562978cdebcd0c0c06769419fac29f1dd3b38957
[libav.git] / libavcodec / mpegaudiodec.c
1 /*
2 * MPEG Audio decoder
3 * Copyright (c) 2001, 2002 Fabrice Bellard.
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19
20 /**
21 * @file mpegaudiodec.c
22 * MPEG Audio decoder.
23 */
24
25 //#define DEBUG
26 #include "avcodec.h"
27 #include "bitstream.h"
28 #include "dsputil.h"
29
30 /*
31 * TODO:
32 * - in low precision mode, use more 16 bit multiplies in synth filter
33 * - test lsf / mpeg25 extensively.
34 */
35
36 /* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
37 audio decoder */
38 #ifdef CONFIG_MPEGAUDIO_HP
39 # define USE_HIGHPRECISION
40 #endif
41
42 #include "mpegaudio.h"
43
44 #define FRAC_ONE (1 << FRAC_BITS)
45
46 #ifdef ARCH_X86
47 # define MULL(ra, rb) \
48 ({ int rt, dummy; asm (\
49 "imull %3 \n\t"\
50 "shrdl %4, %%edx, %%eax \n\t"\
51 : "=a"(rt), "=d"(dummy)\
52 : "a" (ra), "rm" (rb), "i"(FRAC_BITS));\
53 rt; })
54 # define MUL64(ra, rb) \
55 ({ int64_t rt; asm ("imull %2\n\t" : "=A"(rt) : "a" (ra), "g" (rb)); rt; })
56 # define MULH(ra, rb) \
57 ({ int rt, dummy; asm ("imull %3\n\t" : "=d"(rt), "=a"(dummy): "a" (ra), "rm" (rb)); rt; })
58 #elif defined(ARCH_ARMV4L)
59 # define MULL(a, b) \
60 ({ int lo, hi;\
61 asm("smull %0, %1, %2, %3 \n\t"\
62 "mov %0, %0, lsr %4\n\t"\
63 "add %1, %0, %1, lsl %5\n\t"\
64 : "=&r"(lo), "=&r"(hi)\
65 : "r"(b), "r"(a), "i"(FRAC_BITS), "i"(32-FRAC_BITS));\
66 hi; })
67 # define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))
68 # define MULH(a, b) ({ int lo, hi; asm ("smull %0, %1, %2, %3" : "=&r"(lo), "=&r"(hi) : "r"(b), "r"(a)); hi; })
69 #else
70 # define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
71 # define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))
72 //#define MULH(a,b) (((int64_t)(a) * (int64_t)(b))>>32) //gcc 3.4 creates an incredibly bloated mess out of this
73 static always_inline int MULH(int a, int b){
74 return ((int64_t)(a) * (int64_t)(b))>>32;
75 }
76 #endif
77 #define FIX(a) ((int)((a) * FRAC_ONE))
78 /* WARNING: only correct for posititive numbers */
79 #define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
80 #define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
81
82 #define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
83
84 /****************/
85
86 #define HEADER_SIZE 4
87 #define BACKSTEP_SIZE 512
88
89 struct GranuleDef;
90
91 typedef struct MPADecodeContext {
92 uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE]; /* input buffer */
93 int inbuf_index;
94 uint8_t *inbuf_ptr, *inbuf;
95 int frame_size;
96 int free_format_frame_size; /* frame size in case of free format
97 (zero if currently unknown) */
98 /* next header (used in free format parsing) */
99 uint32_t free_format_next_header;
100 int error_protection;
101 int layer;
102 int sample_rate;
103 int sample_rate_index; /* between 0 and 8 */
104 int bit_rate;
105 int old_frame_size;
106 GetBitContext gb;
107 int nb_channels;
108 int mode;
109 int mode_ext;
110 int lsf;
111 MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));
112 int synth_buf_offset[MPA_MAX_CHANNELS];
113 int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));
114 int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
115 #ifdef DEBUG
116 int frame_count;
117 #endif
118 void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);
119 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
120 unsigned int dither_state;
121 } MPADecodeContext;
122
123 /**
124 * Context for MP3On4 decoder
125 */
126 typedef struct MP3On4DecodeContext {
127 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
128 int chan_cfg; ///< channel config number
129 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
130 } MP3On4DecodeContext;
131
132 /* layer 3 "granule" */
133 typedef struct GranuleDef {
134 uint8_t scfsi;
135 int part2_3_length;
136 int big_values;
137 int global_gain;
138 int scalefac_compress;
139 uint8_t block_type;
140 uint8_t switch_point;
141 int table_select[3];
142 int subblock_gain[3];
143 uint8_t scalefac_scale;
144 uint8_t count1table_select;
145 int region_size[3]; /* number of huffman codes in each region */
146 int preflag;
147 int short_start, long_end; /* long/short band indexes */
148 uint8_t scale_factors[40];
149 int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
150 } GranuleDef;
151
152 #define MODE_EXT_MS_STEREO 2
153 #define MODE_EXT_I_STEREO 1
154
155 /* layer 3 huffman tables */
156 typedef struct HuffTable {
157 int xsize;
158 const uint8_t *bits;
159 const uint16_t *codes;
160 } HuffTable;
161
162 #include "mpegaudiodectab.h"
163
164 static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
165 static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
166
167 /* vlc structure for decoding layer 3 huffman tables */
168 static VLC huff_vlc[16];
169 static VLC huff_quad_vlc[2];
170 /* computed from band_size_long */
171 static uint16_t band_index_long[9][23];
172 /* XXX: free when all decoders are closed */
173 #define TABLE_4_3_SIZE (8191 + 16)*4
174 static int8_t *table_4_3_exp;
175 static uint32_t *table_4_3_value;
176 /* intensity stereo coef table */
177 static int32_t is_table[2][16];
178 static int32_t is_table_lsf[2][2][16];
179 static int32_t csa_table[8][4];
180 static float csa_table_float[8][4];
181 static int32_t mdct_win[8][36];
182
183 /* lower 2 bits: modulo 3, higher bits: shift */
184 static uint16_t scale_factor_modshift[64];
185 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
186 static int32_t scale_factor_mult[15][3];
187 /* mult table for layer 2 group quantization */
188
189 #define SCALE_GEN(v) \
190 { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
191
192 static const int32_t scale_factor_mult2[3][3] = {
193 SCALE_GEN(4.0 / 3.0), /* 3 steps */
194 SCALE_GEN(4.0 / 5.0), /* 5 steps */
195 SCALE_GEN(4.0 / 9.0), /* 9 steps */
196 };
197
198 void ff_mpa_synth_init(MPA_INT *window);
199 static MPA_INT window[512] __attribute__((aligned(16)));
200
201 /* layer 1 unscaling */
202 /* n = number of bits of the mantissa minus 1 */
203 static inline int l1_unscale(int n, int mant, int scale_factor)
204 {
205 int shift, mod;
206 int64_t val;
207
208 shift = scale_factor_modshift[scale_factor];
209 mod = shift & 3;
210 shift >>= 2;
211 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
212 shift += n;
213 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
214 return (int)((val + (1LL << (shift - 1))) >> shift);
215 }
216
217 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
218 {
219 int shift, mod, val;
220
221 shift = scale_factor_modshift[scale_factor];
222 mod = shift & 3;
223 shift >>= 2;
224
225 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
226 /* NOTE: at this point, 0 <= shift <= 21 */
227 if (shift > 0)
228 val = (val + (1 << (shift - 1))) >> shift;
229 return val;
230 }
231
232 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
233 static inline int l3_unscale(int value, int exponent)
234 {
235 unsigned int m;
236 int e;
237
238 e = table_4_3_exp [4*value + (exponent&3)];
239 m = table_4_3_value[4*value + (exponent&3)];
240 e -= (exponent >> 2);
241 assert(e>=1);
242 if (e > 31)
243 return 0;
244 m = (m + (1 << (e-1))) >> e;
245
246 return m;
247 }
248
249 /* all integer n^(4/3) computation code */
250 #define DEV_ORDER 13
251
252 #define POW_FRAC_BITS 24
253 #define POW_FRAC_ONE (1 << POW_FRAC_BITS)
254 #define POW_FIX(a) ((int)((a) * POW_FRAC_ONE))
255 #define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
256
257 static int dev_4_3_coefs[DEV_ORDER];
258
259 #if 0 /* unused */
260 static int pow_mult3[3] = {
261 POW_FIX(1.0),
262 POW_FIX(1.25992104989487316476),
263 POW_FIX(1.58740105196819947474),
264 };
265 #endif
266
267 static void int_pow_init(void)
268 {
269 int i, a;
270
271 a = POW_FIX(1.0);
272 for(i=0;i<DEV_ORDER;i++) {
273 a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
274 dev_4_3_coefs[i] = a;
275 }
276 }
277
278 #if 0 /* unused, remove? */
279 /* return the mantissa and the binary exponent */
280 static int int_pow(int i, int *exp_ptr)
281 {
282 int e, er, eq, j;
283 int a, a1;
284
285 /* renormalize */
286 a = i;
287 e = POW_FRAC_BITS;
288 while (a < (1 << (POW_FRAC_BITS - 1))) {
289 a = a << 1;
290 e--;
291 }
292 a -= (1 << POW_FRAC_BITS);
293 a1 = 0;
294 for(j = DEV_ORDER - 1; j >= 0; j--)
295 a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
296 a = (1 << POW_FRAC_BITS) + a1;
297 /* exponent compute (exact) */
298 e = e * 4;
299 er = e % 3;
300 eq = e / 3;
301 a = POW_MULL(a, pow_mult3[er]);
302 while (a >= 2 * POW_FRAC_ONE) {
303 a = a >> 1;
304 eq++;
305 }
306 /* convert to float */
307 while (a < POW_FRAC_ONE) {
308 a = a << 1;
309 eq--;
310 }
311 /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
312 #if POW_FRAC_BITS > FRAC_BITS
313 a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
314 /* correct overflow */
315 if (a >= 2 * (1 << FRAC_BITS)) {
316 a = a >> 1;
317 eq++;
318 }
319 #endif
320 *exp_ptr = eq;
321 return a;
322 }
323 #endif
324
325 static int decode_init(AVCodecContext * avctx)
326 {
327 MPADecodeContext *s = avctx->priv_data;
328 static int init=0;
329 int i, j, k;
330
331 #if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
332 avctx->sample_fmt= SAMPLE_FMT_S32;
333 #else
334 avctx->sample_fmt= SAMPLE_FMT_S16;
335 #endif
336
337 if(avctx->antialias_algo != FF_AA_FLOAT)
338 s->compute_antialias= compute_antialias_integer;
339 else
340 s->compute_antialias= compute_antialias_float;
341
342 if (!init && !avctx->parse_only) {
343 /* scale factors table for layer 1/2 */
344 for(i=0;i<64;i++) {
345 int shift, mod;
346 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
347 shift = (i / 3);
348 mod = i % 3;
349 scale_factor_modshift[i] = mod | (shift << 2);
350 }
351
352 /* scale factor multiply for layer 1 */
353 for(i=0;i<15;i++) {
354 int n, norm;
355 n = i + 2;
356 norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
357 scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
358 scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
359 scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
360 dprintf("%d: norm=%x s=%x %x %x\n",
361 i, norm,
362 scale_factor_mult[i][0],
363 scale_factor_mult[i][1],
364 scale_factor_mult[i][2]);
365 }
366
367 ff_mpa_synth_init(window);
368
369 /* huffman decode tables */
370 for(i=1;i<16;i++) {
371 const HuffTable *h = &mpa_huff_tables[i];
372 int xsize, x, y;
373 unsigned int n;
374 uint8_t tmp_bits [256];
375 uint16_t tmp_codes[256];
376
377 memset(tmp_bits , 0, sizeof(tmp_bits ));
378 memset(tmp_codes, 0, sizeof(tmp_codes));
379
380 xsize = h->xsize;
381 n = xsize * xsize;
382
383 j = 0;
384 for(x=0;x<xsize;x++) {
385 for(y=0;y<xsize;y++){
386 tmp_bits [(x << 4) | y]= h->bits [j ];
387 tmp_codes[(x << 4) | y]= h->codes[j++];
388 }
389 }
390
391 /* XXX: fail test */
392 init_vlc(&huff_vlc[i], 8, 256,
393 tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
394 }
395 for(i=0;i<2;i++) {
396 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
397 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
398 }
399
400 for(i=0;i<9;i++) {
401 k = 0;
402 for(j=0;j<22;j++) {
403 band_index_long[i][j] = k;
404 k += band_size_long[i][j];
405 }
406 band_index_long[i][22] = k;
407 }
408
409 /* compute n ^ (4/3) and store it in mantissa/exp format */
410 table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));
411 if(!table_4_3_exp)
412 return -1;
413 table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));
414 if(!table_4_3_value)
415 return -1;
416
417 int_pow_init();
418 for(i=1;i<TABLE_4_3_SIZE;i++) {
419 double f, fm;
420 int e, m;
421 f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
422 fm = frexp(f, &e);
423 m = (uint32_t)(fm*(1LL<<31) + 0.5);
424 e+= FRAC_BITS - 31 + 5;
425
426 /* normalized to FRAC_BITS */
427 table_4_3_value[i] = m;
428 // av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
429 table_4_3_exp[i] = -e;
430 }
431
432 for(i=0;i<7;i++) {
433 float f;
434 int v;
435 if (i != 6) {
436 f = tan((double)i * M_PI / 12.0);
437 v = FIXR(f / (1.0 + f));
438 } else {
439 v = FIXR(1.0);
440 }
441 is_table[0][i] = v;
442 is_table[1][6 - i] = v;
443 }
444 /* invalid values */
445 for(i=7;i<16;i++)
446 is_table[0][i] = is_table[1][i] = 0.0;
447
448 for(i=0;i<16;i++) {
449 double f;
450 int e, k;
451
452 for(j=0;j<2;j++) {
453 e = -(j + 1) * ((i + 1) >> 1);
454 f = pow(2.0, e / 4.0);
455 k = i & 1;
456 is_table_lsf[j][k ^ 1][i] = FIXR(f);
457 is_table_lsf[j][k][i] = FIXR(1.0);
458 dprintf("is_table_lsf %d %d: %x %x\n",
459 i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
460 }
461 }
462
463 for(i=0;i<8;i++) {
464 float ci, cs, ca;
465 ci = ci_table[i];
466 cs = 1.0 / sqrt(1.0 + ci * ci);
467 ca = cs * ci;
468 csa_table[i][0] = FIXHR(cs/4);
469 csa_table[i][1] = FIXHR(ca/4);
470 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
471 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
472 csa_table_float[i][0] = cs;
473 csa_table_float[i][1] = ca;
474 csa_table_float[i][2] = ca + cs;
475 csa_table_float[i][3] = ca - cs;
476 // printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
477 // av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
478 }
479
480 /* compute mdct windows */
481 for(i=0;i<36;i++) {
482 for(j=0; j<4; j++){
483 double d;
484
485 if(j==2 && i%3 != 1)
486 continue;
487
488 d= sin(M_PI * (i + 0.5) / 36.0);
489 if(j==1){
490 if (i>=30) d= 0;
491 else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
492 else if(i>=18) d= 1;
493 }else if(j==3){
494 if (i< 6) d= 0;
495 else if(i< 12) d= sin(M_PI * (i - 6 + 0.5) / 12.0);
496 else if(i< 18) d= 1;
497 }
498 //merge last stage of imdct into the window coefficients
499 d*= 0.5 / cos(M_PI*(2*i + 19)/72);
500
501 if(j==2)
502 mdct_win[j][i/3] = FIXHR((d / (1<<5)));
503 else
504 mdct_win[j][i ] = FIXHR((d / (1<<5)));
505 // av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
506 }
507 }
508
509 /* NOTE: we do frequency inversion adter the MDCT by changing
510 the sign of the right window coefs */
511 for(j=0;j<4;j++) {
512 for(i=0;i<36;i+=2) {
513 mdct_win[j + 4][i] = mdct_win[j][i];
514 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
515 }
516 }
517
518 #if defined(DEBUG)
519 for(j=0;j<8;j++) {
520 av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
521 for(i=0;i<36;i++)
522 av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
523 av_log(avctx, AV_LOG_DEBUG, "\n");
524 }
525 #endif
526 init = 1;
527 }
528
529 s->inbuf_index = 0;
530 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
531 s->inbuf_ptr = s->inbuf;
532 #ifdef DEBUG
533 s->frame_count = 0;
534 #endif
535 if (avctx->codec_id == CODEC_ID_MP3ADU)
536 s->adu_mode = 1;
537 return 0;
538 }
539
540 /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
541
542 /* cos(i*pi/64) */
543
544 #define COS0_0 FIXHR(0.50060299823519630134/2)
545 #define COS0_1 FIXHR(0.50547095989754365998/2)
546 #define COS0_2 FIXHR(0.51544730992262454697/2)
547 #define COS0_3 FIXHR(0.53104259108978417447/2)
548 #define COS0_4 FIXHR(0.55310389603444452782/2)
549 #define COS0_5 FIXHR(0.58293496820613387367/2)
550 #define COS0_6 FIXHR(0.62250412303566481615/2)
551 #define COS0_7 FIXHR(0.67480834145500574602/2)
552 #define COS0_8 FIXHR(0.74453627100229844977/2)
553 #define COS0_9 FIXHR(0.83934964541552703873/2)
554 #define COS0_10 FIXHR(0.97256823786196069369/2)
555 #define COS0_11 FIXHR(1.16943993343288495515/4)
556 #define COS0_12 FIXHR(1.48416461631416627724/4)
557 #define COS0_13 FIXHR(2.05778100995341155085/8)
558 #define COS0_14 FIXHR(3.40760841846871878570/8)
559 #define COS0_15 FIXHR(10.19000812354805681150/32)
560
561 #define COS1_0 FIXHR(0.50241928618815570551/2)
562 #define COS1_1 FIXHR(0.52249861493968888062/2)
563 #define COS1_2 FIXHR(0.56694403481635770368/2)
564 #define COS1_3 FIXHR(0.64682178335999012954/2)
565 #define COS1_4 FIXHR(0.78815462345125022473/2)
566 #define COS1_5 FIXHR(1.06067768599034747134/4)
567 #define COS1_6 FIXHR(1.72244709823833392782/4)
568 #define COS1_7 FIXHR(5.10114861868916385802/16)
569
570 #define COS2_0 FIXHR(0.50979557910415916894/2)
571 #define COS2_1 FIXHR(0.60134488693504528054/2)
572 #define COS2_2 FIXHR(0.89997622313641570463/2)
573 #define COS2_3 FIXHR(2.56291544774150617881/8)
574
575 #define COS3_0 FIXHR(0.54119610014619698439/2)
576 #define COS3_1 FIXHR(1.30656296487637652785/4)
577
578 #define COS4_0 FIXHR(0.70710678118654752439/2)
579
580 /* butterfly operator */
581 #define BF(a, b, c, s)\
582 {\
583 tmp0 = tab[a] + tab[b];\
584 tmp1 = tab[a] - tab[b];\
585 tab[a] = tmp0;\
586 tab[b] = MULH(tmp1<<(s), c);\
587 }
588
589 #define BF1(a, b, c, d)\
590 {\
591 BF(a, b, COS4_0, 1);\
592 BF(c, d,-COS4_0, 1);\
593 tab[c] += tab[d];\
594 }
595
596 #define BF2(a, b, c, d)\
597 {\
598 BF(a, b, COS4_0, 1);\
599 BF(c, d,-COS4_0, 1);\
600 tab[c] += tab[d];\
601 tab[a] += tab[c];\
602 tab[c] += tab[b];\
603 tab[b] += tab[d];\
604 }
605
606 #define ADD(a, b) tab[a] += tab[b]
607
608 /* DCT32 without 1/sqrt(2) coef zero scaling. */
609 static void dct32(int32_t *out, int32_t *tab)
610 {
611 int tmp0, tmp1;
612
613 /* pass 1 */
614 BF( 0, 31, COS0_0 , 1);
615 BF(15, 16, COS0_15, 5);
616 /* pass 2 */
617 BF( 0, 15, COS1_0 , 1);
618 BF(16, 31,-COS1_0 , 1);
619 /* pass 1 */
620 BF( 7, 24, COS0_7 , 1);
621 BF( 8, 23, COS0_8 , 1);
622 /* pass 2 */
623 BF( 7, 8, COS1_7 , 4);
624 BF(23, 24,-COS1_7 , 4);
625 /* pass 3 */
626 BF( 0, 7, COS2_0 , 1);
627 BF( 8, 15,-COS2_0 , 1);
628 BF(16, 23, COS2_0 , 1);
629 BF(24, 31,-COS2_0 , 1);
630 /* pass 1 */
631 BF( 3, 28, COS0_3 , 1);
632 BF(12, 19, COS0_12, 2);
633 /* pass 2 */
634 BF( 3, 12, COS1_3 , 1);
635 BF(19, 28,-COS1_3 , 1);
636 /* pass 1 */
637 BF( 4, 27, COS0_4 , 1);
638 BF(11, 20, COS0_11, 2);
639 /* pass 2 */
640 BF( 4, 11, COS1_4 , 1);
641 BF(20, 27,-COS1_4 , 1);
642 /* pass 3 */
643 BF( 3, 4, COS2_3 , 3);
644 BF(11, 12,-COS2_3 , 3);
645 BF(19, 20, COS2_3 , 3);
646 BF(27, 28,-COS2_3 , 3);
647 /* pass 4 */
648 BF( 0, 3, COS3_0 , 1);
649 BF( 4, 7,-COS3_0 , 1);
650 BF( 8, 11, COS3_0 , 1);
651 BF(12, 15,-COS3_0 , 1);
652 BF(16, 19, COS3_0 , 1);
653 BF(20, 23,-COS3_0 , 1);
654 BF(24, 27, COS3_0 , 1);
655 BF(28, 31,-COS3_0 , 1);
656
657
658
659 /* pass 1 */
660 BF( 1, 30, COS0_1 , 1);
661 BF(14, 17, COS0_14, 3);
662 /* pass 2 */
663 BF( 1, 14, COS1_1 , 1);
664 BF(17, 30,-COS1_1 , 1);
665 /* pass 1 */
666 BF( 6, 25, COS0_6 , 1);
667 BF( 9, 22, COS0_9 , 1);
668 /* pass 2 */
669 BF( 6, 9, COS1_6 , 2);
670 BF(22, 25,-COS1_6 , 2);
671 /* pass 3 */
672 BF( 1, 6, COS2_1 , 1);
673 BF( 9, 14,-COS2_1 , 1);
674 BF(17, 22, COS2_1 , 1);
675 BF(25, 30,-COS2_1 , 1);
676
677 /* pass 1 */
678 BF( 2, 29, COS0_2 , 1);
679 BF(13, 18, COS0_13, 3);
680 /* pass 2 */
681 BF( 2, 13, COS1_2 , 1);
682 BF(18, 29,-COS1_2 , 1);
683 /* pass 1 */
684 BF( 5, 26, COS0_5 , 1);
685 BF(10, 21, COS0_10, 1);
686 /* pass 2 */
687 BF( 5, 10, COS1_5 , 2);
688 BF(21, 26,-COS1_5 , 2);
689 /* pass 3 */
690 BF( 2, 5, COS2_2 , 1);
691 BF(10, 13,-COS2_2 , 1);
692 BF(18, 21, COS2_2 , 1);
693 BF(26, 29,-COS2_2 , 1);
694 /* pass 4 */
695 BF( 1, 2, COS3_1 , 2);
696 BF( 5, 6,-COS3_1 , 2);
697 BF( 9, 10, COS3_1 , 2);
698 BF(13, 14,-COS3_1 , 2);
699 BF(17, 18, COS3_1 , 2);
700 BF(21, 22,-COS3_1 , 2);
701 BF(25, 26, COS3_1 , 2);
702 BF(29, 30,-COS3_1 , 2);
703
704 /* pass 5 */
705 BF1( 0, 1, 2, 3);
706 BF2( 4, 5, 6, 7);
707 BF1( 8, 9, 10, 11);
708 BF2(12, 13, 14, 15);
709 BF1(16, 17, 18, 19);
710 BF2(20, 21, 22, 23);
711 BF1(24, 25, 26, 27);
712 BF2(28, 29, 30, 31);
713
714 /* pass 6 */
715
716 ADD( 8, 12);
717 ADD(12, 10);
718 ADD(10, 14);
719 ADD(14, 9);
720 ADD( 9, 13);
721 ADD(13, 11);
722 ADD(11, 15);
723
724 out[ 0] = tab[0];
725 out[16] = tab[1];
726 out[ 8] = tab[2];
727 out[24] = tab[3];
728 out[ 4] = tab[4];
729 out[20] = tab[5];
730 out[12] = tab[6];
731 out[28] = tab[7];
732 out[ 2] = tab[8];
733 out[18] = tab[9];
734 out[10] = tab[10];
735 out[26] = tab[11];
736 out[ 6] = tab[12];
737 out[22] = tab[13];
738 out[14] = tab[14];
739 out[30] = tab[15];
740
741 ADD(24, 28);
742 ADD(28, 26);
743 ADD(26, 30);
744 ADD(30, 25);
745 ADD(25, 29);
746 ADD(29, 27);
747 ADD(27, 31);
748
749 out[ 1] = tab[16] + tab[24];
750 out[17] = tab[17] + tab[25];
751 out[ 9] = tab[18] + tab[26];
752 out[25] = tab[19] + tab[27];
753 out[ 5] = tab[20] + tab[28];
754 out[21] = tab[21] + tab[29];
755 out[13] = tab[22] + tab[30];
756 out[29] = tab[23] + tab[31];
757 out[ 3] = tab[24] + tab[20];
758 out[19] = tab[25] + tab[21];
759 out[11] = tab[26] + tab[22];
760 out[27] = tab[27] + tab[23];
761 out[ 7] = tab[28] + tab[18];
762 out[23] = tab[29] + tab[19];
763 out[15] = tab[30] + tab[17];
764 out[31] = tab[31];
765 }
766
767 #if FRAC_BITS <= 15
768
769 static inline int round_sample(int *sum)
770 {
771 int sum1;
772 sum1 = (*sum) >> OUT_SHIFT;
773 *sum &= (1<<OUT_SHIFT)-1;
774 if (sum1 < OUT_MIN)
775 sum1 = OUT_MIN;
776 else if (sum1 > OUT_MAX)
777 sum1 = OUT_MAX;
778 return sum1;
779 }
780
781 # if defined(ARCH_POWERPC_405)
782 /* signed 16x16 -> 32 multiply add accumulate */
783 # define MACS(rt, ra, rb) \
784 asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
785
786 /* signed 16x16 -> 32 multiply */
787 # define MULS(ra, rb) \
788 ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })
789 # else
790 /* signed 16x16 -> 32 multiply add accumulate */
791 # define MACS(rt, ra, rb) rt += (ra) * (rb)
792
793 /* signed 16x16 -> 32 multiply */
794 # define MULS(ra, rb) ((ra) * (rb))
795 # endif
796 #else
797
798 static inline int round_sample(int64_t *sum)
799 {
800 int sum1;
801 sum1 = (int)((*sum) >> OUT_SHIFT);
802 *sum &= (1<<OUT_SHIFT)-1;
803 if (sum1 < OUT_MIN)
804 sum1 = OUT_MIN;
805 else if (sum1 > OUT_MAX)
806 sum1 = OUT_MAX;
807 return sum1;
808 }
809
810 # define MULS(ra, rb) MUL64(ra, rb)
811 #endif
812
813 #define SUM8(sum, op, w, p) \
814 { \
815 sum op MULS((w)[0 * 64], p[0 * 64]);\
816 sum op MULS((w)[1 * 64], p[1 * 64]);\
817 sum op MULS((w)[2 * 64], p[2 * 64]);\
818 sum op MULS((w)[3 * 64], p[3 * 64]);\
819 sum op MULS((w)[4 * 64], p[4 * 64]);\
820 sum op MULS((w)[5 * 64], p[5 * 64]);\
821 sum op MULS((w)[6 * 64], p[6 * 64]);\
822 sum op MULS((w)[7 * 64], p[7 * 64]);\
823 }
824
825 #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
826 { \
827 int tmp;\
828 tmp = p[0 * 64];\
829 sum1 op1 MULS((w1)[0 * 64], tmp);\
830 sum2 op2 MULS((w2)[0 * 64], tmp);\
831 tmp = p[1 * 64];\
832 sum1 op1 MULS((w1)[1 * 64], tmp);\
833 sum2 op2 MULS((w2)[1 * 64], tmp);\
834 tmp = p[2 * 64];\
835 sum1 op1 MULS((w1)[2 * 64], tmp);\
836 sum2 op2 MULS((w2)[2 * 64], tmp);\
837 tmp = p[3 * 64];\
838 sum1 op1 MULS((w1)[3 * 64], tmp);\
839 sum2 op2 MULS((w2)[3 * 64], tmp);\
840 tmp = p[4 * 64];\
841 sum1 op1 MULS((w1)[4 * 64], tmp);\
842 sum2 op2 MULS((w2)[4 * 64], tmp);\
843 tmp = p[5 * 64];\
844 sum1 op1 MULS((w1)[5 * 64], tmp);\
845 sum2 op2 MULS((w2)[5 * 64], tmp);\
846 tmp = p[6 * 64];\
847 sum1 op1 MULS((w1)[6 * 64], tmp);\
848 sum2 op2 MULS((w2)[6 * 64], tmp);\
849 tmp = p[7 * 64];\
850 sum1 op1 MULS((w1)[7 * 64], tmp);\
851 sum2 op2 MULS((w2)[7 * 64], tmp);\
852 }
853
854 void ff_mpa_synth_init(MPA_INT *window)
855 {
856 int i;
857
858 /* max = 18760, max sum over all 16 coefs : 44736 */
859 for(i=0;i<257;i++) {
860 int v;
861 v = mpa_enwindow[i];
862 #if WFRAC_BITS < 16
863 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
864 #endif
865 window[i] = v;
866 if ((i & 63) != 0)
867 v = -v;
868 if (i != 0)
869 window[512 - i] = v;
870 }
871 }
872
873 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
874 32 samples. */
875 /* XXX: optimize by avoiding ring buffer usage */
876 void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
877 MPA_INT *window, int *dither_state,
878 OUT_INT *samples, int incr,
879 int32_t sb_samples[SBLIMIT])
880 {
881 int32_t tmp[32];
882 register MPA_INT *synth_buf;
883 register const MPA_INT *w, *w2, *p;
884 int j, offset, v;
885 OUT_INT *samples2;
886 #if FRAC_BITS <= 15
887 int sum, sum2;
888 #else
889 int64_t sum, sum2;
890 #endif
891
892 dct32(tmp, sb_samples);
893
894 offset = *synth_buf_offset;
895 synth_buf = synth_buf_ptr + offset;
896
897 for(j=0;j<32;j++) {
898 v = tmp[j];
899 #if FRAC_BITS <= 15
900 /* NOTE: can cause a loss in precision if very high amplitude
901 sound */
902 if (v > 32767)
903 v = 32767;
904 else if (v < -32768)
905 v = -32768;
906 #endif
907 synth_buf[j] = v;
908 }
909 /* copy to avoid wrap */
910 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
911
912 samples2 = samples + 31 * incr;
913 w = window;
914 w2 = window + 31;
915
916 sum = *dither_state;
917 p = synth_buf + 16;
918 SUM8(sum, +=, w, p);
919 p = synth_buf + 48;
920 SUM8(sum, -=, w + 32, p);
921 *samples = round_sample(&sum);
922 samples += incr;
923 w++;
924
925 /* we calculate two samples at the same time to avoid one memory
926 access per two sample */
927 for(j=1;j<16;j++) {
928 sum2 = 0;
929 p = synth_buf + 16 + j;
930 SUM8P2(sum, +=, sum2, -=, w, w2, p);
931 p = synth_buf + 48 - j;
932 SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
933
934 *samples = round_sample(&sum);
935 samples += incr;
936 sum += sum2;
937 *samples2 = round_sample(&sum);
938 samples2 -= incr;
939 w++;
940 w2--;
941 }
942
943 p = synth_buf + 32;
944 SUM8(sum, -=, w + 32, p);
945 *samples = round_sample(&sum);
946 *dither_state= sum;
947
948 offset = (offset - 32) & 511;
949 *synth_buf_offset = offset;
950 }
951
952 #define C3 FIXHR(0.86602540378443864676/2)
953
954 /* 0.5 / cos(pi*(2*i+1)/36) */
955 static const int icos36[9] = {
956 FIXR(0.50190991877167369479),
957 FIXR(0.51763809020504152469), //0
958 FIXR(0.55168895948124587824),
959 FIXR(0.61038729438072803416),
960 FIXR(0.70710678118654752439), //1
961 FIXR(0.87172339781054900991),
962 FIXR(1.18310079157624925896),
963 FIXR(1.93185165257813657349), //2
964 FIXR(5.73685662283492756461),
965 };
966
967 /* 0.5 / cos(pi*(2*i+1)/36) */
968 static const int icos36h[9] = {
969 FIXHR(0.50190991877167369479/2),
970 FIXHR(0.51763809020504152469/2), //0
971 FIXHR(0.55168895948124587824/2),
972 FIXHR(0.61038729438072803416/2),
973 FIXHR(0.70710678118654752439/2), //1
974 FIXHR(0.87172339781054900991/2),
975 FIXHR(1.18310079157624925896/4),
976 FIXHR(1.93185165257813657349/4), //2
977 // FIXHR(5.73685662283492756461),
978 };
979
980 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
981 cases. */
982 static void imdct12(int *out, int *in)
983 {
984 int in0, in1, in2, in3, in4, in5, t1, t2;
985
986 in0= in[0*3];
987 in1= in[1*3] + in[0*3];
988 in2= in[2*3] + in[1*3];
989 in3= in[3*3] + in[2*3];
990 in4= in[4*3] + in[3*3];
991 in5= in[5*3] + in[4*3];
992 in5 += in3;
993 in3 += in1;
994
995 in2= MULH(2*in2, C3);
996 in3= MULH(4*in3, C3);
997
998 t1 = in0 - in4;
999 t2 = MULH(2*(in1 - in5), icos36h[4]);
1000
1001 out[ 7]=
1002 out[10]= t1 + t2;
1003 out[ 1]=
1004 out[ 4]= t1 - t2;
1005
1006 in0 += in4>>1;
1007 in4 = in0 + in2;
1008 in5 += 2*in1;
1009 in1 = MULH(in5 + in3, icos36h[1]);
1010 out[ 8]=
1011 out[ 9]= in4 + in1;
1012 out[ 2]=
1013 out[ 3]= in4 - in1;
1014
1015 in0 -= in2;
1016 in5 = MULH(2*(in5 - in3), icos36h[7]);
1017 out[ 0]=
1018 out[ 5]= in0 - in5;
1019 out[ 6]=
1020 out[11]= in0 + in5;
1021 }
1022
1023 /* cos(pi*i/18) */
1024 #define C1 FIXHR(0.98480775301220805936/2)
1025 #define C2 FIXHR(0.93969262078590838405/2)
1026 #define C3 FIXHR(0.86602540378443864676/2)
1027 #define C4 FIXHR(0.76604444311897803520/2)
1028 #define C5 FIXHR(0.64278760968653932632/2)
1029 #define C6 FIXHR(0.5/2)
1030 #define C7 FIXHR(0.34202014332566873304/2)
1031 #define C8 FIXHR(0.17364817766693034885/2)
1032
1033
1034 /* using Lee like decomposition followed by hand coded 9 points DCT */
1035 static void imdct36(int *out, int *buf, int *in, int *win)
1036 {
1037 int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1038 int tmp[18], *tmp1, *in1;
1039
1040 for(i=17;i>=1;i--)
1041 in[i] += in[i-1];
1042 for(i=17;i>=3;i-=2)
1043 in[i] += in[i-2];
1044
1045 for(j=0;j<2;j++) {
1046 tmp1 = tmp + j;
1047 in1 = in + j;
1048 #if 0
1049 //more accurate but slower
1050 int64_t t0, t1, t2, t3;
1051 t2 = in1[2*4] + in1[2*8] - in1[2*2];
1052
1053 t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1054 t1 = in1[2*0] - in1[2*6];
1055 tmp1[ 6] = t1 - (t2>>1);
1056 tmp1[16] = t1 + t2;
1057
1058 t0 = MUL64(2*(in1[2*2] + in1[2*4]), C2);
1059 t1 = MUL64( in1[2*4] - in1[2*8] , -2*C8);
1060 t2 = MUL64(2*(in1[2*2] + in1[2*8]), -C4);
1061
1062 tmp1[10] = (t3 - t0 - t2) >> 32;
1063 tmp1[ 2] = (t3 + t0 + t1) >> 32;
1064 tmp1[14] = (t3 + t2 - t1) >> 32;
1065
1066 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1067 t2 = MUL64(2*(in1[2*1] + in1[2*5]), C1);
1068 t3 = MUL64( in1[2*5] - in1[2*7] , -2*C7);
1069 t0 = MUL64(2*in1[2*3], C3);
1070
1071 t1 = MUL64(2*(in1[2*1] + in1[2*7]), -C5);
1072
1073 tmp1[ 0] = (t2 + t3 + t0) >> 32;
1074 tmp1[12] = (t2 + t1 - t0) >> 32;
1075 tmp1[ 8] = (t3 - t1 - t0) >> 32;
1076 #else
1077 t2 = in1[2*4] + in1[2*8] - in1[2*2];
1078
1079 t3 = in1[2*0] + (in1[2*6]>>1);
1080 t1 = in1[2*0] - in1[2*6];
1081 tmp1[ 6] = t1 - (t2>>1);
1082 tmp1[16] = t1 + t2;
1083
1084 t0 = MULH(2*(in1[2*2] + in1[2*4]), C2);
1085 t1 = MULH( in1[2*4] - in1[2*8] , -2*C8);
1086 t2 = MULH(2*(in1[2*2] + in1[2*8]), -C4);
1087
1088 tmp1[10] = t3 - t0 - t2;
1089 tmp1[ 2] = t3 + t0 + t1;
1090 tmp1[14] = t3 + t2 - t1;
1091
1092 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1093 t2 = MULH(2*(in1[2*1] + in1[2*5]), C1);
1094 t3 = MULH( in1[2*5] - in1[2*7] , -2*C7);
1095 t0 = MULH(2*in1[2*3], C3);
1096
1097 t1 = MULH(2*(in1[2*1] + in1[2*7]), -C5);
1098
1099 tmp1[ 0] = t2 + t3 + t0;
1100 tmp1[12] = t2 + t1 - t0;
1101 tmp1[ 8] = t3 - t1 - t0;
1102 #endif
1103 }
1104
1105 i = 0;
1106 for(j=0;j<4;j++) {
1107 t0 = tmp[i];
1108 t1 = tmp[i + 2];
1109 s0 = t1 + t0;
1110 s2 = t1 - t0;
1111
1112 t2 = tmp[i + 1];
1113 t3 = tmp[i + 3];
1114 s1 = MULH(2*(t3 + t2), icos36h[j]);
1115 s3 = MULL(t3 - t2, icos36[8 - j]);
1116
1117 t0 = s0 + s1;
1118 t1 = s0 - s1;
1119 out[(9 + j)*SBLIMIT] = MULH(t1, win[9 + j]) + buf[9 + j];
1120 out[(8 - j)*SBLIMIT] = MULH(t1, win[8 - j]) + buf[8 - j];
1121 buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1122 buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1123
1124 t0 = s2 + s3;
1125 t1 = s2 - s3;
1126 out[(9 + 8 - j)*SBLIMIT] = MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1127 out[( j)*SBLIMIT] = MULH(t1, win[ j]) + buf[ j];
1128 buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1129 buf[ + j] = MULH(t0, win[18 + j]);
1130 i += 4;
1131 }
1132
1133 s0 = tmp[16];
1134 s1 = MULH(2*tmp[17], icos36h[4]);
1135 t0 = s0 + s1;
1136 t1 = s0 - s1;
1137 out[(9 + 4)*SBLIMIT] = MULH(t1, win[9 + 4]) + buf[9 + 4];
1138 out[(8 - 4)*SBLIMIT] = MULH(t1, win[8 - 4]) + buf[8 - 4];
1139 buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1140 buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1141 }
1142
1143 /* header decoding. MUST check the header before because no
1144 consistency check is done there. Return 1 if free format found and
1145 that the frame size must be computed externally */
1146 static int decode_header(MPADecodeContext *s, uint32_t header)
1147 {
1148 int sample_rate, frame_size, mpeg25, padding;
1149 int sample_rate_index, bitrate_index;
1150 if (header & (1<<20)) {
1151 s->lsf = (header & (1<<19)) ? 0 : 1;
1152 mpeg25 = 0;
1153 } else {
1154 s->lsf = 1;
1155 mpeg25 = 1;
1156 }
1157
1158 s->layer = 4 - ((header >> 17) & 3);
1159 /* extract frequency */
1160 sample_rate_index = (header >> 10) & 3;
1161 sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1162 sample_rate_index += 3 * (s->lsf + mpeg25);
1163 s->sample_rate_index = sample_rate_index;
1164 s->error_protection = ((header >> 16) & 1) ^ 1;
1165 s->sample_rate = sample_rate;
1166
1167 bitrate_index = (header >> 12) & 0xf;
1168 padding = (header >> 9) & 1;
1169 //extension = (header >> 8) & 1;
1170 s->mode = (header >> 6) & 3;
1171 s->mode_ext = (header >> 4) & 3;
1172 //copyright = (header >> 3) & 1;
1173 //original = (header >> 2) & 1;
1174 //emphasis = header & 3;
1175
1176 if (s->mode == MPA_MONO)
1177 s->nb_channels = 1;
1178 else
1179 s->nb_channels = 2;
1180
1181 if (bitrate_index != 0) {
1182 frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1183 s->bit_rate = frame_size * 1000;
1184 switch(s->layer) {
1185 case 1:
1186 frame_size = (frame_size * 12000) / sample_rate;
1187 frame_size = (frame_size + padding) * 4;
1188 break;
1189 case 2:
1190 frame_size = (frame_size * 144000) / sample_rate;
1191 frame_size += padding;
1192 break;
1193 default:
1194 case 3:
1195 frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1196 frame_size += padding;
1197 break;
1198 }
1199 s->frame_size = frame_size;
1200 } else {
1201 /* if no frame size computed, signal it */
1202 if (!s->free_format_frame_size)
1203 return 1;
1204 /* free format: compute bitrate and real frame size from the
1205 frame size we extracted by reading the bitstream */
1206 s->frame_size = s->free_format_frame_size;
1207 switch(s->layer) {
1208 case 1:
1209 s->frame_size += padding * 4;
1210 s->bit_rate = (s->frame_size * sample_rate) / 48000;
1211 break;
1212 case 2:
1213 s->frame_size += padding;
1214 s->bit_rate = (s->frame_size * sample_rate) / 144000;
1215 break;
1216 default:
1217 case 3:
1218 s->frame_size += padding;
1219 s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1220 break;
1221 }
1222 }
1223
1224 #if defined(DEBUG)
1225 dprintf("layer%d, %d Hz, %d kbits/s, ",
1226 s->layer, s->sample_rate, s->bit_rate);
1227 if (s->nb_channels == 2) {
1228 if (s->layer == 3) {
1229 if (s->mode_ext & MODE_EXT_MS_STEREO)
1230 dprintf("ms-");
1231 if (s->mode_ext & MODE_EXT_I_STEREO)
1232 dprintf("i-");
1233 }
1234 dprintf("stereo");
1235 } else {
1236 dprintf("mono");
1237 }
1238 dprintf("\n");
1239 #endif
1240 return 0;
1241 }
1242
1243 /* useful helper to get mpeg audio stream infos. Return -1 if error in
1244 header, otherwise the coded frame size in bytes */
1245 int mpa_decode_header(AVCodecContext *avctx, uint32_t head)
1246 {
1247 MPADecodeContext s1, *s = &s1;
1248 memset( s, 0, sizeof(MPADecodeContext) );
1249
1250 if (ff_mpa_check_header(head) != 0)
1251 return -1;
1252
1253 if (decode_header(s, head) != 0) {
1254 return -1;
1255 }
1256
1257 switch(s->layer) {
1258 case 1:
1259 avctx->frame_size = 384;
1260 break;
1261 case 2:
1262 avctx->frame_size = 1152;
1263 break;
1264 default:
1265 case 3:
1266 if (s->lsf)
1267 avctx->frame_size = 576;
1268 else
1269 avctx->frame_size = 1152;
1270 break;
1271 }
1272
1273 avctx->sample_rate = s->sample_rate;
1274 avctx->channels = s->nb_channels;
1275 avctx->bit_rate = s->bit_rate;
1276 avctx->sub_id = s->layer;
1277 return s->frame_size;
1278 }
1279
1280 /* return the number of decoded frames */
1281 static int mp_decode_layer1(MPADecodeContext *s)
1282 {
1283 int bound, i, v, n, ch, j, mant;
1284 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1285 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1286
1287 if (s->mode == MPA_JSTEREO)
1288 bound = (s->mode_ext + 1) * 4;
1289 else
1290 bound = SBLIMIT;
1291
1292 /* allocation bits */
1293 for(i=0;i<bound;i++) {
1294 for(ch=0;ch<s->nb_channels;ch++) {
1295 allocation[ch][i] = get_bits(&s->gb, 4);
1296 }
1297 }
1298 for(i=bound;i<SBLIMIT;i++) {
1299 allocation[0][i] = get_bits(&s->gb, 4);
1300 }
1301
1302 /* scale factors */
1303 for(i=0;i<bound;i++) {
1304 for(ch=0;ch<s->nb_channels;ch++) {
1305 if (allocation[ch][i])
1306 scale_factors[ch][i] = get_bits(&s->gb, 6);
1307 }
1308 }
1309 for(i=bound;i<SBLIMIT;i++) {
1310 if (allocation[0][i]) {
1311 scale_factors[0][i] = get_bits(&s->gb, 6);
1312 scale_factors[1][i] = get_bits(&s->gb, 6);
1313 }
1314 }
1315
1316 /* compute samples */
1317 for(j=0;j<12;j++) {
1318 for(i=0;i<bound;i++) {
1319 for(ch=0;ch<s->nb_channels;ch++) {
1320 n = allocation[ch][i];
1321 if (n) {
1322 mant = get_bits(&s->gb, n + 1);
1323 v = l1_unscale(n, mant, scale_factors[ch][i]);
1324 } else {
1325 v = 0;
1326 }
1327 s->sb_samples[ch][j][i] = v;
1328 }
1329 }
1330 for(i=bound;i<SBLIMIT;i++) {
1331 n = allocation[0][i];
1332 if (n) {
1333 mant = get_bits(&s->gb, n + 1);
1334 v = l1_unscale(n, mant, scale_factors[0][i]);
1335 s->sb_samples[0][j][i] = v;
1336 v = l1_unscale(n, mant, scale_factors[1][i]);
1337 s->sb_samples[1][j][i] = v;
1338 } else {
1339 s->sb_samples[0][j][i] = 0;
1340 s->sb_samples[1][j][i] = 0;
1341 }
1342 }
1343 }
1344 return 12;
1345 }
1346
1347 /* bitrate is in kb/s */
1348 int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1349 {
1350 int ch_bitrate, table;
1351
1352 ch_bitrate = bitrate / nb_channels;
1353 if (!lsf) {
1354 if ((freq == 48000 && ch_bitrate >= 56) ||
1355 (ch_bitrate >= 56 && ch_bitrate <= 80))
1356 table = 0;
1357 else if (freq != 48000 && ch_bitrate >= 96)
1358 table = 1;
1359 else if (freq != 32000 && ch_bitrate <= 48)
1360 table = 2;
1361 else
1362 table = 3;
1363 } else {
1364 table = 4;
1365 }
1366 return table;
1367 }
1368
1369 static int mp_decode_layer2(MPADecodeContext *s)
1370 {
1371 int sblimit; /* number of used subbands */
1372 const unsigned char *alloc_table;
1373 int table, bit_alloc_bits, i, j, ch, bound, v;
1374 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1375 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1376 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1377 int scale, qindex, bits, steps, k, l, m, b;
1378
1379 /* select decoding table */
1380 table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1381 s->sample_rate, s->lsf);
1382 sblimit = sblimit_table[table];
1383 alloc_table = alloc_tables[table];
1384
1385 if (s->mode == MPA_JSTEREO)
1386 bound = (s->mode_ext + 1) * 4;
1387 else
1388 bound = sblimit;
1389
1390 dprintf("bound=%d sblimit=%d\n", bound, sblimit);
1391
1392 /* sanity check */
1393 if( bound > sblimit ) bound = sblimit;
1394
1395 /* parse bit allocation */
1396 j = 0;
1397 for(i=0;i<bound;i++) {
1398 bit_alloc_bits = alloc_table[j];
1399 for(ch=0;ch<s->nb_channels;ch++) {
1400 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1401 }
1402 j += 1 << bit_alloc_bits;
1403 }
1404 for(i=bound;i<sblimit;i++) {
1405 bit_alloc_bits = alloc_table[j];
1406 v = get_bits(&s->gb, bit_alloc_bits);
1407 bit_alloc[0][i] = v;
1408 bit_alloc[1][i] = v;
1409 j += 1 << bit_alloc_bits;
1410 }
1411
1412 #ifdef DEBUG
1413 {
1414 for(ch=0;ch<s->nb_channels;ch++) {
1415 for(i=0;i<sblimit;i++)
1416 dprintf(" %d", bit_alloc[ch][i]);
1417 dprintf("\n");
1418 }
1419 }
1420 #endif
1421
1422 /* scale codes */
1423 for(i=0;i<sblimit;i++) {
1424 for(ch=0;ch<s->nb_channels;ch++) {
1425 if (bit_alloc[ch][i])
1426 scale_code[ch][i] = get_bits(&s->gb, 2);
1427 }
1428 }
1429
1430 /* scale factors */
1431 for(i=0;i<sblimit;i++) {
1432 for(ch=0;ch<s->nb_channels;ch++) {
1433 if (bit_alloc[ch][i]) {
1434 sf = scale_factors[ch][i];
1435 switch(scale_code[ch][i]) {
1436 default:
1437 case 0:
1438 sf[0] = get_bits(&s->gb, 6);
1439 sf[1] = get_bits(&s->gb, 6);
1440 sf[2] = get_bits(&s->gb, 6);
1441 break;
1442 case 2:
1443 sf[0] = get_bits(&s->gb, 6);
1444 sf[1] = sf[0];
1445 sf[2] = sf[0];
1446 break;
1447 case 1:
1448 sf[0] = get_bits(&s->gb, 6);
1449 sf[2] = get_bits(&s->gb, 6);
1450 sf[1] = sf[0];
1451 break;
1452 case 3:
1453 sf[0] = get_bits(&s->gb, 6);
1454 sf[2] = get_bits(&s->gb, 6);
1455 sf[1] = sf[2];
1456 break;
1457 }
1458 }
1459 }
1460 }
1461
1462 #ifdef DEBUG
1463 for(ch=0;ch<s->nb_channels;ch++) {
1464 for(i=0;i<sblimit;i++) {
1465 if (bit_alloc[ch][i]) {
1466 sf = scale_factors[ch][i];
1467 dprintf(" %d %d %d", sf[0], sf[1], sf[2]);
1468 } else {
1469 dprintf(" -");
1470 }
1471 }
1472 dprintf("\n");
1473 }
1474 #endif
1475
1476 /* samples */
1477 for(k=0;k<3;k++) {
1478 for(l=0;l<12;l+=3) {
1479 j = 0;
1480 for(i=0;i<bound;i++) {
1481 bit_alloc_bits = alloc_table[j];
1482 for(ch=0;ch<s->nb_channels;ch++) {
1483 b = bit_alloc[ch][i];
1484 if (b) {
1485 scale = scale_factors[ch][i][k];
1486 qindex = alloc_table[j+b];
1487 bits = quant_bits[qindex];
1488 if (bits < 0) {
1489 /* 3 values at the same time */
1490 v = get_bits(&s->gb, -bits);
1491 steps = quant_steps[qindex];
1492 s->sb_samples[ch][k * 12 + l + 0][i] =
1493 l2_unscale_group(steps, v % steps, scale);
1494 v = v / steps;
1495 s->sb_samples[ch][k * 12 + l + 1][i] =
1496 l2_unscale_group(steps, v % steps, scale);
1497 v = v / steps;
1498 s->sb_samples[ch][k * 12 + l + 2][i] =
1499 l2_unscale_group(steps, v, scale);
1500 } else {
1501 for(m=0;m<3;m++) {
1502 v = get_bits(&s->gb, bits);
1503 v = l1_unscale(bits - 1, v, scale);
1504 s->sb_samples[ch][k * 12 + l + m][i] = v;
1505 }
1506 }
1507 } else {
1508 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1509 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1510 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1511 }
1512 }
1513 /* next subband in alloc table */
1514 j += 1 << bit_alloc_bits;
1515 }
1516 /* XXX: find a way to avoid this duplication of code */
1517 for(i=bound;i<sblimit;i++) {
1518 bit_alloc_bits = alloc_table[j];
1519 b = bit_alloc[0][i];
1520 if (b) {
1521 int mant, scale0, scale1;
1522 scale0 = scale_factors[0][i][k];
1523 scale1 = scale_factors[1][i][k];
1524 qindex = alloc_table[j+b];
1525 bits = quant_bits[qindex];
1526 if (bits < 0) {
1527 /* 3 values at the same time */
1528 v = get_bits(&s->gb, -bits);
1529 steps = quant_steps[qindex];
1530 mant = v % steps;
1531 v = v / steps;
1532 s->sb_samples[0][k * 12 + l + 0][i] =
1533 l2_unscale_group(steps, mant, scale0);
1534 s->sb_samples[1][k * 12 + l + 0][i] =
1535 l2_unscale_group(steps, mant, scale1);
1536 mant = v % steps;
1537 v = v / steps;
1538 s->sb_samples[0][k * 12 + l + 1][i] =
1539 l2_unscale_group(steps, mant, scale0);
1540 s->sb_samples[1][k * 12 + l + 1][i] =
1541 l2_unscale_group(steps, mant, scale1);
1542 s->sb_samples[0][k * 12 + l + 2][i] =
1543 l2_unscale_group(steps, v, scale0);
1544 s->sb_samples[1][k * 12 + l + 2][i] =
1545 l2_unscale_group(steps, v, scale1);
1546 } else {
1547 for(m=0;m<3;m++) {
1548 mant = get_bits(&s->gb, bits);
1549 s->sb_samples[0][k * 12 + l + m][i] =
1550 l1_unscale(bits - 1, mant, scale0);
1551 s->sb_samples[1][k * 12 + l + m][i] =
1552 l1_unscale(bits - 1, mant, scale1);
1553 }
1554 }
1555 } else {
1556 s->sb_samples[0][k * 12 + l + 0][i] = 0;
1557 s->sb_samples[0][k * 12 + l + 1][i] = 0;
1558 s->sb_samples[0][k * 12 + l + 2][i] = 0;
1559 s->sb_samples[1][k * 12 + l + 0][i] = 0;
1560 s->sb_samples[1][k * 12 + l + 1][i] = 0;
1561 s->sb_samples[1][k * 12 + l + 2][i] = 0;
1562 }
1563 /* next subband in alloc table */
1564 j += 1 << bit_alloc_bits;
1565 }
1566 /* fill remaining samples to zero */
1567 for(i=sblimit;i<SBLIMIT;i++) {
1568 for(ch=0;ch<s->nb_channels;ch++) {
1569 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1570 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1571 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1572 }
1573 }
1574 }
1575 }
1576 return 3 * 12;
1577 }
1578
1579 /*
1580 * Seek back in the stream for backstep bytes (at most 511 bytes)
1581 */
1582 static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)
1583 {
1584 uint8_t *ptr;
1585
1586 /* compute current position in stream */
1587 ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));
1588
1589 /* copy old data before current one */
1590 ptr -= backstep;
1591 memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
1592 BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
1593 /* init get bits again */
1594 init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);
1595
1596 /* prepare next buffer */
1597 s->inbuf_index ^= 1;
1598 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1599 s->old_frame_size = s->frame_size;
1600 }
1601
1602 static inline void lsf_sf_expand(int *slen,
1603 int sf, int n1, int n2, int n3)
1604 {
1605 if (n3) {
1606 slen[3] = sf % n3;
1607 sf /= n3;
1608 } else {
1609 slen[3] = 0;
1610 }
1611 if (n2) {
1612 slen[2] = sf % n2;
1613 sf /= n2;
1614 } else {
1615 slen[2] = 0;
1616 }
1617 slen[1] = sf % n1;
1618 sf /= n1;
1619 slen[0] = sf;
1620 }
1621
1622 static void exponents_from_scale_factors(MPADecodeContext *s,
1623 GranuleDef *g,
1624 int16_t *exponents)
1625 {
1626 const uint8_t *bstab, *pretab;
1627 int len, i, j, k, l, v0, shift, gain, gains[3];
1628 int16_t *exp_ptr;
1629
1630 exp_ptr = exponents;
1631 gain = g->global_gain - 210;
1632 shift = g->scalefac_scale + 1;
1633
1634 bstab = band_size_long[s->sample_rate_index];
1635 pretab = mpa_pretab[g->preflag];
1636 for(i=0;i<g->long_end;i++) {
1637 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
1638 len = bstab[i];
1639 for(j=len;j>0;j--)
1640 *exp_ptr++ = v0;
1641 }
1642
1643 if (g->short_start < 13) {
1644 bstab = band_size_short[s->sample_rate_index];
1645 gains[0] = gain - (g->subblock_gain[0] << 3);
1646 gains[1] = gain - (g->subblock_gain[1] << 3);
1647 gains[2] = gain - (g->subblock_gain[2] << 3);
1648 k = g->long_end;
1649 for(i=g->short_start;i<13;i++) {
1650 len = bstab[i];
1651 for(l=0;l<3;l++) {
1652 v0 = gains[l] - (g->scale_factors[k++] << shift);
1653 for(j=len;j>0;j--)
1654 *exp_ptr++ = v0;
1655 }
1656 }
1657 }
1658 }
1659
1660 /* handle n = 0 too */
1661 static inline int get_bitsz(GetBitContext *s, int n)
1662 {
1663 if (n == 0)
1664 return 0;
1665 else
1666 return get_bits(s, n);
1667 }
1668
1669 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1670 int16_t *exponents, int end_pos)
1671 {
1672 int s_index;
1673 int linbits, code, x, y, l, v, i, j, k, pos;
1674 int last_pos;
1675 VLC *vlc;
1676
1677 /* low frequencies (called big values) */
1678 s_index = 0;
1679 for(i=0;i<3;i++) {
1680 j = g->region_size[i];
1681 if (j == 0)
1682 continue;
1683 /* select vlc table */
1684 k = g->table_select[i];
1685 l = mpa_huff_data[k][0];
1686 linbits = mpa_huff_data[k][1];
1687 vlc = &huff_vlc[l];
1688
1689 if(!l){
1690 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*j);
1691 s_index += 2*j;
1692 continue;
1693 }
1694
1695 /* read huffcode and compute each couple */
1696 for(;j>0;j--) {
1697 if (get_bits_count(&s->gb) >= end_pos)
1698 break;
1699 y = get_vlc2(&s->gb, vlc->table, 8, 3);
1700
1701 if(!y){
1702 g->sb_hybrid[s_index ] =
1703 g->sb_hybrid[s_index+1] = 0;
1704 s_index += 2;
1705 continue;
1706 }
1707
1708 x = y >> 4;
1709 y = y & 0x0f;
1710
1711 dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
1712 i, g->region_size[i] - j, x, y, exponents[s_index]);
1713 if (x) {
1714 if (x == 15)
1715 x += get_bitsz(&s->gb, linbits);
1716 v = l3_unscale(x, exponents[s_index]);
1717 if (get_bits1(&s->gb))
1718 v = -v;
1719 } else {
1720 v = 0;
1721 }
1722 g->sb_hybrid[s_index++] = v;
1723 if (y) {
1724 if (y == 15)
1725 y += get_bitsz(&s->gb, linbits);
1726 v = l3_unscale(y, exponents[s_index]);
1727 if (get_bits1(&s->gb))
1728 v = -v;
1729 } else {
1730 v = 0;
1731 }
1732 g->sb_hybrid[s_index++] = v;
1733 }
1734 }
1735
1736 /* high frequencies */
1737 vlc = &huff_quad_vlc[g->count1table_select];
1738 last_pos=0;
1739 while (s_index <= 572) {
1740 pos = get_bits_count(&s->gb);
1741 if (pos >= end_pos) {
1742 if (pos > end_pos && last_pos){
1743 /* some encoders generate an incorrect size for this
1744 part. We must go back into the data */
1745 s_index -= 4;
1746 init_get_bits(&s->gb, s->gb.buffer + 4*(last_pos>>5), s->gb.size_in_bits - (last_pos&(~31)));
1747 skip_bits(&s->gb, last_pos&31);
1748 }
1749 break;
1750 }
1751 last_pos= pos;
1752
1753 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1754 dprintf("t=%d code=%d\n", g->count1table_select, code);
1755 g->sb_hybrid[s_index+0]=
1756 g->sb_hybrid[s_index+1]=
1757 g->sb_hybrid[s_index+2]=
1758 g->sb_hybrid[s_index+3]= 0;
1759 while(code){
1760 const static int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1761 int pos= s_index+idxtab[code];
1762 code ^= 8>>idxtab[code];
1763 v = l3_unscale(1, exponents[pos]);
1764 if(get_bits1(&s->gb))
1765 v = -v;
1766 g->sb_hybrid[pos] = v;
1767 }
1768 s_index+=4;
1769 }
1770 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1771 return 0;
1772 }
1773
1774 /* Reorder short blocks from bitstream order to interleaved order. It
1775 would be faster to do it in parsing, but the code would be far more
1776 complicated */
1777 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1778 {
1779 int i, j, len;
1780 int32_t *ptr, *dst, *ptr1;
1781 int32_t tmp[576];
1782
1783 if (g->block_type != 2)
1784 return;
1785
1786 if (g->switch_point) {
1787 if (s->sample_rate_index != 8) {
1788 ptr = g->sb_hybrid + 36;
1789 } else {
1790 ptr = g->sb_hybrid + 48;
1791 }
1792 } else {
1793 ptr = g->sb_hybrid;
1794 }
1795
1796 for(i=g->short_start;i<13;i++) {
1797 len = band_size_short[s->sample_rate_index][i];
1798 ptr1 = ptr;
1799 dst = tmp;
1800 for(j=len;j>0;j--) {
1801 *dst++ = ptr[0*len];
1802 *dst++ = ptr[1*len];
1803 *dst++ = ptr[2*len];
1804 ptr++;
1805 }
1806 ptr+=2*len;
1807 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1808 }
1809 }
1810
1811 #define ISQRT2 FIXR(0.70710678118654752440)
1812
1813 static void compute_stereo(MPADecodeContext *s,
1814 GranuleDef *g0, GranuleDef *g1)
1815 {
1816 int i, j, k, l;
1817 int32_t v1, v2;
1818 int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1819 int32_t (*is_tab)[16];
1820 int32_t *tab0, *tab1;
1821 int non_zero_found_short[3];
1822
1823 /* intensity stereo */
1824 if (s->mode_ext & MODE_EXT_I_STEREO) {
1825 if (!s->lsf) {
1826 is_tab = is_table;
1827 sf_max = 7;
1828 } else {
1829 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1830 sf_max = 16;
1831 }
1832
1833 tab0 = g0->sb_hybrid + 576;
1834 tab1 = g1->sb_hybrid + 576;
1835
1836 non_zero_found_short[0] = 0;
1837 non_zero_found_short[1] = 0;
1838 non_zero_found_short[2] = 0;
1839 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1840 for(i = 12;i >= g1->short_start;i--) {
1841 /* for last band, use previous scale factor */
1842 if (i != 11)
1843 k -= 3;
1844 len = band_size_short[s->sample_rate_index][i];
1845 for(l=2;l>=0;l--) {
1846 tab0 -= len;
1847 tab1 -= len;
1848 if (!non_zero_found_short[l]) {
1849 /* test if non zero band. if so, stop doing i-stereo */
1850 for(j=0;j<len;j++) {
1851 if (tab1[j] != 0) {
1852 non_zero_found_short[l] = 1;
1853 goto found1;
1854 }
1855 }
1856 sf = g1->scale_factors[k + l];
1857 if (sf >= sf_max)
1858 goto found1;
1859
1860 v1 = is_tab[0][sf];
1861 v2 = is_tab[1][sf];
1862 for(j=0;j<len;j++) {
1863 tmp0 = tab0[j];
1864 tab0[j] = MULL(tmp0, v1);
1865 tab1[j] = MULL(tmp0, v2);
1866 }
1867 } else {
1868 found1:
1869 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1870 /* lower part of the spectrum : do ms stereo
1871 if enabled */
1872 for(j=0;j<len;j++) {
1873 tmp0 = tab0[j];
1874 tmp1 = tab1[j];
1875 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1876 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1877 }
1878 }
1879 }
1880 }
1881 }
1882
1883 non_zero_found = non_zero_found_short[0] |
1884 non_zero_found_short[1] |
1885 non_zero_found_short[2];
1886
1887 for(i = g1->long_end - 1;i >= 0;i--) {
1888 len = band_size_long[s->sample_rate_index][i];
1889 tab0 -= len;
1890 tab1 -= len;
1891 /* test if non zero band. if so, stop doing i-stereo */
1892 if (!non_zero_found) {
1893 for(j=0;j<len;j++) {
1894 if (tab1[j] != 0) {
1895 non_zero_found = 1;
1896 goto found2;
1897 }
1898 }
1899 /* for last band, use previous scale factor */
1900 k = (i == 21) ? 20 : i;
1901 sf = g1->scale_factors[k];
1902 if (sf >= sf_max)
1903 goto found2;
1904 v1 = is_tab[0][sf];
1905 v2 = is_tab[1][sf];
1906 for(j=0;j<len;j++) {
1907 tmp0 = tab0[j];
1908 tab0[j] = MULL(tmp0, v1);
1909 tab1[j] = MULL(tmp0, v2);
1910 }
1911 } else {
1912 found2:
1913 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1914 /* lower part of the spectrum : do ms stereo
1915 if enabled */
1916 for(j=0;j<len;j++) {
1917 tmp0 = tab0[j];
1918 tmp1 = tab1[j];
1919 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1920 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1921 }
1922 }
1923 }
1924 }
1925 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1926 /* ms stereo ONLY */
1927 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1928 global gain */
1929 tab0 = g0->sb_hybrid;
1930 tab1 = g1->sb_hybrid;
1931 for(i=0;i<576;i++) {
1932 tmp0 = tab0[i];
1933 tmp1 = tab1[i];
1934 tab0[i] = tmp0 + tmp1;
1935 tab1[i] = tmp0 - tmp1;
1936 }
1937 }
1938 }
1939
1940 static void compute_antialias_integer(MPADecodeContext *s,
1941 GranuleDef *g)
1942 {
1943 int32_t *ptr, *csa;
1944 int n, i;
1945
1946 /* we antialias only "long" bands */
1947 if (g->block_type == 2) {
1948 if (!g->switch_point)
1949 return;
1950 /* XXX: check this for 8000Hz case */
1951 n = 1;
1952 } else {
1953 n = SBLIMIT - 1;
1954 }
1955
1956 ptr = g->sb_hybrid + 18;
1957 for(i = n;i > 0;i--) {
1958 int tmp0, tmp1, tmp2;
1959 csa = &csa_table[0][0];
1960 #define INT_AA(j) \
1961 tmp0 = ptr[-1-j];\
1962 tmp1 = ptr[ j];\
1963 tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1964 ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1965 ptr[ j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1966
1967 INT_AA(0)
1968 INT_AA(1)
1969 INT_AA(2)
1970 INT_AA(3)
1971 INT_AA(4)
1972 INT_AA(5)
1973 INT_AA(6)
1974 INT_AA(7)
1975
1976 ptr += 18;
1977 }
1978 }
1979
1980 static void compute_antialias_float(MPADecodeContext *s,
1981 GranuleDef *g)
1982 {
1983 int32_t *ptr;
1984 int n, i;
1985
1986 /* we antialias only "long" bands */
1987 if (g->block_type == 2) {
1988 if (!g->switch_point)
1989 return;
1990 /* XXX: check this for 8000Hz case */
1991 n = 1;
1992 } else {
1993 n = SBLIMIT - 1;
1994 }
1995
1996 ptr = g->sb_hybrid + 18;
1997 for(i = n;i > 0;i--) {
1998 float tmp0, tmp1;
1999 float *csa = &csa_table_float[0][0];
2000 #define FLOAT_AA(j)\
2001 tmp0= ptr[-1-j];\
2002 tmp1= ptr[ j];\
2003 ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
2004 ptr[ j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
2005
2006 FLOAT_AA(0)
2007 FLOAT_AA(1)
2008 FLOAT_AA(2)
2009 FLOAT_AA(3)
2010 FLOAT_AA(4)
2011 FLOAT_AA(5)
2012 FLOAT_AA(6)
2013 FLOAT_AA(7)
2014
2015 ptr += 18;
2016 }
2017 }
2018
2019 static void compute_imdct(MPADecodeContext *s,
2020 GranuleDef *g,
2021 int32_t *sb_samples,
2022 int32_t *mdct_buf)
2023 {
2024 int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
2025 int32_t out2[12];
2026 int i, j, mdct_long_end, v, sblimit;
2027
2028 /* find last non zero block */
2029 ptr = g->sb_hybrid + 576;
2030 ptr1 = g->sb_hybrid + 2 * 18;
2031 while (ptr >= ptr1) {
2032 ptr -= 6;
2033 v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
2034 if (v != 0)
2035 break;
2036 }
2037 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
2038
2039 if (g->block_type == 2) {
2040 /* XXX: check for 8000 Hz */
2041 if (g->switch_point)
2042 mdct_long_end = 2;
2043 else
2044 mdct_long_end = 0;
2045 } else {
2046 mdct_long_end = sblimit;
2047 }
2048
2049 buf = mdct_buf;
2050 ptr = g->sb_hybrid;
2051 for(j=0;j<mdct_long_end;j++) {
2052 /* apply window & overlap with previous buffer */
2053 out_ptr = sb_samples + j;
2054 /* select window */
2055 if (g->switch_point && j < 2)
2056 win1 = mdct_win[0];
2057 else
2058 win1 = mdct_win[g->block_type];
2059 /* select frequency inversion */
2060 win = win1 + ((4 * 36) & -(j & 1));
2061 imdct36(out_ptr, buf, ptr, win);
2062 out_ptr += 18*SBLIMIT;
2063 ptr += 18;
2064 buf += 18;
2065 }
2066 for(j=mdct_long_end;j<sblimit;j++) {
2067 /* select frequency inversion */
2068 win = mdct_win[2] + ((4 * 36) & -(j & 1));
2069 out_ptr = sb_samples + j;
2070
2071 for(i=0; i<6; i++){
2072 *out_ptr = buf[i];
2073 out_ptr += SBLIMIT;
2074 }
2075 imdct12(out2, ptr + 0);
2076 for(i=0;i<6;i++) {
2077 *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
2078 buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
2079 out_ptr += SBLIMIT;
2080 }
2081 imdct12(out2, ptr + 1);
2082 for(i=0;i<6;i++) {
2083 *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
2084 buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
2085 out_ptr += SBLIMIT;
2086 }
2087 imdct12(out2, ptr + 2);
2088 for(i=0;i<6;i++) {
2089 buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
2090 buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
2091 buf[i + 6*2] = 0;
2092 }
2093 ptr += 18;
2094 buf += 18;
2095 }
2096 /* zero bands */
2097 for(j=sblimit;j<SBLIMIT;j++) {
2098 /* overlap */
2099 out_ptr = sb_samples + j;
2100 for(i=0;i<18;i++) {
2101 *out_ptr = buf[i];
2102 buf[i] = 0;
2103 out_ptr += SBLIMIT;
2104 }
2105 buf += 18;
2106 }
2107 }
2108
2109 #if defined(DEBUG)
2110 void sample_dump(int fnum, int32_t *tab, int n)
2111 {
2112 static FILE *files[16], *f;
2113 char buf[512];
2114 int i;
2115 int32_t v;
2116
2117 f = files[fnum];
2118 if (!f) {
2119 snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
2120 fnum,
2121 #ifdef USE_HIGHPRECISION
2122 "hp"
2123 #else
2124 "lp"
2125 #endif
2126 );
2127 f = fopen(buf, "w");
2128 if (!f)
2129 return;
2130 files[fnum] = f;
2131 }
2132
2133 if (fnum == 0) {
2134 static int pos = 0;
2135 av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
2136 for(i=0;i<n;i++) {
2137 av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
2138 if ((i % 18) == 17)
2139 av_log(NULL, AV_LOG_DEBUG, "\n");
2140 }
2141 pos += n;
2142 }
2143 for(i=0;i<n;i++) {
2144 /* normalize to 23 frac bits */
2145 v = tab[i] << (23 - FRAC_BITS);
2146 fwrite(&v, 1, sizeof(int32_t), f);
2147 }
2148 }
2149 #endif
2150
2151
2152 /* main layer3 decoding function */
2153 static int mp_decode_layer3(MPADecodeContext *s)
2154 {
2155 int nb_granules, main_data_begin, private_bits;
2156 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
2157 GranuleDef granules[2][2], *g;
2158 int16_t exponents[576];
2159
2160 /* read side info */
2161 if (s->lsf) {
2162 main_data_begin = get_bits(&s->gb, 8);
2163 private_bits = get_bits(&s->gb, s->nb_channels);
2164 nb_granules = 1;
2165 } else {
2166 main_data_begin = get_bits(&s->gb, 9);
2167 if (s->nb_channels == 2)
2168 private_bits = get_bits(&s->gb, 3);
2169 else
2170 private_bits = get_bits(&s->gb, 5);
2171 nb_granules = 2;
2172 for(ch=0;ch<s->nb_channels;ch++) {
2173 granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2174 granules[ch][1].scfsi = get_bits(&s->gb, 4);
2175 }
2176 }
2177
2178 for(gr=0;gr<nb_granules;gr++) {
2179 for(ch=0;ch<s->nb_channels;ch++) {
2180 dprintf("gr=%d ch=%d: side_info\n", gr, ch);
2181 g = &granules[ch][gr];
2182 g->part2_3_length = get_bits(&s->gb, 12);
2183 g->big_values = get_bits(&s->gb, 9);
2184 g->global_gain = get_bits(&s->gb, 8);
2185 /* if MS stereo only is selected, we precompute the
2186 1/sqrt(2) renormalization factor */
2187 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2188 MODE_EXT_MS_STEREO)
2189 g->global_gain -= 2;
2190 if (s->lsf)
2191 g->scalefac_compress = get_bits(&s->gb, 9);
2192 else
2193 g->scalefac_compress = get_bits(&s->gb, 4);
2194 blocksplit_flag = get_bits(&s->gb, 1);
2195 if (blocksplit_flag) {
2196 g->block_type = get_bits(&s->gb, 2);
2197 if (g->block_type == 0)
2198 return -1;
2199 g->switch_point = get_bits(&s->gb, 1);
2200 for(i=0;i<2;i++)
2201 g->table_select[i] = get_bits(&s->gb, 5);
2202 for(i=0;i<3;i++)
2203 g->subblock_gain[i] = get_bits(&s->gb, 3);
2204 /* compute huffman coded region sizes */
2205 if (g->block_type == 2)
2206 g->region_size[0] = (36 / 2);
2207 else {
2208 if (s->sample_rate_index <= 2)
2209 g->region_size[0] = (36 / 2);
2210 else if (s->sample_rate_index != 8)
2211 g->region_size[0] = (54 / 2);
2212 else
2213 g->region_size[0] = (108 / 2);
2214 }
2215 g->region_size[1] = (576 / 2);
2216 } else {
2217 int region_address1, region_address2, l;
2218 g->block_type = 0;
2219 g->switch_point = 0;
2220 for(i=0;i<3;i++)
2221 g->table_select[i] = get_bits(&s->gb, 5);
2222 /* compute huffman coded region sizes */
2223 region_address1 = get_bits(&s->gb, 4);
2224 region_address2 = get_bits(&s->gb, 3);
2225 dprintf("region1=%d region2=%d\n",
2226 region_address1, region_address2);
2227 g->region_size[0] =
2228 band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2229 l = region_address1 + region_address2 + 2;
2230 /* should not overflow */
2231 if (l > 22)
2232 l = 22;
2233 g->region_size[1] =
2234 band_index_long[s->sample_rate_index][l] >> 1;
2235 }
2236 /* convert region offsets to region sizes and truncate
2237 size to big_values */
2238 g->region_size[2] = (576 / 2);
2239 j = 0;
2240 for(i=0;i<3;i++) {
2241 k = FFMIN(g->region_size[i], g->big_values);
2242 g->region_size[i] = k - j;
2243 j = k;
2244 }
2245
2246 /* compute band indexes */
2247 if (g->block_type == 2) {
2248 if (g->switch_point) {
2249 /* if switched mode, we handle the 36 first samples as
2250 long blocks. For 8000Hz, we handle the 48 first
2251 exponents as long blocks (XXX: check this!) */
2252 if (s->sample_rate_index <= 2)
2253 g->long_end = 8;
2254 else if (s->sample_rate_index != 8)
2255 g->long_end = 6;
2256 else
2257 g->long_end = 4; /* 8000 Hz */
2258
2259 g->short_start = 2 + (s->sample_rate_index != 8);
2260 } else {
2261 g->long_end = 0;
2262 g->short_start = 0;
2263 }
2264 } else {
2265 g->short_start = 13;
2266 g->long_end = 22;
2267 }
2268
2269 g->preflag = 0;
2270 if (!s->lsf)
2271 g->preflag = get_bits(&s->gb, 1);
2272 g->scalefac_scale = get_bits(&s->gb, 1);
2273 g->count1table_select = get_bits(&s->gb, 1);
2274 dprintf("block_type=%d switch_point=%d\n",
2275 g->block_type, g->switch_point);
2276 }
2277 }
2278
2279 if (!s->adu_mode) {
2280 /* now we get bits from the main_data_begin offset */
2281 dprintf("seekback: %d\n", main_data_begin);
2282 seek_to_maindata(s, main_data_begin);
2283 }
2284
2285 for(gr=0;gr<nb_granules;gr++) {
2286 for(ch=0;ch<s->nb_channels;ch++) {
2287 g = &granules[ch][gr];
2288
2289 bits_pos = get_bits_count(&s->gb);
2290
2291 if (!s->lsf) {
2292 uint8_t *sc;
2293 int slen, slen1, slen2;
2294
2295 /* MPEG1 scale factors */
2296 slen1 = slen_table[0][g->scalefac_compress];
2297 slen2 = slen_table[1][g->scalefac_compress];
2298 dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2299 if (g->block_type == 2) {
2300 n = g->switch_point ? 17 : 18;
2301 j = 0;
2302 if(slen1){
2303 for(i=0;i<n;i++)
2304 g->scale_factors[j++] = get_bits(&s->gb, slen1);
2305 }else{
2306 for(i=0;i<n;i++)
2307 g->scale_factors[j++] = 0;
2308 }
2309 if(slen2){
2310 for(i=0;i<18;i++)
2311 g->scale_factors[j++] = get_bits(&s->gb, slen2);
2312 for(i=0;i<3;i++)
2313 g->scale_factors[j++] = 0;
2314 }else{
2315 for(i=0;i<21;i++)
2316 g->scale_factors[j++] = 0;
2317 }
2318 } else {
2319 sc = granules[ch][0].scale_factors;
2320 j = 0;
2321 for(k=0;k<4;k++) {
2322 n = (k == 0 ? 6 : 5);
2323 if ((g->scfsi & (0x8 >> k)) == 0) {
2324 slen = (k < 2) ? slen1 : slen2;
2325 if(slen){
2326 for(i=0;i<n;i++)
2327 g->scale_factors[j++] = get_bits(&s->gb, slen);
2328 }else{
2329 for(i=0;i<n;i++)
2330 g->scale_factors[j++] = 0;
2331 }
2332 } else {
2333 /* simply copy from last granule */
2334 for(i=0;i<n;i++) {
2335 g->scale_factors[j] = sc[j];
2336 j++;
2337 }
2338 }
2339 }
2340 g->scale_factors[j++] = 0;
2341 }
2342 #if defined(DEBUG)
2343 {
2344 dprintf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2345 g->scfsi, gr, ch);
2346 for(i=0;i<j;i++)
2347 dprintf(" %d", g->scale_factors[i]);
2348 dprintf("\n");
2349 }
2350 #endif
2351 } else {
2352 int tindex, tindex2, slen[4], sl, sf;
2353
2354 /* LSF scale factors */
2355 if (g->block_type == 2) {
2356 tindex = g->switch_point ? 2 : 1;
2357 } else {
2358 tindex = 0;
2359 }
2360 sf = g->scalefac_compress;
2361 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2362 /* intensity stereo case */
2363 sf >>= 1;
2364 if (sf < 180) {
2365 lsf_sf_expand(slen, sf, 6, 6, 0);
2366 tindex2 = 3;
2367 } else if (sf < 244) {
2368 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2369 tindex2 = 4;
2370 } else {
2371 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2372 tindex2 = 5;
2373 }
2374 } else {
2375 /* normal case */
2376 if (sf < 400) {
2377 lsf_sf_expand(slen, sf, 5, 4, 4);
2378 tindex2 = 0;
2379 } else if (sf < 500) {
2380 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2381 tindex2 = 1;
2382 } else {
2383 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2384 tindex2 = 2;
2385 g->preflag = 1;
2386 }
2387 }
2388
2389 j = 0;
2390 for(k=0;k<4;k++) {
2391 n = lsf_nsf_table[tindex2][tindex][k];
2392 sl = slen[k];
2393 if(sl){
2394 for(i=0;i<n;i++)
2395 g->scale_factors[j++] = get_bits(&s->gb, sl);
2396 }else{
2397 for(i=0;i<n;i++)
2398 g->scale_factors[j++] = 0;
2399 }
2400 }
2401 /* XXX: should compute exact size */
2402 for(;j<40;j++)
2403 g->scale_factors[j] = 0;
2404 #if defined(DEBUG)
2405 {
2406 dprintf("gr=%d ch=%d scale_factors:\n",
2407 gr, ch);
2408 for(i=0;i<40;i++)
2409 dprintf(" %d", g->scale_factors[i]);
2410 dprintf("\n");
2411 }
2412 #endif
2413 }
2414
2415 exponents_from_scale_factors(s, g, exponents);
2416
2417 /* read Huffman coded residue */
2418 if (huffman_decode(s, g, exponents,
2419 bits_pos + g->part2_3_length) < 0)
2420 return -1;
2421 #if defined(DEBUG)
2422 sample_dump(0, g->sb_hybrid, 576);
2423 #endif
2424
2425 /* skip extension bits */
2426 bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2427 if (bits_left < 0) {
2428 dprintf("bits_left=%d\n", bits_left);
2429 return -1;
2430 }
2431 while (bits_left >= 16) {
2432 skip_bits(&s->gb, 16);
2433 bits_left -= 16;
2434 }
2435 if (bits_left > 0)
2436 skip_bits(&s->gb, bits_left);
2437 } /* ch */
2438
2439 if (s->nb_channels == 2)
2440 compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2441
2442 for(ch=0;ch<s->nb_channels;ch++) {
2443 g = &granules[ch][gr];
2444
2445 reorder_block(s, g);
2446 #if defined(DEBUG)
2447 sample_dump(0, g->sb_hybrid, 576);
2448 #endif
2449 s->compute_antialias(s, g);
2450 #if defined(DEBUG)
2451 sample_dump(1, g->sb_hybrid, 576);
2452 #endif
2453 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2454 #if defined(DEBUG)
2455 sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2456 #endif
2457 }
2458 } /* gr */
2459 return nb_granules * 18;
2460 }
2461
2462 static int mp_decode_frame(MPADecodeContext *s,
2463 OUT_INT *samples)
2464 {
2465 int i, nb_frames, ch;
2466 OUT_INT *samples_ptr;
2467
2468 init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
2469 (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);
2470
2471 /* skip error protection field */
2472 if (s->error_protection)
2473 get_bits(&s->gb, 16);
2474
2475 dprintf("frame %d:\n", s->frame_count);
2476 switch(s->layer) {
2477 case 1:
2478 nb_frames = mp_decode_layer1(s);
2479 break;
2480 case 2:
2481 nb_frames = mp_decode_layer2(s);
2482 break;
2483 case 3:
2484 default:
2485 nb_frames = mp_decode_layer3(s);
2486 break;
2487 }
2488 #if defined(DEBUG)
2489 for(i=0;i<nb_frames;i++) {
2490 for(ch=0;ch<s->nb_channels;ch++) {
2491 int j;
2492 dprintf("%d-%d:", i, ch);
2493 for(j=0;j<SBLIMIT;j++)
2494 dprintf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2495 dprintf("\n");
2496 }
2497 }
2498 #endif
2499 /* apply the synthesis filter */
2500 for(ch=0;ch<s->nb_channels;ch++) {
2501 samples_ptr = samples + ch;
2502 for(i=0;i<nb_frames;i++) {
2503 ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2504 window, &s->dither_state,
2505 samples_ptr, s->nb_channels,
2506 s->sb_samples[ch][i]);
2507 samples_ptr += 32 * s->nb_channels;
2508 }
2509 }
2510 #ifdef DEBUG
2511 s->frame_count++;
2512 #endif
2513 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2514 }
2515
2516 static int decode_frame(AVCodecContext * avctx,
2517 void *data, int *data_size,
2518 uint8_t * buf, int buf_size)
2519 {
2520 MPADecodeContext *s = avctx->priv_data;
2521 uint32_t header;
2522 uint8_t *buf_ptr;
2523 int len, out_size;
2524 OUT_INT *out_samples = data;
2525
2526 buf_ptr = buf;
2527 while (buf_size > 0) {
2528 len = s->inbuf_ptr - s->inbuf;
2529 if (s->frame_size == 0) {
2530 /* special case for next header for first frame in free
2531 format case (XXX: find a simpler method) */
2532 if (s->free_format_next_header != 0) {
2533 s->inbuf[0] = s->free_format_next_header >> 24;
2534 s->inbuf[1] = s->free_format_next_header >> 16;
2535 s->inbuf[2] = s->free_format_next_header >> 8;
2536 s->inbuf[3] = s->free_format_next_header;
2537 s->inbuf_ptr = s->inbuf + 4;
2538 s->free_format_next_header = 0;
2539 goto got_header;
2540 }
2541 /* no header seen : find one. We need at least HEADER_SIZE
2542 bytes to parse it */
2543 len = HEADER_SIZE - len;
2544 if (len > buf_size)
2545 len = buf_size;
2546 if (len > 0) {
2547 memcpy(s->inbuf_ptr, buf_ptr, len);
2548 buf_ptr += len;
2549 buf_size -= len;
2550 s->inbuf_ptr += len;
2551 }
2552 if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
2553 got_header:
2554 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2555 (s->inbuf[2] << 8) | s->inbuf[3];
2556
2557 if (ff_mpa_check_header(header) < 0) {
2558 /* no sync found : move by one byte (inefficient, but simple!) */
2559 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2560 s->inbuf_ptr--;
2561 dprintf("skip %x\n", header);
2562 /* reset free format frame size to give a chance
2563 to get a new bitrate */
2564 s->free_format_frame_size = 0;
2565 } else {
2566 if (decode_header(s, header) == 1) {
2567 /* free format: prepare to compute frame size */
2568 s->frame_size = -1;
2569 }
2570 /* update codec info */
2571 avctx->sample_rate = s->sample_rate;
2572 avctx->channels = s->nb_channels;
2573 avctx->bit_rate = s->bit_rate;
2574 avctx->sub_id = s->layer;
2575 switch(s->layer) {
2576 case 1:
2577 avctx->frame_size = 384;
2578 break;
2579 case 2:
2580 avctx->frame_size = 1152;
2581 break;
2582 case 3:
2583 if (s->lsf)
2584 avctx->frame_size = 576;
2585 else
2586 avctx->frame_size = 1152;
2587 break;
2588 }
2589 }
2590 }
2591 } else if (s->frame_size == -1) {
2592 /* free format : find next sync to compute frame size */
2593 len = MPA_MAX_CODED_FRAME_SIZE - len;
2594 if (len > buf_size)
2595 len = buf_size;
2596 if (len == 0) {
2597 /* frame too long: resync */
2598 s->frame_size = 0;
2599 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2600 s->inbuf_ptr--;
2601 } else {
2602 uint8_t *p, *pend;
2603 uint32_t header1;
2604 int padding;
2605
2606 memcpy(s->inbuf_ptr, buf_ptr, len);
2607 /* check for header */
2608 p = s->inbuf_ptr - 3;
2609 pend = s->inbuf_ptr + len - 4;
2610 while (p <= pend) {
2611 header = (p[0] << 24) | (p[1] << 16) |
2612 (p[2] << 8) | p[3];
2613 header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2614 (s->inbuf[2] << 8) | s->inbuf[3];
2615 /* check with high probability that we have a
2616 valid header */
2617 if ((header & SAME_HEADER_MASK) ==
2618 (header1 & SAME_HEADER_MASK)) {
2619 /* header found: update pointers */
2620 len = (p + 4) - s->inbuf_ptr;
2621 buf_ptr += len;
2622 buf_size -= len;
2623 s->inbuf_ptr = p;
2624 /* compute frame size */
2625 s->free_format_next_header = header;
2626 s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2627 padding = (header1 >> 9) & 1;
2628 if (s->layer == 1)
2629 s->free_format_frame_size -= padding * 4;
2630 else
2631 s->free_format_frame_size -= padding;
2632 dprintf("free frame size=%d padding=%d\n",
2633 s->free_format_frame_size, padding);
2634 decode_header(s, header1);
2635 goto next_data;
2636 }
2637 p++;
2638 }
2639 /* not found: simply increase pointers */
2640 buf_ptr += len;
2641 s->inbuf_ptr += len;
2642 buf_size -= len;
2643 }
2644 } else if (len < s->frame_size) {
2645 if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2646 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
2647 len = s->frame_size - len;
2648 if (len > buf_size)
2649 len = buf_size;
2650 memcpy(s->inbuf_ptr, buf_ptr, len);
2651 buf_ptr += len;
2652 s->inbuf_ptr += len;
2653 buf_size -= len;
2654 }
2655 next_data:
2656 if (s->frame_size > 0 &&
2657 (s->inbuf_ptr - s->inbuf) >= s->frame_size) {
2658 if (avctx->parse_only) {
2659 /* simply return the frame data */
2660 *(uint8_t **)data = s->inbuf;
2661 out_size = s->inbuf_ptr - s->inbuf;
2662 } else {
2663 out_size = mp_decode_frame(s, out_samples);
2664 }
2665 s->inbuf_ptr = s->inbuf;
2666 s->frame_size = 0;
2667 if(out_size>=0)
2668 *data_size = out_size;
2669 else
2670 av_log(avctx, AV_LOG_DEBUG, "Error while decoding mpeg audio frame\n"); //FIXME return -1 / but also return the number of bytes consumed
2671 break;
2672 }
2673 }
2674 return buf_ptr - buf;
2675 }
2676
2677
2678 static int decode_frame_adu(AVCodecContext * avctx,
2679 void *data, int *data_size,
2680 uint8_t * buf, int buf_size)
2681 {
2682 MPADecodeContext *s = avctx->priv_data;
2683 uint32_t header;
2684 int len, out_size;
2685 OUT_INT *out_samples = data;
2686
2687 len = buf_size;
2688
2689 // Discard too short frames
2690 if (buf_size < HEADER_SIZE) {
2691 *data_size = 0;
2692 return buf_size;
2693 }
2694
2695
2696 if (len > MPA_MAX_CODED_FRAME_SIZE)
2697 len = MPA_MAX_CODED_FRAME_SIZE;
2698
2699 memcpy(s->inbuf, buf, len);
2700 s->inbuf_ptr = s->inbuf + len;
2701
2702 // Get header and restore sync word
2703 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2704 (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;
2705
2706 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2707 *data_size = 0;
2708 return buf_size;
2709 }
2710
2711 decode_header(s, header);
2712 /* update codec info */
2713 avctx->sample_rate = s->sample_rate;
2714 avctx->channels = s->nb_channels;
2715 avctx->bit_rate = s->bit_rate;
2716 avctx->sub_id = s->layer;
2717
2718 avctx->frame_size=s->frame_size = len;
2719
2720 if (avctx->parse_only) {
2721 /* simply return the frame data */
2722 *(uint8_t **)data = s->inbuf;
2723 out_size = s->inbuf_ptr - s->inbuf;
2724 } else {
2725 out_size = mp_decode_frame(s, out_samples);
2726 }
2727
2728 *data_size = out_size;
2729 return buf_size;
2730 }
2731
2732
2733 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2734 static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2}; /* number of mp3 decoder instances */
2735 static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2736 /* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2737 static int chan_offset[9][5] = {
2738 {0},
2739 {0}, // C
2740 {0}, // FLR
2741 {2,0}, // C FLR
2742 {2,0,3}, // C FLR BS
2743 {4,0,2}, // C FLR BLRS
2744 {4,0,2,5}, // C FLR BLRS LFE
2745 {4,0,2,6,5}, // C FLR BLRS BLR LFE
2746 {0,2} // FLR BLRS
2747 };
2748
2749
2750 static int decode_init_mp3on4(AVCodecContext * avctx)
2751 {
2752 MP3On4DecodeContext *s = avctx->priv_data;
2753 int i;
2754
2755 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2756 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2757 return -1;
2758 }
2759
2760 s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2761 s->frames = mp3Frames[s->chan_cfg];
2762 if(!s->frames) {
2763 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2764 return -1;
2765 }
2766 avctx->channels = mp3Channels[s->chan_cfg];
2767
2768 /* Init the first mp3 decoder in standard way, so that all tables get builded
2769 * We replace avctx->priv_data with the context of the first decoder so that
2770 * decode_init() does not have to be changed.
2771 * Other decoders will be inited here copying data from the first context
2772 */
2773 // Allocate zeroed memory for the first decoder context
2774 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2775 // Put decoder context in place to make init_decode() happy
2776 avctx->priv_data = s->mp3decctx[0];
2777 decode_init(avctx);
2778 // Restore mp3on4 context pointer
2779 avctx->priv_data = s;
2780 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2781
2782 /* Create a separate codec/context for each frame (first is already ok).
2783 * Each frame is 1 or 2 channels - up to 5 frames allowed
2784 */
2785 for (i = 1; i < s->frames; i++) {
2786 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2787 s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2788 s->mp3decctx[i]->inbuf = &s->mp3decctx[i]->inbuf1[0][BACKSTEP_SIZE];
2789 s->mp3decctx[i]->inbuf_ptr = s->mp3decctx[i]->inbuf;
2790 s->mp3decctx[i]->adu_mode = 1;
2791 }
2792
2793 return 0;
2794 }
2795
2796
2797 static int decode_close_mp3on4(AVCodecContext * avctx)
2798 {
2799 MP3On4DecodeContext *s = avctx->priv_data;
2800 int i;
2801
2802 for (i = 0; i < s->frames; i++)
2803 if (s->mp3decctx[i])
2804 av_free(s->mp3decctx[i]);
2805
2806 return 0;
2807 }
2808
2809
2810 static int decode_frame_mp3on4(AVCodecContext * avctx,
2811 void *data, int *data_size,
2812 uint8_t * buf, int buf_size)
2813 {
2814 MP3On4DecodeContext *s = avctx->priv_data;
2815 MPADecodeContext *m;
2816 int len, out_size = 0;
2817 uint32_t header;
2818 OUT_INT *out_samples = data;
2819 OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2820 OUT_INT *outptr, *bp;
2821 int fsize;
2822 unsigned char *start2 = buf, *start;
2823 int fr, i, j, n;
2824 int off = avctx->channels;
2825 int *coff = chan_offset[s->chan_cfg];
2826
2827 len = buf_size;
2828
2829 // Discard too short frames
2830 if (buf_size < HEADER_SIZE) {
2831 *data_size = 0;
2832 return buf_size;
2833 }
2834
2835 // If only one decoder interleave is not needed
2836 outptr = s->frames == 1 ? out_samples : decoded_buf;
2837
2838 for (fr = 0; fr < s->frames; fr++) {
2839 start = start2;
2840 fsize = (start[0] << 4) | (start[1] >> 4);
2841 start2 += fsize;
2842 if (fsize > len)
2843 fsize = len;
2844 len -= fsize;
2845 if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2846 fsize = MPA_MAX_CODED_FRAME_SIZE;
2847 m = s->mp3decctx[fr];
2848 assert (m != NULL);
2849 /* copy original to new */
2850 m->inbuf_ptr = m->inbuf + fsize;
2851 memcpy(m->inbuf, start, fsize);
2852
2853 // Get header
2854 header = (m->inbuf[0] << 24) | (m->inbuf[1] << 16) |
2855 (m->inbuf[2] << 8) | m->inbuf[3] | 0xfff00000;
2856
2857 if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2858 *data_size = 0;
2859 return buf_size;
2860 }
2861
2862 decode_header(m, header);
2863 mp_decode_frame(m, decoded_buf);
2864
2865 n = MPA_FRAME_SIZE * m->nb_channels;
2866 out_size += n * sizeof(OUT_INT);
2867 if(s->frames > 1) {
2868 /* interleave output data */
2869 bp = out_samples + coff[fr];
2870 if(m->nb_channels == 1) {
2871 for(j = 0; j < n; j++) {
2872 *bp = decoded_buf[j];
2873 bp += off;
2874 }
2875 } else {
2876 for(j = 0; j < n; j++) {
2877 bp[0] = decoded_buf[j++];
2878 bp[1] = decoded_buf[j];
2879 bp += off;
2880 }
2881 }
2882 }
2883 }
2884
2885 /* update codec info */
2886 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2887 avctx->frame_size= buf_size;
2888 avctx->bit_rate = 0;
2889 for (i = 0; i < s->frames; i++)
2890 avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2891
2892 *data_size = out_size;
2893 return buf_size;
2894 }
2895
2896
2897 AVCodec mp2_decoder =
2898 {
2899 "mp2",
2900 CODEC_TYPE_AUDIO,
2901 CODEC_ID_MP2,
2902 sizeof(MPADecodeContext),
2903 decode_init,
2904 NULL,
2905 NULL,
2906 decode_frame,
2907 CODEC_CAP_PARSE_ONLY,
2908 };
2909
2910 AVCodec mp3_decoder =
2911 {
2912 "mp3",
2913 CODEC_TYPE_AUDIO,
2914 CODEC_ID_MP3,
2915 sizeof(MPADecodeContext),
2916 decode_init,
2917 NULL,
2918 NULL,
2919 decode_frame,
2920 CODEC_CAP_PARSE_ONLY,
2921 };
2922
2923 AVCodec mp3adu_decoder =
2924 {
2925 "mp3adu",
2926 CODEC_TYPE_AUDIO,
2927 CODEC_ID_MP3ADU,
2928 sizeof(MPADecodeContext),
2929 decode_init,
2930 NULL,
2931 NULL,
2932 decode_frame_adu,
2933 CODEC_CAP_PARSE_ONLY,
2934 };
2935
2936 AVCodec mp3on4_decoder =
2937 {
2938 "mp3on4",
2939 CODEC_TYPE_AUDIO,
2940 CODEC_ID_MP3ON4,
2941 sizeof(MP3On4DecodeContext),
2942 decode_init_mp3on4,
2943 NULL,
2944 decode_close_mp3on4,
2945 decode_frame_mp3on4,
2946 0
2947 };