Mercurial > audlegacy-plugins

comparison src/ffmpeg/libavcodec/mpegaudiodec.c @ 808:e8776388b02a trunk

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
[svn] - add ffmpeg
author nenolod
date Mon, 12 Mar 2007 11:18:54 -0700
parents
children
comparison
equal deleted inserted replaced
807:0f9c8d4d3ac4 808:e8776388b02a
1 /*
2 * MPEG Audio decoder
3 * Copyright (c) 2001, 2002 Fabrice Bellard.
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file mpegaudiodec.c
24 * MPEG Audio decoder.
25 */
26
27 //#define DEBUG
28 #include "avcodec.h"
29 #include "bitstream.h"
30 #include "dsputil.h"
31
32 /*
33 * TODO:
34 * - in low precision mode, use more 16 bit multiplies in synth filter
35 * - test lsf / mpeg25 extensively.
36 */
37
38 /* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
39 audio decoder */
40 #ifdef CONFIG_MPEGAUDIO_HP
41 # define USE_HIGHPRECISION
42 #endif
43
44 #include "mpegaudio.h"
45
46 #include "mathops.h"
47
48 #define FRAC_ONE (1 << FRAC_BITS)
49
50 #define FIX(a) ((int)((a) * FRAC_ONE))
51 /* WARNING: only correct for posititive numbers */
52 #define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
53 #define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
54
55 #define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
56
57 /****************/
58
59 #define HEADER_SIZE 4
60 #define BACKSTEP_SIZE 512
61 #define EXTRABYTES 24
62
63 struct GranuleDef;
64
65 typedef struct MPADecodeContext {
66 DECLARE_ALIGNED_8(uint8_t, last_buf[2*BACKSTEP_SIZE + EXTRABYTES]);
67 int last_buf_size;
68 int frame_size;
69 /* next header (used in free format parsing) */
70 uint32_t free_format_next_header;
71 int error_protection;
72 int layer;
73 int sample_rate;
74 int sample_rate_index; /* between 0 and 8 */
75 int bit_rate;
76 GetBitContext gb;
77 GetBitContext in_gb;
78 int nb_channels;
79 int mode;
80 int mode_ext;
81 int lsf;
82 MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));
83 int synth_buf_offset[MPA_MAX_CHANNELS];
84 int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));
85 int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
86 #ifdef DEBUG
87 int frame_count;
88 #endif
89 void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);
90 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
91 int dither_state;
92 } MPADecodeContext;
93
94 /**
95 * Context for MP3On4 decoder
96 */
97 typedef struct MP3On4DecodeContext {
98 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
99 int chan_cfg; ///< channel config number
100 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
101 } MP3On4DecodeContext;
102
103 /* layer 3 "granule" */
104 typedef struct GranuleDef {
105 uint8_t scfsi;
106 int part2_3_length;
107 int big_values;
108 int global_gain;
109 int scalefac_compress;
110 uint8_t block_type;
111 uint8_t switch_point;
112 int table_select[3];
113 int subblock_gain[3];
114 uint8_t scalefac_scale;
115 uint8_t count1table_select;
116 int region_size[3]; /* number of huffman codes in each region */
117 int preflag;
118 int short_start, long_end; /* long/short band indexes */
119 uint8_t scale_factors[40];
120 int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
121 } GranuleDef;
122
123 #define MODE_EXT_MS_STEREO 2
124 #define MODE_EXT_I_STEREO 1
125
126 /* layer 3 huffman tables */
127 typedef struct HuffTable {
128 int xsize;
129 const uint8_t *bits;
130 const uint16_t *codes;
131 } HuffTable;
132
133 #include "mpegaudiodectab.h"
134
135 static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
136 static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
137
138 /* vlc structure for decoding layer 3 huffman tables */
139 static VLC huff_vlc[16];
140 static VLC huff_quad_vlc[2];
141 /* computed from band_size_long */
142 static uint16_t band_index_long[9][23];
143 /* XXX: free when all decoders are closed */
144 #define TABLE_4_3_SIZE (8191 + 16)*4
145 static int8_t *table_4_3_exp;
146 static uint32_t *table_4_3_value;
147 static uint32_t exp_table[512];
148 static uint32_t expval_table[512][16];
149 /* intensity stereo coef table */
150 static int32_t is_table[2][16];
151 static int32_t is_table_lsf[2][2][16];
152 static int32_t csa_table[8][4];
153 static float csa_table_float[8][4];
154 static int32_t mdct_win[8][36];
155
156 /* lower 2 bits: modulo 3, higher bits: shift */
157 static uint16_t scale_factor_modshift[64];
158 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
159 static int32_t scale_factor_mult[15][3];
160 /* mult table for layer 2 group quantization */
161
162 #define SCALE_GEN(v) \
163 { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
164
165 static const int32_t scale_factor_mult2[3][3] = {
166 SCALE_GEN(4.0 / 3.0), /* 3 steps */
167 SCALE_GEN(4.0 / 5.0), /* 5 steps */
168 SCALE_GEN(4.0 / 9.0), /* 9 steps */
169 };
170
171 static MPA_INT window[512] __attribute__((aligned(16)));
172
173 /* layer 1 unscaling */
174 /* n = number of bits of the mantissa minus 1 */
175 static inline int l1_unscale(int n, int mant, int scale_factor)
176 {
177 int shift, mod;
178 int64_t val;
179
180 shift = scale_factor_modshift[scale_factor];
181 mod = shift & 3;
182 shift >>= 2;
183 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
184 shift += n;
185 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
186 return (int)((val + (1LL << (shift - 1))) >> shift);
187 }
188
189 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
190 {
191 int shift, mod, val;
192
193 shift = scale_factor_modshift[scale_factor];
194 mod = shift & 3;
195 shift >>= 2;
196
197 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
198 /* NOTE: at this point, 0 <= shift <= 21 */
199 if (shift > 0)
200 val = (val + (1 << (shift - 1))) >> shift;
201 return val;
202 }
203
204 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
205 static inline int l3_unscale(int value, int exponent)
206 {
207 unsigned int m;
208 int e;
209
210 e = table_4_3_exp [4*value + (exponent&3)];
211 m = table_4_3_value[4*value + (exponent&3)];
212 e -= (exponent >> 2);
213 assert(e>=1);
214 if (e > 31)
215 return 0;
216 m = (m + (1 << (e-1))) >> e;
217
218 return m;
219 }
220
221 /* all integer n^(4/3) computation code */
222 #define DEV_ORDER 13
223
224 #define POW_FRAC_BITS 24
225 #define POW_FRAC_ONE (1 << POW_FRAC_BITS)
226 #define POW_FIX(a) ((int)((a) * POW_FRAC_ONE))
227 #define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
228
229 static int dev_4_3_coefs[DEV_ORDER];
230
231 #if 0 /* unused */
232 static int pow_mult3[3] = {
233 POW_FIX(1.0),
234 POW_FIX(1.25992104989487316476),
235 POW_FIX(1.58740105196819947474),
236 };
237 #endif
238
239 static void int_pow_init(void)
240 {
241 int i, a;
242
243 a = POW_FIX(1.0);
244 for(i=0;i<DEV_ORDER;i++) {
245 a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
246 dev_4_3_coefs[i] = a;
247 }
248 }
249
250 #if 0 /* unused, remove? */
251 /* return the mantissa and the binary exponent */
252 static int int_pow(int i, int *exp_ptr)
253 {
254 int e, er, eq, j;
255 int a, a1;
256
257 /* renormalize */
258 a = i;
259 e = POW_FRAC_BITS;
260 while (a < (1 << (POW_FRAC_BITS - 1))) {
261 a = a << 1;
262 e--;
263 }
264 a -= (1 << POW_FRAC_BITS);
265 a1 = 0;
266 for(j = DEV_ORDER - 1; j >= 0; j--)
267 a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
268 a = (1 << POW_FRAC_BITS) + a1;
269 /* exponent compute (exact) */
270 e = e * 4;
271 er = e % 3;
272 eq = e / 3;
273 a = POW_MULL(a, pow_mult3[er]);
274 while (a >= 2 * POW_FRAC_ONE) {
275 a = a >> 1;
276 eq++;
277 }
278 /* convert to float */
279 while (a < POW_FRAC_ONE) {
280 a = a << 1;
281 eq--;
282 }
283 /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
284 #if POW_FRAC_BITS > FRAC_BITS
285 a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
286 /* correct overflow */
287 if (a >= 2 * (1 << FRAC_BITS)) {
288 a = a >> 1;
289 eq++;
290 }
291 #endif
292 *exp_ptr = eq;
293 return a;
294 }
295 #endif
296
297 static int decode_init(AVCodecContext * avctx)
298 {
299 MPADecodeContext *s = avctx->priv_data;
300 static int init=0;
301 int i, j, k;
302
303 #if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
304 avctx->sample_fmt= SAMPLE_FMT_S32;
305 #else
306 avctx->sample_fmt= SAMPLE_FMT_S16;
307 #endif
308
309 if(avctx->antialias_algo != FF_AA_FLOAT)
310 s->compute_antialias= compute_antialias_integer;
311 else
312 s->compute_antialias= compute_antialias_float;
313
314 if (!init && !avctx->parse_only) {
315 /* scale factors table for layer 1/2 */
316 for(i=0;i<64;i++) {
317 int shift, mod;
318 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
319 shift = (i / 3);
320 mod = i % 3;
321 scale_factor_modshift[i] = mod | (shift << 2);
322 }
323
324 /* scale factor multiply for layer 1 */
325 for(i=0;i<15;i++) {
326 int n, norm;
327 n = i + 2;
328 norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
329 scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
330 scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
331 scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
332 dprintf("%d: norm=%x s=%x %x %x\n",
333 i, norm,
334 scale_factor_mult[i][0],
335 scale_factor_mult[i][1],
336 scale_factor_mult[i][2]);
337 }
338
339 ff_mpa_synth_init(window);
340
341 /* huffman decode tables */
342 for(i=1;i<16;i++) {
343 const HuffTable *h = &mpa_huff_tables[i];
344 int xsize, x, y;
345 unsigned int n;
346 uint8_t tmp_bits [512];
347 uint16_t tmp_codes[512];
348
349 memset(tmp_bits , 0, sizeof(tmp_bits ));
350 memset(tmp_codes, 0, sizeof(tmp_codes));
351
352 xsize = h->xsize;
353 n = xsize * xsize;
354
355 j = 0;
356 for(x=0;x<xsize;x++) {
357 for(y=0;y<xsize;y++){
358 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
359 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
360 }
361 }
362
363 /* XXX: fail test */
364 init_vlc(&huff_vlc[i], 7, 512,
365 tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
366 }
367 for(i=0;i<2;i++) {
368 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
369 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
370 }
371
372 for(i=0;i<9;i++) {
373 k = 0;
374 for(j=0;j<22;j++) {
375 band_index_long[i][j] = k;
376 k += band_size_long[i][j];
377 }
378 band_index_long[i][22] = k;
379 }
380
381 /* compute n ^ (4/3) and store it in mantissa/exp format */
382 table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));
383 if(!table_4_3_exp)
384 return -1;
385 table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));
386 if(!table_4_3_value)
387 return -1;
388
389 int_pow_init();
390 for(i=1;i<TABLE_4_3_SIZE;i++) {
391 double f, fm;
392 int e, m;
393 f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
394 fm = frexp(f, &e);
395 m = (uint32_t)(fm*(1LL<<31) + 0.5);
396 e+= FRAC_BITS - 31 + 5 - 100;
397
398 /* normalized to FRAC_BITS */
399 table_4_3_value[i] = m;
400 // av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
401 table_4_3_exp[i] = -e;
402 }
403 for(i=0; i<512*16; i++){
404 int exponent= (i>>4);
405 double f= pow(i&15, 4.0 / 3.0) * pow(2, (exponent-400)*0.25 + FRAC_BITS + 5);
406 expval_table[exponent][i&15]= llrint(f);
407 if((i&15)==1)
408 exp_table[exponent]= llrint(f);
409 }
410
411 for(i=0;i<7;i++) {
412 float f;
413 int v;
414 if (i != 6) {
415 f = tan((double)i * M_PI / 12.0);
416 v = FIXR(f / (1.0 + f));
417 } else {
418 v = FIXR(1.0);
419 }
420 is_table[0][i] = v;
421 is_table[1][6 - i] = v;
422 }
423 /* invalid values */
424 for(i=7;i<16;i++)
425 is_table[0][i] = is_table[1][i] = 0.0;
426
427 for(i=0;i<16;i++) {
428 double f;
429 int e, k;
430
431 for(j=0;j<2;j++) {
432 e = -(j + 1) * ((i + 1) >> 1);
433 f = pow(2.0, e / 4.0);
434 k = i & 1;
435 is_table_lsf[j][k ^ 1][i] = FIXR(f);
436 is_table_lsf[j][k][i] = FIXR(1.0);
437 dprintf("is_table_lsf %d %d: %x %x\n",
438 i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
439 }
440 }
441
442 for(i=0;i<8;i++) {
443 float ci, cs, ca;
444 ci = ci_table[i];
445 cs = 1.0 / sqrt(1.0 + ci * ci);
446 ca = cs * ci;
447 csa_table[i][0] = FIXHR(cs/4);
448 csa_table[i][1] = FIXHR(ca/4);
449 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
450 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
451 csa_table_float[i][0] = cs;
452 csa_table_float[i][1] = ca;
453 csa_table_float[i][2] = ca + cs;
454 csa_table_float[i][3] = ca - cs;
455 // printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
456 // av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
457 }
458
459 /* compute mdct windows */
460 for(i=0;i<36;i++) {
461 for(j=0; j<4; j++){
462 double d;
463
464 if(j==2 && i%3 != 1)
465 continue;
466
467 d= sin(M_PI * (i + 0.5) / 36.0);
468 if(j==1){
469 if (i>=30) d= 0;
470 else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
471 else if(i>=18) d= 1;
472 }else if(j==3){
473 if (i< 6) d= 0;
474 else if(i< 12) d= sin(M_PI * (i - 6 + 0.5) / 12.0);
475 else if(i< 18) d= 1;
476 }
477 //merge last stage of imdct into the window coefficients
478 d*= 0.5 / cos(M_PI*(2*i + 19)/72);
479
480 if(j==2)
481 mdct_win[j][i/3] = FIXHR((d / (1<<5)));
482 else
483 mdct_win[j][i ] = FIXHR((d / (1<<5)));
484 // av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
485 }
486 }
487
488 /* NOTE: we do frequency inversion adter the MDCT by changing
489 the sign of the right window coefs */
490 for(j=0;j<4;j++) {
491 for(i=0;i<36;i+=2) {
492 mdct_win[j + 4][i] = mdct_win[j][i];
493 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
494 }
495 }
496
497 #if defined(DEBUG)
498 for(j=0;j<8;j++) {
499 av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
500 for(i=0;i<36;i++)
501 av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
502 av_log(avctx, AV_LOG_DEBUG, "\n");
503 }
504 #endif
505 init = 1;
506 }
507
508 #ifdef DEBUG
509 s->frame_count = 0;
510 #endif
511 if (avctx->codec_id == CODEC_ID_MP3ADU)
512 s->adu_mode = 1;
513 return 0;
514 }
515
516 /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
517
518 /* cos(i*pi/64) */
519
520 #define COS0_0 FIXHR(0.50060299823519630134/2)
521 #define COS0_1 FIXHR(0.50547095989754365998/2)
522 #define COS0_2 FIXHR(0.51544730992262454697/2)
523 #define COS0_3 FIXHR(0.53104259108978417447/2)
524 #define COS0_4 FIXHR(0.55310389603444452782/2)
525 #define COS0_5 FIXHR(0.58293496820613387367/2)
526 #define COS0_6 FIXHR(0.62250412303566481615/2)
527 #define COS0_7 FIXHR(0.67480834145500574602/2)
528 #define COS0_8 FIXHR(0.74453627100229844977/2)
529 #define COS0_9 FIXHR(0.83934964541552703873/2)
530 #define COS0_10 FIXHR(0.97256823786196069369/2)
531 #define COS0_11 FIXHR(1.16943993343288495515/4)
532 #define COS0_12 FIXHR(1.48416461631416627724/4)
533 #define COS0_13 FIXHR(2.05778100995341155085/8)
534 #define COS0_14 FIXHR(3.40760841846871878570/8)
535 #define COS0_15 FIXHR(10.19000812354805681150/32)
536
537 #define COS1_0 FIXHR(0.50241928618815570551/2)
538 #define COS1_1 FIXHR(0.52249861493968888062/2)
539 #define COS1_2 FIXHR(0.56694403481635770368/2)
540 #define COS1_3 FIXHR(0.64682178335999012954/2)
541 #define COS1_4 FIXHR(0.78815462345125022473/2)
542 #define COS1_5 FIXHR(1.06067768599034747134/4)
543 #define COS1_6 FIXHR(1.72244709823833392782/4)
544 #define COS1_7 FIXHR(5.10114861868916385802/16)
545
546 #define COS2_0 FIXHR(0.50979557910415916894/2)
547 #define COS2_1 FIXHR(0.60134488693504528054/2)
548 #define COS2_2 FIXHR(0.89997622313641570463/2)
549 #define COS2_3 FIXHR(2.56291544774150617881/8)
550
551 #define COS3_0 FIXHR(0.54119610014619698439/2)
552 #define COS3_1 FIXHR(1.30656296487637652785/4)
553
554 #define COS4_0 FIXHR(0.70710678118654752439/2)
555
556 /* butterfly operator */
557 #define BF(a, b, c, s)\
558 {\
559 tmp0 = tab[a] + tab[b];\
560 tmp1 = tab[a] - tab[b];\
561 tab[a] = tmp0;\
562 tab[b] = MULH(tmp1<<(s), c);\
563 }
564
565 #define BF1(a, b, c, d)\
566 {\
567 BF(a, b, COS4_0, 1);\
568 BF(c, d,-COS4_0, 1);\
569 tab[c] += tab[d];\
570 }
571
572 #define BF2(a, b, c, d)\
573 {\
574 BF(a, b, COS4_0, 1);\
575 BF(c, d,-COS4_0, 1);\
576 tab[c] += tab[d];\
577 tab[a] += tab[c];\
578 tab[c] += tab[b];\
579 tab[b] += tab[d];\
580 }
581
582 #define ADD(a, b) tab[a] += tab[b]
583
584 /* DCT32 without 1/sqrt(2) coef zero scaling. */
585 static void dct32(int32_t *out, int32_t *tab)
586 {
587 int tmp0, tmp1;
588
589 /* pass 1 */
590 BF( 0, 31, COS0_0 , 1);
591 BF(15, 16, COS0_15, 5);
592 /* pass 2 */
593 BF( 0, 15, COS1_0 , 1);
594 BF(16, 31,-COS1_0 , 1);
595 /* pass 1 */
596 BF( 7, 24, COS0_7 , 1);
597 BF( 8, 23, COS0_8 , 1);
598 /* pass 2 */
599 BF( 7, 8, COS1_7 , 4);
600 BF(23, 24,-COS1_7 , 4);
601 /* pass 3 */
602 BF( 0, 7, COS2_0 , 1);
603 BF( 8, 15,-COS2_0 , 1);
604 BF(16, 23, COS2_0 , 1);
605 BF(24, 31,-COS2_0 , 1);
606 /* pass 1 */
607 BF( 3, 28, COS0_3 , 1);
608 BF(12, 19, COS0_12, 2);
609 /* pass 2 */
610 BF( 3, 12, COS1_3 , 1);
611 BF(19, 28,-COS1_3 , 1);
612 /* pass 1 */
613 BF( 4, 27, COS0_4 , 1);
614 BF(11, 20, COS0_11, 2);
615 /* pass 2 */
616 BF( 4, 11, COS1_4 , 1);
617 BF(20, 27,-COS1_4 , 1);
618 /* pass 3 */
619 BF( 3, 4, COS2_3 , 3);
620 BF(11, 12,-COS2_3 , 3);
621 BF(19, 20, COS2_3 , 3);
622 BF(27, 28,-COS2_3 , 3);
623 /* pass 4 */
624 BF( 0, 3, COS3_0 , 1);
625 BF( 4, 7,-COS3_0 , 1);
626 BF( 8, 11, COS3_0 , 1);
627 BF(12, 15,-COS3_0 , 1);
628 BF(16, 19, COS3_0 , 1);
629 BF(20, 23,-COS3_0 , 1);
630 BF(24, 27, COS3_0 , 1);
631 BF(28, 31,-COS3_0 , 1);
632
633
634
635 /* pass 1 */
636 BF( 1, 30, COS0_1 , 1);
637 BF(14, 17, COS0_14, 3);
638 /* pass 2 */
639 BF( 1, 14, COS1_1 , 1);
640 BF(17, 30,-COS1_1 , 1);
641 /* pass 1 */
642 BF( 6, 25, COS0_6 , 1);
643 BF( 9, 22, COS0_9 , 1);
644 /* pass 2 */
645 BF( 6, 9, COS1_6 , 2);
646 BF(22, 25,-COS1_6 , 2);
647 /* pass 3 */
648 BF( 1, 6, COS2_1 , 1);
649 BF( 9, 14,-COS2_1 , 1);
650 BF(17, 22, COS2_1 , 1);
651 BF(25, 30,-COS2_1 , 1);
652
653 /* pass 1 */
654 BF( 2, 29, COS0_2 , 1);
655 BF(13, 18, COS0_13, 3);
656 /* pass 2 */
657 BF( 2, 13, COS1_2 , 1);
658 BF(18, 29,-COS1_2 , 1);
659 /* pass 1 */
660 BF( 5, 26, COS0_5 , 1);
661 BF(10, 21, COS0_10, 1);
662 /* pass 2 */
663 BF( 5, 10, COS1_5 , 2);
664 BF(21, 26,-COS1_5 , 2);
665 /* pass 3 */
666 BF( 2, 5, COS2_2 , 1);
667 BF(10, 13,-COS2_2 , 1);
668 BF(18, 21, COS2_2 , 1);
669 BF(26, 29,-COS2_2 , 1);
670 /* pass 4 */
671 BF( 1, 2, COS3_1 , 2);
672 BF( 5, 6,-COS3_1 , 2);
673 BF( 9, 10, COS3_1 , 2);
674 BF(13, 14,-COS3_1 , 2);
675 BF(17, 18, COS3_1 , 2);
676 BF(21, 22,-COS3_1 , 2);
677 BF(25, 26, COS3_1 , 2);
678 BF(29, 30,-COS3_1 , 2);
679
680 /* pass 5 */
681 BF1( 0, 1, 2, 3);
682 BF2( 4, 5, 6, 7);
683 BF1( 8, 9, 10, 11);
684 BF2(12, 13, 14, 15);
685 BF1(16, 17, 18, 19);
686 BF2(20, 21, 22, 23);
687 BF1(24, 25, 26, 27);
688 BF2(28, 29, 30, 31);
689
690 /* pass 6 */
691
692 ADD( 8, 12);
693 ADD(12, 10);
694 ADD(10, 14);
695 ADD(14, 9);
696 ADD( 9, 13);
697 ADD(13, 11);
698 ADD(11, 15);
699
700 out[ 0] = tab[0];
701 out[16] = tab[1];
702 out[ 8] = tab[2];
703 out[24] = tab[3];
704 out[ 4] = tab[4];
705 out[20] = tab[5];
706 out[12] = tab[6];
707 out[28] = tab[7];
708 out[ 2] = tab[8];
709 out[18] = tab[9];
710 out[10] = tab[10];
711 out[26] = tab[11];
712 out[ 6] = tab[12];
713 out[22] = tab[13];
714 out[14] = tab[14];
715 out[30] = tab[15];
716
717 ADD(24, 28);
718 ADD(28, 26);
719 ADD(26, 30);
720 ADD(30, 25);
721 ADD(25, 29);
722 ADD(29, 27);
723 ADD(27, 31);
724
725 out[ 1] = tab[16] + tab[24];
726 out[17] = tab[17] + tab[25];
727 out[ 9] = tab[18] + tab[26];
728 out[25] = tab[19] + tab[27];
729 out[ 5] = tab[20] + tab[28];
730 out[21] = tab[21] + tab[29];
731 out[13] = tab[22] + tab[30];
732 out[29] = tab[23] + tab[31];
733 out[ 3] = tab[24] + tab[20];
734 out[19] = tab[25] + tab[21];
735 out[11] = tab[26] + tab[22];
736 out[27] = tab[27] + tab[23];
737 out[ 7] = tab[28] + tab[18];
738 out[23] = tab[29] + tab[19];
739 out[15] = tab[30] + tab[17];
740 out[31] = tab[31];
741 }
742
743 #if FRAC_BITS <= 15
744
745 static inline int round_sample(int *sum)
746 {
747 int sum1;
748 sum1 = (*sum) >> OUT_SHIFT;
749 *sum &= (1<<OUT_SHIFT)-1;
750 if (sum1 < OUT_MIN)
751 sum1 = OUT_MIN;
752 else if (sum1 > OUT_MAX)
753 sum1 = OUT_MAX;
754 return sum1;
755 }
756
757 /* signed 16x16 -> 32 multiply add accumulate */
758 #define MACS(rt, ra, rb) MAC16(rt, ra, rb)
759
760 /* signed 16x16 -> 32 multiply */
761 #define MULS(ra, rb) MUL16(ra, rb)
762
763 #else
764
765 static inline int round_sample(int64_t *sum)
766 {
767 int sum1;
768 sum1 = (int)((*sum) >> OUT_SHIFT);
769 *sum &= (1<<OUT_SHIFT)-1;
770 if (sum1 < OUT_MIN)
771 sum1 = OUT_MIN;
772 else if (sum1 > OUT_MAX)
773 sum1 = OUT_MAX;
774 return sum1;
775 }
776
777 # define MULS(ra, rb) MUL64(ra, rb)
778 #endif
779
780 #define SUM8(sum, op, w, p) \
781 { \
782 sum op MULS((w)[0 * 64], p[0 * 64]);\
783 sum op MULS((w)[1 * 64], p[1 * 64]);\
784 sum op MULS((w)[2 * 64], p[2 * 64]);\
785 sum op MULS((w)[3 * 64], p[3 * 64]);\
786 sum op MULS((w)[4 * 64], p[4 * 64]);\
787 sum op MULS((w)[5 * 64], p[5 * 64]);\
788 sum op MULS((w)[6 * 64], p[6 * 64]);\
789 sum op MULS((w)[7 * 64], p[7 * 64]);\
790 }
791
792 #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
793 { \
794 int tmp;\
795 tmp = p[0 * 64];\
796 sum1 op1 MULS((w1)[0 * 64], tmp);\
797 sum2 op2 MULS((w2)[0 * 64], tmp);\
798 tmp = p[1 * 64];\
799 sum1 op1 MULS((w1)[1 * 64], tmp);\
800 sum2 op2 MULS((w2)[1 * 64], tmp);\
801 tmp = p[2 * 64];\
802 sum1 op1 MULS((w1)[2 * 64], tmp);\
803 sum2 op2 MULS((w2)[2 * 64], tmp);\
804 tmp = p[3 * 64];\
805 sum1 op1 MULS((w1)[3 * 64], tmp);\
806 sum2 op2 MULS((w2)[3 * 64], tmp);\
807 tmp = p[4 * 64];\
808 sum1 op1 MULS((w1)[4 * 64], tmp);\
809 sum2 op2 MULS((w2)[4 * 64], tmp);\
810 tmp = p[5 * 64];\
811 sum1 op1 MULS((w1)[5 * 64], tmp);\
812 sum2 op2 MULS((w2)[5 * 64], tmp);\
813 tmp = p[6 * 64];\
814 sum1 op1 MULS((w1)[6 * 64], tmp);\
815 sum2 op2 MULS((w2)[6 * 64], tmp);\
816 tmp = p[7 * 64];\
817 sum1 op1 MULS((w1)[7 * 64], tmp);\
818 sum2 op2 MULS((w2)[7 * 64], tmp);\
819 }
820
821 void ff_mpa_synth_init(MPA_INT *window)
822 {
823 int i;
824
825 /* max = 18760, max sum over all 16 coefs : 44736 */
826 for(i=0;i<257;i++) {
827 int v;
828 v = mpa_enwindow[i];
829 #if WFRAC_BITS < 16
830 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
831 #endif
832 window[i] = v;
833 if ((i & 63) != 0)
834 v = -v;
835 if (i != 0)
836 window[512 - i] = v;
837 }
838 }
839
840 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
841 32 samples. */
842 /* XXX: optimize by avoiding ring buffer usage */
843 void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
844 MPA_INT *window, int *dither_state,
845 OUT_INT *samples, int incr,
846 int32_t sb_samples[SBLIMIT])
847 {
848 int32_t tmp[32];
849 register MPA_INT *synth_buf;
850 register const MPA_INT *w, *w2, *p;
851 int j, offset, v;
852 OUT_INT *samples2;
853 #if FRAC_BITS <= 15
854 int sum, sum2;
855 #else
856 int64_t sum, sum2;
857 #endif
858
859 dct32(tmp, sb_samples);
860
861 offset = *synth_buf_offset;
862 synth_buf = synth_buf_ptr + offset;
863
864 for(j=0;j<32;j++) {
865 v = tmp[j];
866 #if FRAC_BITS <= 15
867 /* NOTE: can cause a loss in precision if very high amplitude
868 sound */
869 if (v > 32767)
870 v = 32767;
871 else if (v < -32768)
872 v = -32768;
873 #endif
874 synth_buf[j] = v;
875 }
876 /* copy to avoid wrap */
877 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
878
879 samples2 = samples + 31 * incr;
880 w = window;
881 w2 = window + 31;
882
883 sum = *dither_state;
884 p = synth_buf + 16;
885 SUM8(sum, +=, w, p);
886 p = synth_buf + 48;
887 SUM8(sum, -=, w + 32, p);
888 *samples = round_sample(&sum);
889 samples += incr;
890 w++;
891
892 /* we calculate two samples at the same time to avoid one memory
893 access per two sample */
894 for(j=1;j<16;j++) {
895 sum2 = 0;
896 p = synth_buf + 16 + j;
897 SUM8P2(sum, +=, sum2, -=, w, w2, p);
898 p = synth_buf + 48 - j;
899 SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
900
901 *samples = round_sample(&sum);
902 samples += incr;
903 sum += sum2;
904 *samples2 = round_sample(&sum);
905 samples2 -= incr;
906 w++;
907 w2--;
908 }
909
910 p = synth_buf + 32;
911 SUM8(sum, -=, w + 32, p);
912 *samples = round_sample(&sum);
913 *dither_state= sum;
914
915 offset = (offset - 32) & 511;
916 *synth_buf_offset = offset;
917 }
918
919 #define C3 FIXHR(0.86602540378443864676/2)
920
921 /* 0.5 / cos(pi*(2*i+1)/36) */
922 static const int icos36[9] = {
923 FIXR(0.50190991877167369479),
924 FIXR(0.51763809020504152469), //0
925 FIXR(0.55168895948124587824),
926 FIXR(0.61038729438072803416),
927 FIXR(0.70710678118654752439), //1
928 FIXR(0.87172339781054900991),
929 FIXR(1.18310079157624925896),
930 FIXR(1.93185165257813657349), //2
931 FIXR(5.73685662283492756461),
932 };
933
934 /* 0.5 / cos(pi*(2*i+1)/36) */
935 static const int icos36h[9] = {
936 FIXHR(0.50190991877167369479/2),
937 FIXHR(0.51763809020504152469/2), //0
938 FIXHR(0.55168895948124587824/2),
939 FIXHR(0.61038729438072803416/2),
940 FIXHR(0.70710678118654752439/2), //1
941 FIXHR(0.87172339781054900991/2),
942 FIXHR(1.18310079157624925896/4),
943 FIXHR(1.93185165257813657349/4), //2
944 // FIXHR(5.73685662283492756461),
945 };
946
947 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
948 cases. */
949 static void imdct12(int *out, int *in)
950 {
951 int in0, in1, in2, in3, in4, in5, t1, t2;
952
953 in0= in[0*3];
954 in1= in[1*3] + in[0*3];
955 in2= in[2*3] + in[1*3];
956 in3= in[3*3] + in[2*3];
957 in4= in[4*3] + in[3*3];
958 in5= in[5*3] + in[4*3];
959 in5 += in3;
960 in3 += in1;
961
962 in2= MULH(2*in2, C3);
963 in3= MULH(4*in3, C3);
964
965 t1 = in0 - in4;
966 t2 = MULH(2*(in1 - in5), icos36h[4]);
967
968 out[ 7]=
969 out[10]= t1 + t2;
970 out[ 1]=
971 out[ 4]= t1 - t2;
972
973 in0 += in4>>1;
974 in4 = in0 + in2;
975 in5 += 2*in1;
976 in1 = MULH(in5 + in3, icos36h[1]);
977 out[ 8]=
978 out[ 9]= in4 + in1;
979 out[ 2]=
980 out[ 3]= in4 - in1;
981
982 in0 -= in2;
983 in5 = MULH(2*(in5 - in3), icos36h[7]);
984 out[ 0]=
985 out[ 5]= in0 - in5;
986 out[ 6]=
987 out[11]= in0 + in5;
988 }
989
990 /* cos(pi*i/18) */
991 #define C1 FIXHR(0.98480775301220805936/2)
992 #define C2 FIXHR(0.93969262078590838405/2)
993 #define C3 FIXHR(0.86602540378443864676/2)
994 #define C4 FIXHR(0.76604444311897803520/2)
995 #define C5 FIXHR(0.64278760968653932632/2)
996 #define C6 FIXHR(0.5/2)
997 #define C7 FIXHR(0.34202014332566873304/2)
998 #define C8 FIXHR(0.17364817766693034885/2)
999
1000
1001 /* using Lee like decomposition followed by hand coded 9 points DCT */
1002 static void imdct36(int *out, int *buf, int *in, int *win)
1003 {
1004 int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1005 int tmp[18], *tmp1, *in1;
1006
1007 for(i=17;i>=1;i--)
1008 in[i] += in[i-1];
1009 for(i=17;i>=3;i-=2)
1010 in[i] += in[i-2];
1011
1012 for(j=0;j<2;j++) {
1013 tmp1 = tmp + j;
1014 in1 = in + j;
1015 #if 0
1016 //more accurate but slower
1017 int64_t t0, t1, t2, t3;
1018 t2 = in1[2*4] + in1[2*8] - in1[2*2];
1019
1020 t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1021 t1 = in1[2*0] - in1[2*6];
1022 tmp1[ 6] = t1 - (t2>>1);
1023 tmp1[16] = t1 + t2;
1024
1025 t0 = MUL64(2*(in1[2*2] + in1[2*4]), C2);
1026 t1 = MUL64( in1[2*4] - in1[2*8] , -2*C8);
1027 t2 = MUL64(2*(in1[2*2] + in1[2*8]), -C4);
1028
1029 tmp1[10] = (t3 - t0 - t2) >> 32;
1030 tmp1[ 2] = (t3 + t0 + t1) >> 32;
1031 tmp1[14] = (t3 + t2 - t1) >> 32;
1032
1033 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1034 t2 = MUL64(2*(in1[2*1] + in1[2*5]), C1);
1035 t3 = MUL64( in1[2*5] - in1[2*7] , -2*C7);
1036 t0 = MUL64(2*in1[2*3], C3);
1037
1038 t1 = MUL64(2*(in1[2*1] + in1[2*7]), -C5);
1039
1040 tmp1[ 0] = (t2 + t3 + t0) >> 32;
1041 tmp1[12] = (t2 + t1 - t0) >> 32;
1042 tmp1[ 8] = (t3 - t1 - t0) >> 32;
1043 #else
1044 t2 = in1[2*4] + in1[2*8] - in1[2*2];
1045
1046 t3 = in1[2*0] + (in1[2*6]>>1);
1047 t1 = in1[2*0] - in1[2*6];
1048 tmp1[ 6] = t1 - (t2>>1);
1049 tmp1[16] = t1 + t2;
1050
1051 t0 = MULH(2*(in1[2*2] + in1[2*4]), C2);
1052 t1 = MULH( in1[2*4] - in1[2*8] , -2*C8);
1053 t2 = MULH(2*(in1[2*2] + in1[2*8]), -C4);
1054
1055 tmp1[10] = t3 - t0 - t2;
1056 tmp1[ 2] = t3 + t0 + t1;
1057 tmp1[14] = t3 + t2 - t1;
1058
1059 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1060 t2 = MULH(2*(in1[2*1] + in1[2*5]), C1);
1061 t3 = MULH( in1[2*5] - in1[2*7] , -2*C7);
1062 t0 = MULH(2*in1[2*3], C3);
1063
1064 t1 = MULH(2*(in1[2*1] + in1[2*7]), -C5);
1065
1066 tmp1[ 0] = t2 + t3 + t0;
1067 tmp1[12] = t2 + t1 - t0;
1068 tmp1[ 8] = t3 - t1 - t0;
1069 #endif
1070 }
1071
1072 i = 0;
1073 for(j=0;j<4;j++) {
1074 t0 = tmp[i];
1075 t1 = tmp[i + 2];
1076 s0 = t1 + t0;
1077 s2 = t1 - t0;
1078
1079 t2 = tmp[i + 1];
1080 t3 = tmp[i + 3];
1081 s1 = MULH(2*(t3 + t2), icos36h[j]);
1082 s3 = MULL(t3 - t2, icos36[8 - j]);
1083
1084 t0 = s0 + s1;
1085 t1 = s0 - s1;
1086 out[(9 + j)*SBLIMIT] = MULH(t1, win[9 + j]) + buf[9 + j];
1087 out[(8 - j)*SBLIMIT] = MULH(t1, win[8 - j]) + buf[8 - j];
1088 buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1089 buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1090
1091 t0 = s2 + s3;
1092 t1 = s2 - s3;
1093 out[(9 + 8 - j)*SBLIMIT] = MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1094 out[( j)*SBLIMIT] = MULH(t1, win[ j]) + buf[ j];
1095 buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1096 buf[ + j] = MULH(t0, win[18 + j]);
1097 i += 4;
1098 }
1099
1100 s0 = tmp[16];
1101 s1 = MULH(2*tmp[17], icos36h[4]);
1102 t0 = s0 + s1;
1103 t1 = s0 - s1;
1104 out[(9 + 4)*SBLIMIT] = MULH(t1, win[9 + 4]) + buf[9 + 4];
1105 out[(8 - 4)*SBLIMIT] = MULH(t1, win[8 - 4]) + buf[8 - 4];
1106 buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1107 buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1108 }
1109
1110 /* header decoding. MUST check the header before because no
1111 consistency check is done there. Return 1 if free format found and
1112 that the frame size must be computed externally */
1113 static int decode_header(MPADecodeContext *s, uint32_t header)
1114 {
1115 int sample_rate, frame_size, mpeg25, padding;
1116 int sample_rate_index, bitrate_index;
1117 if (header & (1<<20)) {
1118 s->lsf = (header & (1<<19)) ? 0 : 1;
1119 mpeg25 = 0;
1120 } else {
1121 s->lsf = 1;
1122 mpeg25 = 1;
1123 }
1124
1125 s->layer = 4 - ((header >> 17) & 3);
1126 /* extract frequency */
1127 sample_rate_index = (header >> 10) & 3;
1128 sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1129 sample_rate_index += 3 * (s->lsf + mpeg25);
1130 s->sample_rate_index = sample_rate_index;
1131 s->error_protection = ((header >> 16) & 1) ^ 1;
1132 s->sample_rate = sample_rate;
1133
1134 bitrate_index = (header >> 12) & 0xf;
1135 padding = (header >> 9) & 1;
1136 //extension = (header >> 8) & 1;
1137 s->mode = (header >> 6) & 3;
1138 s->mode_ext = (header >> 4) & 3;
1139 //copyright = (header >> 3) & 1;
1140 //original = (header >> 2) & 1;
1141 //emphasis = header & 3;
1142
1143 if (s->mode == MPA_MONO)
1144 s->nb_channels = 1;
1145 else
1146 s->nb_channels = 2;
1147
1148 if (bitrate_index != 0) {
1149 frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1150 s->bit_rate = frame_size * 1000;
1151 switch(s->layer) {
1152 case 1:
1153 frame_size = (frame_size * 12000) / sample_rate;
1154 frame_size = (frame_size + padding) * 4;
1155 break;
1156 case 2:
1157 frame_size = (frame_size * 144000) / sample_rate;
1158 frame_size += padding;
1159 break;
1160 default:
1161 case 3:
1162 frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1163 frame_size += padding;
1164 break;
1165 }
1166 s->frame_size = frame_size;
1167 } else {
1168 /* if no frame size computed, signal it */
1169 return 1;
1170 }
1171
1172 #if defined(DEBUG)
1173 dprintf("layer%d, %d Hz, %d kbits/s, ",
1174 s->layer, s->sample_rate, s->bit_rate);
1175 if (s->nb_channels == 2) {
1176 if (s->layer == 3) {
1177 if (s->mode_ext & MODE_EXT_MS_STEREO)
1178 dprintf("ms-");
1179 if (s->mode_ext & MODE_EXT_I_STEREO)
1180 dprintf("i-");
1181 }
1182 dprintf("stereo");
1183 } else {
1184 dprintf("mono");
1185 }
1186 dprintf("\n");
1187 #endif
1188 return 0;
1189 }
1190
1191 /* useful helper to get mpeg audio stream infos. Return -1 if error in
1192 header, otherwise the coded frame size in bytes */
1193 int mpa_decode_header(AVCodecContext *avctx, uint32_t head)
1194 {
1195 MPADecodeContext s1, *s = &s1;
1196
1197 if (ff_mpa_check_header(head) != 0)
1198 return -1;
1199
1200 if (decode_header(s, head) != 0) {
1201 return -1;
1202 }
1203
1204 switch(s->layer) {
1205 case 1:
1206 avctx->frame_size = 384;
1207 break;
1208 case 2:
1209 avctx->frame_size = 1152;
1210 break;
1211 default:
1212 case 3:
1213 if (s->lsf)
1214 avctx->frame_size = 576;
1215 else
1216 avctx->frame_size = 1152;
1217 break;
1218 }
1219
1220 avctx->sample_rate = s->sample_rate;
1221 avctx->channels = s->nb_channels;
1222 avctx->bit_rate = s->bit_rate;
1223 avctx->sub_id = s->layer;
1224 return s->frame_size;
1225 }
1226
1227 /* return the number of decoded frames */
1228 static int mp_decode_layer1(MPADecodeContext *s)
1229 {
1230 int bound, i, v, n, ch, j, mant;
1231 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1232 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1233
1234 if (s->mode == MPA_JSTEREO)
1235 bound = (s->mode_ext + 1) * 4;
1236 else
1237 bound = SBLIMIT;
1238
1239 /* allocation bits */
1240 for(i=0;i<bound;i++) {
1241 for(ch=0;ch<s->nb_channels;ch++) {
1242 allocation[ch][i] = get_bits(&s->gb, 4);
1243 }
1244 }
1245 for(i=bound;i<SBLIMIT;i++) {
1246 allocation[0][i] = get_bits(&s->gb, 4);
1247 }
1248
1249 /* scale factors */
1250 for(i=0;i<bound;i++) {
1251 for(ch=0;ch<s->nb_channels;ch++) {
1252 if (allocation[ch][i])
1253 scale_factors[ch][i] = get_bits(&s->gb, 6);
1254 }
1255 }
1256 for(i=bound;i<SBLIMIT;i++) {
1257 if (allocation[0][i]) {
1258 scale_factors[0][i] = get_bits(&s->gb, 6);
1259 scale_factors[1][i] = get_bits(&s->gb, 6);
1260 }
1261 }
1262
1263 /* compute samples */
1264 for(j=0;j<12;j++) {
1265 for(i=0;i<bound;i++) {
1266 for(ch=0;ch<s->nb_channels;ch++) {
1267 n = allocation[ch][i];
1268 if (n) {
1269 mant = get_bits(&s->gb, n + 1);
1270 v = l1_unscale(n, mant, scale_factors[ch][i]);
1271 } else {
1272 v = 0;
1273 }
1274 s->sb_samples[ch][j][i] = v;
1275 }
1276 }
1277 for(i=bound;i<SBLIMIT;i++) {
1278 n = allocation[0][i];
1279 if (n) {
1280 mant = get_bits(&s->gb, n + 1);
1281 v = l1_unscale(n, mant, scale_factors[0][i]);
1282 s->sb_samples[0][j][i] = v;
1283 v = l1_unscale(n, mant, scale_factors[1][i]);
1284 s->sb_samples[1][j][i] = v;
1285 } else {
1286 s->sb_samples[0][j][i] = 0;
1287 s->sb_samples[1][j][i] = 0;
1288 }
1289 }
1290 }
1291 return 12;
1292 }
1293
1294 /* bitrate is in kb/s */
1295 int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1296 {
1297 int ch_bitrate, table;
1298
1299 ch_bitrate = bitrate / nb_channels;
1300 if (!lsf) {
1301 if ((freq == 48000 && ch_bitrate >= 56) ||
1302 (ch_bitrate >= 56 && ch_bitrate <= 80))
1303 table = 0;
1304 else if (freq != 48000 && ch_bitrate >= 96)
1305 table = 1;
1306 else if (freq != 32000 && ch_bitrate <= 48)
1307 table = 2;
1308 else
1309 table = 3;
1310 } else {
1311 table = 4;
1312 }
1313 return table;
1314 }
1315
1316 static int mp_decode_layer2(MPADecodeContext *s)
1317 {
1318 int sblimit; /* number of used subbands */
1319 const unsigned char *alloc_table;
1320 int table, bit_alloc_bits, i, j, ch, bound, v;
1321 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1322 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1323 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1324 int scale, qindex, bits, steps, k, l, m, b;
1325
1326 /* select decoding table */
1327 table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1328 s->sample_rate, s->lsf);
1329 sblimit = sblimit_table[table];
1330 alloc_table = alloc_tables[table];
1331
1332 if (s->mode == MPA_JSTEREO)
1333 bound = (s->mode_ext + 1) * 4;
1334 else
1335 bound = sblimit;
1336
1337 dprintf("bound=%d sblimit=%d\n", bound, sblimit);
1338
1339 /* sanity check */
1340 if( bound > sblimit ) bound = sblimit;
1341
1342 /* parse bit allocation */
1343 j = 0;
1344 for(i=0;i<bound;i++) {
1345 bit_alloc_bits = alloc_table[j];
1346 for(ch=0;ch<s->nb_channels;ch++) {
1347 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1348 }
1349 j += 1 << bit_alloc_bits;
1350 }
1351 for(i=bound;i<sblimit;i++) {
1352 bit_alloc_bits = alloc_table[j];
1353 v = get_bits(&s->gb, bit_alloc_bits);
1354 bit_alloc[0][i] = v;
1355 bit_alloc[1][i] = v;
1356 j += 1 << bit_alloc_bits;
1357 }
1358
1359 #ifdef DEBUG
1360 {
1361 for(ch=0;ch<s->nb_channels;ch++) {
1362 for(i=0;i<sblimit;i++)
1363 dprintf(" %d", bit_alloc[ch][i]);
1364 dprintf("\n");
1365 }
1366 }
1367 #endif
1368
1369 /* scale codes */
1370 for(i=0;i<sblimit;i++) {
1371 for(ch=0;ch<s->nb_channels;ch++) {
1372 if (bit_alloc[ch][i])
1373 scale_code[ch][i] = get_bits(&s->gb, 2);
1374 }
1375 }
1376
1377 /* scale factors */
1378 for(i=0;i<sblimit;i++) {
1379 for(ch=0;ch<s->nb_channels;ch++) {
1380 if (bit_alloc[ch][i]) {
1381 sf = scale_factors[ch][i];
1382 switch(scale_code[ch][i]) {
1383 default:
1384 case 0:
1385 sf[0] = get_bits(&s->gb, 6);
1386 sf[1] = get_bits(&s->gb, 6);
1387 sf[2] = get_bits(&s->gb, 6);
1388 break;
1389 case 2:
1390 sf[0] = get_bits(&s->gb, 6);
1391 sf[1] = sf[0];
1392 sf[2] = sf[0];
1393 break;
1394 case 1:
1395 sf[0] = get_bits(&s->gb, 6);
1396 sf[2] = get_bits(&s->gb, 6);
1397 sf[1] = sf[0];
1398 break;
1399 case 3:
1400 sf[0] = get_bits(&s->gb, 6);
1401 sf[2] = get_bits(&s->gb, 6);
1402 sf[1] = sf[2];
1403 break;
1404 }
1405 }
1406 }
1407 }
1408
1409 #ifdef DEBUG
1410 for(ch=0;ch<s->nb_channels;ch++) {
1411 for(i=0;i<sblimit;i++) {
1412 if (bit_alloc[ch][i]) {
1413 sf = scale_factors[ch][i];
1414 dprintf(" %d %d %d", sf[0], sf[1], sf[2]);
1415 } else {
1416 dprintf(" -");
1417 }
1418 }
1419 dprintf("\n");
1420 }
1421 #endif
1422
1423 /* samples */
1424 for(k=0;k<3;k++) {
1425 for(l=0;l<12;l+=3) {
1426 j = 0;
1427 for(i=0;i<bound;i++) {
1428 bit_alloc_bits = alloc_table[j];
1429 for(ch=0;ch<s->nb_channels;ch++) {
1430 b = bit_alloc[ch][i];
1431 if (b) {
1432 scale = scale_factors[ch][i][k];
1433 qindex = alloc_table[j+b];
1434 bits = quant_bits[qindex];
1435 if (bits < 0) {
1436 /* 3 values at the same time */
1437 v = get_bits(&s->gb, -bits);
1438 steps = quant_steps[qindex];
1439 s->sb_samples[ch][k * 12 + l + 0][i] =
1440 l2_unscale_group(steps, v % steps, scale);
1441 v = v / steps;
1442 s->sb_samples[ch][k * 12 + l + 1][i] =
1443 l2_unscale_group(steps, v % steps, scale);
1444 v = v / steps;
1445 s->sb_samples[ch][k * 12 + l + 2][i] =
1446 l2_unscale_group(steps, v, scale);
1447 } else {
1448 for(m=0;m<3;m++) {
1449 v = get_bits(&s->gb, bits);
1450 v = l1_unscale(bits - 1, v, scale);
1451 s->sb_samples[ch][k * 12 + l + m][i] = v;
1452 }
1453 }
1454 } else {
1455 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1456 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1457 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1458 }
1459 }
1460 /* next subband in alloc table */
1461 j += 1 << bit_alloc_bits;
1462 }
1463 /* XXX: find a way to avoid this duplication of code */
1464 for(i=bound;i<sblimit;i++) {
1465 bit_alloc_bits = alloc_table[j];
1466 b = bit_alloc[0][i];
1467 if (b) {
1468 int mant, scale0, scale1;
1469 scale0 = scale_factors[0][i][k];
1470 scale1 = scale_factors[1][i][k];
1471 qindex = alloc_table[j+b];
1472 bits = quant_bits[qindex];
1473 if (bits < 0) {
1474 /* 3 values at the same time */
1475 v = get_bits(&s->gb, -bits);
1476 steps = quant_steps[qindex];
1477 mant = v % steps;
1478 v = v / steps;
1479 s->sb_samples[0][k * 12 + l + 0][i] =
1480 l2_unscale_group(steps, mant, scale0);
1481 s->sb_samples[1][k * 12 + l + 0][i] =
1482 l2_unscale_group(steps, mant, scale1);
1483 mant = v % steps;
1484 v = v / steps;
1485 s->sb_samples[0][k * 12 + l + 1][i] =
1486 l2_unscale_group(steps, mant, scale0);
1487 s->sb_samples[1][k * 12 + l + 1][i] =
1488 l2_unscale_group(steps, mant, scale1);
1489 s->sb_samples[0][k * 12 + l + 2][i] =
1490 l2_unscale_group(steps, v, scale0);
1491 s->sb_samples[1][k * 12 + l + 2][i] =
1492 l2_unscale_group(steps, v, scale1);
1493 } else {
1494 for(m=0;m<3;m++) {
1495 mant = get_bits(&s->gb, bits);
1496 s->sb_samples[0][k * 12 + l + m][i] =
1497 l1_unscale(bits - 1, mant, scale0);
1498 s->sb_samples[1][k * 12 + l + m][i] =
1499 l1_unscale(bits - 1, mant, scale1);
1500 }
1501 }
1502 } else {
1503 s->sb_samples[0][k * 12 + l + 0][i] = 0;
1504 s->sb_samples[0][k * 12 + l + 1][i] = 0;
1505 s->sb_samples[0][k * 12 + l + 2][i] = 0;
1506 s->sb_samples[1][k * 12 + l + 0][i] = 0;
1507 s->sb_samples[1][k * 12 + l + 1][i] = 0;
1508 s->sb_samples[1][k * 12 + l + 2][i] = 0;
1509 }
1510 /* next subband in alloc table */
1511 j += 1 << bit_alloc_bits;
1512 }
1513 /* fill remaining samples to zero */
1514 for(i=sblimit;i<SBLIMIT;i++) {
1515 for(ch=0;ch<s->nb_channels;ch++) {
1516 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1517 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1518 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1519 }
1520 }
1521 }
1522 }
1523 return 3 * 12;
1524 }
1525
1526 static inline void lsf_sf_expand(int *slen,
1527 int sf, int n1, int n2, int n3)
1528 {
1529 if (n3) {
1530 slen[3] = sf % n3;
1531 sf /= n3;
1532 } else {
1533 slen[3] = 0;
1534 }
1535 if (n2) {
1536 slen[2] = sf % n2;
1537 sf /= n2;
1538 } else {
1539 slen[2] = 0;
1540 }
1541 slen[1] = sf % n1;
1542 sf /= n1;
1543 slen[0] = sf;
1544 }
1545
1546 static void exponents_from_scale_factors(MPADecodeContext *s,
1547 GranuleDef *g,
1548 int16_t *exponents)
1549 {
1550 const uint8_t *bstab, *pretab;
1551 int len, i, j, k, l, v0, shift, gain, gains[3];
1552 int16_t *exp_ptr;
1553
1554 exp_ptr = exponents;
1555 gain = g->global_gain - 210;
1556 shift = g->scalefac_scale + 1;
1557
1558 bstab = band_size_long[s->sample_rate_index];
1559 pretab = mpa_pretab[g->preflag];
1560 for(i=0;i<g->long_end;i++) {
1561 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1562 len = bstab[i];
1563 for(j=len;j>0;j--)
1564 *exp_ptr++ = v0;
1565 }
1566
1567 if (g->short_start < 13) {
1568 bstab = band_size_short[s->sample_rate_index];
1569 gains[0] = gain - (g->subblock_gain[0] << 3);
1570 gains[1] = gain - (g->subblock_gain[1] << 3);
1571 gains[2] = gain - (g->subblock_gain[2] << 3);
1572 k = g->long_end;
1573 for(i=g->short_start;i<13;i++) {
1574 len = bstab[i];
1575 for(l=0;l<3;l++) {
1576 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1577 for(j=len;j>0;j--)
1578 *exp_ptr++ = v0;
1579 }
1580 }
1581 }
1582 }
1583
1584 /* handle n = 0 too */
1585 static inline int get_bitsz(GetBitContext *s, int n)
1586 {
1587 if (n == 0)
1588 return 0;
1589 else
1590 return get_bits(s, n);
1591 }
1592
1593 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1594 int16_t *exponents, int end_pos2)
1595 {
1596 int s_index;
1597 int i;
1598 int last_pos, bits_left;
1599 VLC *vlc;
1600 int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
1601
1602 /* low frequencies (called big values) */
1603 s_index = 0;
1604 for(i=0;i<3;i++) {
1605 int j, k, l, linbits;
1606 j = g->region_size[i];
1607 if (j == 0)
1608 continue;
1609 /* select vlc table */
1610 k = g->table_select[i];
1611 l = mpa_huff_data[k][0];
1612 linbits = mpa_huff_data[k][1];
1613 vlc = &huff_vlc[l];
1614
1615 if(!l){
1616 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
1617 s_index += 2*j;
1618 continue;
1619 }
1620
1621 /* read huffcode and compute each couple */
1622 for(;j>0;j--) {
1623 int exponent, x, y, v;
1624 int pos= get_bits_count(&s->gb);
1625
1626 if (pos >= end_pos){
1627 // av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1628 if(s->in_gb.buffer && pos >= s->gb.size_in_bits){
1629 s->gb= s->in_gb;
1630 s->in_gb.buffer=NULL;
1631 assert((get_bits_count(&s->gb) & 7) == 0);
1632 skip_bits_long(&s->gb, pos - end_pos);
1633 end_pos2=
1634 end_pos= end_pos2 + get_bits_count(&s->gb) - pos;
1635 pos= get_bits_count(&s->gb);
1636 }
1637 // av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
1638 if(pos >= end_pos)
1639 break;
1640 }
1641 y = get_vlc2(&s->gb, vlc->table, 7, 3);
1642
1643 if(!y){
1644 g->sb_hybrid[s_index ] =
1645 g->sb_hybrid[s_index+1] = 0;
1646 s_index += 2;
1647 continue;
1648 }
1649
1650 exponent= exponents[s_index];
1651
1652 dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
1653 i, g->region_size[i] - j, x, y, exponent);
1654 if(y&16){
1655 x = y >> 5;
1656 y = y & 0x0f;
1657 if (x < 15){
1658 v = expval_table[ exponent ][ x ];
1659 // v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);
1660 }else{
1661 x += get_bitsz(&s->gb, linbits);
1662 v = l3_unscale(x, exponent);
1663 }
1664 if (get_bits1(&s->gb))
1665 v = -v;
1666 g->sb_hybrid[s_index] = v;
1667 if (y < 15){
1668 v = expval_table[ exponent ][ y ];
1669 }else{
1670 y += get_bitsz(&s->gb, linbits);
1671 v = l3_unscale(y, exponent);
1672 }
1673 if (get_bits1(&s->gb))
1674 v = -v;
1675 g->sb_hybrid[s_index+1] = v;
1676 }else{
1677 x = y >> 5;
1678 y = y & 0x0f;
1679 x += y;
1680 if (x < 15){
1681 v = expval_table[ exponent ][ x ];
1682 }else{
1683 x += get_bitsz(&s->gb, linbits);
1684 v = l3_unscale(x, exponent);
1685 }
1686 if (get_bits1(&s->gb))
1687 v = -v;
1688 g->sb_hybrid[s_index+!!y] = v;
1689 g->sb_hybrid[s_index+ !y] = 0;
1690 }
1691 s_index+=2;
1692 }
1693 }
1694
1695 /* high frequencies */
1696 vlc = &huff_quad_vlc[g->count1table_select];
1697 last_pos=0;
1698 while (s_index <= 572) {
1699 int pos, code;
1700 pos = get_bits_count(&s->gb);
1701 if (pos >= end_pos) {
1702 if (pos > end_pos2 && last_pos){
1703 /* some encoders generate an incorrect size for this
1704 part. We must go back into the data */
1705 s_index -= 4;
1706 skip_bits_long(&s->gb, last_pos - pos);
1707 av_log(NULL, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
1708 break;
1709 }
1710 // av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
1711 if(s->in_gb.buffer && pos >= s->gb.size_in_bits){
1712 s->gb= s->in_gb;
1713 s->in_gb.buffer=NULL;
1714 assert((get_bits_count(&s->gb) & 7) == 0);
1715 skip_bits_long(&s->gb, pos - end_pos);
1716 end_pos2=
1717 end_pos= end_pos2 + get_bits_count(&s->gb) - pos;
1718 pos= get_bits_count(&s->gb);
1719 }
1720 // av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
1721 if(pos >= end_pos)
1722 break;
1723 }
1724 last_pos= pos;
1725
1726 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1727 dprintf("t=%d code=%d\n", g->count1table_select, code);
1728 g->sb_hybrid[s_index+0]=
1729 g->sb_hybrid[s_index+1]=
1730 g->sb_hybrid[s_index+2]=
1731 g->sb_hybrid[s_index+3]= 0;
1732 while(code){
1733 const static int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1734 int v;
1735 int pos= s_index+idxtab[code];
1736 code ^= 8>>idxtab[code];
1737 v = exp_table[ exponents[pos] ];
1738 // v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);
1739 if(get_bits1(&s->gb))
1740 v = -v;
1741 g->sb_hybrid[pos] = v;
1742 }
1743 s_index+=4;
1744 }
1745 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1746
1747 /* skip extension bits */
1748 bits_left = end_pos - get_bits_count(&s->gb);
1749 //av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
1750 if (bits_left < 0) {
1751 dprintf("bits_left=%d\n", bits_left);
1752 return -1;
1753 }
1754 skip_bits_long(&s->gb, bits_left);
1755
1756 return 0;
1757 }
1758
1759 /* Reorder short blocks from bitstream order to interleaved order. It
1760 would be faster to do it in parsing, but the code would be far more
1761 complicated */
1762 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1763 {
1764 int i, j, len;
1765 int32_t *ptr, *dst, *ptr1;
1766 int32_t tmp[576];
1767
1768 if (g->block_type != 2)
1769 return;
1770
1771 if (g->switch_point) {
1772 if (s->sample_rate_index != 8) {
1773 ptr = g->sb_hybrid + 36;
1774 } else {
1775 ptr = g->sb_hybrid + 48;
1776 }
1777 } else {
1778 ptr = g->sb_hybrid;
1779 }
1780
1781 for(i=g->short_start;i<13;i++) {
1782 len = band_size_short[s->sample_rate_index][i];
1783 ptr1 = ptr;
1784 dst = tmp;
1785 for(j=len;j>0;j--) {
1786 *dst++ = ptr[0*len];
1787 *dst++ = ptr[1*len];
1788 *dst++ = ptr[2*len];
1789 ptr++;
1790 }
1791 ptr+=2*len;
1792 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1793 }
1794 }
1795
1796 #define ISQRT2 FIXR(0.70710678118654752440)
1797
1798 static void compute_stereo(MPADecodeContext *s,
1799 GranuleDef *g0, GranuleDef *g1)
1800 {
1801 int i, j, k, l;
1802 int32_t v1, v2;
1803 int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1804 int32_t (*is_tab)[16];
1805 int32_t *tab0, *tab1;
1806 int non_zero_found_short[3];
1807
1808 /* intensity stereo */
1809 if (s->mode_ext & MODE_EXT_I_STEREO) {
1810 if (!s->lsf) {
1811 is_tab = is_table;
1812 sf_max = 7;
1813 } else {
1814 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1815 sf_max = 16;
1816 }
1817
1818 tab0 = g0->sb_hybrid + 576;
1819 tab1 = g1->sb_hybrid + 576;
1820
1821 non_zero_found_short[0] = 0;
1822 non_zero_found_short[1] = 0;
1823 non_zero_found_short[2] = 0;
1824 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1825 for(i = 12;i >= g1->short_start;i--) {
1826 /* for last band, use previous scale factor */
1827 if (i != 11)
1828 k -= 3;
1829 len = band_size_short[s->sample_rate_index][i];
1830 for(l=2;l>=0;l--) {
1831 tab0 -= len;
1832 tab1 -= len;
1833 if (!non_zero_found_short[l]) {
1834 /* test if non zero band. if so, stop doing i-stereo */
1835 for(j=0;j<len;j++) {
1836 if (tab1[j] != 0) {
1837 non_zero_found_short[l] = 1;
1838 goto found1;
1839 }
1840 }
1841 sf = g1->scale_factors[k + l];
1842 if (sf >= sf_max)
1843 goto found1;
1844
1845 v1 = is_tab[0][sf];
1846 v2 = is_tab[1][sf];
1847 for(j=0;j<len;j++) {
1848 tmp0 = tab0[j];
1849 tab0[j] = MULL(tmp0, v1);
1850 tab1[j] = MULL(tmp0, v2);
1851 }
1852 } else {
1853 found1:
1854 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1855 /* lower part of the spectrum : do ms stereo
1856 if enabled */
1857 for(j=0;j<len;j++) {
1858 tmp0 = tab0[j];
1859 tmp1 = tab1[j];
1860 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1861 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1862 }
1863 }
1864 }
1865 }
1866 }
1867
1868 non_zero_found = non_zero_found_short[0] |
1869 non_zero_found_short[1] |
1870 non_zero_found_short[2];
1871
1872 for(i = g1->long_end - 1;i >= 0;i--) {
1873 len = band_size_long[s->sample_rate_index][i];
1874 tab0 -= len;
1875 tab1 -= len;
1876 /* test if non zero band. if so, stop doing i-stereo */
1877 if (!non_zero_found) {
1878 for(j=0;j<len;j++) {
1879 if (tab1[j] != 0) {
1880 non_zero_found = 1;
1881 goto found2;
1882 }
1883 }
1884 /* for last band, use previous scale factor */
1885 k = (i == 21) ? 20 : i;
1886 sf = g1->scale_factors[k];
1887 if (sf >= sf_max)
1888 goto found2;
1889 v1 = is_tab[0][sf];
1890 v2 = is_tab[1][sf];
1891 for(j=0;j<len;j++) {
1892 tmp0 = tab0[j];
1893 tab0[j] = MULL(tmp0, v1);
1894 tab1[j] = MULL(tmp0, v2);
1895 }
1896 } else {
1897 found2:
1898 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1899 /* lower part of the spectrum : do ms stereo
1900 if enabled */
1901 for(j=0;j<len;j++) {
1902 tmp0 = tab0[j];
1903 tmp1 = tab1[j];
1904 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1905 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1906 }
1907 }
1908 }
1909 }
1910 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1911 /* ms stereo ONLY */
1912 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1913 global gain */
1914 tab0 = g0->sb_hybrid;
1915 tab1 = g1->sb_hybrid;
1916 for(i=0;i<576;i++) {
1917 tmp0 = tab0[i];
1918 tmp1 = tab1[i];
1919 tab0[i] = tmp0 + tmp1;
1920 tab1[i] = tmp0 - tmp1;
1921 }
1922 }
1923 }
1924
1925 static void compute_antialias_integer(MPADecodeContext *s,
1926 GranuleDef *g)
1927 {
1928 int32_t *ptr, *csa;
1929 int n, i;
1930
1931 /* we antialias only "long" bands */
1932 if (g->block_type == 2) {
1933 if (!g->switch_point)
1934 return;
1935 /* XXX: check this for 8000Hz case */
1936 n = 1;
1937 } else {
1938 n = SBLIMIT - 1;
1939 }
1940
1941 ptr = g->sb_hybrid + 18;
1942 for(i = n;i > 0;i--) {
1943 int tmp0, tmp1, tmp2;
1944 csa = &csa_table[0][0];
1945 #define INT_AA(j) \
1946 tmp0 = ptr[-1-j];\
1947 tmp1 = ptr[ j];\
1948 tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1949 ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1950 ptr[ j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1951
1952 INT_AA(0)
1953 INT_AA(1)
1954 INT_AA(2)
1955 INT_AA(3)
1956 INT_AA(4)
1957 INT_AA(5)
1958 INT_AA(6)
1959 INT_AA(7)
1960
1961 ptr += 18;
1962 }
1963 }
1964
1965 static void compute_antialias_float(MPADecodeContext *s,
1966 GranuleDef *g)
1967 {
1968 int32_t *ptr;
1969 int n, i;
1970
1971 /* we antialias only "long" bands */
1972 if (g->block_type == 2) {
1973 if (!g->switch_point)
1974 return;
1975 /* XXX: check this for 8000Hz case */
1976 n = 1;
1977 } else {
1978 n = SBLIMIT - 1;
1979 }
1980
1981 ptr = g->sb_hybrid + 18;
1982 for(i = n;i > 0;i--) {
1983 float tmp0, tmp1;
1984 float *csa = &csa_table_float[0][0];
1985 #define FLOAT_AA(j)\
1986 tmp0= ptr[-1-j];\
1987 tmp1= ptr[ j];\
1988 ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
1989 ptr[ j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
1990
1991 FLOAT_AA(0)
1992 FLOAT_AA(1)
1993 FLOAT_AA(2)
1994 FLOAT_AA(3)
1995 FLOAT_AA(4)
1996 FLOAT_AA(5)
1997 FLOAT_AA(6)
1998 FLOAT_AA(7)
1999
2000 ptr += 18;
2001 }
2002 }
2003
2004 static void compute_imdct(MPADecodeContext *s,
2005 GranuleDef *g,
2006 int32_t *sb_samples,
2007 int32_t *mdct_buf)
2008 {
2009 int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
2010 int32_t out2[12];
2011 int i, j, mdct_long_end, v, sblimit;
2012
2013 /* find last non zero block */
2014 ptr = g->sb_hybrid + 576;
2015 ptr1 = g->sb_hybrid + 2 * 18;
2016 while (ptr >= ptr1) {
2017 ptr -= 6;
2018 v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
2019 if (v != 0)
2020 break;
2021 }
2022 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
2023
2024 if (g->block_type == 2) {
2025 /* XXX: check for 8000 Hz */
2026 if (g->switch_point)
2027 mdct_long_end = 2;
2028 else
2029 mdct_long_end = 0;
2030 } else {
2031 mdct_long_end = sblimit;
2032 }
2033
2034 buf = mdct_buf;
2035 ptr = g->sb_hybrid;
2036 for(j=0;j<mdct_long_end;j++) {
2037 /* apply window & overlap with previous buffer */
2038 out_ptr = sb_samples + j;
2039 /* select window */
2040 if (g->switch_point && j < 2)
2041 win1 = mdct_win[0];
2042 else
2043 win1 = mdct_win[g->block_type];
2044 /* select frequency inversion */
2045 win = win1 + ((4 * 36) & -(j & 1));
2046 imdct36(out_ptr, buf, ptr, win);
2047 out_ptr += 18*SBLIMIT;
2048 ptr += 18;
2049 buf += 18;
2050 }
2051 for(j=mdct_long_end;j<sblimit;j++) {
2052 /* select frequency inversion */
2053 win = mdct_win[2] + ((4 * 36) & -(j & 1));
2054 out_ptr = sb_samples + j;
2055
2056 for(i=0; i<6; i++){
2057 *out_ptr = buf[i];
2058 out_ptr += SBLIMIT;
2059 }
2060 imdct12(out2, ptr + 0);
2061 for(i=0;i<6;i++) {
2062 *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
2063 buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
2064 out_ptr += SBLIMIT;
2065 }
2066 imdct12(out2, ptr + 1);
2067 for(i=0;i<6;i++) {
2068 *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
2069 buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
2070 out_ptr += SBLIMIT;
2071 }
2072 imdct12(out2, ptr + 2);
2073 for(i=0;i<6;i++) {
2074 buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
2075 buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
2076 buf[i + 6*2] = 0;
2077 }
2078 ptr += 18;
2079 buf += 18;
2080 }
2081 /* zero bands */
2082 for(j=sblimit;j<SBLIMIT;j++) {
2083 /* overlap */
2084 out_ptr = sb_samples + j;
2085 for(i=0;i<18;i++) {
2086 *out_ptr = buf[i];
2087 buf[i] = 0;
2088 out_ptr += SBLIMIT;
2089 }
2090 buf += 18;
2091 }
2092 }
2093
2094 #if defined(DEBUG)
2095 void sample_dump(int fnum, int32_t *tab, int n)
2096 {
2097 static FILE *files[16], *f;
2098 char buf[512];
2099 int i;
2100 int32_t v;
2101
2102 f = files[fnum];
2103 if (!f) {
2104 snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
2105 fnum,
2106 #ifdef USE_HIGHPRECISION
2107 "hp"
2108 #else
2109 "lp"
2110 #endif
2111 );
2112 f = fopen(buf, "w");
2113 if (!f)
2114 return;
2115 files[fnum] = f;
2116 }
2117
2118 if (fnum == 0) {
2119 static int pos = 0;
2120 av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
2121 for(i=0;i<n;i++) {
2122 av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
2123 if ((i % 18) == 17)
2124 av_log(NULL, AV_LOG_DEBUG, "\n");
2125 }
2126 pos += n;
2127 }
2128 for(i=0;i<n;i++) {
2129 /* normalize to 23 frac bits */
2130 v = tab[i] << (23 - FRAC_BITS);
2131 fwrite(&v, 1, sizeof(int32_t), f);
2132 }
2133 }
2134 #endif
2135
2136
2137 /* main layer3 decoding function */
2138 static int mp_decode_layer3(MPADecodeContext *s)
2139 {
2140 int nb_granules, main_data_begin, private_bits;
2141 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
2142 GranuleDef granules[2][2], *g;
2143 int16_t exponents[576];
2144
2145 /* read side info */
2146 if (s->lsf) {
2147 main_data_begin = get_bits(&s->gb, 8);
2148 private_bits = get_bits(&s->gb, s->nb_channels);
2149 nb_granules = 1;
2150 } else {
2151 main_data_begin = get_bits(&s->gb, 9);
2152 if (s->nb_channels == 2)
2153 private_bits = get_bits(&s->gb, 3);
2154 else
2155 private_bits = get_bits(&s->gb, 5);
2156 nb_granules = 2;
2157 for(ch=0;ch<s->nb_channels;ch++) {
2158 granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2159 granules[ch][1].scfsi = get_bits(&s->gb, 4);
2160 }
2161 }
2162
2163 for(gr=0;gr<nb_granules;gr++) {
2164 for(ch=0;ch<s->nb_channels;ch++) {
2165 dprintf("gr=%d ch=%d: side_info\n", gr, ch);
2166 g = &granules[ch][gr];
2167 g->part2_3_length = get_bits(&s->gb, 12);
2168 g->big_values = get_bits(&s->gb, 9);
2169 g->global_gain = get_bits(&s->gb, 8);
2170 /* if MS stereo only is selected, we precompute the
2171 1/sqrt(2) renormalization factor */
2172 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2173 MODE_EXT_MS_STEREO)
2174 g->global_gain -= 2;
2175 if (s->lsf)
2176 g->scalefac_compress = get_bits(&s->gb, 9);
2177 else
2178 g->scalefac_compress = get_bits(&s->gb, 4);
2179 blocksplit_flag = get_bits(&s->gb, 1);
2180 if (blocksplit_flag) {
2181 g->block_type = get_bits(&s->gb, 2);
2182 if (g->block_type == 0)
2183 return -1;
2184 g->switch_point = get_bits(&s->gb, 1);
2185 for(i=0;i<2;i++)
2186 g->table_select[i] = get_bits(&s->gb, 5);
2187 for(i=0;i<3;i++)
2188 g->subblock_gain[i] = get_bits(&s->gb, 3);
2189 /* compute huffman coded region sizes */
2190 if (g->block_type == 2)
2191 g->region_size[0] = (36 / 2);
2192 else {
2193 if (s->sample_rate_index <= 2)
2194 g->region_size[0] = (36 / 2);
2195 else if (s->sample_rate_index != 8)
2196 g->region_size[0] = (54 / 2);
2197 else
2198 g->region_size[0] = (108 / 2);
2199 }
2200 g->region_size[1] = (576 / 2);
2201 } else {
2202 int region_address1, region_address2, l;
2203 g->block_type = 0;
2204 g->switch_point = 0;
2205 for(i=0;i<3;i++)
2206 g->table_select[i] = get_bits(&s->gb, 5);
2207 /* compute huffman coded region sizes */
2208 region_address1 = get_bits(&s->gb, 4);
2209 region_address2 = get_bits(&s->gb, 3);
2210 dprintf("region1=%d region2=%d\n",
2211 region_address1, region_address2);
2212 g->region_size[0] =
2213 band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2214 l = region_address1 + region_address2 + 2;
2215 /* should not overflow */
2216 if (l > 22)
2217 l = 22;
2218 g->region_size[1] =
2219 band_index_long[s->sample_rate_index][l] >> 1;
2220 }
2221 /* convert region offsets to region sizes and truncate
2222 size to big_values */
2223 g->region_size[2] = (576 / 2);
2224 j = 0;
2225 for(i=0;i<3;i++) {
2226 k = FFMIN(g->region_size[i], g->big_values);
2227 g->region_size[i] = k - j;
2228 j = k;
2229 }
2230
2231 /* compute band indexes */
2232 if (g->block_type == 2) {
2233 if (g->switch_point) {
2234 /* if switched mode, we handle the 36 first samples as
2235 long blocks. For 8000Hz, we handle the 48 first
2236 exponents as long blocks (XXX: check this!) */
2237 if (s->sample_rate_index <= 2)
2238 g->long_end = 8;
2239 else if (s->sample_rate_index != 8)
2240 g->long_end = 6;
2241 else
2242 g->long_end = 4; /* 8000 Hz */
2243
2244 g->short_start = 2 + (s->sample_rate_index != 8);
2245 } else {
2246 g->long_end = 0;
2247 g->short_start = 0;
2248 }
2249 } else {
2250 g->short_start = 13;
2251 g->long_end = 22;
2252 }
2253
2254 g->preflag = 0;
2255 if (!s->lsf)
2256 g->preflag = get_bits(&s->gb, 1);
2257 g->scalefac_scale = get_bits(&s->gb, 1);
2258 g->count1table_select = get_bits(&s->gb, 1);
2259 dprintf("block_type=%d switch_point=%d\n",
2260 g->block_type, g->switch_point);
2261 }
2262 }
2263
2264 if (!s->adu_mode) {
2265 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
2266 assert((get_bits_count(&s->gb) & 7) == 0);
2267 /* now we get bits from the main_data_begin offset */
2268 dprintf("seekback: %d\n", main_data_begin);
2269 //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
2270 if(main_data_begin > s->last_buf_size){
2271 av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
2272 s->last_buf_size= main_data_begin;
2273 }
2274
2275 memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
2276 s->in_gb= s->gb;
2277 init_get_bits(&s->gb, s->last_buf + s->last_buf_size - main_data_begin, main_data_begin*8);
2278 }
2279
2280 for(gr=0;gr<nb_granules;gr++) {
2281 for(ch=0;ch<s->nb_channels;ch++) {
2282 g = &granules[ch][gr];
2283
2284 bits_pos = get_bits_count(&s->gb);
2285
2286 if (!s->lsf) {
2287 uint8_t *sc;
2288 int slen, slen1, slen2;
2289
2290 /* MPEG1 scale factors */
2291 slen1 = slen_table[0][g->scalefac_compress];
2292 slen2 = slen_table[1][g->scalefac_compress];
2293 dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2294 if (g->block_type == 2) {
2295 n = g->switch_point ? 17 : 18;
2296 j = 0;
2297 if(slen1){
2298 for(i=0;i<n;i++)
2299 g->scale_factors[j++] = get_bits(&s->gb, slen1);
2300 }else{
2301 for(i=0;i<n;i++)
2302 g->scale_factors[j++] = 0;
2303 }
2304 if(slen2){
2305 for(i=0;i<18;i++)
2306 g->scale_factors[j++] = get_bits(&s->gb, slen2);
2307 for(i=0;i<3;i++)
2308 g->scale_factors[j++] = 0;
2309 }else{
2310 for(i=0;i<21;i++)
2311 g->scale_factors[j++] = 0;
2312 }
2313 } else {
2314 sc = granules[ch][0].scale_factors;
2315 j = 0;
2316 for(k=0;k<4;k++) {
2317 n = (k == 0 ? 6 : 5);
2318 if ((g->scfsi & (0x8 >> k)) == 0) {
2319 slen = (k < 2) ? slen1 : slen2;
2320 if(slen){
2321 for(i=0;i<n;i++)
2322 g->scale_factors[j++] = get_bits(&s->gb, slen);
2323 }else{
2324 for(i=0;i<n;i++)
2325 g->scale_factors[j++] = 0;
2326 }
2327 } else {
2328 /* simply copy from last granule */
2329 for(i=0;i<n;i++) {
2330 g->scale_factors[j] = sc[j];
2331 j++;
2332 }
2333 }
2334 }
2335 g->scale_factors[j++] = 0;
2336 }
2337 #if defined(DEBUG)
2338 {
2339 dprintf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2340 g->scfsi, gr, ch);
2341 for(i=0;i<j;i++)
2342 dprintf(" %d", g->scale_factors[i]);
2343 dprintf("\n");
2344 }
2345 #endif
2346 } else {
2347 int tindex, tindex2, slen[4], sl, sf;
2348
2349 /* LSF scale factors */
2350 if (g->block_type == 2) {
2351 tindex = g->switch_point ? 2 : 1;
2352 } else {
2353 tindex = 0;
2354 }
2355 sf = g->scalefac_compress;
2356 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2357 /* intensity stereo case */
2358 sf >>= 1;
2359 if (sf < 180) {
2360 lsf_sf_expand(slen, sf, 6, 6, 0);
2361 tindex2 = 3;
2362 } else if (sf < 244) {
2363 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2364 tindex2 = 4;
2365 } else {
2366 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2367 tindex2 = 5;
2368 }
2369 } else {
2370 /* normal case */
2371 if (sf < 400) {
2372 lsf_sf_expand(slen, sf, 5, 4, 4);
2373 tindex2 = 0;
2374 } else if (sf < 500) {
2375 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2376 tindex2 = 1;
2377 } else {
2378 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2379 tindex2 = 2;
2380 g->preflag = 1;
2381 }
2382 }
2383
2384 j = 0;
2385 for(k=0;k<4;k++) {
2386 n = lsf_nsf_table[tindex2][tindex][k];
2387 sl = slen[k];
2388 if(sl){
2389 for(i=0;i<n;i++)
2390 g->scale_factors[j++] = get_bits(&s->gb, sl);
2391 }else{
2392 for(i=0;i<n;i++)
2393 g->scale_factors[j++] = 0;
2394 }
2395 }
2396 /* XXX: should compute exact size */
2397 for(;j<40;j++)
2398 g->scale_factors[j] = 0;
2399 #if defined(DEBUG)
2400 {
2401 dprintf("gr=%d ch=%d scale_factors:\n",
2402 gr, ch);
2403 for(i=0;i<40;i++)
2404 dprintf(" %d", g->scale_factors[i]);
2405 dprintf("\n");
2406 }
2407 #endif
2408 }
2409
2410 exponents_from_scale_factors(s, g, exponents);
2411
2412 /* read Huffman coded residue */
2413 if (huffman_decode(s, g, exponents,
2414 bits_pos + g->part2_3_length) < 0)
2415 return -1;
2416 #if defined(DEBUG)
2417 sample_dump(0, g->sb_hybrid, 576);
2418 #endif
2419 } /* ch */
2420
2421 if (s->nb_channels == 2)
2422 compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2423
2424 for(ch=0;ch<s->nb_channels;ch++) {
2425 g = &granules[ch][gr];
2426
2427 reorder_block(s, g);
2428 #if defined(DEBUG)
2429 sample_dump(0, g->sb_hybrid, 576);
2430 #endif
2431 s->compute_antialias(s, g);
2432 #if defined(DEBUG)
2433 sample_dump(1, g->sb_hybrid, 576);
2434 #endif
2435 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2436 #if defined(DEBUG)
2437 sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2438 #endif
2439 }
2440 } /* gr */
2441 return nb_granules * 18;
2442 }
2443
2444 static int mp_decode_frame(MPADecodeContext *s,
2445 OUT_INT *samples, const uint8_t *buf, int buf_size)
2446 {
2447 int i, nb_frames, ch;
2448 OUT_INT *samples_ptr;
2449
2450 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
2451
2452 /* skip error protection field */
2453 if (s->error_protection)
2454 get_bits(&s->gb, 16);
2455
2456 dprintf("frame %d:\n", s->frame_count);
2457 switch(s->layer) {
2458 case 1:
2459 nb_frames = mp_decode_layer1(s);
2460 break;
2461 case 2:
2462 nb_frames = mp_decode_layer2(s);
2463 break;
2464 case 3:
2465 default:
2466 nb_frames = mp_decode_layer3(s);
2467
2468 s->last_buf_size=0;
2469 if(s->in_gb.buffer){
2470 align_get_bits(&s->gb);
2471 i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2472 if(i >= 0 && i <= BACKSTEP_SIZE){
2473 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
2474 s->last_buf_size=i;
2475 }else
2476 av_log(NULL, AV_LOG_ERROR, "invalid old backstep %d\n", i);
2477 s->gb= s->in_gb;
2478 }
2479
2480 align_get_bits(&s->gb);
2481 assert((get_bits_count(&s->gb) & 7) == 0);
2482 i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;
2483
2484 if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
2485 av_log(NULL, AV_LOG_ERROR, "invalid new backstep %d\n", i);
2486 i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
2487 }
2488 assert(i <= buf_size - HEADER_SIZE && i>= 0);
2489 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
2490 s->last_buf_size += i;
2491
2492 break;
2493 }
2494 #if defined(DEBUG)
2495 for(i=0;i<nb_frames;i++) {
2496 for(ch=0;ch<s->nb_channels;ch++) {
2497 int j;
2498 dprintf("%d-%d:", i, ch);
2499 for(j=0;j<SBLIMIT;j++)
2500 dprintf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2501 dprintf("\n");
2502 }
2503 }
2504 #endif
2505 /* apply the synthesis filter */
2506 for(ch=0;ch<s->nb_channels;ch++) {
2507 samples_ptr = samples + ch;
2508 for(i=0;i<nb_frames;i++) {
2509 ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2510 window, &s->dither_state,
2511 samples_ptr, s->nb_channels,
2512 s->sb_samples[ch][i]);
2513 samples_ptr += 32 * s->nb_channels;
2514 }
2515 }
2516 #ifdef DEBUG
2517 s->frame_count++;
2518 #endif
2519 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2520 }
2521
2522 static int decode_frame(AVCodecContext * avctx,
2523 void *data, int *data_size,
2524 uint8_t * buf, int buf_size)
2525 {
2526 MPADecodeContext *s = avctx->priv_data;
2527 uint32_t header;
2528 int out_size;
2529 OUT_INT *out_samples = data;
2530
2531 retry:
2532 if(buf_size < HEADER_SIZE)
2533 return -1;
2534
2535 header = (buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3];
2536 if(ff_mpa_check_header(header) < 0){
2537 buf++;
2538 // buf_size--;
2539 av_log(avctx, AV_LOG_ERROR, "Header missing skipping one byte.\n");
2540 goto retry;
2541 }
2542
2543 if (decode_header(s, header) == 1) {
2544 /* free format: prepare to compute frame size */
2545 s->frame_size = -1;
2546 return -1;
2547 }
2548 /* update codec info */
2549 avctx->sample_rate = s->sample_rate;
2550 avctx->channels = s->nb_channels;
2551 avctx->bit_rate = s->bit_rate;
2552 avctx->sub_id = s->layer;
2553 switch(s->layer) {
2554 case 1:
2555 avctx->frame_size = 384;
2556 break;
2557 case 2:
2558 avctx->frame_size = 1152;
2559 break;
2560 case 3:
2561 if (s->lsf)
2562 avctx->frame_size = 576;
2563 else
2564 avctx->frame_size = 1152;
2565 break;
2566 }
2567
2568 if(s->frame_size<=0 || s->frame_size > buf_size){
2569 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
2570 return -1;
2571 }else if(s->frame_size < buf_size){
2572 av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
2573 }
2574
2575 out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2576 if(out_size>=0)
2577 *data_size = out_size;
2578 else
2579 av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
2580 s->frame_size = 0;
2581 return buf_size;
2582 }
2583
2584 #ifdef CONFIG_MP3ADU_DECODER
2585 static int decode_frame_adu(AVCodecContext * avctx,
2586 void *data, int *data_size,
2587 uint8_t * buf, int buf_size)
2588 {
2589 MPADecodeContext *s = avctx->priv_data;
2590 uint32_t header;
2591 int len, out_size;
2592 OUT_INT *out_samples = data;
2593
2594 len = buf_size;
2595
2596 // Discard too short frames
2597 if (buf_size < HEADER_SIZE) {
2598 *data_size = 0;
2599 return buf_size;
2600 }
2601
2602
2603 if (len > MPA_MAX_CODED_FRAME_SIZE)
2604 len = MPA_MAX_CODED_FRAME_SIZE;
2605
2606 // Get header and restore sync word
2607 header = (buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3] | 0xffe00000;
2608
2609 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2610 *data_size = 0;
2611 return buf_size;
2612 }
2613
2614 decode_header(s, header);
2615 /* update codec info */
2616 avctx->sample_rate = s->sample_rate;
2617 avctx->channels = s->nb_channels;
2618 avctx->bit_rate = s->bit_rate;
2619 avctx->sub_id = s->layer;
2620
2621 avctx->frame_size=s->frame_size = len;
2622
2623 if (avctx->parse_only) {
2624 out_size = buf_size;
2625 } else {
2626 out_size = mp_decode_frame(s, out_samples, buf, buf_size);
2627 }
2628
2629 *data_size = out_size;
2630 return buf_size;
2631 }
2632 #endif /* CONFIG_MP3ADU_DECODER */
2633
2634 #ifdef CONFIG_MP3ON4_DECODER
2635 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2636 static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2}; /* number of mp3 decoder instances */
2637 static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2638 /* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2639 static int chan_offset[9][5] = {
2640 {0},
2641 {0}, // C
2642 {0}, // FLR
2643 {2,0}, // C FLR
2644 {2,0,3}, // C FLR BS
2645 {4,0,2}, // C FLR BLRS
2646 {4,0,2,5}, // C FLR BLRS LFE
2647 {4,0,2,6,5}, // C FLR BLRS BLR LFE
2648 {0,2} // FLR BLRS
2649 };
2650
2651
2652 static int decode_init_mp3on4(AVCodecContext * avctx)
2653 {
2654 MP3On4DecodeContext *s = avctx->priv_data;
2655 int i;
2656
2657 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2658 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2659 return -1;
2660 }
2661
2662 s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2663 s->frames = mp3Frames[s->chan_cfg];
2664 if(!s->frames) {
2665 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2666 return -1;
2667 }
2668 avctx->channels = mp3Channels[s->chan_cfg];
2669
2670 /* Init the first mp3 decoder in standard way, so that all tables get builded
2671 * We replace avctx->priv_data with the context of the first decoder so that
2672 * decode_init() does not have to be changed.
2673 * Other decoders will be inited here copying data from the first context
2674 */
2675 // Allocate zeroed memory for the first decoder context
2676 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2677 // Put decoder context in place to make init_decode() happy
2678 avctx->priv_data = s->mp3decctx[0];
2679 decode_init(avctx);
2680 // Restore mp3on4 context pointer
2681 avctx->priv_data = s;
2682 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2683
2684 /* Create a separate codec/context for each frame (first is already ok).
2685 * Each frame is 1 or 2 channels - up to 5 frames allowed
2686 */
2687 for (i = 1; i < s->frames; i++) {
2688 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2689 s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2690 s->mp3decctx[i]->adu_mode = 1;
2691 }
2692
2693 return 0;
2694 }
2695
2696
2697 static int decode_close_mp3on4(AVCodecContext * avctx)
2698 {
2699 MP3On4DecodeContext *s = avctx->priv_data;
2700 int i;
2701
2702 for (i = 0; i < s->frames; i++)
2703 if (s->mp3decctx[i])
2704 av_free(s->mp3decctx[i]);
2705
2706 return 0;
2707 }
2708
2709
2710 static int decode_frame_mp3on4(AVCodecContext * avctx,
2711 void *data, int *data_size,
2712 uint8_t * buf, int buf_size)
2713 {
2714 MP3On4DecodeContext *s = avctx->priv_data;
2715 MPADecodeContext *m;
2716 int len, out_size = 0;
2717 uint32_t header;
2718 OUT_INT *out_samples = data;
2719 OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2720 OUT_INT *outptr, *bp;
2721 int fsize;
2722 unsigned char *start2 = buf, *start;
2723 int fr, i, j, n;
2724 int off = avctx->channels;
2725 int *coff = chan_offset[s->chan_cfg];
2726
2727 len = buf_size;
2728
2729 // Discard too short frames
2730 if (buf_size < HEADER_SIZE) {
2731 *data_size = 0;
2732 return buf_size;
2733 }
2734
2735 // If only one decoder interleave is not needed
2736 outptr = s->frames == 1 ? out_samples : decoded_buf;
2737
2738 for (fr = 0; fr < s->frames; fr++) {
2739 start = start2;
2740 fsize = (start[0] << 4) | (start[1] >> 4);
2741 start2 += fsize;
2742 if (fsize > len)
2743 fsize = len;
2744 len -= fsize;
2745 if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2746 fsize = MPA_MAX_CODED_FRAME_SIZE;
2747 m = s->mp3decctx[fr];
2748 assert (m != NULL);
2749
2750 // Get header
2751 header = (start[0] << 24) | (start[1] << 16) | (start[2] << 8) | start[3] | 0xfff00000;
2752
2753 if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2754 *data_size = 0;
2755 return buf_size;
2756 }
2757
2758 decode_header(m, header);
2759 mp_decode_frame(m, decoded_buf, start, fsize);
2760
2761 n = MPA_FRAME_SIZE * m->nb_channels;
2762 out_size += n * sizeof(OUT_INT);
2763 if(s->frames > 1) {
2764 /* interleave output data */
2765 bp = out_samples + coff[fr];
2766 if(m->nb_channels == 1) {
2767 for(j = 0; j < n; j++) {
2768 *bp = decoded_buf[j];
2769 bp += off;
2770 }
2771 } else {
2772 for(j = 0; j < n; j++) {
2773 bp[0] = decoded_buf[j++];
2774 bp[1] = decoded_buf[j];
2775 bp += off;
2776 }
2777 }
2778 }
2779 }
2780
2781 /* update codec info */
2782 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2783 avctx->frame_size= buf_size;
2784 avctx->bit_rate = 0;
2785 for (i = 0; i < s->frames; i++)
2786 avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2787
2788 *data_size = out_size;
2789 return buf_size;
2790 }
2791 #endif /* CONFIG_MP3ON4_DECODER */
2792
2793 #ifdef CONFIG_MP2_DECODER
2794 AVCodec mp2_decoder =
2795 {
2796 "mp2",
2797 CODEC_TYPE_AUDIO,
2798 CODEC_ID_MP2,
2799 sizeof(MPADecodeContext),
2800 decode_init,
2801 NULL,
2802 NULL,
2803 decode_frame,
2804 CODEC_CAP_PARSE_ONLY,
2805 };
2806 #endif
2807 #ifdef CONFIG_MP3_DECODER
2808 AVCodec mp3_decoder =
2809 {
2810 "mp3",
2811 CODEC_TYPE_AUDIO,
2812 CODEC_ID_MP3,
2813 sizeof(MPADecodeContext),
2814 decode_init,
2815 NULL,
2816 NULL,
2817 decode_frame,
2818 CODEC_CAP_PARSE_ONLY,
2819 };
2820 #endif
2821 #ifdef CONFIG_MP3ADU_DECODER
2822 AVCodec mp3adu_decoder =
2823 {
2824 "mp3adu",
2825 CODEC_TYPE_AUDIO,
2826 CODEC_ID_MP3ADU,
2827 sizeof(MPADecodeContext),
2828 decode_init,
2829 NULL,
2830 NULL,
2831 decode_frame_adu,
2832 CODEC_CAP_PARSE_ONLY,
2833 };
2834 #endif
2835 #ifdef CONFIG_MP3ON4_DECODER
2836 AVCodec mp3on4_decoder =
2837 {
2838 "mp3on4",
2839 CODEC_TYPE_AUDIO,
2840 CODEC_ID_MP3ON4,
2841 sizeof(MP3On4DecodeContext),
2842 decode_init_mp3on4,
2843 NULL,
2844 decode_close_mp3on4,
2845 decode_frame_mp3on4,
2846 0
2847 };
2848 #endif