLOAT *out, SUINTFLOAT *in)
322 SUINTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
325 in1 = in[1*3] + in[0*3];
326 in2 = in[2*3] + in[1*3];
327 in3 = in[3*3] + in[2*3];
328 in4 = in[4*3] + in[3*3];
329 in5 = in[5*3] + in[4*3];
333 in2 = MULH3(in2, C3, 2);
334 in3 = MULH3(in3, C3, 4);
337 t2 = MULH3(in1 - in5, C4, 2);
347 in1 = MULH3(in5 + in3, C5, 1);
354 in5 = MULH3(in5 - in3, C6, 2);
361 static int handle_crc(MPADecodeContext *s, int sec_len)
363 if (s->error_protection && (s->err_recognition & AV_EF_CRCCHECK)) {
364 const uint8_t *buf = s->gb.buffer - HEADER_SIZE;
365 int sec_byte_len = sec_len >> 3;
366 int sec_rem_bits = sec_len & 7;
367 const AVCRC *crc_tab = av_crc_get_table(AV_CRC_16_ANSI);
369 uint32_t crc_val = av_crc(crc_tab, UINT16_MAX, &buf[2], 2);
370 crc_val = av_crc(crc_tab, crc_val, &buf[6], sec_byte_len);
373 ((buf[6 + sec_byte_len] & (0xFF00 >> sec_rem_bits)) << 24) +
374 ((s->crc << 16) >> sec_rem_bits));
376 crc_val = av_crc(crc_tab, crc_val, tmp_buf, 3);
379 av_log(s->avctx, AV_LOG_ERROR, "CRC mismatch %X!\n", crc_val);
380 if (s->err_recognition & AV_EF_EXPLODE)
381 return AVERROR_INVALIDDATA;
387 /* return the number of decoded frames */
388 static int mp_decode_layer1(MPADecodeContext *s)
390 int bound, i, v, n, ch, j, mant;
391 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
392 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
395 ret = handle_crc(s, (s->nb_channels == 1) ? 8*16 : 8*32);
399 if (s->mode == MPA_JSTEREO)
400 bound = (s->mode_ext + 1) * 4;
404 /* allocation bits */
405 for (i = 0; i < bound; i++) {
406 for (ch = 0; ch < s->nb_channels; ch++) {
407 allocation[ch][i] = get_bits(&s->gb, 4);
410 for (i = bound; i < SBLIMIT; i++)
411 allocation[0][i] = get_bits(&s->gb, 4);
414 for (i = 0; i < bound; i++) {
415 for (ch = 0; ch < s->nb_channels; ch++) {
416 if (allocation[ch][i])
417 scale_factors[ch][i] = get_bits(&s->gb, 6);
420 for (i = bound; i < SBLIMIT; i++) {
421 if (allocation[0][i]) {
422 scale_factors[0][i] = get_bits(&s->gb, 6);
423 scale_factors[1][i] = get_bits(&s->gb, 6);
427 /* compute samples */
428 for (j = 0; j < 12; j++) {
429 for (i = 0; i < bound; i++) {
430 for (ch = 0; ch < s->nb_channels; ch++) {
431 n = allocation[ch][i];
433 mant = get_bits(&s->gb, n + 1);
434 v = l1_unscale(n, mant, scale_factors[ch][i]);
438 s->sb_samples[ch][j][i] = v;
441 for (i = bound; i < SBLIMIT; i++) {
442 n = allocation[0][i];
444 mant = get_bits(&s->gb, n + 1);
445 v = l1_unscale(n, mant, scale_factors[0][i]);
446 s->sb_samples[0][j][i] = v;
447 v = l1_unscale(n, mant, scale_factors[1][i]);
448 s->sb_samples[1][j][i] = v;
450 s->sb_samples[0][j][i] = 0;
451 s->sb_samples[1][j][i] = 0;
458 static int mp_decode_layer2(MPADecodeContext *s)
460 int sblimit; /* number of used subbands */
461 const unsigned char *alloc_table;
462 int table, bit_alloc_bits, i, j, ch, bound, v;
463 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
464 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
465 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
466 int scale, qindex, bits, steps, k, l, m, b;
469 /* select decoding table */
470 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
471 s->sample_rate, s->lsf);
472 sblimit = ff_mpa_sblimit_table[table];
473 alloc_table = ff_mpa_alloc_tables[table];
475 if (s->mode == MPA_JSTEREO)
476 bound = (s->mode_ext + 1) * 4;
480 ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
486 /* parse bit allocation */
488 for (i = 0; i < bound; i++) {
489 bit_alloc_bits = alloc_table[j];
490 for (ch = 0; ch < s->nb_channels; ch++)
491 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
492 j += 1 << bit_alloc_bits;
494 for (i = bound; i < sblimit; i++) {
495 bit_alloc_bits = alloc_table[j];
496 v = get_bits(&s->gb, bit_alloc_bits);
499 j += 1 << bit_alloc_bits;
503 for (i = 0; i < sblimit; i++) {
504 for (ch = 0; ch < s->nb_channels; ch++) {
505 if (bit_alloc[ch][i])
506 scale_code[ch][i] = get_bits(&s->gb, 2);
510 ret = handle_crc(s, get_bits_count(&s->gb) - 16);
515 for (i = 0; i < sblimit; i++) {
516 for (ch = 0; ch < s->nb_channels; ch++) {
517 if (bit_alloc[ch][i]) {
518 sf = scale_factors[ch][i];
519 switch (scale_code[ch][i]) {
522 sf[0] = get_bits(&s->gb, 6);
523 sf[1] = get_bits(&s->gb, 6);
524 sf[2] = get_bits(&s->gb, 6);
527 sf[0] = get_bits(&s->gb, 6);
532 sf[0] = get_bits(&s->gb, 6);
533 sf[2] = get_bits(&s->gb, 6);
537 sf[0] = get_bits(&s->gb, 6);
538 sf[2] = get_bits(&s->gb, 6);
547 for (k = 0; k < 3; k++) {
548 for (l = 0; l < 12; l += 3) {
550 for (i = 0; i < bound; i++) {
551 bit_alloc_bits = alloc_table[j];
552 for (ch = 0; ch < s->nb_channels; ch++) {
553 b = bit_alloc[ch][i];
555 scale = scale_factors[ch][i][k];
556 qindex = alloc_table[j+b];
557 bits = ff_mpa_quant_bits[qindex];
560 /* 3 values at the same time */
561 v = get_bits(&s->gb, -bits);
562 v2 = ff_division_tabs[qindex][v];
563 steps = ff_mpa_quant_steps[qindex];
565 s->sb_samples[ch][k * 12 + l + 0][i] =
566 l2_unscale_group(steps, v2 & 15, scale);
567 s->sb_samples[ch][k * 12 + l + 1][i] =
568 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
569 s->sb_samples[ch][k * 12 + l + 2][i] =
570 l2_unscale_group(steps, v2 >> 8 , scale);
572 for (m = 0; m < 3; m++) {
573 v = get_bits(&s->gb, bits);
574 v = l1_unscale(bits - 1, v, scale);
575 s->sb_samples[ch][k * 12 + l + m][i] = v;
579 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
580 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
581 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
584 /* next subband in alloc table */
585 j += 1 << bit_alloc_bits;
587 /* XXX: find a way to avoid this duplication of code */
588 for (i = bound; i < sblimit; i++) {
589 bit_alloc_bits = alloc_table[j];
592 int mant, scale0, scale1;
593 scale0 = scale_factors[0][i][k];
594 scale1 = scale_factors[1][i][k];
595 qindex = alloc_table[j + b];
596 bits = ff_mpa_quant_bits[qindex];
598 /* 3 values at the same time */
599 v = get_bits(&s->gb, -bits);
600 steps = ff_mpa_quant_steps[qindex];
603 s->sb_samples[0][k * 12 + l + 0][i] =
604 l2_unscale_group(steps, mant, scale0);
605 s->sb_samples[1][k * 12 + l + 0][i] =
606 l2_unscale_group(steps, mant, scale1);
609 s->sb_samples[0][k * 12 + l + 1][i] =
610 l2_unscale_group(steps, mant, scale0);
611 s->sb_samples[1][k * 12 + l + 1][i] =
612 l2_unscale_group(steps, mant, scale1);
613 s->sb_samples[0][k * 12 + l + 2][i] =
614 l2_unscale_group(steps, v, scale0);
615 s->sb_samples[1][k * 12 + l + 2][i] =
616 l2_unscale_group(steps, v, scale1);
618 for (m = 0; m < 3; m++) {
619 mant = get_bits(&s->gb, bits);
620 s->sb_samples[0][k * 12 + l + m][i] =
621 l1_unscale(bits - 1, mant, scale0);
622 s->sb_samples[1][k * 12 + l + m][i] =
623 l1_unscale(bits - 1, mant, scale1);
627 s->sb_samples[0][k * 12 + l + 0][i] = 0;
628 s->sb_samples[0][k * 12 + l + 1][i] = 0;
629 s->sb_samples[0][k * 12 + l + 2][i] = 0;
630 s->sb_samples[1][k * 12 + l + 0][i] = 0;
631 s->sb_samples[1][k * 12 + l + 1][i] = 0;
632 s->sb_samples[1][k * 12 + l + 2][i] = 0;
634 /* next subband in alloc table */
635 j += 1 << bit_alloc_bits;
637 /* fill remaining samples to zero */
638 for (i = sblimit; i < SBLIMIT; i++) {
639 for (ch = 0; ch < s->nb_channels; ch++) {
640 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
641 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
642 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
650 #define SPLIT(dst,sf,n) \
652 int m = (sf * 171) >> 9; \
655 } else if (n == 4) { \
658 } else if (n == 5) { \
659 int m = (sf * 205) >> 10; \
662 } else if (n == 6) { \
663 int m = (sf * 171) >> 10; \
670 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
673 SPLIT(slen[3], sf, n3)
674 SPLIT(slen[2], sf, n2)
675 SPLIT(slen[1], sf, n1)
679 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
682 const uint8_t *bstab, *pretab;
683 int len, i, j, k, l, v0, shift, gain, gains[3];
687 gain = g->global_gain - 210;
688 shift = g->scalefac_scale + 1;
690 bstab = ff_band_size_long[s->sample_rate_index];
691 pretab = ff_mpa_pretab[g->preflag];
692 for (i = 0; i < g->long_end; i++) {
693 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
695 for (j = len; j > 0; j--)
699 if (g->short_start < 13) {
700 bstab = ff_band_size_short[s->sample_rate_index];
701 gains[0] = gain - (g->subblock_gain[0] << 3);
702 gains[1] = gain - (g->subblock_gain[1] << 3);
703 gains[2] = gain - (g->subblock_gain[2] << 3);
705 for (i = g->short_start; i < 13; i++) {
707 for (l = 0; l < 3; l++) {
708 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
709 for (j = len; j > 0; j--)
716 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
719 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits - s->extrasize * 8) {
721 s->in_gb.buffer = NULL;
723 av_assert2((get_bits_count(&s->gb) & 7) == 0);
724 skip_bits_long(&s->gb, *pos - *end_pos);
726 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
727 *pos = get_bits_count(&s->gb);
731 /* Following is an optimized code for
733 if(get_bits1(&s->gb))
738 #define READ_FLIP_SIGN(dst,src) \
739 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
742 #define READ_FLIP_SIGN(dst,src) \
743 v = -get_bits1(&s->gb); \
744 *(dst) = (*(src) ^ v) - v;
747 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
748 int16_t *exponents, int end_pos2)
752 int last_pos, bits_left;
754 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits - s->extrasize * 8);
756 /* low frequencies (called big values) */
758 for (i = 0; i < 3; i++) {
759 int j, k, l, linbits;
760 j = g->region_size[i];
763 /* select vlc table */
764 k = g->table_select[i];
765 l = ff_mpa_huff_data[k][0];
766 linbits = ff_mpa_huff_data[k][1];
767 vlc = &ff_huff_vlc[l];
770 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
775 /* read huffcode and compute each couple */
779 int pos = get_bits_count(&s->gb);
782 switch_buffer(s, &pos, &end_pos, &end_pos2);
786 y = get_vlc2(&s->gb, vlc->table, 7, 3);
789 g->sb_hybrid[s_index ] =
790 g->sb_hybrid[s_index + 1] = 0;
795 exponent= exponents[s_index];
797 ff_dlog(s->avctx, "region=%d n=%d y=%d exp=%d\n",
798 i, g->region_size[i] - j, y, exponent);
803 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
805 x += get_bitsz(&s->gb, linbits);
806 v = l3_unscale(x, exponent);
807 if (get_bits1(&s->gb))
809 g->sb_hybrid[s_index] = v;
812 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
814 y += get_bitsz(&s->gb, linbits);
815 v = l3_unscale(y, exponent);
816 if (get_bits1(&s->gb))
818 g->sb_hybrid[s_index + 1] = v;
825 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
827 x += get_bitsz(&s->gb, linbits);
828 v = l3_unscale(x, exponent);
829 if (get_bits1(&s->gb))
831 g->sb_hybrid[s_index+!!y] = v;
833 g->sb_hybrid[s_index + !y] = 0;
839 /* high frequencies */
840 vlc = &ff_huff_quad_vlc[g->count1table_select];
842 while (s_index <= 572) {
844 pos = get_bits_count(&s->gb);
845 if (pos >= end_pos) {
846 if (pos > end_pos2 && last_pos) {
847 /* some encoders generate an incorrect size for this
848 part. We must go back into the data */
850 skip_bits_long(&s->gb, last_pos - pos);
851 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
852 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
856 switch_buffer(s, &pos, &end_pos, &end_pos2);
862 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
863 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
864 g->sb_hybrid[s_index + 0] =
865 g->sb_hybrid[s_index + 1] =
866 g->sb_hybrid[s_index + 2] =
867 g->sb_hybrid[s_index + 3] = 0;
869 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
871 int pos = s_index + idxtab[code];
872 code ^= 8 >> idxtab[code];
873 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
877 /* skip extension bits */
878 bits_left = end_pos2 - get_bits_count(&s->gb);
879 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
880 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
882 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
883 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
886 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
887 skip_bits_long(&s->gb, bits_left);
889 i = get_bits_count(&s->gb);
890 switch_buffer(s, &i, &end_pos, &end_pos2);
895 /* Reorder short blocks from bitstream order to interleaved order. It
896 would be faster to do it in parsing, but the code would be far more
898 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
901 INTFLOAT *ptr, *dst, *ptr1;
904 if (g->block_type != 2)
907 if (g->switch_point) {
908 if (s->sample_rate_index != 8)
909 ptr = g->sb_hybrid + 36;
911 ptr = g->sb_hybrid + 72;
916 for (i = g->short_start; i < 13; i++) {
917 len = ff_band_size_short[s->sample_rate_index][i];
920 for (j = len; j > 0; j--) {
927 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
931 #define ISQRT2 FIXR(0.70710678118654752440)
933 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
936 int sf_max, sf, len, non_zero_found;
937 INTFLOAT *tab0, *tab1, v1, v2;
938 const INTFLOAT (*is_tab)[16];
939 SUINTFLOAT tmp0, tmp1;
940 int non_zero_found_short[3];
942 /* intensity stereo */
943 if (s->mode_ext & MODE_EXT_I_STEREO) {
948 is_tab = is_table_lsf[g1->scalefac_compress & 1];
952 tab0 = g0->sb_hybrid + 576;
953 tab1 = g1->sb_hybrid + 576;
955 non_zero_found_short[0] = 0;
956 non_zero_found_short[1] = 0;
957 non_zero_found_short[2] = 0;
958 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
959 for (i = 12; i >= g1->short_start; i--) {
960 /* for last band, use previous scale factor */
963 len = ff_band_size_short[s->sample_rate_index][i];
964 for (l = 2; l >= 0; l--) {
967 if (!non_zero_found_short[l]) {
968 /* test if non zero band. if so, stop doing i-stereo */
969 for (j = 0; j < len; j++) {
971 non_zero_found_short[l] = 1;
975 sf = g1->scale_factors[k + l];
981 for (j = 0; j < len; j++) {
983 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
984 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
988 if (s->mode_ext & MODE_EXT_MS_STEREO) {
989 /* lower part of the spectrum : do ms stereo
991 for (j = 0; j < len; j++) {
994 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
995 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1002 non_zero_found = non_zero_found_short[0] |
1003 non_zero_found_short[1] |
1004 non_zero_found_short[2];
1006 for (i = g1->long_end - 1;i >= 0;i--) {
1007 len = ff_band_size_long[s->sample_rate_index][i];
1010 /* test if non zero band. if so, stop doing i-stereo */
1011 if (!non_zero_found) {
1012 for (j = 0; j < len; j++) {
1018 /* for last band, use previous scale factor */
1019 k = (i == 21) ? 20 : i;
1020 sf = g1->scale_factors[k];
1025 for (j = 0; j < len; j++) {
1027 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1028 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1032 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1033 /* lower part of the spectrum : do ms stereo
1035 for (j = 0; j < len; j++) {
1038 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1039 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1044 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1045 /* ms stereo ONLY */
1046 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1049 s->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1051 tab0 = g0->sb_hybrid;
1052 tab1 = g1->sb_hybrid;
1053 for (i = 0; i < 576; i++) {
1056 tab0[i] = tmp0 + tmp1;
1057 tab1[i] = tmp0 - tmp1;
1065 # include "mips/compute_antialias_float.h"
1066 #endif /* HAVE_MIPSFPU */
1069 # include "mips/compute_antialias_fixed.h"
1070 #endif /* HAVE_MIPSDSP */
1071 #endif /* USE_FLOATS */
1073 #ifndef compute_antialias
1075 #define AA(j) do { \
1076 float tmp0 = ptr[-1-j]; \
1077 float tmp1 = ptr[ j]; \
1078 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1079 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1082 #define AA(j) do { \
1083 SUINT tmp0 = ptr[-1-j]; \
1084 SUINT tmp1 = ptr[ j]; \
1085 SUINT tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1086 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1087 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1091 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1096 /* we antialias only "long" bands */
1097 if (g->block_type == 2) {
1098 if (!g->switch_point)
1100 /* XXX: check this for 8000Hz case */
1106 ptr = g->sb_hybrid + 18;
1107 for (i = n; i > 0; i--) {
1120 #endif /* compute_antialias */
1122 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1123 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1125 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1127 int i, j, mdct_long_end, sblimit;
1129 /* find last non zero block */
1130 ptr = g->sb_hybrid + 576;
1131 ptr1 = g->sb_hybrid + 2 * 18;
1132 while (ptr >= ptr1) {
1136 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1139 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1141 if (g->block_type == 2) {
1142 /* XXX: check for 8000 Hz */
1143 if (g->switch_point)
1148 mdct_long_end = sblimit;
1151 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1152 mdct_long_end, g->switch_point,
1155 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1156 ptr = g->sb_hybrid + 18 * mdct_long_end;
1158 for (j = mdct_long_end; j < sblimit; j++) {
1159 /* select frequency inversion */
1160 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1161 out_ptr = sb_samples + j;
1163 for (i = 0; i < 6; i++) {
1164 *out_ptr = buf[4*i];
1167 imdct12(out2, ptr + 0);
1168 for (i = 0; i < 6; i++) {
1169 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1170 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1173 imdct12(out2, ptr + 1);
1174 for (i = 0; i < 6; i++) {
1175 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1176 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1179 imdct12(out2, ptr + 2);
1180 for (i = 0; i < 6; i++) {
1181 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1182 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1183 buf[4*(i + 6*2)] = 0;
1186 buf += (j&3) != 3 ? 1 : (4*18-3);
1189 for (j = sblimit; j < SBLIMIT; j++) {
1191 out_ptr = sb_samples + j;
1192 for (i = 0; i < 18; i++) {
1193 *out_ptr = buf[4*i];
1197 buf += (j&3) != 3 ? 1 : (4*18-3);
1201 /* main layer3 decoding function */
1202 static int mp_decode_layer3(MPADecodeContext *s)
1204 int nb_granules, main_data_begin;
1205 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1207 int16_t exponents[576]; //FIXME try INTFLOAT
1210 /* read side info */
1212 ret = handle_crc(s, ((s->nb_channels == 1) ? 8*9 : 8*17));
1213 main_data_begin = get_bits(&s->gb, 8);
1214 skip_bits(&s->gb, s->nb_channels);
1217 ret = handle_crc(s, ((s->nb_channels == 1) ? 8*17 : 8*32));
1218 main_data_begin = get_bits(&s->gb, 9);
1219 if (s->nb_channels == 2)
1220 skip_bits(&s->gb, 3);
1222 skip_bits(&s->gb, 5);
1224 for (ch = 0; ch < s->nb_channels; ch++) {
1225 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1226 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1232 for (gr = 0; gr < nb_granules; gr++) {
1233 for (ch = 0; ch < s->nb_channels; ch++) {
1234 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1235 g = &s->granules[ch][gr];
1236 g->part2_3_length = get_bits(&s->gb, 12);
1237 g->big_values = get_bits(&s->gb, 9);
1238 if (g->big_values > 288) {
1239 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1240 return AVERROR_INVALIDDATA;
1243 g->global_gain = get_bits(&s->gb, 8);
1244 /* if MS stereo only is selected, we precompute the
1245 1/sqrt(2) renormalization factor */
1246 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1248 g->global_gain -= 2;
1250 g->scalefac_compress = get_bits(&s->gb, 9);
1252 g->scalefac_compress = get_bits(&s->gb, 4);
1253 blocksplit_flag = get_bits1(&s->gb);
1254 if (blocksplit_flag) {
1255 g->block_type = get_bits(&s->gb, 2);
1256 if (g->block_type == 0) {
1257 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1258 return AVERROR_INVALIDDATA;
1260 g->switch_point = get_bits1(&s->gb);
1261 for (i = 0; i < 2; i++)
1262 g->table_select[i] = get_bits(&s->gb, 5);
1263 for (i = 0; i < 3; i++)
1264 g->subblock_gain[i] = get_bits(&s->gb, 3);
1265 init_short_region(s, g);
1267 int region_address1, region_address2;
1269 g->switch_point = 0;
1270 for (i = 0; i < 3; i++)
1271 g->table_select[i] = get_bits(&s->gb, 5);
1272 /* compute huffman coded region sizes */
1273 region_address1 = get_bits(&s->gb, 4);
1274 region_address2 = get_bits(&s->gb, 3);
1275 ff_dlog(s->avctx, "region1=%d region2=%d\n",
1276 region_address1, region_address2);
1277 init_long_region(s, g, region_address1, region_address2);
1279 region_offset2size(g);
1280 compute_band_indexes(s, g);
1284 g->preflag = get_bits1(&s->gb);
1285 g->scalefac_scale = get_bits1(&s->gb);
1286 g->count1table_select = get_bits1(&s->gb);
1287 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1288 g->block_type, g->switch_point);
1294 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb) >> 3);
1295 s->extrasize = av_clip((get_bits_left(&s->gb) >> 3) - s->extrasize, 0,
1296 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1297 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1298 /* now we get bits from the main_data_begin offset */
1299 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1300 main_data_begin, s->last_buf_size);
1302 memcpy(s->last_buf + s->last_buf_size, ptr, s->extrasize);
1304 init_get_bits(&s->gb, s->last_buf, (s->last_buf_size + s->extrasize) * 8);
1305 s->last_buf_size <<= 3;
1306 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1307 for (ch = 0; ch < s->nb_channels; ch++) {
1308 g = &s->granules[ch][gr];
1309 s->last_buf_size += g->part2_3_length;
1310 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1311 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1314 skip = s->last_buf_size - 8 * main_data_begin;
1315 if (skip >= s->gb.size_in_bits - s->extrasize * 8 && s->in_gb.buffer) {
1316 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits + s->extrasize * 8);
1318 s->in_gb.buffer = NULL;
1321 skip_bits_long(&s->gb, skip);
1328 for (; gr < nb_granules; gr++) {
1329 for (ch = 0; ch < s->nb_channels; ch++) {
1330 g = &s->granules[ch][gr];
1331 bits_pos = get_bits_count(&s->gb);
1335 int slen, slen1, slen2;
1337 /* MPEG-1 scale factors */
1338 slen1 = ff_slen_table[0][g->scalefac_compress];
1339 slen2 = ff_slen_table[1][g->scalefac_compress];
1340 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1341 if (g->block_type == 2) {
1342 n = g->switch_point ? 17 : 18;
1345 for (i = 0; i < n; i++)
1346 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1348 for (i = 0; i < n; i++)
1349 g->scale_factors[j++] = 0;
1352 for (i = 0; i < 18; i++)
1353 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1354 for (i = 0; i < 3; i++)
1355 g->scale_factors[j++] = 0;
1357 for (i = 0; i < 21; i++)
1358 g->scale_factors[j++] = 0;
1361 sc = s->granules[ch][0].scale_factors;
1363 for (k = 0; k < 4; k++) {
1365 if ((g->scfsi & (0x8 >> k)) == 0) {
1366 slen = (k < 2) ? slen1 : slen2;
1368 for (i = 0; i < n; i++)
1369 g->scale_factors[j++] = get_bits(&s->gb, slen);
1371 for (i = 0; i < n; i++)
1372 g->scale_factors[j++] = 0;
1375 /* simply copy from last granule */
1376 for (i = 0; i < n; i++) {
1377 g->scale_factors[j] = sc[j];
1382 g->scale_factors[j++] = 0;
1385 int tindex, tindex2, slen[4], sl, sf;
1387 /* LSF scale factors */
1388 if (g->block_type == 2)
1389 tindex = g->switch_point ? 2 : 1;
1393 sf = g->scalefac_compress;
1394 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1395 /* intensity stereo case */
1398 lsf_sf_expand(slen, sf, 6, 6, 0);
1400 } else if (sf < 244) {
1401 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1404 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1410 lsf_sf_expand(slen, sf, 5, 4, 4);
1412 } else if (sf < 500) {
1413 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1416 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1423 for (k = 0; k < 4; k++) {
1424 n = ff_lsf_nsf_table[tindex2][tindex][k];
1427 for (i = 0; i < n; i++)
1428 g->scale_factors[j++] = get_bits(&s->gb, sl);
1430 for (i = 0; i < n; i++)
1431 g->scale_factors[j++] = 0;
1434 /* XXX: should compute exact size */
1436 g->scale_factors[j] = 0;
1439 exponents_from_scale_factors(s, g, exponents);
1441 /* read Huffman coded residue */
1442 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1445 if (s->mode == MPA_JSTEREO)
1446 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1448 for (ch = 0; ch < s->nb_channels; ch++) {
1449 g = &s->granules[ch][gr];
1451 reorder_block(s, g);
1452 compute_antialias(s, g);
1453 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1456 if (get_bits_count(&s->gb) < 0)
1457 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1458 return nb_granules * 18;
1461 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1462 const uint8_t *buf, int buf_size)
1464 int i, nb_frames, ch, ret;
1465 OUT_INT *samples_ptr;
1467 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1468 if (s->error_protection)
1469 s->crc = get_bits(&s->gb, 16);
1473 s->avctx->frame_size = 384;
1474 nb_frames = mp_decode_layer1(s);
1477 s->avctx->frame_size = 1152;
1478 nb_frames = mp_decode_layer2(s);
1481 s->avctx->frame_size = s->lsf ? 576 : 1152;
1483 nb_frames = mp_decode_layer3(s);
1486 if (s->in_gb.buffer) {
1487 align_get_bits(&s->gb);
1488 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1489 if (i >= 0 && i <= BACKSTEP_SIZE) {
1490 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb) >> 3), i);
1493 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1495 s->in_gb.buffer = NULL;
1499 align_get_bits(&s->gb);
1500 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1501 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1502 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1504 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1505 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1507 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1508 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1509 s->last_buf_size += i;
1515 /* get output buffer */
1517 av_assert0(s->frame);
1518 s->frame->nb_samples = s->avctx->frame_size;
1519 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)
1521 samples = (OUT_INT **)s->frame->extended_data;
1524 /* apply the synthesis filter */
1525 for (ch = 0; ch < s->nb_channels; ch++) {
1527 if (s->avctx->sample_fmt == OUT_FMT_P) {
1528 samples_ptr = samples[ch];
1531 samples_ptr = samples[0] + ch;
1532 sample_stride = s->nb_channels;
1534 for (i = 0; i < nb_frames; i++) {
1535 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1536 &(s->synth_buf_offset[ch]),
1537 RENAME(ff_mpa_synth_window),
1538 &s->dither_state, samples_ptr,
1539 sample_stride, s->sb_samples[ch][i]);
1540 samples_ptr += 32 * sample_stride;
1544 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1547 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1550 const uint8_t *buf = avpkt->data;
1551 int buf_size = avpkt->size;
1552 MPADecodeContext *s = avctx->priv_data;
1557 while(buf_size && !*buf){
1563 if (buf_size < HEADER_SIZE)
1564 return AVERROR_INVALIDDATA;
1566 header = AV_RB32(buf);
1567 if (header >> 8 == AV_RB32("TAG") >> 8) {
1568 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1569 return buf_size + skipped;
1571 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1573 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1574 return AVERROR_INVALIDDATA;
1575 } else if (ret == 1) {
1576 /* free format: prepare to compute frame size */
1578 return AVERROR_INVALIDDATA;
1580 /* update codec info */
1581 avctx->channels = s->nb_channels;
1582 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1583 if (!avctx->bit_rate)
1584 avctx->bit_rate = s->bit_rate;
1586 if (s->frame_size <= 0) {
1587 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1588 return AVERROR_INVALIDDATA;
1589 } else if (s->frame_size < buf_size) {
1590 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1591 buf_size= s->frame_size;
1596 ret = mp_decode_frame(s, NULL, buf, buf_size);
1598 s->frame->nb_samples = avctx->frame_size;
1600 avctx->sample_rate = s->sample_rate;
1601 //FIXME maybe move the other codec info stuff from above here too
1603 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1604 /* Only return an error if the bad frame makes up the whole packet or
1605 * the error is related to buffer management.
1606 * If there is more data in the packet, just consume the bad frame
1607 * instead of returning an error, which would discard the whole
1610 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1614 return buf_size + skipped;
1617 static void mp_flush(MPADecodeContext *ctx)
1619 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1620 memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));
1621 ctx->last_buf_size = 0;
1622 ctx->dither_state = 0;
1625 static void flush(AVCodecContext *avctx)
1627 mp_flush(avctx->priv_data);
1630 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1631 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1632 int *got_frame_ptr, AVPacket *avpkt)
1634 const uint8_t *buf = avpkt->data;
1635 int buf_size = avpkt->size;
1636 MPADecodeContext *s = avctx->priv_data;
1639 int av_unused out_size;
1643 // Discard too short frames
1644 if (buf_size < HEADER_SIZE) {
1645 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1646 return AVERROR_INVALIDDATA;
1650 if (len > MPA_MAX_CODED_FRAME_SIZE)
1651 len = MPA_MAX_CODED_FRAME_SIZE;
1653 // Get header and restore sync word
1654 header = AV_RB32(buf) | 0xffe00000;
1656 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1658 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1661 /* update codec info */
1662 avctx->sample_rate = s->sample_rate;
1663 avctx->channels = s->nb_channels;
1664 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1665 if (!avctx->bit_rate)
1666 avctx->bit_rate = s->bit_rate;
1668 s->frame_size = len;
1672 ret = mp_decode_frame(s, NULL, buf, buf_size);
1674 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1682 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1684 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1687 * Context for MP3On4 decoder
1689 typedef struct MP3On4DecodeContext {
1690 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1691 int syncword; ///< syncword patch
1692 const uint8_t *coff; ///< channel offsets in output buffer
1693 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1694 } MP3On4DecodeContext;
1696 #include "mpeg4audio.h"
1698 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1700 /* number of mp3 decoder instances */
1701 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1703 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1704 static const uint8_t chan_offset[8][5] = {
1709 { 2, 0, 3 }, // C FLR BS
1710 { 2, 0, 3 }, // C FLR BLRS
1711 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1712 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1715 /* mp3on4 channel layouts */
1716 static const int16_t chan_layout[8] = {
1719 AV_CH_LAYOUT_STEREO,
1720 AV_CH_LAYOUT_SURROUND,
1721 AV_CH_LAYOUT_4POINT0,
1722 AV_CH_LAYOUT_5POINT0,
1723 AV_CH_LAYOUT_5POINT1,
1724 AV_CH_LAYOUT_7POINT1
1727 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1729 MP3On4DecodeContext *s = avctx->priv_data;
1732 for (i = 0; i < s->frames; i++)
1733 av_freep(&s->mp3decctx[i]);
1739 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1741 MP3On4DecodeContext *s = avctx->priv_data;
1742 MPEG4AudioConfig cfg;
1745 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1746 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1747 return AVERROR_INVALIDDATA;
1750 avpriv_mpeg4audio_get_config2(&cfg, avctx->extradata,
1751 avctx->extradata_size, 1, avctx);
1752 if (!cfg.chan_config || cfg.chan_config > 7) {
1753 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1754 return AVERROR_INVALIDDATA;
1756 s->frames = mp3Frames[cfg.chan_config];
1757 s->coff = chan_offset[cfg.chan_config];
1758 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1759 avctx->channel_layout = chan_layout[cfg.chan_config];
1761 if (cfg.sample_rate < 16000)
1762 s->syncword = 0xffe00000;
1764 s->syncword = 0xfff00000;
1766 /* Init the first mp3 decoder in standard way, so that all tables get builded
1767 * We replace avctx->priv_data with the context of the first decoder so that
1768 * decode_init() does not have to be changed.
1769 * Other decoders will be initialized here copying data from the first context
1771 // Allocate zeroed memory for the first decoder context
1772 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1773 if (!s->mp3decctx[0])
1774 return AVERROR(ENOMEM);
1775 // Put decoder context in place to make init_decode() happy
1776 avctx->priv_data = s->mp3decctx[0];
1777 ret = decode_init(avctx);
1778 // Restore mp3on4 context pointer
1779 avctx->priv_data = s;
1782 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1784 /* Create a separate codec/context for each frame (first is already ok).
1785 * Each frame is 1 or 2 channels - up to 5 frames allowed
1787 for (i = 1; i < s->frames; i++) {
1788 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1789 if (!s->mp3decctx[i])
1790 return AVERROR(ENOMEM);
1791 s->mp3decctx[i]->adu_mode = 1;
1792 s->mp3decctx[i]->avctx = avctx;
1793 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1794 s->mp3decctx[i]->butterflies_float = s->mp3decctx[0]->butterflies_float;
1801 static void flush_mp3on4(AVCodecContext *avctx)
1804 MP3On4DecodeContext *s = avctx->priv_data;
1806 for (i = 0; i < s->frames; i++)
1807 mp_flush(s->mp3decctx[i]);
1811 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1812 int *got_frame_ptr, AVPacket *avpkt)
1814 AVFrame *frame = data;
1815 const uint8_t *buf = avpkt->data;
1816 int buf_size = avpkt->size;
1817 MP3On4DecodeContext *s = avctx->priv_data;
1818 MPADecodeContext *m;
1819 int fsize, len = buf_size, out_size = 0;
1821 OUT_INT **out_samples;
1825 /* get output buffer */
1826 frame->nb_samples = MPA_FRAME_SIZE;
1827 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1829 out_samples = (OUT_INT **)frame->extended_data;
1831 // Discard too short frames
1832 if (buf_size < HEADER_SIZE)
1833 return AVERROR_INVALIDDATA;
1835 avctx->bit_rate = 0;
1838 for (fr = 0; fr < s->frames; fr++) {
1839 fsize = AV_RB16(buf) >> 4;
1840 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1841 m = s->mp3decctx[fr];
1844 if (fsize < HEADER_SIZE) {
1845 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1846 return AVERROR_INVALIDDATA;
1848 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1850 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1852 av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n");
1853 return AVERROR_INVALIDDATA;
1856 if (ch + m->nb_channels > avctx->channels ||
1857 s->coff[fr] + m->nb_channels > avctx->channels) {
1858 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1860 return AVERROR_INVALIDDATA;
1862 ch += m->nb_channels;
1864 outptr[0] = out_samples[s->coff[fr]];
1865 if (m->nb_channels > 1)
1866 outptr[1] = out_samples[s->coff[fr] + 1];
1868 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) {
1869 av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch);
1870 memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1871 if (m->nb_channels > 1)
1872 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1873 ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT);
1880 avctx->bit_rate += m->bit_rate;
1882 if (ch != avctx->channels) {
1883 av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n");
1884 return AVERROR_INVALIDDATA;
1887 /* update codec info */
1888 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1890 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1895 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */