3 * Copyright (c) 2001 Gerard Lantau.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 #include "mpegaudio.h"
26 * - in low precision mode, use more 16 bit multiplies in synth filter
27 * - test lsf / mpeg25 extensively.
30 /* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
32 //#define USE_HIGHPRECISION
34 #ifdef USE_HIGHPRECISION
35 #define FRAC_BITS 23 /* fractional bits for sb_samples and dct */
36 #define WFRAC_BITS 16 /* fractional bits for window */
38 #define FRAC_BITS 15 /* fractional bits for sb_samples and dct */
39 #define WFRAC_BITS 14 /* fractional bits for window */
42 #define FRAC_ONE (1 << FRAC_BITS)
44 #define MULL(a,b) (((INT64)(a) * (INT64)(b)) >> FRAC_BITS)
45 #define MUL64(a,b) ((INT64)(a) * (INT64)(b))
46 #define FIX(a) ((int)((a) * FRAC_ONE))
47 /* WARNING: only correct for posititive numbers */
48 #define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
49 #define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
52 typedef INT16 MPA_INT;
54 typedef INT32 MPA_INT;
60 #define BACKSTEP_SIZE 512
62 typedef struct MPADecodeContext {
63 UINT8 inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE]; /* input buffer */
65 UINT8 *inbuf_ptr, *inbuf;
67 int free_format_frame_size; /* frame size in case of free format
68 (zero if currently unknown) */
69 /* next header (used in free format parsing) */
70 UINT32 free_format_next_header;
74 int sample_rate_index; /* between 0 and 8 */
82 MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2];
83 int synth_buf_offset[MPA_MAX_CHANNELS];
84 INT32 sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT];
85 INT32 mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
91 /* layer 3 "granule" */
92 typedef struct GranuleDef {
97 int scalefac_compress;
101 int subblock_gain[3];
102 UINT8 scalefac_scale;
103 UINT8 count1table_select;
104 int region_size[3]; /* number of huffman codes in each region */
106 int short_start, long_end; /* long/short band indexes */
107 UINT8 scale_factors[40];
108 INT32 sb_hybrid[SBLIMIT * 18]; /* 576 samples */
111 #define MODE_EXT_MS_STEREO 2
112 #define MODE_EXT_I_STEREO 1
114 /* layer 3 huffman tables */
115 typedef struct HuffTable {
121 #include "mpegaudiodectab.h"
123 /* vlc structure for decoding layer 3 huffman tables */
124 static VLC huff_vlc[16];
125 static UINT8 *huff_code_table[16];
126 static VLC huff_quad_vlc[2];
127 /* computed from band_size_long */
128 static UINT16 band_index_long[9][23];
129 /* XXX: free when all decoders are closed */
130 #define TABLE_4_3_SIZE (8191 + 16)
131 static UINT8 *table_4_3_exp;
133 static UINT16 *table_4_3_value;
135 static UINT32 *table_4_3_value;
137 /* intensity stereo coef table */
138 static INT32 is_table[2][16];
139 static INT32 is_table_lsf[2][2][16];
140 static INT32 csa_table[8][2];
141 static INT32 mdct_win[8][36];
143 /* lower 2 bits: modulo 3, higher bits: shift */
144 static UINT16 scale_factor_modshift[64];
145 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
146 static INT32 scale_factor_mult[15][3];
147 /* mult table for layer 2 group quantization */
149 #define SCALE_GEN(v) \
150 { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
152 static INT32 scale_factor_mult2[3][3] = {
153 SCALE_GEN(1.0 / 3.0), /* 3 steps */
154 SCALE_GEN(1.0 / 5.0), /* 5 steps */
155 SCALE_GEN(1.0 / 9.0), /* 9 steps */
159 static UINT32 scale_factor_mult3[4] = {
161 FIXR(1.18920711500272106671),
162 FIXR(1.41421356237309504880),
163 FIXR(1.68179283050742908605),
166 static MPA_INT window[512];
168 /* layer 1 unscaling */
169 /* n = number of bits of the mantissa minus 1 */
170 static inline int l1_unscale(int n, int mant, int scale_factor)
175 shift = scale_factor_modshift[scale_factor];
178 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
180 return (int)((val + (1 << (shift - 1))) >> shift);
183 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
187 shift = scale_factor_modshift[scale_factor];
190 /* XXX: store the result directly */
191 val = (2 * (mant - (steps >> 1))) * scale_factor_mult2[steps >> 2][mod];
192 return (val + (1 << (shift - 1))) >> shift;
195 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
196 static inline int l3_unscale(int value, int exponent)
205 e = table_4_3_exp[value];
206 e += (exponent >> 2);
212 m = table_4_3_value[value];
214 m = (m * scale_factor_mult3[exponent & 3]);
215 m = (m + (1 << (e-1))) >> e;
218 m = MUL64(m, scale_factor_mult3[exponent & 3]);
219 m = (m + (UINT64_C(1) << (e-1))) >> e;
225 static int decode_init(AVCodecContext * avctx)
227 MPADecodeContext *s = avctx->priv_data;
232 /* scale factors table for layer 1/2 */
235 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
242 scale_factor_modshift[i] = mod | (shift << 2);
245 /* scale factor multiply for layer 1 */
249 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
250 scale_factor_mult[i][0] = MULL(FIXR(1.0), norm);
251 scale_factor_mult[i][1] = MULL(FIXR(0.7937005259), norm);
252 scale_factor_mult[i][2] = MULL(FIXR(0.6299605249), norm);
253 dprintf("%d: norm=%x s=%x %x %x\n",
255 scale_factor_mult[i][0],
256 scale_factor_mult[i][1],
257 scale_factor_mult[i][2]);
261 /* max = 18760, max sum over all 16 coefs : 44736 */
266 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
275 /* huffman decode tables */
276 huff_code_table[0] = NULL;
278 const HuffTable *h = &mpa_huff_tables[i];
285 init_vlc(&huff_vlc[i], 8, n,
286 h->bits, 1, 1, h->codes, 2, 2);
288 code_table = av_mallocz(n);
290 for(x=0;x<xsize;x++) {
292 code_table[j++] = (x << 4) | y;
294 huff_code_table[i] = code_table;
297 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
298 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1);
304 band_index_long[i][j] = k;
305 k += band_size_long[i][j];
307 band_index_long[i][22] = k;
310 /* compute n ^ (4/3) and store it in mantissa/exp format */
311 table_4_3_exp = av_mallocz(TABLE_4_3_SIZE *
312 sizeof(table_4_3_exp[0]));
315 table_4_3_value = av_mallocz(TABLE_4_3_SIZE *
316 sizeof(table_4_3_value[0]));
317 if (!table_4_3_value) {
322 for(i=1;i<TABLE_4_3_SIZE;i++) {
325 f = pow((double)i, 4.0 / 3.0);
329 if ((unsigned short)m != m)
332 /* normalized to FRAC_BITS */
333 table_4_3_value[i] = m;
334 table_4_3_exp[i] = e - 1;
342 f = tan((double)i * M_PI / 12.0);
343 v = FIXR(f / (1.0 + f));
348 is_table[1][6 - i] = v;
352 is_table[0][i] = is_table[1][i] = 0.0;
359 e = -(j + 1) * ((i + 1) >> 1);
360 f = pow(2.0, e / 4.0);
362 is_table_lsf[j][k ^ 1][i] = FIXR(f);
363 is_table_lsf[j][k][i] = FIXR(1.0);
364 dprintf("is_table_lsf %d %d: %x %x\n",
365 i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
372 cs = 1.0 / sqrt(1.0 + ci * ci);
374 csa_table[i][0] = FIX(cs);
375 csa_table[i][1] = FIX(ca);
378 /* compute mdct windows */
381 v = FIXR(sin(M_PI * (i + 0.5) / 36.0));
387 mdct_win[1][18 + i] = FIXR(1.0);
388 mdct_win[1][24 + i] = FIXR(sin(M_PI * ((i + 6) + 0.5) / 12.0));
389 mdct_win[1][30 + i] = FIXR(0.0);
391 mdct_win[3][i] = FIXR(0.0);
392 mdct_win[3][6 + i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));
393 mdct_win[3][12 + i] = FIXR(1.0);
397 mdct_win[2][i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));
399 /* NOTE: we do frequency inversion adter the MDCT by changing
400 the sign of the right window coefs */
403 mdct_win[j + 4][i] = mdct_win[j][i];
404 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
410 printf("win%d=\n", j);
412 printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);
420 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
421 s->inbuf_ptr = s->inbuf;
428 /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */;
432 #define COS0_0 FIXR(0.50060299823519630134)
433 #define COS0_1 FIXR(0.50547095989754365998)
434 #define COS0_2 FIXR(0.51544730992262454697)
435 #define COS0_3 FIXR(0.53104259108978417447)
436 #define COS0_4 FIXR(0.55310389603444452782)
437 #define COS0_5 FIXR(0.58293496820613387367)
438 #define COS0_6 FIXR(0.62250412303566481615)
439 #define COS0_7 FIXR(0.67480834145500574602)
440 #define COS0_8 FIXR(0.74453627100229844977)
441 #define COS0_9 FIXR(0.83934964541552703873)
442 #define COS0_10 FIXR(0.97256823786196069369)
443 #define COS0_11 FIXR(1.16943993343288495515)
444 #define COS0_12 FIXR(1.48416461631416627724)
445 #define COS0_13 FIXR(2.05778100995341155085)
446 #define COS0_14 FIXR(3.40760841846871878570)
447 #define COS0_15 FIXR(10.19000812354805681150)
449 #define COS1_0 FIXR(0.50241928618815570551)
450 #define COS1_1 FIXR(0.52249861493968888062)
451 #define COS1_2 FIXR(0.56694403481635770368)
452 #define COS1_3 FIXR(0.64682178335999012954)
453 #define COS1_4 FIXR(0.78815462345125022473)
454 #define COS1_5 FIXR(1.06067768599034747134)
455 #define COS1_6 FIXR(1.72244709823833392782)
456 #define COS1_7 FIXR(5.10114861868916385802)
458 #define COS2_0 FIXR(0.50979557910415916894)
459 #define COS2_1 FIXR(0.60134488693504528054)
460 #define COS2_2 FIXR(0.89997622313641570463)
461 #define COS2_3 FIXR(2.56291544774150617881)
463 #define COS3_0 FIXR(0.54119610014619698439)
464 #define COS3_1 FIXR(1.30656296487637652785)
466 #define COS4_0 FIXR(0.70710678118654752439)
468 /* butterfly operator */
471 tmp0 = tab[a] + tab[b];\
472 tmp1 = tab[a] - tab[b];\
474 tab[b] = MULL(tmp1, c);\
477 #define BF1(a, b, c, d)\
484 #define BF2(a, b, c, d)\
494 #define ADD(a, b) tab[a] += tab[b]
496 /* DCT32 without 1/sqrt(2) coef zero scaling. */
497 static void dct32(INT32 *out, INT32 *tab)
629 out[ 1] = tab[16] + tab[24];
630 out[17] = tab[17] + tab[25];
631 out[ 9] = tab[18] + tab[26];
632 out[25] = tab[19] + tab[27];
633 out[ 5] = tab[20] + tab[28];
634 out[21] = tab[21] + tab[29];
635 out[13] = tab[22] + tab[30];
636 out[29] = tab[23] + tab[31];
637 out[ 3] = tab[24] + tab[20];
638 out[19] = tab[25] + tab[21];
639 out[11] = tab[26] + tab[22];
640 out[27] = tab[27] + tab[23];
641 out[ 7] = tab[28] + tab[18];
642 out[23] = tab[29] + tab[19];
643 out[15] = tab[30] + tab[17];
647 #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
651 #define OUT_SAMPLE(sum)\
654 sum1 = (sum + (1 << (OUT_SHIFT - 1))) >> OUT_SHIFT;\
657 else if (sum1 > 32767)\
663 #define SUM8(off, op) \
665 sum op w[0 * 64 + off] * p[0 * 64];\
666 sum op w[1 * 64 + off] * p[1 * 64];\
667 sum op w[2 * 64 + off] * p[2 * 64];\
668 sum op w[3 * 64 + off] * p[3 * 64];\
669 sum op w[4 * 64 + off] * p[4 * 64];\
670 sum op w[5 * 64 + off] * p[5 * 64];\
671 sum op w[6 * 64 + off] * p[6 * 64];\
672 sum op w[7 * 64 + off] * p[7 * 64];\
677 #define OUT_SAMPLE(sum)\
680 sum1 = (int)((sum + (INT64_C(1) << (OUT_SHIFT - 1))) >> OUT_SHIFT);\
683 else if (sum1 > 32767)\
689 #define SUM8(off, op) \
691 sum op MUL64(w[0 * 64 + off], p[0 * 64]);\
692 sum op MUL64(w[1 * 64 + off], p[1 * 64]);\
693 sum op MUL64(w[2 * 64 + off], p[2 * 64]);\
694 sum op MUL64(w[3 * 64 + off], p[3 * 64]);\
695 sum op MUL64(w[4 * 64 + off], p[4 * 64]);\
696 sum op MUL64(w[5 * 64 + off], p[5 * 64]);\
697 sum op MUL64(w[6 * 64 + off], p[6 * 64]);\
698 sum op MUL64(w[7 * 64 + off], p[7 * 64]);\
703 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
705 /* XXX: optimize by avoiding ring buffer usage */
706 static void synth_filter(MPADecodeContext *s1,
707 int ch, INT16 *samples, int incr,
708 INT32 sb_samples[SBLIMIT])
711 register MPA_INT *synth_buf, *p;
720 dct32(tmp, sb_samples);
722 offset = s1->synth_buf_offset[ch];
723 synth_buf = s1->synth_buf[ch] + offset;
735 /* copy to avoid wrap */
736 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
741 p = synth_buf + 16 + j; /* 0-15 */
743 p = synth_buf + 48 - j; /* 32-47 */
749 p = synth_buf + 32; /* 48 */
757 p = synth_buf + 48 - j; /* 17-31 */
759 p = synth_buf + 16 + j; /* 49-63 */
764 offset = (offset - 32) & 511;
765 s1->synth_buf_offset[ch] = offset;
769 #define C1 FIXR(0.99144486137381041114)
770 #define C3 FIXR(0.92387953251128675612)
771 #define C5 FIXR(0.79335334029123516458)
772 #define C7 FIXR(0.60876142900872063941)
773 #define C9 FIXR(0.38268343236508977173)
774 #define C11 FIXR(0.13052619222005159154)
776 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
778 static void imdct12(int *out, int *in)
781 INT64 in1_3, in1_9, in4_3, in4_9;
783 in1_3 = MUL64(in[1], C3);
784 in1_9 = MUL64(in[1], C9);
785 in4_3 = MUL64(in[4], C3);
786 in4_9 = MUL64(in[4], C9);
788 tmp = FRAC_RND(MUL64(in[0], C7) - in1_3 - MUL64(in[2], C11) +
789 MUL64(in[3], C1) - in4_9 - MUL64(in[5], C5));
792 tmp = FRAC_RND(MUL64(in[0] - in[3], C9) - in1_3 +
793 MUL64(in[2] + in[5], C3) - in4_9);
796 tmp = FRAC_RND(MUL64(in[0], C11) - in1_9 + MUL64(in[2], C7) -
797 MUL64(in[3], C5) + in4_3 - MUL64(in[5], C1));
800 tmp = FRAC_RND(MUL64(-in[0], C5) + in1_9 + MUL64(in[2], C1) +
801 MUL64(in[3], C11) - in4_3 - MUL64(in[5], C7));
804 tmp = FRAC_RND(MUL64(-in[0] + in[3], C3) - in1_9 +
805 MUL64(in[2] + in[5], C9) + in4_3);
808 tmp = FRAC_RND(-MUL64(in[0], C1) - in1_3 - MUL64(in[2], C5) -
809 MUL64(in[3], C7) - in4_9 - MUL64(in[5], C11));
822 #define C1 FIXR(0.98480775301220805936)
823 #define C2 FIXR(0.93969262078590838405)
824 #define C3 FIXR(0.86602540378443864676)
825 #define C4 FIXR(0.76604444311897803520)
826 #define C5 FIXR(0.64278760968653932632)
828 #define C7 FIXR(0.34202014332566873304)
829 #define C8 FIXR(0.17364817766693034885)
831 /* 0.5 / cos(pi*(2*i+1)/36) */
832 static const int icos36[9] = {
833 FIXR(0.50190991877167369479),
834 FIXR(0.51763809020504152469),
835 FIXR(0.55168895948124587824),
836 FIXR(0.61038729438072803416),
837 FIXR(0.70710678118654752439),
838 FIXR(0.87172339781054900991),
839 FIXR(1.18310079157624925896),
840 FIXR(1.93185165257813657349),
841 FIXR(5.73685662283492756461),
844 static const int icos72[18] = {
845 /* 0.5 / cos(pi*(2*i+19)/72) */
846 FIXR(0.74009361646113053152),
847 FIXR(0.82133981585229078570),
848 FIXR(0.93057949835178895673),
849 FIXR(1.08284028510010010928),
850 FIXR(1.30656296487637652785),
851 FIXR(1.66275476171152078719),
852 FIXR(2.31011315767264929558),
853 FIXR(3.83064878777019433457),
854 FIXR(11.46279281302667383546),
856 /* 0.5 / cos(pi*(2*(i + 18) +19)/72) */
857 FIXR(-0.67817085245462840086),
858 FIXR(-0.63023620700513223342),
859 FIXR(-0.59284452371708034528),
860 FIXR(-0.56369097343317117734),
861 FIXR(-0.54119610014619698439),
862 FIXR(-0.52426456257040533932),
863 FIXR(-0.51213975715725461845),
864 FIXR(-0.50431448029007636036),
865 FIXR(-0.50047634258165998492),
868 /* using Lee like decomposition followed by hand coded 9 points DCT */
869 static void imdct36(int *out, int *in)
871 int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
872 int tmp[18], *tmp1, *in1;
884 in3_3 = MUL64(in1[2*3], C3);
885 in6_6 = MUL64(in1[2*6], C6);
887 tmp1[0] = FRAC_RND(MUL64(in1[2*1], C1) + in3_3 +
888 MUL64(in1[2*5], C5) + MUL64(in1[2*7], C7));
889 tmp1[2] = in1[2*0] + FRAC_RND(MUL64(in1[2*2], C2) +
890 MUL64(in1[2*4], C4) + in6_6 +
891 MUL64(in1[2*8], C8));
892 tmp1[4] = FRAC_RND(MUL64(in1[2*1] - in1[2*5] - in1[2*7], C3));
893 tmp1[6] = FRAC_RND(MUL64(in1[2*2] - in1[2*4] - in1[2*8], C6)) -
895 tmp1[8] = FRAC_RND(MUL64(in1[2*1], C5) - in3_3 -
896 MUL64(in1[2*5], C7) + MUL64(in1[2*7], C1));
897 tmp1[10] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C8) -
898 MUL64(in1[2*4], C2) + in6_6 +
899 MUL64(in1[2*8], C4));
900 tmp1[12] = FRAC_RND(MUL64(in1[2*1], C7) - in3_3 +
901 MUL64(in1[2*5], C1) -
902 MUL64(in1[2*7], C5));
903 tmp1[14] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C4) +
904 MUL64(in1[2*4], C8) + in6_6 -
905 MUL64(in1[2*8], C2));
906 tmp1[16] = in1[2*0] - in1[2*2] + in1[2*4] - in1[2*6] + in1[2*8];
918 s1 = MULL(t3 + t2, icos36[j]);
919 s3 = MULL(t3 - t2, icos36[8 - j]);
921 t0 = MULL(s0 + s1, icos72[9 + 8 - j]);
922 t1 = MULL(s0 - s1, icos72[8 - j]);
923 out[18 + 9 + j] = t0;
924 out[18 + 8 - j] = t0;
928 t0 = MULL(s2 + s3, icos72[9+j]);
929 t1 = MULL(s2 - s3, icos72[j]);
930 out[18 + 9 + (8 - j)] = t0;
932 out[9 + (8 - j)] = -t1;
938 s1 = MULL(tmp[17], icos36[4]);
939 t0 = MULL(s0 + s1, icos72[9 + 4]);
940 t1 = MULL(s0 - s1, icos72[4]);
941 out[18 + 9 + 4] = t0;
942 out[18 + 8 - 4] = t0;
947 /* fast header check for resync */
948 static int check_header(UINT32 header)
951 if ((header & 0xffe00000) != 0xffe00000)
954 if (((header >> 17) & 3) == 0)
957 if (((header >> 12) & 0xf) == 0xf)
960 if (((header >> 10) & 3) == 3)
965 /* header + layer + bitrate + freq + lsf/mpeg25 */
966 #define SAME_HEADER_MASK \
967 (0xffe00000 | (3 << 17) | (0xf << 12) | (3 << 10) | (3 << 19))
969 /* header decoding. MUST check the header before because no
970 consistency check is done there. Return 1 if free format found and
971 that the frame size must be computed externally */
972 static int decode_header(MPADecodeContext *s, UINT32 header)
974 int sample_rate, frame_size, mpeg25, padding;
975 int sample_rate_index, bitrate_index;
976 if (header & (1<<20)) {
977 s->lsf = (header & (1<<19)) ? 0 : 1;
984 s->layer = 4 - ((header >> 17) & 3);
985 /* extract frequency */
986 sample_rate_index = (header >> 10) & 3;
987 sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
988 if (sample_rate == 0)
990 sample_rate_index += 3 * (s->lsf + mpeg25);
991 s->sample_rate_index = sample_rate_index;
992 s->error_protection = ((header >> 16) & 1) ^ 1;
994 bitrate_index = (header >> 12) & 0xf;
995 padding = (header >> 9) & 1;
996 //extension = (header >> 8) & 1;
997 s->mode = (header >> 6) & 3;
998 s->mode_ext = (header >> 4) & 3;
999 //copyright = (header >> 3) & 1;
1000 //original = (header >> 2) & 1;
1001 //emphasis = header & 3;
1003 if (s->mode == MPA_MONO)
1008 if (bitrate_index != 0) {
1009 frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1010 s->bit_rate = frame_size * 1000;
1013 frame_size = (frame_size * 12000) / sample_rate;
1014 frame_size = (frame_size + padding) * 4;
1017 frame_size = (frame_size * 144000) / sample_rate;
1018 frame_size += padding;
1022 frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1023 frame_size += padding;
1026 s->frame_size = frame_size;
1028 /* if no frame size computed, signal it */
1029 if (!s->free_format_frame_size)
1031 /* free format: compute bitrate and real frame size from the
1032 frame size we extracted by reading the bitstream */
1033 s->frame_size = s->free_format_frame_size;
1036 s->frame_size += padding * 4;
1037 s->bit_rate = (s->frame_size * sample_rate) / 48000;
1040 s->frame_size += padding;
1041 s->bit_rate = (s->frame_size * sample_rate) / 144000;
1045 s->frame_size += padding;
1046 s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1050 s->sample_rate = sample_rate;
1053 printf("layer%d, %d Hz, %d kbits/s, ",
1054 s->layer, s->sample_rate, s->bit_rate);
1055 if (s->nb_channels == 2) {
1056 if (s->layer == 3) {
1057 if (s->mode_ext & MODE_EXT_MS_STEREO)
1059 if (s->mode_ext & MODE_EXT_I_STEREO)
1071 /* return the number of decoded frames */
1072 static int mp_decode_layer1(MPADecodeContext *s)
1074 int bound, i, v, n, ch, j, mant;
1075 UINT8 allocation[MPA_MAX_CHANNELS][SBLIMIT];
1076 UINT8 scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1078 if (s->mode == MPA_JSTEREO)
1079 bound = (s->mode_ext + 1) * 4;
1083 /* allocation bits */
1084 for(i=0;i<bound;i++) {
1085 for(ch=0;ch<s->nb_channels;ch++) {
1086 allocation[ch][i] = get_bits(&s->gb, 4);
1089 for(i=bound;i<SBLIMIT;i++) {
1090 allocation[0][i] = get_bits(&s->gb, 4);
1094 for(i=0;i<bound;i++) {
1095 for(ch=0;ch<s->nb_channels;ch++) {
1096 if (allocation[ch][i])
1097 scale_factors[ch][i] = get_bits(&s->gb, 6);
1100 for(i=bound;i<SBLIMIT;i++) {
1101 if (allocation[0][i]) {
1102 scale_factors[0][i] = get_bits(&s->gb, 6);
1103 scale_factors[1][i] = get_bits(&s->gb, 6);
1107 /* compute samples */
1109 for(i=0;i<bound;i++) {
1110 for(ch=0;ch<s->nb_channels;ch++) {
1111 n = allocation[ch][i];
1113 mant = get_bits(&s->gb, n + 1);
1114 v = l1_unscale(n, mant, scale_factors[ch][i]);
1118 s->sb_samples[ch][j][i] = v;
1121 for(i=bound;i<SBLIMIT;i++) {
1122 n = allocation[0][i];
1124 mant = get_bits(&s->gb, n + 1);
1125 v = l1_unscale(n, mant, scale_factors[0][i]);
1126 s->sb_samples[0][j][i] = v;
1127 v = l1_unscale(n, mant, scale_factors[1][i]);
1128 s->sb_samples[1][j][i] = v;
1130 s->sb_samples[0][j][i] = 0;
1131 s->sb_samples[1][j][i] = 0;
1138 /* bitrate is in kb/s */
1139 int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1141 int ch_bitrate, table;
1143 ch_bitrate = bitrate / nb_channels;
1145 if ((freq == 48000 && ch_bitrate >= 56) ||
1146 (ch_bitrate >= 56 && ch_bitrate <= 80))
1148 else if (freq != 48000 && ch_bitrate >= 96)
1150 else if (freq != 32000 && ch_bitrate <= 48)
1160 static int mp_decode_layer2(MPADecodeContext *s)
1162 int sblimit; /* number of used subbands */
1163 const unsigned char *alloc_table;
1164 int table, bit_alloc_bits, i, j, ch, bound, v;
1165 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1166 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1167 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1168 int scale, qindex, bits, steps, k, l, m, b;
1170 /* select decoding table */
1171 table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1172 s->sample_rate, s->lsf);
1173 sblimit = sblimit_table[table];
1174 alloc_table = alloc_tables[table];
1176 if (s->mode == MPA_JSTEREO)
1177 bound = (s->mode_ext + 1) * 4;
1181 dprintf("bound=%d sblimit=%d\n", bound, sblimit);
1182 /* parse bit allocation */
1184 for(i=0;i<bound;i++) {
1185 bit_alloc_bits = alloc_table[j];
1186 for(ch=0;ch<s->nb_channels;ch++) {
1187 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1189 j += 1 << bit_alloc_bits;
1191 for(i=bound;i<sblimit;i++) {
1192 bit_alloc_bits = alloc_table[j];
1193 v = get_bits(&s->gb, bit_alloc_bits);
1194 bit_alloc[0][i] = v;
1195 bit_alloc[1][i] = v;
1196 j += 1 << bit_alloc_bits;
1201 for(ch=0;ch<s->nb_channels;ch++) {
1202 for(i=0;i<sblimit;i++)
1203 printf(" %d", bit_alloc[ch][i]);
1210 for(i=0;i<sblimit;i++) {
1211 for(ch=0;ch<s->nb_channels;ch++) {
1212 if (bit_alloc[ch][i])
1213 scale_code[ch][i] = get_bits(&s->gb, 2);
1218 for(i=0;i<sblimit;i++) {
1219 for(ch=0;ch<s->nb_channels;ch++) {
1220 if (bit_alloc[ch][i]) {
1221 sf = scale_factors[ch][i];
1222 switch(scale_code[ch][i]) {
1225 sf[0] = get_bits(&s->gb, 6);
1226 sf[1] = get_bits(&s->gb, 6);
1227 sf[2] = get_bits(&s->gb, 6);
1230 sf[0] = get_bits(&s->gb, 6);
1235 sf[0] = get_bits(&s->gb, 6);
1236 sf[2] = get_bits(&s->gb, 6);
1240 sf[0] = get_bits(&s->gb, 6);
1241 sf[2] = get_bits(&s->gb, 6);
1250 for(ch=0;ch<s->nb_channels;ch++) {
1251 for(i=0;i<sblimit;i++) {
1252 if (bit_alloc[ch][i]) {
1253 sf = scale_factors[ch][i];
1254 printf(" %d %d %d", sf[0], sf[1], sf[2]);
1265 for(l=0;l<12;l+=3) {
1267 for(i=0;i<bound;i++) {
1268 bit_alloc_bits = alloc_table[j];
1269 for(ch=0;ch<s->nb_channels;ch++) {
1270 b = bit_alloc[ch][i];
1272 scale = scale_factors[ch][i][k];
1273 qindex = alloc_table[j+b];
1274 bits = quant_bits[qindex];
1276 /* 3 values at the same time */
1277 v = get_bits(&s->gb, -bits);
1278 steps = quant_steps[qindex];
1279 s->sb_samples[ch][k * 12 + l + 0][i] =
1280 l2_unscale_group(steps, v % steps, scale);
1282 s->sb_samples[ch][k * 12 + l + 1][i] =
1283 l2_unscale_group(steps, v % steps, scale);
1285 s->sb_samples[ch][k * 12 + l + 2][i] =
1286 l2_unscale_group(steps, v, scale);
1289 v = get_bits(&s->gb, bits);
1290 v = l1_unscale(bits - 1, v, scale);
1291 s->sb_samples[ch][k * 12 + l + m][i] = v;
1295 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1296 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1297 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1300 /* next subband in alloc table */
1301 j += 1 << bit_alloc_bits;
1303 /* XXX: find a way to avoid this duplication of code */
1304 for(i=bound;i<sblimit;i++) {
1305 bit_alloc_bits = alloc_table[j];
1306 b = bit_alloc[0][i];
1308 int mant, scale0, scale1;
1309 scale0 = scale_factors[0][i][k];
1310 scale1 = scale_factors[1][i][k];
1311 qindex = alloc_table[j+b];
1312 bits = quant_bits[qindex];
1314 /* 3 values at the same time */
1315 v = get_bits(&s->gb, -bits);
1316 steps = quant_steps[qindex];
1319 s->sb_samples[0][k * 12 + l + 0][i] =
1320 l2_unscale_group(steps, mant, scale0);
1321 s->sb_samples[1][k * 12 + l + 0][i] =
1322 l2_unscale_group(steps, mant, scale1);
1325 s->sb_samples[0][k * 12 + l + 1][i] =
1326 l2_unscale_group(steps, mant, scale0);
1327 s->sb_samples[1][k * 12 + l + 1][i] =
1328 l2_unscale_group(steps, mant, scale1);
1329 s->sb_samples[0][k * 12 + l + 2][i] =
1330 l2_unscale_group(steps, v, scale0);
1331 s->sb_samples[1][k * 12 + l + 2][i] =
1332 l2_unscale_group(steps, v, scale1);
1335 mant = get_bits(&s->gb, bits);
1336 s->sb_samples[0][k * 12 + l + m][i] =
1337 l1_unscale(bits - 1, mant, scale0);
1338 s->sb_samples[1][k * 12 + l + m][i] =
1339 l1_unscale(bits - 1, mant, scale1);
1343 s->sb_samples[0][k * 12 + l + 0][i] = 0;
1344 s->sb_samples[0][k * 12 + l + 1][i] = 0;
1345 s->sb_samples[0][k * 12 + l + 2][i] = 0;
1346 s->sb_samples[1][k * 12 + l + 0][i] = 0;
1347 s->sb_samples[1][k * 12 + l + 1][i] = 0;
1348 s->sb_samples[1][k * 12 + l + 2][i] = 0;
1350 /* next subband in alloc table */
1351 j += 1 << bit_alloc_bits;
1353 /* fill remaining samples to zero */
1354 for(i=sblimit;i<SBLIMIT;i++) {
1355 for(ch=0;ch<s->nb_channels;ch++) {
1356 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1357 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1358 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1367 * Seek back in the stream for backstep bytes (at most 511 bytes)
1369 static void seek_to_maindata(MPADecodeContext *s, long backstep)
1373 /* compute current position in stream */
1374 #ifdef ALT_BITSTREAM_READER
1375 ptr = s->gb.buffer + (s->gb.index>>3);
1377 ptr = s->gb.buf_ptr - (s->gb.bit_cnt >> 3);
1379 /* copy old data before current one */
1381 memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
1382 BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
1383 /* init get bits again */
1384 init_get_bits(&s->gb, ptr, s->frame_size + backstep);
1386 /* prepare next buffer */
1387 s->inbuf_index ^= 1;
1388 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1389 s->old_frame_size = s->frame_size;
1392 static inline void lsf_sf_expand(int *slen,
1393 int sf, int n1, int n2, int n3)
1412 static void exponents_from_scale_factors(MPADecodeContext *s,
1416 const UINT8 *bstab, *pretab;
1417 int len, i, j, k, l, v0, shift, gain, gains[3];
1420 exp_ptr = exponents;
1421 gain = g->global_gain - 210;
1422 shift = g->scalefac_scale + 1;
1424 bstab = band_size_long[s->sample_rate_index];
1425 pretab = mpa_pretab[g->preflag];
1426 for(i=0;i<g->long_end;i++) {
1427 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
1433 if (g->short_start < 13) {
1434 bstab = band_size_short[s->sample_rate_index];
1435 gains[0] = gain - (g->subblock_gain[0] << 3);
1436 gains[1] = gain - (g->subblock_gain[1] << 3);
1437 gains[2] = gain - (g->subblock_gain[2] << 3);
1439 for(i=g->short_start;i<13;i++) {
1442 v0 = gains[l] - (g->scale_factors[k++] << shift);
1450 /* handle n = 0 too */
1451 static inline int get_bitsz(GetBitContext *s, int n)
1456 return get_bits(s, n);
1459 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1460 INT16 *exponents, int end_pos)
1463 int linbits, code, x, y, l, v, i, j, k, pos;
1464 UINT8 *last_buf_ptr;
1465 UINT32 last_bit_buf;
1470 /* low frequencies (called big values) */
1473 j = g->region_size[i];
1476 /* select vlc table */
1477 k = g->table_select[i];
1478 l = mpa_huff_data[k][0];
1479 linbits = mpa_huff_data[k][1];
1481 code_table = huff_code_table[l];
1483 /* read huffcode and compute each couple */
1485 if (get_bits_count(&s->gb) >= end_pos)
1488 code = get_vlc(&s->gb, vlc);
1491 y = code_table[code];
1498 dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
1499 i, g->region_size[i] - j, x, y, exponents[s_index]);
1502 x += get_bitsz(&s->gb, linbits);
1503 v = l3_unscale(x, exponents[s_index]);
1504 if (get_bits1(&s->gb))
1509 g->sb_hybrid[s_index++] = v;
1512 y += get_bitsz(&s->gb, linbits);
1513 v = l3_unscale(y, exponents[s_index]);
1514 if (get_bits1(&s->gb))
1519 g->sb_hybrid[s_index++] = v;
1523 /* high frequencies */
1524 vlc = &huff_quad_vlc[g->count1table_select];
1525 last_buf_ptr = NULL;
1528 while (s_index <= 572) {
1529 pos = get_bits_count(&s->gb);
1530 if (pos >= end_pos) {
1531 if (pos > end_pos && last_buf_ptr != NULL) {
1532 /* some encoders generate an incorrect size for this
1533 part. We must go back into the data */
1535 #ifdef ALT_BITSTREAM_READER
1536 s->gb.buffer = last_buf_ptr;
1537 s->gb.index = last_bit_cnt;
1539 s->gb.buf_ptr = last_buf_ptr;
1540 s->gb.bit_buf = last_bit_buf;
1541 s->gb.bit_cnt = last_bit_cnt;
1546 #ifdef ALT_BITSTREAM_READER
1547 last_buf_ptr = s->gb.buffer;
1548 last_bit_cnt = s->gb.index;
1550 last_buf_ptr = s->gb.buf_ptr;
1551 last_bit_buf = s->gb.bit_buf;
1552 last_bit_cnt = s->gb.bit_cnt;
1555 code = get_vlc(&s->gb, vlc);
1556 dprintf("t=%d code=%d\n", g->count1table_select, code);
1560 if (code & (8 >> i)) {
1561 /* non zero value. Could use a hand coded function for
1563 v = l3_unscale(1, exponents[s_index]);
1564 if(get_bits1(&s->gb))
1569 g->sb_hybrid[s_index++] = v;
1572 while (s_index < 576)
1573 g->sb_hybrid[s_index++] = 0;
1577 /* Reorder short blocks from bitstream order to interleaved order. It
1578 would be faster to do it in parsing, but the code would be far more
1580 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1583 INT32 *ptr, *dst, *ptr1;
1586 if (g->block_type != 2)
1589 if (g->switch_point) {
1590 if (s->sample_rate_index != 8) {
1591 ptr = g->sb_hybrid + 36;
1593 ptr = g->sb_hybrid + 48;
1599 for(i=g->short_start;i<13;i++) {
1600 len = band_size_short[s->sample_rate_index][i];
1604 for(j=len;j>0;j--) {
1609 memcpy(ptr1, tmp, len * 3 * sizeof(INT32));
1613 #define ISQRT2 FIXR(0.70710678118654752440)
1615 static void compute_stereo(MPADecodeContext *s,
1616 GranuleDef *g0, GranuleDef *g1)
1620 int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1621 INT32 (*is_tab)[16];
1623 int non_zero_found_short[3];
1625 /* intensity stereo */
1626 if (s->mode_ext & MODE_EXT_I_STEREO) {
1631 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1635 tab0 = g0->sb_hybrid + 576;
1636 tab1 = g1->sb_hybrid + 576;
1638 non_zero_found_short[0] = 0;
1639 non_zero_found_short[1] = 0;
1640 non_zero_found_short[2] = 0;
1641 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1642 for(i = 12;i >= g1->short_start;i--) {
1643 /* for last band, use previous scale factor */
1646 len = band_size_short[s->sample_rate_index][i];
1650 if (!non_zero_found_short[l]) {
1651 /* test if non zero band. if so, stop doing i-stereo */
1652 for(j=0;j<len;j++) {
1654 non_zero_found_short[l] = 1;
1658 sf = g1->scale_factors[k + l];
1664 for(j=0;j<len;j++) {
1666 tab0[j] = MULL(tmp0, v1);
1667 tab1[j] = MULL(tmp0, v2);
1671 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1672 /* lower part of the spectrum : do ms stereo
1674 for(j=0;j<len;j++) {
1677 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1678 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1685 non_zero_found = non_zero_found_short[0] |
1686 non_zero_found_short[1] |
1687 non_zero_found_short[2];
1689 for(i = g1->long_end - 1;i >= 0;i--) {
1690 len = band_size_long[s->sample_rate_index][i];
1693 /* test if non zero band. if so, stop doing i-stereo */
1694 if (!non_zero_found) {
1695 for(j=0;j<len;j++) {
1701 /* for last band, use previous scale factor */
1702 k = (i == 21) ? 20 : i;
1703 sf = g1->scale_factors[k];
1708 for(j=0;j<len;j++) {
1710 tab0[j] = MULL(tmp0, v1);
1711 tab1[j] = MULL(tmp0, v2);
1715 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1716 /* lower part of the spectrum : do ms stereo
1718 for(j=0;j<len;j++) {
1721 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1722 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1727 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1728 /* ms stereo ONLY */
1729 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1731 tab0 = g0->sb_hybrid;
1732 tab1 = g1->sb_hybrid;
1733 for(i=0;i<576;i++) {
1736 tab0[i] = tmp0 + tmp1;
1737 tab1[i] = tmp0 - tmp1;
1742 static void compute_antialias(MPADecodeContext *s,
1745 INT32 *ptr, *p0, *p1, *csa;
1746 int n, tmp0, tmp1, i, j;
1748 /* we antialias only "long" bands */
1749 if (g->block_type == 2) {
1750 if (!g->switch_point)
1752 /* XXX: check this for 8000Hz case */
1758 ptr = g->sb_hybrid + 18;
1759 for(i = n;i > 0;i--) {
1762 csa = &csa_table[0][0];
1766 *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));
1767 *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));
1776 static void compute_imdct(MPADecodeContext *s,
1781 INT32 *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1;
1785 int i, j, k, mdct_long_end, v, sblimit;
1787 /* find last non zero block */
1788 ptr = g->sb_hybrid + 576;
1789 ptr1 = g->sb_hybrid + 2 * 18;
1790 while (ptr >= ptr1) {
1792 v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
1796 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1798 if (g->block_type == 2) {
1799 /* XXX: check for 8000 Hz */
1800 if (g->switch_point)
1805 mdct_long_end = sblimit;
1810 for(j=0;j<mdct_long_end;j++) {
1812 /* apply window & overlap with previous buffer */
1813 out_ptr = sb_samples + j;
1815 if (g->switch_point && j < 2)
1818 win1 = mdct_win[g->block_type];
1819 /* select frequency inversion */
1820 win = win1 + ((4 * 36) & -(j & 1));
1822 *out_ptr = MULL(out[i], win[i]) + buf[i];
1823 buf[i] = MULL(out[i + 18], win[i + 18]);
1829 for(j=mdct_long_end;j<sblimit;j++) {
1835 /* select frequency inversion */
1836 win = mdct_win[2] + ((4 * 36) & -(j & 1));
1839 /* reorder input for short mdct */
1846 /* apply 12 point window and do small overlap */
1848 buf2[i] = MULL(out2[i], win[i]) + buf2[i];
1849 buf2[i + 6] = MULL(out2[i + 6], win[i + 6]);
1854 out_ptr = sb_samples + j;
1856 *out_ptr = out[i] + buf[i];
1857 buf[i] = out[i + 18];
1864 for(j=sblimit;j<SBLIMIT;j++) {
1866 out_ptr = sb_samples + j;
1877 void sample_dump(int fnum, INT32 *tab, int n)
1879 static FILE *files[16], *f;
1884 sprintf(buf, "/tmp/out%d.pcm", fnum);
1885 f = fopen(buf, "w");
1894 printf("pos=%d\n", pos);
1896 printf(" %f", (double)tab[i] / 32768.0);
1903 fwrite(tab, 1, n * sizeof(INT32), f);
1908 /* main layer3 decoding function */
1909 static int mp_decode_layer3(MPADecodeContext *s)
1911 int nb_granules, main_data_begin, private_bits;
1912 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
1913 GranuleDef granules[2][2], *g;
1914 INT16 exponents[576];
1916 /* read side info */
1918 main_data_begin = get_bits(&s->gb, 8);
1919 if (s->nb_channels == 2)
1920 private_bits = get_bits(&s->gb, 2);
1922 private_bits = get_bits(&s->gb, 1);
1925 main_data_begin = get_bits(&s->gb, 9);
1926 if (s->nb_channels == 2)
1927 private_bits = get_bits(&s->gb, 3);
1929 private_bits = get_bits(&s->gb, 5);
1931 for(ch=0;ch<s->nb_channels;ch++) {
1932 granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
1933 granules[ch][1].scfsi = get_bits(&s->gb, 4);
1937 for(gr=0;gr<nb_granules;gr++) {
1938 for(ch=0;ch<s->nb_channels;ch++) {
1939 dprintf("gr=%d ch=%d: side_info\n", gr, ch);
1940 g = &granules[ch][gr];
1941 g->part2_3_length = get_bits(&s->gb, 12);
1942 g->big_values = get_bits(&s->gb, 9);
1943 g->global_gain = get_bits(&s->gb, 8);
1944 /* if MS stereo only is selected, we precompute the
1945 1/sqrt(2) renormalization factor */
1946 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1948 g->global_gain -= 2;
1950 g->scalefac_compress = get_bits(&s->gb, 9);
1952 g->scalefac_compress = get_bits(&s->gb, 4);
1953 blocksplit_flag = get_bits(&s->gb, 1);
1954 if (blocksplit_flag) {
1955 g->block_type = get_bits(&s->gb, 2);
1956 if (g->block_type == 0)
1958 g->switch_point = get_bits(&s->gb, 1);
1960 g->table_select[i] = get_bits(&s->gb, 5);
1962 g->subblock_gain[i] = get_bits(&s->gb, 3);
1963 /* compute huffman coded region sizes */
1964 if (g->block_type == 2)
1965 g->region_size[0] = (36 / 2);
1967 if (s->sample_rate_index <= 2)
1968 g->region_size[0] = (36 / 2);
1969 else if (s->sample_rate_index != 8)
1970 g->region_size[0] = (54 / 2);
1972 g->region_size[0] = (108 / 2);
1974 g->region_size[1] = (576 / 2);
1976 int region_address1, region_address2, l;
1978 g->switch_point = 0;
1980 g->table_select[i] = get_bits(&s->gb, 5);
1981 /* compute huffman coded region sizes */
1982 region_address1 = get_bits(&s->gb, 4);
1983 region_address2 = get_bits(&s->gb, 3);
1984 dprintf("region1=%d region2=%d\n",
1985 region_address1, region_address2);
1987 band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
1988 l = region_address1 + region_address2 + 2;
1989 /* should not overflow */
1993 band_index_long[s->sample_rate_index][l] >> 1;
1995 /* convert region offsets to region sizes and truncate
1996 size to big_values */
1997 g->region_size[2] = (576 / 2);
2000 k = g->region_size[i];
2001 if (k > g->big_values)
2003 g->region_size[i] = k - j;
2007 /* compute band indexes */
2008 if (g->block_type == 2) {
2009 if (g->switch_point) {
2010 /* if switched mode, we handle the 36 first samples as
2011 long blocks. For 8000Hz, we handle the 48 first
2012 exponents as long blocks (XXX: check this!) */
2013 if (s->sample_rate_index <= 2)
2015 else if (s->sample_rate_index != 8)
2018 g->long_end = 4; /* 8000 Hz */
2020 if (s->sample_rate_index != 8)
2029 g->short_start = 13;
2035 g->preflag = get_bits(&s->gb, 1);
2036 g->scalefac_scale = get_bits(&s->gb, 1);
2037 g->count1table_select = get_bits(&s->gb, 1);
2038 dprintf("block_type=%d switch_point=%d\n",
2039 g->block_type, g->switch_point);
2043 /* now we get bits from the main_data_begin offset */
2044 dprintf("seekback: %d\n", main_data_begin);
2045 seek_to_maindata(s, main_data_begin);
2047 for(gr=0;gr<nb_granules;gr++) {
2048 for(ch=0;ch<s->nb_channels;ch++) {
2049 g = &granules[ch][gr];
2051 bits_pos = get_bits_count(&s->gb);
2055 int slen, slen1, slen2;
2057 /* MPEG1 scale factors */
2058 slen1 = slen_table[0][g->scalefac_compress];
2059 slen2 = slen_table[1][g->scalefac_compress];
2060 dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2061 if (g->block_type == 2) {
2062 n = g->switch_point ? 17 : 18;
2065 g->scale_factors[j++] = get_bitsz(&s->gb, slen1);
2067 g->scale_factors[j++] = get_bitsz(&s->gb, slen2);
2069 g->scale_factors[j++] = 0;
2071 sc = granules[ch][0].scale_factors;
2074 n = (k == 0 ? 6 : 5);
2075 if ((g->scfsi & (0x8 >> k)) == 0) {
2076 slen = (k < 2) ? slen1 : slen2;
2078 g->scale_factors[j++] = get_bitsz(&s->gb, slen);
2080 /* simply copy from last granule */
2082 g->scale_factors[j] = sc[j];
2087 g->scale_factors[j++] = 0;
2091 printf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2094 printf(" %d", g->scale_factors[i]);
2099 int tindex, tindex2, slen[4], sl, sf;
2101 /* LSF scale factors */
2102 if (g->block_type == 2) {
2103 tindex = g->switch_point ? 2 : 1;
2107 sf = g->scalefac_compress;
2108 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2109 /* intensity stereo case */
2112 lsf_sf_expand(slen, sf, 6, 6, 0);
2114 } else if (sf < 244) {
2115 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2118 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2124 lsf_sf_expand(slen, sf, 5, 4, 4);
2126 } else if (sf < 500) {
2127 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2130 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2138 n = lsf_nsf_table[tindex2][tindex][k];
2141 g->scale_factors[j++] = get_bitsz(&s->gb, sl);
2143 /* XXX: should compute exact size */
2145 g->scale_factors[j] = 0;
2148 printf("gr=%d ch=%d scale_factors:\n",
2151 printf(" %d", g->scale_factors[i]);
2157 exponents_from_scale_factors(s, g, exponents);
2159 /* read Huffman coded residue */
2160 if (huffman_decode(s, g, exponents,
2161 bits_pos + g->part2_3_length) < 0)
2163 #if defined(DEBUG) && 0
2164 sample_dump(3, g->sb_hybrid, 576);
2167 /* skip extension bits */
2168 bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2169 if (bits_left < 0) {
2170 dprintf("bits_left=%d\n", bits_left);
2173 while (bits_left >= 16) {
2174 skip_bits(&s->gb, 16);
2178 skip_bits(&s->gb, bits_left);
2181 if (s->nb_channels == 2)
2182 compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2184 for(ch=0;ch<s->nb_channels;ch++) {
2185 g = &granules[ch][gr];
2187 reorder_block(s, g);
2189 sample_dump(0, g->sb_hybrid, 576);
2191 compute_antialias(s, g);
2193 sample_dump(1, g->sb_hybrid, 576);
2195 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2197 sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2201 return nb_granules * 18;
2204 static int mp_decode_frame(MPADecodeContext *s,
2207 int i, nb_frames, ch;
2210 init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
2211 s->inbuf_ptr - s->inbuf - HEADER_SIZE);
2213 /* skip error protection field */
2214 if (s->error_protection)
2215 get_bits(&s->gb, 16);
2217 dprintf("frame %d:\n", s->frame_count);
2220 nb_frames = mp_decode_layer1(s);
2223 nb_frames = mp_decode_layer2(s);
2227 nb_frames = mp_decode_layer3(s);
2231 for(i=0;i<nb_frames;i++) {
2232 for(ch=0;ch<s->nb_channels;ch++) {
2234 printf("%d-%d:", i, ch);
2235 for(j=0;j<SBLIMIT;j++)
2236 printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2241 /* apply the synthesis filter */
2242 for(ch=0;ch<s->nb_channels;ch++) {
2243 samples_ptr = samples + ch;
2244 for(i=0;i<nb_frames;i++) {
2245 synth_filter(s, ch, samples_ptr, s->nb_channels,
2246 s->sb_samples[ch][i]);
2247 samples_ptr += 32 * s->nb_channels;
2253 return nb_frames * 32 * sizeof(short) * s->nb_channels;
2256 static int decode_frame(AVCodecContext * avctx,
2257 void *data, int *data_size,
2258 UINT8 * buf, int buf_size)
2260 MPADecodeContext *s = avctx->priv_data;
2264 short *out_samples = data;
2268 while (buf_size > 0) {
2269 len = s->inbuf_ptr - s->inbuf;
2270 if (s->frame_size == 0) {
2271 /* special case for next header for first frame in free
2272 format case (XXX: find a simpler method) */
2273 if (s->free_format_next_header != 0) {
2274 s->inbuf[0] = s->free_format_next_header >> 24;
2275 s->inbuf[1] = s->free_format_next_header >> 16;
2276 s->inbuf[2] = s->free_format_next_header >> 8;
2277 s->inbuf[3] = s->free_format_next_header;
2278 s->inbuf_ptr = s->inbuf + 4;
2279 s->free_format_next_header = 0;
2282 /* no header seen : find one. We need at least HEADER_SIZE
2283 bytes to parse it */
2284 len = HEADER_SIZE - len;
2288 memcpy(s->inbuf_ptr, buf_ptr, len);
2291 s->inbuf_ptr += len;
2293 if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
2295 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2296 (s->inbuf[2] << 8) | s->inbuf[3];
2298 if (check_header(header) < 0) {
2299 /* no sync found : move by one byte (inefficient, but simple!) */
2300 memcpy(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2302 dprintf("skip %x\n", header);
2303 /* reset free format frame size to give a chance
2304 to get a new bitrate */
2305 s->free_format_frame_size = 0;
2307 if (decode_header(s, header) == 1) {
2308 /* free format: compute frame size */
2310 memcpy(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2313 /* update codec info */
2314 avctx->sample_rate = s->sample_rate;
2315 avctx->channels = s->nb_channels;
2316 avctx->bit_rate = s->bit_rate;
2320 } else if (s->frame_size == -1) {
2321 /* free format : find next sync to compute frame size */
2322 len = MPA_MAX_CODED_FRAME_SIZE - len;
2326 /* frame too long: resync */
2333 memcpy(s->inbuf_ptr, buf_ptr, len);
2334 /* check for header */
2335 p = s->inbuf_ptr - 3;
2336 pend = s->inbuf_ptr + len - 4;
2338 header = (p[0] << 24) | (p[1] << 16) |
2340 header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2341 (s->inbuf[2] << 8) | s->inbuf[3];
2342 /* check with high probability that we have a
2344 if ((header & SAME_HEADER_MASK) ==
2345 (header1 & SAME_HEADER_MASK)) {
2346 /* header found: update pointers */
2347 len = (p + 4) - s->inbuf_ptr;
2351 /* compute frame size */
2352 s->free_format_next_header = header;
2353 s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2354 padding = (header1 >> 9) & 1;
2356 s->free_format_frame_size -= padding * 4;
2358 s->free_format_frame_size -= padding;
2359 dprintf("free frame size=%d padding=%d\n",
2360 s->free_format_frame_size, padding);
2361 decode_header(s, header1);
2366 /* not found: simply increase pointers */
2368 s->inbuf_ptr += len;
2371 } else if (len < s->frame_size) {
2372 if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2373 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
2374 len = s->frame_size - len;
2378 len = buf_size > 4 ? 4 : buf_size;
2379 memcpy(s->inbuf_ptr, buf_ptr, len);
2381 s->inbuf_ptr += len;
2384 out_size = mp_decode_frame(s, out_samples);
2385 s->inbuf_ptr = s->inbuf;
2387 *data_size = out_size;
2392 return buf_ptr - buf;
2395 AVCodec mp3_decoder =
2400 sizeof(MPADecodeContext),