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/avassert.h"
28 #include "libavutil/channel_layout.h"
29 #include "libavutil/libm.h"
33 #include "mpegaudiodsp.h"
38 * - test lsf / mpeg25 extensively.
41 #include "mpegaudio.h"
42 #include "mpegaudiodecheader.h"
44 #define BACKSTEP_SIZE 512
46 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES
48 /* layer 3 "granule" */
49 typedef struct GranuleDef {
54 int scalefac_compress;
59 uint8_t scalefac_scale;
60 uint8_t count1table_select;
61 int region_size[3]; /* number of huffman codes in each region */
63 int short_start, long_end; /* long/short band indexes */
64 uint8_t scale_factors[40];
65 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
68 typedef struct MPADecodeContext {
70 uint8_t last_buf[LAST_BUF_SIZE];
72 /* next header (used in free format parsing) */
73 uint32_t free_format_next_header;
76 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
77 int synth_buf_offset[MPA_MAX_CHANNELS];
78 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
79 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
80 GranuleDef granules[2][2]; /* Used in Layer 3 */
81 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
84 AVCodecContext* avctx;
91 # define SHR(a,b) ((a)*(1.0f/(1<<(b))))
92 # define FIXR_OLD(a) ((int)((a) * FRAC_ONE + 0.5))
93 # define FIXR(x) ((float)(x))
94 # define FIXHR(x) ((float)(x))
95 # define MULH3(x, y, s) ((s)*(y)*(x))
96 # define MULLx(x, y, s) ((y)*(x))
97 # define RENAME(a) a ## _float
98 # define OUT_FMT AV_SAMPLE_FMT_FLT
99 # define OUT_FMT_P AV_SAMPLE_FMT_FLTP
101 # define SHR(a,b) ((a)>>(b))
102 /* WARNING: only correct for positive numbers */
103 # define FIXR_OLD(a) ((int)((a) * FRAC_ONE + 0.5))
104 # define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
105 # define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
106 # define MULH3(x, y, s) MULH((s)*(x), y)
107 # define MULLx(x, y, s) MULL(x,y,s)
108 # define RENAME(a) a ## _fixed
109 # define OUT_FMT AV_SAMPLE_FMT_S16
110 # define OUT_FMT_P AV_SAMPLE_FMT_S16P
115 #define HEADER_SIZE 4
117 #include "mpegaudiodata.h"
118 #include "mpegaudiodectab.h"
120 /* vlc structure for decoding layer 3 huffman tables */
121 static VLC huff_vlc[16];
122 static VLC_TYPE huff_vlc_tables[
123 0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +
124 142 + 204 + 190 + 170 + 542 + 460 + 662 + 414
126 static const int huff_vlc_tables_sizes[16] = {
127 0, 128, 128, 128, 130, 128, 154, 166,
128 142, 204, 190, 170, 542, 460, 662, 414
130 static VLC huff_quad_vlc[2];
131 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
132 static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };
133 /* computed from band_size_long */
134 static uint16_t band_index_long[9][23];
135 #include "mpegaudio_tablegen.h"
136 /* intensity stereo coef table */
137 static INTFLOAT is_table[2][16];
138 static INTFLOAT is_table_lsf[2][2][16];
139 static INTFLOAT csa_table[8][4];
141 static int16_t division_tab3[1<<6 ];
142 static int16_t division_tab5[1<<8 ];
143 static int16_t division_tab9[1<<11];
145 static int16_t * const division_tabs[4] = {
146 division_tab3, division_tab5, NULL, division_tab9
149 /* lower 2 bits: modulo 3, higher bits: shift */
150 static uint16_t scale_factor_modshift[64];
151 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
152 static int32_t scale_factor_mult[15][3];
153 /* mult table for layer 2 group quantization */
155 #define SCALE_GEN(v) \
156 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
158 static const int32_t scale_factor_mult2[3][3] = {
159 SCALE_GEN(4.0 / 3.0), /* 3 steps */
160 SCALE_GEN(4.0 / 5.0), /* 5 steps */
161 SCALE_GEN(4.0 / 9.0), /* 9 steps */
165 * Convert region offsets to region sizes and truncate
166 * size to big_values.
168 static void ff_region_offset2size(GranuleDef *g)
171 g->region_size[2] = 576 / 2;
172 for (i = 0; i < 3; i++) {
173 k = FFMIN(g->region_size[i], g->big_values);
174 g->region_size[i] = k - j;
179 static void ff_init_short_region(MPADecodeContext *s, GranuleDef *g)
181 if (g->block_type == 2) {
182 if (s->sample_rate_index != 8)
183 g->region_size[0] = (36 / 2);
185 g->region_size[0] = (72 / 2);
187 if (s->sample_rate_index <= 2)
188 g->region_size[0] = (36 / 2);
189 else if (s->sample_rate_index != 8)
190 g->region_size[0] = (54 / 2);
192 g->region_size[0] = (108 / 2);
194 g->region_size[1] = (576 / 2);
197 static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2)
200 g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
201 /* should not overflow */
202 l = FFMIN(ra1 + ra2 + 2, 22);
203 g->region_size[1] = band_index_long[s->sample_rate_index][ l] >> 1;
206 static void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
208 if (g->block_type == 2) {
209 if (g->switch_point) {
210 if(s->sample_rate_index == 8)
211 av_log_ask_for_sample(s->avctx, "switch point in 8khz\n");
212 /* if switched mode, we handle the 36 first samples as
213 long blocks. For 8000Hz, we handle the 72 first
214 exponents as long blocks */
215 if (s->sample_rate_index <= 2)
231 /* layer 1 unscaling */
232 /* n = number of bits of the mantissa minus 1 */
233 static inline int l1_unscale(int n, int mant, int scale_factor)
238 shift = scale_factor_modshift[scale_factor];
241 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
243 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
244 return (int)((val + (1LL << (shift - 1))) >> shift);
247 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
251 shift = scale_factor_modshift[scale_factor];
255 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
256 /* NOTE: at this point, 0 <= shift <= 21 */
258 val = (val + (1 << (shift - 1))) >> shift;
262 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
263 static inline int l3_unscale(int value, int exponent)
268 e = table_4_3_exp [4 * value + (exponent & 3)];
269 m = table_4_3_value[4 * value + (exponent & 3)];
273 av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e);
277 m = (m + (1 << (e - 1))) >> e;
282 static av_cold void decode_init_static(void)
287 /* scale factors table for layer 1/2 */
288 for (i = 0; i < 64; i++) {
290 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
293 scale_factor_modshift[i] = mod | (shift << 2);
296 /* scale factor multiply for layer 1 */
297 for (i = 0; i < 15; i++) {
300 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
301 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
302 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
303 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
304 av_dlog(NULL, "%d: norm=%x s=%x %x %x\n", i, norm,
305 scale_factor_mult[i][0],
306 scale_factor_mult[i][1],
307 scale_factor_mult[i][2]);
310 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
312 /* huffman decode tables */
314 for (i = 1; i < 16; i++) {
315 const HuffTable *h = &mpa_huff_tables[i];
317 uint8_t tmp_bits [512] = { 0 };
318 uint16_t tmp_codes[512] = { 0 };
323 for (x = 0; x < xsize; x++) {
324 for (y = 0; y < xsize; y++) {
325 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
326 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
331 huff_vlc[i].table = huff_vlc_tables+offset;
332 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
333 init_vlc(&huff_vlc[i], 7, 512,
334 tmp_bits, 1, 1, tmp_codes, 2, 2,
335 INIT_VLC_USE_NEW_STATIC);
336 offset += huff_vlc_tables_sizes[i];
338 av_assert0(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
341 for (i = 0; i < 2; i++) {
342 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
343 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
344 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
345 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
346 INIT_VLC_USE_NEW_STATIC);
347 offset += huff_quad_vlc_tables_sizes[i];
349 av_assert0(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
351 for (i = 0; i < 9; i++) {
353 for (j = 0; j < 22; j++) {
354 band_index_long[i][j] = k;
355 k += band_size_long[i][j];
357 band_index_long[i][22] = k;
360 /* compute n ^ (4/3) and store it in mantissa/exp format */
362 mpegaudio_tableinit();
364 for (i = 0; i < 4; i++) {
365 if (ff_mpa_quant_bits[i] < 0) {
366 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
367 int val1, val2, val3, steps;
369 steps = ff_mpa_quant_steps[i];
374 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
380 for (i = 0; i < 7; i++) {
384 f = tan((double)i * M_PI / 12.0);
385 v = FIXR(f / (1.0 + f));
390 is_table[1][6 - i] = v;
393 for (i = 7; i < 16; i++)
394 is_table[0][i] = is_table[1][i] = 0.0;
396 for (i = 0; i < 16; i++) {
400 for (j = 0; j < 2; j++) {
401 e = -(j + 1) * ((i + 1) >> 1);
404 is_table_lsf[j][k ^ 1][i] = FIXR(f);
405 is_table_lsf[j][k ][i] = FIXR(1.0);
406 av_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
407 i, j, (float) is_table_lsf[j][0][i],
408 (float) is_table_lsf[j][1][i]);
412 for (i = 0; i < 8; i++) {
415 cs = 1.0 / sqrt(1.0 + ci * ci);
418 csa_table[i][0] = FIXHR(cs/4);
419 csa_table[i][1] = FIXHR(ca/4);
420 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
421 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
423 csa_table[i][0] = cs;
424 csa_table[i][1] = ca;
425 csa_table[i][2] = ca + cs;
426 csa_table[i][3] = ca - cs;
431 static av_cold int decode_init(AVCodecContext * avctx)
433 static int initialized_tables = 0;
434 MPADecodeContext *s = avctx->priv_data;
436 if (!initialized_tables) {
437 decode_init_static();
438 initialized_tables = 1;
443 ff_mpadsp_init(&s->mpadsp);
444 ff_dsputil_init(&s->dsp, avctx);
446 if (avctx->request_sample_fmt == OUT_FMT &&
447 avctx->codec_id != AV_CODEC_ID_MP3ON4)
448 avctx->sample_fmt = OUT_FMT;
450 avctx->sample_fmt = OUT_FMT_P;
451 s->err_recognition = avctx->err_recognition;
453 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
456 avcodec_get_frame_defaults(&s->frame);
457 avctx->coded_frame = &s->frame;
462 #define C3 FIXHR(0.86602540378443864676/2)
463 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
464 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
465 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
467 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
469 static void imdct12(INTFLOAT *out, INTFLOAT *in)
471 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
474 in1 = in[1*3] + in[0*3];
475 in2 = in[2*3] + in[1*3];
476 in3 = in[3*3] + in[2*3];
477 in4 = in[4*3] + in[3*3];
478 in5 = in[5*3] + in[4*3];
482 in2 = MULH3(in2, C3, 2);
483 in3 = MULH3(in3, C3, 4);
486 t2 = MULH3(in1 - in5, C4, 2);
496 in1 = MULH3(in5 + in3, C5, 1);
503 in5 = MULH3(in5 - in3, C6, 2);
510 /* return the number of decoded frames */
511 static int mp_decode_layer1(MPADecodeContext *s)
513 int bound, i, v, n, ch, j, mant;
514 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
515 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
517 if (s->mode == MPA_JSTEREO)
518 bound = (s->mode_ext + 1) * 4;
522 /* allocation bits */
523 for (i = 0; i < bound; i++) {
524 for (ch = 0; ch < s->nb_channels; ch++) {
525 allocation[ch][i] = get_bits(&s->gb, 4);
528 for (i = bound; i < SBLIMIT; i++)
529 allocation[0][i] = get_bits(&s->gb, 4);
532 for (i = 0; i < bound; i++) {
533 for (ch = 0; ch < s->nb_channels; ch++) {
534 if (allocation[ch][i])
535 scale_factors[ch][i] = get_bits(&s->gb, 6);
538 for (i = bound; i < SBLIMIT; i++) {
539 if (allocation[0][i]) {
540 scale_factors[0][i] = get_bits(&s->gb, 6);
541 scale_factors[1][i] = get_bits(&s->gb, 6);
545 /* compute samples */
546 for (j = 0; j < 12; j++) {
547 for (i = 0; i < bound; i++) {
548 for (ch = 0; ch < s->nb_channels; ch++) {
549 n = allocation[ch][i];
551 mant = get_bits(&s->gb, n + 1);
552 v = l1_unscale(n, mant, scale_factors[ch][i]);
556 s->sb_samples[ch][j][i] = v;
559 for (i = bound; i < SBLIMIT; i++) {
560 n = allocation[0][i];
562 mant = get_bits(&s->gb, n + 1);
563 v = l1_unscale(n, mant, scale_factors[0][i]);
564 s->sb_samples[0][j][i] = v;
565 v = l1_unscale(n, mant, scale_factors[1][i]);
566 s->sb_samples[1][j][i] = v;
568 s->sb_samples[0][j][i] = 0;
569 s->sb_samples[1][j][i] = 0;
576 static int mp_decode_layer2(MPADecodeContext *s)
578 int sblimit; /* number of used subbands */
579 const unsigned char *alloc_table;
580 int table, bit_alloc_bits, i, j, ch, bound, v;
581 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
582 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
583 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
584 int scale, qindex, bits, steps, k, l, m, b;
586 /* select decoding table */
587 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
588 s->sample_rate, s->lsf);
589 sblimit = ff_mpa_sblimit_table[table];
590 alloc_table = ff_mpa_alloc_tables[table];
592 if (s->mode == MPA_JSTEREO)
593 bound = (s->mode_ext + 1) * 4;
597 av_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
603 /* parse bit allocation */
605 for (i = 0; i < bound; i++) {
606 bit_alloc_bits = alloc_table[j];
607 for (ch = 0; ch < s->nb_channels; ch++)
608 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
609 j += 1 << bit_alloc_bits;
611 for (i = bound; i < sblimit; i++) {
612 bit_alloc_bits = alloc_table[j];
613 v = get_bits(&s->gb, bit_alloc_bits);
616 j += 1 << bit_alloc_bits;
620 for (i = 0; i < sblimit; i++) {
621 for (ch = 0; ch < s->nb_channels; ch++) {
622 if (bit_alloc[ch][i])
623 scale_code[ch][i] = get_bits(&s->gb, 2);
628 for (i = 0; i < sblimit; i++) {
629 for (ch = 0; ch < s->nb_channels; ch++) {
630 if (bit_alloc[ch][i]) {
631 sf = scale_factors[ch][i];
632 switch (scale_code[ch][i]) {
635 sf[0] = get_bits(&s->gb, 6);
636 sf[1] = get_bits(&s->gb, 6);
637 sf[2] = get_bits(&s->gb, 6);
640 sf[0] = get_bits(&s->gb, 6);
645 sf[0] = get_bits(&s->gb, 6);
646 sf[2] = get_bits(&s->gb, 6);
650 sf[0] = get_bits(&s->gb, 6);
651 sf[2] = get_bits(&s->gb, 6);
660 for (k = 0; k < 3; k++) {
661 for (l = 0; l < 12; l += 3) {
663 for (i = 0; i < bound; i++) {
664 bit_alloc_bits = alloc_table[j];
665 for (ch = 0; ch < s->nb_channels; ch++) {
666 b = bit_alloc[ch][i];
668 scale = scale_factors[ch][i][k];
669 qindex = alloc_table[j+b];
670 bits = ff_mpa_quant_bits[qindex];
673 /* 3 values at the same time */
674 v = get_bits(&s->gb, -bits);
675 v2 = division_tabs[qindex][v];
676 steps = ff_mpa_quant_steps[qindex];
678 s->sb_samples[ch][k * 12 + l + 0][i] =
679 l2_unscale_group(steps, v2 & 15, scale);
680 s->sb_samples[ch][k * 12 + l + 1][i] =
681 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
682 s->sb_samples[ch][k * 12 + l + 2][i] =
683 l2_unscale_group(steps, v2 >> 8 , scale);
685 for (m = 0; m < 3; m++) {
686 v = get_bits(&s->gb, bits);
687 v = l1_unscale(bits - 1, v, scale);
688 s->sb_samples[ch][k * 12 + l + m][i] = v;
692 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
693 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
694 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
697 /* next subband in alloc table */
698 j += 1 << bit_alloc_bits;
700 /* XXX: find a way to avoid this duplication of code */
701 for (i = bound; i < sblimit; i++) {
702 bit_alloc_bits = alloc_table[j];
705 int mant, scale0, scale1;
706 scale0 = scale_factors[0][i][k];
707 scale1 = scale_factors[1][i][k];
708 qindex = alloc_table[j+b];
709 bits = ff_mpa_quant_bits[qindex];
711 /* 3 values at the same time */
712 v = get_bits(&s->gb, -bits);
713 steps = ff_mpa_quant_steps[qindex];
716 s->sb_samples[0][k * 12 + l + 0][i] =
717 l2_unscale_group(steps, mant, scale0);
718 s->sb_samples[1][k * 12 + l + 0][i] =
719 l2_unscale_group(steps, mant, scale1);
722 s->sb_samples[0][k * 12 + l + 1][i] =
723 l2_unscale_group(steps, mant, scale0);
724 s->sb_samples[1][k * 12 + l + 1][i] =
725 l2_unscale_group(steps, mant, scale1);
726 s->sb_samples[0][k * 12 + l + 2][i] =
727 l2_unscale_group(steps, v, scale0);
728 s->sb_samples[1][k * 12 + l + 2][i] =
729 l2_unscale_group(steps, v, scale1);
731 for (m = 0; m < 3; m++) {
732 mant = get_bits(&s->gb, bits);
733 s->sb_samples[0][k * 12 + l + m][i] =
734 l1_unscale(bits - 1, mant, scale0);
735 s->sb_samples[1][k * 12 + l + m][i] =
736 l1_unscale(bits - 1, mant, scale1);
740 s->sb_samples[0][k * 12 + l + 0][i] = 0;
741 s->sb_samples[0][k * 12 + l + 1][i] = 0;
742 s->sb_samples[0][k * 12 + l + 2][i] = 0;
743 s->sb_samples[1][k * 12 + l + 0][i] = 0;
744 s->sb_samples[1][k * 12 + l + 1][i] = 0;
745 s->sb_samples[1][k * 12 + l + 2][i] = 0;
747 /* next subband in alloc table */
748 j += 1 << bit_alloc_bits;
750 /* fill remaining samples to zero */
751 for (i = sblimit; i < SBLIMIT; i++) {
752 for (ch = 0; ch < s->nb_channels; ch++) {
753 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
754 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
755 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
763 #define SPLIT(dst,sf,n) \
765 int m = (sf * 171) >> 9; \
768 } else if (n == 4) { \
771 } else if (n == 5) { \
772 int m = (sf * 205) >> 10; \
775 } else if (n == 6) { \
776 int m = (sf * 171) >> 10; \
783 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
786 SPLIT(slen[3], sf, n3)
787 SPLIT(slen[2], sf, n2)
788 SPLIT(slen[1], sf, n1)
792 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
795 const uint8_t *bstab, *pretab;
796 int len, i, j, k, l, v0, shift, gain, gains[3];
800 gain = g->global_gain - 210;
801 shift = g->scalefac_scale + 1;
803 bstab = band_size_long[s->sample_rate_index];
804 pretab = mpa_pretab[g->preflag];
805 for (i = 0; i < g->long_end; i++) {
806 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
808 for (j = len; j > 0; j--)
812 if (g->short_start < 13) {
813 bstab = band_size_short[s->sample_rate_index];
814 gains[0] = gain - (g->subblock_gain[0] << 3);
815 gains[1] = gain - (g->subblock_gain[1] << 3);
816 gains[2] = gain - (g->subblock_gain[2] << 3);
818 for (i = g->short_start; i < 13; i++) {
820 for (l = 0; l < 3; l++) {
821 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
822 for (j = len; j > 0; j--)
829 /* handle n = 0 too */
830 static inline int get_bitsz(GetBitContext *s, int n)
832 return n ? get_bits(s, n) : 0;
836 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
839 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits) {
841 s->in_gb.buffer = NULL;
842 av_assert2((get_bits_count(&s->gb) & 7) == 0);
843 skip_bits_long(&s->gb, *pos - *end_pos);
845 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
846 *pos = get_bits_count(&s->gb);
850 /* Following is a optimized code for
852 if(get_bits1(&s->gb))
857 #define READ_FLIP_SIGN(dst,src) \
858 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
861 #define READ_FLIP_SIGN(dst,src) \
862 v = -get_bits1(&s->gb); \
863 *(dst) = (*(src) ^ v) - v;
866 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
867 int16_t *exponents, int end_pos2)
871 int last_pos, bits_left;
873 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits);
875 /* low frequencies (called big values) */
877 for (i = 0; i < 3; i++) {
878 int j, k, l, linbits;
879 j = g->region_size[i];
882 /* select vlc table */
883 k = g->table_select[i];
884 l = mpa_huff_data[k][0];
885 linbits = mpa_huff_data[k][1];
889 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
894 /* read huffcode and compute each couple */
898 int pos = get_bits_count(&s->gb);
901 switch_buffer(s, &pos, &end_pos, &end_pos2);
905 y = get_vlc2(&s->gb, vlc->table, 7, 3);
908 g->sb_hybrid[s_index ] =
909 g->sb_hybrid[s_index+1] = 0;
914 exponent= exponents[s_index];
916 av_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
917 i, g->region_size[i] - j, x, y, exponent);
922 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
924 x += get_bitsz(&s->gb, linbits);
925 v = l3_unscale(x, exponent);
926 if (get_bits1(&s->gb))
928 g->sb_hybrid[s_index] = v;
931 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
933 y += get_bitsz(&s->gb, linbits);
934 v = l3_unscale(y, exponent);
935 if (get_bits1(&s->gb))
937 g->sb_hybrid[s_index+1] = v;
944 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
946 x += get_bitsz(&s->gb, linbits);
947 v = l3_unscale(x, exponent);
948 if (get_bits1(&s->gb))
950 g->sb_hybrid[s_index+!!y] = v;
952 g->sb_hybrid[s_index + !y] = 0;
958 /* high frequencies */
959 vlc = &huff_quad_vlc[g->count1table_select];
961 while (s_index <= 572) {
963 pos = get_bits_count(&s->gb);
964 if (pos >= end_pos) {
965 if (pos > end_pos2 && last_pos) {
966 /* some encoders generate an incorrect size for this
967 part. We must go back into the data */
969 skip_bits_long(&s->gb, last_pos - pos);
970 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
971 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
975 switch_buffer(s, &pos, &end_pos, &end_pos2);
981 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
982 av_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
983 g->sb_hybrid[s_index+0] =
984 g->sb_hybrid[s_index+1] =
985 g->sb_hybrid[s_index+2] =
986 g->sb_hybrid[s_index+3] = 0;
988 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
990 int pos = s_index + idxtab[code];
991 code ^= 8 >> idxtab[code];
992 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
996 /* skip extension bits */
997 bits_left = end_pos2 - get_bits_count(&s->gb);
998 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
999 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1001 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
1002 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1005 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
1006 skip_bits_long(&s->gb, bits_left);
1008 i = get_bits_count(&s->gb);
1009 switch_buffer(s, &i, &end_pos, &end_pos2);
1014 /* Reorder short blocks from bitstream order to interleaved order. It
1015 would be faster to do it in parsing, but the code would be far more
1017 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1020 INTFLOAT *ptr, *dst, *ptr1;
1023 if (g->block_type != 2)
1026 if (g->switch_point) {
1027 if (s->sample_rate_index != 8)
1028 ptr = g->sb_hybrid + 36;
1030 ptr = g->sb_hybrid + 72;
1035 for (i = g->short_start; i < 13; i++) {
1036 len = band_size_short[s->sample_rate_index][i];
1039 for (j = len; j > 0; j--) {
1040 *dst++ = ptr[0*len];
1041 *dst++ = ptr[1*len];
1042 *dst++ = ptr[2*len];
1046 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1050 #define ISQRT2 FIXR(0.70710678118654752440)
1052 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1055 int sf_max, sf, len, non_zero_found;
1056 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1057 int non_zero_found_short[3];
1059 /* intensity stereo */
1060 if (s->mode_ext & MODE_EXT_I_STEREO) {
1065 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1069 tab0 = g0->sb_hybrid + 576;
1070 tab1 = g1->sb_hybrid + 576;
1072 non_zero_found_short[0] = 0;
1073 non_zero_found_short[1] = 0;
1074 non_zero_found_short[2] = 0;
1075 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1076 for (i = 12; i >= g1->short_start; i--) {
1077 /* for last band, use previous scale factor */
1080 len = band_size_short[s->sample_rate_index][i];
1081 for (l = 2; l >= 0; l--) {
1084 if (!non_zero_found_short[l]) {
1085 /* test if non zero band. if so, stop doing i-stereo */
1086 for (j = 0; j < len; j++) {
1088 non_zero_found_short[l] = 1;
1092 sf = g1->scale_factors[k + l];
1098 for (j = 0; j < len; j++) {
1100 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1101 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1105 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1106 /* lower part of the spectrum : do ms stereo
1108 for (j = 0; j < len; j++) {
1111 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1112 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1119 non_zero_found = non_zero_found_short[0] |
1120 non_zero_found_short[1] |
1121 non_zero_found_short[2];
1123 for (i = g1->long_end - 1;i >= 0;i--) {
1124 len = band_size_long[s->sample_rate_index][i];
1127 /* test if non zero band. if so, stop doing i-stereo */
1128 if (!non_zero_found) {
1129 for (j = 0; j < len; j++) {
1135 /* for last band, use previous scale factor */
1136 k = (i == 21) ? 20 : i;
1137 sf = g1->scale_factors[k];
1142 for (j = 0; j < len; j++) {
1144 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1145 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1149 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1150 /* lower part of the spectrum : do ms stereo
1152 for (j = 0; j < len; j++) {
1155 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1156 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1161 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1162 /* ms stereo ONLY */
1163 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1166 s-> dsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1168 tab0 = g0->sb_hybrid;
1169 tab1 = g1->sb_hybrid;
1170 for (i = 0; i < 576; i++) {
1173 tab0[i] = tmp0 + tmp1;
1174 tab1[i] = tmp0 - tmp1;
1182 # include "mips/compute_antialias_float.h"
1183 #endif /* HAVE_MIPSFPU */
1186 # include "mips/compute_antialias_fixed.h"
1187 #endif /* HAVE_MIPSDSPR1 */
1188 #endif /* CONFIG_FLOAT */
1190 #ifndef compute_antialias
1192 #define AA(j) do { \
1193 float tmp0 = ptr[-1-j]; \
1194 float tmp1 = ptr[ j]; \
1195 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1196 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1199 #define AA(j) do { \
1200 int tmp0 = ptr[-1-j]; \
1201 int tmp1 = ptr[ j]; \
1202 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1203 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1204 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1208 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1213 /* we antialias only "long" bands */
1214 if (g->block_type == 2) {
1215 if (!g->switch_point)
1217 /* XXX: check this for 8000Hz case */
1223 ptr = g->sb_hybrid + 18;
1224 for (i = n; i > 0; i--) {
1237 #endif /* compute_antialias */
1239 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1240 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1242 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1244 int i, j, mdct_long_end, sblimit;
1246 /* find last non zero block */
1247 ptr = g->sb_hybrid + 576;
1248 ptr1 = g->sb_hybrid + 2 * 18;
1249 while (ptr >= ptr1) {
1253 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1256 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1258 if (g->block_type == 2) {
1259 /* XXX: check for 8000 Hz */
1260 if (g->switch_point)
1265 mdct_long_end = sblimit;
1268 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1269 mdct_long_end, g->switch_point,
1272 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1273 ptr = g->sb_hybrid + 18 * mdct_long_end;
1275 for (j = mdct_long_end; j < sblimit; j++) {
1276 /* select frequency inversion */
1277 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1278 out_ptr = sb_samples + j;
1280 for (i = 0; i < 6; i++) {
1281 *out_ptr = buf[4*i];
1284 imdct12(out2, ptr + 0);
1285 for (i = 0; i < 6; i++) {
1286 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1287 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1290 imdct12(out2, ptr + 1);
1291 for (i = 0; i < 6; i++) {
1292 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1293 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1296 imdct12(out2, ptr + 2);
1297 for (i = 0; i < 6; i++) {
1298 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1299 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1300 buf[4*(i + 6*2)] = 0;
1303 buf += (j&3) != 3 ? 1 : (4*18-3);
1306 for (j = sblimit; j < SBLIMIT; j++) {
1308 out_ptr = sb_samples + j;
1309 for (i = 0; i < 18; i++) {
1310 *out_ptr = buf[4*i];
1314 buf += (j&3) != 3 ? 1 : (4*18-3);
1318 /* main layer3 decoding function */
1319 static int mp_decode_layer3(MPADecodeContext *s)
1321 int nb_granules, main_data_begin;
1322 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1324 int16_t exponents[576]; //FIXME try INTFLOAT
1326 /* read side info */
1328 main_data_begin = get_bits(&s->gb, 8);
1329 skip_bits(&s->gb, s->nb_channels);
1332 main_data_begin = get_bits(&s->gb, 9);
1333 if (s->nb_channels == 2)
1334 skip_bits(&s->gb, 3);
1336 skip_bits(&s->gb, 5);
1338 for (ch = 0; ch < s->nb_channels; ch++) {
1339 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1340 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1344 for (gr = 0; gr < nb_granules; gr++) {
1345 for (ch = 0; ch < s->nb_channels; ch++) {
1346 av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1347 g = &s->granules[ch][gr];
1348 g->part2_3_length = get_bits(&s->gb, 12);
1349 g->big_values = get_bits(&s->gb, 9);
1350 if (g->big_values > 288) {
1351 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1352 return AVERROR_INVALIDDATA;
1355 g->global_gain = get_bits(&s->gb, 8);
1356 /* if MS stereo only is selected, we precompute the
1357 1/sqrt(2) renormalization factor */
1358 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1360 g->global_gain -= 2;
1362 g->scalefac_compress = get_bits(&s->gb, 9);
1364 g->scalefac_compress = get_bits(&s->gb, 4);
1365 blocksplit_flag = get_bits1(&s->gb);
1366 if (blocksplit_flag) {
1367 g->block_type = get_bits(&s->gb, 2);
1368 if (g->block_type == 0) {
1369 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1370 return AVERROR_INVALIDDATA;
1372 g->switch_point = get_bits1(&s->gb);
1373 for (i = 0; i < 2; i++)
1374 g->table_select[i] = get_bits(&s->gb, 5);
1375 for (i = 0; i < 3; i++)
1376 g->subblock_gain[i] = get_bits(&s->gb, 3);
1377 ff_init_short_region(s, g);
1379 int region_address1, region_address2;
1381 g->switch_point = 0;
1382 for (i = 0; i < 3; i++)
1383 g->table_select[i] = get_bits(&s->gb, 5);
1384 /* compute huffman coded region sizes */
1385 region_address1 = get_bits(&s->gb, 4);
1386 region_address2 = get_bits(&s->gb, 3);
1387 av_dlog(s->avctx, "region1=%d region2=%d\n",
1388 region_address1, region_address2);
1389 ff_init_long_region(s, g, region_address1, region_address2);
1391 ff_region_offset2size(g);
1392 ff_compute_band_indexes(s, g);
1396 g->preflag = get_bits1(&s->gb);
1397 g->scalefac_scale = get_bits1(&s->gb);
1398 g->count1table_select = get_bits1(&s->gb);
1399 av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1400 g->block_type, g->switch_point);
1406 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1407 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0, EXTRABYTES);
1408 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1409 /* now we get bits from the main_data_begin offset */
1410 av_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1411 main_data_begin, s->last_buf_size);
1413 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1415 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1416 #if !UNCHECKED_BITSTREAM_READER
1417 s->gb.size_in_bits_plus8 += FFMAX(extrasize, LAST_BUF_SIZE - s->last_buf_size) * 8;
1419 s->last_buf_size <<= 3;
1420 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1421 for (ch = 0; ch < s->nb_channels; ch++) {
1422 g = &s->granules[ch][gr];
1423 s->last_buf_size += g->part2_3_length;
1424 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1425 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1428 skip = s->last_buf_size - 8 * main_data_begin;
1429 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1430 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1432 s->in_gb.buffer = NULL;
1434 skip_bits_long(&s->gb, skip);
1440 for (; gr < nb_granules; gr++) {
1441 for (ch = 0; ch < s->nb_channels; ch++) {
1442 g = &s->granules[ch][gr];
1443 bits_pos = get_bits_count(&s->gb);
1447 int slen, slen1, slen2;
1449 /* MPEG1 scale factors */
1450 slen1 = slen_table[0][g->scalefac_compress];
1451 slen2 = slen_table[1][g->scalefac_compress];
1452 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1453 if (g->block_type == 2) {
1454 n = g->switch_point ? 17 : 18;
1457 for (i = 0; i < n; i++)
1458 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1460 for (i = 0; i < n; i++)
1461 g->scale_factors[j++] = 0;
1464 for (i = 0; i < 18; i++)
1465 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1466 for (i = 0; i < 3; i++)
1467 g->scale_factors[j++] = 0;
1469 for (i = 0; i < 21; i++)
1470 g->scale_factors[j++] = 0;
1473 sc = s->granules[ch][0].scale_factors;
1475 for (k = 0; k < 4; k++) {
1477 if ((g->scfsi & (0x8 >> k)) == 0) {
1478 slen = (k < 2) ? slen1 : slen2;
1480 for (i = 0; i < n; i++)
1481 g->scale_factors[j++] = get_bits(&s->gb, slen);
1483 for (i = 0; i < n; i++)
1484 g->scale_factors[j++] = 0;
1487 /* simply copy from last granule */
1488 for (i = 0; i < n; i++) {
1489 g->scale_factors[j] = sc[j];
1494 g->scale_factors[j++] = 0;
1497 int tindex, tindex2, slen[4], sl, sf;
1499 /* LSF scale factors */
1500 if (g->block_type == 2)
1501 tindex = g->switch_point ? 2 : 1;
1505 sf = g->scalefac_compress;
1506 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1507 /* intensity stereo case */
1510 lsf_sf_expand(slen, sf, 6, 6, 0);
1512 } else if (sf < 244) {
1513 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1516 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1522 lsf_sf_expand(slen, sf, 5, 4, 4);
1524 } else if (sf < 500) {
1525 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1528 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1535 for (k = 0; k < 4; k++) {
1536 n = lsf_nsf_table[tindex2][tindex][k];
1539 for (i = 0; i < n; i++)
1540 g->scale_factors[j++] = get_bits(&s->gb, sl);
1542 for (i = 0; i < n; i++)
1543 g->scale_factors[j++] = 0;
1546 /* XXX: should compute exact size */
1548 g->scale_factors[j] = 0;
1551 exponents_from_scale_factors(s, g, exponents);
1553 /* read Huffman coded residue */
1554 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1557 if (s->mode == MPA_JSTEREO)
1558 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1560 for (ch = 0; ch < s->nb_channels; ch++) {
1561 g = &s->granules[ch][gr];
1563 reorder_block(s, g);
1564 compute_antialias(s, g);
1565 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1568 if (get_bits_count(&s->gb) < 0)
1569 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1570 return nb_granules * 18;
1573 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1574 const uint8_t *buf, int buf_size)
1576 int i, nb_frames, ch, ret;
1577 OUT_INT *samples_ptr;
1579 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1581 /* skip error protection field */
1582 if (s->error_protection)
1583 skip_bits(&s->gb, 16);
1587 s->avctx->frame_size = 384;
1588 nb_frames = mp_decode_layer1(s);
1591 s->avctx->frame_size = 1152;
1592 nb_frames = mp_decode_layer2(s);
1595 s->avctx->frame_size = s->lsf ? 576 : 1152;
1597 nb_frames = mp_decode_layer3(s);
1600 if (s->in_gb.buffer) {
1601 align_get_bits(&s->gb);
1602 i = get_bits_left(&s->gb)>>3;
1603 if (i >= 0 && i <= BACKSTEP_SIZE) {
1604 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1607 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1609 s->in_gb.buffer = NULL;
1612 align_get_bits(&s->gb);
1613 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1614 i = get_bits_left(&s->gb) >> 3;
1616 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1618 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1619 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1621 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1622 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1623 s->last_buf_size += i;
1629 /* get output buffer */
1631 s->frame.nb_samples = s->avctx->frame_size;
1632 if ((ret = s->avctx->get_buffer(s->avctx, &s->frame)) < 0) {
1633 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1636 samples = (OUT_INT **)s->frame.extended_data;
1639 /* apply the synthesis filter */
1640 for (ch = 0; ch < s->nb_channels; ch++) {
1642 if (s->avctx->sample_fmt == OUT_FMT_P) {
1643 samples_ptr = samples[ch];
1646 samples_ptr = samples[0] + ch;
1647 sample_stride = s->nb_channels;
1649 for (i = 0; i < nb_frames; i++) {
1650 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1651 &(s->synth_buf_offset[ch]),
1652 RENAME(ff_mpa_synth_window),
1653 &s->dither_state, samples_ptr,
1654 sample_stride, s->sb_samples[ch][i]);
1655 samples_ptr += 32 * sample_stride;
1659 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1662 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1665 const uint8_t *buf = avpkt->data;
1666 int buf_size = avpkt->size;
1667 MPADecodeContext *s = avctx->priv_data;
1671 while(buf_size && !*buf){
1676 if (buf_size < HEADER_SIZE)
1677 return AVERROR_INVALIDDATA;
1679 header = AV_RB32(buf);
1680 if (header>>8 == AV_RB32("TAG")>>8) {
1681 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1684 if (ff_mpa_check_header(header) < 0) {
1685 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1686 return AVERROR_INVALIDDATA;
1689 if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
1690 /* free format: prepare to compute frame size */
1692 return AVERROR_INVALIDDATA;
1694 /* update codec info */
1695 avctx->channels = s->nb_channels;
1696 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1697 if (!avctx->bit_rate)
1698 avctx->bit_rate = s->bit_rate;
1700 if (s->frame_size <= 0 || s->frame_size > buf_size) {
1701 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1702 return AVERROR_INVALIDDATA;
1703 } else if (s->frame_size < buf_size) {
1704 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1705 buf_size= s->frame_size;
1708 ret = mp_decode_frame(s, NULL, buf, buf_size);
1711 *(AVFrame *)data = s->frame;
1712 avctx->sample_rate = s->sample_rate;
1713 //FIXME maybe move the other codec info stuff from above here too
1715 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1716 /* Only return an error if the bad frame makes up the whole packet or
1717 * the error is related to buffer management.
1718 * If there is more data in the packet, just consume the bad frame
1719 * instead of returning an error, which would discard the whole
1722 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1729 static void mp_flush(MPADecodeContext *ctx)
1731 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1732 ctx->last_buf_size = 0;
1735 static void flush(AVCodecContext *avctx)
1737 mp_flush(avctx->priv_data);
1740 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1741 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1742 int *got_frame_ptr, AVPacket *avpkt)
1744 const uint8_t *buf = avpkt->data;
1745 int buf_size = avpkt->size;
1746 MPADecodeContext *s = avctx->priv_data;
1749 int av_unused out_size;
1753 // Discard too short frames
1754 if (buf_size < HEADER_SIZE) {
1755 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1756 return AVERROR_INVALIDDATA;
1760 if (len > MPA_MAX_CODED_FRAME_SIZE)
1761 len = MPA_MAX_CODED_FRAME_SIZE;
1763 // Get header and restore sync word
1764 header = AV_RB32(buf) | 0xffe00000;
1766 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1767 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1768 return AVERROR_INVALIDDATA;
1771 avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1772 /* update codec info */
1773 avctx->sample_rate = s->sample_rate;
1774 avctx->channels = s->nb_channels;
1775 if (!avctx->bit_rate)
1776 avctx->bit_rate = s->bit_rate;
1778 s->frame_size = len;
1780 ret = mp_decode_frame(s, NULL, buf, buf_size);
1782 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1787 *(AVFrame *)data = s->frame;
1791 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1793 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1796 * Context for MP3On4 decoder
1798 typedef struct MP3On4DecodeContext {
1800 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1801 int syncword; ///< syncword patch
1802 const uint8_t *coff; ///< channel offsets in output buffer
1803 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1804 } MP3On4DecodeContext;
1806 #include "mpeg4audio.h"
1808 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1810 /* number of mp3 decoder instances */
1811 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1813 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1814 static const uint8_t chan_offset[8][5] = {
1819 { 2, 0, 3 }, // C FLR BS
1820 { 2, 0, 3 }, // C FLR BLRS
1821 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1822 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1825 /* mp3on4 channel layouts */
1826 static const int16_t chan_layout[8] = {
1829 AV_CH_LAYOUT_STEREO,
1830 AV_CH_LAYOUT_SURROUND,
1831 AV_CH_LAYOUT_4POINT0,
1832 AV_CH_LAYOUT_5POINT0,
1833 AV_CH_LAYOUT_5POINT1,
1834 AV_CH_LAYOUT_7POINT1
1837 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1839 MP3On4DecodeContext *s = avctx->priv_data;
1842 for (i = 0; i < s->frames; i++)
1843 av_free(s->mp3decctx[i]);
1849 static int decode_init_mp3on4(AVCodecContext * avctx)
1851 MP3On4DecodeContext *s = avctx->priv_data;
1852 MPEG4AudioConfig cfg;
1855 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
1856 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1857 return AVERROR_INVALIDDATA;
1860 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1861 avctx->extradata_size * 8, 1);
1862 if (!cfg.chan_config || cfg.chan_config > 7) {
1863 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1864 return AVERROR_INVALIDDATA;
1866 s->frames = mp3Frames[cfg.chan_config];
1867 s->coff = chan_offset[cfg.chan_config];
1868 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1869 avctx->channel_layout = chan_layout[cfg.chan_config];
1871 if (cfg.sample_rate < 16000)
1872 s->syncword = 0xffe00000;
1874 s->syncword = 0xfff00000;
1876 /* Init the first mp3 decoder in standard way, so that all tables get builded
1877 * We replace avctx->priv_data with the context of the first decoder so that
1878 * decode_init() does not have to be changed.
1879 * Other decoders will be initialized here copying data from the first context
1881 // Allocate zeroed memory for the first decoder context
1882 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1883 if (!s->mp3decctx[0])
1885 // Put decoder context in place to make init_decode() happy
1886 avctx->priv_data = s->mp3decctx[0];
1888 s->frame = avctx->coded_frame;
1889 // Restore mp3on4 context pointer
1890 avctx->priv_data = s;
1891 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1893 /* Create a separate codec/context for each frame (first is already ok).
1894 * Each frame is 1 or 2 channels - up to 5 frames allowed
1896 for (i = 1; i < s->frames; i++) {
1897 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1898 if (!s->mp3decctx[i])
1900 s->mp3decctx[i]->adu_mode = 1;
1901 s->mp3decctx[i]->avctx = avctx;
1902 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1907 decode_close_mp3on4(avctx);
1908 return AVERROR(ENOMEM);
1912 static void flush_mp3on4(AVCodecContext *avctx)
1915 MP3On4DecodeContext *s = avctx->priv_data;
1917 for (i = 0; i < s->frames; i++)
1918 mp_flush(s->mp3decctx[i]);
1922 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1923 int *got_frame_ptr, AVPacket *avpkt)
1925 const uint8_t *buf = avpkt->data;
1926 int buf_size = avpkt->size;
1927 MP3On4DecodeContext *s = avctx->priv_data;
1928 MPADecodeContext *m;
1929 int fsize, len = buf_size, out_size = 0;
1931 OUT_INT **out_samples;
1935 /* get output buffer */
1936 s->frame->nb_samples = s->frames * MPA_FRAME_SIZE;
1937 if ((ret = avctx->get_buffer(avctx, s->frame)) < 0) {
1938 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1941 out_samples = (OUT_INT **)s->frame->extended_data;
1943 // Discard too short frames
1944 if (buf_size < HEADER_SIZE)
1945 return AVERROR_INVALIDDATA;
1947 avctx->bit_rate = 0;
1950 for (fr = 0; fr < s->frames; fr++) {
1951 fsize = AV_RB16(buf) >> 4;
1952 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1953 m = s->mp3decctx[fr];
1956 if (fsize < HEADER_SIZE) {
1957 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1958 return AVERROR_INVALIDDATA;
1960 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1962 if (ff_mpa_check_header(header) < 0) // Bad header, discard block
1965 avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1967 if (ch + m->nb_channels > avctx->channels) {
1968 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1970 return AVERROR_INVALIDDATA;
1972 ch += m->nb_channels;
1974 outptr[0] = out_samples[s->coff[fr]];
1975 if (m->nb_channels > 1)
1976 outptr[1] = out_samples[s->coff[fr] + 1];
1978 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0)
1985 avctx->bit_rate += m->bit_rate;
1988 /* update codec info */
1989 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1991 s->frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1993 *(AVFrame *)data = *s->frame;
1997 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */
2000 #if CONFIG_MP1_DECODER
2001 AVCodec ff_mp1_decoder = {
2003 .type = AVMEDIA_TYPE_AUDIO,
2004 .id = AV_CODEC_ID_MP1,
2005 .priv_data_size = sizeof(MPADecodeContext),
2006 .init = decode_init,
2007 .decode = decode_frame,
2008 .capabilities = CODEC_CAP_DR1,
2010 .long_name = NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
2011 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2013 AV_SAMPLE_FMT_NONE },
2016 #if CONFIG_MP2_DECODER
2017 AVCodec ff_mp2_decoder = {
2019 .type = AVMEDIA_TYPE_AUDIO,
2020 .id = AV_CODEC_ID_MP2,
2021 .priv_data_size = sizeof(MPADecodeContext),
2022 .init = decode_init,
2023 .decode = decode_frame,
2024 .capabilities = CODEC_CAP_DR1,
2026 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
2027 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2029 AV_SAMPLE_FMT_NONE },
2032 #if CONFIG_MP3_DECODER
2033 AVCodec ff_mp3_decoder = {
2035 .type = AVMEDIA_TYPE_AUDIO,
2036 .id = AV_CODEC_ID_MP3,
2037 .priv_data_size = sizeof(MPADecodeContext),
2038 .init = decode_init,
2039 .decode = decode_frame,
2040 .capabilities = CODEC_CAP_DR1,
2042 .long_name = NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),
2043 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2045 AV_SAMPLE_FMT_NONE },
2048 #if CONFIG_MP3ADU_DECODER
2049 AVCodec ff_mp3adu_decoder = {
2051 .type = AVMEDIA_TYPE_AUDIO,
2052 .id = AV_CODEC_ID_MP3ADU,
2053 .priv_data_size = sizeof(MPADecodeContext),
2054 .init = decode_init,
2055 .decode = decode_frame_adu,
2056 .capabilities = CODEC_CAP_DR1,
2058 .long_name = NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),
2059 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2061 AV_SAMPLE_FMT_NONE },
2064 #if CONFIG_MP3ON4_DECODER
2065 AVCodec ff_mp3on4_decoder = {
2067 .type = AVMEDIA_TYPE_AUDIO,
2068 .id = AV_CODEC_ID_MP3ON4,
2069 .priv_data_size = sizeof(MP3On4DecodeContext),
2070 .init = decode_init_mp3on4,
2071 .close = decode_close_mp3on4,
2072 .decode = decode_frame_mp3on4,
2073 .capabilities = CODEC_CAP_DR1,
2074 .flush = flush_mp3on4,
2075 .long_name = NULL_IF_CONFIG_SMALL("MP3onMP4"),
2076 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2077 AV_SAMPLE_FMT_NONE },