fixed layer1/2 overflow if very loud sound - fixed broken free format decoding to...
[libav.git] / libavcodec / mpegaudiodec.c
1 /*
2 * MPEG Audio decoder
3 * Copyright (c) 2001, 2002 Fabrice Bellard.
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library 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 GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19 //#define DEBUG
20 #include "avcodec.h"
21 #include "mpegaudio.h"
22
23 /*
24 * TODO:
25 * - in low precision mode, use more 16 bit multiplies in synth filter
26 * - test lsf / mpeg25 extensively.
27 */
28
29 /* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
30 audio decoder */
31 #ifdef CONFIG_MPEGAUDIO_HP
32 #define USE_HIGHPRECISION
33 #endif
34
35 #ifdef USE_HIGHPRECISION
36 #define FRAC_BITS 23 /* fractional bits for sb_samples and dct */
37 #define WFRAC_BITS 16 /* fractional bits for window */
38 #else
39 #define FRAC_BITS 15 /* fractional bits for sb_samples and dct */
40 #define WFRAC_BITS 14 /* fractional bits for window */
41 #endif
42
43 #define FRAC_ONE (1 << FRAC_BITS)
44
45 #define MULL(a,b) (((INT64)(a) * (INT64)(b)) >> FRAC_BITS)
46 #define MUL64(a,b) ((INT64)(a) * (INT64)(b))
47 #define FIX(a) ((int)((a) * FRAC_ONE))
48 /* WARNING: only correct for posititive numbers */
49 #define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
50 #define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
51
52 #if FRAC_BITS <= 15
53 typedef INT16 MPA_INT;
54 #else
55 typedef INT32 MPA_INT;
56 #endif
57
58 /****************/
59
60 #define HEADER_SIZE 4
61 #define BACKSTEP_SIZE 512
62
63 typedef struct MPADecodeContext {
64 UINT8 inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE]; /* input buffer */
65 int inbuf_index;
66 UINT8 *inbuf_ptr, *inbuf;
67 int frame_size;
68 int free_format_frame_size; /* frame size in case of free format
69 (zero if currently unknown) */
70 /* next header (used in free format parsing) */
71 UINT32 free_format_next_header;
72 int error_protection;
73 int layer;
74 int sample_rate;
75 int sample_rate_index; /* between 0 and 8 */
76 int bit_rate;
77 int old_frame_size;
78 GetBitContext gb;
79 int nb_channels;
80 int mode;
81 int mode_ext;
82 int lsf;
83 MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2];
84 int synth_buf_offset[MPA_MAX_CHANNELS];
85 INT32 sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT];
86 INT32 mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
87 #ifdef DEBUG
88 int frame_count;
89 #endif
90 } MPADecodeContext;
91
92 /* layer 3 "granule" */
93 typedef struct GranuleDef {
94 UINT8 scfsi;
95 int part2_3_length;
96 int big_values;
97 int global_gain;
98 int scalefac_compress;
99 UINT8 block_type;
100 UINT8 switch_point;
101 int table_select[3];
102 int subblock_gain[3];
103 UINT8 scalefac_scale;
104 UINT8 count1table_select;
105 int region_size[3]; /* number of huffman codes in each region */
106 int preflag;
107 int short_start, long_end; /* long/short band indexes */
108 UINT8 scale_factors[40];
109 INT32 sb_hybrid[SBLIMIT * 18]; /* 576 samples */
110 } GranuleDef;
111
112 #define MODE_EXT_MS_STEREO 2
113 #define MODE_EXT_I_STEREO 1
114
115 /* layer 3 huffman tables */
116 typedef struct HuffTable {
117 int xsize;
118 const UINT8 *bits;
119 const UINT16 *codes;
120 } HuffTable;
121
122 #include "mpegaudiodectab.h"
123
124 /* vlc structure for decoding layer 3 huffman tables */
125 static VLC huff_vlc[16];
126 static UINT8 *huff_code_table[16];
127 static VLC huff_quad_vlc[2];
128 /* computed from band_size_long */
129 static UINT16 band_index_long[9][23];
130 /* XXX: free when all decoders are closed */
131 #define TABLE_4_3_SIZE (8191 + 16)
132 static INT8 *table_4_3_exp;
133 #if FRAC_BITS <= 15
134 static UINT16 *table_4_3_value;
135 #else
136 static UINT32 *table_4_3_value;
137 #endif
138 /* intensity stereo coef table */
139 static INT32 is_table[2][16];
140 static INT32 is_table_lsf[2][2][16];
141 static INT32 csa_table[8][2];
142 static INT32 mdct_win[8][36];
143
144 /* lower 2 bits: modulo 3, higher bits: shift */
145 static UINT16 scale_factor_modshift[64];
146 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
147 static INT32 scale_factor_mult[15][3];
148 /* mult table for layer 2 group quantization */
149
150 #define SCALE_GEN(v) \
151 { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
152
153 static INT32 scale_factor_mult2[3][3] = {
154 SCALE_GEN(4.0 / 3.0), /* 3 steps */
155 SCALE_GEN(4.0 / 5.0), /* 5 steps */
156 SCALE_GEN(4.0 / 9.0), /* 9 steps */
157 };
158
159 /* 2^(n/4) */
160 static UINT32 scale_factor_mult3[4] = {
161 FIXR(1.0),
162 FIXR(1.18920711500272106671),
163 FIXR(1.41421356237309504880),
164 FIXR(1.68179283050742908605),
165 };
166
167 static MPA_INT window[512];
168
169 /* layer 1 unscaling */
170 /* n = number of bits of the mantissa minus 1 */
171 static inline int l1_unscale(int n, int mant, int scale_factor)
172 {
173 int shift, mod;
174 INT64 val;
175
176 shift = scale_factor_modshift[scale_factor];
177 mod = shift & 3;
178 shift >>= 2;
179 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
180 shift += n;
181 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
182 return (int)((val + (1LL << (shift - 1))) >> shift);
183 }
184
185 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
186 {
187 int shift, mod, val;
188
189 shift = scale_factor_modshift[scale_factor];
190 mod = shift & 3;
191 shift >>= 2;
192
193 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
194 /* NOTE: at this point, 0 <= shift <= 21 */
195 if (shift > 0)
196 val = (val + (1 << (shift - 1))) >> shift;
197 return val;
198 }
199
200 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
201 static inline int l3_unscale(int value, int exponent)
202 {
203 #if FRAC_BITS <= 15
204 unsigned int m;
205 #else
206 UINT64 m;
207 #endif
208 int e;
209
210 e = table_4_3_exp[value];
211 e += (exponent >> 2);
212 e = FRAC_BITS - e;
213 #if FRAC_BITS <= 15
214 if (e > 31)
215 e = 31;
216 #endif
217 m = table_4_3_value[value];
218 #if FRAC_BITS <= 15
219 m = (m * scale_factor_mult3[exponent & 3]);
220 m = (m + (1 << (e-1))) >> e;
221 return m;
222 #else
223 m = MUL64(m, scale_factor_mult3[exponent & 3]);
224 m = (m + (UINT64_C(1) << (e-1))) >> e;
225 return m;
226 #endif
227 }
228
229 /* all integer n^(4/3) computation code */
230 #define DEV_ORDER 13
231
232 #define POW_FRAC_BITS 24
233 #define POW_FRAC_ONE (1 << POW_FRAC_BITS)
234 #define POW_FIX(a) ((int)((a) * POW_FRAC_ONE))
235 #define POW_MULL(a,b) (((INT64)(a) * (INT64)(b)) >> POW_FRAC_BITS)
236
237 static int dev_4_3_coefs[DEV_ORDER];
238
239 static int pow_mult3[3] = {
240 POW_FIX(1.0),
241 POW_FIX(1.25992104989487316476),
242 POW_FIX(1.58740105196819947474),
243 };
244
245 static void int_pow_init(void)
246 {
247 int i, a;
248
249 a = POW_FIX(1.0);
250 for(i=0;i<DEV_ORDER;i++) {
251 a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
252 dev_4_3_coefs[i] = a;
253 }
254 }
255
256 /* return the mantissa and the binary exponent */
257 static int int_pow(int i, int *exp_ptr)
258 {
259 int e, er, eq, j;
260 int a, a1;
261
262 /* renormalize */
263 a = i;
264 e = POW_FRAC_BITS;
265 while (a < (1 << (POW_FRAC_BITS - 1))) {
266 a = a << 1;
267 e--;
268 }
269 a -= (1 << POW_FRAC_BITS);
270 a1 = 0;
271 for(j = DEV_ORDER - 1; j >= 0; j--)
272 a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
273 a = (1 << POW_FRAC_BITS) + a1;
274 /* exponent compute (exact) */
275 e = e * 4;
276 er = e % 3;
277 eq = e / 3;
278 a = POW_MULL(a, pow_mult3[er]);
279 while (a >= 2 * POW_FRAC_ONE) {
280 a = a >> 1;
281 eq++;
282 }
283 /* convert to float */
284 while (a < POW_FRAC_ONE) {
285 a = a << 1;
286 eq--;
287 }
288 /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
289 #if POW_FRAC_BITS > FRAC_BITS
290 a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
291 /* correct overflow */
292 if (a >= 2 * (1 << FRAC_BITS)) {
293 a = a >> 1;
294 eq++;
295 }
296 #endif
297 *exp_ptr = eq;
298 return a;
299 }
300
301 static int decode_init(AVCodecContext * avctx)
302 {
303 MPADecodeContext *s = avctx->priv_data;
304 static int init;
305 int i, j, k;
306
307 if(!init) {
308 /* scale factors table for layer 1/2 */
309 for(i=0;i<64;i++) {
310 int shift, mod;
311 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
312 shift = (i / 3);
313 mod = i % 3;
314 scale_factor_modshift[i] = mod | (shift << 2);
315 }
316
317 /* scale factor multiply for layer 1 */
318 for(i=0;i<15;i++) {
319 int n, norm;
320 n = i + 2;
321 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
322 scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
323 scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
324 scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
325 dprintf("%d: norm=%x s=%x %x %x\n",
326 i, norm,
327 scale_factor_mult[i][0],
328 scale_factor_mult[i][1],
329 scale_factor_mult[i][2]);
330 }
331
332 /* window */
333 /* max = 18760, max sum over all 16 coefs : 44736 */
334 for(i=0;i<257;i++) {
335 int v;
336 v = mpa_enwindow[i];
337 #if WFRAC_BITS < 16
338 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
339 #endif
340 window[i] = v;
341 if ((i & 63) != 0)
342 v = -v;
343 if (i != 0)
344 window[512 - i] = v;
345 }
346
347 /* huffman decode tables */
348 huff_code_table[0] = NULL;
349 for(i=1;i<16;i++) {
350 const HuffTable *h = &mpa_huff_tables[i];
351 int xsize, n, x, y;
352 UINT8 *code_table;
353
354 xsize = h->xsize;
355 n = xsize * xsize;
356 /* XXX: fail test */
357 init_vlc(&huff_vlc[i], 8, n,
358 h->bits, 1, 1, h->codes, 2, 2);
359
360 code_table = av_mallocz(n);
361 j = 0;
362 for(x=0;x<xsize;x++) {
363 for(y=0;y<xsize;y++)
364 code_table[j++] = (x << 4) | y;
365 }
366 huff_code_table[i] = code_table;
367 }
368 for(i=0;i<2;i++) {
369 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
370 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1);
371 }
372
373 for(i=0;i<9;i++) {
374 k = 0;
375 for(j=0;j<22;j++) {
376 band_index_long[i][j] = k;
377 k += band_size_long[i][j];
378 }
379 band_index_long[i][22] = k;
380 }
381
382 /* compute n ^ (4/3) and store it in mantissa/exp format */
383 table_4_3_exp = av_mallocz(TABLE_4_3_SIZE *
384 sizeof(table_4_3_exp[0]));
385 if (!table_4_3_exp)
386 return -1;
387 table_4_3_value = av_mallocz(TABLE_4_3_SIZE *
388 sizeof(table_4_3_value[0]));
389 if (!table_4_3_value) {
390 av_free(table_4_3_exp);
391 return -1;
392 }
393
394 int_pow_init();
395 for(i=1;i<TABLE_4_3_SIZE;i++) {
396 int e, m;
397 m = int_pow(i, &e);
398 #if 0
399 /* test code */
400 {
401 double f, fm;
402 int e1, m1;
403 f = pow((double)i, 4.0 / 3.0);
404 fm = frexp(f, &e1);
405 m1 = FIXR(2 * fm);
406 #if FRAC_BITS <= 15
407 if ((unsigned short)m1 != m1) {
408 m1 = m1 >> 1;
409 e1++;
410 }
411 #endif
412 e1--;
413 if (m != m1 || e != e1) {
414 printf("%4d: m=%x m1=%x e=%d e1=%d\n",
415 i, m, m1, e, e1);
416 }
417 }
418 #endif
419 /* normalized to FRAC_BITS */
420 table_4_3_value[i] = m;
421 table_4_3_exp[i] = e;
422 }
423
424 for(i=0;i<7;i++) {
425 float f;
426 int v;
427 if (i != 6) {
428 f = tan((double)i * M_PI / 12.0);
429 v = FIXR(f / (1.0 + f));
430 } else {
431 v = FIXR(1.0);
432 }
433 is_table[0][i] = v;
434 is_table[1][6 - i] = v;
435 }
436 /* invalid values */
437 for(i=7;i<16;i++)
438 is_table[0][i] = is_table[1][i] = 0.0;
439
440 for(i=0;i<16;i++) {
441 double f;
442 int e, k;
443
444 for(j=0;j<2;j++) {
445 e = -(j + 1) * ((i + 1) >> 1);
446 f = pow(2.0, e / 4.0);
447 k = i & 1;
448 is_table_lsf[j][k ^ 1][i] = FIXR(f);
449 is_table_lsf[j][k][i] = FIXR(1.0);
450 dprintf("is_table_lsf %d %d: %x %x\n",
451 i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
452 }
453 }
454
455 for(i=0;i<8;i++) {
456 float ci, cs, ca;
457 ci = ci_table[i];
458 cs = 1.0 / sqrt(1.0 + ci * ci);
459 ca = cs * ci;
460 csa_table[i][0] = FIX(cs);
461 csa_table[i][1] = FIX(ca);
462 }
463
464 /* compute mdct windows */
465 for(i=0;i<36;i++) {
466 int v;
467 v = FIXR(sin(M_PI * (i + 0.5) / 36.0));
468 mdct_win[0][i] = v;
469 mdct_win[1][i] = v;
470 mdct_win[3][i] = v;
471 }
472 for(i=0;i<6;i++) {
473 mdct_win[1][18 + i] = FIXR(1.0);
474 mdct_win[1][24 + i] = FIXR(sin(M_PI * ((i + 6) + 0.5) / 12.0));
475 mdct_win[1][30 + i] = FIXR(0.0);
476
477 mdct_win[3][i] = FIXR(0.0);
478 mdct_win[3][6 + i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));
479 mdct_win[3][12 + i] = FIXR(1.0);
480 }
481
482 for(i=0;i<12;i++)
483 mdct_win[2][i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));
484
485 /* NOTE: we do frequency inversion adter the MDCT by changing
486 the sign of the right window coefs */
487 for(j=0;j<4;j++) {
488 for(i=0;i<36;i+=2) {
489 mdct_win[j + 4][i] = mdct_win[j][i];
490 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
491 }
492 }
493
494 #if defined(DEBUG)
495 for(j=0;j<8;j++) {
496 printf("win%d=\n", j);
497 for(i=0;i<36;i++)
498 printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);
499 printf("\n");
500 }
501 #endif
502 init = 1;
503 }
504
505 s->inbuf_index = 0;
506 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
507 s->inbuf_ptr = s->inbuf;
508 #ifdef DEBUG
509 s->frame_count = 0;
510 #endif
511 return 0;
512 }
513
514 /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */;
515
516 /* cos(i*pi/64) */
517
518 #define COS0_0 FIXR(0.50060299823519630134)
519 #define COS0_1 FIXR(0.50547095989754365998)
520 #define COS0_2 FIXR(0.51544730992262454697)
521 #define COS0_3 FIXR(0.53104259108978417447)
522 #define COS0_4 FIXR(0.55310389603444452782)
523 #define COS0_5 FIXR(0.58293496820613387367)
524 #define COS0_6 FIXR(0.62250412303566481615)
525 #define COS0_7 FIXR(0.67480834145500574602)
526 #define COS0_8 FIXR(0.74453627100229844977)
527 #define COS0_9 FIXR(0.83934964541552703873)
528 #define COS0_10 FIXR(0.97256823786196069369)
529 #define COS0_11 FIXR(1.16943993343288495515)
530 #define COS0_12 FIXR(1.48416461631416627724)
531 #define COS0_13 FIXR(2.05778100995341155085)
532 #define COS0_14 FIXR(3.40760841846871878570)
533 #define COS0_15 FIXR(10.19000812354805681150)
534
535 #define COS1_0 FIXR(0.50241928618815570551)
536 #define COS1_1 FIXR(0.52249861493968888062)
537 #define COS1_2 FIXR(0.56694403481635770368)
538 #define COS1_3 FIXR(0.64682178335999012954)
539 #define COS1_4 FIXR(0.78815462345125022473)
540 #define COS1_5 FIXR(1.06067768599034747134)
541 #define COS1_6 FIXR(1.72244709823833392782)
542 #define COS1_7 FIXR(5.10114861868916385802)
543
544 #define COS2_0 FIXR(0.50979557910415916894)
545 #define COS2_1 FIXR(0.60134488693504528054)
546 #define COS2_2 FIXR(0.89997622313641570463)
547 #define COS2_3 FIXR(2.56291544774150617881)
548
549 #define COS3_0 FIXR(0.54119610014619698439)
550 #define COS3_1 FIXR(1.30656296487637652785)
551
552 #define COS4_0 FIXR(0.70710678118654752439)
553
554 /* butterfly operator */
555 #define BF(a, b, c)\
556 {\
557 tmp0 = tab[a] + tab[b];\
558 tmp1 = tab[a] - tab[b];\
559 tab[a] = tmp0;\
560 tab[b] = MULL(tmp1, c);\
561 }
562
563 #define BF1(a, b, c, d)\
564 {\
565 BF(a, b, COS4_0);\
566 BF(c, d, -COS4_0);\
567 tab[c] += tab[d];\
568 }
569
570 #define BF2(a, b, c, d)\
571 {\
572 BF(a, b, COS4_0);\
573 BF(c, d, -COS4_0);\
574 tab[c] += tab[d];\
575 tab[a] += tab[c];\
576 tab[c] += tab[b];\
577 tab[b] += tab[d];\
578 }
579
580 #define ADD(a, b) tab[a] += tab[b]
581
582 /* DCT32 without 1/sqrt(2) coef zero scaling. */
583 static void dct32(INT32 *out, INT32 *tab)
584 {
585 int tmp0, tmp1;
586
587 /* pass 1 */
588 BF(0, 31, COS0_0);
589 BF(1, 30, COS0_1);
590 BF(2, 29, COS0_2);
591 BF(3, 28, COS0_3);
592 BF(4, 27, COS0_4);
593 BF(5, 26, COS0_5);
594 BF(6, 25, COS0_6);
595 BF(7, 24, COS0_7);
596 BF(8, 23, COS0_8);
597 BF(9, 22, COS0_9);
598 BF(10, 21, COS0_10);
599 BF(11, 20, COS0_11);
600 BF(12, 19, COS0_12);
601 BF(13, 18, COS0_13);
602 BF(14, 17, COS0_14);
603 BF(15, 16, COS0_15);
604
605 /* pass 2 */
606 BF(0, 15, COS1_0);
607 BF(1, 14, COS1_1);
608 BF(2, 13, COS1_2);
609 BF(3, 12, COS1_3);
610 BF(4, 11, COS1_4);
611 BF(5, 10, COS1_5);
612 BF(6, 9, COS1_6);
613 BF(7, 8, COS1_7);
614
615 BF(16, 31, -COS1_0);
616 BF(17, 30, -COS1_1);
617 BF(18, 29, -COS1_2);
618 BF(19, 28, -COS1_3);
619 BF(20, 27, -COS1_4);
620 BF(21, 26, -COS1_5);
621 BF(22, 25, -COS1_6);
622 BF(23, 24, -COS1_7);
623
624 /* pass 3 */
625 BF(0, 7, COS2_0);
626 BF(1, 6, COS2_1);
627 BF(2, 5, COS2_2);
628 BF(3, 4, COS2_3);
629
630 BF(8, 15, -COS2_0);
631 BF(9, 14, -COS2_1);
632 BF(10, 13, -COS2_2);
633 BF(11, 12, -COS2_3);
634
635 BF(16, 23, COS2_0);
636 BF(17, 22, COS2_1);
637 BF(18, 21, COS2_2);
638 BF(19, 20, COS2_3);
639
640 BF(24, 31, -COS2_0);
641 BF(25, 30, -COS2_1);
642 BF(26, 29, -COS2_2);
643 BF(27, 28, -COS2_3);
644
645 /* pass 4 */
646 BF(0, 3, COS3_0);
647 BF(1, 2, COS3_1);
648
649 BF(4, 7, -COS3_0);
650 BF(5, 6, -COS3_1);
651
652 BF(8, 11, COS3_0);
653 BF(9, 10, COS3_1);
654
655 BF(12, 15, -COS3_0);
656 BF(13, 14, -COS3_1);
657
658 BF(16, 19, COS3_0);
659 BF(17, 18, COS3_1);
660
661 BF(20, 23, -COS3_0);
662 BF(21, 22, -COS3_1);
663
664 BF(24, 27, COS3_0);
665 BF(25, 26, COS3_1);
666
667 BF(28, 31, -COS3_0);
668 BF(29, 30, -COS3_1);
669
670 /* pass 5 */
671 BF1(0, 1, 2, 3);
672 BF2(4, 5, 6, 7);
673 BF1(8, 9, 10, 11);
674 BF2(12, 13, 14, 15);
675 BF1(16, 17, 18, 19);
676 BF2(20, 21, 22, 23);
677 BF1(24, 25, 26, 27);
678 BF2(28, 29, 30, 31);
679
680 /* pass 6 */
681
682 ADD( 8, 12);
683 ADD(12, 10);
684 ADD(10, 14);
685 ADD(14, 9);
686 ADD( 9, 13);
687 ADD(13, 11);
688 ADD(11, 15);
689
690 out[ 0] = tab[0];
691 out[16] = tab[1];
692 out[ 8] = tab[2];
693 out[24] = tab[3];
694 out[ 4] = tab[4];
695 out[20] = tab[5];
696 out[12] = tab[6];
697 out[28] = tab[7];
698 out[ 2] = tab[8];
699 out[18] = tab[9];
700 out[10] = tab[10];
701 out[26] = tab[11];
702 out[ 6] = tab[12];
703 out[22] = tab[13];
704 out[14] = tab[14];
705 out[30] = tab[15];
706
707 ADD(24, 28);
708 ADD(28, 26);
709 ADD(26, 30);
710 ADD(30, 25);
711 ADD(25, 29);
712 ADD(29, 27);
713 ADD(27, 31);
714
715 out[ 1] = tab[16] + tab[24];
716 out[17] = tab[17] + tab[25];
717 out[ 9] = tab[18] + tab[26];
718 out[25] = tab[19] + tab[27];
719 out[ 5] = tab[20] + tab[28];
720 out[21] = tab[21] + tab[29];
721 out[13] = tab[22] + tab[30];
722 out[29] = tab[23] + tab[31];
723 out[ 3] = tab[24] + tab[20];
724 out[19] = tab[25] + tab[21];
725 out[11] = tab[26] + tab[22];
726 out[27] = tab[27] + tab[23];
727 out[ 7] = tab[28] + tab[18];
728 out[23] = tab[29] + tab[19];
729 out[15] = tab[30] + tab[17];
730 out[31] = tab[31];
731 }
732
733 #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
734
735 #if FRAC_BITS <= 15
736
737 #define OUT_SAMPLE(sum)\
738 {\
739 int sum1;\
740 sum1 = (sum + (1 << (OUT_SHIFT - 1))) >> OUT_SHIFT;\
741 if (sum1 < -32768)\
742 sum1 = -32768;\
743 else if (sum1 > 32767)\
744 sum1 = 32767;\
745 *samples = sum1;\
746 samples += incr;\
747 }
748
749 #define SUM8(off, op) \
750 { \
751 sum op w[0 * 64 + off] * p[0 * 64];\
752 sum op w[1 * 64 + off] * p[1 * 64];\
753 sum op w[2 * 64 + off] * p[2 * 64];\
754 sum op w[3 * 64 + off] * p[3 * 64];\
755 sum op w[4 * 64 + off] * p[4 * 64];\
756 sum op w[5 * 64 + off] * p[5 * 64];\
757 sum op w[6 * 64 + off] * p[6 * 64];\
758 sum op w[7 * 64 + off] * p[7 * 64];\
759 }
760
761 #else
762
763 #define OUT_SAMPLE(sum)\
764 {\
765 int sum1;\
766 sum1 = (int)((sum + (INT64_C(1) << (OUT_SHIFT - 1))) >> OUT_SHIFT);\
767 if (sum1 < -32768)\
768 sum1 = -32768;\
769 else if (sum1 > 32767)\
770 sum1 = 32767;\
771 *samples = sum1;\
772 samples += incr;\
773 }
774
775 #define SUM8(off, op) \
776 { \
777 sum op MUL64(w[0 * 64 + off], p[0 * 64]);\
778 sum op MUL64(w[1 * 64 + off], p[1 * 64]);\
779 sum op MUL64(w[2 * 64 + off], p[2 * 64]);\
780 sum op MUL64(w[3 * 64 + off], p[3 * 64]);\
781 sum op MUL64(w[4 * 64 + off], p[4 * 64]);\
782 sum op MUL64(w[5 * 64 + off], p[5 * 64]);\
783 sum op MUL64(w[6 * 64 + off], p[6 * 64]);\
784 sum op MUL64(w[7 * 64 + off], p[7 * 64]);\
785 }
786
787 #endif
788
789 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
790 32 samples. */
791 /* XXX: optimize by avoiding ring buffer usage */
792 static void synth_filter(MPADecodeContext *s1,
793 int ch, INT16 *samples, int incr,
794 INT32 sb_samples[SBLIMIT])
795 {
796 INT32 tmp[32];
797 register MPA_INT *synth_buf, *p;
798 register MPA_INT *w;
799 int j, offset, v;
800 #if FRAC_BITS <= 15
801 int sum;
802 #else
803 INT64 sum;
804 #endif
805
806 dct32(tmp, sb_samples);
807
808 offset = s1->synth_buf_offset[ch];
809 synth_buf = s1->synth_buf[ch] + offset;
810
811 for(j=0;j<32;j++) {
812 v = tmp[j];
813 #if FRAC_BITS <= 15
814 /* NOTE: can cause a loss in precision if very high amplitude
815 sound */
816 if (v > 32767)
817 v = 32767;
818 else if (v < -32768)
819 v = -32768;
820 #endif
821 synth_buf[j] = v;
822 }
823 /* copy to avoid wrap */
824 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
825
826 w = window;
827 for(j=0;j<16;j++) {
828 sum = 0;
829 p = synth_buf + 16 + j; /* 0-15 */
830 SUM8(0, +=);
831 p = synth_buf + 48 - j; /* 32-47 */
832 SUM8(32, -=);
833 OUT_SAMPLE(sum);
834 w++;
835 }
836
837 p = synth_buf + 32; /* 48 */
838 sum = 0;
839 SUM8(32, -=);
840 OUT_SAMPLE(sum);
841 w++;
842
843 for(j=17;j<32;j++) {
844 sum = 0;
845 p = synth_buf + 48 - j; /* 17-31 */
846 SUM8(0, -=);
847 p = synth_buf + 16 + j; /* 49-63 */
848 SUM8(32, -=);
849 OUT_SAMPLE(sum);
850 w++;
851 }
852 offset = (offset - 32) & 511;
853 s1->synth_buf_offset[ch] = offset;
854 }
855
856 /* cos(pi*i/24) */
857 #define C1 FIXR(0.99144486137381041114)
858 #define C3 FIXR(0.92387953251128675612)
859 #define C5 FIXR(0.79335334029123516458)
860 #define C7 FIXR(0.60876142900872063941)
861 #define C9 FIXR(0.38268343236508977173)
862 #define C11 FIXR(0.13052619222005159154)
863
864 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
865 cases. */
866 static void imdct12(int *out, int *in)
867 {
868 int tmp;
869 INT64 in1_3, in1_9, in4_3, in4_9;
870
871 in1_3 = MUL64(in[1], C3);
872 in1_9 = MUL64(in[1], C9);
873 in4_3 = MUL64(in[4], C3);
874 in4_9 = MUL64(in[4], C9);
875
876 tmp = FRAC_RND(MUL64(in[0], C7) - in1_3 - MUL64(in[2], C11) +
877 MUL64(in[3], C1) - in4_9 - MUL64(in[5], C5));
878 out[0] = tmp;
879 out[5] = -tmp;
880 tmp = FRAC_RND(MUL64(in[0] - in[3], C9) - in1_3 +
881 MUL64(in[2] + in[5], C3) - in4_9);
882 out[1] = tmp;
883 out[4] = -tmp;
884 tmp = FRAC_RND(MUL64(in[0], C11) - in1_9 + MUL64(in[2], C7) -
885 MUL64(in[3], C5) + in4_3 - MUL64(in[5], C1));
886 out[2] = tmp;
887 out[3] = -tmp;
888 tmp = FRAC_RND(MUL64(-in[0], C5) + in1_9 + MUL64(in[2], C1) +
889 MUL64(in[3], C11) - in4_3 - MUL64(in[5], C7));
890 out[6] = tmp;
891 out[11] = tmp;
892 tmp = FRAC_RND(MUL64(-in[0] + in[3], C3) - in1_9 +
893 MUL64(in[2] + in[5], C9) + in4_3);
894 out[7] = tmp;
895 out[10] = tmp;
896 tmp = FRAC_RND(-MUL64(in[0], C1) - in1_3 - MUL64(in[2], C5) -
897 MUL64(in[3], C7) - in4_9 - MUL64(in[5], C11));
898 out[8] = tmp;
899 out[9] = tmp;
900 }
901
902 #undef C1
903 #undef C3
904 #undef C5
905 #undef C7
906 #undef C9
907 #undef C11
908
909 /* cos(pi*i/18) */
910 #define C1 FIXR(0.98480775301220805936)
911 #define C2 FIXR(0.93969262078590838405)
912 #define C3 FIXR(0.86602540378443864676)
913 #define C4 FIXR(0.76604444311897803520)
914 #define C5 FIXR(0.64278760968653932632)
915 #define C6 FIXR(0.5)
916 #define C7 FIXR(0.34202014332566873304)
917 #define C8 FIXR(0.17364817766693034885)
918
919 /* 0.5 / cos(pi*(2*i+1)/36) */
920 static const int icos36[9] = {
921 FIXR(0.50190991877167369479),
922 FIXR(0.51763809020504152469),
923 FIXR(0.55168895948124587824),
924 FIXR(0.61038729438072803416),
925 FIXR(0.70710678118654752439),
926 FIXR(0.87172339781054900991),
927 FIXR(1.18310079157624925896),
928 FIXR(1.93185165257813657349),
929 FIXR(5.73685662283492756461),
930 };
931
932 static const int icos72[18] = {
933 /* 0.5 / cos(pi*(2*i+19)/72) */
934 FIXR(0.74009361646113053152),
935 FIXR(0.82133981585229078570),
936 FIXR(0.93057949835178895673),
937 FIXR(1.08284028510010010928),
938 FIXR(1.30656296487637652785),
939 FIXR(1.66275476171152078719),
940 FIXR(2.31011315767264929558),
941 FIXR(3.83064878777019433457),
942 FIXR(11.46279281302667383546),
943
944 /* 0.5 / cos(pi*(2*(i + 18) +19)/72) */
945 FIXR(-0.67817085245462840086),
946 FIXR(-0.63023620700513223342),
947 FIXR(-0.59284452371708034528),
948 FIXR(-0.56369097343317117734),
949 FIXR(-0.54119610014619698439),
950 FIXR(-0.52426456257040533932),
951 FIXR(-0.51213975715725461845),
952 FIXR(-0.50431448029007636036),
953 FIXR(-0.50047634258165998492),
954 };
955
956 /* using Lee like decomposition followed by hand coded 9 points DCT */
957 static void imdct36(int *out, int *in)
958 {
959 int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
960 int tmp[18], *tmp1, *in1;
961 INT64 in3_3, in6_6;
962
963 for(i=17;i>=1;i--)
964 in[i] += in[i-1];
965 for(i=17;i>=3;i-=2)
966 in[i] += in[i-2];
967
968 for(j=0;j<2;j++) {
969 tmp1 = tmp + j;
970 in1 = in + j;
971
972 in3_3 = MUL64(in1[2*3], C3);
973 in6_6 = MUL64(in1[2*6], C6);
974
975 tmp1[0] = FRAC_RND(MUL64(in1[2*1], C1) + in3_3 +
976 MUL64(in1[2*5], C5) + MUL64(in1[2*7], C7));
977 tmp1[2] = in1[2*0] + FRAC_RND(MUL64(in1[2*2], C2) +
978 MUL64(in1[2*4], C4) + in6_6 +
979 MUL64(in1[2*8], C8));
980 tmp1[4] = FRAC_RND(MUL64(in1[2*1] - in1[2*5] - in1[2*7], C3));
981 tmp1[6] = FRAC_RND(MUL64(in1[2*2] - in1[2*4] - in1[2*8], C6)) -
982 in1[2*6] + in1[2*0];
983 tmp1[8] = FRAC_RND(MUL64(in1[2*1], C5) - in3_3 -
984 MUL64(in1[2*5], C7) + MUL64(in1[2*7], C1));
985 tmp1[10] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C8) -
986 MUL64(in1[2*4], C2) + in6_6 +
987 MUL64(in1[2*8], C4));
988 tmp1[12] = FRAC_RND(MUL64(in1[2*1], C7) - in3_3 +
989 MUL64(in1[2*5], C1) -
990 MUL64(in1[2*7], C5));
991 tmp1[14] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C4) +
992 MUL64(in1[2*4], C8) + in6_6 -
993 MUL64(in1[2*8], C2));
994 tmp1[16] = in1[2*0] - in1[2*2] + in1[2*4] - in1[2*6] + in1[2*8];
995 }
996
997 i = 0;
998 for(j=0;j<4;j++) {
999 t0 = tmp[i];
1000 t1 = tmp[i + 2];
1001 s0 = t1 + t0;
1002 s2 = t1 - t0;
1003
1004 t2 = tmp[i + 1];
1005 t3 = tmp[i + 3];
1006 s1 = MULL(t3 + t2, icos36[j]);
1007 s3 = MULL(t3 - t2, icos36[8 - j]);
1008
1009 t0 = MULL(s0 + s1, icos72[9 + 8 - j]);
1010 t1 = MULL(s0 - s1, icos72[8 - j]);
1011 out[18 + 9 + j] = t0;
1012 out[18 + 8 - j] = t0;
1013 out[9 + j] = -t1;
1014 out[8 - j] = t1;
1015
1016 t0 = MULL(s2 + s3, icos72[9+j]);
1017 t1 = MULL(s2 - s3, icos72[j]);
1018 out[18 + 9 + (8 - j)] = t0;
1019 out[18 + j] = t0;
1020 out[9 + (8 - j)] = -t1;
1021 out[j] = t1;
1022 i += 4;
1023 }
1024
1025 s0 = tmp[16];
1026 s1 = MULL(tmp[17], icos36[4]);
1027 t0 = MULL(s0 + s1, icos72[9 + 4]);
1028 t1 = MULL(s0 - s1, icos72[4]);
1029 out[18 + 9 + 4] = t0;
1030 out[18 + 8 - 4] = t0;
1031 out[9 + 4] = -t1;
1032 out[8 - 4] = t1;
1033 }
1034
1035 /* fast header check for resync */
1036 static int check_header(UINT32 header)
1037 {
1038 /* header */
1039 if ((header & 0xffe00000) != 0xffe00000)
1040 return -1;
1041 /* layer check */
1042 if (((header >> 17) & 3) == 0)
1043 return -1;
1044 /* bit rate */
1045 if (((header >> 12) & 0xf) == 0xf)
1046 return -1;
1047 /* frequency */
1048 if (((header >> 10) & 3) == 3)
1049 return -1;
1050 return 0;
1051 }
1052
1053 /* header + layer + bitrate + freq + lsf/mpeg25 */
1054 #define SAME_HEADER_MASK \
1055 (0xffe00000 | (3 << 17) | (0xf << 12) | (3 << 10) | (3 << 19))
1056
1057 /* header decoding. MUST check the header before because no
1058 consistency check is done there. Return 1 if free format found and
1059 that the frame size must be computed externally */
1060 static int decode_header(MPADecodeContext *s, UINT32 header)
1061 {
1062 int sample_rate, frame_size, mpeg25, padding;
1063 int sample_rate_index, bitrate_index;
1064 if (header & (1<<20)) {
1065 s->lsf = (header & (1<<19)) ? 0 : 1;
1066 mpeg25 = 0;
1067 } else {
1068 s->lsf = 1;
1069 mpeg25 = 1;
1070 }
1071
1072 s->layer = 4 - ((header >> 17) & 3);
1073 /* extract frequency */
1074 sample_rate_index = (header >> 10) & 3;
1075 sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1076 sample_rate_index += 3 * (s->lsf + mpeg25);
1077 s->sample_rate_index = sample_rate_index;
1078 s->error_protection = ((header >> 16) & 1) ^ 1;
1079 s->sample_rate = sample_rate;
1080
1081 bitrate_index = (header >> 12) & 0xf;
1082 padding = (header >> 9) & 1;
1083 //extension = (header >> 8) & 1;
1084 s->mode = (header >> 6) & 3;
1085 s->mode_ext = (header >> 4) & 3;
1086 //copyright = (header >> 3) & 1;
1087 //original = (header >> 2) & 1;
1088 //emphasis = header & 3;
1089
1090 if (s->mode == MPA_MONO)
1091 s->nb_channels = 1;
1092 else
1093 s->nb_channels = 2;
1094
1095 if (bitrate_index != 0) {
1096 frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1097 s->bit_rate = frame_size * 1000;
1098 switch(s->layer) {
1099 case 1:
1100 frame_size = (frame_size * 12000) / sample_rate;
1101 frame_size = (frame_size + padding) * 4;
1102 break;
1103 case 2:
1104 frame_size = (frame_size * 144000) / sample_rate;
1105 frame_size += padding;
1106 break;
1107 default:
1108 case 3:
1109 frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1110 frame_size += padding;
1111 break;
1112 }
1113 s->frame_size = frame_size;
1114 } else {
1115 /* if no frame size computed, signal it */
1116 if (!s->free_format_frame_size)
1117 return 1;
1118 /* free format: compute bitrate and real frame size from the
1119 frame size we extracted by reading the bitstream */
1120 s->frame_size = s->free_format_frame_size;
1121 switch(s->layer) {
1122 case 1:
1123 s->frame_size += padding * 4;
1124 s->bit_rate = (s->frame_size * sample_rate) / 48000;
1125 break;
1126 case 2:
1127 s->frame_size += padding;
1128 s->bit_rate = (s->frame_size * sample_rate) / 144000;
1129 break;
1130 default:
1131 case 3:
1132 s->frame_size += padding;
1133 s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1134 break;
1135 }
1136 }
1137
1138 #if defined(DEBUG)
1139 printf("layer%d, %d Hz, %d kbits/s, ",
1140 s->layer, s->sample_rate, s->bit_rate);
1141 if (s->nb_channels == 2) {
1142 if (s->layer == 3) {
1143 if (s->mode_ext & MODE_EXT_MS_STEREO)
1144 printf("ms-");
1145 if (s->mode_ext & MODE_EXT_I_STEREO)
1146 printf("i-");
1147 }
1148 printf("stereo");
1149 } else {
1150 printf("mono");
1151 }
1152 printf("\n");
1153 #endif
1154 return 0;
1155 }
1156
1157 /* return the number of decoded frames */
1158 static int mp_decode_layer1(MPADecodeContext *s)
1159 {
1160 int bound, i, v, n, ch, j, mant;
1161 UINT8 allocation[MPA_MAX_CHANNELS][SBLIMIT];
1162 UINT8 scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1163
1164 if (s->mode == MPA_JSTEREO)
1165 bound = (s->mode_ext + 1) * 4;
1166 else
1167 bound = SBLIMIT;
1168
1169 /* allocation bits */
1170 for(i=0;i<bound;i++) {
1171 for(ch=0;ch<s->nb_channels;ch++) {
1172 allocation[ch][i] = get_bits(&s->gb, 4);
1173 }
1174 }
1175 for(i=bound;i<SBLIMIT;i++) {
1176 allocation[0][i] = get_bits(&s->gb, 4);
1177 }
1178
1179 /* scale factors */
1180 for(i=0;i<bound;i++) {
1181 for(ch=0;ch<s->nb_channels;ch++) {
1182 if (allocation[ch][i])
1183 scale_factors[ch][i] = get_bits(&s->gb, 6);
1184 }
1185 }
1186 for(i=bound;i<SBLIMIT;i++) {
1187 if (allocation[0][i]) {
1188 scale_factors[0][i] = get_bits(&s->gb, 6);
1189 scale_factors[1][i] = get_bits(&s->gb, 6);
1190 }
1191 }
1192
1193 /* compute samples */
1194 for(j=0;j<12;j++) {
1195 for(i=0;i<bound;i++) {
1196 for(ch=0;ch<s->nb_channels;ch++) {
1197 n = allocation[ch][i];
1198 if (n) {
1199 mant = get_bits(&s->gb, n + 1);
1200 v = l1_unscale(n, mant, scale_factors[ch][i]);
1201 } else {
1202 v = 0;
1203 }
1204 s->sb_samples[ch][j][i] = v;
1205 }
1206 }
1207 for(i=bound;i<SBLIMIT;i++) {
1208 n = allocation[0][i];
1209 if (n) {
1210 mant = get_bits(&s->gb, n + 1);
1211 v = l1_unscale(n, mant, scale_factors[0][i]);
1212 s->sb_samples[0][j][i] = v;
1213 v = l1_unscale(n, mant, scale_factors[1][i]);
1214 s->sb_samples[1][j][i] = v;
1215 } else {
1216 s->sb_samples[0][j][i] = 0;
1217 s->sb_samples[1][j][i] = 0;
1218 }
1219 }
1220 }
1221 return 12;
1222 }
1223
1224 /* bitrate is in kb/s */
1225 int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1226 {
1227 int ch_bitrate, table;
1228
1229 ch_bitrate = bitrate / nb_channels;
1230 if (!lsf) {
1231 if ((freq == 48000 && ch_bitrate >= 56) ||
1232 (ch_bitrate >= 56 && ch_bitrate <= 80))
1233 table = 0;
1234 else if (freq != 48000 && ch_bitrate >= 96)
1235 table = 1;
1236 else if (freq != 32000 && ch_bitrate <= 48)
1237 table = 2;
1238 else
1239 table = 3;
1240 } else {
1241 table = 4;
1242 }
1243 return table;
1244 }
1245
1246 static int mp_decode_layer2(MPADecodeContext *s)
1247 {
1248 int sblimit; /* number of used subbands */
1249 const unsigned char *alloc_table;
1250 int table, bit_alloc_bits, i, j, ch, bound, v;
1251 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1252 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1253 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1254 int scale, qindex, bits, steps, k, l, m, b;
1255
1256 /* select decoding table */
1257 table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1258 s->sample_rate, s->lsf);
1259 sblimit = sblimit_table[table];
1260 alloc_table = alloc_tables[table];
1261
1262 if (s->mode == MPA_JSTEREO)
1263 bound = (s->mode_ext + 1) * 4;
1264 else
1265 bound = sblimit;
1266
1267 dprintf("bound=%d sblimit=%d\n", bound, sblimit);
1268 /* parse bit allocation */
1269 j = 0;
1270 for(i=0;i<bound;i++) {
1271 bit_alloc_bits = alloc_table[j];
1272 for(ch=0;ch<s->nb_channels;ch++) {
1273 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1274 }
1275 j += 1 << bit_alloc_bits;
1276 }
1277 for(i=bound;i<sblimit;i++) {
1278 bit_alloc_bits = alloc_table[j];
1279 v = get_bits(&s->gb, bit_alloc_bits);
1280 bit_alloc[0][i] = v;
1281 bit_alloc[1][i] = v;
1282 j += 1 << bit_alloc_bits;
1283 }
1284
1285 #ifdef DEBUG
1286 {
1287 for(ch=0;ch<s->nb_channels;ch++) {
1288 for(i=0;i<sblimit;i++)
1289 printf(" %d", bit_alloc[ch][i]);
1290 printf("\n");
1291 }
1292 }
1293 #endif
1294
1295 /* scale codes */
1296 for(i=0;i<sblimit;i++) {
1297 for(ch=0;ch<s->nb_channels;ch++) {
1298 if (bit_alloc[ch][i])
1299 scale_code[ch][i] = get_bits(&s->gb, 2);
1300 }
1301 }
1302
1303 /* scale factors */
1304 for(i=0;i<sblimit;i++) {
1305 for(ch=0;ch<s->nb_channels;ch++) {
1306 if (bit_alloc[ch][i]) {
1307 sf = scale_factors[ch][i];
1308 switch(scale_code[ch][i]) {
1309 default:
1310 case 0:
1311 sf[0] = get_bits(&s->gb, 6);
1312 sf[1] = get_bits(&s->gb, 6);
1313 sf[2] = get_bits(&s->gb, 6);
1314 break;
1315 case 2:
1316 sf[0] = get_bits(&s->gb, 6);
1317 sf[1] = sf[0];
1318 sf[2] = sf[0];
1319 break;
1320 case 1:
1321 sf[0] = get_bits(&s->gb, 6);
1322 sf[2] = get_bits(&s->gb, 6);
1323 sf[1] = sf[0];
1324 break;
1325 case 3:
1326 sf[0] = get_bits(&s->gb, 6);
1327 sf[2] = get_bits(&s->gb, 6);
1328 sf[1] = sf[2];
1329 break;
1330 }
1331 }
1332 }
1333 }
1334
1335 #ifdef DEBUG
1336 for(ch=0;ch<s->nb_channels;ch++) {
1337 for(i=0;i<sblimit;i++) {
1338 if (bit_alloc[ch][i]) {
1339 sf = scale_factors[ch][i];
1340 printf(" %d %d %d", sf[0], sf[1], sf[2]);
1341 } else {
1342 printf(" -");
1343 }
1344 }
1345 printf("\n");
1346 }
1347 #endif
1348
1349 /* samples */
1350 for(k=0;k<3;k++) {
1351 for(l=0;l<12;l+=3) {
1352 j = 0;
1353 for(i=0;i<bound;i++) {
1354 bit_alloc_bits = alloc_table[j];
1355 for(ch=0;ch<s->nb_channels;ch++) {
1356 b = bit_alloc[ch][i];
1357 if (b) {
1358 scale = scale_factors[ch][i][k];
1359 qindex = alloc_table[j+b];
1360 bits = quant_bits[qindex];
1361 if (bits < 0) {
1362 /* 3 values at the same time */
1363 v = get_bits(&s->gb, -bits);
1364 steps = quant_steps[qindex];
1365 s->sb_samples[ch][k * 12 + l + 0][i] =
1366 l2_unscale_group(steps, v % steps, scale);
1367 v = v / steps;
1368 s->sb_samples[ch][k * 12 + l + 1][i] =
1369 l2_unscale_group(steps, v % steps, scale);
1370 v = v / steps;
1371 s->sb_samples[ch][k * 12 + l + 2][i] =
1372 l2_unscale_group(steps, v, scale);
1373 } else {
1374 for(m=0;m<3;m++) {
1375 v = get_bits(&s->gb, bits);
1376 v = l1_unscale(bits - 1, v, scale);
1377 s->sb_samples[ch][k * 12 + l + m][i] = v;
1378 }
1379 }
1380 } else {
1381 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1382 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1383 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1384 }
1385 }
1386 /* next subband in alloc table */
1387 j += 1 << bit_alloc_bits;
1388 }
1389 /* XXX: find a way to avoid this duplication of code */
1390 for(i=bound;i<sblimit;i++) {
1391 bit_alloc_bits = alloc_table[j];
1392 b = bit_alloc[0][i];
1393 if (b) {
1394 int mant, scale0, scale1;
1395 scale0 = scale_factors[0][i][k];
1396 scale1 = scale_factors[1][i][k];
1397 qindex = alloc_table[j+b];
1398 bits = quant_bits[qindex];
1399 if (bits < 0) {
1400 /* 3 values at the same time */
1401 v = get_bits(&s->gb, -bits);
1402 steps = quant_steps[qindex];
1403 mant = v % steps;
1404 v = v / steps;
1405 s->sb_samples[0][k * 12 + l + 0][i] =
1406 l2_unscale_group(steps, mant, scale0);
1407 s->sb_samples[1][k * 12 + l + 0][i] =
1408 l2_unscale_group(steps, mant, scale1);
1409 mant = v % steps;
1410 v = v / steps;
1411 s->sb_samples[0][k * 12 + l + 1][i] =
1412 l2_unscale_group(steps, mant, scale0);
1413 s->sb_samples[1][k * 12 + l + 1][i] =
1414 l2_unscale_group(steps, mant, scale1);
1415 s->sb_samples[0][k * 12 + l + 2][i] =
1416 l2_unscale_group(steps, v, scale0);
1417 s->sb_samples[1][k * 12 + l + 2][i] =
1418 l2_unscale_group(steps, v, scale1);
1419 } else {
1420 for(m=0;m<3;m++) {
1421 mant = get_bits(&s->gb, bits);
1422 s->sb_samples[0][k * 12 + l + m][i] =
1423 l1_unscale(bits - 1, mant, scale0);
1424 s->sb_samples[1][k * 12 + l + m][i] =
1425 l1_unscale(bits - 1, mant, scale1);
1426 }
1427 }
1428 } else {
1429 s->sb_samples[0][k * 12 + l + 0][i] = 0;
1430 s->sb_samples[0][k * 12 + l + 1][i] = 0;
1431 s->sb_samples[0][k * 12 + l + 2][i] = 0;
1432 s->sb_samples[1][k * 12 + l + 0][i] = 0;
1433 s->sb_samples[1][k * 12 + l + 1][i] = 0;
1434 s->sb_samples[1][k * 12 + l + 2][i] = 0;
1435 }
1436 /* next subband in alloc table */
1437 j += 1 << bit_alloc_bits;
1438 }
1439 /* fill remaining samples to zero */
1440 for(i=sblimit;i<SBLIMIT;i++) {
1441 for(ch=0;ch<s->nb_channels;ch++) {
1442 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1443 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1444 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1445 }
1446 }
1447 }
1448 }
1449 return 3 * 12;
1450 }
1451
1452 /*
1453 * Seek back in the stream for backstep bytes (at most 511 bytes)
1454 */
1455 static void seek_to_maindata(MPADecodeContext *s, long backstep)
1456 {
1457 UINT8 *ptr;
1458
1459 /* compute current position in stream */
1460 #ifdef ALT_BITSTREAM_READER
1461 ptr = s->gb.buffer + (s->gb.index>>3);
1462 #else
1463 ptr = s->gb.buf_ptr - (s->gb.bit_cnt >> 3);
1464 #endif
1465 /* copy old data before current one */
1466 ptr -= backstep;
1467 memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
1468 BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
1469 /* init get bits again */
1470 init_get_bits(&s->gb, ptr, s->frame_size + backstep);
1471
1472 /* prepare next buffer */
1473 s->inbuf_index ^= 1;
1474 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1475 s->old_frame_size = s->frame_size;
1476 }
1477
1478 static inline void lsf_sf_expand(int *slen,
1479 int sf, int n1, int n2, int n3)
1480 {
1481 if (n3) {
1482 slen[3] = sf % n3;
1483 sf /= n3;
1484 } else {
1485 slen[3] = 0;
1486 }
1487 if (n2) {
1488 slen[2] = sf % n2;
1489 sf /= n2;
1490 } else {
1491 slen[2] = 0;
1492 }
1493 slen[1] = sf % n1;
1494 sf /= n1;
1495 slen[0] = sf;
1496 }
1497
1498 static void exponents_from_scale_factors(MPADecodeContext *s,
1499 GranuleDef *g,
1500 INT16 *exponents)
1501 {
1502 const UINT8 *bstab, *pretab;
1503 int len, i, j, k, l, v0, shift, gain, gains[3];
1504 INT16 *exp_ptr;
1505
1506 exp_ptr = exponents;
1507 gain = g->global_gain - 210;
1508 shift = g->scalefac_scale + 1;
1509
1510 bstab = band_size_long[s->sample_rate_index];
1511 pretab = mpa_pretab[g->preflag];
1512 for(i=0;i<g->long_end;i++) {
1513 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
1514 len = bstab[i];
1515 for(j=len;j>0;j--)
1516 *exp_ptr++ = v0;
1517 }
1518
1519 if (g->short_start < 13) {
1520 bstab = band_size_short[s->sample_rate_index];
1521 gains[0] = gain - (g->subblock_gain[0] << 3);
1522 gains[1] = gain - (g->subblock_gain[1] << 3);
1523 gains[2] = gain - (g->subblock_gain[2] << 3);
1524 k = g->long_end;
1525 for(i=g->short_start;i<13;i++) {
1526 len = bstab[i];
1527 for(l=0;l<3;l++) {
1528 v0 = gains[l] - (g->scale_factors[k++] << shift);
1529 for(j=len;j>0;j--)
1530 *exp_ptr++ = v0;
1531 }
1532 }
1533 }
1534 }
1535
1536 /* handle n = 0 too */
1537 static inline int get_bitsz(GetBitContext *s, int n)
1538 {
1539 if (n == 0)
1540 return 0;
1541 else
1542 return get_bits(s, n);
1543 }
1544
1545 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1546 INT16 *exponents, int end_pos)
1547 {
1548 int s_index;
1549 int linbits, code, x, y, l, v, i, j, k, pos;
1550 UINT8 *last_buf_ptr;
1551 UINT32 last_bit_buf;
1552 int last_bit_cnt;
1553 VLC *vlc;
1554 UINT8 *code_table;
1555
1556 /* low frequencies (called big values) */
1557 s_index = 0;
1558 for(i=0;i<3;i++) {
1559 j = g->region_size[i];
1560 if (j == 0)
1561 continue;
1562 /* select vlc table */
1563 k = g->table_select[i];
1564 l = mpa_huff_data[k][0];
1565 linbits = mpa_huff_data[k][1];
1566 vlc = &huff_vlc[l];
1567 code_table = huff_code_table[l];
1568
1569 /* read huffcode and compute each couple */
1570 for(;j>0;j--) {
1571 if (get_bits_count(&s->gb) >= end_pos)
1572 break;
1573 if (code_table) {
1574 code = get_vlc(&s->gb, vlc);
1575 if (code < 0)
1576 return -1;
1577 y = code_table[code];
1578 x = y >> 4;
1579 y = y & 0x0f;
1580 } else {
1581 x = 0;
1582 y = 0;
1583 }
1584 dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
1585 i, g->region_size[i] - j, x, y, exponents[s_index]);
1586 if (x) {
1587 if (x == 15)
1588 x += get_bitsz(&s->gb, linbits);
1589 v = l3_unscale(x, exponents[s_index]);
1590 if (get_bits1(&s->gb))
1591 v = -v;
1592 } else {
1593 v = 0;
1594 }
1595 g->sb_hybrid[s_index++] = v;
1596 if (y) {
1597 if (y == 15)
1598 y += get_bitsz(&s->gb, linbits);
1599 v = l3_unscale(y, exponents[s_index]);
1600 if (get_bits1(&s->gb))
1601 v = -v;
1602 } else {
1603 v = 0;
1604 }
1605 g->sb_hybrid[s_index++] = v;
1606 }
1607 }
1608
1609 /* high frequencies */
1610 vlc = &huff_quad_vlc[g->count1table_select];
1611 last_buf_ptr = NULL;
1612 last_bit_buf = 0;
1613 last_bit_cnt = 0;
1614 while (s_index <= 572) {
1615 pos = get_bits_count(&s->gb);
1616 if (pos >= end_pos) {
1617 if (pos > end_pos && last_buf_ptr != NULL) {
1618 /* some encoders generate an incorrect size for this
1619 part. We must go back into the data */
1620 s_index -= 4;
1621 #ifdef ALT_BITSTREAM_READER
1622 s->gb.buffer = last_buf_ptr;
1623 s->gb.index = last_bit_cnt;
1624 #else
1625 s->gb.buf_ptr = last_buf_ptr;
1626 s->gb.bit_buf = last_bit_buf;
1627 s->gb.bit_cnt = last_bit_cnt;
1628 #endif
1629 }
1630 break;
1631 }
1632 #ifdef ALT_BITSTREAM_READER
1633 last_buf_ptr = s->gb.buffer;
1634 last_bit_cnt = s->gb.index;
1635 #else
1636 last_buf_ptr = s->gb.buf_ptr;
1637 last_bit_buf = s->gb.bit_buf;
1638 last_bit_cnt = s->gb.bit_cnt;
1639 #endif
1640
1641 code = get_vlc(&s->gb, vlc);
1642 dprintf("t=%d code=%d\n", g->count1table_select, code);
1643 if (code < 0)
1644 return -1;
1645 for(i=0;i<4;i++) {
1646 if (code & (8 >> i)) {
1647 /* non zero value. Could use a hand coded function for
1648 'one' value */
1649 v = l3_unscale(1, exponents[s_index]);
1650 if(get_bits1(&s->gb))
1651 v = -v;
1652 } else {
1653 v = 0;
1654 }
1655 g->sb_hybrid[s_index++] = v;
1656 }
1657 }
1658 while (s_index < 576)
1659 g->sb_hybrid[s_index++] = 0;
1660 return 0;
1661 }
1662
1663 /* Reorder short blocks from bitstream order to interleaved order. It
1664 would be faster to do it in parsing, but the code would be far more
1665 complicated */
1666 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1667 {
1668 int i, j, k, len;
1669 INT32 *ptr, *dst, *ptr1;
1670 INT32 tmp[576];
1671
1672 if (g->block_type != 2)
1673 return;
1674
1675 if (g->switch_point) {
1676 if (s->sample_rate_index != 8) {
1677 ptr = g->sb_hybrid + 36;
1678 } else {
1679 ptr = g->sb_hybrid + 48;
1680 }
1681 } else {
1682 ptr = g->sb_hybrid;
1683 }
1684
1685 for(i=g->short_start;i<13;i++) {
1686 len = band_size_short[s->sample_rate_index][i];
1687 ptr1 = ptr;
1688 for(k=0;k<3;k++) {
1689 dst = tmp + k;
1690 for(j=len;j>0;j--) {
1691 *dst = *ptr++;
1692 dst += 3;
1693 }
1694 }
1695 memcpy(ptr1, tmp, len * 3 * sizeof(INT32));
1696 }
1697 }
1698
1699 #define ISQRT2 FIXR(0.70710678118654752440)
1700
1701 static void compute_stereo(MPADecodeContext *s,
1702 GranuleDef *g0, GranuleDef *g1)
1703 {
1704 int i, j, k, l;
1705 INT32 v1, v2;
1706 int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1707 INT32 (*is_tab)[16];
1708 INT32 *tab0, *tab1;
1709 int non_zero_found_short[3];
1710
1711 /* intensity stereo */
1712 if (s->mode_ext & MODE_EXT_I_STEREO) {
1713 if (!s->lsf) {
1714 is_tab = is_table;
1715 sf_max = 7;
1716 } else {
1717 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1718 sf_max = 16;
1719 }
1720
1721 tab0 = g0->sb_hybrid + 576;
1722 tab1 = g1->sb_hybrid + 576;
1723
1724 non_zero_found_short[0] = 0;
1725 non_zero_found_short[1] = 0;
1726 non_zero_found_short[2] = 0;
1727 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1728 for(i = 12;i >= g1->short_start;i--) {
1729 /* for last band, use previous scale factor */
1730 if (i != 11)
1731 k -= 3;
1732 len = band_size_short[s->sample_rate_index][i];
1733 for(l=2;l>=0;l--) {
1734 tab0 -= len;
1735 tab1 -= len;
1736 if (!non_zero_found_short[l]) {
1737 /* test if non zero band. if so, stop doing i-stereo */
1738 for(j=0;j<len;j++) {
1739 if (tab1[j] != 0) {
1740 non_zero_found_short[l] = 1;
1741 goto found1;
1742 }
1743 }
1744 sf = g1->scale_factors[k + l];
1745 if (sf >= sf_max)
1746 goto found1;
1747
1748 v1 = is_tab[0][sf];
1749 v2 = is_tab[1][sf];
1750 for(j=0;j<len;j++) {
1751 tmp0 = tab0[j];
1752 tab0[j] = MULL(tmp0, v1);
1753 tab1[j] = MULL(tmp0, v2);
1754 }
1755 } else {
1756 found1:
1757 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1758 /* lower part of the spectrum : do ms stereo
1759 if enabled */
1760 for(j=0;j<len;j++) {
1761 tmp0 = tab0[j];
1762 tmp1 = tab1[j];
1763 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1764 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1765 }
1766 }
1767 }
1768 }
1769 }
1770
1771 non_zero_found = non_zero_found_short[0] |
1772 non_zero_found_short[1] |
1773 non_zero_found_short[2];
1774
1775 for(i = g1->long_end - 1;i >= 0;i--) {
1776 len = band_size_long[s->sample_rate_index][i];
1777 tab0 -= len;
1778 tab1 -= len;
1779 /* test if non zero band. if so, stop doing i-stereo */
1780 if (!non_zero_found) {
1781 for(j=0;j<len;j++) {
1782 if (tab1[j] != 0) {
1783 non_zero_found = 1;
1784 goto found2;
1785 }
1786 }
1787 /* for last band, use previous scale factor */
1788 k = (i == 21) ? 20 : i;
1789 sf = g1->scale_factors[k];
1790 if (sf >= sf_max)
1791 goto found2;
1792 v1 = is_tab[0][sf];
1793 v2 = is_tab[1][sf];
1794 for(j=0;j<len;j++) {
1795 tmp0 = tab0[j];
1796 tab0[j] = MULL(tmp0, v1);
1797 tab1[j] = MULL(tmp0, v2);
1798 }
1799 } else {
1800 found2:
1801 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1802 /* lower part of the spectrum : do ms stereo
1803 if enabled */
1804 for(j=0;j<len;j++) {
1805 tmp0 = tab0[j];
1806 tmp1 = tab1[j];
1807 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1808 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1809 }
1810 }
1811 }
1812 }
1813 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1814 /* ms stereo ONLY */
1815 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1816 global gain */
1817 tab0 = g0->sb_hybrid;
1818 tab1 = g1->sb_hybrid;
1819 for(i=0;i<576;i++) {
1820 tmp0 = tab0[i];
1821 tmp1 = tab1[i];
1822 tab0[i] = tmp0 + tmp1;
1823 tab1[i] = tmp0 - tmp1;
1824 }
1825 }
1826 }
1827
1828 static void compute_antialias(MPADecodeContext *s,
1829 GranuleDef *g)
1830 {
1831 INT32 *ptr, *p0, *p1, *csa;
1832 int n, tmp0, tmp1, i, j;
1833
1834 /* we antialias only "long" bands */
1835 if (g->block_type == 2) {
1836 if (!g->switch_point)
1837 return;
1838 /* XXX: check this for 8000Hz case */
1839 n = 1;
1840 } else {
1841 n = SBLIMIT - 1;
1842 }
1843
1844 ptr = g->sb_hybrid + 18;
1845 for(i = n;i > 0;i--) {
1846 p0 = ptr - 1;
1847 p1 = ptr;
1848 csa = &csa_table[0][0];
1849 for(j=0;j<8;j++) {
1850 tmp0 = *p0;
1851 tmp1 = *p1;
1852 *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));
1853 *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));
1854 p0--;
1855 p1++;
1856 csa += 2;
1857 }
1858 ptr += 18;
1859 }
1860 }
1861
1862 static void compute_imdct(MPADecodeContext *s,
1863 GranuleDef *g,
1864 INT32 *sb_samples,
1865 INT32 *mdct_buf)
1866 {
1867 INT32 *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1;
1868 INT32 in[6];
1869 INT32 out[36];
1870 INT32 out2[12];
1871 int i, j, k, mdct_long_end, v, sblimit;
1872
1873 /* find last non zero block */
1874 ptr = g->sb_hybrid + 576;
1875 ptr1 = g->sb_hybrid + 2 * 18;
1876 while (ptr >= ptr1) {
1877 ptr -= 6;
1878 v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
1879 if (v != 0)
1880 break;
1881 }
1882 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1883
1884 if (g->block_type == 2) {
1885 /* XXX: check for 8000 Hz */
1886 if (g->switch_point)
1887 mdct_long_end = 2;
1888 else
1889 mdct_long_end = 0;
1890 } else {
1891 mdct_long_end = sblimit;
1892 }
1893
1894 buf = mdct_buf;
1895 ptr = g->sb_hybrid;
1896 for(j=0;j<mdct_long_end;j++) {
1897 imdct36(out, ptr);
1898 /* apply window & overlap with previous buffer */
1899 out_ptr = sb_samples + j;
1900 /* select window */
1901 if (g->switch_point && j < 2)
1902 win1 = mdct_win[0];
1903 else
1904 win1 = mdct_win[g->block_type];
1905 /* select frequency inversion */
1906 win = win1 + ((4 * 36) & -(j & 1));
1907 for(i=0;i<18;i++) {
1908 *out_ptr = MULL(out[i], win[i]) + buf[i];
1909 buf[i] = MULL(out[i + 18], win[i + 18]);
1910 out_ptr += SBLIMIT;
1911 }
1912 ptr += 18;
1913 buf += 18;
1914 }
1915 for(j=mdct_long_end;j<sblimit;j++) {
1916 for(i=0;i<6;i++) {
1917 out[i] = 0;
1918 out[6 + i] = 0;
1919 out[30+i] = 0;
1920 }
1921 /* select frequency inversion */
1922 win = mdct_win[2] + ((4 * 36) & -(j & 1));
1923 buf2 = out + 6;
1924 for(k=0;k<3;k++) {
1925 /* reorder input for short mdct */
1926 ptr1 = ptr + k;
1927 for(i=0;i<6;i++) {
1928 in[i] = *ptr1;
1929 ptr1 += 3;
1930 }
1931 imdct12(out2, in);
1932 /* apply 12 point window and do small overlap */
1933 for(i=0;i<6;i++) {
1934 buf2[i] = MULL(out2[i], win[i]) + buf2[i];
1935 buf2[i + 6] = MULL(out2[i + 6], win[i + 6]);
1936 }
1937 buf2 += 6;
1938 }
1939 /* overlap */
1940 out_ptr = sb_samples + j;
1941 for(i=0;i<18;i++) {
1942 *out_ptr = out[i] + buf[i];
1943 buf[i] = out[i + 18];
1944 out_ptr += SBLIMIT;
1945 }
1946 ptr += 18;
1947 buf += 18;
1948 }
1949 /* zero bands */
1950 for(j=sblimit;j<SBLIMIT;j++) {
1951 /* overlap */
1952 out_ptr = sb_samples + j;
1953 for(i=0;i<18;i++) {
1954 *out_ptr = buf[i];
1955 buf[i] = 0;
1956 out_ptr += SBLIMIT;
1957 }
1958 buf += 18;
1959 }
1960 }
1961
1962 #if defined(DEBUG)
1963 void sample_dump(int fnum, INT32 *tab, int n)
1964 {
1965 static FILE *files[16], *f;
1966 char buf[512];
1967 int i;
1968 INT32 v;
1969
1970 f = files[fnum];
1971 if (!f) {
1972 sprintf(buf, "/tmp/out%d.%s.pcm",
1973 fnum,
1974 #ifdef USE_HIGHPRECISION
1975 "hp"
1976 #else
1977 "lp"
1978 #endif
1979 );
1980 f = fopen(buf, "w");
1981 if (!f)
1982 return;
1983 files[fnum] = f;
1984 }
1985
1986 if (fnum == 0) {
1987 static int pos = 0;
1988 printf("pos=%d\n", pos);
1989 for(i=0;i<n;i++) {
1990 printf(" %0.4f", (double)tab[i] / FRAC_ONE);
1991 if ((i % 18) == 17)
1992 printf("\n");
1993 }
1994 pos += n;
1995 }
1996 for(i=0;i<n;i++) {
1997 /* normalize to 23 frac bits */
1998 v = tab[i] << (23 - FRAC_BITS);
1999 fwrite(&v, 1, sizeof(INT32), f);
2000 }
2001 }
2002 #endif
2003
2004
2005 /* main layer3 decoding function */
2006 static int mp_decode_layer3(MPADecodeContext *s)
2007 {
2008 int nb_granules, main_data_begin, private_bits;
2009 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
2010 GranuleDef granules[2][2], *g;
2011 INT16 exponents[576];
2012
2013 /* read side info */
2014 if (s->lsf) {
2015 main_data_begin = get_bits(&s->gb, 8);
2016 if (s->nb_channels == 2)
2017 private_bits = get_bits(&s->gb, 2);
2018 else
2019 private_bits = get_bits(&s->gb, 1);
2020 nb_granules = 1;
2021 } else {
2022 main_data_begin = get_bits(&s->gb, 9);
2023 if (s->nb_channels == 2)
2024 private_bits = get_bits(&s->gb, 3);
2025 else
2026 private_bits = get_bits(&s->gb, 5);
2027 nb_granules = 2;
2028 for(ch=0;ch<s->nb_channels;ch++) {
2029 granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2030 granules[ch][1].scfsi = get_bits(&s->gb, 4);
2031 }
2032 }
2033
2034 for(gr=0;gr<nb_granules;gr++) {
2035 for(ch=0;ch<s->nb_channels;ch++) {
2036 dprintf("gr=%d ch=%d: side_info\n", gr, ch);
2037 g = &granules[ch][gr];
2038 g->part2_3_length = get_bits(&s->gb, 12);
2039 g->big_values = get_bits(&s->gb, 9);
2040 g->global_gain = get_bits(&s->gb, 8);
2041 /* if MS stereo only is selected, we precompute the
2042 1/sqrt(2) renormalization factor */
2043 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2044 MODE_EXT_MS_STEREO)
2045 g->global_gain -= 2;
2046 if (s->lsf)
2047 g->scalefac_compress = get_bits(&s->gb, 9);
2048 else
2049 g->scalefac_compress = get_bits(&s->gb, 4);
2050 blocksplit_flag = get_bits(&s->gb, 1);
2051 if (blocksplit_flag) {
2052 g->block_type = get_bits(&s->gb, 2);
2053 if (g->block_type == 0)
2054 return -1;
2055 g->switch_point = get_bits(&s->gb, 1);
2056 for(i=0;i<2;i++)
2057 g->table_select[i] = get_bits(&s->gb, 5);
2058 for(i=0;i<3;i++)
2059 g->subblock_gain[i] = get_bits(&s->gb, 3);
2060 /* compute huffman coded region sizes */
2061 if (g->block_type == 2)
2062 g->region_size[0] = (36 / 2);
2063 else {
2064 if (s->sample_rate_index <= 2)
2065 g->region_size[0] = (36 / 2);
2066 else if (s->sample_rate_index != 8)
2067 g->region_size[0] = (54 / 2);
2068 else
2069 g->region_size[0] = (108 / 2);
2070 }
2071 g->region_size[1] = (576 / 2);
2072 } else {
2073 int region_address1, region_address2, l;
2074 g->block_type = 0;
2075 g->switch_point = 0;
2076 for(i=0;i<3;i++)
2077 g->table_select[i] = get_bits(&s->gb, 5);
2078 /* compute huffman coded region sizes */
2079 region_address1 = get_bits(&s->gb, 4);
2080 region_address2 = get_bits(&s->gb, 3);
2081 dprintf("region1=%d region2=%d\n",
2082 region_address1, region_address2);
2083 g->region_size[0] =
2084 band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2085 l = region_address1 + region_address2 + 2;
2086 /* should not overflow */
2087 if (l > 22)
2088 l = 22;
2089 g->region_size[1] =
2090 band_index_long[s->sample_rate_index][l] >> 1;
2091 }
2092 /* convert region offsets to region sizes and truncate
2093 size to big_values */
2094 g->region_size[2] = (576 / 2);
2095 j = 0;
2096 for(i=0;i<3;i++) {
2097 k = g->region_size[i];
2098 if (k > g->big_values)
2099 k = g->big_values;
2100 g->region_size[i] = k - j;
2101 j = k;
2102 }
2103
2104 /* compute band indexes */
2105 if (g->block_type == 2) {
2106 if (g->switch_point) {
2107 /* if switched mode, we handle the 36 first samples as
2108 long blocks. For 8000Hz, we handle the 48 first
2109 exponents as long blocks (XXX: check this!) */
2110 if (s->sample_rate_index <= 2)
2111 g->long_end = 8;
2112 else if (s->sample_rate_index != 8)
2113 g->long_end = 6;
2114 else
2115 g->long_end = 4; /* 8000 Hz */
2116
2117 if (s->sample_rate_index != 8)
2118 g->short_start = 3;
2119 else
2120 g->short_start = 2;
2121 } else {
2122 g->long_end = 0;
2123 g->short_start = 0;
2124 }
2125 } else {
2126 g->short_start = 13;
2127 g->long_end = 22;
2128 }
2129
2130 g->preflag = 0;
2131 if (!s->lsf)
2132 g->preflag = get_bits(&s->gb, 1);
2133 g->scalefac_scale = get_bits(&s->gb, 1);
2134 g->count1table_select = get_bits(&s->gb, 1);
2135 dprintf("block_type=%d switch_point=%d\n",
2136 g->block_type, g->switch_point);
2137 }
2138 }
2139
2140 /* now we get bits from the main_data_begin offset */
2141 dprintf("seekback: %d\n", main_data_begin);
2142 seek_to_maindata(s, main_data_begin);
2143
2144 for(gr=0;gr<nb_granules;gr++) {
2145 for(ch=0;ch<s->nb_channels;ch++) {
2146 g = &granules[ch][gr];
2147
2148 bits_pos = get_bits_count(&s->gb);
2149
2150 if (!s->lsf) {
2151 UINT8 *sc;
2152 int slen, slen1, slen2;
2153
2154 /* MPEG1 scale factors */
2155 slen1 = slen_table[0][g->scalefac_compress];
2156 slen2 = slen_table[1][g->scalefac_compress];
2157 dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2158 if (g->block_type == 2) {
2159 n = g->switch_point ? 17 : 18;
2160 j = 0;
2161 for(i=0;i<n;i++)
2162 g->scale_factors[j++] = get_bitsz(&s->gb, slen1);
2163 for(i=0;i<18;i++)
2164 g->scale_factors[j++] = get_bitsz(&s->gb, slen2);
2165 for(i=0;i<3;i++)
2166 g->scale_factors[j++] = 0;
2167 } else {
2168 sc = granules[ch][0].scale_factors;
2169 j = 0;
2170 for(k=0;k<4;k++) {
2171 n = (k == 0 ? 6 : 5);
2172 if ((g->scfsi & (0x8 >> k)) == 0) {
2173 slen = (k < 2) ? slen1 : slen2;
2174 for(i=0;i<n;i++)
2175 g->scale_factors[j++] = get_bitsz(&s->gb, slen);
2176 } else {
2177 /* simply copy from last granule */
2178 for(i=0;i<n;i++) {
2179 g->scale_factors[j] = sc[j];
2180 j++;
2181 }
2182 }
2183 }
2184 g->scale_factors[j++] = 0;
2185 }
2186 #if defined(DEBUG)
2187 {
2188 printf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2189 g->scfsi, gr, ch);
2190 for(i=0;i<j;i++)
2191 printf(" %d", g->scale_factors[i]);
2192 printf("\n");
2193 }
2194 #endif
2195 } else {
2196 int tindex, tindex2, slen[4], sl, sf;
2197
2198 /* LSF scale factors */
2199 if (g->block_type == 2) {
2200 tindex = g->switch_point ? 2 : 1;
2201 } else {
2202 tindex = 0;
2203 }
2204 sf = g->scalefac_compress;
2205 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2206 /* intensity stereo case */
2207 sf >>= 1;
2208 if (sf < 180) {
2209 lsf_sf_expand(slen, sf, 6, 6, 0);
2210 tindex2 = 3;
2211 } else if (sf < 244) {
2212 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2213 tindex2 = 4;
2214 } else {
2215 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2216 tindex2 = 5;
2217 }
2218 } else {
2219 /* normal case */
2220 if (sf < 400) {
2221 lsf_sf_expand(slen, sf, 5, 4, 4);
2222 tindex2 = 0;
2223 } else if (sf < 500) {
2224 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2225 tindex2 = 1;
2226 } else {
2227 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2228 tindex2 = 2;
2229 g->preflag = 1;
2230 }
2231 }
2232
2233 j = 0;
2234 for(k=0;k<4;k++) {
2235 n = lsf_nsf_table[tindex2][tindex][k];
2236 sl = slen[k];
2237 for(i=0;i<n;i++)
2238 g->scale_factors[j++] = get_bitsz(&s->gb, sl);
2239 }
2240 /* XXX: should compute exact size */
2241 for(;j<40;j++)
2242 g->scale_factors[j] = 0;
2243 #if defined(DEBUG)
2244 {
2245 printf("gr=%d ch=%d scale_factors:\n",
2246 gr, ch);
2247 for(i=0;i<40;i++)
2248 printf(" %d", g->scale_factors[i]);
2249 printf("\n");
2250 }
2251 #endif
2252 }
2253
2254 exponents_from_scale_factors(s, g, exponents);
2255
2256 /* read Huffman coded residue */
2257 if (huffman_decode(s, g, exponents,
2258 bits_pos + g->part2_3_length) < 0)
2259 return -1;
2260 #if defined(DEBUG)
2261 sample_dump(0, g->sb_hybrid, 576);
2262 #endif
2263
2264 /* skip extension bits */
2265 bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2266 if (bits_left < 0) {
2267 dprintf("bits_left=%d\n", bits_left);
2268 return -1;
2269 }
2270 while (bits_left >= 16) {
2271 skip_bits(&s->gb, 16);
2272 bits_left -= 16;
2273 }
2274 if (bits_left > 0)
2275 skip_bits(&s->gb, bits_left);
2276 } /* ch */
2277
2278 if (s->nb_channels == 2)
2279 compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2280
2281 for(ch=0;ch<s->nb_channels;ch++) {
2282 g = &granules[ch][gr];
2283
2284 reorder_block(s, g);
2285 #if defined(DEBUG)
2286 sample_dump(0, g->sb_hybrid, 576);
2287 #endif
2288 compute_antialias(s, g);
2289 #if defined(DEBUG)
2290 sample_dump(1, g->sb_hybrid, 576);
2291 #endif
2292 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2293 #if defined(DEBUG)
2294 sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2295 #endif
2296 }
2297 } /* gr */
2298 return nb_granules * 18;
2299 }
2300
2301 static int mp_decode_frame(MPADecodeContext *s,
2302 short *samples)
2303 {
2304 int i, nb_frames, ch;
2305 short *samples_ptr;
2306
2307 init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
2308 s->inbuf_ptr - s->inbuf - HEADER_SIZE);
2309
2310 /* skip error protection field */
2311 if (s->error_protection)
2312 get_bits(&s->gb, 16);
2313
2314 dprintf("frame %d:\n", s->frame_count);
2315 switch(s->layer) {
2316 case 1:
2317 nb_frames = mp_decode_layer1(s);
2318 break;
2319 case 2:
2320 nb_frames = mp_decode_layer2(s);
2321 break;
2322 case 3:
2323 default:
2324 nb_frames = mp_decode_layer3(s);
2325 break;
2326 }
2327 #if defined(DEBUG)
2328 for(i=0;i<nb_frames;i++) {
2329 for(ch=0;ch<s->nb_channels;ch++) {
2330 int j;
2331 printf("%d-%d:", i, ch);
2332 for(j=0;j<SBLIMIT;j++)
2333 printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2334 printf("\n");
2335 }
2336 }
2337 #endif
2338 /* apply the synthesis filter */
2339 for(ch=0;ch<s->nb_channels;ch++) {
2340 samples_ptr = samples + ch;
2341 for(i=0;i<nb_frames;i++) {
2342 synth_filter(s, ch, samples_ptr, s->nb_channels,
2343 s->sb_samples[ch][i]);
2344 samples_ptr += 32 * s->nb_channels;
2345 }
2346 }
2347 #ifdef DEBUG
2348 s->frame_count++;
2349 #endif
2350 return nb_frames * 32 * sizeof(short) * s->nb_channels;
2351 }
2352
2353 static int decode_frame(AVCodecContext * avctx,
2354 void *data, int *data_size,
2355 UINT8 * buf, int buf_size)
2356 {
2357 MPADecodeContext *s = avctx->priv_data;
2358 UINT32 header;
2359 UINT8 *buf_ptr;
2360 int len, out_size;
2361 short *out_samples = data;
2362
2363 *data_size = 0;
2364 buf_ptr = buf;
2365 while (buf_size > 0) {
2366 len = s->inbuf_ptr - s->inbuf;
2367 if (s->frame_size == 0) {
2368 /* special case for next header for first frame in free
2369 format case (XXX: find a simpler method) */
2370 if (s->free_format_next_header != 0) {
2371 s->inbuf[0] = s->free_format_next_header >> 24;
2372 s->inbuf[1] = s->free_format_next_header >> 16;
2373 s->inbuf[2] = s->free_format_next_header >> 8;
2374 s->inbuf[3] = s->free_format_next_header;
2375 s->inbuf_ptr = s->inbuf + 4;
2376 s->free_format_next_header = 0;
2377 goto got_header;
2378 }
2379 /* no header seen : find one. We need at least HEADER_SIZE
2380 bytes to parse it */
2381 len = HEADER_SIZE - len;
2382 if (len > buf_size)
2383 len = buf_size;
2384 if (len > 0) {
2385 memcpy(s->inbuf_ptr, buf_ptr, len);
2386 buf_ptr += len;
2387 buf_size -= len;
2388 s->inbuf_ptr += len;
2389 }
2390 if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
2391 got_header:
2392 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2393 (s->inbuf[2] << 8) | s->inbuf[3];
2394
2395 if (check_header(header) < 0) {
2396 /* no sync found : move by one byte (inefficient, but simple!) */
2397 memcpy(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2398 s->inbuf_ptr--;
2399 dprintf("skip %x\n", header);
2400 /* reset free format frame size to give a chance
2401 to get a new bitrate */
2402 s->free_format_frame_size = 0;
2403 } else {
2404 if (decode_header(s, header) == 1) {
2405 /* free format: prepare to compute frame size */
2406 s->frame_size = -1;
2407 }
2408 /* update codec info */
2409 avctx->sample_rate = s->sample_rate;
2410 avctx->channels = s->nb_channels;
2411 avctx->bit_rate = s->bit_rate;
2412 avctx->frame_size = s->frame_size;
2413 }
2414 }
2415 } else if (s->frame_size == -1) {
2416 /* free format : find next sync to compute frame size */
2417 len = MPA_MAX_CODED_FRAME_SIZE - len;
2418 if (len > buf_size)
2419 len = buf_size;
2420 if (len == 0) {
2421 /* frame too long: resync */
2422 s->frame_size = 0;
2423 } else {
2424 UINT8 *p, *pend;
2425 UINT32 header1;
2426 int padding;
2427
2428 memcpy(s->inbuf_ptr, buf_ptr, len);
2429 /* check for header */
2430 p = s->inbuf_ptr - 3;
2431 pend = s->inbuf_ptr + len - 4;
2432 while (p <= pend) {
2433 header = (p[0] << 24) | (p[1] << 16) |
2434 (p[2] << 8) | p[3];
2435 header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2436 (s->inbuf[2] << 8) | s->inbuf[3];
2437 /* check with high probability that we have a
2438 valid header */
2439 if ((header & SAME_HEADER_MASK) ==
2440 (header1 & SAME_HEADER_MASK)) {
2441 /* header found: update pointers */
2442 len = (p + 4) - s->inbuf_ptr;
2443 buf_ptr += len;
2444 buf_size -= len;
2445 s->inbuf_ptr = p;
2446 /* compute frame size */
2447 s->free_format_next_header = header;
2448 s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2449 padding = (header1 >> 9) & 1;
2450 if (s->layer == 1)
2451 s->free_format_frame_size -= padding * 4;
2452 else
2453 s->free_format_frame_size -= padding;
2454 dprintf("free frame size=%d padding=%d\n",
2455 s->free_format_frame_size, padding);
2456 decode_header(s, header1);
2457 goto next_data;
2458 }
2459 p++;
2460 }
2461 /* not found: simply increase pointers */
2462 buf_ptr += len;
2463 s->inbuf_ptr += len;
2464 buf_size -= len;
2465 }
2466 } else if (len < s->frame_size) {
2467 if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2468 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
2469 len = s->frame_size - len;
2470 if (len > buf_size)
2471 len = buf_size;
2472 memcpy(s->inbuf_ptr, buf_ptr, len);
2473 buf_ptr += len;
2474 s->inbuf_ptr += len;
2475 buf_size -= len;
2476 } else {
2477 out_size = mp_decode_frame(s, out_samples);
2478 s->inbuf_ptr = s->inbuf;
2479 s->frame_size = 0;
2480 *data_size = out_size;
2481 break;
2482 }
2483 next_data:
2484 }
2485 return buf_ptr - buf;
2486 }
2487
2488 AVCodec mp2_decoder =
2489 {
2490 "mp2",
2491 CODEC_TYPE_AUDIO,
2492 CODEC_ID_MP2,
2493 sizeof(MPADecodeContext),
2494 decode_init,
2495 NULL,
2496 NULL,
2497 decode_frame,
2498 };
2499
2500 AVCodec mp3_decoder =
2501 {
2502 "mp3",
2503 CODEC_TYPE_AUDIO,
2504 CODEC_ID_MP3LAME,
2505 sizeof(MPADecodeContext),
2506 decode_init,
2507 NULL,
2508 NULL,
2509 decode_frame,
2510 };
2511
2512 #undef C1
2513 #undef C2
2514 #undef C3
2515 #undef C4
2516 #undef C5
2517 #undef C6
2518 #undef C7
2519 #undef C8
2520 #undef FRAC_BITS
2521 #undef HEADER_SIZE