move the +400 out of the innermost loop
[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 static uint32_t exp_table[512];
177 static uint32_t expval_table[512][16];
178 /* intensity stereo coef table */
179 static int32_t is_table[2][16];
180 static int32_t is_table_lsf[2][2][16];
181 static int32_t csa_table[8][4];
182 static float csa_table_float[8][4];
183 static int32_t mdct_win[8][36];
184
185 /* lower 2 bits: modulo 3, higher bits: shift */
186 static uint16_t scale_factor_modshift[64];
187 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
188 static int32_t scale_factor_mult[15][3];
189 /* mult table for layer 2 group quantization */
190
191 #define SCALE_GEN(v) \
192 { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
193
194 static const int32_t scale_factor_mult2[3][3] = {
195 SCALE_GEN(4.0 / 3.0), /* 3 steps */
196 SCALE_GEN(4.0 / 5.0), /* 5 steps */
197 SCALE_GEN(4.0 / 9.0), /* 9 steps */
198 };
199
200 void ff_mpa_synth_init(MPA_INT *window);
201 static MPA_INT window[512] __attribute__((aligned(16)));
202
203 /* layer 1 unscaling */
204 /* n = number of bits of the mantissa minus 1 */
205 static inline int l1_unscale(int n, int mant, int scale_factor)
206 {
207 int shift, mod;
208 int64_t val;
209
210 shift = scale_factor_modshift[scale_factor];
211 mod = shift & 3;
212 shift >>= 2;
213 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
214 shift += n;
215 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
216 return (int)((val + (1LL << (shift - 1))) >> shift);
217 }
218
219 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
220 {
221 int shift, mod, val;
222
223 shift = scale_factor_modshift[scale_factor];
224 mod = shift & 3;
225 shift >>= 2;
226
227 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
228 /* NOTE: at this point, 0 <= shift <= 21 */
229 if (shift > 0)
230 val = (val + (1 << (shift - 1))) >> shift;
231 return val;
232 }
233
234 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
235 static inline int l3_unscale(int value, int exponent)
236 {
237 unsigned int m;
238 int e;
239
240 e = table_4_3_exp [4*value + (exponent&3)];
241 m = table_4_3_value[4*value + (exponent&3)];
242 e -= (exponent >> 2);
243 assert(e>=1);
244 if (e > 31)
245 return 0;
246 m = (m + (1 << (e-1))) >> e;
247
248 return m;
249 }
250
251 /* all integer n^(4/3) computation code */
252 #define DEV_ORDER 13
253
254 #define POW_FRAC_BITS 24
255 #define POW_FRAC_ONE (1 << POW_FRAC_BITS)
256 #define POW_FIX(a) ((int)((a) * POW_FRAC_ONE))
257 #define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
258
259 static int dev_4_3_coefs[DEV_ORDER];
260
261 #if 0 /* unused */
262 static int pow_mult3[3] = {
263 POW_FIX(1.0),
264 POW_FIX(1.25992104989487316476),
265 POW_FIX(1.58740105196819947474),
266 };
267 #endif
268
269 static void int_pow_init(void)
270 {
271 int i, a;
272
273 a = POW_FIX(1.0);
274 for(i=0;i<DEV_ORDER;i++) {
275 a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
276 dev_4_3_coefs[i] = a;
277 }
278 }
279
280 #if 0 /* unused, remove? */
281 /* return the mantissa and the binary exponent */
282 static int int_pow(int i, int *exp_ptr)
283 {
284 int e, er, eq, j;
285 int a, a1;
286
287 /* renormalize */
288 a = i;
289 e = POW_FRAC_BITS;
290 while (a < (1 << (POW_FRAC_BITS - 1))) {
291 a = a << 1;
292 e--;
293 }
294 a -= (1 << POW_FRAC_BITS);
295 a1 = 0;
296 for(j = DEV_ORDER - 1; j >= 0; j--)
297 a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
298 a = (1 << POW_FRAC_BITS) + a1;
299 /* exponent compute (exact) */
300 e = e * 4;
301 er = e % 3;
302 eq = e / 3;
303 a = POW_MULL(a, pow_mult3[er]);
304 while (a >= 2 * POW_FRAC_ONE) {
305 a = a >> 1;
306 eq++;
307 }
308 /* convert to float */
309 while (a < POW_FRAC_ONE) {
310 a = a << 1;
311 eq--;
312 }
313 /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
314 #if POW_FRAC_BITS > FRAC_BITS
315 a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
316 /* correct overflow */
317 if (a >= 2 * (1 << FRAC_BITS)) {
318 a = a >> 1;
319 eq++;
320 }
321 #endif
322 *exp_ptr = eq;
323 return a;
324 }
325 #endif
326
327 static int decode_init(AVCodecContext * avctx)
328 {
329 MPADecodeContext *s = avctx->priv_data;
330 static int init=0;
331 int i, j, k;
332
333 #if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
334 avctx->sample_fmt= SAMPLE_FMT_S32;
335 #else
336 avctx->sample_fmt= SAMPLE_FMT_S16;
337 #endif
338
339 if(avctx->antialias_algo != FF_AA_FLOAT)
340 s->compute_antialias= compute_antialias_integer;
341 else
342 s->compute_antialias= compute_antialias_float;
343
344 if (!init && !avctx->parse_only) {
345 /* scale factors table for layer 1/2 */
346 for(i=0;i<64;i++) {
347 int shift, mod;
348 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
349 shift = (i / 3);
350 mod = i % 3;
351 scale_factor_modshift[i] = mod | (shift << 2);
352 }
353
354 /* scale factor multiply for layer 1 */
355 for(i=0;i<15;i++) {
356 int n, norm;
357 n = i + 2;
358 norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
359 scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
360 scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
361 scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
362 dprintf("%d: norm=%x s=%x %x %x\n",
363 i, norm,
364 scale_factor_mult[i][0],
365 scale_factor_mult[i][1],
366 scale_factor_mult[i][2]);
367 }
368
369 ff_mpa_synth_init(window);
370
371 /* huffman decode tables */
372 for(i=1;i<16;i++) {
373 const HuffTable *h = &mpa_huff_tables[i];
374 int xsize, x, y;
375 unsigned int n;
376 uint8_t tmp_bits [256];
377 uint16_t tmp_codes[256];
378
379 memset(tmp_bits , 0, sizeof(tmp_bits ));
380 memset(tmp_codes, 0, sizeof(tmp_codes));
381
382 xsize = h->xsize;
383 n = xsize * xsize;
384
385 j = 0;
386 for(x=0;x<xsize;x++) {
387 for(y=0;y<xsize;y++){
388 tmp_bits [(x << 4) | y]= h->bits [j ];
389 tmp_codes[(x << 4) | y]= h->codes[j++];
390 }
391 }
392
393 /* XXX: fail test */
394 init_vlc(&huff_vlc[i], 7, 256,
395 tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
396 }
397 for(i=0;i<2;i++) {
398 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
399 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
400 }
401
402 for(i=0;i<9;i++) {
403 k = 0;
404 for(j=0;j<22;j++) {
405 band_index_long[i][j] = k;
406 k += band_size_long[i][j];
407 }
408 band_index_long[i][22] = k;
409 }
410
411 /* compute n ^ (4/3) and store it in mantissa/exp format */
412 table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));
413 if(!table_4_3_exp)
414 return -1;
415 table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));
416 if(!table_4_3_value)
417 return -1;
418
419 int_pow_init();
420 for(i=1;i<TABLE_4_3_SIZE;i++) {
421 double f, fm;
422 int e, m;
423 f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
424 fm = frexp(f, &e);
425 m = (uint32_t)(fm*(1LL<<31) + 0.5);
426 e+= FRAC_BITS - 31 + 5 - 100;
427
428 /* normalized to FRAC_BITS */
429 table_4_3_value[i] = m;
430 // av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
431 table_4_3_exp[i] = -e;
432 }
433 for(i=0; i<512*16; i++){
434 int exponent= (i>>4);
435 double f= pow(i&15, 4.0 / 3.0) * pow(2, (exponent-400)*0.25 + FRAC_BITS + 5);
436 expval_table[exponent][i&15]= lrintf(f);
437 if((i&15)==1)
438 exp_table[exponent]= lrintf(f);
439 }
440
441 for(i=0;i<7;i++) {
442 float f;
443 int v;
444 if (i != 6) {
445 f = tan((double)i * M_PI / 12.0);
446 v = FIXR(f / (1.0 + f));
447 } else {
448 v = FIXR(1.0);
449 }
450 is_table[0][i] = v;
451 is_table[1][6 - i] = v;
452 }
453 /* invalid values */
454 for(i=7;i<16;i++)
455 is_table[0][i] = is_table[1][i] = 0.0;
456
457 for(i=0;i<16;i++) {
458 double f;
459 int e, k;
460
461 for(j=0;j<2;j++) {
462 e = -(j + 1) * ((i + 1) >> 1);
463 f = pow(2.0, e / 4.0);
464 k = i & 1;
465 is_table_lsf[j][k ^ 1][i] = FIXR(f);
466 is_table_lsf[j][k][i] = FIXR(1.0);
467 dprintf("is_table_lsf %d %d: %x %x\n",
468 i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
469 }
470 }
471
472 for(i=0;i<8;i++) {
473 float ci, cs, ca;
474 ci = ci_table[i];
475 cs = 1.0 / sqrt(1.0 + ci * ci);
476 ca = cs * ci;
477 csa_table[i][0] = FIXHR(cs/4);
478 csa_table[i][1] = FIXHR(ca/4);
479 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
480 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
481 csa_table_float[i][0] = cs;
482 csa_table_float[i][1] = ca;
483 csa_table_float[i][2] = ca + cs;
484 csa_table_float[i][3] = ca - cs;
485 // printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
486 // av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
487 }
488
489 /* compute mdct windows */
490 for(i=0;i<36;i++) {
491 for(j=0; j<4; j++){
492 double d;
493
494 if(j==2 && i%3 != 1)
495 continue;
496
497 d= sin(M_PI * (i + 0.5) / 36.0);
498 if(j==1){
499 if (i>=30) d= 0;
500 else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
501 else if(i>=18) d= 1;
502 }else if(j==3){
503 if (i< 6) d= 0;
504 else if(i< 12) d= sin(M_PI * (i - 6 + 0.5) / 12.0);
505 else if(i< 18) d= 1;
506 }
507 //merge last stage of imdct into the window coefficients
508 d*= 0.5 / cos(M_PI*(2*i + 19)/72);
509
510 if(j==2)
511 mdct_win[j][i/3] = FIXHR((d / (1<<5)));
512 else
513 mdct_win[j][i ] = FIXHR((d / (1<<5)));
514 // av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
515 }
516 }
517
518 /* NOTE: we do frequency inversion adter the MDCT by changing
519 the sign of the right window coefs */
520 for(j=0;j<4;j++) {
521 for(i=0;i<36;i+=2) {
522 mdct_win[j + 4][i] = mdct_win[j][i];
523 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
524 }
525 }
526
527 #if defined(DEBUG)
528 for(j=0;j<8;j++) {
529 av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
530 for(i=0;i<36;i++)
531 av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
532 av_log(avctx, AV_LOG_DEBUG, "\n");
533 }
534 #endif
535 init = 1;
536 }
537
538 s->inbuf_index = 0;
539 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
540 s->inbuf_ptr = s->inbuf;
541 #ifdef DEBUG
542 s->frame_count = 0;
543 #endif
544 if (avctx->codec_id == CODEC_ID_MP3ADU)
545 s->adu_mode = 1;
546 return 0;
547 }
548
549 /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
550
551 /* cos(i*pi/64) */
552
553 #define COS0_0 FIXHR(0.50060299823519630134/2)
554 #define COS0_1 FIXHR(0.50547095989754365998/2)
555 #define COS0_2 FIXHR(0.51544730992262454697/2)
556 #define COS0_3 FIXHR(0.53104259108978417447/2)
557 #define COS0_4 FIXHR(0.55310389603444452782/2)
558 #define COS0_5 FIXHR(0.58293496820613387367/2)
559 #define COS0_6 FIXHR(0.62250412303566481615/2)
560 #define COS0_7 FIXHR(0.67480834145500574602/2)
561 #define COS0_8 FIXHR(0.74453627100229844977/2)
562 #define COS0_9 FIXHR(0.83934964541552703873/2)
563 #define COS0_10 FIXHR(0.97256823786196069369/2)
564 #define COS0_11 FIXHR(1.16943993343288495515/4)
565 #define COS0_12 FIXHR(1.48416461631416627724/4)
566 #define COS0_13 FIXHR(2.05778100995341155085/8)
567 #define COS0_14 FIXHR(3.40760841846871878570/8)
568 #define COS0_15 FIXHR(10.19000812354805681150/32)
569
570 #define COS1_0 FIXHR(0.50241928618815570551/2)
571 #define COS1_1 FIXHR(0.52249861493968888062/2)
572 #define COS1_2 FIXHR(0.56694403481635770368/2)
573 #define COS1_3 FIXHR(0.64682178335999012954/2)
574 #define COS1_4 FIXHR(0.78815462345125022473/2)
575 #define COS1_5 FIXHR(1.06067768599034747134/4)
576 #define COS1_6 FIXHR(1.72244709823833392782/4)
577 #define COS1_7 FIXHR(5.10114861868916385802/16)
578
579 #define COS2_0 FIXHR(0.50979557910415916894/2)
580 #define COS2_1 FIXHR(0.60134488693504528054/2)
581 #define COS2_2 FIXHR(0.89997622313641570463/2)
582 #define COS2_3 FIXHR(2.56291544774150617881/8)
583
584 #define COS3_0 FIXHR(0.54119610014619698439/2)
585 #define COS3_1 FIXHR(1.30656296487637652785/4)
586
587 #define COS4_0 FIXHR(0.70710678118654752439/2)
588
589 /* butterfly operator */
590 #define BF(a, b, c, s)\
591 {\
592 tmp0 = tab[a] + tab[b];\
593 tmp1 = tab[a] - tab[b];\
594 tab[a] = tmp0;\
595 tab[b] = MULH(tmp1<<(s), c);\
596 }
597
598 #define BF1(a, b, c, d)\
599 {\
600 BF(a, b, COS4_0, 1);\
601 BF(c, d,-COS4_0, 1);\
602 tab[c] += tab[d];\
603 }
604
605 #define BF2(a, b, c, d)\
606 {\
607 BF(a, b, COS4_0, 1);\
608 BF(c, d,-COS4_0, 1);\
609 tab[c] += tab[d];\
610 tab[a] += tab[c];\
611 tab[c] += tab[b];\
612 tab[b] += tab[d];\
613 }
614
615 #define ADD(a, b) tab[a] += tab[b]
616
617 /* DCT32 without 1/sqrt(2) coef zero scaling. */
618 static void dct32(int32_t *out, int32_t *tab)
619 {
620 int tmp0, tmp1;
621
622 /* pass 1 */
623 BF( 0, 31, COS0_0 , 1);
624 BF(15, 16, COS0_15, 5);
625 /* pass 2 */
626 BF( 0, 15, COS1_0 , 1);
627 BF(16, 31,-COS1_0 , 1);
628 /* pass 1 */
629 BF( 7, 24, COS0_7 , 1);
630 BF( 8, 23, COS0_8 , 1);
631 /* pass 2 */
632 BF( 7, 8, COS1_7 , 4);
633 BF(23, 24,-COS1_7 , 4);
634 /* pass 3 */
635 BF( 0, 7, COS2_0 , 1);
636 BF( 8, 15,-COS2_0 , 1);
637 BF(16, 23, COS2_0 , 1);
638 BF(24, 31,-COS2_0 , 1);
639 /* pass 1 */
640 BF( 3, 28, COS0_3 , 1);
641 BF(12, 19, COS0_12, 2);
642 /* pass 2 */
643 BF( 3, 12, COS1_3 , 1);
644 BF(19, 28,-COS1_3 , 1);
645 /* pass 1 */
646 BF( 4, 27, COS0_4 , 1);
647 BF(11, 20, COS0_11, 2);
648 /* pass 2 */
649 BF( 4, 11, COS1_4 , 1);
650 BF(20, 27,-COS1_4 , 1);
651 /* pass 3 */
652 BF( 3, 4, COS2_3 , 3);
653 BF(11, 12,-COS2_3 , 3);
654 BF(19, 20, COS2_3 , 3);
655 BF(27, 28,-COS2_3 , 3);
656 /* pass 4 */
657 BF( 0, 3, COS3_0 , 1);
658 BF( 4, 7,-COS3_0 , 1);
659 BF( 8, 11, COS3_0 , 1);
660 BF(12, 15,-COS3_0 , 1);
661 BF(16, 19, COS3_0 , 1);
662 BF(20, 23,-COS3_0 , 1);
663 BF(24, 27, COS3_0 , 1);
664 BF(28, 31,-COS3_0 , 1);
665
666
667
668 /* pass 1 */
669 BF( 1, 30, COS0_1 , 1);
670 BF(14, 17, COS0_14, 3);
671 /* pass 2 */
672 BF( 1, 14, COS1_1 , 1);
673 BF(17, 30,-COS1_1 , 1);
674 /* pass 1 */
675 BF( 6, 25, COS0_6 , 1);
676 BF( 9, 22, COS0_9 , 1);
677 /* pass 2 */
678 BF( 6, 9, COS1_6 , 2);
679 BF(22, 25,-COS1_6 , 2);
680 /* pass 3 */
681 BF( 1, 6, COS2_1 , 1);
682 BF( 9, 14,-COS2_1 , 1);
683 BF(17, 22, COS2_1 , 1);
684 BF(25, 30,-COS2_1 , 1);
685
686 /* pass 1 */
687 BF( 2, 29, COS0_2 , 1);
688 BF(13, 18, COS0_13, 3);
689 /* pass 2 */
690 BF( 2, 13, COS1_2 , 1);
691 BF(18, 29,-COS1_2 , 1);
692 /* pass 1 */
693 BF( 5, 26, COS0_5 , 1);
694 BF(10, 21, COS0_10, 1);
695 /* pass 2 */
696 BF( 5, 10, COS1_5 , 2);
697 BF(21, 26,-COS1_5 , 2);
698 /* pass 3 */
699 BF( 2, 5, COS2_2 , 1);
700 BF(10, 13,-COS2_2 , 1);
701 BF(18, 21, COS2_2 , 1);
702 BF(26, 29,-COS2_2 , 1);
703 /* pass 4 */
704 BF( 1, 2, COS3_1 , 2);
705 BF( 5, 6,-COS3_1 , 2);
706 BF( 9, 10, COS3_1 , 2);
707 BF(13, 14,-COS3_1 , 2);
708 BF(17, 18, COS3_1 , 2);
709 BF(21, 22,-COS3_1 , 2);
710 BF(25, 26, COS3_1 , 2);
711 BF(29, 30,-COS3_1 , 2);
712
713 /* pass 5 */
714 BF1( 0, 1, 2, 3);
715 BF2( 4, 5, 6, 7);
716 BF1( 8, 9, 10, 11);
717 BF2(12, 13, 14, 15);
718 BF1(16, 17, 18, 19);
719 BF2(20, 21, 22, 23);
720 BF1(24, 25, 26, 27);
721 BF2(28, 29, 30, 31);
722
723 /* pass 6 */
724
725 ADD( 8, 12);
726 ADD(12, 10);
727 ADD(10, 14);
728 ADD(14, 9);
729 ADD( 9, 13);
730 ADD(13, 11);
731 ADD(11, 15);
732
733 out[ 0] = tab[0];
734 out[16] = tab[1];
735 out[ 8] = tab[2];
736 out[24] = tab[3];
737 out[ 4] = tab[4];
738 out[20] = tab[5];
739 out[12] = tab[6];
740 out[28] = tab[7];
741 out[ 2] = tab[8];
742 out[18] = tab[9];
743 out[10] = tab[10];
744 out[26] = tab[11];
745 out[ 6] = tab[12];
746 out[22] = tab[13];
747 out[14] = tab[14];
748 out[30] = tab[15];
749
750 ADD(24, 28);
751 ADD(28, 26);
752 ADD(26, 30);
753 ADD(30, 25);
754 ADD(25, 29);
755 ADD(29, 27);
756 ADD(27, 31);
757
758 out[ 1] = tab[16] + tab[24];
759 out[17] = tab[17] + tab[25];
760 out[ 9] = tab[18] + tab[26];
761 out[25] = tab[19] + tab[27];
762 out[ 5] = tab[20] + tab[28];
763 out[21] = tab[21] + tab[29];
764 out[13] = tab[22] + tab[30];
765 out[29] = tab[23] + tab[31];
766 out[ 3] = tab[24] + tab[20];
767 out[19] = tab[25] + tab[21];
768 out[11] = tab[26] + tab[22];
769 out[27] = tab[27] + tab[23];
770 out[ 7] = tab[28] + tab[18];
771 out[23] = tab[29] + tab[19];
772 out[15] = tab[30] + tab[17];
773 out[31] = tab[31];
774 }
775
776 #if FRAC_BITS <= 15
777
778 static inline int round_sample(int *sum)
779 {
780 int sum1;
781 sum1 = (*sum) >> OUT_SHIFT;
782 *sum &= (1<<OUT_SHIFT)-1;
783 if (sum1 < OUT_MIN)
784 sum1 = OUT_MIN;
785 else if (sum1 > OUT_MAX)
786 sum1 = OUT_MAX;
787 return sum1;
788 }
789
790 # if defined(ARCH_POWERPC_405)
791 /* signed 16x16 -> 32 multiply add accumulate */
792 # define MACS(rt, ra, rb) \
793 asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
794
795 /* signed 16x16 -> 32 multiply */
796 # define MULS(ra, rb) \
797 ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })
798 # else
799 /* signed 16x16 -> 32 multiply add accumulate */
800 # define MACS(rt, ra, rb) rt += (ra) * (rb)
801
802 /* signed 16x16 -> 32 multiply */
803 # define MULS(ra, rb) ((ra) * (rb))
804 # endif
805 #else
806
807 static inline int round_sample(int64_t *sum)
808 {
809 int sum1;
810 sum1 = (int)((*sum) >> OUT_SHIFT);
811 *sum &= (1<<OUT_SHIFT)-1;
812 if (sum1 < OUT_MIN)
813 sum1 = OUT_MIN;
814 else if (sum1 > OUT_MAX)
815 sum1 = OUT_MAX;
816 return sum1;
817 }
818
819 # define MULS(ra, rb) MUL64(ra, rb)
820 #endif
821
822 #define SUM8(sum, op, w, p) \
823 { \
824 sum op MULS((w)[0 * 64], p[0 * 64]);\
825 sum op MULS((w)[1 * 64], p[1 * 64]);\
826 sum op MULS((w)[2 * 64], p[2 * 64]);\
827 sum op MULS((w)[3 * 64], p[3 * 64]);\
828 sum op MULS((w)[4 * 64], p[4 * 64]);\
829 sum op MULS((w)[5 * 64], p[5 * 64]);\
830 sum op MULS((w)[6 * 64], p[6 * 64]);\
831 sum op MULS((w)[7 * 64], p[7 * 64]);\
832 }
833
834 #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
835 { \
836 int tmp;\
837 tmp = p[0 * 64];\
838 sum1 op1 MULS((w1)[0 * 64], tmp);\
839 sum2 op2 MULS((w2)[0 * 64], tmp);\
840 tmp = p[1 * 64];\
841 sum1 op1 MULS((w1)[1 * 64], tmp);\
842 sum2 op2 MULS((w2)[1 * 64], tmp);\
843 tmp = p[2 * 64];\
844 sum1 op1 MULS((w1)[2 * 64], tmp);\
845 sum2 op2 MULS((w2)[2 * 64], tmp);\
846 tmp = p[3 * 64];\
847 sum1 op1 MULS((w1)[3 * 64], tmp);\
848 sum2 op2 MULS((w2)[3 * 64], tmp);\
849 tmp = p[4 * 64];\
850 sum1 op1 MULS((w1)[4 * 64], tmp);\
851 sum2 op2 MULS((w2)[4 * 64], tmp);\
852 tmp = p[5 * 64];\
853 sum1 op1 MULS((w1)[5 * 64], tmp);\
854 sum2 op2 MULS((w2)[5 * 64], tmp);\
855 tmp = p[6 * 64];\
856 sum1 op1 MULS((w1)[6 * 64], tmp);\
857 sum2 op2 MULS((w2)[6 * 64], tmp);\
858 tmp = p[7 * 64];\
859 sum1 op1 MULS((w1)[7 * 64], tmp);\
860 sum2 op2 MULS((w2)[7 * 64], tmp);\
861 }
862
863 void ff_mpa_synth_init(MPA_INT *window)
864 {
865 int i;
866
867 /* max = 18760, max sum over all 16 coefs : 44736 */
868 for(i=0;i<257;i++) {
869 int v;
870 v = mpa_enwindow[i];
871 #if WFRAC_BITS < 16
872 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
873 #endif
874 window[i] = v;
875 if ((i & 63) != 0)
876 v = -v;
877 if (i != 0)
878 window[512 - i] = v;
879 }
880 }
881
882 /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
883 32 samples. */
884 /* XXX: optimize by avoiding ring buffer usage */
885 void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
886 MPA_INT *window, int *dither_state,
887 OUT_INT *samples, int incr,
888 int32_t sb_samples[SBLIMIT])
889 {
890 int32_t tmp[32];
891 register MPA_INT *synth_buf;
892 register const MPA_INT *w, *w2, *p;
893 int j, offset, v;
894 OUT_INT *samples2;
895 #if FRAC_BITS <= 15
896 int sum, sum2;
897 #else
898 int64_t sum, sum2;
899 #endif
900
901 dct32(tmp, sb_samples);
902
903 offset = *synth_buf_offset;
904 synth_buf = synth_buf_ptr + offset;
905
906 for(j=0;j<32;j++) {
907 v = tmp[j];
908 #if FRAC_BITS <= 15
909 /* NOTE: can cause a loss in precision if very high amplitude
910 sound */
911 if (v > 32767)
912 v = 32767;
913 else if (v < -32768)
914 v = -32768;
915 #endif
916 synth_buf[j] = v;
917 }
918 /* copy to avoid wrap */
919 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
920
921 samples2 = samples + 31 * incr;
922 w = window;
923 w2 = window + 31;
924
925 sum = *dither_state;
926 p = synth_buf + 16;
927 SUM8(sum, +=, w, p);
928 p = synth_buf + 48;
929 SUM8(sum, -=, w + 32, p);
930 *samples = round_sample(&sum);
931 samples += incr;
932 w++;
933
934 /* we calculate two samples at the same time to avoid one memory
935 access per two sample */
936 for(j=1;j<16;j++) {
937 sum2 = 0;
938 p = synth_buf + 16 + j;
939 SUM8P2(sum, +=, sum2, -=, w, w2, p);
940 p = synth_buf + 48 - j;
941 SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
942
943 *samples = round_sample(&sum);
944 samples += incr;
945 sum += sum2;
946 *samples2 = round_sample(&sum);
947 samples2 -= incr;
948 w++;
949 w2--;
950 }
951
952 p = synth_buf + 32;
953 SUM8(sum, -=, w + 32, p);
954 *samples = round_sample(&sum);
955 *dither_state= sum;
956
957 offset = (offset - 32) & 511;
958 *synth_buf_offset = offset;
959 }
960
961 #define C3 FIXHR(0.86602540378443864676/2)
962
963 /* 0.5 / cos(pi*(2*i+1)/36) */
964 static const int icos36[9] = {
965 FIXR(0.50190991877167369479),
966 FIXR(0.51763809020504152469), //0
967 FIXR(0.55168895948124587824),
968 FIXR(0.61038729438072803416),
969 FIXR(0.70710678118654752439), //1
970 FIXR(0.87172339781054900991),
971 FIXR(1.18310079157624925896),
972 FIXR(1.93185165257813657349), //2
973 FIXR(5.73685662283492756461),
974 };
975
976 /* 0.5 / cos(pi*(2*i+1)/36) */
977 static const int icos36h[9] = {
978 FIXHR(0.50190991877167369479/2),
979 FIXHR(0.51763809020504152469/2), //0
980 FIXHR(0.55168895948124587824/2),
981 FIXHR(0.61038729438072803416/2),
982 FIXHR(0.70710678118654752439/2), //1
983 FIXHR(0.87172339781054900991/2),
984 FIXHR(1.18310079157624925896/4),
985 FIXHR(1.93185165257813657349/4), //2
986 // FIXHR(5.73685662283492756461),
987 };
988
989 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
990 cases. */
991 static void imdct12(int *out, int *in)
992 {
993 int in0, in1, in2, in3, in4, in5, t1, t2;
994
995 in0= in[0*3];
996 in1= in[1*3] + in[0*3];
997 in2= in[2*3] + in[1*3];
998 in3= in[3*3] + in[2*3];
999 in4= in[4*3] + in[3*3];
1000 in5= in[5*3] + in[4*3];
1001 in5 += in3;
1002 in3 += in1;
1003
1004 in2= MULH(2*in2, C3);
1005 in3= MULH(4*in3, C3);
1006
1007 t1 = in0 - in4;
1008 t2 = MULH(2*(in1 - in5), icos36h[4]);
1009
1010 out[ 7]=
1011 out[10]= t1 + t2;
1012 out[ 1]=
1013 out[ 4]= t1 - t2;
1014
1015 in0 += in4>>1;
1016 in4 = in0 + in2;
1017 in5 += 2*in1;
1018 in1 = MULH(in5 + in3, icos36h[1]);
1019 out[ 8]=
1020 out[ 9]= in4 + in1;
1021 out[ 2]=
1022 out[ 3]= in4 - in1;
1023
1024 in0 -= in2;
1025 in5 = MULH(2*(in5 - in3), icos36h[7]);
1026 out[ 0]=
1027 out[ 5]= in0 - in5;
1028 out[ 6]=
1029 out[11]= in0 + in5;
1030 }
1031
1032 /* cos(pi*i/18) */
1033 #define C1 FIXHR(0.98480775301220805936/2)
1034 #define C2 FIXHR(0.93969262078590838405/2)
1035 #define C3 FIXHR(0.86602540378443864676/2)
1036 #define C4 FIXHR(0.76604444311897803520/2)
1037 #define C5 FIXHR(0.64278760968653932632/2)
1038 #define C6 FIXHR(0.5/2)
1039 #define C7 FIXHR(0.34202014332566873304/2)
1040 #define C8 FIXHR(0.17364817766693034885/2)
1041
1042
1043 /* using Lee like decomposition followed by hand coded 9 points DCT */
1044 static void imdct36(int *out, int *buf, int *in, int *win)
1045 {
1046 int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1047 int tmp[18], *tmp1, *in1;
1048
1049 for(i=17;i>=1;i--)
1050 in[i] += in[i-1];
1051 for(i=17;i>=3;i-=2)
1052 in[i] += in[i-2];
1053
1054 for(j=0;j<2;j++) {
1055 tmp1 = tmp + j;
1056 in1 = in + j;
1057 #if 0
1058 //more accurate but slower
1059 int64_t t0, t1, t2, t3;
1060 t2 = in1[2*4] + in1[2*8] - in1[2*2];
1061
1062 t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1063 t1 = in1[2*0] - in1[2*6];
1064 tmp1[ 6] = t1 - (t2>>1);
1065 tmp1[16] = t1 + t2;
1066
1067 t0 = MUL64(2*(in1[2*2] + in1[2*4]), C2);
1068 t1 = MUL64( in1[2*4] - in1[2*8] , -2*C8);
1069 t2 = MUL64(2*(in1[2*2] + in1[2*8]), -C4);
1070
1071 tmp1[10] = (t3 - t0 - t2) >> 32;
1072 tmp1[ 2] = (t3 + t0 + t1) >> 32;
1073 tmp1[14] = (t3 + t2 - t1) >> 32;
1074
1075 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1076 t2 = MUL64(2*(in1[2*1] + in1[2*5]), C1);
1077 t3 = MUL64( in1[2*5] - in1[2*7] , -2*C7);
1078 t0 = MUL64(2*in1[2*3], C3);
1079
1080 t1 = MUL64(2*(in1[2*1] + in1[2*7]), -C5);
1081
1082 tmp1[ 0] = (t2 + t3 + t0) >> 32;
1083 tmp1[12] = (t2 + t1 - t0) >> 32;
1084 tmp1[ 8] = (t3 - t1 - t0) >> 32;
1085 #else
1086 t2 = in1[2*4] + in1[2*8] - in1[2*2];
1087
1088 t3 = in1[2*0] + (in1[2*6]>>1);
1089 t1 = in1[2*0] - in1[2*6];
1090 tmp1[ 6] = t1 - (t2>>1);
1091 tmp1[16] = t1 + t2;
1092
1093 t0 = MULH(2*(in1[2*2] + in1[2*4]), C2);
1094 t1 = MULH( in1[2*4] - in1[2*8] , -2*C8);
1095 t2 = MULH(2*(in1[2*2] + in1[2*8]), -C4);
1096
1097 tmp1[10] = t3 - t0 - t2;
1098 tmp1[ 2] = t3 + t0 + t1;
1099 tmp1[14] = t3 + t2 - t1;
1100
1101 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1102 t2 = MULH(2*(in1[2*1] + in1[2*5]), C1);
1103 t3 = MULH( in1[2*5] - in1[2*7] , -2*C7);
1104 t0 = MULH(2*in1[2*3], C3);
1105
1106 t1 = MULH(2*(in1[2*1] + in1[2*7]), -C5);
1107
1108 tmp1[ 0] = t2 + t3 + t0;
1109 tmp1[12] = t2 + t1 - t0;
1110 tmp1[ 8] = t3 - t1 - t0;
1111 #endif
1112 }
1113
1114 i = 0;
1115 for(j=0;j<4;j++) {
1116 t0 = tmp[i];
1117 t1 = tmp[i + 2];
1118 s0 = t1 + t0;
1119 s2 = t1 - t0;
1120
1121 t2 = tmp[i + 1];
1122 t3 = tmp[i + 3];
1123 s1 = MULH(2*(t3 + t2), icos36h[j]);
1124 s3 = MULL(t3 - t2, icos36[8 - j]);
1125
1126 t0 = s0 + s1;
1127 t1 = s0 - s1;
1128 out[(9 + j)*SBLIMIT] = MULH(t1, win[9 + j]) + buf[9 + j];
1129 out[(8 - j)*SBLIMIT] = MULH(t1, win[8 - j]) + buf[8 - j];
1130 buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1131 buf[8 - j] = MULH(t0, win[18 + 8 - j]);
1132
1133 t0 = s2 + s3;
1134 t1 = s2 - s3;
1135 out[(9 + 8 - j)*SBLIMIT] = MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1136 out[( j)*SBLIMIT] = MULH(t1, win[ j]) + buf[ j];
1137 buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1138 buf[ + j] = MULH(t0, win[18 + j]);
1139 i += 4;
1140 }
1141
1142 s0 = tmp[16];
1143 s1 = MULH(2*tmp[17], icos36h[4]);
1144 t0 = s0 + s1;
1145 t1 = s0 - s1;
1146 out[(9 + 4)*SBLIMIT] = MULH(t1, win[9 + 4]) + buf[9 + 4];
1147 out[(8 - 4)*SBLIMIT] = MULH(t1, win[8 - 4]) + buf[8 - 4];
1148 buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1149 buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
1150 }
1151
1152 /* header decoding. MUST check the header before because no
1153 consistency check is done there. Return 1 if free format found and
1154 that the frame size must be computed externally */
1155 static int decode_header(MPADecodeContext *s, uint32_t header)
1156 {
1157 int sample_rate, frame_size, mpeg25, padding;
1158 int sample_rate_index, bitrate_index;
1159 if (header & (1<<20)) {
1160 s->lsf = (header & (1<<19)) ? 0 : 1;
1161 mpeg25 = 0;
1162 } else {
1163 s->lsf = 1;
1164 mpeg25 = 1;
1165 }
1166
1167 s->layer = 4 - ((header >> 17) & 3);
1168 /* extract frequency */
1169 sample_rate_index = (header >> 10) & 3;
1170 sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1171 sample_rate_index += 3 * (s->lsf + mpeg25);
1172 s->sample_rate_index = sample_rate_index;
1173 s->error_protection = ((header >> 16) & 1) ^ 1;
1174 s->sample_rate = sample_rate;
1175
1176 bitrate_index = (header >> 12) & 0xf;
1177 padding = (header >> 9) & 1;
1178 //extension = (header >> 8) & 1;
1179 s->mode = (header >> 6) & 3;
1180 s->mode_ext = (header >> 4) & 3;
1181 //copyright = (header >> 3) & 1;
1182 //original = (header >> 2) & 1;
1183 //emphasis = header & 3;
1184
1185 if (s->mode == MPA_MONO)
1186 s->nb_channels = 1;
1187 else
1188 s->nb_channels = 2;
1189
1190 if (bitrate_index != 0) {
1191 frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1192 s->bit_rate = frame_size * 1000;
1193 switch(s->layer) {
1194 case 1:
1195 frame_size = (frame_size * 12000) / sample_rate;
1196 frame_size = (frame_size + padding) * 4;
1197 break;
1198 case 2:
1199 frame_size = (frame_size * 144000) / sample_rate;
1200 frame_size += padding;
1201 break;
1202 default:
1203 case 3:
1204 frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1205 frame_size += padding;
1206 break;
1207 }
1208 s->frame_size = frame_size;
1209 } else {
1210 /* if no frame size computed, signal it */
1211 if (!s->free_format_frame_size)
1212 return 1;
1213 /* free format: compute bitrate and real frame size from the
1214 frame size we extracted by reading the bitstream */
1215 s->frame_size = s->free_format_frame_size;
1216 switch(s->layer) {
1217 case 1:
1218 s->frame_size += padding * 4;
1219 s->bit_rate = (s->frame_size * sample_rate) / 48000;
1220 break;
1221 case 2:
1222 s->frame_size += padding;
1223 s->bit_rate = (s->frame_size * sample_rate) / 144000;
1224 break;
1225 default:
1226 case 3:
1227 s->frame_size += padding;
1228 s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1229 break;
1230 }
1231 }
1232
1233 #if defined(DEBUG)
1234 dprintf("layer%d, %d Hz, %d kbits/s, ",
1235 s->layer, s->sample_rate, s->bit_rate);
1236 if (s->nb_channels == 2) {
1237 if (s->layer == 3) {
1238 if (s->mode_ext & MODE_EXT_MS_STEREO)
1239 dprintf("ms-");
1240 if (s->mode_ext & MODE_EXT_I_STEREO)
1241 dprintf("i-");
1242 }
1243 dprintf("stereo");
1244 } else {
1245 dprintf("mono");
1246 }
1247 dprintf("\n");
1248 #endif
1249 return 0;
1250 }
1251
1252 /* useful helper to get mpeg audio stream infos. Return -1 if error in
1253 header, otherwise the coded frame size in bytes */
1254 int mpa_decode_header(AVCodecContext *avctx, uint32_t head)
1255 {
1256 MPADecodeContext s1, *s = &s1;
1257 memset( s, 0, sizeof(MPADecodeContext) );
1258
1259 if (ff_mpa_check_header(head) != 0)
1260 return -1;
1261
1262 if (decode_header(s, head) != 0) {
1263 return -1;
1264 }
1265
1266 switch(s->layer) {
1267 case 1:
1268 avctx->frame_size = 384;
1269 break;
1270 case 2:
1271 avctx->frame_size = 1152;
1272 break;
1273 default:
1274 case 3:
1275 if (s->lsf)
1276 avctx->frame_size = 576;
1277 else
1278 avctx->frame_size = 1152;
1279 break;
1280 }
1281
1282 avctx->sample_rate = s->sample_rate;
1283 avctx->channels = s->nb_channels;
1284 avctx->bit_rate = s->bit_rate;
1285 avctx->sub_id = s->layer;
1286 return s->frame_size;
1287 }
1288
1289 /* return the number of decoded frames */
1290 static int mp_decode_layer1(MPADecodeContext *s)
1291 {
1292 int bound, i, v, n, ch, j, mant;
1293 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1294 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
1295
1296 if (s->mode == MPA_JSTEREO)
1297 bound = (s->mode_ext + 1) * 4;
1298 else
1299 bound = SBLIMIT;
1300
1301 /* allocation bits */
1302 for(i=0;i<bound;i++) {
1303 for(ch=0;ch<s->nb_channels;ch++) {
1304 allocation[ch][i] = get_bits(&s->gb, 4);
1305 }
1306 }
1307 for(i=bound;i<SBLIMIT;i++) {
1308 allocation[0][i] = get_bits(&s->gb, 4);
1309 }
1310
1311 /* scale factors */
1312 for(i=0;i<bound;i++) {
1313 for(ch=0;ch<s->nb_channels;ch++) {
1314 if (allocation[ch][i])
1315 scale_factors[ch][i] = get_bits(&s->gb, 6);
1316 }
1317 }
1318 for(i=bound;i<SBLIMIT;i++) {
1319 if (allocation[0][i]) {
1320 scale_factors[0][i] = get_bits(&s->gb, 6);
1321 scale_factors[1][i] = get_bits(&s->gb, 6);
1322 }
1323 }
1324
1325 /* compute samples */
1326 for(j=0;j<12;j++) {
1327 for(i=0;i<bound;i++) {
1328 for(ch=0;ch<s->nb_channels;ch++) {
1329 n = allocation[ch][i];
1330 if (n) {
1331 mant = get_bits(&s->gb, n + 1);
1332 v = l1_unscale(n, mant, scale_factors[ch][i]);
1333 } else {
1334 v = 0;
1335 }
1336 s->sb_samples[ch][j][i] = v;
1337 }
1338 }
1339 for(i=bound;i<SBLIMIT;i++) {
1340 n = allocation[0][i];
1341 if (n) {
1342 mant = get_bits(&s->gb, n + 1);
1343 v = l1_unscale(n, mant, scale_factors[0][i]);
1344 s->sb_samples[0][j][i] = v;
1345 v = l1_unscale(n, mant, scale_factors[1][i]);
1346 s->sb_samples[1][j][i] = v;
1347 } else {
1348 s->sb_samples[0][j][i] = 0;
1349 s->sb_samples[1][j][i] = 0;
1350 }
1351 }
1352 }
1353 return 12;
1354 }
1355
1356 /* bitrate is in kb/s */
1357 int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1358 {
1359 int ch_bitrate, table;
1360
1361 ch_bitrate = bitrate / nb_channels;
1362 if (!lsf) {
1363 if ((freq == 48000 && ch_bitrate >= 56) ||
1364 (ch_bitrate >= 56 && ch_bitrate <= 80))
1365 table = 0;
1366 else if (freq != 48000 && ch_bitrate >= 96)
1367 table = 1;
1368 else if (freq != 32000 && ch_bitrate <= 48)
1369 table = 2;
1370 else
1371 table = 3;
1372 } else {
1373 table = 4;
1374 }
1375 return table;
1376 }
1377
1378 static int mp_decode_layer2(MPADecodeContext *s)
1379 {
1380 int sblimit; /* number of used subbands */
1381 const unsigned char *alloc_table;
1382 int table, bit_alloc_bits, i, j, ch, bound, v;
1383 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1384 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1385 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1386 int scale, qindex, bits, steps, k, l, m, b;
1387
1388 /* select decoding table */
1389 table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1390 s->sample_rate, s->lsf);
1391 sblimit = sblimit_table[table];
1392 alloc_table = alloc_tables[table];
1393
1394 if (s->mode == MPA_JSTEREO)
1395 bound = (s->mode_ext + 1) * 4;
1396 else
1397 bound = sblimit;
1398
1399 dprintf("bound=%d sblimit=%d\n", bound, sblimit);
1400
1401 /* sanity check */
1402 if( bound > sblimit ) bound = sblimit;
1403
1404 /* parse bit allocation */
1405 j = 0;
1406 for(i=0;i<bound;i++) {
1407 bit_alloc_bits = alloc_table[j];
1408 for(ch=0;ch<s->nb_channels;ch++) {
1409 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1410 }
1411 j += 1 << bit_alloc_bits;
1412 }
1413 for(i=bound;i<sblimit;i++) {
1414 bit_alloc_bits = alloc_table[j];
1415 v = get_bits(&s->gb, bit_alloc_bits);
1416 bit_alloc[0][i] = v;
1417 bit_alloc[1][i] = v;
1418 j += 1 << bit_alloc_bits;
1419 }
1420
1421 #ifdef DEBUG
1422 {
1423 for(ch=0;ch<s->nb_channels;ch++) {
1424 for(i=0;i<sblimit;i++)
1425 dprintf(" %d", bit_alloc[ch][i]);
1426 dprintf("\n");
1427 }
1428 }
1429 #endif
1430
1431 /* scale codes */
1432 for(i=0;i<sblimit;i++) {
1433 for(ch=0;ch<s->nb_channels;ch++) {
1434 if (bit_alloc[ch][i])
1435 scale_code[ch][i] = get_bits(&s->gb, 2);
1436 }
1437 }
1438
1439 /* scale factors */
1440 for(i=0;i<sblimit;i++) {
1441 for(ch=0;ch<s->nb_channels;ch++) {
1442 if (bit_alloc[ch][i]) {
1443 sf = scale_factors[ch][i];
1444 switch(scale_code[ch][i]) {
1445 default:
1446 case 0:
1447 sf[0] = get_bits(&s->gb, 6);
1448 sf[1] = get_bits(&s->gb, 6);
1449 sf[2] = get_bits(&s->gb, 6);
1450 break;
1451 case 2:
1452 sf[0] = get_bits(&s->gb, 6);
1453 sf[1] = sf[0];
1454 sf[2] = sf[0];
1455 break;
1456 case 1:
1457 sf[0] = get_bits(&s->gb, 6);
1458 sf[2] = get_bits(&s->gb, 6);
1459 sf[1] = sf[0];
1460 break;
1461 case 3:
1462 sf[0] = get_bits(&s->gb, 6);
1463 sf[2] = get_bits(&s->gb, 6);
1464 sf[1] = sf[2];
1465 break;
1466 }
1467 }
1468 }
1469 }
1470
1471 #ifdef DEBUG
1472 for(ch=0;ch<s->nb_channels;ch++) {
1473 for(i=0;i<sblimit;i++) {
1474 if (bit_alloc[ch][i]) {
1475 sf = scale_factors[ch][i];
1476 dprintf(" %d %d %d", sf[0], sf[1], sf[2]);
1477 } else {
1478 dprintf(" -");
1479 }
1480 }
1481 dprintf("\n");
1482 }
1483 #endif
1484
1485 /* samples */
1486 for(k=0;k<3;k++) {
1487 for(l=0;l<12;l+=3) {
1488 j = 0;
1489 for(i=0;i<bound;i++) {
1490 bit_alloc_bits = alloc_table[j];
1491 for(ch=0;ch<s->nb_channels;ch++) {
1492 b = bit_alloc[ch][i];
1493 if (b) {
1494 scale = scale_factors[ch][i][k];
1495 qindex = alloc_table[j+b];
1496 bits = quant_bits[qindex];
1497 if (bits < 0) {
1498 /* 3 values at the same time */
1499 v = get_bits(&s->gb, -bits);
1500 steps = quant_steps[qindex];
1501 s->sb_samples[ch][k * 12 + l + 0][i] =
1502 l2_unscale_group(steps, v % steps, scale);
1503 v = v / steps;
1504 s->sb_samples[ch][k * 12 + l + 1][i] =
1505 l2_unscale_group(steps, v % steps, scale);
1506 v = v / steps;
1507 s->sb_samples[ch][k * 12 + l + 2][i] =
1508 l2_unscale_group(steps, v, scale);
1509 } else {
1510 for(m=0;m<3;m++) {
1511 v = get_bits(&s->gb, bits);
1512 v = l1_unscale(bits - 1, v, scale);
1513 s->sb_samples[ch][k * 12 + l + m][i] = v;
1514 }
1515 }
1516 } else {
1517 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1518 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1519 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1520 }
1521 }
1522 /* next subband in alloc table */
1523 j += 1 << bit_alloc_bits;
1524 }
1525 /* XXX: find a way to avoid this duplication of code */
1526 for(i=bound;i<sblimit;i++) {
1527 bit_alloc_bits = alloc_table[j];
1528 b = bit_alloc[0][i];
1529 if (b) {
1530 int mant, scale0, scale1;
1531 scale0 = scale_factors[0][i][k];
1532 scale1 = scale_factors[1][i][k];
1533 qindex = alloc_table[j+b];
1534 bits = quant_bits[qindex];
1535 if (bits < 0) {
1536 /* 3 values at the same time */
1537 v = get_bits(&s->gb, -bits);
1538 steps = quant_steps[qindex];
1539 mant = v % steps;
1540 v = v / steps;
1541 s->sb_samples[0][k * 12 + l + 0][i] =
1542 l2_unscale_group(steps, mant, scale0);
1543 s->sb_samples[1][k * 12 + l + 0][i] =
1544 l2_unscale_group(steps, mant, scale1);
1545 mant = v % steps;
1546 v = v / steps;
1547 s->sb_samples[0][k * 12 + l + 1][i] =
1548 l2_unscale_group(steps, mant, scale0);
1549 s->sb_samples[1][k * 12 + l + 1][i] =
1550 l2_unscale_group(steps, mant, scale1);
1551 s->sb_samples[0][k * 12 + l + 2][i] =
1552 l2_unscale_group(steps, v, scale0);
1553 s->sb_samples[1][k * 12 + l + 2][i] =
1554 l2_unscale_group(steps, v, scale1);
1555 } else {
1556 for(m=0;m<3;m++) {
1557 mant = get_bits(&s->gb, bits);
1558 s->sb_samples[0][k * 12 + l + m][i] =
1559 l1_unscale(bits - 1, mant, scale0);
1560 s->sb_samples[1][k * 12 + l + m][i] =
1561 l1_unscale(bits - 1, mant, scale1);
1562 }
1563 }
1564 } else {
1565 s->sb_samples[0][k * 12 + l + 0][i] = 0;
1566 s->sb_samples[0][k * 12 + l + 1][i] = 0;
1567 s->sb_samples[0][k * 12 + l + 2][i] = 0;
1568 s->sb_samples[1][k * 12 + l + 0][i] = 0;
1569 s->sb_samples[1][k * 12 + l + 1][i] = 0;
1570 s->sb_samples[1][k * 12 + l + 2][i] = 0;
1571 }
1572 /* next subband in alloc table */
1573 j += 1 << bit_alloc_bits;
1574 }
1575 /* fill remaining samples to zero */
1576 for(i=sblimit;i<SBLIMIT;i++) {
1577 for(ch=0;ch<s->nb_channels;ch++) {
1578 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1579 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1580 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1581 }
1582 }
1583 }
1584 }
1585 return 3 * 12;
1586 }
1587
1588 /*
1589 * Seek back in the stream for backstep bytes (at most 511 bytes)
1590 */
1591 static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)
1592 {
1593 uint8_t *ptr;
1594
1595 /* compute current position in stream */
1596 ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));
1597
1598 /* copy old data before current one */
1599 ptr -= backstep;
1600 memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
1601 BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
1602 /* init get bits again */
1603 init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);
1604
1605 /* prepare next buffer */
1606 s->inbuf_index ^= 1;
1607 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1608 s->old_frame_size = s->frame_size;
1609 }
1610
1611 static inline void lsf_sf_expand(int *slen,
1612 int sf, int n1, int n2, int n3)
1613 {
1614 if (n3) {
1615 slen[3] = sf % n3;
1616 sf /= n3;
1617 } else {
1618 slen[3] = 0;
1619 }
1620 if (n2) {
1621 slen[2] = sf % n2;
1622 sf /= n2;
1623 } else {
1624 slen[2] = 0;
1625 }
1626 slen[1] = sf % n1;
1627 sf /= n1;
1628 slen[0] = sf;
1629 }
1630
1631 static void exponents_from_scale_factors(MPADecodeContext *s,
1632 GranuleDef *g,
1633 int16_t *exponents)
1634 {
1635 const uint8_t *bstab, *pretab;
1636 int len, i, j, k, l, v0, shift, gain, gains[3];
1637 int16_t *exp_ptr;
1638
1639 exp_ptr = exponents;
1640 gain = g->global_gain - 210;
1641 shift = g->scalefac_scale + 1;
1642
1643 bstab = band_size_long[s->sample_rate_index];
1644 pretab = mpa_pretab[g->preflag];
1645 for(i=0;i<g->long_end;i++) {
1646 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
1647 len = bstab[i];
1648 for(j=len;j>0;j--)
1649 *exp_ptr++ = v0;
1650 }
1651
1652 if (g->short_start < 13) {
1653 bstab = band_size_short[s->sample_rate_index];
1654 gains[0] = gain - (g->subblock_gain[0] << 3);
1655 gains[1] = gain - (g->subblock_gain[1] << 3);
1656 gains[2] = gain - (g->subblock_gain[2] << 3);
1657 k = g->long_end;
1658 for(i=g->short_start;i<13;i++) {
1659 len = bstab[i];
1660 for(l=0;l<3;l++) {
1661 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
1662 for(j=len;j>0;j--)
1663 *exp_ptr++ = v0;
1664 }
1665 }
1666 }
1667 }
1668
1669 /* handle n = 0 too */
1670 static inline int get_bitsz(GetBitContext *s, int n)
1671 {
1672 if (n == 0)
1673 return 0;
1674 else
1675 return get_bits(s, n);
1676 }
1677
1678 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
1679 int16_t *exponents, int end_pos)
1680 {
1681 int s_index;
1682 int linbits, code, x, y, l, v, i, j, k, pos;
1683 int last_pos;
1684 VLC *vlc;
1685
1686 /* low frequencies (called big values) */
1687 s_index = 0;
1688 for(i=0;i<3;i++) {
1689 j = g->region_size[i];
1690 if (j == 0)
1691 continue;
1692 /* select vlc table */
1693 k = g->table_select[i];
1694 l = mpa_huff_data[k][0];
1695 linbits = mpa_huff_data[k][1];
1696 vlc = &huff_vlc[l];
1697
1698 if(!l){
1699 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*j);
1700 s_index += 2*j;
1701 continue;
1702 }
1703
1704 /* read huffcode and compute each couple */
1705 for(;j>0;j--) {
1706 int exponent;
1707
1708 if (get_bits_count(&s->gb) >= end_pos)
1709 break;
1710 y = get_vlc2(&s->gb, vlc->table, 7, 3);
1711
1712 if(!y){
1713 g->sb_hybrid[s_index ] =
1714 g->sb_hybrid[s_index+1] = 0;
1715 s_index += 2;
1716 continue;
1717 }
1718
1719 x = y >> 4;
1720 y = y & 0x0f;
1721 exponent= exponents[s_index];
1722
1723 dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
1724 i, g->region_size[i] - j, x, y, exponent);
1725 if (x) {
1726 #if 0
1727 if (x == 15)
1728 x += get_bitsz(&s->gb, linbits);
1729 v = l3_unscale(x, exponent);
1730 #else
1731 if (x < 15){
1732 v = expval_table[ exponent ][ x ];
1733 // v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);
1734 }else{
1735 x += get_bitsz(&s->gb, linbits);
1736 v = l3_unscale(x, exponent);
1737 }
1738 #endif
1739 if (get_bits1(&s->gb))
1740 v = -v;
1741 } else {
1742 v = 0;
1743 }
1744 g->sb_hybrid[s_index++] = v;
1745 if (y) {
1746 #if 0
1747 if (y == 15)
1748 y += get_bitsz(&s->gb, linbits);
1749 v = l3_unscale(y, exponent);
1750 #else
1751 if (y < 15){
1752 v = expval_table[ exponent ][ y ];
1753 }else{
1754 y += get_bitsz(&s->gb, linbits);
1755 v = l3_unscale(y, exponent);
1756 }
1757 #endif
1758 if (get_bits1(&s->gb))
1759 v = -v;
1760 } else {
1761 v = 0;
1762 }
1763 g->sb_hybrid[s_index++] = v;
1764 }
1765 }
1766
1767 /* high frequencies */
1768 vlc = &huff_quad_vlc[g->count1table_select];
1769 last_pos=0;
1770 while (s_index <= 572) {
1771 pos = get_bits_count(&s->gb);
1772 if (pos >= end_pos) {
1773 if (pos > end_pos && last_pos){
1774 /* some encoders generate an incorrect size for this
1775 part. We must go back into the data */
1776 s_index -= 4;
1777 init_get_bits(&s->gb, s->gb.buffer + 4*(last_pos>>5), s->gb.size_in_bits - (last_pos&(~31)));
1778 skip_bits(&s->gb, last_pos&31);
1779 }
1780 break;
1781 }
1782 last_pos= pos;
1783
1784 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
1785 dprintf("t=%d code=%d\n", g->count1table_select, code);
1786 g->sb_hybrid[s_index+0]=
1787 g->sb_hybrid[s_index+1]=
1788 g->sb_hybrid[s_index+2]=
1789 g->sb_hybrid[s_index+3]= 0;
1790 while(code){
1791 const static int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1792 int pos= s_index+idxtab[code];
1793 code ^= 8>>idxtab[code];
1794 v = exp_table[ exponents[pos] ];
1795 // v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);
1796 if(get_bits1(&s->gb))
1797 v = -v;
1798 g->sb_hybrid[pos] = v;
1799 }
1800 s_index+=4;
1801 }
1802 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
1803 return 0;
1804 }
1805
1806 /* Reorder short blocks from bitstream order to interleaved order. It
1807 would be faster to do it in parsing, but the code would be far more
1808 complicated */
1809 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1810 {
1811 int i, j, len;
1812 int32_t *ptr, *dst, *ptr1;
1813 int32_t tmp[576];
1814
1815 if (g->block_type != 2)
1816 return;
1817
1818 if (g->switch_point) {
1819 if (s->sample_rate_index != 8) {
1820 ptr = g->sb_hybrid + 36;
1821 } else {
1822 ptr = g->sb_hybrid + 48;
1823 }
1824 } else {
1825 ptr = g->sb_hybrid;
1826 }
1827
1828 for(i=g->short_start;i<13;i++) {
1829 len = band_size_short[s->sample_rate_index][i];
1830 ptr1 = ptr;
1831 dst = tmp;
1832 for(j=len;j>0;j--) {
1833 *dst++ = ptr[0*len];
1834 *dst++ = ptr[1*len];
1835 *dst++ = ptr[2*len];
1836 ptr++;
1837 }
1838 ptr+=2*len;
1839 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1840 }
1841 }
1842
1843 #define ISQRT2 FIXR(0.70710678118654752440)
1844
1845 static void compute_stereo(MPADecodeContext *s,
1846 GranuleDef *g0, GranuleDef *g1)
1847 {
1848 int i, j, k, l;
1849 int32_t v1, v2;
1850 int sf_max, tmp0, tmp1, sf, len, non_zero_found;
1851 int32_t (*is_tab)[16];
1852 int32_t *tab0, *tab1;
1853 int non_zero_found_short[3];
1854
1855 /* intensity stereo */
1856 if (s->mode_ext & MODE_EXT_I_STEREO) {
1857 if (!s->lsf) {
1858 is_tab = is_table;
1859 sf_max = 7;
1860 } else {
1861 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1862 sf_max = 16;
1863 }
1864
1865 tab0 = g0->sb_hybrid + 576;
1866 tab1 = g1->sb_hybrid + 576;
1867
1868 non_zero_found_short[0] = 0;
1869 non_zero_found_short[1] = 0;
1870 non_zero_found_short[2] = 0;
1871 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1872 for(i = 12;i >= g1->short_start;i--) {
1873 /* for last band, use previous scale factor */
1874 if (i != 11)
1875 k -= 3;
1876 len = band_size_short[s->sample_rate_index][i];
1877 for(l=2;l>=0;l--) {
1878 tab0 -= len;
1879 tab1 -= len;
1880 if (!non_zero_found_short[l]) {
1881 /* test if non zero band. if so, stop doing i-stereo */
1882 for(j=0;j<len;j++) {
1883 if (tab1[j] != 0) {
1884 non_zero_found_short[l] = 1;
1885 goto found1;
1886 }
1887 }
1888 sf = g1->scale_factors[k + l];
1889 if (sf >= sf_max)
1890 goto found1;
1891
1892 v1 = is_tab[0][sf];
1893 v2 = is_tab[1][sf];
1894 for(j=0;j<len;j++) {
1895 tmp0 = tab0[j];
1896 tab0[j] = MULL(tmp0, v1);
1897 tab1[j] = MULL(tmp0, v2);
1898 }
1899 } else {
1900 found1:
1901 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1902 /* lower part of the spectrum : do ms stereo
1903 if enabled */
1904 for(j=0;j<len;j++) {
1905 tmp0 = tab0[j];
1906 tmp1 = tab1[j];
1907 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1908 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1909 }
1910 }
1911 }
1912 }
1913 }
1914
1915 non_zero_found = non_zero_found_short[0] |
1916 non_zero_found_short[1] |
1917 non_zero_found_short[2];
1918
1919 for(i = g1->long_end - 1;i >= 0;i--) {
1920 len = band_size_long[s->sample_rate_index][i];
1921 tab0 -= len;
1922 tab1 -= len;
1923 /* test if non zero band. if so, stop doing i-stereo */
1924 if (!non_zero_found) {
1925 for(j=0;j<len;j++) {
1926 if (tab1[j] != 0) {
1927 non_zero_found = 1;
1928 goto found2;
1929 }
1930 }
1931 /* for last band, use previous scale factor */
1932 k = (i == 21) ? 20 : i;
1933 sf = g1->scale_factors[k];
1934 if (sf >= sf_max)
1935 goto found2;
1936 v1 = is_tab[0][sf];
1937 v2 = is_tab[1][sf];
1938 for(j=0;j<len;j++) {
1939 tmp0 = tab0[j];
1940 tab0[j] = MULL(tmp0, v1);
1941 tab1[j] = MULL(tmp0, v2);
1942 }
1943 } else {
1944 found2:
1945 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1946 /* lower part of the spectrum : do ms stereo
1947 if enabled */
1948 for(j=0;j<len;j++) {
1949 tmp0 = tab0[j];
1950 tmp1 = tab1[j];
1951 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1952 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1953 }
1954 }
1955 }
1956 }
1957 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1958 /* ms stereo ONLY */
1959 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1960 global gain */
1961 tab0 = g0->sb_hybrid;
1962 tab1 = g1->sb_hybrid;
1963 for(i=0;i<576;i++) {
1964 tmp0 = tab0[i];
1965 tmp1 = tab1[i];
1966 tab0[i] = tmp0 + tmp1;
1967 tab1[i] = tmp0 - tmp1;
1968 }
1969 }
1970 }
1971
1972 static void compute_antialias_integer(MPADecodeContext *s,
1973 GranuleDef *g)
1974 {
1975 int32_t *ptr, *csa;
1976 int n, i;
1977
1978 /* we antialias only "long" bands */
1979 if (g->block_type == 2) {
1980 if (!g->switch_point)
1981 return;
1982 /* XXX: check this for 8000Hz case */
1983 n = 1;
1984 } else {
1985 n = SBLIMIT - 1;
1986 }
1987
1988 ptr = g->sb_hybrid + 18;
1989 for(i = n;i > 0;i--) {
1990 int tmp0, tmp1, tmp2;
1991 csa = &csa_table[0][0];
1992 #define INT_AA(j) \
1993 tmp0 = ptr[-1-j];\
1994 tmp1 = ptr[ j];\
1995 tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
1996 ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1997 ptr[ j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
1998
1999 INT_AA(0)
2000 INT_AA(1)
2001 INT_AA(2)
2002 INT_AA(3)
2003 INT_AA(4)
2004 INT_AA(5)
2005 INT_AA(6)
2006 INT_AA(7)
2007
2008 ptr += 18;
2009 }
2010 }
2011
2012 static void compute_antialias_float(MPADecodeContext *s,
2013 GranuleDef *g)
2014 {
2015 int32_t *ptr;
2016 int n, i;
2017
2018 /* we antialias only "long" bands */
2019 if (g->block_type == 2) {
2020 if (!g->switch_point)
2021 return;
2022 /* XXX: check this for 8000Hz case */
2023 n = 1;
2024 } else {
2025 n = SBLIMIT - 1;
2026 }
2027
2028 ptr = g->sb_hybrid + 18;
2029 for(i = n;i > 0;i--) {
2030 float tmp0, tmp1;
2031 float *csa = &csa_table_float[0][0];
2032 #define FLOAT_AA(j)\
2033 tmp0= ptr[-1-j];\
2034 tmp1= ptr[ j];\
2035 ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
2036 ptr[ j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
2037
2038 FLOAT_AA(0)
2039 FLOAT_AA(1)
2040 FLOAT_AA(2)
2041 FLOAT_AA(3)
2042 FLOAT_AA(4)
2043 FLOAT_AA(5)
2044 FLOAT_AA(6)
2045 FLOAT_AA(7)
2046
2047 ptr += 18;
2048 }
2049 }
2050
2051 static void compute_imdct(MPADecodeContext *s,
2052 GranuleDef *g,
2053 int32_t *sb_samples,
2054 int32_t *mdct_buf)
2055 {
2056 int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
2057 int32_t out2[12];
2058 int i, j, mdct_long_end, v, sblimit;
2059
2060 /* find last non zero block */
2061 ptr = g->sb_hybrid + 576;
2062 ptr1 = g->sb_hybrid + 2 * 18;
2063 while (ptr >= ptr1) {
2064 ptr -= 6;
2065 v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
2066 if (v != 0)
2067 break;
2068 }
2069 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
2070
2071 if (g->block_type == 2) {
2072 /* XXX: check for 8000 Hz */
2073 if (g->switch_point)
2074 mdct_long_end = 2;
2075 else
2076 mdct_long_end = 0;
2077 } else {
2078 mdct_long_end = sblimit;
2079 }
2080
2081 buf = mdct_buf;
2082 ptr = g->sb_hybrid;
2083 for(j=0;j<mdct_long_end;j++) {
2084 /* apply window & overlap with previous buffer */
2085 out_ptr = sb_samples + j;
2086 /* select window */
2087 if (g->switch_point && j < 2)
2088 win1 = mdct_win[0];
2089 else
2090 win1 = mdct_win[g->block_type];
2091 /* select frequency inversion */
2092 win = win1 + ((4 * 36) & -(j & 1));
2093 imdct36(out_ptr, buf, ptr, win);
2094 out_ptr += 18*SBLIMIT;
2095 ptr += 18;
2096 buf += 18;
2097 }
2098 for(j=mdct_long_end;j<sblimit;j++) {
2099 /* select frequency inversion */
2100 win = mdct_win[2] + ((4 * 36) & -(j & 1));
2101 out_ptr = sb_samples + j;
2102
2103 for(i=0; i<6; i++){
2104 *out_ptr = buf[i];
2105 out_ptr += SBLIMIT;
2106 }
2107 imdct12(out2, ptr + 0);
2108 for(i=0;i<6;i++) {
2109 *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
2110 buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
2111 out_ptr += SBLIMIT;
2112 }
2113 imdct12(out2, ptr + 1);
2114 for(i=0;i<6;i++) {
2115 *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
2116 buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
2117 out_ptr += SBLIMIT;
2118 }
2119 imdct12(out2, ptr + 2);
2120 for(i=0;i<6;i++) {
2121 buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
2122 buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
2123 buf[i + 6*2] = 0;
2124 }
2125 ptr += 18;
2126 buf += 18;
2127 }
2128 /* zero bands */
2129 for(j=sblimit;j<SBLIMIT;j++) {
2130 /* overlap */
2131 out_ptr = sb_samples + j;
2132 for(i=0;i<18;i++) {
2133 *out_ptr = buf[i];
2134 buf[i] = 0;
2135 out_ptr += SBLIMIT;
2136 }
2137 buf += 18;
2138 }
2139 }
2140
2141 #if defined(DEBUG)
2142 void sample_dump(int fnum, int32_t *tab, int n)
2143 {
2144 static FILE *files[16], *f;
2145 char buf[512];
2146 int i;
2147 int32_t v;
2148
2149 f = files[fnum];
2150 if (!f) {
2151 snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
2152 fnum,
2153 #ifdef USE_HIGHPRECISION
2154 "hp"
2155 #else
2156 "lp"
2157 #endif
2158 );
2159 f = fopen(buf, "w");
2160 if (!f)
2161 return;
2162 files[fnum] = f;
2163 }
2164
2165 if (fnum == 0) {
2166 static int pos = 0;
2167 av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
2168 for(i=0;i<n;i++) {
2169 av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
2170 if ((i % 18) == 17)
2171 av_log(NULL, AV_LOG_DEBUG, "\n");
2172 }
2173 pos += n;
2174 }
2175 for(i=0;i<n;i++) {
2176 /* normalize to 23 frac bits */
2177 v = tab[i] << (23 - FRAC_BITS);
2178 fwrite(&v, 1, sizeof(int32_t), f);
2179 }
2180 }
2181 #endif
2182
2183
2184 /* main layer3 decoding function */
2185 static int mp_decode_layer3(MPADecodeContext *s)
2186 {
2187 int nb_granules, main_data_begin, private_bits;
2188 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
2189 GranuleDef granules[2][2], *g;
2190 int16_t exponents[576];
2191
2192 /* read side info */
2193 if (s->lsf) {
2194 main_data_begin = get_bits(&s->gb, 8);
2195 private_bits = get_bits(&s->gb, s->nb_channels);
2196 nb_granules = 1;
2197 } else {
2198 main_data_begin = get_bits(&s->gb, 9);
2199 if (s->nb_channels == 2)
2200 private_bits = get_bits(&s->gb, 3);
2201 else
2202 private_bits = get_bits(&s->gb, 5);
2203 nb_granules = 2;
2204 for(ch=0;ch<s->nb_channels;ch++) {
2205 granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2206 granules[ch][1].scfsi = get_bits(&s->gb, 4);
2207 }
2208 }
2209
2210 for(gr=0;gr<nb_granules;gr++) {
2211 for(ch=0;ch<s->nb_channels;ch++) {
2212 dprintf("gr=%d ch=%d: side_info\n", gr, ch);
2213 g = &granules[ch][gr];
2214 g->part2_3_length = get_bits(&s->gb, 12);
2215 g->big_values = get_bits(&s->gb, 9);
2216 g->global_gain = get_bits(&s->gb, 8);
2217 /* if MS stereo only is selected, we precompute the
2218 1/sqrt(2) renormalization factor */
2219 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2220 MODE_EXT_MS_STEREO)
2221 g->global_gain -= 2;
2222 if (s->lsf)
2223 g->scalefac_compress = get_bits(&s->gb, 9);
2224 else
2225 g->scalefac_compress = get_bits(&s->gb, 4);
2226 blocksplit_flag = get_bits(&s->gb, 1);
2227 if (blocksplit_flag) {
2228 g->block_type = get_bits(&s->gb, 2);
2229 if (g->block_type == 0)
2230 return -1;
2231 g->switch_point = get_bits(&s->gb, 1);
2232 for(i=0;i<2;i++)
2233 g->table_select[i] = get_bits(&s->gb, 5);
2234 for(i=0;i<3;i++)
2235 g->subblock_gain[i] = get_bits(&s->gb, 3);
2236 /* compute huffman coded region sizes */
2237 if (g->block_type == 2)
2238 g->region_size[0] = (36 / 2);
2239 else {
2240 if (s->sample_rate_index <= 2)
2241 g->region_size[0] = (36 / 2);
2242 else if (s->sample_rate_index != 8)
2243 g->region_size[0] = (54 / 2);
2244 else
2245 g->region_size[0] = (108 / 2);
2246 }
2247 g->region_size[1] = (576 / 2);
2248 } else {
2249 int region_address1, region_address2, l;
2250 g->block_type = 0;
2251 g->switch_point = 0;
2252 for(i=0;i<3;i++)
2253 g->table_select[i] = get_bits(&s->gb, 5);
2254 /* compute huffman coded region sizes */
2255 region_address1 = get_bits(&s->gb, 4);
2256 region_address2 = get_bits(&s->gb, 3);
2257 dprintf("region1=%d region2=%d\n",
2258 region_address1, region_address2);
2259 g->region_size[0] =
2260 band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2261 l = region_address1 + region_address2 + 2;
2262 /* should not overflow */
2263 if (l > 22)
2264 l = 22;
2265 g->region_size[1] =
2266 band_index_long[s->sample_rate_index][l] >> 1;
2267 }
2268 /* convert region offsets to region sizes and truncate
2269 size to big_values */
2270 g->region_size[2] = (576 / 2);
2271 j = 0;
2272 for(i=0;i<3;i++) {
2273 k = FFMIN(g->region_size[i], g->big_values);
2274 g->region_size[i] = k - j;
2275 j = k;
2276 }
2277
2278 /* compute band indexes */
2279 if (g->block_type == 2) {
2280 if (g->switch_point) {
2281 /* if switched mode, we handle the 36 first samples as
2282 long blocks. For 8000Hz, we handle the 48 first
2283 exponents as long blocks (XXX: check this!) */
2284 if (s->sample_rate_index <= 2)
2285 g->long_end = 8;
2286 else if (s->sample_rate_index != 8)
2287 g->long_end = 6;
2288 else
2289 g->long_end = 4; /* 8000 Hz */
2290
2291 g->short_start = 2 + (s->sample_rate_index != 8);
2292 } else {
2293 g->long_end = 0;
2294 g->short_start = 0;
2295 }
2296 } else {
2297 g->short_start = 13;
2298 g->long_end = 22;
2299 }
2300
2301 g->preflag = 0;
2302 if (!s->lsf)
2303 g->preflag = get_bits(&s->gb, 1);
2304 g->scalefac_scale = get_bits(&s->gb, 1);
2305 g->count1table_select = get_bits(&s->gb, 1);
2306 dprintf("block_type=%d switch_point=%d\n",
2307 g->block_type, g->switch_point);
2308 }
2309 }
2310
2311 if (!s->adu_mode) {
2312 /* now we get bits from the main_data_begin offset */
2313 dprintf("seekback: %d\n", main_data_begin);
2314 seek_to_maindata(s, main_data_begin);
2315 }
2316
2317 for(gr=0;gr<nb_granules;gr++) {
2318 for(ch=0;ch<s->nb_channels;ch++) {
2319 g = &granules[ch][gr];
2320
2321 bits_pos = get_bits_count(&s->gb);
2322
2323 if (!s->lsf) {
2324 uint8_t *sc;
2325 int slen, slen1, slen2;
2326
2327 /* MPEG1 scale factors */
2328 slen1 = slen_table[0][g->scalefac_compress];
2329 slen2 = slen_table[1][g->scalefac_compress];
2330 dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2331 if (g->block_type == 2) {
2332 n = g->switch_point ? 17 : 18;
2333 j = 0;
2334 if(slen1){
2335 for(i=0;i<n;i++)
2336 g->scale_factors[j++] = get_bits(&s->gb, slen1);
2337 }else{
2338 for(i=0;i<n;i++)
2339 g->scale_factors[j++] = 0;
2340 }
2341 if(slen2){
2342 for(i=0;i<18;i++)
2343 g->scale_factors[j++] = get_bits(&s->gb, slen2);
2344 for(i=0;i<3;i++)
2345 g->scale_factors[j++] = 0;
2346 }else{
2347 for(i=0;i<21;i++)
2348 g->scale_factors[j++] = 0;
2349 }
2350 } else {
2351 sc = granules[ch][0].scale_factors;
2352 j = 0;
2353 for(k=0;k<4;k++) {
2354 n = (k == 0 ? 6 : 5);
2355 if ((g->scfsi & (0x8 >> k)) == 0) {
2356 slen = (k < 2) ? slen1 : slen2;
2357 if(slen){
2358 for(i=0;i<n;i++)
2359 g->scale_factors[j++] = get_bits(&s->gb, slen);
2360 }else{
2361 for(i=0;i<n;i++)
2362 g->scale_factors[j++] = 0;
2363 }
2364 } else {
2365 /* simply copy from last granule */
2366 for(i=0;i<n;i++) {
2367 g->scale_factors[j] = sc[j];
2368 j++;
2369 }
2370 }
2371 }
2372 g->scale_factors[j++] = 0;
2373 }
2374 #if defined(DEBUG)
2375 {
2376 dprintf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2377 g->scfsi, gr, ch);
2378 for(i=0;i<j;i++)
2379 dprintf(" %d", g->scale_factors[i]);
2380 dprintf("\n");
2381 }
2382 #endif
2383 } else {
2384 int tindex, tindex2, slen[4], sl, sf;
2385
2386 /* LSF scale factors */
2387 if (g->block_type == 2) {
2388 tindex = g->switch_point ? 2 : 1;
2389 } else {
2390 tindex = 0;
2391 }
2392 sf = g->scalefac_compress;
2393 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2394 /* intensity stereo case */
2395 sf >>= 1;
2396 if (sf < 180) {
2397 lsf_sf_expand(slen, sf, 6, 6, 0);
2398 tindex2 = 3;
2399 } else if (sf < 244) {
2400 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2401 tindex2 = 4;
2402 } else {
2403 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2404 tindex2 = 5;
2405 }
2406 } else {
2407 /* normal case */
2408 if (sf < 400) {
2409 lsf_sf_expand(slen, sf, 5, 4, 4);
2410 tindex2 = 0;
2411 } else if (sf < 500) {
2412 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2413 tindex2 = 1;
2414 } else {
2415 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2416 tindex2 = 2;
2417 g->preflag = 1;
2418 }
2419 }
2420
2421 j = 0;
2422 for(k=0;k<4;k++) {
2423 n = lsf_nsf_table[tindex2][tindex][k];
2424 sl = slen[k];
2425 if(sl){
2426 for(i=0;i<n;i++)
2427 g->scale_factors[j++] = get_bits(&s->gb, sl);
2428 }else{
2429 for(i=0;i<n;i++)
2430 g->scale_factors[j++] = 0;
2431 }
2432 }
2433 /* XXX: should compute exact size */
2434 for(;j<40;j++)
2435 g->scale_factors[j] = 0;
2436 #if defined(DEBUG)
2437 {
2438 dprintf("gr=%d ch=%d scale_factors:\n",
2439 gr, ch);
2440 for(i=0;i<40;i++)
2441 dprintf(" %d", g->scale_factors[i]);
2442 dprintf("\n");
2443 }
2444 #endif
2445 }
2446
2447 exponents_from_scale_factors(s, g, exponents);
2448
2449 /* read Huffman coded residue */
2450 if (huffman_decode(s, g, exponents,
2451 bits_pos + g->part2_3_length) < 0)
2452 return -1;
2453 #if defined(DEBUG)
2454 sample_dump(0, g->sb_hybrid, 576);
2455 #endif
2456
2457 /* skip extension bits */
2458 bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2459 if (bits_left < 0) {
2460 dprintf("bits_left=%d\n", bits_left);
2461 return -1;
2462 }
2463 while (bits_left >= 16) {
2464 skip_bits(&s->gb, 16);
2465 bits_left -= 16;
2466 }
2467 if (bits_left > 0)
2468 skip_bits(&s->gb, bits_left);
2469 } /* ch */
2470
2471 if (s->nb_channels == 2)
2472 compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2473
2474 for(ch=0;ch<s->nb_channels;ch++) {
2475 g = &granules[ch][gr];
2476
2477 reorder_block(s, g);
2478 #if defined(DEBUG)
2479 sample_dump(0, g->sb_hybrid, 576);
2480 #endif
2481 s->compute_antialias(s, g);
2482 #if defined(DEBUG)
2483 sample_dump(1, g->sb_hybrid, 576);
2484 #endif
2485 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
2486 #if defined(DEBUG)
2487 sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2488 #endif
2489 }
2490 } /* gr */
2491 return nb_granules * 18;
2492 }
2493
2494 static int mp_decode_frame(MPADecodeContext *s,
2495 OUT_INT *samples)
2496 {
2497 int i, nb_frames, ch;
2498 OUT_INT *samples_ptr;
2499
2500 init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
2501 (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);
2502
2503 /* skip error protection field */
2504 if (s->error_protection)
2505 get_bits(&s->gb, 16);
2506
2507 dprintf("frame %d:\n", s->frame_count);
2508 switch(s->layer) {
2509 case 1:
2510 nb_frames = mp_decode_layer1(s);
2511 break;
2512 case 2:
2513 nb_frames = mp_decode_layer2(s);
2514 break;
2515 case 3:
2516 default:
2517 nb_frames = mp_decode_layer3(s);
2518 break;
2519 }
2520 #if defined(DEBUG)
2521 for(i=0;i<nb_frames;i++) {
2522 for(ch=0;ch<s->nb_channels;ch++) {
2523 int j;
2524 dprintf("%d-%d:", i, ch);
2525 for(j=0;j<SBLIMIT;j++)
2526 dprintf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2527 dprintf("\n");
2528 }
2529 }
2530 #endif
2531 /* apply the synthesis filter */
2532 for(ch=0;ch<s->nb_channels;ch++) {
2533 samples_ptr = samples + ch;
2534 for(i=0;i<nb_frames;i++) {
2535 ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
2536 window, &s->dither_state,
2537 samples_ptr, s->nb_channels,
2538 s->sb_samples[ch][i]);
2539 samples_ptr += 32 * s->nb_channels;
2540 }
2541 }
2542 #ifdef DEBUG
2543 s->frame_count++;
2544 #endif
2545 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
2546 }
2547
2548 static int decode_frame(AVCodecContext * avctx,
2549 void *data, int *data_size,
2550 uint8_t * buf, int buf_size)
2551 {
2552 MPADecodeContext *s = avctx->priv_data;
2553 uint32_t header;
2554 uint8_t *buf_ptr;
2555 int len, out_size;
2556 OUT_INT *out_samples = data;
2557
2558 buf_ptr = buf;
2559 while (buf_size > 0) {
2560 len = s->inbuf_ptr - s->inbuf;
2561 if (s->frame_size == 0) {
2562 /* special case for next header for first frame in free
2563 format case (XXX: find a simpler method) */
2564 if (s->free_format_next_header != 0) {
2565 s->inbuf[0] = s->free_format_next_header >> 24;
2566 s->inbuf[1] = s->free_format_next_header >> 16;
2567 s->inbuf[2] = s->free_format_next_header >> 8;
2568 s->inbuf[3] = s->free_format_next_header;
2569 s->inbuf_ptr = s->inbuf + 4;
2570 s->free_format_next_header = 0;
2571 goto got_header;
2572 }
2573 /* no header seen : find one. We need at least HEADER_SIZE
2574 bytes to parse it */
2575 len = HEADER_SIZE - len;
2576 if (len > buf_size)
2577 len = buf_size;
2578 if (len > 0) {
2579 memcpy(s->inbuf_ptr, buf_ptr, len);
2580 buf_ptr += len;
2581 buf_size -= len;
2582 s->inbuf_ptr += len;
2583 }
2584 if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
2585 got_header:
2586 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2587 (s->inbuf[2] << 8) | s->inbuf[3];
2588
2589 if (ff_mpa_check_header(header) < 0) {
2590 /* no sync found : move by one byte (inefficient, but simple!) */
2591 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2592 s->inbuf_ptr--;
2593 dprintf("skip %x\n", header);
2594 /* reset free format frame size to give a chance
2595 to get a new bitrate */
2596 s->free_format_frame_size = 0;
2597 } else {
2598 if (decode_header(s, header) == 1) {
2599 /* free format: prepare to compute frame size */
2600 s->frame_size = -1;
2601 }
2602 /* update codec info */
2603 avctx->sample_rate = s->sample_rate;
2604 avctx->channels = s->nb_channels;
2605 avctx->bit_rate = s->bit_rate;
2606 avctx->sub_id = s->layer;
2607 switch(s->layer) {
2608 case 1:
2609 avctx->frame_size = 384;
2610 break;
2611 case 2:
2612 avctx->frame_size = 1152;
2613 break;
2614 case 3:
2615 if (s->lsf)
2616 avctx->frame_size = 576;
2617 else
2618 avctx->frame_size = 1152;
2619 break;
2620 }
2621 }
2622 }
2623 } else if (s->frame_size == -1) {
2624 /* free format : find next sync to compute frame size */
2625 len = MPA_MAX_CODED_FRAME_SIZE - len;
2626 if (len > buf_size)
2627 len = buf_size;
2628 if (len == 0) {
2629 /* frame too long: resync */
2630 s->frame_size = 0;
2631 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2632 s->inbuf_ptr--;
2633 } else {
2634 uint8_t *p, *pend;
2635 uint32_t header1;
2636 int padding;
2637
2638 memcpy(s->inbuf_ptr, buf_ptr, len);
2639 /* check for header */
2640 p = s->inbuf_ptr - 3;
2641 pend = s->inbuf_ptr + len - 4;
2642 while (p <= pend) {
2643 header = (p[0] << 24) | (p[1] << 16) |
2644 (p[2] << 8) | p[3];
2645 header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2646 (s->inbuf[2] << 8) | s->inbuf[3];
2647 /* check with high probability that we have a
2648 valid header */
2649 if ((header & SAME_HEADER_MASK) ==
2650 (header1 & SAME_HEADER_MASK)) {
2651 /* header found: update pointers */
2652 len = (p + 4) - s->inbuf_ptr;
2653 buf_ptr += len;
2654 buf_size -= len;
2655 s->inbuf_ptr = p;
2656 /* compute frame size */
2657 s->free_format_next_header = header;
2658 s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2659 padding = (header1 >> 9) & 1;
2660 if (s->layer == 1)
2661 s->free_format_frame_size -= padding * 4;
2662 else
2663 s->free_format_frame_size -= padding;
2664 dprintf("free frame size=%d padding=%d\n",
2665 s->free_format_frame_size, padding);
2666 decode_header(s, header1);
2667 goto next_data;
2668 }
2669 p++;
2670 }
2671 /* not found: simply increase pointers */
2672 buf_ptr += len;
2673 s->inbuf_ptr += len;
2674 buf_size -= len;
2675 }
2676 } else if (len < s->frame_size) {
2677 if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2678 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
2679 len = s->frame_size - len;
2680 if (len > buf_size)
2681 len = buf_size;
2682 memcpy(s->inbuf_ptr, buf_ptr, len);
2683 buf_ptr += len;
2684 s->inbuf_ptr += len;
2685 buf_size -= len;
2686 }
2687 next_data:
2688 if (s->frame_size > 0 &&
2689 (s->inbuf_ptr - s->inbuf) >= s->frame_size) {
2690 if (avctx->parse_only) {
2691 /* simply return the frame data */
2692 *(uint8_t **)data = s->inbuf;
2693 out_size = s->inbuf_ptr - s->inbuf;
2694 } else {
2695 out_size = mp_decode_frame(s, out_samples);
2696 }
2697 s->inbuf_ptr = s->inbuf;
2698 s->frame_size = 0;
2699 if(out_size>=0)
2700 *data_size = out_size;
2701 else
2702 av_log(avctx, AV_LOG_DEBUG, "Error while decoding mpeg audio frame\n"); //FIXME return -1 / but also return the number of bytes consumed
2703 break;
2704 }
2705 }
2706 return buf_ptr - buf;
2707 }
2708
2709
2710 static int decode_frame_adu(AVCodecContext * avctx,
2711 void *data, int *data_size,
2712 uint8_t * buf, int buf_size)
2713 {
2714 MPADecodeContext *s = avctx->priv_data;
2715 uint32_t header;
2716 int len, out_size;
2717 OUT_INT *out_samples = data;
2718
2719 len = buf_size;
2720
2721 // Discard too short frames
2722 if (buf_size < HEADER_SIZE) {
2723 *data_size = 0;
2724 return buf_size;
2725 }
2726
2727
2728 if (len > MPA_MAX_CODED_FRAME_SIZE)
2729 len = MPA_MAX_CODED_FRAME_SIZE;
2730
2731 memcpy(s->inbuf, buf, len);
2732 s->inbuf_ptr = s->inbuf + len;
2733
2734 // Get header and restore sync word
2735 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2736 (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;
2737
2738 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
2739 *data_size = 0;
2740 return buf_size;
2741 }
2742
2743 decode_header(s, header);
2744 /* update codec info */
2745 avctx->sample_rate = s->sample_rate;
2746 avctx->channels = s->nb_channels;
2747 avctx->bit_rate = s->bit_rate;
2748 avctx->sub_id = s->layer;
2749
2750 avctx->frame_size=s->frame_size = len;
2751
2752 if (avctx->parse_only) {
2753 /* simply return the frame data */
2754 *(uint8_t **)data = s->inbuf;
2755 out_size = s->inbuf_ptr - s->inbuf;
2756 } else {
2757 out_size = mp_decode_frame(s, out_samples);
2758 }
2759
2760 *data_size = out_size;
2761 return buf_size;
2762 }
2763
2764
2765 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2766 static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2}; /* number of mp3 decoder instances */
2767 static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2768 /* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2769 static int chan_offset[9][5] = {
2770 {0},
2771 {0}, // C
2772 {0}, // FLR
2773 {2,0}, // C FLR
2774 {2,0,3}, // C FLR BS
2775 {4,0,2}, // C FLR BLRS
2776 {4,0,2,5}, // C FLR BLRS LFE
2777 {4,0,2,6,5}, // C FLR BLRS BLR LFE
2778 {0,2} // FLR BLRS
2779 };
2780
2781
2782 static int decode_init_mp3on4(AVCodecContext * avctx)
2783 {
2784 MP3On4DecodeContext *s = avctx->priv_data;
2785 int i;
2786
2787 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2788 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2789 return -1;
2790 }
2791
2792 s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2793 s->frames = mp3Frames[s->chan_cfg];
2794 if(!s->frames) {
2795 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2796 return -1;
2797 }
2798 avctx->channels = mp3Channels[s->chan_cfg];
2799
2800 /* Init the first mp3 decoder in standard way, so that all tables get builded
2801 * We replace avctx->priv_data with the context of the first decoder so that
2802 * decode_init() does not have to be changed.
2803 * Other decoders will be inited here copying data from the first context
2804 */
2805 // Allocate zeroed memory for the first decoder context
2806 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2807 // Put decoder context in place to make init_decode() happy
2808 avctx->priv_data = s->mp3decctx[0];
2809 decode_init(avctx);
2810 // Restore mp3on4 context pointer
2811 avctx->priv_data = s;
2812 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2813
2814 /* Create a separate codec/context for each frame (first is already ok).
2815 * Each frame is 1 or 2 channels - up to 5 frames allowed
2816 */
2817 for (i = 1; i < s->frames; i++) {
2818 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2819 s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2820 s->mp3decctx[i]->inbuf = &s->mp3decctx[i]->inbuf1[0][BACKSTEP_SIZE];
2821 s->mp3decctx[i]->inbuf_ptr = s->mp3decctx[i]->inbuf;
2822 s->mp3decctx[i]->adu_mode = 1;
2823 }
2824
2825 return 0;
2826 }
2827
2828
2829 static int decode_close_mp3on4(AVCodecContext * avctx)
2830 {
2831 MP3On4DecodeContext *s = avctx->priv_data;
2832 int i;
2833
2834 for (i = 0; i < s->frames; i++)
2835 if (s->mp3decctx[i])
2836 av_free(s->mp3decctx[i]);
2837
2838 return 0;
2839 }
2840
2841
2842 static int decode_frame_mp3on4(AVCodecContext * avctx,
2843 void *data, int *data_size,
2844 uint8_t * buf, int buf_size)
2845 {
2846 MP3On4DecodeContext *s = avctx->priv_data;
2847 MPADecodeContext *m;
2848 int len, out_size = 0;
2849 uint32_t header;
2850 OUT_INT *out_samples = data;
2851 OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2852 OUT_INT *outptr, *bp;
2853 int fsize;
2854 unsigned char *start2 = buf, *start;
2855 int fr, i, j, n;
2856 int off = avctx->channels;
2857 int *coff = chan_offset[s->chan_cfg];
2858
2859 len = buf_size;
2860
2861 // Discard too short frames
2862 if (buf_size < HEADER_SIZE) {
2863 *data_size = 0;
2864 return buf_size;
2865 }
2866
2867 // If only one decoder interleave is not needed
2868 outptr = s->frames == 1 ? out_samples : decoded_buf;
2869
2870 for (fr = 0; fr < s->frames; fr++) {
2871 start = start2;
2872 fsize = (start[0] << 4) | (start[1] >> 4);
2873 start2 += fsize;
2874 if (fsize > len)
2875 fsize = len;
2876 len -= fsize;
2877 if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2878 fsize = MPA_MAX_CODED_FRAME_SIZE;
2879 m = s->mp3decctx[fr];
2880 assert (m != NULL);
2881 /* copy original to new */
2882 m->inbuf_ptr = m->inbuf + fsize;
2883 memcpy(m->inbuf, start, fsize);
2884
2885 // Get header
2886 header = (m->inbuf[0] << 24) | (m->inbuf[1] << 16) |
2887 (m->inbuf[2] << 8) | m->inbuf[3] | 0xfff00000;
2888
2889 if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2890 *data_size = 0;
2891 return buf_size;
2892 }
2893
2894 decode_header(m, header);
2895 mp_decode_frame(m, decoded_buf);
2896
2897 n = MPA_FRAME_SIZE * m->nb_channels;
2898 out_size += n * sizeof(OUT_INT);
2899 if(s->frames > 1) {
2900 /* interleave output data */
2901 bp = out_samples + coff[fr];
2902 if(m->nb_channels == 1) {
2903 for(j = 0; j < n; j++) {
2904 *bp = decoded_buf[j];
2905 bp += off;
2906 }
2907 } else {
2908 for(j = 0; j < n; j++) {
2909 bp[0] = decoded_buf[j++];
2910 bp[1] = decoded_buf[j];
2911 bp += off;
2912 }
2913 }
2914 }
2915 }
2916
2917 /* update codec info */
2918 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2919 avctx->frame_size= buf_size;
2920 avctx->bit_rate = 0;
2921 for (i = 0; i < s->frames; i++)
2922 avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2923
2924 *data_size = out_size;
2925 return buf_size;
2926 }
2927
2928
2929 AVCodec mp2_decoder =
2930 {
2931 "mp2",
2932 CODEC_TYPE_AUDIO,
2933 CODEC_ID_MP2,
2934 sizeof(MPADecodeContext),
2935 decode_init,
2936 NULL,
2937 NULL,
2938 decode_frame,
2939 CODEC_CAP_PARSE_ONLY,
2940 };
2941
2942 AVCodec mp3_decoder =
2943 {
2944 "mp3",
2945 CODEC_TYPE_AUDIO,
2946 CODEC_ID_MP3,
2947 sizeof(MPADecodeContext),
2948 decode_init,
2949 NULL,
2950 NULL,
2951 decode_frame,
2952 CODEC_CAP_PARSE_ONLY,
2953 };
2954
2955 AVCodec mp3adu_decoder =
2956 {
2957 "mp3adu",
2958 CODEC_TYPE_AUDIO,
2959 CODEC_ID_MP3ADU,
2960 sizeof(MPADecodeContext),
2961 decode_init,
2962 NULL,
2963 NULL,
2964 decode_frame_adu,
2965 CODEC_CAP_PARSE_ONLY,
2966 };
2967
2968 AVCodec mp3on4_decoder =
2969 {
2970 "mp3on4",
2971 CODEC_TYPE_AUDIO,
2972 CODEC_ID_MP3ON4,
2973 sizeof(MP3On4DecodeContext),
2974 decode_init_mp3on4,
2975 NULL,
2976 decode_close_mp3on4,