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[2][16];
103 static INTFLOAT is_table_lsf[2][2][16];
104 static INTFLOAT csa_table[8][4];
106 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
107 static int32_t scale_factor_mult[15][3];
108 /* mult table for layer 2 group quantization */
110 #define SCALE_GEN(v) \
111 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
113 static const int32_t scale_factor_mult2[3][3] = {
114 SCALE_GEN(4.0 / 3.0), /* 3 steps */
115 SCALE_GEN(4.0 / 5.0), /* 5 steps */
116 SCALE_GEN(4.0 / 9.0), /* 9 steps */
120 * Convert region offsets to region sizes and truncate
121 * size to big_values.
123 static void region_offset2size(GranuleDef *g)
126 g->region_size[2] = 576 / 2;
127 for (i = 0; i < 3; i++) {
128 k = FFMIN(g->region_size[i], g->big_values);
129 g->region_size[i] = k - j;
134 static void init_short_region(MPADecodeContext *s, GranuleDef *g)
136 if (g->block_type == 2) {
137 if (s->sample_rate_index != 8)
138 g->region_size[0] = (36 / 2);
140 g->region_size[0] = (72 / 2);
142 if (s->sample_rate_index <= 2)
143 g->region_size[0] = (36 / 2);
144 else if (s->sample_rate_index != 8)
145 g->region_size[0] = (54 / 2);
147 g->region_size[0] = (108 / 2);
149 g->region_size[1] = (576 / 2);
152 static void init_long_region(MPADecodeContext *s, GranuleDef *g,
156 g->region_size[0] = ff_band_index_long[s->sample_rate_index][ra1 + 1];
157 /* should not overflow */
158 l = FFMIN(ra1 + ra2 + 2, 22);
159 g->region_size[1] = ff_band_index_long[s->sample_rate_index][ l];
162 static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
164 if (g->block_type == 2) {
165 if (g->switch_point) {
166 if(s->sample_rate_index == 8)
167 avpriv_request_sample(s->avctx, "switch point in 8khz");
168 /* if switched mode, we handle the 36 first samples as
169 long blocks. For 8000Hz, we handle the 72 first
170 exponents as long blocks */
171 if (s->sample_rate_index <= 2)
187 /* layer 1 unscaling */
188 /* n = number of bits of the mantissa minus 1 */
189 static inline int l1_unscale(int n, int mant, int scale_factor)
194 shift = ff_scale_factor_modshift[scale_factor];
197 val = MUL64((int)(mant + (-1U << n) + 1), scale_factor_mult[n-1][mod]);
199 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
200 return (int)((val + (1LL << (shift - 1))) >> shift);
203 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
207 shift = ff_scale_factor_modshift[scale_factor];
211 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
212 /* NOTE: at this point, 0 <= shift <= 21 */
214 val = (val + (1 << (shift - 1))) >> shift;
218 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
219 static inline int l3_unscale(int value, int exponent)
224 e = table_4_3_exp [4 * value + (exponent & 3)];
225 m = table_4_3_value[4 * value + (exponent & 3)];
229 av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e);
233 m = (m + ((1U << e) >> 1)) >> e;
238 static av_cold void decode_init_static(void)
242 /* scale factor multiply for layer 1 */
243 for (i = 0; i < 15; i++) {
246 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
247 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
248 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
249 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
250 ff_dlog(NULL, "%d: norm=%x s=%"PRIx32" %"PRIx32" %"PRIx32"\n", i,
252 scale_factor_mult[i][0],
253 scale_factor_mult[i][1],
254 scale_factor_mult[i][2]);
257 /* compute n ^ (4/3) and store it in mantissa/exp format */
259 mpegaudio_tableinit();
261 for (i = 0; i < 7; i++) {
265 f = tan((double)i * M_PI / 12.0);
266 v = FIXR(f / (1.0 + f));
271 is_table[1][6 - i] = v;
274 for (i = 7; i < 16; i++)
275 is_table[0][i] = is_table[1][i] = 0.0;
277 for (i = 0; i < 16; i++) {
281 for (j = 0; j < 2; j++) {
282 e = -(j + 1) * ((i + 1) >> 1);
285 is_table_lsf[j][k ^ 1][i] = FIXR(f);
286 is_table_lsf[j][k ][i] = FIXR(1.0);
287 ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
288 i, j, (float) is_table_lsf[j][0][i],
289 (float) is_table_lsf[j][1][i]);
293 for (i = 0; i < 8; i++) {
296 cs = 1.0 / sqrt(1.0 + ci * ci);
299 csa_table[i][0] = FIXHR(cs/4);
300 csa_table[i][1] = FIXHR(ca/4);
301 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
302 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
304 csa_table[i][0] = cs;
305 csa_table[i][1] = ca;
306 csa_table[i][2] = ca + cs;
307 csa_table[i][3] = ca - cs;
310 RENAME(ff_mpa_synth_init)();
311 ff_mpegaudiodec_common_init_static();
314 static av_cold int decode_init(AVCodecContext * avctx)
316 static AVOnce init_static_once = AV_ONCE_INIT;
317 MPADecodeContext *s = avctx->priv_data;
323 AVFloatDSPContext *fdsp;
324 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
326 return AVERROR(ENOMEM);
327 s->butterflies_float = fdsp->butterflies_float;
332 ff_mpadsp_init(&s->mpadsp);
334 if (avctx->request_sample_fmt == OUT_FMT &&
335 avctx->codec_id != AV_CODEC_ID_MP3ON4)
336 avctx->sample_fmt = OUT_FMT;
338 avctx->sample_fmt = OUT_FMT_P;
339 s->err_recognition = avctx->err_recognition;
341 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
344 ff_thread_once(&init_static_once, decode_init_static);
349 #define C3 FIXHR(0.86602540378443864676/2)
350 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
351 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
352 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
354 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
356 static void imdct12(INTFLOAT *out, SUINTFLOAT *in)
358 SUINTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
361 in1 = in[1*3] + in[0*3];
362 in2 = in[2*3] + in[1*3];
363 in3 = in[3*3] + in[2*3];
364 in4 = in[4*3] + in[3*3];
365 in5 = in[5*3] + in[4*3];
369 in2 = MULH3(in2, C3, 2);
370 in3 = MULH3(in3, C3, 4);
373 t2 = MULH3(in1 - in5, C4, 2);
383 in1 = MULH3(in5 + in3, C5, 1);
390 in5 = MULH3(in5 - in3, C6, 2);
397 static int handle_crc(MPADecodeContext *s, int sec_len)
399 if (s->error_protection && (s->err_recognition & AV_EF_CRCCHECK)) {
400 const uint8_t *buf = s->gb.buffer - HEADER_SIZE;
401 int sec_byte_len = sec_len >> 3;
402 int sec_rem_bits = sec_len & 7;
403 const AVCRC *crc_tab = av_crc_get_table(AV_CRC_16_ANSI);
405 uint32_t crc_val = av_crc(crc_tab, UINT16_MAX, &buf[2], 2);
406 crc_val = av_crc(crc_tab, crc_val, &buf[6], sec_byte_len);
409 ((buf[6 + sec_byte_len] & (0xFF00 >> sec_rem_bits)) << 24) +
410 ((s->crc << 16) >> sec_rem_bits));
412 crc_val = av_crc(crc_tab, crc_val, tmp_buf, 3);
415 av_log(s->avctx, AV_LOG_ERROR, "CRC mismatch %X!\n", crc_val);
416 if (s->err_recognition & AV_EF_EXPLODE)
417 return AVERROR_INVALIDDATA;
423 /* return the number of decoded frames */
424 static int mp_decode_layer1(MPADecodeContext *s)
426 int bound, i, v, n, ch, j, mant;
427 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
428 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
431 ret = handle_crc(s, (s->nb_channels == 1) ? 8*16 : 8*32);
435 if (s->mode == MPA_JSTEREO)
436 bound = (s->mode_ext + 1) * 4;
440 /* allocation bits */
441 for (i = 0; i < bound; i++) {
442 for (ch = 0; ch < s->nb_channels; ch++) {
443 allocation[ch][i] = get_bits(&s->gb, 4);
446 for (i = bound; i < SBLIMIT; i++)
447 allocation[0][i] = get_bits(&s->gb, 4);
450 for (i = 0; i < bound; i++) {
451 for (ch = 0; ch < s->nb_channels; ch++) {
452 if (allocation[ch][i])
453 scale_factors[ch][i] = get_bits(&s->gb, 6);
456 for (i = bound; i < SBLIMIT; i++) {
457 if (allocation[0][i]) {
458 scale_factors[0][i] = get_bits(&s->gb, 6);
459 scale_factors[1][i] = get_bits(&s->gb, 6);
463 /* compute samples */
464 for (j = 0; j < 12; j++) {
465 for (i = 0; i < bound; i++) {
466 for (ch = 0; ch < s->nb_channels; ch++) {
467 n = allocation[ch][i];
469 mant = get_bits(&s->gb, n + 1);
470 v = l1_unscale(n, mant, scale_factors[ch][i]);
474 s->sb_samples[ch][j][i] = v;
477 for (i = bound; i < SBLIMIT; i++) {
478 n = allocation[0][i];
480 mant = get_bits(&s->gb, n + 1);
481 v = l1_unscale(n, mant, scale_factors[0][i]);
482 s->sb_samples[0][j][i] = v;
483 v = l1_unscale(n, mant, scale_factors[1][i]);
484 s->sb_samples[1][j][i] = v;
486 s->sb_samples[0][j][i] = 0;
487 s->sb_samples[1][j][i] = 0;
494 static int mp_decode_layer2(MPADecodeContext *s)
496 int sblimit; /* number of used subbands */
497 const unsigned char *alloc_table;
498 int table, bit_alloc_bits, i, j, ch, bound, v;
499 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
500 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
501 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
502 int scale, qindex, bits, steps, k, l, m, b;
505 /* select decoding table */
506 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
507 s->sample_rate, s->lsf);
508 sblimit = ff_mpa_sblimit_table[table];
509 alloc_table = ff_mpa_alloc_tables[table];
511 if (s->mode == MPA_JSTEREO)
512 bound = (s->mode_ext + 1) * 4;
516 ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
522 /* parse bit allocation */
524 for (i = 0; i < bound; i++) {
525 bit_alloc_bits = alloc_table[j];
526 for (ch = 0; ch < s->nb_channels; ch++)
527 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
528 j += 1 << bit_alloc_bits;
530 for (i = bound; i < sblimit; i++) {
531 bit_alloc_bits = alloc_table[j];
532 v = get_bits(&s->gb, bit_alloc_bits);
535 j += 1 << bit_alloc_bits;
539 for (i = 0; i < sblimit; i++) {
540 for (ch = 0; ch < s->nb_channels; ch++) {
541 if (bit_alloc[ch][i])
542 scale_code[ch][i] = get_bits(&s->gb, 2);
546 ret = handle_crc(s, get_bits_count(&s->gb) - 16);
551 for (i = 0; i < sblimit; i++) {
552 for (ch = 0; ch < s->nb_channels; ch++) {
553 if (bit_alloc[ch][i]) {
554 sf = scale_factors[ch][i];
555 switch (scale_code[ch][i]) {
558 sf[0] = get_bits(&s->gb, 6);
559 sf[1] = get_bits(&s->gb, 6);
560 sf[2] = get_bits(&s->gb, 6);
563 sf[0] = get_bits(&s->gb, 6);
568 sf[0] = get_bits(&s->gb, 6);
569 sf[2] = get_bits(&s->gb, 6);
573 sf[0] = get_bits(&s->gb, 6);
574 sf[2] = get_bits(&s->gb, 6);
583 for (k = 0; k < 3; k++) {
584 for (l = 0; l < 12; l += 3) {
586 for (i = 0; i < bound; i++) {
587 bit_alloc_bits = alloc_table[j];
588 for (ch = 0; ch < s->nb_channels; ch++) {
589 b = bit_alloc[ch][i];
591 scale = scale_factors[ch][i][k];
592 qindex = alloc_table[j+b];
593 bits = ff_mpa_quant_bits[qindex];
596 /* 3 values at the same time */
597 v = get_bits(&s->gb, -bits);
598 v2 = ff_division_tabs[qindex][v];
599 steps = ff_mpa_quant_steps[qindex];
601 s->sb_samples[ch][k * 12 + l + 0][i] =
602 l2_unscale_group(steps, v2 & 15, scale);
603 s->sb_samples[ch][k * 12 + l + 1][i] =
604 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
605 s->sb_samples[ch][k * 12 + l + 2][i] =
606 l2_unscale_group(steps, v2 >> 8 , scale);
608 for (m = 0; m < 3; m++) {
609 v = get_bits(&s->gb, bits);
610 v = l1_unscale(bits - 1, v, scale);
611 s->sb_samples[ch][k * 12 + l + m][i] = v;
615 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
616 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
617 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
620 /* next subband in alloc table */
621 j += 1 << bit_alloc_bits;
623 /* XXX: find a way to avoid this duplication of code */
624 for (i = bound; i < sblimit; i++) {
625 bit_alloc_bits = alloc_table[j];
628 int mant, scale0, scale1;
629 scale0 = scale_factors[0][i][k];
630 scale1 = scale_factors[1][i][k];
631 qindex = alloc_table[j + b];
632 bits = ff_mpa_quant_bits[qindex];
634 /* 3 values at the same time */
635 v = get_bits(&s->gb, -bits);
636 steps = ff_mpa_quant_steps[qindex];
639 s->sb_samples[0][k * 12 + l + 0][i] =
640 l2_unscale_group(steps, mant, scale0);
641 s->sb_samples[1][k * 12 + l + 0][i] =
642 l2_unscale_group(steps, mant, scale1);
645 s->sb_samples[0][k * 12 + l + 1][i] =
646 l2_unscale_group(steps, mant, scale0);
647 s->sb_samples[1][k * 12 + l + 1][i] =
648 l2_unscale_group(steps, mant, scale1);
649 s->sb_samples[0][k * 12 + l + 2][i] =
650 l2_unscale_group(steps, v, scale0);
651 s->sb_samples[1][k * 12 + l + 2][i] =
652 l2_unscale_group(steps, v, scale1);
654 for (m = 0; m < 3; m++) {
655 mant = get_bits(&s->gb, bits);
656 s->sb_samples[0][k * 12 + l + m][i] =
657 l1_unscale(bits - 1, mant, scale0);
658 s->sb_samples[1][k * 12 + l + m][i] =
659 l1_unscale(bits - 1, mant, scale1);
663 s->sb_samples[0][k * 12 + l + 0][i] = 0;
664 s->sb_samples[0][k * 12 + l + 1][i] = 0;
665 s->sb_samples[0][k * 12 + l + 2][i] = 0;
666 s->sb_samples[1][k * 12 + l + 0][i] = 0;
667 s->sb_samples[1][k * 12 + l + 1][i] = 0;
668 s->sb_samples[1][k * 12 + l + 2][i] = 0;
670 /* next subband in alloc table */
671 j += 1 << bit_alloc_bits;
673 /* fill remaining samples to zero */
674 for (i = sblimit; i < SBLIMIT; i++) {
675 for (ch = 0; ch < s->nb_channels; ch++) {
676 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
677 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
678 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
686 #define SPLIT(dst,sf,n) \
688 int m = (sf * 171) >> 9; \
691 } else if (n == 4) { \
694 } else if (n == 5) { \
695 int m = (sf * 205) >> 10; \
698 } else if (n == 6) { \
699 int m = (sf * 171) >> 10; \
706 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
709 SPLIT(slen[3], sf, n3)
710 SPLIT(slen[2], sf, n2)
711 SPLIT(slen[1], sf, n1)
715 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
718 const uint8_t *bstab, *pretab;
719 int len, i, j, k, l, v0, shift, gain, gains[3];
723 gain = g->global_gain - 210;
724 shift = g->scalefac_scale + 1;
726 bstab = ff_band_size_long[s->sample_rate_index];
727 pretab = ff_mpa_pretab[g->preflag];
728 for (i = 0; i < g->long_end; i++) {
729 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
731 for (j = len; j > 0; j--)
735 if (g->short_start < 13) {
736 bstab = ff_band_size_short[s->sample_rate_index];
737 gains[0] = gain - (g->subblock_gain[0] << 3);
738 gains[1] = gain - (g->subblock_gain[1] << 3);
739 gains[2] = gain - (g->subblock_gain[2] << 3);
741 for (i = g->short_start; i < 13; i++) {
743 for (l = 0; l < 3; l++) {
744 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
745 for (j = len; j > 0; j--)
752 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
755 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits - s->extrasize * 8) {
757 s->in_gb.buffer = NULL;
759 av_assert2((get_bits_count(&s->gb) & 7) == 0);
760 skip_bits_long(&s->gb, *pos - *end_pos);
762 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
763 *pos = get_bits_count(&s->gb);
767 /* Following is an optimized code for
769 if(get_bits1(&s->gb))
774 #define READ_FLIP_SIGN(dst,src) \
775 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
778 #define READ_FLIP_SIGN(dst,src) \
779 v = -get_bits1(&s->gb); \
780 *(dst) = (*(src) ^ v) - v;
783 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
784 int16_t *exponents, int end_pos2)
788 int last_pos, bits_left;
790 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits - s->extrasize * 8);
792 /* low frequencies (called big values) */
794 for (i = 0; i < 3; i++) {
795 int j, k, l, linbits;
796 j = g->region_size[i];
799 /* select vlc table */
800 k = g->table_select[i];
801 l = ff_mpa_huff_data[k][0];
802 linbits = ff_mpa_huff_data[k][1];
803 vlc = &ff_huff_vlc[l];
806 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
811 /* read huffcode and compute each couple */
815 int pos = get_bits_count(&s->gb);
818 switch_buffer(s, &pos, &end_pos, &end_pos2);
822 y = get_vlc2(&s->gb, vlc->table, 7, 3);
825 g->sb_hybrid[s_index ] =
826 g->sb_hybrid[s_index + 1] = 0;
831 exponent= exponents[s_index];
833 ff_dlog(s->avctx, "region=%d n=%d y=%d exp=%d\n",
834 i, g->region_size[i] - j, y, exponent);
839 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
841 x += get_bitsz(&s->gb, linbits);
842 v = l3_unscale(x, exponent);
843 if (get_bits1(&s->gb))
845 g->sb_hybrid[s_index] = v;
848 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
850 y += get_bitsz(&s->gb, linbits);
851 v = l3_unscale(y, exponent);
852 if (get_bits1(&s->gb))
854 g->sb_hybrid[s_index + 1] = v;
861 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
863 x += get_bitsz(&s->gb, linbits);
864 v = l3_unscale(x, exponent);
865 if (get_bits1(&s->gb))
867 g->sb_hybrid[s_index+!!y] = v;
869 g->sb_hybrid[s_index + !y] = 0;
875 /* high frequencies */
876 vlc = &ff_huff_quad_vlc[g->count1table_select];
878 while (s_index <= 572) {
880 pos = get_bits_count(&s->gb);
881 if (pos >= end_pos) {
882 if (pos > end_pos2 && last_pos) {
883 /* some encoders generate an incorrect size for this
884 part. We must go back into the data */
886 skip_bits_long(&s->gb, last_pos - pos);
887 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
888 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
892 switch_buffer(s, &pos, &end_pos, &end_pos2);
898 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
899 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
900 g->sb_hybrid[s_index + 0] =
901 g->sb_hybrid[s_index + 1] =
902 g->sb_hybrid[s_index + 2] =
903 g->sb_hybrid[s_index + 3] = 0;
905 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
907 int pos = s_index + idxtab[code];
908 code ^= 8 >> idxtab[code];
909 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
913 /* skip extension bits */
914 bits_left = end_pos2 - get_bits_count(&s->gb);
915 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
916 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
918 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
919 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
922 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
923 skip_bits_long(&s->gb, bits_left);
925 i = get_bits_count(&s->gb);
926 switch_buffer(s, &i, &end_pos, &end_pos2);
931 /* Reorder short blocks from bitstream order to interleaved order. It
932 would be faster to do it in parsing, but the code would be far more
934 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
937 INTFLOAT *ptr, *dst, *ptr1;
940 if (g->block_type != 2)
943 if (g->switch_point) {
944 if (s->sample_rate_index != 8)
945 ptr = g->sb_hybrid + 36;
947 ptr = g->sb_hybrid + 72;
952 for (i = g->short_start; i < 13; i++) {
953 len = ff_band_size_short[s->sample_rate_index][i];
956 for (j = len; j > 0; j--) {
963 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
967 #define ISQRT2 FIXR(0.70710678118654752440)
969 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
972 int sf_max, sf, len, non_zero_found;
973 INTFLOAT (*is_tab)[16], *tab0, *tab1, v1, v2;
974 SUINTFLOAT tmp0, tmp1;
975 int non_zero_found_short[3];
977 /* intensity stereo */
978 if (s->mode_ext & MODE_EXT_I_STEREO) {
983 is_tab = is_table_lsf[g1->scalefac_compress & 1];
987 tab0 = g0->sb_hybrid + 576;
988 tab1 = g1->sb_hybrid + 576;
990 non_zero_found_short[0] = 0;
991 non_zero_found_short[1] = 0;
992 non_zero_found_short[2] = 0;
993 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
994 for (i = 12; i >= g1->short_start; i--) {
995 /* for last band, use previous scale factor */
998 len = ff_band_size_short[s->sample_rate_index][i];
999 for (l = 2; l >= 0; l--) {
1002 if (!non_zero_found_short[l]) {
1003 /* test if non zero band. if so, stop doing i-stereo */
1004 for (j = 0; j < len; j++) {
1006 non_zero_found_short[l] = 1;
1010 sf = g1->scale_factors[k + l];
1016 for (j = 0; j < len; j++) {
1018 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1019 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1023 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1024 /* lower part of the spectrum : do ms stereo
1026 for (j = 0; j < len; j++) {
1029 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1030 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1037 non_zero_found = non_zero_found_short[0] |
1038 non_zero_found_short[1] |
1039 non_zero_found_short[2];
1041 for (i = g1->long_end - 1;i >= 0;i--) {
1042 len = ff_band_size_long[s->sample_rate_index][i];
1045 /* test if non zero band. if so, stop doing i-stereo */
1046 if (!non_zero_found) {
1047 for (j = 0; j < len; j++) {
1053 /* for last band, use previous scale factor */
1054 k = (i == 21) ? 20 : i;
1055 sf = g1->scale_factors[k];
1060 for (j = 0; j < len; j++) {
1062 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1063 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1067 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1068 /* lower part of the spectrum : do ms stereo
1070 for (j = 0; j < len; j++) {
1073 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1074 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1079 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1080 /* ms stereo ONLY */
1081 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1084 s->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1086 tab0 = g0->sb_hybrid;
1087 tab1 = g1->sb_hybrid;
1088 for (i = 0; i < 576; i++) {
1091 tab0[i] = tmp0 + tmp1;
1092 tab1[i] = tmp0 - tmp1;
1100 # include "mips/compute_antialias_float.h"
1101 #endif /* HAVE_MIPSFPU */
1104 # include "mips/compute_antialias_fixed.h"
1105 #endif /* HAVE_MIPSDSP */
1106 #endif /* USE_FLOATS */
1108 #ifndef compute_antialias
1110 #define AA(j) do { \
1111 float tmp0 = ptr[-1-j]; \
1112 float tmp1 = ptr[ j]; \
1113 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1114 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1117 #define AA(j) do { \
1118 SUINT tmp0 = ptr[-1-j]; \
1119 SUINT tmp1 = ptr[ j]; \
1120 SUINT tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1121 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1122 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1126 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1131 /* we antialias only "long" bands */
1132 if (g->block_type == 2) {
1133 if (!g->switch_point)
1135 /* XXX: check this for 8000Hz case */
1141 ptr = g->sb_hybrid + 18;
1142 for (i = n; i > 0; i--) {
1155 #endif /* compute_antialias */
1157 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1158 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1160 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1162 int i, j, mdct_long_end, sblimit;
1164 /* find last non zero block */
1165 ptr = g->sb_hybrid + 576;
1166 ptr1 = g->sb_hybrid + 2 * 18;
1167 while (ptr >= ptr1) {
1171 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1174 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1176 if (g->block_type == 2) {
1177 /* XXX: check for 8000 Hz */
1178 if (g->switch_point)
1183 mdct_long_end = sblimit;
1186 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1187 mdct_long_end, g->switch_point,
1190 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1191 ptr = g->sb_hybrid + 18 * mdct_long_end;
1193 for (j = mdct_long_end; j < sblimit; j++) {
1194 /* select frequency inversion */
1195 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1196 out_ptr = sb_samples + j;
1198 for (i = 0; i < 6; i++) {
1199 *out_ptr = buf[4*i];
1202 imdct12(out2, ptr + 0);
1203 for (i = 0; i < 6; i++) {
1204 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1205 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1208 imdct12(out2, ptr + 1);
1209 for (i = 0; i < 6; i++) {
1210 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1211 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1214 imdct12(out2, ptr + 2);
1215 for (i = 0; i < 6; i++) {
1216 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1217 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1218 buf[4*(i + 6*2)] = 0;
1221 buf += (j&3) != 3 ? 1 : (4*18-3);
1224 for (j = sblimit; j < SBLIMIT; j++) {
1226 out_ptr = sb_samples + j;
1227 for (i = 0; i < 18; i++) {
1228 *out_ptr = buf[4*i];
1232 buf += (j&3) != 3 ? 1 : (4*18-3);
1236 /* main layer3 decoding function */
1237 static int mp_decode_layer3(MPADecodeContext *s)
1239 int nb_granules, main_data_begin;
1240 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1242 int16_t exponents[576]; //FIXME try INTFLOAT
1245 /* read side info */
1247 ret = handle_crc(s, ((s->nb_channels == 1) ? 8*9 : 8*17));
1248 main_data_begin = get_bits(&s->gb, 8);
1249 skip_bits(&s->gb, s->nb_channels);
1252 ret = handle_crc(s, ((s->nb_channels == 1) ? 8*17 : 8*32));
1253 main_data_begin = get_bits(&s->gb, 9);
1254 if (s->nb_channels == 2)
1255 skip_bits(&s->gb, 3);
1257 skip_bits(&s->gb, 5);
1259 for (ch = 0; ch < s->nb_channels; ch++) {
1260 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1261 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1267 for (gr = 0; gr < nb_granules; gr++) {
1268 for (ch = 0; ch < s->nb_channels; ch++) {
1269 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1270 g = &s->granules[ch][gr];
1271 g->part2_3_length = get_bits(&s->gb, 12);
1272 g->big_values = get_bits(&s->gb, 9);
1273 if (g->big_values > 288) {
1274 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1275 return AVERROR_INVALIDDATA;
1278 g->global_gain = get_bits(&s->gb, 8);
1279 /* if MS stereo only is selected, we precompute the
1280 1/sqrt(2) renormalization factor */
1281 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1283 g->global_gain -= 2;
1285 g->scalefac_compress = get_bits(&s->gb, 9);
1287 g->scalefac_compress = get_bits(&s->gb, 4);
1288 blocksplit_flag = get_bits1(&s->gb);
1289 if (blocksplit_flag) {
1290 g->block_type = get_bits(&s->gb, 2);
1291 if (g->block_type == 0) {
1292 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1293 return AVERROR_INVALIDDATA;
1295 g->switch_point = get_bits1(&s->gb);
1296 for (i = 0; i < 2; i++)
1297 g->table_select[i] = get_bits(&s->gb, 5);
1298 for (i = 0; i < 3; i++)
1299 g->subblock_gain[i] = get_bits(&s->gb, 3);
1300 init_short_region(s, g);
1302 int region_address1, region_address2;
1304 g->switch_point = 0;
1305 for (i = 0; i < 3; i++)
1306 g->table_select[i] = get_bits(&s->gb, 5);
1307 /* compute huffman coded region sizes */
1308 region_address1 = get_bits(&s->gb, 4);
1309 region_address2 = get_bits(&s->gb, 3);
1310 ff_dlog(s->avctx, "region1=%d region2=%d\n",
1311 region_address1, region_address2);
1312 init_long_region(s, g, region_address1, region_address2);
1314 region_offset2size(g);
1315 compute_band_indexes(s, g);
1319 g->preflag = get_bits1(&s->gb);
1320 g->scalefac_scale = get_bits1(&s->gb);
1321 g->count1table_select = get_bits1(&s->gb);
1322 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1323 g->block_type, g->switch_point);
1329 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb) >> 3);
1330 s->extrasize = av_clip((get_bits_left(&s->gb) >> 3) - s->extrasize, 0,
1331 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1332 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1333 /* now we get bits from the main_data_begin offset */
1334 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1335 main_data_begin, s->last_buf_size);
1337 memcpy(s->last_buf + s->last_buf_size, ptr, s->extrasize);
1339 init_get_bits(&s->gb, s->last_buf, (s->last_buf_size + s->extrasize) * 8);
1340 s->last_buf_size <<= 3;
1341 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1342 for (ch = 0; ch < s->nb_channels; ch++) {
1343 g = &s->granules[ch][gr];
1344 s->last_buf_size += g->part2_3_length;
1345 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1346 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1349 skip = s->last_buf_size - 8 * main_data_begin;
1350 if (skip >= s->gb.size_in_bits - s->extrasize * 8 && s->in_gb.buffer) {
1351 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits + s->extrasize * 8);
1353 s->in_gb.buffer = NULL;
1356 skip_bits_long(&s->gb, skip);
1363 for (; gr < nb_granules; gr++) {
1364 for (ch = 0; ch < s->nb_channels; ch++) {
1365 g = &s->granules[ch][gr];
1366 bits_pos = get_bits_count(&s->gb);
1370 int slen, slen1, slen2;
1372 /* MPEG-1 scale factors */
1373 slen1 = ff_slen_table[0][g->scalefac_compress];
1374 slen2 = ff_slen_table[1][g->scalefac_compress];
1375 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1376 if (g->block_type == 2) {
1377 n = g->switch_point ? 17 : 18;
1380 for (i = 0; i < n; i++)
1381 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1383 for (i = 0; i < n; i++)
1384 g->scale_factors[j++] = 0;
1387 for (i = 0; i < 18; i++)
1388 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1389 for (i = 0; i < 3; i++)
1390 g->scale_factors[j++] = 0;
1392 for (i = 0; i < 21; i++)
1393 g->scale_factors[j++] = 0;
1396 sc = s->granules[ch][0].scale_factors;
1398 for (k = 0; k < 4; k++) {
1400 if ((g->scfsi & (0x8 >> k)) == 0) {
1401 slen = (k < 2) ? slen1 : slen2;
1403 for (i = 0; i < n; i++)
1404 g->scale_factors[j++] = get_bits(&s->gb, slen);
1406 for (i = 0; i < n; i++)
1407 g->scale_factors[j++] = 0;
1410 /* simply copy from last granule */
1411 for (i = 0; i < n; i++) {
1412 g->scale_factors[j] = sc[j];
1417 g->scale_factors[j++] = 0;
1420 int tindex, tindex2, slen[4], sl, sf;
1422 /* LSF scale factors */
1423 if (g->block_type == 2)
1424 tindex = g->switch_point ? 2 : 1;
1428 sf = g->scalefac_compress;
1429 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1430 /* intensity stereo case */
1433 lsf_sf_expand(slen, sf, 6, 6, 0);
1435 } else if (sf < 244) {
1436 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1439 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1445 lsf_sf_expand(slen, sf, 5, 4, 4);
1447 } else if (sf < 500) {
1448 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1451 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1458 for (k = 0; k < 4; k++) {
1459 n = ff_lsf_nsf_table[tindex2][tindex][k];
1462 for (i = 0; i < n; i++)
1463 g->scale_factors[j++] = get_bits(&s->gb, sl);
1465 for (i = 0; i < n; i++)
1466 g->scale_factors[j++] = 0;
1469 /* XXX: should compute exact size */
1471 g->scale_factors[j] = 0;
1474 exponents_from_scale_factors(s, g, exponents);
1476 /* read Huffman coded residue */
1477 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1480 if (s->mode == MPA_JSTEREO)
1481 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1483 for (ch = 0; ch < s->nb_channels; ch++) {
1484 g = &s->granules[ch][gr];
1486 reorder_block(s, g);
1487 compute_antialias(s, g);
1488 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1491 if (get_bits_count(&s->gb) < 0)
1492 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1493 return nb_granules * 18;
1496 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1497 const uint8_t *buf, int buf_size)
1499 int i, nb_frames, ch, ret;
1500 OUT_INT *samples_ptr;
1502 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1503 if (s->error_protection)
1504 s->crc = get_bits(&s->gb, 16);
1508 s->avctx->frame_size = 384;
1509 nb_frames = mp_decode_layer1(s);
1512 s->avctx->frame_size = 1152;
1513 nb_frames = mp_decode_layer2(s);
1516 s->avctx->frame_size = s->lsf ? 576 : 1152;
1518 nb_frames = mp_decode_layer3(s);
1521 if (s->in_gb.buffer) {
1522 align_get_bits(&s->gb);
1523 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1524 if (i >= 0 && i <= BACKSTEP_SIZE) {
1525 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb) >> 3), i);
1528 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1530 s->in_gb.buffer = NULL;
1534 align_get_bits(&s->gb);
1535 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1536 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1537 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1539 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1540 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1542 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1543 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1544 s->last_buf_size += i;
1550 /* get output buffer */
1552 av_assert0(s->frame);
1553 s->frame->nb_samples = s->avctx->frame_size;
1554 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)
1556 samples = (OUT_INT **)s->frame->extended_data;
1559 /* apply the synthesis filter */
1560 for (ch = 0; ch < s->nb_channels; ch++) {
1562 if (s->avctx->sample_fmt == OUT_FMT_P) {
1563 samples_ptr = samples[ch];
1566 samples_ptr = samples[0] + ch;
1567 sample_stride = s->nb_channels;
1569 for (i = 0; i < nb_frames; i++) {
1570 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1571 &(s->synth_buf_offset[ch]),
1572 RENAME(ff_mpa_synth_window),
1573 &s->dither_state, samples_ptr,
1574 sample_stride, s->sb_samples[ch][i]);
1575 samples_ptr += 32 * sample_stride;
1579 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1582 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1585 const uint8_t *buf = avpkt->data;
1586 int buf_size = avpkt->size;
1587 MPADecodeContext *s = avctx->priv_data;
1592 while(buf_size && !*buf){
1598 if (buf_size < HEADER_SIZE)
1599 return AVERROR_INVALIDDATA;
1601 header = AV_RB32(buf);
1602 if (header >> 8 == AV_RB32("TAG") >> 8) {
1603 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1604 return buf_size + skipped;
1606 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1608 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1609 return AVERROR_INVALIDDATA;
1610 } else if (ret == 1) {
1611 /* free format: prepare to compute frame size */
1613 return AVERROR_INVALIDDATA;
1615 /* update codec info */
1616 avctx->channels = s->nb_channels;
1617 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1618 if (!avctx->bit_rate)
1619 avctx->bit_rate = s->bit_rate;
1621 if (s->frame_size <= 0) {
1622 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1623 return AVERROR_INVALIDDATA;
1624 } else if (s->frame_size < buf_size) {
1625 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1626 buf_size= s->frame_size;
1631 ret = mp_decode_frame(s, NULL, buf, buf_size);
1633 s->frame->nb_samples = avctx->frame_size;
1635 avctx->sample_rate = s->sample_rate;
1636 //FIXME maybe move the other codec info stuff from above here too
1638 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1639 /* Only return an error if the bad frame makes up the whole packet or
1640 * the error is related to buffer management.
1641 * If there is more data in the packet, just consume the bad frame
1642 * instead of returning an error, which would discard the whole
1645 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1649 return buf_size + skipped;
1652 static void mp_flush(MPADecodeContext *ctx)
1654 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1655 memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));
1656 ctx->last_buf_size = 0;
1657 ctx->dither_state = 0;
1660 static void flush(AVCodecContext *avctx)
1662 mp_flush(avctx->priv_data);
1665 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1666 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1667 int *got_frame_ptr, AVPacket *avpkt)
1669 const uint8_t *buf = avpkt->data;
1670 int buf_size = avpkt->size;
1671 MPADecodeContext *s = avctx->priv_data;
1674 int av_unused out_size;
1678 // Discard too short frames
1679 if (buf_size < HEADER_SIZE) {
1680 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1681 return AVERROR_INVALIDDATA;
1685 if (len > MPA_MAX_CODED_FRAME_SIZE)
1686 len = MPA_MAX_CODED_FRAME_SIZE;
1688 // Get header and restore sync word
1689 header = AV_RB32(buf) | 0xffe00000;
1691 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1693 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1696 /* update codec info */
1697 avctx->sample_rate = s->sample_rate;
1698 avctx->channels = s->nb_channels;
1699 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1700 if (!avctx->bit_rate)
1701 avctx->bit_rate = s->bit_rate;
1703 s->frame_size = len;
1707 ret = mp_decode_frame(s, NULL, buf, buf_size);
1709 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1717 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1719 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1722 * Context for MP3On4 decoder
1724 typedef struct MP3On4DecodeContext {
1725 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1726 int syncword; ///< syncword patch
1727 const uint8_t *coff; ///< channel offsets in output buffer
1728 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1729 } MP3On4DecodeContext;
1731 #include "mpeg4audio.h"
1733 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1735 /* number of mp3 decoder instances */
1736 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1738 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1739 static const uint8_t chan_offset[8][5] = {
1744 { 2, 0, 3 }, // C FLR BS
1745 { 2, 0, 3 }, // C FLR BLRS
1746 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1747 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1750 /* mp3on4 channel layouts */
1751 static const int16_t chan_layout[8] = {
1754 AV_CH_LAYOUT_STEREO,
1755 AV_CH_LAYOUT_SURROUND,
1756 AV_CH_LAYOUT_4POINT0,
1757 AV_CH_LAYOUT_5POINT0,
1758 AV_CH_LAYOUT_5POINT1,
1759 AV_CH_LAYOUT_7POINT1
1762 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1764 MP3On4DecodeContext *s = avctx->priv_data;
1767 for (i = 0; i < s->frames; i++)
1768 av_freep(&s->mp3decctx[i]);
1774 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1776 MP3On4DecodeContext *s = avctx->priv_data;
1777 MPEG4AudioConfig cfg;
1780 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1781 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1782 return AVERROR_INVALIDDATA;
1785 avpriv_mpeg4audio_get_config2(&cfg, avctx->extradata,
1786 avctx->extradata_size, 1, avctx);
1787 if (!cfg.chan_config || cfg.chan_config > 7) {
1788 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1789 return AVERROR_INVALIDDATA;
1791 s->frames = mp3Frames[cfg.chan_config];
1792 s->coff = chan_offset[cfg.chan_config];
1793 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1794 avctx->channel_layout = chan_layout[cfg.chan_config];
1796 if (cfg.sample_rate < 16000)
1797 s->syncword = 0xffe00000;
1799 s->syncword = 0xfff00000;
1801 /* Init the first mp3 decoder in standard way, so that all tables get builded
1802 * We replace avctx->priv_data with the context of the first decoder so that
1803 * decode_init() does not have to be changed.
1804 * Other decoders will be initialized here copying data from the first context
1806 // Allocate zeroed memory for the first decoder context
1807 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1808 if (!s->mp3decctx[0])
1809 return AVERROR(ENOMEM);
1810 // Put decoder context in place to make init_decode() happy
1811 avctx->priv_data = s->mp3decctx[0];
1812 ret = decode_init(avctx);
1813 // Restore mp3on4 context pointer
1814 avctx->priv_data = s;
1817 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1819 /* Create a separate codec/context for each frame (first is already ok).
1820 * Each frame is 1 or 2 channels - up to 5 frames allowed
1822 for (i = 1; i < s->frames; i++) {
1823 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1824 if (!s->mp3decctx[i])
1825 return AVERROR(ENOMEM);
1826 s->mp3decctx[i]->adu_mode = 1;
1827 s->mp3decctx[i]->avctx = avctx;
1828 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1829 s->mp3decctx[i]->butterflies_float = s->mp3decctx[0]->butterflies_float;
1836 static void flush_mp3on4(AVCodecContext *avctx)
1839 MP3On4DecodeContext *s = avctx->priv_data;
1841 for (i = 0; i < s->frames; i++)
1842 mp_flush(s->mp3decctx[i]);
1846 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1847 int *got_frame_ptr, AVPacket *avpkt)
1849 AVFrame *frame = data;
1850 const uint8_t *buf = avpkt->data;
1851 int buf_size = avpkt->size;
1852 MP3On4DecodeContext *s = avctx->priv_data;
1853 MPADecodeContext *m;
1854 int fsize, len = buf_size, out_size = 0;
1856 OUT_INT **out_samples;
1860 /* get output buffer */
1861 frame->nb_samples = MPA_FRAME_SIZE;
1862 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1864 out_samples = (OUT_INT **)frame->extended_data;
1866 // Discard too short frames
1867 if (buf_size < HEADER_SIZE)
1868 return AVERROR_INVALIDDATA;
1870 avctx->bit_rate = 0;
1873 for (fr = 0; fr < s->frames; fr++) {
1874 fsize = AV_RB16(buf) >> 4;
1875 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1876 m = s->mp3decctx[fr];
1879 if (fsize < HEADER_SIZE) {
1880 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1881 return AVERROR_INVALIDDATA;
1883 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1885 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1887 av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n");
1888 return AVERROR_INVALIDDATA;
1891 if (ch + m->nb_channels > avctx->channels ||
1892 s->coff[fr] + m->nb_channels > avctx->channels) {
1893 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1895 return AVERROR_INVALIDDATA;
1897 ch += m->nb_channels;
1899 outptr[0] = out_samples[s->coff[fr]];
1900 if (m->nb_channels > 1)
1901 outptr[1] = out_samples[s->coff[fr] + 1];
1903 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) {
1904 av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch);
1905 memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1906 if (m->nb_channels > 1)
1907 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1908 ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT);
1915 avctx->bit_rate += m->bit_rate;
1917 if (ch != avctx->channels) {
1918 av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n");
1919 return AVERROR_INVALIDDATA;
1922 /* update codec info */
1923 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1925 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1930 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */