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 * Modified heavily by Matt Campbell <mattcampbell@pobox.com> for the
20 * stand-alone mpaudec library. Based on mpegaudiodec.c from libavcodec.
21 */
22
23 /*#define DEBUG*/
24 #include "internal.h"
25 #include "mpegaudio.h"
26
27 #ifdef _MSC_VER
28 #pragma warning(disable : 4244)
29 #endif
30
31 /*
32 * TODO:
33 * - in low precision mode, use more 16 bit multiplies in synth filter
34 * - test lsf / mpeg25 extensively.
35 */
36
37 /* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
38 audio decoder */
39 #define USE_HIGHPRECISION
40
41 #ifdef USE_HIGHPRECISION
42 #define FRAC_BITS 23 /* fractional bits for sb_samples and dct */
43 #define WFRAC_BITS 16 /* fractional bits for window */
44 #else
45 #define FRAC_BITS 15 /* fractional bits for sb_samples and dct */
46 #define WFRAC_BITS 14 /* fractional bits for window */
47 #endif
48
49 #define FRAC_ONE (1 << FRAC_BITS)
50
51 #define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
52 #define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))
53 #define FIX(a) ((int)((a) * FRAC_ONE))
54 /* WARNING: only correct for posititive numbers */
55 #define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
56 #define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
57
58 #if FRAC_BITS <= 15
59 typedef int16_t MPA_INT;
60 #else
61 typedef int32_t MPA_INT;
62 #endif
63
64 /****************/
65
66 #define HEADER_SIZE 4
67 #define BACKSTEP_SIZE 512
68
69 typedef struct MPADecodeContext {
70 uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE]; /* input buffer */
71 int inbuf_index;
72 uint8_t *inbuf_ptr, *inbuf;
73 int frame_size;
74 int free_format_frame_size; /* frame size in case of free format
75 (zero if currently unknown) */
76 /* next header (used in free format parsing) */
77 int error_protection;
78 int layer;
79 int sample_rate;
80 int sample_rate_index; /* between 0 and 8 */
81 int bit_rate;
82 int old_frame_size;
83 GetBitContext gb;
84 int nb_channels;
85 int mode;
86 int mode_ext;
87 int lsf;
88 MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2];
89 int synth_buf_offset[MPA_MAX_CHANNELS];
90 int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT];
91 int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
92 #ifdef DEBUG
93 int frame_count;
94 #endif
95 } MPADecodeContext;
96
97 /* layer 3 "granule" */
98 typedef struct GranuleDef {
99 uint8_t scfsi;
100 int part2_3_length;
101 int big_values;
102 int global_gain;
103 int scalefac_compress;
104 uint8_t block_type;
105 uint8_t switch_point;
106 int table_select[3];
107 int subblock_gain[3];
108 uint8_t scalefac_scale;
109 uint8_t count1table_select;
110 int region_size[3]; /* number of huffman codes in each region */
111 int preflag;
112 int short_start, long_end; /* long/short band indexes */
113 uint8_t scale_factors[40];
114 int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
115 } GranuleDef;
116
117 #define MODE_EXT_MS_STEREO 2
118 #define MODE_EXT_I_STEREO 1
119
120 /* layer 3 huffman tables */
121 typedef struct HuffTable {
122 int xsize;
123 const uint8_t *bits;
124 const uint16_t *codes;
125 } HuffTable;
126
127 #include "mpaudectab.h"
128
129 /* vlc structure for decoding layer 3 huffman tables */
130 static VLC huff_vlc[16];
131 static uint8_t *huff_code_table[16];
132 static VLC huff_quad_vlc[2];
133 /* computed from band_size_long */
134 static uint16_t band_index_long[9][23];
135 /* XXX: free when all decoders are closed */
136 #define TABLE_4_3_SIZE (8191 + 16)
137 static int8_t table_4_3_exp[TABLE_4_3_SIZE];
138 #if FRAC_BITS <= 15
139 static uint16_t table_4_3_value[TABLE_4_3_SIZE];
140 #else
141 static uint32_t table_4_3_value[TABLE_4_3_SIZE];
142 #endif
143 /* intensity stereo coef table */
144 static int32_t is_table[2][16];
145 static int32_t is_table_lsf[2][2][16];
146 static int32_t csa_table[8][2];
147 static int32_t mdct_win[8][36];
148
149 /* lower 2 bits: modulo 3, higher bits: shift */
150 static uint16_t scale_factor_modshift[64];
151 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
152 static int32_t scale_factor_mult[15][3];
153 /* mult table for layer 2 group quantization */
154
155 #define SCALE_GEN(v) \
156 { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
157
158 static int32_t scale_factor_mult2[3][3] = {
159 SCALE_GEN(4.0 / 3.0), /* 3 steps */
160 SCALE_GEN(4.0 / 5.0), /* 5 steps */
161 SCALE_GEN(4.0 / 9.0), /* 9 steps */
162 };
163
164 /* 2^(n/4) */
165 static uint32_t scale_factor_mult3[4] = {
166 FIXR(1.0),
167 FIXR(1.18920711500272106671),
168 FIXR(1.41421356237309504880),
169 FIXR(1.68179283050742908605),
170 };
171
172 static MPA_INT window[512];
173
174 /* layer 1 unscaling */
175 /* n = number of bits of the mantissa minus 1 */
l1_unscale(int n,int mant,int scale_factor)176 static int l1_unscale(int n, int mant, int scale_factor)
177 {
178 int shift, mod;
179 int64_t val;
180
181 shift = scale_factor_modshift[scale_factor];
182 mod = shift & 3;
183 shift >>= 2;
184 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
185 shift += n;
186 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
187 return (int)((val + ((int64_t)(1) << (shift - 1))) >> shift);
188 }
189
l2_unscale_group(int steps,int mant,int scale_factor)190 static int l2_unscale_group(int steps, int mant, int scale_factor)
191 {
192 int shift, mod, val;
193
194 shift = scale_factor_modshift[scale_factor];
195 mod = shift & 3;
196 shift >>= 2;
197
198 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
199 /* NOTE: at this point, 0 <= shift <= 21 */
200 if (shift > 0)
201 val = (val + (1 << (shift - 1))) >> shift;
202 return val;
203 }
204
205 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
l3_unscale(int value,int exponent)206 static int l3_unscale(int value, int exponent)
207 {
208 #if FRAC_BITS <= 15
209 unsigned int m;
210 #else
211 uint64_t m;
212 #endif
213 int e;
214
215 e = table_4_3_exp[value];
216 e += (exponent >> 2);
217 e = FRAC_BITS - e;
218 #if FRAC_BITS <= 15
219 if (e > 31)
220 e = 31;
221 #endif
222 m = table_4_3_value[value];
223 #if FRAC_BITS <= 15
224 m = (m * scale_factor_mult3[exponent & 3]);
225 m = (m + (1 << (e-1))) >> e;
226 return m;
227 #else
228 m = MUL64(m, scale_factor_mult3[exponent & 3]);
229 m = (m + ((uint64_t)(1) << (e-1))) >> e;
230 return (int)m;
231 #endif
232 }
233
234 /* all integer n^(4/3) computation code */
235 #define DEV_ORDER 13
236
237 #define POW_FRAC_BITS 24
238 #define POW_FRAC_ONE (1 << POW_FRAC_BITS)
239 #define POW_FIX(a) ((int)((a) * POW_FRAC_ONE))
240 #define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
241
242 static int dev_4_3_coefs[DEV_ORDER];
243
244 static int pow_mult3[3] = {
245 POW_FIX(1.0),
246 POW_FIX(1.25992104989487316476),
247 POW_FIX(1.58740105196819947474),
248 };
249
int_pow_init(void)250 static void int_pow_init(void)
251 {
252 int i, a;
253
254 a = POW_FIX(1.0);
255 for(i=0;i<DEV_ORDER;i++) {
256 a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
257 dev_4_3_coefs[i] = a;
258 }
259 }
260
261 /* return the mantissa and the binary exponent */
int_pow(int i,int * exp_ptr)262 static int int_pow(int i, int *exp_ptr)
263 {
264 int e, er, eq, j;
265 int a, a1;
266
267 /* renormalize */
268 a = i;
269 e = POW_FRAC_BITS;
270 while (a < (1 << (POW_FRAC_BITS - 1))) {
271 a = a << 1;
272 e--;
273 }
274 a -= (1 << POW_FRAC_BITS);
275 a1 = 0;
276 for(j = DEV_ORDER - 1; j >= 0; j--)
277 a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
278 a = (1 << POW_FRAC_BITS) + a1;
279 /* exponent compute (exact) */
280 e = e * 4;
281 er = e % 3;
282 eq = e / 3;
283 a = POW_MULL(a, pow_mult3[er]);
284 while (a >= 2 * POW_FRAC_ONE) {
285 a = a >> 1;
286 eq++;
287 }
288 /* convert to float */
289 while (a < POW_FRAC_ONE) {
290 a = a << 1;
291 eq--;
292 }
293 /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
294 #if POW_FRAC_BITS > FRAC_BITS
295 a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
296 /* correct overflow */
297 if (a >= 2 * (1 << FRAC_BITS)) {
298 a = a >> 1;
299 eq++;
300 }
301 #endif
302 *exp_ptr = eq;
303 return a;
304 }
305
mpaudec_init(MPAuDecContext * mpctx)306 int mpaudec_init(MPAuDecContext * mpctx)
307 {
308 MPADecodeContext *s;
309 static int init=0;
310 int i, j, k;
311 assert(mpctx != NULL);
312 memset(mpctx, 0, sizeof(MPAuDecContext));
313 mpctx->priv_data = calloc(1, sizeof(MPADecodeContext));
314 if (mpctx->priv_data == NULL)
315 return -1;
316 s = mpctx->priv_data;
317
318 if (!init && !mpctx->parse_only) {
319 /* scale factors table for layer 1/2 */
320 for(i=0;i<64;i++) {
321 int shift, mod;
322 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
323 shift = (i / 3);
324 mod = i % 3;
325 scale_factor_modshift[i] = mod | (shift << 2);
326 }
327
328 /* scale factor multiply for layer 1 */
329 for(i=0;i<15;i++) {
330 int n, norm;
331 n = i + 2;
332 norm = (((int64_t)(1) << n) * FRAC_ONE) / ((1 << n) - 1);
333 scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
334 scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
335 scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
336 #ifdef DEBUG
337 printf("%d: norm=%x s=%x %x %x\n",
338 i, norm,
339 scale_factor_mult[i][0],
340 scale_factor_mult[i][1],
341 scale_factor_mult[i][2]);
342 #endif
343 }
344
345 /* window */
346 /* max = 18760, max sum over all 16 coefs : 44736 */
347 for(i=0;i<257;i++) {
348 int v;
349 v = mpa_enwindow[i];
350 #if WFRAC_BITS < 16
351 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
352 #endif
353 window[i] = v;
354 if ((i & 63) != 0)
355 v = -v;
356 if (i != 0)
357 window[512 - i] = v;
358 }
359
360 /* huffman decode tables */
361 huff_code_table[0] = NULL;
362 for(i=1;i<16;i++) {
363 const HuffTable *h = &mpa_huff_tables[i];
364 int xsize, x, y;
365 unsigned int n;
366 uint8_t *code_table;
367
368 xsize = h->xsize;
369 n = xsize * xsize;
370 /* XXX: fail test */
371 init_vlc(&huff_vlc[i], 8, n,
372 h->bits, 1, 1, h->codes, 2, 2);
373
374 code_table = calloc(n, 1);
375 j = 0;
376 for(x=0;x<xsize;x++) {
377 for(y=0;y<xsize;y++)
378 code_table[j++] = (x << 4) | y;
379 }
380 huff_code_table[i] = code_table;
381 }
382 for(i=0;i<2;i++) {
383 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
384 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1);
385 }
386
387 for(i=0;i<9;i++) {
388 k = 0;
389 for(j=0;j<22;j++) {
390 band_index_long[i][j] = k;
391 k += band_size_long[i][j];
392 }
393 band_index_long[i][22] = k;
394 }
395
396 /* compute n ^ (4/3) and store it in mantissa/exp format */
397 int_pow_init();
398 for(i=1;i<TABLE_4_3_SIZE;i++) {
399 int e, m;
400 m = int_pow(i, &e);
401 /* normalized to FRAC_BITS */
402 table_4_3_value[i] = m;
403 table_4_3_exp[i] = e;
404 }
405
406 for(i=0;i<7;i++) {
407 float f;
408 int v;
409 if (i != 6) {
410 f = tan((double)i * M_PI / 12.0);
411 v = FIXR(f / (1.0 + f));
412 } else {
413 v = FIXR(1.0);
414 }
415 is_table[0][i] = v;
416 is_table[1][6 - i] = v;
417 }
418 /* invalid values */
419 for(i=7;i<16;i++)
420 is_table[0][i] = is_table[1][i] = 0.0;
421
422 for(i=0;i<16;i++) {
423 double f;
424 int e, k;
425
426 for(j=0;j<2;j++) {
427 e = -(j + 1) * ((i + 1) >> 1);
428 f = pow(2.0, e / 4.0);
429 k = i & 1;
430 is_table_lsf[j][k ^ 1][i] = FIXR(f);
431 is_table_lsf[j][k][i] = FIXR(1.0);
432 #ifdef DEBUG
433 printf("is_table_lsf %d %d: %x %x\n",
434 i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
435 #endif
436 }
437 }
438
439 for(i=0;i<8;i++) {
440 float ci, cs, ca;
441 ci = ci_table[i];
442 cs = 1.0 / sqrt(1.0 + ci * ci);
443 ca = cs * ci;
444 csa_table[i][0] = FIX(cs);
445 csa_table[i][1] = FIX(ca);
446 }
447
448 /* compute mdct windows */
449 for(i=0;i<36;i++) {
450 int v;
451 v = FIXR(sin(M_PI * (i + 0.5) / 36.0));
452 mdct_win[0][i] = v;
453 mdct_win[1][i] = v;
454 mdct_win[3][i] = v;
455 }
456 for(i=0;i<6;i++) {
457 mdct_win[1][18 + i] = FIXR(1.0);
458 mdct_win[1][24 + i] = FIXR(sin(M_PI * ((i + 6) + 0.5) / 12.0));
459 mdct_win[1][30 + i] = FIXR(0.0);
460
461 mdct_win[3][i] = FIXR(0.0);
462 mdct_win[3][6 + i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));
463 mdct_win[3][12 + i] = FIXR(1.0);
464 }
465
466 for(i=0;i<12;i++)
467 mdct_win[2][i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));
468
469 /* NOTE: we do frequency inversion adter the MDCT by changing
470 the sign of the right window coefs */
471 for(j=0;j<4;j++) {
472 for(i=0;i<36;i+=2) {
473 mdct_win[j + 4][i] = mdct_win[j][i];
474 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
475 }
476 }
477
478 #if defined(DEBUG)
479 for(j=0;j<8;j++) {
480 printf("win%d=\n", j);
481 for(i=0;i<36;i++)
482 printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);
483 printf("\n");
484 }
485 #endif
486 init = 1;
487 }
488
489 s->inbuf_index = 0;
490 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
491 s->inbuf_ptr = s->inbuf;
492 #ifdef DEBUG
493 s->frame_count = 0;
494 #endif
495 return 0;
496 }
497
498 /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
499
500 /* cos(i*pi/64) */
501
502 #define COS0_0 FIXR(0.50060299823519630134)
503 #define COS0_1 FIXR(0.50547095989754365998)
504 #define COS0_2 FIXR(0.51544730992262454697)
505 #define COS0_3 FIXR(0.53104259108978417447)
506 #define COS0_4 FIXR(0.55310389603444452782)
507 #define COS0_5 FIXR(0.58293496820613387367)
508 #define COS0_6 FIXR(0.62250412303566481615)
509 #define COS0_7 FIXR(0.67480834145500574602)
510 #define COS0_8 FIXR(0.74453627100229844977)
511 #define COS0_9 FIXR(0.83934964541552703873)
512 #define COS0_10 FIXR(0.97256823786196069369)
513 #define COS0_11 FIXR(1.16943993343288495515)
514 #define COS0_12 FIXR(1.48416461631416627724)
515 #define COS0_13 FIXR(2.05778100995341155085)
516 #define COS0_14 FIXR(3.40760841846871878570)
517 #define COS0_15 FIXR(10.19000812354805681150)
518
519 #define COS1_0 FIXR(0.50241928618815570551)
520 #define COS1_1 FIXR(0.52249861493968888062)
521 #define COS1_2 FIXR(0.56694403481635770368)
522 #define COS1_3 FIXR(0.64682178335999012954)
523 #define COS1_4 FIXR(0.78815462345125022473)
524 #define COS1_5 FIXR(1.06067768599034747134)
525 #define COS1_6 FIXR(1.72244709823833392782)
526 #define COS1_7 FIXR(5.10114861868916385802)
527
528 #define COS2_0 FIXR(0.50979557910415916894)
529 #define COS2_1 FIXR(0.60134488693504528054)
530 #define COS2_2 FIXR(0.89997622313641570463)
531 #define COS2_3 FIXR(2.56291544774150617881)
532
533 #define COS3_0 FIXR(0.54119610014619698439)
534 #define COS3_1 FIXR(1.30656296487637652785)
535
536 #define COS4_0 FIXR(0.70710678118654752439)
537
538 /* butterfly operator */
539 #define BF(a, b, c)\
540 {\
541 tmp0 = tab[a] + tab[b];\
542 tmp1 = tab[a] - tab[b];\
543 tab[a] = tmp0;\
544 tab[b] = MULL(tmp1, c);\
545 }
546
547 #define BF1(a, b, c, d)\
548 {\
549 BF(a, b, COS4_0);\
550 BF(c, d, -COS4_0);\
551 tab[c] += tab[d];\
552 }
553
554 #define BF2(a, b, c, d)\
555 {\
556 BF(a, b, COS4_0);\
557 BF(c, d, -COS4_0);\
558 tab[c] += tab[d];\
559 tab[a] += tab[c];\
560 tab[c] += tab[b];\
561 tab[b] += tab[d];\
562 }
563
564 #define ADD(a, b) tab[a] += tab[b]
565
566 /* DCT32 without 1/sqrt(2) coef zero scaling. */
dct32(int32_t * out,int32_t * tab)567 static void dct32(int32_t *out, int32_t *tab)
568 {
569 int tmp0, tmp1;
570
571 /* pass 1 */
572 BF(0, 31, COS0_0);
573 BF(1, 30, COS0_1);
574 BF(2, 29, COS0_2);
575 BF(3, 28, COS0_3);
576 BF(4, 27, COS0_4);
577 BF(5, 26, COS0_5);
578 BF(6, 25, COS0_6);
579 BF(7, 24, COS0_7);
580 BF(8, 23, COS0_8);
581 BF(9, 22, COS0_9);
582 BF(10, 21, COS0_10);
583 BF(11, 20, COS0_11);
584 BF(12, 19, COS0_12);
585 BF(13, 18, COS0_13);
586 BF(14, 17, COS0_14);
587 BF(15, 16, COS0_15);
588
589 /* pass 2 */
590 BF(0, 15, COS1_0);
591 BF(1, 14, COS1_1);
592 BF(2, 13, COS1_2);
593 BF(3, 12, COS1_3);
594 BF(4, 11, COS1_4);
595 BF(5, 10, COS1_5);
596 BF(6, 9, COS1_6);
597 BF(7, 8, COS1_7);
598
599 BF(16, 31, -COS1_0);
600 BF(17, 30, -COS1_1);
601 BF(18, 29, -COS1_2);
602 BF(19, 28, -COS1_3);
603 BF(20, 27, -COS1_4);
604 BF(21, 26, -COS1_5);
605 BF(22, 25, -COS1_6);
606 BF(23, 24, -COS1_7);
607
608 /* pass 3 */
609 BF(0, 7, COS2_0);
610 BF(1, 6, COS2_1);
611 BF(2, 5, COS2_2);
612 BF(3, 4, COS2_3);
613
614 BF(8, 15, -COS2_0);
615 BF(9, 14, -COS2_1);
616 BF(10, 13, -COS2_2);
617 BF(11, 12, -COS2_3);
618
619 BF(16, 23, COS2_0);
620 BF(17, 22, COS2_1);
621 BF(18, 21, COS2_2);
622 BF(19, 20, COS2_3);
623
624 BF(24, 31, -COS2_0);
625 BF(25, 30, -COS2_1);
626 BF(26, 29, -COS2_2);
627 BF(27, 28, -COS2_3);
628
629 /* pass 4 */
630 BF(0, 3, COS3_0);
631 BF(1, 2, COS3_1);
632
633 BF(4, 7, -COS3_0);
634 BF(5, 6, -COS3_1);
635
636 BF(8, 11, COS3_0);
637 BF(9, 10, COS3_1);
638
639 BF(12, 15, -COS3_0);
640 BF(13, 14, -COS3_1);
641
642 BF(16, 19, COS3_0);
643 BF(17, 18, COS3_1);
644
645 BF(20, 23, -COS3_0);
646 BF(21, 22, -COS3_1);
647
648 BF(24, 27, COS3_0);
649 BF(25, 26, COS3_1);
650
651 BF(28, 31, -COS3_0);
652 BF(29, 30, -COS3_1);
653
654 /* pass 5 */
655 BF1(0, 1, 2, 3);
656 BF2(4, 5, 6, 7);
657 BF1(8, 9, 10, 11);
658 BF2(12, 13, 14, 15);
659 BF1(16, 17, 18, 19);
660 BF2(20, 21, 22, 23);
661 BF1(24, 25, 26, 27);
662 BF2(28, 29, 30, 31);
663
664 /* pass 6 */
665
666 ADD( 8, 12);
667 ADD(12, 10);
668 ADD(10, 14);
669 ADD(14, 9);
670 ADD( 9, 13);
671 ADD(13, 11);
672 ADD(11, 15);
673
674 out[ 0] = tab[0];
675 out[16] = tab[1];
676 out[ 8] = tab[2];
677 out[24] = tab[3];
678 out[ 4] = tab[4];
679 out[20] = tab[5];
680 out[12] = tab[6];
681 out[28] = tab[7];
682 out[ 2] = tab[8];
683 out[18] = tab[9];
684 out[10] = tab[10];
685 out[26] = tab[11];
686 out[ 6] = tab[12];
687 out[22] = tab[13];
688 out[14] = tab[14];
689 out[30] = tab[15];
690
691 ADD(24, 28);
692 ADD(28, 26);
693 ADD(26, 30);
694 ADD(30, 25);
695 ADD(25, 29);
696 ADD(29, 27);
697 ADD(27, 31);
698
699 out[ 1] = tab[16] + tab[24];
700 out[17] = tab[17] + tab[25];
701 out[ 9] = tab[18] + tab[26];
702 out[25] = tab[19] + tab[27];
703 out[ 5] = tab[20] + tab[28];
704 out[21] = tab[21] + tab[29];
705 out[13] = tab[22] + tab[30];
706 out[29] = tab[23] + tab[31];
707 out[ 3] = tab[24] + tab[20];
708 out[19] = tab[25] + tab[21];
709 out[11] = tab[26] + tab[22];
710 out[27] = tab[27] + tab[23];
711 out[ 7] = tab[28] + tab[18];
712 out[23] = tab[29] + tab[19];
713 out[15] = tab[30] + tab[17];
714 out[31] = tab[31];
715 }
716
717 #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
718
719 #if FRAC_BITS <= 15
720
round_sample(int sum)721 static int round_sample(int sum)
722 {
723 int sum1;
724 sum1 = (sum + (1 << (OUT_SHIFT - 1))) >> OUT_SHIFT;
725 if (sum1 < -32768)
726 sum1 = -32768;
727 else if (sum1 > 32767)
728 sum1 = 32767;
729 return sum1;
730 }
731
732 /* signed 16x16 -> 32 multiply add accumulate */
733 #define MACS(rt, ra, rb) rt += (ra) * (rb)
734
735 /* signed 16x16 -> 32 multiply */
736 #define MULS(ra, rb) ((ra) * (rb))
737
738 #else
739
round_sample(int64_t sum)740 static int round_sample(int64_t sum)
741 {
742 int sum1;
743 sum1 = (int)((sum + ((int64_t)(1) << (OUT_SHIFT - 1))) >> OUT_SHIFT);
744 if (sum1 < -32768)
745 sum1 = -32768;
746 else if (sum1 > 32767)
747 sum1 = 32767;
748 return sum1;
749 }
750
751 #define MULS(ra, rb) MUL64(ra, rb)
752
753 #endif
754
755 #define SUM8(sum, op, w, p) \
756 { \
757 sum op MULS((w)[0 * 64], p[0 * 64]);\
758 sum op MULS((w)[1 * 64], p[1 * 64]);\
759 sum op MULS((w)[2 * 64], p[2 * 64]);\
760 sum op MULS((w)[3 * 64], p[3 * 64]);\
761 sum op MULS((w)[4 * 64], p[4 * 64]);\
762 sum op MULS((w)[5 * 64], p[5 * 64]);\
763 sum op MULS((w)[6 * 64], p[6 * 64]);\
764 sum op MULS((w)[7 * 64], p[7 * 64]);\
765 }
766
767 #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
768 { \
769 int tmp;\
770 tmp = p[0 * 64];\
771 sum1 op1 MULS((w1)[0 * 64], tmp);\
772 sum2 op2 MULS((w2)[0 * 64], tmp);\
773 tmp = p[1 * 64];\
774 sum1 op1 MULS((w1)[1 * 64], tmp);\
775 sum2 op2 MULS((w2)[1 * 64], tmp);\
776 tmp = p[2 * 64];\
777 sum1 op1 MULS((w1)[2 * 64], tmp);\
778 sum2 op2 MULS((w2)[2 * 64], tmp);\
779 tmp = p[3 * 64];\
780 sum1 op1 MULS((w1)[3 * 64], tmp);\
781 sum2 op2 MULS((w2)[3 * 64], tmp);\
782 tmp = p[4 * 64];\
783 sum1 op1 MULS((w1)[4 * 64], tmp);\
784 sum2 op2 MULS((w2)[4 * 64], tmp);\
785 tmp = p[5 * 64];\
786 sum1 op1 MULS((w1)[5 * 64], tmp);\
787 sum2 op2 MULS((w2)[5 * 64], tmp);\
788 tmp = p[6 * 64];\
789 sum1 op1 MULS((w1)[6 * 64], tmp);\
790 sum2 op2 MULS((w2)[6 * 64], tmp);\
791 tmp = p[7 * 64];\
792 sum1 op1 MULS((w1)[7 * 64], tmp);\
793 sum2 op2 MULS((w2)[7 * 64], tmp);\
794 }
795
796
797 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
798 32 samples. */
799 /* XXX: optimize by avoiding ring buffer usage */
synth_filter(MPADecodeContext * s1,int ch,int16_t * samples,int incr,int32_t sb_samples[SBLIMIT])800 static void synth_filter(MPADecodeContext *s1,
801 int ch, int16_t *samples, int incr,
802 int32_t sb_samples[SBLIMIT])
803 {
804 int32_t tmp[32];
805 MPA_INT *synth_buf;
806 const MPA_INT *w, *w2, *p;
807 int j, offset, v;
808 int16_t *samples2;
809 #if FRAC_BITS <= 15
810 int32_t sum, sum2;
811 #else
812 int64_t sum, sum2;
813 #endif
814
815 dct32(tmp, sb_samples);
816
817 offset = s1->synth_buf_offset[ch];
818 synth_buf = s1->synth_buf[ch] + offset;
819
820 for(j=0;j<32;j++) {
821 v = tmp[j];
822 #if FRAC_BITS <= 15
823 /* NOTE: can cause a loss in precision if very high amplitude
824 sound */
825 if (v > 32767)
826 v = 32767;
827 else if (v < -32768)
828 v = -32768;
829 #endif
830 synth_buf[j] = v;
831 }
832 /* copy to avoid wrap */
833 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
834
835 samples2 = samples + 31 * incr;
836 w = window;
837 w2 = window + 31;
838
839 sum = 0;
840 p = synth_buf + 16;
841 SUM8(sum, +=, w, p);
842 p = synth_buf + 48;
843 SUM8(sum, -=, w + 32, p);
844 *samples = round_sample(sum);
845 samples += incr;
846 w++;
847
848 /* we calculate two samples at the same time to avoid one memory
849 access per two sample */
850 for(j=1;j<16;j++) {
851 sum = 0;
852 sum2 = 0;
853 p = synth_buf + 16 + j;
854 SUM8P2(sum, +=, sum2, -=, w, w2, p);
855 p = synth_buf + 48 - j;
856 SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
857
858 *samples = round_sample(sum);
859 samples += incr;
860 *samples2 = round_sample(sum2);
861 samples2 -= incr;
862 w++;
863 w2--;
864 }
865
866 p = synth_buf + 32;
867 sum = 0;
868 SUM8(sum, -=, w + 32, p);
869 *samples = round_sample(sum);
870
871 offset = (offset - 32) & 511;
872 s1->synth_buf_offset[ch] = offset;
873 }
874
875 /* cos(pi*i/24) */
876 #define C1 FIXR(0.99144486137381041114)
877 #define C3 FIXR(0.92387953251128675612)
878 #define C5 FIXR(0.79335334029123516458)
879 #define C7 FIXR(0.60876142900872063941)
880 #define C9 FIXR(0.38268343236508977173)
881 #define C11 FIXR(0.13052619222005159154)
882
883 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
884 cases. */
imdct12(int * out,int * in)885 static void imdct12(int *out, int *in)
886 {
887 int tmp;
888 int64_t in1_3, in1_9, in4_3, in4_9;
889
890 in1_3 = MUL64(in[1], C3);
891 in1_9 = MUL64(in[1], C9);
892 in4_3 = MUL64(in[4], C3);
893 in4_9 = MUL64(in[4], C9);
894
895 tmp = FRAC_RND(MUL64(in[0], C7) - in1_3 - MUL64(in[2], C11) +
896 MUL64(in[3], C1) - in4_9 - MUL64(in[5], C5));
897 out[0] = tmp;
898 out[5] = -tmp;
899 tmp = FRAC_RND(MUL64(in[0] - in[3], C9) - in1_3 +
900 MUL64(in[2] + in[5], C3) - in4_9);
901 out[1] = tmp;
902 out[4] = -tmp;
903 tmp = FRAC_RND(MUL64(in[0], C11) - in1_9 + MUL64(in[2], C7) -
904 MUL64(in[3], C5) + in4_3 - MUL64(in[5], C1));
905 out[2] = tmp;
906 out[3] = -tmp;
907 tmp = FRAC_RND(MUL64(-in[0], C5) + in1_9 + MUL64(in[2], C1) +
908 MUL64(in[3], C11) - in4_3 - MUL64(in[5], C7));
909 out[6] = tmp;
910 out[11] = tmp;
911 tmp = FRAC_RND(MUL64(-in[0] + in[3], C3) - in1_9 +
912 MUL64(in[2] + in[5], C9) + in4_3);
913 out[7] = tmp;
914 out[10] = tmp;
915 tmp = FRAC_RND(-MUL64(in[0], C1) - in1_3 - MUL64(in[2], C5) -
916 MUL64(in[3], C7) - in4_9 - MUL64(in[5], C11));
917 out[8] = tmp;
918 out[9] = tmp;
919 }
920
921 #undef C1
922 #undef C3
923 #undef C5
924 #undef C7
925 #undef C9
926 #undef C11
927
928 /* cos(pi*i/18) */
929 #define C1 FIXR(0.98480775301220805936)
930 #define C2 FIXR(0.93969262078590838405)
931 #define C3 FIXR(0.86602540378443864676)
932 #define C4 FIXR(0.76604444311897803520)
933 #define C5 FIXR(0.64278760968653932632)
934 #define C6 FIXR(0.5)
935 #define C7 FIXR(0.34202014332566873304)
936 #define C8 FIXR(0.17364817766693034885)
937
938 /* 0.5 / cos(pi*(2*i+1)/36) */
939 static const int icos36[9] = {
940 FIXR(0.50190991877167369479),
941 FIXR(0.51763809020504152469),
942 FIXR(0.55168895948124587824),
943 FIXR(0.61038729438072803416),
944 FIXR(0.70710678118654752439),
945 FIXR(0.87172339781054900991),
946 FIXR(1.18310079157624925896),
947 FIXR(1.93185165257813657349),
948 FIXR(5.73685662283492756461),
949 };
950
951 static const int icos72[18] = {
952 /* 0.5 / cos(pi*(2*i+19)/72) */
953 FIXR(0.74009361646113053152),
954 FIXR(0.82133981585229078570),
955 FIXR(0.93057949835178895673),
956 FIXR(1.08284028510010010928),
957 FIXR(1.30656296487637652785),
958 FIXR(1.66275476171152078719),
959 FIXR(2.31011315767264929558),
960 FIXR(3.83064878777019433457),
961 FIXR(11.46279281302667383546),
962
963 /* 0.5 / cos(pi*(2*(i + 18) +19)/72) */
964 FIXR(-0.67817085245462840086),
965 FIXR(-0.63023620700513223342),
966 FIXR(-0.59284452371708034528),
967 FIXR(-0.56369097343317117734),
968 FIXR(-0.54119610014619698439),
969 FIXR(-0.52426456257040533932),
970 FIXR(-0.51213975715725461845),
971 FIXR(-0.50431448029007636036),
972 FIXR(-0.50047634258165998492),
973 };
974
975 /* using Lee like decomposition followed by hand coded 9 points DCT */
imdct36(int * out,int * in)976 static void imdct36(int *out, int *in)
977 {
978 int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
979 int tmp[18], *tmp1, *in1;
980 int64_t in3_3, in6_6;
981
982 for(i=17;i>=1;i--)
983 in[i] += in[i-1];
984 for(i=17;i>=3;i-=2)
985 in[i] += in[i-2];
986
987 for(j=0;j<2;j++) {
988 tmp1 = tmp + j;
989 in1 = in + j;
990
991 in3_3 = MUL64(in1[2*3], C3);
992 in6_6 = MUL64(in1[2*6], C6);
993
994 tmp1[0] = FRAC_RND(MUL64(in1[2*1], C1) + in3_3 +
995 MUL64(in1[2*5], C5) + MUL64(in1[2*7], C7));
996 tmp1[2] = in1[2*0] + FRAC_RND(MUL64(in1[2*2], C2) +
997 MUL64(in1[2*4], C4) + in6_6 +
998 MUL64(in1[2*8], C8));
999 tmp1[4] = FRAC_RND(MUL64(in1[2*1] - in1[2*5] - in1[2*7], C3));
1000 tmp1[6] = FRAC_RND(MUL64(in1[2*2] - in1[2*4] - in1[2*8], C6)) -
1001 in1[2*6] + in1[2*0];
1002 tmp1[8] = FRAC_RND(MUL64(in1[2*1], C5) - in3_3 -
1003 MUL64(in1[2*5], C7) + MUL64(in1[2*7], C1));
1004 tmp1[10] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C8) -
1005 MUL64(in1[2*4], C2) + in6_6 +
1006 MUL64(in1[2*8], C4));
1007 tmp1[12] = FRAC_RND(MUL64(in1[2*1], C7) - in3_3 +
1008 MUL64(in1[2*5], C1) -
1009 MUL64(in1[2*7], C5));
1010 tmp1[14] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C4) +
1011 MUL64(in1[2*4], C8) + in6_6 -
1012 MUL64(in1[2*8], C2));
1013 tmp1[16] = in1[2*0] - in1[2*2] + in1[2*4] - in1[2*6] + in1[2*8];
1014 }
1015
1016 i = 0;
1017 for(j=0;j<4;j++) {
1018 t0 = tmp[i];
1019 t1 = tmp[i + 2];
1020 s0 = t1 + t0;
1021 s2 = t1 - t0;
1022
1023 t2 = tmp[i + 1];
1024 t3 = tmp[i + 3];
1025 s1 = MULL(t3 + t2, icos36[j]);
1026 s3 = MULL(t3 - t2, icos36[8 - j]);
1027
1028 t0 = MULL(s0 + s1, icos72[9 + 8 - j]);
1029 t1 = MULL(s0 - s1, icos72[8 - j]);
1030 out[18 + 9 + j] = t0;
1031 out[18 + 8 - j] = t0;
1032 out[9 + j] = -t1;
1033 out[8 - j] = t1;
1034
1035 t0 = MULL(s2 + s3, icos72[9+j]);
1036 t1 = MULL(s2 - s3, icos72[j]);
1037 out[18 + 9 + (8 - j)] = t0;
1038 out[18 + j] = t0;
1039 out[9 + (8 - j)] = -t1;
1040 out[j] = t1;
1041 i += 4;
1042 }
1043
1044 s0 = tmp[16];
1045 s1 = MULL(tmp[17], icos36[4]);
1046 t0 = MULL(s0 + s1, icos72[9 + 4]);
1047 t1 = MULL(s0 - s1, icos72[4]);
1048 out[18 + 9 + 4] = t0;
1049 out[18 + 8 - 4] = t0;
1050 out[9 + 4] = -t1;
1051 out[8 - 4] = t1;
1052 }
1053
1054 /* fast header check for resync */
check_header(uint32_t header)1055 static int check_header(uint32_t header)
1056 {
1057 /* header */
1058 if ((header & 0xffe00000) != 0xffe00000)
1059 return -1;
1060 /* layer check */
1061 if (((header >> 17) & 3) == 0)
1062 return -1;
1063 /* bit rate */
1064 if (((header >> 12) & 0xf) == 0xf)
1065 return -1;
1066 /* frequency */
1067 if (((header >> 10) & 3) == 3)
1068 return -1;
1069 return 0;
1070 }
1071
1072 /* header + layer + bitrate + freq + lsf/mpeg25 */
1073 #define SAME_HEADER_MASK \
1074 (0xffe00000 | (3 << 17) | (0xf << 12) | (3 << 10) | (3 << 19))
1075
1076 /* header decoding. MUST check the header before because no
1077 consistency check is done there. Return 1 if free format found and
1078 that the frame size must be computed externally */
decode_header(MPADecodeContext * s,uint32_t header)1079 static int decode_header(MPADecodeContext *s, uint32_t header)
1080 {
1081 int sample_rate, frame_size, mpeg25, padding;
1082 int sample_rate_index, bitrate_index;
1083 if (header & (1<<20)) {
1084 s->lsf = (header & (1<<19)) ? 0 : 1;
1085 mpeg25 = 0;
1086 } else {
1087 s->lsf = 1;
1088 mpeg25 = 1;
1089 }
1090
1091 s->layer = 4 - ((header >> 17) & 3);
1092 /* extract frequency */
1093 sample_rate_index = (header >> 10) & 3;
1094 sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1095 sample_rate_index += 3 * (s->lsf + mpeg25);
1096 s->sample_rate_index = sample_rate_index;
1097 s->error_protection = ((header >> 16) & 1) ^ 1;
1098 s->sample_rate = sample_rate;
1099
1100 bitrate_index = (header >> 12) & 0xf;
1101 padding = (header >> 9) & 1;
1102 s->mode = (header >> 6) & 3;
1103 s->mode_ext = (header >> 4) & 3;
1104
1105 if (s->mode == MPA_MONO)
1106 s->nb_channels = 1;
1107 else
1108 s->nb_channels = 2;
1109
1110 if (bitrate_index != 0) {
1111 frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1112 s->bit_rate = frame_size * 1000;
1113 switch(s->layer) {
1114 case 1:
1115 frame_size = (frame_size * 12000) / sample_rate;
1116 frame_size = (frame_size + padding) * 4;
1117 break;
1118 case 2:
1119 frame_size = (frame_size * 144000) / sample_rate;
1120 frame_size += padding;
1121 break;
1122 default:
1123 case 3:
1124 frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1125 frame_size += padding;
1126 break;
1127 }
1128 s->frame_size = frame_size;
1129 } else {
1130 /* if no frame size computed, signal it */
1131 if (!s->free_format_frame_size)
1132 return 1;
1133 /* free format: compute bitrate and real frame size from the
1134 frame size we extracted by reading the bitstream */
1135 s->frame_size = s->free_format_frame_size;
1136 switch(s->layer) {
1137 case 1:
1138 s->frame_size += padding * 4;
1139 s->bit_rate = (s->frame_size * sample_rate) / 48000;
1140 break;
1141 case 2:
1142 s->frame_size += padding;
1143 s->bit_rate = (s->frame_size * sample_rate) / 144000;
1144 break;
1145 default:
1146 case 3:
1147 s->frame_size += padding;
1148 s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1149 break;
1150 }
1151 }
1152
1153 #if defined(DEBUG)
1154 printf("layer%d, %d Hz, %d kbits/s, ",
1155 s->layer, s->sample_rate, s->bit_rate);
1156 if (s->nb_channels == 2) {
1157 if (s->layer == 3) {
1158 if (s->mode_ext & MODE_EXT_MS_STEREO)
1159 printf("ms-");
1160 if (s->mode_ext & MODE_EXT_I_STEREO)
1161 printf("i-");
1162 }
1163 printf("stereo");
1164 } else {
1165 printf("mono");
1166 }
1167 printf("\n");
1168 #endif
1169 return 0;
1170 }
1171
1172 /* return the number of decoded frames */
mp_decode_layer1(MPADecodeContext * s)1173 static int mp_decode_layer1(MPADecodeContext *s)
1174 {
1175 int bound, i, v, n, ch, j, mant;
1176 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1177 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1178
1179 if (s->mode == MPA_JSTEREO)
1180 bound = (s->mode_ext + 1) * 4;
1181 else
1182 bound = SBLIMIT;
1183
1184 /* allocation bits */
1185 for(i=0;i<bound;i++) {
1186 for(ch=0;ch<s->nb_channels;ch++) {
1187 allocation[ch][i] = get_bits(&s->gb, 4);
1188 }
1189 }
1190 for(i=bound;i<SBLIMIT;i++) {
1191 allocation[0][i] = get_bits(&s->gb, 4);
1192 }
1193
1194 /* scale factors */
1195 for(i=0;i<bound;i++) {
1196 for(ch=0;ch<s->nb_channels;ch++) {
1197 if (allocation[ch][i])
1198 scale_factors[ch][i] = get_bits(&s->gb, 6);
1199 }
1200 }
1201 for(i=bound;i<SBLIMIT;i++) {
1202 if (allocation[0][i]) {
1203 scale_factors[0][i] = get_bits(&s->gb, 6);
1204 scale_factors[1][i] = get_bits(&s->gb, 6);
1205 }
1206 }
1207
1208 /* compute samples */
1209 for(j=0;j<12;j++) {
1210 for(i=0;i<bound;i++) {
1211 for(ch=0;ch<s->nb_channels;ch++) {
1212 n = allocation[ch][i];
1213 if (n) {
1214 mant = get_bits(&s->gb, n + 1);
1215 v = l1_unscale(n, mant, scale_factors[ch][i]);
1216 } else {
1217 v = 0;
1218 }
1219 s->sb_samples[ch][j][i] = v;
1220 }
1221 }
1222 for(i=bound;i<SBLIMIT;i++) {
1223 n = allocation[0][i];
1224 if (n) {
1225 mant = get_bits(&s->gb, n + 1);
1226 v = l1_unscale(n, mant, scale_factors[0][i]);
1227 s->sb_samples[0][j][i] = v;
1228 v = l1_unscale(n, mant, scale_factors[1][i]);
1229 s->sb_samples[1][j][i] = v;
1230 } else {
1231 s->sb_samples[0][j][i] = 0;
1232 s->sb_samples[1][j][i] = 0;
1233 }
1234 }
1235 }
1236 return 12;
1237 }
1238
1239 /* bitrate is in kb/s */
l2_select_table(int bitrate,int nb_channels,int freq,int lsf)1240 static int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1241 {
1242 int ch_bitrate, table;
1243
1244 ch_bitrate = bitrate / nb_channels;
1245 if (!lsf) {
1246 if ((freq == 48000 && ch_bitrate >= 56) ||
1247 (ch_bitrate >= 56 && ch_bitrate <= 80))
1248 table = 0;
1249 else if (freq != 48000 && ch_bitrate >= 96)
1250 table = 1;
1251 else if (freq != 32000 && ch_bitrate <= 48)
1252 table = 2;
1253 else
1254 table = 3;
1255 } else {
1256 table = 4;
1257 }
1258 return table;
1259 }
1260
mp_decode_layer2(MPADecodeContext * s)1261 static int mp_decode_layer2(MPADecodeContext *s)
1262 {
1263 int sblimit; /* number of used subbands */
1264 const unsigned char *alloc_table;
1265 int table, bit_alloc_bits, i, j, ch, bound, v;
1266 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1267 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1268 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1269 int scale, qindex, bits, steps, k, l, m, b;
1270
1271 /* select decoding table */
1272 table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1273 s->sample_rate, s->lsf);
1274 sblimit = sblimit_table[table];
1275 alloc_table = alloc_tables[table];
1276
1277 if (s->mode == MPA_JSTEREO)
1278 bound = (s->mode_ext + 1) * 4;
1279 else
1280 bound = sblimit;
1281
1282 #ifdef DEBUG
1283 printf("bound=%d sblimit=%d\n", bound, sblimit);
1284 #endif
1285 /* parse bit allocation */
1286 j = 0;
1287 for(i=0;i<bound;i++) {
1288 bit_alloc_bits = alloc_table[j];
1289 for(ch=0;ch<s->nb_channels;ch++) {
1290 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1291 }
1292 j += 1 << bit_alloc_bits;
1293 }
1294 for(i=bound;i<sblimit;i++) {
1295 bit_alloc_bits = alloc_table[j];
1296 v = get_bits(&s->gb, bit_alloc_bits);
1297 bit_alloc[0][i] = v;
1298 bit_alloc[1][i] = v;
1299 j += 1 << bit_alloc_bits;
1300 }
1301
1302 #ifdef DEBUG
1303 {
1304 for(ch=0;ch<s->nb_channels;ch++) {
1305 for(i=0;i<sblimit;i++)
1306 printf(" %d", bit_alloc[ch][i]);
1307 printf("\n");
1308 }
1309 }
1310 #endif
1311
1312 /* scale codes */
1313 for(i=0;i<sblimit;i++) {
1314 for(ch=0;ch<s->nb_channels;ch++) {
1315 if (bit_alloc[ch][i])
1316 scale_code[ch][i] = get_bits(&s->gb, 2);
1317 }
1318 }
1319
1320 /* scale factors */
1321 for(i=0;i<sblimit;i++) {
1322 for(ch=0;ch<s->nb_channels;ch++) {
1323 if (bit_alloc[ch][i]) {
1324 sf = scale_factors[ch][i];
1325 switch(scale_code[ch][i]) {
1326 default:
1327 case 0:
1328 sf[0] = get_bits(&s->gb, 6);
1329 sf[1] = get_bits(&s->gb, 6);
1330 sf[2] = get_bits(&s->gb, 6);
1331 break;
1332 case 2:
1333 sf[0] = get_bits(&s->gb, 6);
1334 sf[1] = sf[0];
1335 sf[2] = sf[0];
1336 break;
1337 case 1:
1338 sf[0] = get_bits(&s->gb, 6);
1339 sf[2] = get_bits(&s->gb, 6);
1340 sf[1] = sf[0];
1341 break;
1342 case 3:
1343 sf[0] = get_bits(&s->gb, 6);
1344 sf[2] = get_bits(&s->gb, 6);
1345 sf[1] = sf[2];
1346 break;
1347 }
1348 }
1349 }
1350 }
1351
1352 #ifdef DEBUG
1353 for(ch=0;ch<s->nb_channels;ch++) {
1354 for(i=0;i<sblimit;i++) {
1355 if (bit_alloc[ch][i]) {
1356 sf = scale_factors[ch][i];
1357 printf(" %d %d %d", sf[0], sf[1], sf[2]);
1358 } else {
1359 printf(" -");
1360 }
1361 }
1362 printf("\n");
1363 }
1364 #endif
1365
1366 /* samples */
1367 for(k=0;k<3;k++) {
1368 for(l=0;l<12;l+=3) {
1369 j = 0;
1370 for(i=0;i<bound;i++) {
1371 bit_alloc_bits = alloc_table[j];
1372 for(ch=0;ch<s->nb_channels;ch++) {
1373 b = bit_alloc[ch][i];
1374 if (b) {
1375 scale = scale_factors[ch][i][k];
1376 qindex = alloc_table[j+b];
1377 bits = quant_bits[qindex];
1378 if (bits < 0) {
1379 /* 3 values at the same time */
1380 v = get_bits(&s->gb, -bits);
1381 steps = quant_steps[qindex];
1382 s->sb_samples[ch][k * 12 + l + 0][i] =
1383 l2_unscale_group(steps, v % steps, scale);
1384 v = v / steps;
1385 s->sb_samples[ch][k * 12 + l + 1][i] =
1386 l2_unscale_group(steps, v % steps, scale);
1387 v = v / steps;
1388 s->sb_samples[ch][k * 12 + l + 2][i] =
1389 l2_unscale_group(steps, v, scale);
1390 } else {
1391 for(m=0;m<3;m++) {
1392 v = get_bits(&s->gb, bits);
1393 v = l1_unscale(bits - 1, v, scale);
1394 s->sb_samples[ch][k * 12 + l + m][i] = v;
1395 }
1396 }
1397 } else {
1398 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1399 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1400 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1401 }
1402 }
1403 /* next subband in alloc table */
1404 j += 1 << bit_alloc_bits;
1405 }
1406 /* XXX: find a way to avoid this duplication of code */
1407 for(i=bound;i<sblimit;i++) {
1408 bit_alloc_bits = alloc_table[j];
1409 b = bit_alloc[0][i];
1410 if (b) {
1411 int mant, scale0, scale1;
1412 scale0 = scale_factors[0][i][k];
1413 scale1 = scale_factors[1][i][k];
1414 qindex = alloc_table[j+b];
1415 bits = quant_bits[qindex];
1416 if (bits < 0) {
1417 /* 3 values at the same time */
1418 v = get_bits(&s->gb, -bits);
1419 steps = quant_steps[qindex];
1420 mant = v % steps;
1421 v = v / steps;
1422 s->sb_samples[0][k * 12 + l + 0][i] =
1423 l2_unscale_group(steps, mant, scale0);
1424 s->sb_samples[1][k * 12 + l + 0][i] =
1425 l2_unscale_group(steps, mant, scale1);
1426 mant = v % steps;
1427 v = v / steps;
1428 s->sb_samples[0][k * 12 + l + 1][i] =
1429 l2_unscale_group(steps, mant, scale0);
1430 s->sb_samples[1][k * 12 + l + 1][i] =
1431 l2_unscale_group(steps, mant, scale1);
1432 s->sb_samples[0][k * 12 + l + 2][i] =
1433 l2_unscale_group(steps, v, scale0);
1434 s->sb_samples[1][k * 12 + l + 2][i] =
1435 l2_unscale_group(steps, v, scale1);
1436 } else {
1437 for(m=0;m<3;m++) {
1438 mant = get_bits(&s->gb, bits);
1439 s->sb_samples[0][k * 12 + l + m][i] =
1440 l1_unscale(bits - 1, mant, scale0);
1441 s->sb_samples[1][k * 12 + l + m][i] =
1442 l1_unscale(bits - 1, mant, scale1);
1443 }
1444 }
1445 } else {
1446 s->sb_samples[0][k * 12 + l + 0][i] = 0;
1447 s->sb_samples[0][k * 12 + l + 1][i] = 0;
1448 s->sb_samples[0][k * 12 + l + 2][i] = 0;
1449 s->sb_samples[1][k * 12 + l + 0][i] = 0;
1450 s->sb_samples[1][k * 12 + l + 1][i] = 0;
1451 s->sb_samples[1][k * 12 + l + 2][i] = 0;
1452 }
1453 /* next subband in alloc table */
1454 j += 1 << bit_alloc_bits;
1455 }
1456 /* fill remaining samples to zero */
1457 for(i=sblimit;i<SBLIMIT;i++) {
1458 for(ch=0;ch<s->nb_channels;ch++) {
1459 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1460 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1461 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1462 }
1463 }
1464 }
1465 }
1466 return 3 * 12;
1467 }
1468
1469 /*
1470 * Seek back in the stream for backstep bytes (at most 511 bytes)
1471 */
seek_to_maindata(MPADecodeContext * s,unsigned int backstep)1472 static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)
1473 {
1474 uint8_t *ptr;
1475
1476 /* compute current position in stream */
1477 ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));
1478
1479 /* copy old data before current one */
1480 ptr -= backstep;
1481 memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
1482 BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
1483 /* init get bits again */
1484 init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);
1485
1486 /* prepare next buffer */
1487 s->inbuf_index ^= 1;
1488 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1489 s->old_frame_size = s->frame_size;
1490 }
1491
lsf_sf_expand(int * slen,int sf,int n1,int n2,int n3)1492 static void lsf_sf_expand(int *slen,
1493 int sf, int n1, int n2, int n3)
1494 {
1495 if (n3) {
1496 slen[3] = sf % n3;
1497 sf /= n3;
1498 } else {
1499 slen[3] = 0;
1500 }
1501 if (n2) {
1502 slen[2] = sf % n2;
1503 sf /= n2;
1504 } else {
1505 slen[2] = 0;
1506 }
1507 slen[1] = sf % n1;
1508 sf /= n1;
1509 slen[0] = sf;
1510 }
1511
exponents_from_scale_factors(MPADecodeContext * s,GranuleDef * g,int16_t * exponents)1512 static void exponents_from_scale_factors(MPADecodeContext *s,
1513 GranuleDef *g,
1514 int16_t *exponents)
1515 {
1516 const uint8_t *bstab, *pretab;
1517 int len, i, j, k, l, v0, shift, gain, gains[3];
1518 int16_t *exp_ptr;
1519
1520 exp_ptr = exponents;
1521 gain = g->global_gain - 210;
1522 shift = g->scalefac_scale + 1;
1523
1524 bstab = band_size_long[s->sample_rate_index];
1525 pretab = mpa_pretab[g->preflag];
1526 for(i=0;i<g->long_end;i++) {
1527 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
1528 len = bstab[i];
1529 for(j=len;j>0;j--)
1530 *exp_ptr++ = v0;
1531 }
1532
1533 if (g->short_start < 13) {
1534 bstab = band_size_short[s->sample_rate_index];
1535 gains[0] = gain - (g->subblock_gain[0] << 3);
1536 gains[1] = gain - (g->subblock_gain[1] << 3);
1537 gains[2] = gain - (g->subblock_gain[2] << 3);
1538 k = g->long_end;
1539 for(i=g->short_start;i<13;i++) {
1540 len = bstab[i];
1541 for(l=0;l<3;l++) {
1542 v0 = gains[l] - (g->scale_factors[k++] << shift);
1543 for(j=len;j>0;j--)
1544 *exp_ptr++ = v0;
1545 }
1546 }
1547 }
1548 }
1549
1550 /* handle n = 0 too */
get_bitsz(GetBitContext * s,int n)1551 static int get_bitsz(GetBitContext *s, int n)
1552 {
1553 if (n == 0)
1554 return 0;
1555 else
1556 return get_bits(s, n);
1557 }
1558
huffman_decode(MPADecodeContext * s,GranuleDef * g,int16_t * exponents,int end_pos)1559 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1560 int16_t *exponents, int end_pos)
1561 {
1562 int s_index;
1563 int linbits, code, x, y, l, v, i, j, k, pos;
1564 GetBitContext last_gb;
1565 VLC *vlc;
1566 uint8_t *code_table;
1567
1568 /* low frequencies (called big values) */
1569 s_index = 0;
1570 for(i=0;i<3;i++) {
1571 j = g->region_size[i];
1572 if (j == 0)
1573 continue;
1574 /* select vlc table */
1575 k = g->table_select[i];
1576 l = mpa_huff_data[k][0];
1577 linbits = mpa_huff_data[k][1];
1578 vlc = &huff_vlc[l];
1579 code_table = huff_code_table[l];
1580
1581 /* read huffcode and compute each couple */
1582 for(;j>0;j--) {
1583 if (get_bits_count(&s->gb) >= end_pos)
1584 break;
1585 if (code_table) {
1586 code = get_vlc(&s->gb, vlc);
1587 if (code < 0)
1588 return -1;
1589 y = code_table[code];
1590 x = y >> 4;
1591 y = y & 0x0f;
1592 } else {
1593 x = 0;
1594 y = 0;
1595 }
1596 #ifdef DEBUG
1597 printf("region=%d n=%d x=%d y=%d exp=%d\n",
1598 i, g->region_size[i] - j, x, y, exponents[s_index]);
1599 #endif
1600 if (x) {
1601 if (x == 15)
1602 x += get_bitsz(&s->gb, linbits);
1603 v = l3_unscale(x, exponents[s_index]);
1604 if (get_bits(&s->gb, 1))
1605 v = -v;
1606 } else {
1607 v = 0;
1608 }
1609 g->sb_hybrid[s_index++] = v;
1610 if (y) {
1611 if (y == 15)
1612 y += get_bitsz(&s->gb, linbits);
1613 v = l3_unscale(y, exponents[s_index]);
1614 if (get_bits(&s->gb, 1))
1615 v = -v;
1616 } else {
1617 v = 0;
1618 }
1619 g->sb_hybrid[s_index++] = v;
1620 }
1621 }
1622
1623 /* high frequencies */
1624 vlc = &huff_quad_vlc[g->count1table_select];
1625 last_gb.buffer = NULL;
1626 while (s_index <= 572) {
1627 pos = get_bits_count(&s->gb);
1628 if (pos >= end_pos) {
1629 if (pos > end_pos && last_gb.buffer != NULL) {
1630 /* some encoders generate an incorrect size for this
1631 part. We must go back into the data */
1632 s_index -= 4;
1633 s->gb = last_gb;
1634 }
1635 break;
1636 }
1637 last_gb= s->gb;
1638
1639 code = get_vlc(&s->gb, vlc);
1640 #ifdef DEBUG
1641 printf("t=%d code=%d\n", g->count1table_select, code);
1642 #endif
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_bits(&s->gb, 1))
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 */
reorder_block(MPADecodeContext * s,GranuleDef * g)1666 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1667 {
1668 int i, j, k, len;
1669 int32_t *ptr, *dst, *ptr1;
1670 int32_t 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_t));
1696 }
1697 }
1698
1699 #define ISQRT2 FIXR(0.70710678118654752440)
1700
compute_stereo(MPADecodeContext * s,GranuleDef * g0,GranuleDef * g1)1701 static void compute_stereo(MPADecodeContext *s,
1702 GranuleDef *g0, GranuleDef *g1)
1703 {
1704 int i, j, k, l;
1705 int32_t v1, v2;
1706 int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1707 int32_t (*is_tab)[16];
1708 int32_t *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
compute_antialias(MPADecodeContext * s,GranuleDef * g)1828 static void compute_antialias(MPADecodeContext *s,
1829 GranuleDef *g)
1830 {
1831 int32_t *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
compute_imdct(MPADecodeContext * s,GranuleDef * g,int32_t * sb_samples,int32_t * mdct_buf)1862 static void compute_imdct(MPADecodeContext *s,
1863 GranuleDef *g,
1864 int32_t *sb_samples,
1865 int32_t *mdct_buf)
1866 {
1867 int32_t *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1;
1868 int32_t in[6];
1869 int32_t out[36];
1870 int32_t 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 /* main layer3 decoding function */
mp_decode_layer3(MPADecodeContext * s)1963 static int mp_decode_layer3(MPADecodeContext *s)
1964 {
1965 int nb_granules, main_data_begin, private_bits;
1966 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
1967 GranuleDef granules[2][2], *g;
1968 int16_t exponents[576];
1969
1970 /* read side info */
1971 if (s->lsf) {
1972 main_data_begin = get_bits(&s->gb, 8);
1973 if (s->nb_channels == 2)
1974 private_bits = get_bits(&s->gb, 2);
1975 else
1976 private_bits = get_bits(&s->gb, 1);
1977 nb_granules = 1;
1978 } else {
1979 main_data_begin = get_bits(&s->gb, 9);
1980 if (s->nb_channels == 2)
1981 private_bits = get_bits(&s->gb, 3);
1982 else
1983 private_bits = get_bits(&s->gb, 5);
1984 nb_granules = 2;
1985 for(ch=0;ch<s->nb_channels;ch++) {
1986 granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
1987 granules[ch][1].scfsi = get_bits(&s->gb, 4);
1988 }
1989 }
1990
1991 for(gr=0;gr<nb_granules;gr++) {
1992 for(ch=0;ch<s->nb_channels;ch++) {
1993 #ifdef DEBUG
1994 printf("gr=%d ch=%d: side_info\n", gr, ch);
1995 #endif
1996 g = &granules[ch][gr];
1997 g->part2_3_length = get_bits(&s->gb, 12);
1998 g->big_values = get_bits(&s->gb, 9);
1999 g->global_gain = get_bits(&s->gb, 8);
2000 /* if MS stereo only is selected, we precompute the
2001 1/sqrt(2) renormalization factor */
2002 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2003 MODE_EXT_MS_STEREO)
2004 g->global_gain -= 2;
2005 if (s->lsf)
2006 g->scalefac_compress = get_bits(&s->gb, 9);
2007 else
2008 g->scalefac_compress = get_bits(&s->gb, 4);
2009 blocksplit_flag = get_bits(&s->gb, 1);
2010 if (blocksplit_flag) {
2011 g->block_type = get_bits(&s->gb, 2);
2012 if (g->block_type == 0)
2013 return -1;
2014 g->switch_point = get_bits(&s->gb, 1);
2015 for(i=0;i<2;i++)
2016 g->table_select[i] = get_bits(&s->gb, 5);
2017 for(i=0;i<3;i++)
2018 g->subblock_gain[i] = get_bits(&s->gb, 3);
2019 /* compute huffman coded region sizes */
2020 if (g->block_type == 2)
2021 g->region_size[0] = (36 / 2);
2022 else {
2023 if (s->sample_rate_index <= 2)
2024 g->region_size[0] = (36 / 2);
2025 else if (s->sample_rate_index != 8)
2026 g->region_size[0] = (54 / 2);
2027 else
2028 g->region_size[0] = (108 / 2);
2029 }
2030 g->region_size[1] = (576 / 2);
2031 } else {
2032 int region_address1, region_address2, l;
2033 g->block_type = 0;
2034 g->switch_point = 0;
2035 for(i=0;i<3;i++)
2036 g->table_select[i] = get_bits(&s->gb, 5);
2037 /* compute huffman coded region sizes */
2038 region_address1 = get_bits(&s->gb, 4);
2039 region_address2 = get_bits(&s->gb, 3);
2040 #ifdef DEBUG
2041 printf("region1=%d region2=%d\n",
2042 region_address1, region_address2);
2043 #endif
2044 g->region_size[0] =
2045 band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2046 l = region_address1 + region_address2 + 2;
2047 /* should not overflow */
2048 if (l > 22)
2049 l = 22;
2050 g->region_size[1] =
2051 band_index_long[s->sample_rate_index][l] >> 1;
2052 }
2053 /* convert region offsets to region sizes and truncate
2054 size to big_values */
2055 g->region_size[2] = (576 / 2);
2056 j = 0;
2057 for(i=0;i<3;i++) {
2058 k = g->region_size[i];
2059 if (k > g->big_values)
2060 k = g->big_values;
2061 g->region_size[i] = k - j;
2062 j = k;
2063 }
2064
2065 /* compute band indexes */
2066 if (g->block_type == 2) {
2067 if (g->switch_point) {
2068 /* if switched mode, we handle the 36 first samples as
2069 long blocks. For 8000Hz, we handle the 48 first
2070 exponents as long blocks (XXX: check this!) */
2071 if (s->sample_rate_index <= 2)
2072 g->long_end = 8;
2073 else if (s->sample_rate_index != 8)
2074 g->long_end = 6;
2075 else
2076 g->long_end = 4; /* 8000 Hz */
2077
2078 if (s->sample_rate_index != 8)
2079 g->short_start = 3;
2080 else
2081 g->short_start = 2;
2082 } else {
2083 g->long_end = 0;
2084 g->short_start = 0;
2085 }
2086 } else {
2087 g->short_start = 13;
2088 g->long_end = 22;
2089 }
2090
2091 g->preflag = 0;
2092 if (!s->lsf)
2093 g->preflag = get_bits(&s->gb, 1);
2094 g->scalefac_scale = get_bits(&s->gb, 1);
2095 g->count1table_select = get_bits(&s->gb, 1);
2096 #ifdef DEBUG
2097 printf("block_type=%d switch_point=%d\n",
2098 g->block_type, g->switch_point);
2099 #endif
2100 }
2101 }
2102
2103 /* now we get bits from the main_data_begin offset */
2104 #ifdef DEBUG
2105 printf("seekback: %d\n", main_data_begin);
2106 #endif
2107 seek_to_maindata(s, main_data_begin);
2108
2109 for(gr=0;gr<nb_granules;gr++) {
2110 for(ch=0;ch<s->nb_channels;ch++) {
2111 g = &granules[ch][gr];
2112
2113 bits_pos = get_bits_count(&s->gb);
2114
2115 if (!s->lsf) {
2116 uint8_t *sc;
2117 int slen, slen1, slen2;
2118
2119 /* MPEG1 scale factors */
2120 slen1 = slen_table[0][g->scalefac_compress];
2121 slen2 = slen_table[1][g->scalefac_compress];
2122 #ifdef DEBUG
2123 printf("slen1=%d slen2=%d\n", slen1, slen2);
2124 #endif
2125 if (g->block_type == 2) {
2126 n = g->switch_point ? 17 : 18;
2127 j = 0;
2128 for(i=0;i<n;i++)
2129 g->scale_factors[j++] = get_bitsz(&s->gb, slen1);
2130 for(i=0;i<18;i++)
2131 g->scale_factors[j++] = get_bitsz(&s->gb, slen2);
2132 for(i=0;i<3;i++)
2133 g->scale_factors[j++] = 0;
2134 } else {
2135 sc = granules[ch][0].scale_factors;
2136 j = 0;
2137 for(k=0;k<4;k++) {
2138 n = (k == 0 ? 6 : 5);
2139 if ((g->scfsi & (0x8 >> k)) == 0) {
2140 slen = (k < 2) ? slen1 : slen2;
2141 for(i=0;i<n;i++)
2142 g->scale_factors[j++] = get_bitsz(&s->gb, slen);
2143 } else {
2144 /* simply copy from last granule */
2145 for(i=0;i<n;i++) {
2146 g->scale_factors[j] = sc[j];
2147 j++;
2148 }
2149 }
2150 }
2151 g->scale_factors[j++] = 0;
2152 }
2153 #if defined(DEBUG)
2154 {
2155 printf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2156 g->scfsi, gr, ch);
2157 for(i=0;i<j;i++)
2158 printf(" %d", g->scale_factors[i]);
2159 printf("\n");
2160 }
2161 #endif
2162 } else {
2163 int tindex, tindex2, slen[4], sl, sf;
2164
2165 /* LSF scale factors */
2166 if (g->block_type == 2) {
2167 tindex = g->switch_point ? 2 : 1;
2168 } else {
2169 tindex = 0;
2170 }
2171 sf = g->scalefac_compress;
2172 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2173 /* intensity stereo case */
2174 sf >>= 1;
2175 if (sf < 180) {
2176 lsf_sf_expand(slen, sf, 6, 6, 0);
2177 tindex2 = 3;
2178 } else if (sf < 244) {
2179 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2180 tindex2 = 4;
2181 } else {
2182 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2183 tindex2 = 5;
2184 }
2185 } else {
2186 /* normal case */
2187 if (sf < 400) {
2188 lsf_sf_expand(slen, sf, 5, 4, 4);
2189 tindex2 = 0;
2190 } else if (sf < 500) {
2191 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2192 tindex2 = 1;
2193 } else {
2194 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2195 tindex2 = 2;
2196 g->preflag = 1;
2197 }
2198 }
2199
2200 j = 0;
2201 for(k=0;k<4;k++) {
2202 n = lsf_nsf_table[tindex2][tindex][k];
2203 sl = slen[k];
2204 for(i=0;i<n;i++)
2205 g->scale_factors[j++] = get_bitsz(&s->gb, sl);
2206 }
2207 /* XXX: should compute exact size */
2208 for(;j<40;j++)
2209 g->scale_factors[j] = 0;
2210 #if defined(DEBUG)
2211 {
2212 printf("gr=%d ch=%d scale_factors:\n",
2213 gr, ch);
2214 for(i=0;i<40;i++)
2215 printf(" %d", g->scale_factors[i]);
2216 printf("\n");
2217 }
2218 #endif
2219 }
2220
2221 exponents_from_scale_factors(s, g, exponents);
2222
2223 /* read Huffman coded residue */
2224 if (huffman_decode(s, g, exponents,
2225 bits_pos + g->part2_3_length) < 0)
2226 return -1;
2227
2228 /* skip extension bits */
2229 bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2230 if (bits_left < 0) {
2231 #ifdef DEBUG
2232 printf("bits_left=%d\n", bits_left);
2233 #endif
2234 return -1;
2235 }
2236 while (bits_left >= 16) {
2237 skip_bits(&s->gb, 16);
2238 bits_left -= 16;
2239 }
2240 if (bits_left > 0)
2241 skip_bits(&s->gb, bits_left);
2242 } /* ch */
2243
2244 if (s->nb_channels == 2)
2245 compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2246
2247 for(ch=0;ch<s->nb_channels;ch++) {
2248 g = &granules[ch][gr];
2249
2250 reorder_block(s, g);
2251 compute_antialias(s, g);
2252 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2253 }
2254 } /* gr */
2255 return nb_granules * 18;
2256 }
2257
mp_decode_frame(MPADecodeContext * s,int16_t * samples)2258 static int mp_decode_frame(MPADecodeContext *s,
2259 int16_t *samples)
2260 {
2261 int i, nb_frames, ch;
2262 int16_t *samples_ptr;
2263
2264 init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
2265 (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);
2266
2267 /* skip error protection field */
2268 if (s->error_protection)
2269 get_bits(&s->gb, 16);
2270
2271 #ifdef DEBUG
2272 printf("frame %d:\n", s->frame_count);
2273 #endif
2274 switch(s->layer) {
2275 case 1:
2276 nb_frames = mp_decode_layer1(s);
2277 break;
2278 case 2:
2279 nb_frames = mp_decode_layer2(s);
2280 break;
2281 case 3:
2282 default:
2283 nb_frames = mp_decode_layer3(s);
2284 break;
2285 }
2286 #if defined(DEBUG)
2287 for(i=0;i<nb_frames;i++) {
2288 for(ch=0;ch<s->nb_channels;ch++) {
2289 int j;
2290 printf("%d-%d:", i, ch);
2291 for(j=0;j<SBLIMIT;j++)
2292 printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2293 printf("\n");
2294 }
2295 }
2296 #endif
2297 /* apply the synthesis filter */
2298 for(ch=0;ch<s->nb_channels;ch++) {
2299 samples_ptr = samples + ch;
2300 for(i=0;i<nb_frames;i++) {
2301 synth_filter(s, ch, samples_ptr, s->nb_channels,
2302 s->sb_samples[ch][i]);
2303 samples_ptr += 32 * s->nb_channels;
2304 }
2305 }
2306 #ifdef DEBUG
2307 s->frame_count++;
2308 #endif
2309 return nb_frames * 32 * sizeof(short) * s->nb_channels;
2310 }
2311
mpaudec_decode_frame(MPAuDecContext * mpctx,void * data,int * data_size,const uint8_t * buf,int buf_size)2312 int mpaudec_decode_frame(MPAuDecContext * mpctx,
2313 void *data, int *data_size,
2314 const uint8_t * buf, int buf_size)
2315 {
2316 MPADecodeContext *s;
2317 const uint8_t *buf_ptr = buf;
2318 int out_size = 0;
2319 int16_t *out_samples = data;
2320 assert(mpctx != NULL);
2321 assert(mpctx->priv_data != NULL);
2322 s = mpctx->priv_data;
2323
2324 while (buf_size > 0 && out_size == 0) {
2325 uint32_t header;
2326 uint32_t free_format_next_header = 0;
2327 int len = s->inbuf_ptr - s->inbuf;
2328 if (s->frame_size == 0) {
2329 /* no header seen : find one. We need at least HEADER_SIZE
2330 bytes to parse it */
2331 len = HEADER_SIZE - len;
2332 if (len > buf_size)
2333 len = buf_size;
2334 if (len > 0) {
2335 memcpy(s->inbuf_ptr, buf_ptr, len);
2336 buf_ptr += len;
2337 buf_size -= len;
2338 s->inbuf_ptr += len;
2339 }
2340 if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
2341 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2342 (s->inbuf[2] << 8) | s->inbuf[3];
2343
2344 if (check_header(header) < 0) {
2345 /* no sync found : move by one byte (inefficient, but simple!) */
2346 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2347 s->inbuf_ptr--;
2348 #ifdef DEBUG
2349 printf("skip %x\n", header);
2350 #endif
2351 /* reset free format frame size to give a chance
2352 to get a new bitrate */
2353 s->free_format_frame_size = 0;
2354 } else {
2355 if (decode_header(s, header) == 1) {
2356 /* free format: prepare to compute frame size */
2357 s->frame_size = -1;
2358 }
2359 /* update codec info */
2360 mpctx->sample_rate = s->sample_rate;
2361 mpctx->channels = s->nb_channels;
2362 mpctx->bit_rate = s->bit_rate;
2363 mpctx->layer = s->layer;
2364 switch(s->layer) {
2365 case 1:
2366 mpctx->frame_size = 384;
2367 break;
2368 case 2:
2369 mpctx->frame_size = 1152;
2370 break;
2371 case 3:
2372 if (s->lsf)
2373 mpctx->frame_size = 576;
2374 else
2375 mpctx->frame_size = 1152;
2376 break;
2377 }
2378 }
2379 }
2380 } else if (s->frame_size == -1) {
2381 /* free format : find next sync to compute frame size */
2382 len = MPA_MAX_CODED_FRAME_SIZE - len;
2383 if (len > buf_size)
2384 len = buf_size;
2385 if (len == 0) {
2386 /* frame too long: resync */
2387 s->frame_size = 0;
2388 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2389 s->inbuf_ptr--;
2390 } else {
2391 uint8_t *p, *pend;
2392 uint32_t header1;
2393 int padding;
2394
2395 memcpy(s->inbuf_ptr, buf_ptr, len);
2396 /* check for header */
2397 p = s->inbuf_ptr - 3;
2398 pend = s->inbuf_ptr + len - 4;
2399 while (p <= pend && free_format_next_header == 0) {
2400 header = (p[0] << 24) | (p[1] << 16) |
2401 (p[2] << 8) | p[3];
2402 header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2403 (s->inbuf[2] << 8) | s->inbuf[3];
2404 /* check with high probability that we have a
2405 valid header */
2406 if ((header & SAME_HEADER_MASK) ==
2407 (header1 & SAME_HEADER_MASK)) {
2408 /* header found: update pointers */
2409 len = (p + 4) - s->inbuf_ptr;
2410 buf_ptr += len;
2411 buf_size -= len;
2412 s->inbuf_ptr = p;
2413 free_format_next_header = header;
2414 /* compute frame size */
2415 s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2416 padding = (header1 >> 9) & 1;
2417 if (s->layer == 1)
2418 s->free_format_frame_size -= padding * 4;
2419 else
2420 s->free_format_frame_size -= padding;
2421 #ifdef DEBUG
2422 printf("free frame size=%d padding=%d\n",
2423 s->free_format_frame_size, padding);
2424 #endif
2425 decode_header(s, header1);
2426 } else
2427 p++;
2428 }
2429 if (free_format_next_header == 0) {
2430 /* not found: simply increase pointers */
2431 buf_ptr += len;
2432 s->inbuf_ptr += len;
2433 buf_size -= len;
2434 }
2435 }
2436 } else if (len < s->frame_size) {
2437 if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2438 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
2439 len = s->frame_size - len;
2440 if (len > buf_size)
2441 len = buf_size;
2442 memcpy(s->inbuf_ptr, buf_ptr, len);
2443 buf_ptr += len;
2444 s->inbuf_ptr += len;
2445 buf_size -= len;
2446 }
2447 if (s->frame_size > 0 &&
2448 (s->inbuf_ptr - s->inbuf) >= s->frame_size) {
2449 mpctx->coded_frame_size = s->frame_size;
2450 if (mpctx->parse_only) {
2451 /* simply return the frame data */
2452 *(uint8_t **)data = s->inbuf;
2453 out_size = s->inbuf_ptr - s->inbuf;
2454 } else {
2455 out_size = mp_decode_frame(s, out_samples);
2456 }
2457 if (free_format_next_header != 0) {
2458 s->inbuf[0] = free_format_next_header >> 24;
2459 s->inbuf[1] = free_format_next_header >> 16;
2460 s->inbuf[2] = free_format_next_header >> 8;
2461 s->inbuf[3] = free_format_next_header;
2462 s->inbuf_ptr = s->inbuf + 4;
2463 } else
2464 s->inbuf_ptr = s->inbuf;
2465 s->frame_size = 0;
2466 }
2467 }
2468 *data_size = out_size;
2469 return buf_ptr - buf;
2470 }
2471
mpaudec_clear(MPAuDecContext * mpctx)2472 void mpaudec_clear(MPAuDecContext *mpctx)
2473 {
2474 assert(mpctx != NULL);
2475 free(mpctx->priv_data);
2476 memset(mpctx, 0, sizeof(MPAuDecContext));
2477 }
2478