3 * Copyright (c) 2001, 2002 Fabrice Bellard
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include "libavutil/audioconvert.h"
31 #include "mpegaudiodsp.h"
35 * - test lsf / mpeg25 extensively.
38 #include "mpegaudio.h"
39 #include "mpegaudiodecheader.h"
41 #define BACKSTEP_SIZE 512
44 /* layer 3 "granule" */
45 typedef struct GranuleDef {
50 int scalefac_compress;
55 uint8_t scalefac_scale;
56 uint8_t count1table_select;
57 int region_size[3]; /* number of huffman codes in each region */
59 int short_start, long_end; /* long/short band indexes */
60 uint8_t scale_factors[40];
61 INTFLOAT sb_hybrid[SBLIMIT * 18]; /* 576 samples */
64 typedef struct MPADecodeContext {
66 uint8_t last_buf[2*BACKSTEP_SIZE + EXTRABYTES];
68 /* next header (used in free format parsing) */
69 uint32_t free_format_next_header;
72 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
73 int synth_buf_offset[MPA_MAX_CHANNELS];
74 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
75 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
76 GranuleDef granules[2][2]; /* Used in Layer 3 */
80 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
82 int error_recognition;
83 AVCodecContext* avctx;
88 # define SHR(a,b) ((a)*(1.0f/(1<<(b))))
89 # define FIXR_OLD(a) ((int)((a) * FRAC_ONE + 0.5))
90 # define FIXR(x) ((float)(x))
91 # define FIXHR(x) ((float)(x))
92 # define MULH3(x, y, s) ((s)*(y)*(x))
93 # define MULLx(x, y, s) ((y)*(x))
94 # define RENAME(a) a ## _float
95 # define OUT_FMT AV_SAMPLE_FMT_FLT
97 # define SHR(a,b) ((a)>>(b))
98 /* WARNING: only correct for posititive numbers */
99 # define FIXR_OLD(a) ((int)((a) * FRAC_ONE + 0.5))
100 # define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
101 # define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
102 # define MULH3(x, y, s) MULH((s)*(x), y)
103 # define MULLx(x, y, s) MULL(x,y,s)
104 # define RENAME(a) a ## _fixed
105 # define OUT_FMT AV_SAMPLE_FMT_S16
110 #define HEADER_SIZE 4
112 #include "mpegaudiodata.h"
113 #include "mpegaudiodectab.h"
115 /* vlc structure for decoding layer 3 huffman tables */
116 static VLC huff_vlc[16];
117 static VLC_TYPE huff_vlc_tables[
118 0+128+128+128+130+128+154+166+
119 142+204+190+170+542+460+662+414
121 static const int huff_vlc_tables_sizes[16] = {
122 0, 128, 128, 128, 130, 128, 154, 166,
123 142, 204, 190, 170, 542, 460, 662, 414
125 static VLC huff_quad_vlc[2];
126 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
127 static const int huff_quad_vlc_tables_sizes[2] = {
130 /* computed from band_size_long */
131 static uint16_t band_index_long[9][23];
132 #include "mpegaudio_tablegen.h"
133 /* intensity stereo coef table */
134 static INTFLOAT is_table[2][16];
135 static INTFLOAT is_table_lsf[2][2][16];
136 static INTFLOAT csa_table[8][4];
137 static INTFLOAT mdct_win[8][36];
139 static int16_t division_tab3[1<<6 ];
140 static int16_t division_tab5[1<<8 ];
141 static int16_t division_tab9[1<<11];
143 static int16_t * const division_tabs[4] = {
144 division_tab3, division_tab5, NULL, division_tab9
147 /* lower 2 bits: modulo 3, higher bits: shift */
148 static uint16_t scale_factor_modshift[64];
149 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
150 static int32_t scale_factor_mult[15][3];
151 /* mult table for layer 2 group quantization */
153 #define SCALE_GEN(v) \
154 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
156 static const int32_t scale_factor_mult2[3][3] = {
157 SCALE_GEN(4.0 / 3.0), /* 3 steps */
158 SCALE_GEN(4.0 / 5.0), /* 5 steps */
159 SCALE_GEN(4.0 / 9.0), /* 9 steps */
163 * Convert region offsets to region sizes and truncate
164 * size to big_values.
166 static void ff_region_offset2size(GranuleDef *g){
168 g->region_size[2] = (576 / 2);
170 k = FFMIN(g->region_size[i], g->big_values);
171 g->region_size[i] = k - j;
176 static void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){
177 if (g->block_type == 2)
178 g->region_size[0] = (36 / 2);
180 if (s->sample_rate_index <= 2)
181 g->region_size[0] = (36 / 2);
182 else if (s->sample_rate_index != 8)
183 g->region_size[0] = (54 / 2);
185 g->region_size[0] = (108 / 2);
187 g->region_size[1] = (576 / 2);
190 static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){
193 band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
194 /* should not overflow */
195 l = FFMIN(ra1 + ra2 + 2, 22);
197 band_index_long[s->sample_rate_index][l] >> 1;
200 static void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){
201 if (g->block_type == 2) {
202 if (g->switch_point) {
203 /* if switched mode, we handle the 36 first samples as
204 long blocks. For 8000Hz, we handle the 48 first
205 exponents as long blocks (XXX: check this!) */
206 if (s->sample_rate_index <= 2)
208 else if (s->sample_rate_index != 8)
211 g->long_end = 4; /* 8000 Hz */
213 g->short_start = 2 + (s->sample_rate_index != 8);
224 /* layer 1 unscaling */
225 /* n = number of bits of the mantissa minus 1 */
226 static inline int l1_unscale(int n, int mant, int scale_factor)
231 shift = scale_factor_modshift[scale_factor];
234 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
236 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
237 return (int)((val + (1LL << (shift - 1))) >> shift);
240 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
244 shift = scale_factor_modshift[scale_factor];
248 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
249 /* NOTE: at this point, 0 <= shift <= 21 */
251 val = (val + (1 << (shift - 1))) >> shift;
255 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
256 static inline int l3_unscale(int value, int exponent)
261 e = table_4_3_exp [4*value + (exponent&3)];
262 m = table_4_3_value[4*value + (exponent&3)];
263 e -= (exponent >> 2);
267 m = (m + (1 << (e-1))) >> e;
272 static av_cold int decode_init(AVCodecContext * avctx)
274 MPADecodeContext *s = avctx->priv_data;
280 ff_mpadsp_init(&s->mpadsp);
282 avctx->sample_fmt= OUT_FMT;
283 s->error_recognition= avctx->error_recognition;
285 if (!init && !avctx->parse_only) {
288 /* scale factors table for layer 1/2 */
291 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
294 scale_factor_modshift[i] = mod | (shift << 2);
297 /* scale factor multiply for layer 1 */
301 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
302 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
303 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
304 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
305 av_dlog(avctx, "%d: norm=%x s=%x %x %x\n",
307 scale_factor_mult[i][0],
308 scale_factor_mult[i][1],
309 scale_factor_mult[i][2]);
312 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
314 /* huffman decode tables */
317 const HuffTable *h = &mpa_huff_tables[i];
319 uint8_t tmp_bits [512];
320 uint16_t tmp_codes[512];
322 memset(tmp_bits , 0, sizeof(tmp_bits ));
323 memset(tmp_codes, 0, sizeof(tmp_codes));
328 for(x=0;x<xsize;x++) {
329 for(y=0;y<xsize;y++){
330 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
331 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
336 huff_vlc[i].table = huff_vlc_tables+offset;
337 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
338 init_vlc(&huff_vlc[i], 7, 512,
339 tmp_bits, 1, 1, tmp_codes, 2, 2,
340 INIT_VLC_USE_NEW_STATIC);
341 offset += huff_vlc_tables_sizes[i];
343 assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
347 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
348 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
349 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
350 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
351 INIT_VLC_USE_NEW_STATIC);
352 offset += huff_quad_vlc_tables_sizes[i];
354 assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
359 band_index_long[i][j] = k;
360 k += band_size_long[i][j];
362 band_index_long[i][22] = k;
365 /* compute n ^ (4/3) and store it in mantissa/exp format */
367 mpegaudio_tableinit();
369 for (i = 0; i < 4; i++)
370 if (ff_mpa_quant_bits[i] < 0)
371 for (j = 0; j < (1<<(-ff_mpa_quant_bits[i]+1)); j++) {
372 int val1, val2, val3, steps;
374 steps = ff_mpa_quant_steps[i];
379 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
387 f = tan((double)i * M_PI / 12.0);
388 v = FIXR(f / (1.0 + f));
393 is_table[1][6 - i] = v;
397 is_table[0][i] = is_table[1][i] = 0.0;
404 e = -(j + 1) * ((i + 1) >> 1);
405 f = pow(2.0, e / 4.0);
407 is_table_lsf[j][k ^ 1][i] = FIXR(f);
408 is_table_lsf[j][k][i] = FIXR(1.0);
409 av_dlog(avctx, "is_table_lsf %d %d: %f %f\n",
410 i, j, (float) is_table_lsf[j][0][i],
411 (float) is_table_lsf[j][1][i]);
418 cs = 1.0 / sqrt(1.0 + ci * ci);
421 csa_table[i][0] = FIXHR(cs/4);
422 csa_table[i][1] = FIXHR(ca/4);
423 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
424 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
426 csa_table[i][0] = cs;
427 csa_table[i][1] = ca;
428 csa_table[i][2] = ca + cs;
429 csa_table[i][3] = ca - cs;
433 /* compute mdct windows */
441 d= sin(M_PI * (i + 0.5) / 36.0);
444 else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
448 else if(i< 12) d= sin(M_PI * (i - 6 + 0.5) / 12.0);
451 //merge last stage of imdct into the window coefficients
452 d*= 0.5 / cos(M_PI*(2*i + 19)/72);
455 mdct_win[j][i/3] = FIXHR((d / (1<<5)));
457 mdct_win[j][i ] = FIXHR((d / (1<<5)));
461 /* NOTE: we do frequency inversion adter the MDCT by changing
462 the sign of the right window coefs */
465 mdct_win[j + 4][i] = mdct_win[j][i];
466 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
473 if (avctx->codec_id == CODEC_ID_MP3ADU)
478 #define C3 FIXHR(0.86602540378443864676/2)
480 /* 0.5 / cos(pi*(2*i+1)/36) */
481 static const INTFLOAT icos36[9] = {
482 FIXR(0.50190991877167369479),
483 FIXR(0.51763809020504152469), //0
484 FIXR(0.55168895948124587824),
485 FIXR(0.61038729438072803416),
486 FIXR(0.70710678118654752439), //1
487 FIXR(0.87172339781054900991),
488 FIXR(1.18310079157624925896),
489 FIXR(1.93185165257813657349), //2
490 FIXR(5.73685662283492756461),
493 /* 0.5 / cos(pi*(2*i+1)/36) */
494 static const INTFLOAT icos36h[9] = {
495 FIXHR(0.50190991877167369479/2),
496 FIXHR(0.51763809020504152469/2), //0
497 FIXHR(0.55168895948124587824/2),
498 FIXHR(0.61038729438072803416/2),
499 FIXHR(0.70710678118654752439/2), //1
500 FIXHR(0.87172339781054900991/2),
501 FIXHR(1.18310079157624925896/4),
502 FIXHR(1.93185165257813657349/4), //2
503 // FIXHR(5.73685662283492756461),
506 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
508 static void imdct12(INTFLOAT *out, INTFLOAT *in)
510 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
513 in1= in[1*3] + in[0*3];
514 in2= in[2*3] + in[1*3];
515 in3= in[3*3] + in[2*3];
516 in4= in[4*3] + in[3*3];
517 in5= in[5*3] + in[4*3];
521 in2= MULH3(in2, C3, 2);
522 in3= MULH3(in3, C3, 4);
525 t2 = MULH3(in1 - in5, icos36h[4], 2);
535 in1 = MULH3(in5 + in3, icos36h[1], 1);
542 in5 = MULH3(in5 - in3, icos36h[7], 2);
550 #define C1 FIXHR(0.98480775301220805936/2)
551 #define C2 FIXHR(0.93969262078590838405/2)
552 #define C3 FIXHR(0.86602540378443864676/2)
553 #define C4 FIXHR(0.76604444311897803520/2)
554 #define C5 FIXHR(0.64278760968653932632/2)
555 #define C6 FIXHR(0.5/2)
556 #define C7 FIXHR(0.34202014332566873304/2)
557 #define C8 FIXHR(0.17364817766693034885/2)
560 /* using Lee like decomposition followed by hand coded 9 points DCT */
561 static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win)
564 INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
565 INTFLOAT tmp[18], *tmp1, *in1;
576 t2 = in1[2*4] + in1[2*8] - in1[2*2];
578 t3 = in1[2*0] + SHR(in1[2*6],1);
579 t1 = in1[2*0] - in1[2*6];
580 tmp1[ 6] = t1 - SHR(t2,1);
583 t0 = MULH3(in1[2*2] + in1[2*4] , C2, 2);
584 t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
585 t2 = MULH3(in1[2*2] + in1[2*8] , -C4, 2);
587 tmp1[10] = t3 - t0 - t2;
588 tmp1[ 2] = t3 + t0 + t1;
589 tmp1[14] = t3 + t2 - t1;
591 tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
592 t2 = MULH3(in1[2*1] + in1[2*5], C1, 2);
593 t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
594 t0 = MULH3(in1[2*3], C3, 2);
596 t1 = MULH3(in1[2*1] + in1[2*7], -C5, 2);
598 tmp1[ 0] = t2 + t3 + t0;
599 tmp1[12] = t2 + t1 - t0;
600 tmp1[ 8] = t3 - t1 - t0;
612 s1 = MULH3(t3 + t2, icos36h[j], 2);
613 s3 = MULLx(t3 - t2, icos36[8 - j], FRAC_BITS);
617 out[(9 + j)*SBLIMIT] = MULH3(t1, win[9 + j], 1) + buf[9 + j];
618 out[(8 - j)*SBLIMIT] = MULH3(t1, win[8 - j], 1) + buf[8 - j];
619 buf[9 + j] = MULH3(t0, win[18 + 9 + j], 1);
620 buf[8 - j] = MULH3(t0, win[18 + 8 - j], 1);
624 out[(9 + 8 - j)*SBLIMIT] = MULH3(t1, win[9 + 8 - j], 1) + buf[9 + 8 - j];
625 out[( j)*SBLIMIT] = MULH3(t1, win[ j], 1) + buf[ j];
626 buf[9 + 8 - j] = MULH3(t0, win[18 + 9 + 8 - j], 1);
627 buf[ + j] = MULH3(t0, win[18 + j], 1);
632 s1 = MULH3(tmp[17], icos36h[4], 2);
635 out[(9 + 4)*SBLIMIT] = MULH3(t1, win[9 + 4], 1) + buf[9 + 4];
636 out[(8 - 4)*SBLIMIT] = MULH3(t1, win[8 - 4], 1) + buf[8 - 4];
637 buf[9 + 4] = MULH3(t0, win[18 + 9 + 4], 1);
638 buf[8 - 4] = MULH3(t0, win[18 + 8 - 4], 1);
641 /* return the number of decoded frames */
642 static int mp_decode_layer1(MPADecodeContext *s)
644 int bound, i, v, n, ch, j, mant;
645 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
646 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
648 if (s->mode == MPA_JSTEREO)
649 bound = (s->mode_ext + 1) * 4;
653 /* allocation bits */
654 for(i=0;i<bound;i++) {
655 for(ch=0;ch<s->nb_channels;ch++) {
656 allocation[ch][i] = get_bits(&s->gb, 4);
659 for(i=bound;i<SBLIMIT;i++) {
660 allocation[0][i] = get_bits(&s->gb, 4);
664 for(i=0;i<bound;i++) {
665 for(ch=0;ch<s->nb_channels;ch++) {
666 if (allocation[ch][i])
667 scale_factors[ch][i] = get_bits(&s->gb, 6);
670 for(i=bound;i<SBLIMIT;i++) {
671 if (allocation[0][i]) {
672 scale_factors[0][i] = get_bits(&s->gb, 6);
673 scale_factors[1][i] = get_bits(&s->gb, 6);
677 /* compute samples */
679 for(i=0;i<bound;i++) {
680 for(ch=0;ch<s->nb_channels;ch++) {
681 n = allocation[ch][i];
683 mant = get_bits(&s->gb, n + 1);
684 v = l1_unscale(n, mant, scale_factors[ch][i]);
688 s->sb_samples[ch][j][i] = v;
691 for(i=bound;i<SBLIMIT;i++) {
692 n = allocation[0][i];
694 mant = get_bits(&s->gb, n + 1);
695 v = l1_unscale(n, mant, scale_factors[0][i]);
696 s->sb_samples[0][j][i] = v;
697 v = l1_unscale(n, mant, scale_factors[1][i]);
698 s->sb_samples[1][j][i] = v;
700 s->sb_samples[0][j][i] = 0;
701 s->sb_samples[1][j][i] = 0;
708 static int mp_decode_layer2(MPADecodeContext *s)
710 int sblimit; /* number of used subbands */
711 const unsigned char *alloc_table;
712 int table, bit_alloc_bits, i, j, ch, bound, v;
713 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
714 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
715 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
716 int scale, qindex, bits, steps, k, l, m, b;
718 /* select decoding table */
719 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
720 s->sample_rate, s->lsf);
721 sblimit = ff_mpa_sblimit_table[table];
722 alloc_table = ff_mpa_alloc_tables[table];
724 if (s->mode == MPA_JSTEREO)
725 bound = (s->mode_ext + 1) * 4;
729 av_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
732 if( bound > sblimit ) bound = sblimit;
734 /* parse bit allocation */
736 for(i=0;i<bound;i++) {
737 bit_alloc_bits = alloc_table[j];
738 for(ch=0;ch<s->nb_channels;ch++) {
739 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
741 j += 1 << bit_alloc_bits;
743 for(i=bound;i<sblimit;i++) {
744 bit_alloc_bits = alloc_table[j];
745 v = get_bits(&s->gb, bit_alloc_bits);
748 j += 1 << bit_alloc_bits;
752 for(i=0;i<sblimit;i++) {
753 for(ch=0;ch<s->nb_channels;ch++) {
754 if (bit_alloc[ch][i])
755 scale_code[ch][i] = get_bits(&s->gb, 2);
760 for(i=0;i<sblimit;i++) {
761 for(ch=0;ch<s->nb_channels;ch++) {
762 if (bit_alloc[ch][i]) {
763 sf = scale_factors[ch][i];
764 switch(scale_code[ch][i]) {
767 sf[0] = get_bits(&s->gb, 6);
768 sf[1] = get_bits(&s->gb, 6);
769 sf[2] = get_bits(&s->gb, 6);
772 sf[0] = get_bits(&s->gb, 6);
777 sf[0] = get_bits(&s->gb, 6);
778 sf[2] = get_bits(&s->gb, 6);
782 sf[0] = get_bits(&s->gb, 6);
783 sf[2] = get_bits(&s->gb, 6);
795 for(i=0;i<bound;i++) {
796 bit_alloc_bits = alloc_table[j];
797 for(ch=0;ch<s->nb_channels;ch++) {
798 b = bit_alloc[ch][i];
800 scale = scale_factors[ch][i][k];
801 qindex = alloc_table[j+b];
802 bits = ff_mpa_quant_bits[qindex];
805 /* 3 values at the same time */
806 v = get_bits(&s->gb, -bits);
807 v2 = division_tabs[qindex][v];
808 steps = ff_mpa_quant_steps[qindex];
810 s->sb_samples[ch][k * 12 + l + 0][i] =
811 l2_unscale_group(steps, v2 & 15, scale);
812 s->sb_samples[ch][k * 12 + l + 1][i] =
813 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
814 s->sb_samples[ch][k * 12 + l + 2][i] =
815 l2_unscale_group(steps, v2 >> 8 , scale);
818 v = get_bits(&s->gb, bits);
819 v = l1_unscale(bits - 1, v, scale);
820 s->sb_samples[ch][k * 12 + l + m][i] = v;
824 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
825 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
826 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
829 /* next subband in alloc table */
830 j += 1 << bit_alloc_bits;
832 /* XXX: find a way to avoid this duplication of code */
833 for(i=bound;i<sblimit;i++) {
834 bit_alloc_bits = alloc_table[j];
837 int mant, scale0, scale1;
838 scale0 = scale_factors[0][i][k];
839 scale1 = scale_factors[1][i][k];
840 qindex = alloc_table[j+b];
841 bits = ff_mpa_quant_bits[qindex];
843 /* 3 values at the same time */
844 v = get_bits(&s->gb, -bits);
845 steps = ff_mpa_quant_steps[qindex];
848 s->sb_samples[0][k * 12 + l + 0][i] =
849 l2_unscale_group(steps, mant, scale0);
850 s->sb_samples[1][k * 12 + l + 0][i] =
851 l2_unscale_group(steps, mant, scale1);
854 s->sb_samples[0][k * 12 + l + 1][i] =
855 l2_unscale_group(steps, mant, scale0);
856 s->sb_samples[1][k * 12 + l + 1][i] =
857 l2_unscale_group(steps, mant, scale1);
858 s->sb_samples[0][k * 12 + l + 2][i] =
859 l2_unscale_group(steps, v, scale0);
860 s->sb_samples[1][k * 12 + l + 2][i] =
861 l2_unscale_group(steps, v, scale1);
864 mant = get_bits(&s->gb, bits);
865 s->sb_samples[0][k * 12 + l + m][i] =
866 l1_unscale(bits - 1, mant, scale0);
867 s->sb_samples[1][k * 12 + l + m][i] =
868 l1_unscale(bits - 1, mant, scale1);
872 s->sb_samples[0][k * 12 + l + 0][i] = 0;
873 s->sb_samples[0][k * 12 + l + 1][i] = 0;
874 s->sb_samples[0][k * 12 + l + 2][i] = 0;
875 s->sb_samples[1][k * 12 + l + 0][i] = 0;
876 s->sb_samples[1][k * 12 + l + 1][i] = 0;
877 s->sb_samples[1][k * 12 + l + 2][i] = 0;
879 /* next subband in alloc table */
880 j += 1 << bit_alloc_bits;
882 /* fill remaining samples to zero */
883 for(i=sblimit;i<SBLIMIT;i++) {
884 for(ch=0;ch<s->nb_channels;ch++) {
885 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
886 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
887 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
895 #define SPLIT(dst,sf,n)\
904 int m= (sf*205)>>10;\
908 int m= (sf*171)>>10;\
915 static av_always_inline void lsf_sf_expand(int *slen,
916 int sf, int n1, int n2, int n3)
918 SPLIT(slen[3], sf, n3)
919 SPLIT(slen[2], sf, n2)
920 SPLIT(slen[1], sf, n1)
924 static void exponents_from_scale_factors(MPADecodeContext *s,
928 const uint8_t *bstab, *pretab;
929 int len, i, j, k, l, v0, shift, gain, gains[3];
933 gain = g->global_gain - 210;
934 shift = g->scalefac_scale + 1;
936 bstab = band_size_long[s->sample_rate_index];
937 pretab = mpa_pretab[g->preflag];
938 for(i=0;i<g->long_end;i++) {
939 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
945 if (g->short_start < 13) {
946 bstab = band_size_short[s->sample_rate_index];
947 gains[0] = gain - (g->subblock_gain[0] << 3);
948 gains[1] = gain - (g->subblock_gain[1] << 3);
949 gains[2] = gain - (g->subblock_gain[2] << 3);
951 for(i=g->short_start;i<13;i++) {
954 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
962 /* handle n = 0 too */
963 static inline int get_bitsz(GetBitContext *s, int n)
968 return get_bits(s, n);
972 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
973 if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
975 s->in_gb.buffer=NULL;
976 assert((get_bits_count(&s->gb) & 7) == 0);
977 skip_bits_long(&s->gb, *pos - *end_pos);
979 *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
980 *pos= get_bits_count(&s->gb);
984 /* Following is a optimized code for
986 if(get_bits1(&s->gb))
991 #define READ_FLIP_SIGN(dst,src)\
992 v = AV_RN32A(src) ^ (get_bits1(&s->gb)<<31);\
995 #define READ_FLIP_SIGN(dst,src)\
996 v= -get_bits1(&s->gb);\
997 *(dst) = (*(src) ^ v) - v;
1000 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1001 int16_t *exponents, int end_pos2)
1005 int last_pos, bits_left;
1007 int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1009 /* low frequencies (called big values) */
1012 int j, k, l, linbits;
1013 j = g->region_size[i];
1016 /* select vlc table */
1017 k = g->table_select[i];
1018 l = mpa_huff_data[k][0];
1019 linbits = mpa_huff_data[k][1];
1023 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1028 /* read huffcode and compute each couple */
1032 int pos= get_bits_count(&s->gb);
1034 if (pos >= end_pos){
1035 // av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1036 switch_buffer(s, &pos, &end_pos, &end_pos2);
1037 // av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1041 y = get_vlc2(&s->gb, vlc->table, 7, 3);
1044 g->sb_hybrid[s_index ] =
1045 g->sb_hybrid[s_index+1] = 0;
1050 exponent= exponents[s_index];
1052 av_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
1053 i, g->region_size[i] - j, x, y, exponent);
1058 READ_FLIP_SIGN(g->sb_hybrid+s_index, RENAME(expval_table)[ exponent ]+x)
1060 x += get_bitsz(&s->gb, linbits);
1061 v = l3_unscale(x, exponent);
1062 if (get_bits1(&s->gb))
1064 g->sb_hybrid[s_index] = v;
1067 READ_FLIP_SIGN(g->sb_hybrid+s_index+1, RENAME(expval_table)[ exponent ]+y)
1069 y += get_bitsz(&s->gb, linbits);
1070 v = l3_unscale(y, exponent);
1071 if (get_bits1(&s->gb))
1073 g->sb_hybrid[s_index+1] = v;
1080 READ_FLIP_SIGN(g->sb_hybrid+s_index+!!y, RENAME(expval_table)[ exponent ]+x)
1082 x += get_bitsz(&s->gb, linbits);
1083 v = l3_unscale(x, exponent);
1084 if (get_bits1(&s->gb))
1086 g->sb_hybrid[s_index+!!y] = v;
1088 g->sb_hybrid[s_index+ !y] = 0;
1094 /* high frequencies */
1095 vlc = &huff_quad_vlc[g->count1table_select];
1097 while (s_index <= 572) {
1099 pos = get_bits_count(&s->gb);
1100 if (pos >= end_pos) {
1101 if (pos > end_pos2 && last_pos){
1102 /* some encoders generate an incorrect size for this
1103 part. We must go back into the data */
1105 skip_bits_long(&s->gb, last_pos - pos);
1106 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1107 if(s->error_recognition >= FF_ER_COMPLIANT)
1111 // av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1112 switch_buffer(s, &pos, &end_pos, &end_pos2);
1113 // av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1119 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1120 av_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
1121 g->sb_hybrid[s_index+0]=
1122 g->sb_hybrid[s_index+1]=
1123 g->sb_hybrid[s_index+2]=
1124 g->sb_hybrid[s_index+3]= 0;
1126 static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1128 int pos= s_index+idxtab[code];
1129 code ^= 8>>idxtab[code];
1130 READ_FLIP_SIGN(g->sb_hybrid+pos, RENAME(exp_table)+exponents[pos])
1134 /* skip extension bits */
1135 bits_left = end_pos2 - get_bits_count(&s->gb);
1136 //av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
1137 if (bits_left < 0 && s->error_recognition >= FF_ER_COMPLIANT) {
1138 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1140 }else if(bits_left > 0 && s->error_recognition >= FF_ER_AGGRESSIVE){
1141 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1144 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1145 skip_bits_long(&s->gb, bits_left);
1147 i= get_bits_count(&s->gb);
1148 switch_buffer(s, &i, &end_pos, &end_pos2);
1153 /* Reorder short blocks from bitstream order to interleaved order. It
1154 would be faster to do it in parsing, but the code would be far more
1156 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1159 INTFLOAT *ptr, *dst, *ptr1;
1162 if (g->block_type != 2)
1165 if (g->switch_point) {
1166 if (s->sample_rate_index != 8) {
1167 ptr = g->sb_hybrid + 36;
1169 ptr = g->sb_hybrid + 48;
1175 for(i=g->short_start;i<13;i++) {
1176 len = band_size_short[s->sample_rate_index][i];
1179 for(j=len;j>0;j--) {
1180 *dst++ = ptr[0*len];
1181 *dst++ = ptr[1*len];
1182 *dst++ = ptr[2*len];
1186 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1190 #define ISQRT2 FIXR(0.70710678118654752440)
1192 static void compute_stereo(MPADecodeContext *s,
1193 GranuleDef *g0, GranuleDef *g1)
1196 int sf_max, sf, len, non_zero_found;
1197 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1198 int non_zero_found_short[3];
1200 /* intensity stereo */
1201 if (s->mode_ext & MODE_EXT_I_STEREO) {
1206 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1210 tab0 = g0->sb_hybrid + 576;
1211 tab1 = g1->sb_hybrid + 576;
1213 non_zero_found_short[0] = 0;
1214 non_zero_found_short[1] = 0;
1215 non_zero_found_short[2] = 0;
1216 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1217 for(i = 12;i >= g1->short_start;i--) {
1218 /* for last band, use previous scale factor */
1221 len = band_size_short[s->sample_rate_index][i];
1225 if (!non_zero_found_short[l]) {
1226 /* test if non zero band. if so, stop doing i-stereo */
1227 for(j=0;j<len;j++) {
1229 non_zero_found_short[l] = 1;
1233 sf = g1->scale_factors[k + l];
1239 for(j=0;j<len;j++) {
1241 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1242 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1246 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1247 /* lower part of the spectrum : do ms stereo
1249 for(j=0;j<len;j++) {
1252 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1253 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1260 non_zero_found = non_zero_found_short[0] |
1261 non_zero_found_short[1] |
1262 non_zero_found_short[2];
1264 for(i = g1->long_end - 1;i >= 0;i--) {
1265 len = band_size_long[s->sample_rate_index][i];
1268 /* test if non zero band. if so, stop doing i-stereo */
1269 if (!non_zero_found) {
1270 for(j=0;j<len;j++) {
1276 /* for last band, use previous scale factor */
1277 k = (i == 21) ? 20 : i;
1278 sf = g1->scale_factors[k];
1283 for(j=0;j<len;j++) {
1285 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1286 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1290 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1291 /* lower part of the spectrum : do ms stereo
1293 for(j=0;j<len;j++) {
1296 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1297 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1302 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1303 /* ms stereo ONLY */
1304 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1306 tab0 = g0->sb_hybrid;
1307 tab1 = g1->sb_hybrid;
1308 for(i=0;i<576;i++) {
1311 tab0[i] = tmp0 + tmp1;
1312 tab1[i] = tmp0 - tmp1;
1318 #define AA(j) do { \
1319 float tmp0 = ptr[-1-j]; \
1320 float tmp1 = ptr[ j]; \
1321 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1322 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1325 #define AA(j) do { \
1326 int tmp0 = ptr[-1-j]; \
1327 int tmp1 = ptr[ j]; \
1328 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1329 ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa_table[j][2])); \
1330 ptr[ j] = 4*(tmp2 + MULH(tmp0, csa_table[j][3])); \
1334 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1339 /* we antialias only "long" bands */
1340 if (g->block_type == 2) {
1341 if (!g->switch_point)
1343 /* XXX: check this for 8000Hz case */
1349 ptr = g->sb_hybrid + 18;
1350 for(i = n;i > 0;i--) {
1364 static void compute_imdct(MPADecodeContext *s,
1366 INTFLOAT *sb_samples,
1369 INTFLOAT *win, *win1, *out_ptr, *ptr, *buf, *ptr1;
1371 int i, j, mdct_long_end, sblimit;
1373 /* find last non zero block */
1374 ptr = g->sb_hybrid + 576;
1375 ptr1 = g->sb_hybrid + 2 * 18;
1376 while (ptr >= ptr1) {
1380 if(p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1383 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1385 if (g->block_type == 2) {
1386 /* XXX: check for 8000 Hz */
1387 if (g->switch_point)
1392 mdct_long_end = sblimit;
1397 for(j=0;j<mdct_long_end;j++) {
1398 /* apply window & overlap with previous buffer */
1399 out_ptr = sb_samples + j;
1401 if (g->switch_point && j < 2)
1404 win1 = mdct_win[g->block_type];
1405 /* select frequency inversion */
1406 win = win1 + ((4 * 36) & -(j & 1));
1407 imdct36(out_ptr, buf, ptr, win);
1408 out_ptr += 18*SBLIMIT;
1412 for(j=mdct_long_end;j<sblimit;j++) {
1413 /* select frequency inversion */
1414 win = mdct_win[2] + ((4 * 36) & -(j & 1));
1415 out_ptr = sb_samples + j;
1421 imdct12(out2, ptr + 0);
1423 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[i + 6*1];
1424 buf[i + 6*2] = MULH3(out2[i + 6], win[i + 6], 1);
1427 imdct12(out2, ptr + 1);
1429 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[i + 6*2];
1430 buf[i + 6*0] = MULH3(out2[i + 6], win[i + 6], 1);
1433 imdct12(out2, ptr + 2);
1435 buf[i + 6*0] = MULH3(out2[i ], win[i ], 1) + buf[i + 6*0];
1436 buf[i + 6*1] = MULH3(out2[i + 6], win[i + 6], 1);
1443 for(j=sblimit;j<SBLIMIT;j++) {
1445 out_ptr = sb_samples + j;
1455 /* main layer3 decoding function */
1456 static int mp_decode_layer3(MPADecodeContext *s)
1458 int nb_granules, main_data_begin;
1459 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1461 int16_t exponents[576]; //FIXME try INTFLOAT
1463 /* read side info */
1465 main_data_begin = get_bits(&s->gb, 8);
1466 skip_bits(&s->gb, s->nb_channels);
1469 main_data_begin = get_bits(&s->gb, 9);
1470 if (s->nb_channels == 2)
1471 skip_bits(&s->gb, 3);
1473 skip_bits(&s->gb, 5);
1475 for(ch=0;ch<s->nb_channels;ch++) {
1476 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1477 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1481 for(gr=0;gr<nb_granules;gr++) {
1482 for(ch=0;ch<s->nb_channels;ch++) {
1483 av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1484 g = &s->granules[ch][gr];
1485 g->part2_3_length = get_bits(&s->gb, 12);
1486 g->big_values = get_bits(&s->gb, 9);
1487 if(g->big_values > 288){
1488 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1492 g->global_gain = get_bits(&s->gb, 8);
1493 /* if MS stereo only is selected, we precompute the
1494 1/sqrt(2) renormalization factor */
1495 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1497 g->global_gain -= 2;
1499 g->scalefac_compress = get_bits(&s->gb, 9);
1501 g->scalefac_compress = get_bits(&s->gb, 4);
1502 blocksplit_flag = get_bits1(&s->gb);
1503 if (blocksplit_flag) {
1504 g->block_type = get_bits(&s->gb, 2);
1505 if (g->block_type == 0){
1506 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1509 g->switch_point = get_bits1(&s->gb);
1511 g->table_select[i] = get_bits(&s->gb, 5);
1513 g->subblock_gain[i] = get_bits(&s->gb, 3);
1514 ff_init_short_region(s, g);
1516 int region_address1, region_address2;
1518 g->switch_point = 0;
1520 g->table_select[i] = get_bits(&s->gb, 5);
1521 /* compute huffman coded region sizes */
1522 region_address1 = get_bits(&s->gb, 4);
1523 region_address2 = get_bits(&s->gb, 3);
1524 av_dlog(s->avctx, "region1=%d region2=%d\n",
1525 region_address1, region_address2);
1526 ff_init_long_region(s, g, region_address1, region_address2);
1528 ff_region_offset2size(g);
1529 ff_compute_band_indexes(s, g);
1533 g->preflag = get_bits1(&s->gb);
1534 g->scalefac_scale = get_bits1(&s->gb);
1535 g->count1table_select = get_bits1(&s->gb);
1536 av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1537 g->block_type, g->switch_point);
1542 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1543 assert((get_bits_count(&s->gb) & 7) == 0);
1544 /* now we get bits from the main_data_begin offset */
1545 av_dlog(s->avctx, "seekback: %d\n", main_data_begin);
1546 //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
1548 memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
1550 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1551 skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
1554 for(gr=0;gr<nb_granules;gr++) {
1555 for(ch=0;ch<s->nb_channels;ch++) {
1556 g = &s->granules[ch][gr];
1557 if(get_bits_count(&s->gb)<0){
1558 av_log(s->avctx, AV_LOG_DEBUG, "mdb:%d, lastbuf:%d skipping granule %d\n",
1559 main_data_begin, s->last_buf_size, gr);
1560 skip_bits_long(&s->gb, g->part2_3_length);
1561 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1562 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
1563 skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
1565 s->in_gb.buffer=NULL;
1570 bits_pos = get_bits_count(&s->gb);
1574 int slen, slen1, slen2;
1576 /* MPEG1 scale factors */
1577 slen1 = slen_table[0][g->scalefac_compress];
1578 slen2 = slen_table[1][g->scalefac_compress];
1579 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1580 if (g->block_type == 2) {
1581 n = g->switch_point ? 17 : 18;
1585 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1588 g->scale_factors[j++] = 0;
1592 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1594 g->scale_factors[j++] = 0;
1597 g->scale_factors[j++] = 0;
1600 sc = s->granules[ch][0].scale_factors;
1603 n = (k == 0 ? 6 : 5);
1604 if ((g->scfsi & (0x8 >> k)) == 0) {
1605 slen = (k < 2) ? slen1 : slen2;
1608 g->scale_factors[j++] = get_bits(&s->gb, slen);
1611 g->scale_factors[j++] = 0;
1614 /* simply copy from last granule */
1616 g->scale_factors[j] = sc[j];
1621 g->scale_factors[j++] = 0;
1624 int tindex, tindex2, slen[4], sl, sf;
1626 /* LSF scale factors */
1627 if (g->block_type == 2) {
1628 tindex = g->switch_point ? 2 : 1;
1632 sf = g->scalefac_compress;
1633 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1634 /* intensity stereo case */
1637 lsf_sf_expand(slen, sf, 6, 6, 0);
1639 } else if (sf < 244) {
1640 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1643 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1649 lsf_sf_expand(slen, sf, 5, 4, 4);
1651 } else if (sf < 500) {
1652 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1655 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1663 n = lsf_nsf_table[tindex2][tindex][k];
1667 g->scale_factors[j++] = get_bits(&s->gb, sl);
1670 g->scale_factors[j++] = 0;
1673 /* XXX: should compute exact size */
1675 g->scale_factors[j] = 0;
1678 exponents_from_scale_factors(s, g, exponents);
1680 /* read Huffman coded residue */
1681 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1684 if (s->nb_channels == 2)
1685 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1687 for(ch=0;ch<s->nb_channels;ch++) {
1688 g = &s->granules[ch][gr];
1690 reorder_block(s, g);
1691 compute_antialias(s, g);
1692 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1695 if(get_bits_count(&s->gb)<0)
1696 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1697 return nb_granules * 18;
1700 static int mp_decode_frame(MPADecodeContext *s,
1701 OUT_INT *samples, const uint8_t *buf, int buf_size)
1703 int i, nb_frames, ch;
1704 OUT_INT *samples_ptr;
1706 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
1708 /* skip error protection field */
1709 if (s->error_protection)
1710 skip_bits(&s->gb, 16);
1712 av_dlog(s->avctx, "frame %d:\n", s->frame_count);
1715 s->avctx->frame_size = 384;
1716 nb_frames = mp_decode_layer1(s);
1719 s->avctx->frame_size = 1152;
1720 nb_frames = mp_decode_layer2(s);
1723 s->avctx->frame_size = s->lsf ? 576 : 1152;
1725 nb_frames = mp_decode_layer3(s);
1728 if(s->in_gb.buffer){
1729 align_get_bits(&s->gb);
1730 i= get_bits_left(&s->gb)>>3;
1731 if(i >= 0 && i <= BACKSTEP_SIZE){
1732 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1735 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1737 s->in_gb.buffer= NULL;
1740 align_get_bits(&s->gb);
1741 assert((get_bits_count(&s->gb) & 7) == 0);
1742 i= get_bits_left(&s->gb)>>3;
1744 if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
1746 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1747 i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1749 assert(i <= buf_size - HEADER_SIZE && i>= 0);
1750 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1751 s->last_buf_size += i;
1756 /* apply the synthesis filter */
1757 for(ch=0;ch<s->nb_channels;ch++) {
1758 samples_ptr = samples + ch;
1759 for(i=0;i<nb_frames;i++) {
1760 RENAME(ff_mpa_synth_filter)(
1762 s->synth_buf[ch], &(s->synth_buf_offset[ch]),
1763 RENAME(ff_mpa_synth_window), &s->dither_state,
1764 samples_ptr, s->nb_channels,
1765 s->sb_samples[ch][i]);
1766 samples_ptr += 32 * s->nb_channels;
1770 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1773 static int decode_frame(AVCodecContext * avctx,
1774 void *data, int *data_size,
1777 const uint8_t *buf = avpkt->data;
1778 int buf_size = avpkt->size;
1779 MPADecodeContext *s = avctx->priv_data;
1782 OUT_INT *out_samples = data;
1784 if(buf_size < HEADER_SIZE)
1787 header = AV_RB32(buf);
1788 if(ff_mpa_check_header(header) < 0){
1789 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1793 if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
1794 /* free format: prepare to compute frame size */
1798 /* update codec info */
1799 avctx->channels = s->nb_channels;
1800 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1801 if (!avctx->bit_rate)
1802 avctx->bit_rate = s->bit_rate;
1803 avctx->sub_id = s->layer;
1805 if(*data_size < 1152*avctx->channels*sizeof(OUT_INT))
1809 if(s->frame_size<=0 || s->frame_size > buf_size){
1810 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1812 }else if(s->frame_size < buf_size){
1813 av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
1814 buf_size= s->frame_size;
1817 out_size = mp_decode_frame(s, out_samples, buf, buf_size);
1819 *data_size = out_size;
1820 avctx->sample_rate = s->sample_rate;
1821 //FIXME maybe move the other codec info stuff from above here too
1823 av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
1828 static void flush(AVCodecContext *avctx){
1829 MPADecodeContext *s = avctx->priv_data;
1830 memset(s->synth_buf, 0, sizeof(s->synth_buf));
1831 s->last_buf_size= 0;
1834 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1835 static int decode_frame_adu(AVCodecContext * avctx,
1836 void *data, int *data_size,
1839 const uint8_t *buf = avpkt->data;
1840 int buf_size = avpkt->size;
1841 MPADecodeContext *s = avctx->priv_data;
1844 OUT_INT *out_samples = data;
1848 // Discard too short frames
1849 if (buf_size < HEADER_SIZE) {
1855 if (len > MPA_MAX_CODED_FRAME_SIZE)
1856 len = MPA_MAX_CODED_FRAME_SIZE;
1858 // Get header and restore sync word
1859 header = AV_RB32(buf) | 0xffe00000;
1861 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1866 avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1867 /* update codec info */
1868 avctx->sample_rate = s->sample_rate;
1869 avctx->channels = s->nb_channels;
1870 if (!avctx->bit_rate)
1871 avctx->bit_rate = s->bit_rate;
1872 avctx->sub_id = s->layer;
1874 s->frame_size = len;
1876 if (avctx->parse_only) {
1877 out_size = buf_size;
1879 out_size = mp_decode_frame(s, out_samples, buf, buf_size);
1882 *data_size = out_size;
1885 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1887 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1890 * Context for MP3On4 decoder
1892 typedef struct MP3On4DecodeContext {
1893 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1894 int syncword; ///< syncword patch
1895 const uint8_t *coff; ///< channels offsets in output buffer
1896 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1897 OUT_INT *decoded_buf; ///< output buffer for decoded samples
1898 } MP3On4DecodeContext;
1900 #include "mpeg4audio.h"
1902 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1903 static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5}; /* number of mp3 decoder instances */
1904 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1905 static const uint8_t chan_offset[8][5] = {
1910 {2,0,3}, // C FLR BS
1911 {2,0,3}, // C FLR BLRS
1912 {2,0,4,3}, // C FLR BLRS LFE
1913 {2,0,6,4,3}, // C FLR BLRS BLR LFE
1916 /* mp3on4 channel layouts */
1917 static const int16_t chan_layout[8] = {
1920 AV_CH_LAYOUT_STEREO,
1921 AV_CH_LAYOUT_SURROUND,
1922 AV_CH_LAYOUT_4POINT0,
1923 AV_CH_LAYOUT_5POINT0,
1924 AV_CH_LAYOUT_5POINT1,
1925 AV_CH_LAYOUT_7POINT1
1928 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1930 MP3On4DecodeContext *s = avctx->priv_data;
1933 for (i = 0; i < s->frames; i++)
1934 av_free(s->mp3decctx[i]);
1936 av_freep(&s->decoded_buf);
1942 static int decode_init_mp3on4(AVCodecContext * avctx)
1944 MP3On4DecodeContext *s = avctx->priv_data;
1945 MPEG4AudioConfig cfg;
1948 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
1949 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1953 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);
1954 if (!cfg.chan_config || cfg.chan_config > 7) {
1955 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1958 s->frames = mp3Frames[cfg.chan_config];
1959 s->coff = chan_offset[cfg.chan_config];
1960 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1961 avctx->channel_layout = chan_layout[cfg.chan_config];
1963 if (cfg.sample_rate < 16000)
1964 s->syncword = 0xffe00000;
1966 s->syncword = 0xfff00000;
1968 /* Init the first mp3 decoder in standard way, so that all tables get builded
1969 * We replace avctx->priv_data with the context of the first decoder so that
1970 * decode_init() does not have to be changed.
1971 * Other decoders will be initialized here copying data from the first context
1973 // Allocate zeroed memory for the first decoder context
1974 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1975 if (!s->mp3decctx[0])
1977 // Put decoder context in place to make init_decode() happy
1978 avctx->priv_data = s->mp3decctx[0];
1980 // Restore mp3on4 context pointer
1981 avctx->priv_data = s;
1982 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1984 /* Create a separate codec/context for each frame (first is already ok).
1985 * Each frame is 1 or 2 channels - up to 5 frames allowed
1987 for (i = 1; i < s->frames; i++) {
1988 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1989 if (!s->mp3decctx[i])
1991 s->mp3decctx[i]->adu_mode = 1;
1992 s->mp3decctx[i]->avctx = avctx;
1993 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1996 /* Allocate buffer for multi-channel output if needed */
1997 if (s->frames > 1) {
1998 s->decoded_buf = av_malloc(MPA_FRAME_SIZE * MPA_MAX_CHANNELS *
1999 sizeof(*s->decoded_buf));
2000 if (!s->decoded_buf)
2006 decode_close_mp3on4(avctx);
2007 return AVERROR(ENOMEM);
2011 static void flush_mp3on4(AVCodecContext *avctx)
2014 MP3On4DecodeContext *s = avctx->priv_data;
2016 for (i = 0; i < s->frames; i++) {
2017 MPADecodeContext *m = s->mp3decctx[i];
2018 memset(m->synth_buf, 0, sizeof(m->synth_buf));
2019 m->last_buf_size = 0;
2024 static int decode_frame_mp3on4(AVCodecContext * avctx,
2025 void *data, int *data_size,
2028 const uint8_t *buf = avpkt->data;
2029 int buf_size = avpkt->size;
2030 MP3On4DecodeContext *s = avctx->priv_data;
2031 MPADecodeContext *m;
2032 int fsize, len = buf_size, out_size = 0;
2034 OUT_INT *out_samples = data;
2035 OUT_INT *outptr, *bp;
2038 if (*data_size < MPA_FRAME_SIZE * avctx->channels * sizeof(OUT_INT)) {
2039 av_log(avctx, AV_LOG_ERROR, "output buffer is too small\n");
2040 return AVERROR(EINVAL);
2044 // Discard too short frames
2045 if (buf_size < HEADER_SIZE)
2048 // If only one decoder interleave is not needed
2049 outptr = s->frames == 1 ? out_samples : s->decoded_buf;
2051 avctx->bit_rate = 0;
2054 for (fr = 0; fr < s->frames; fr++) {
2055 fsize = AV_RB16(buf) >> 4;
2056 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
2057 m = s->mp3decctx[fr];
2060 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
2062 if (ff_mpa_check_header(header) < 0) // Bad header, discard block
2065 avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
2067 if (ch + m->nb_channels > avctx->channels) {
2068 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
2070 return AVERROR_INVALIDDATA;
2072 ch += m->nb_channels;
2074 out_size += mp_decode_frame(m, outptr, buf, fsize);
2079 n = m->avctx->frame_size*m->nb_channels;
2080 /* interleave output data */
2081 bp = out_samples + s->coff[fr];
2082 if(m->nb_channels == 1) {
2083 for(j = 0; j < n; j++) {
2084 *bp = s->decoded_buf[j];
2085 bp += avctx->channels;
2088 for(j = 0; j < n; j++) {
2089 bp[0] = s->decoded_buf[j++];
2090 bp[1] = s->decoded_buf[j];
2091 bp += avctx->channels;
2095 avctx->bit_rate += m->bit_rate;
2098 /* update codec info */
2099 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2101 *data_size = out_size;
2104 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */
2107 #if CONFIG_MP1_DECODER
2108 AVCodec ff_mp1_decoder = {
2110 .type = AVMEDIA_TYPE_AUDIO,
2112 .priv_data_size = sizeof(MPADecodeContext),
2113 .init = decode_init,
2114 .decode = decode_frame,
2115 .capabilities = CODEC_CAP_PARSE_ONLY,
2117 .long_name = NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
2120 #if CONFIG_MP2_DECODER
2121 AVCodec ff_mp2_decoder = {
2123 .type = AVMEDIA_TYPE_AUDIO,
2125 .priv_data_size = sizeof(MPADecodeContext),
2126 .init = decode_init,
2127 .decode = decode_frame,
2128 .capabilities = CODEC_CAP_PARSE_ONLY,
2130 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
2133 #if CONFIG_MP3_DECODER
2134 AVCodec ff_mp3_decoder = {
2136 .type = AVMEDIA_TYPE_AUDIO,
2138 .priv_data_size = sizeof(MPADecodeContext),
2139 .init = decode_init,
2140 .decode = decode_frame,
2141 .capabilities = CODEC_CAP_PARSE_ONLY,
2143 .long_name = NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),
2146 #if CONFIG_MP3ADU_DECODER
2147 AVCodec ff_mp3adu_decoder = {
2149 .type = AVMEDIA_TYPE_AUDIO,
2150 .id = CODEC_ID_MP3ADU,
2151 .priv_data_size = sizeof(MPADecodeContext),
2152 .init = decode_init,
2153 .decode = decode_frame_adu,
2154 .capabilities = CODEC_CAP_PARSE_ONLY,
2156 .long_name = NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),
2159 #if CONFIG_MP3ON4_DECODER
2160 AVCodec ff_mp3on4_decoder = {
2162 .type = AVMEDIA_TYPE_AUDIO,
2163 .id = CODEC_ID_MP3ON4,
2164 .priv_data_size = sizeof(MP3On4DecodeContext),
2165 .init = decode_init_mp3on4,
2166 .close = decode_close_mp3on4,
2167 .decode = decode_frame_mp3on4,
2168 .flush = flush_mp3on4,
2169 .long_name = NULL_IF_CONFIG_SMALL("MP3onMP4"),