faster, simpler and more accurate l3_unscale()
[libav.git] / libavcodec / mpegaudiodec.c
CommitLineData
de6d9b64
FB
1/*
2 * MPEG Audio decoder
ff4ec49e 3 * Copyright (c) 2001, 2002 Fabrice Bellard.
de6d9b64 4 *
ff4ec49e
FB
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.
de6d9b64 9 *
ff4ec49e 10 * This library is distributed in the hope that it will be useful,
de6d9b64 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
ff4ec49e
FB
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
de6d9b64 14 *
ff4ec49e
FB
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
de6d9b64 18 */
983e3246
MN
19
20/**
21 * @file mpegaudiodec.c
22 * MPEG Audio decoder.
23 */
24
239c2f4c 25//#define DEBUG
de6d9b64 26#include "avcodec.h"
caa336b4 27#include "bitstream.h"
239c2f4c 28#include "mpegaudio.h"
84f986c0 29#include "dsputil.h"
de6d9b64
FB
30
31/*
239c2f4c
FB
32 * TODO:
33 * - in low precision mode, use more 16 bit multiplies in synth filter
34 * - test lsf / mpeg25 extensively.
de6d9b64
FB
35 */
36
239c2f4c
FB
37/* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
38 audio decoder */
81552334
FB
39#ifdef CONFIG_MPEGAUDIO_HP
40#define USE_HIGHPRECISION
41#endif
239c2f4c
FB
42
43#ifdef USE_HIGHPRECISION
44#define FRAC_BITS 23 /* fractional bits for sb_samples and dct */
45#define WFRAC_BITS 16 /* fractional bits for window */
46#else
47#define FRAC_BITS 15 /* fractional bits for sb_samples and dct */
48#define WFRAC_BITS 14 /* fractional bits for window */
49#endif
50
a3a5f4d6
MN
51#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
52typedef int32_t OUT_INT;
53#define OUT_MAX INT32_MAX
54#define OUT_MIN INT32_MIN
55#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 31)
56#else
57typedef int16_t OUT_INT;
58#define OUT_MAX INT16_MAX
59#define OUT_MIN INT16_MIN
60#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
61#endif
62
239c2f4c
FB
63#define FRAC_ONE (1 << FRAC_BITS)
64
0c1a9eda
ZK
65#define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
66#define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))
239c2f4c
FB
67#define FIX(a) ((int)((a) * FRAC_ONE))
68/* WARNING: only correct for posititive numbers */
69#define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
70#define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)
71
711ae726
MN
72#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
73//#define MULH(a,b) (((int64_t)(a) * (int64_t)(b))>>32) //gcc 3.4 creates an incredibly bloated mess out of this
74static always_inline int MULH(int a, int b){
75 return ((int64_t)(a) * (int64_t)(b))>>32;
76}
77
239c2f4c 78#if FRAC_BITS <= 15
0c1a9eda 79typedef int16_t MPA_INT;
239c2f4c 80#else
0c1a9eda 81typedef int32_t MPA_INT;
239c2f4c
FB
82#endif
83
84/****************/
85
de6d9b64
FB
86#define HEADER_SIZE 4
87#define BACKSTEP_SIZE 512
88
a1e257b2
MN
89struct GranuleDef;
90
de6d9b64 91typedef struct MPADecodeContext {
0c1a9eda 92 uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE]; /* input buffer */
de6d9b64 93 int inbuf_index;
0c1a9eda 94 uint8_t *inbuf_ptr, *inbuf;
de6d9b64 95 int frame_size;
239c2f4c
FB
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) */
0c1a9eda 99 uint32_t free_format_next_header;
de6d9b64
FB
100 int error_protection;
101 int layer;
102 int sample_rate;
239c2f4c 103 int sample_rate_index; /* between 0 and 8 */
de6d9b64
FB
104 int bit_rate;
105 int old_frame_size;
106 GetBitContext gb;
239c2f4c
FB
107 int nb_channels;
108 int mode;
109 int mode_ext;
110 int lsf;
a05c8d71 111 MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));
239c2f4c 112 int synth_buf_offset[MPA_MAX_CHANNELS];
a05c8d71 113 int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));
0c1a9eda 114 int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
239c2f4c
FB
115#ifdef DEBUG
116 int frame_count;
117#endif
a1e257b2 118 void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);
1ede228a 119 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
a7a85899 120 unsigned int dither_state;
de6d9b64
FB
121} MPADecodeContext;
122
239c2f4c
FB
123/* layer 3 "granule" */
124typedef struct GranuleDef {
0c1a9eda 125 uint8_t scfsi;
239c2f4c
FB
126 int part2_3_length;
127 int big_values;
128 int global_gain;
129 int scalefac_compress;
0c1a9eda
ZK
130 uint8_t block_type;
131 uint8_t switch_point;
239c2f4c
FB
132 int table_select[3];
133 int subblock_gain[3];
0c1a9eda
ZK
134 uint8_t scalefac_scale;
135 uint8_t count1table_select;
239c2f4c
FB
136 int region_size[3]; /* number of huffman codes in each region */
137 int preflag;
138 int short_start, long_end; /* long/short band indexes */
0c1a9eda
ZK
139 uint8_t scale_factors[40];
140 int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
239c2f4c 141} GranuleDef;
de6d9b64 142
239c2f4c
FB
143#define MODE_EXT_MS_STEREO 2
144#define MODE_EXT_I_STEREO 1
145
146/* layer 3 huffman tables */
147typedef struct HuffTable {
148 int xsize;
0c1a9eda
ZK
149 const uint8_t *bits;
150 const uint16_t *codes;
239c2f4c
FB
151} HuffTable;
152
153#include "mpegaudiodectab.h"
154
a1e257b2
MN
155static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
156static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
157
239c2f4c
FB
158/* vlc structure for decoding layer 3 huffman tables */
159static VLC huff_vlc[16];
0c1a9eda 160static uint8_t *huff_code_table[16];
239c2f4c
FB
161static VLC huff_quad_vlc[2];
162/* computed from band_size_long */
0c1a9eda 163static uint16_t band_index_long[9][23];
239c2f4c 164/* XXX: free when all decoders are closed */
d04728bb 165#define TABLE_4_3_SIZE (8191 + 16)*4
0c1a9eda 166static int8_t *table_4_3_exp;
0c1a9eda 167static uint32_t *table_4_3_value;
239c2f4c 168/* intensity stereo coef table */
0c1a9eda
ZK
169static int32_t is_table[2][16];
170static int32_t is_table_lsf[2][2][16];
a1e257b2
MN
171static int32_t csa_table[8][4];
172static float csa_table_float[8][4];
0c1a9eda 173static int32_t mdct_win[8][36];
239c2f4c
FB
174
175/* lower 2 bits: modulo 3, higher bits: shift */
0c1a9eda 176static uint16_t scale_factor_modshift[64];
239c2f4c 177/* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
0c1a9eda 178static int32_t scale_factor_mult[15][3];
239c2f4c
FB
179/* mult table for layer 2 group quantization */
180
181#define SCALE_GEN(v) \
182{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
183
0c1a9eda 184static int32_t scale_factor_mult2[3][3] = {
81552334
FB
185 SCALE_GEN(4.0 / 3.0), /* 3 steps */
186 SCALE_GEN(4.0 / 5.0), /* 5 steps */
187 SCALE_GEN(4.0 / 9.0), /* 9 steps */
239c2f4c
FB
188};
189
bf1f4da0 190void ff_mpa_synth_init(MPA_INT *window);
a05c8d71 191static MPA_INT window[512] __attribute__((aligned(16)));
239c2f4c
FB
192
193/* layer 1 unscaling */
194/* n = number of bits of the mantissa minus 1 */
195static inline int l1_unscale(int n, int mant, int scale_factor)
196{
197 int shift, mod;
0c1a9eda 198 int64_t val;
239c2f4c
FB
199
200 shift = scale_factor_modshift[scale_factor];
201 mod = shift & 3;
202 shift >>= 2;
203 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
204 shift += n;
81552334
FB
205 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
206 return (int)((val + (1LL << (shift - 1))) >> shift);
239c2f4c
FB
207}
208
209static inline int l2_unscale_group(int steps, int mant, int scale_factor)
210{
211 int shift, mod, val;
212
213 shift = scale_factor_modshift[scale_factor];
214 mod = shift & 3;
215 shift >>= 2;
81552334
FB
216
217 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
218 /* NOTE: at this point, 0 <= shift <= 21 */
219 if (shift > 0)
220 val = (val + (1 << (shift - 1))) >> shift;
221 return val;
239c2f4c
FB
222}
223
224/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
225static inline int l3_unscale(int value, int exponent)
226{
d04728bb 227
239c2f4c 228 unsigned int m;
239c2f4c
FB
229 int e;
230
d04728bb
MN
231 e = table_4_3_exp [4*value + (exponent&3)];
232 m = table_4_3_value[4*value + (exponent&3)];
233 e -= (exponent >> 2);
234 assert(e>=1);
239c2f4c 235 if (e > 31)
b696d2a6 236 return 0;
239c2f4c 237 m = (m + (1 << (e-1))) >> e;
d04728bb 238
239c2f4c 239 return m;
239c2f4c
FB
240}
241
f9ed4f88
FB
242/* all integer n^(4/3) computation code */
243#define DEV_ORDER 13
244
245#define POW_FRAC_BITS 24
246#define POW_FRAC_ONE (1 << POW_FRAC_BITS)
247#define POW_FIX(a) ((int)((a) * POW_FRAC_ONE))
0c1a9eda 248#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
f9ed4f88
FB
249
250static int dev_4_3_coefs[DEV_ORDER];
251
252static int pow_mult3[3] = {
253 POW_FIX(1.0),
254 POW_FIX(1.25992104989487316476),
255 POW_FIX(1.58740105196819947474),
256};
257
258static void int_pow_init(void)
259{
260 int i, a;
261
262 a = POW_FIX(1.0);
263 for(i=0;i<DEV_ORDER;i++) {
264 a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
265 dev_4_3_coefs[i] = a;
266 }
267}
268
269/* return the mantissa and the binary exponent */
270static int int_pow(int i, int *exp_ptr)
271{
272 int e, er, eq, j;
273 int a, a1;
274
275 /* renormalize */
276 a = i;
277 e = POW_FRAC_BITS;
278 while (a < (1 << (POW_FRAC_BITS - 1))) {
279 a = a << 1;
280 e--;
281 }
282 a -= (1 << POW_FRAC_BITS);
283 a1 = 0;
284 for(j = DEV_ORDER - 1; j >= 0; j--)
285 a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
286 a = (1 << POW_FRAC_BITS) + a1;
287 /* exponent compute (exact) */
288 e = e * 4;
289 er = e % 3;
290 eq = e / 3;
291 a = POW_MULL(a, pow_mult3[er]);
292 while (a >= 2 * POW_FRAC_ONE) {
293 a = a >> 1;
294 eq++;
295 }
296 /* convert to float */
297 while (a < POW_FRAC_ONE) {
298 a = a << 1;
299 eq--;
300 }
59d3e367 301 /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
81552334 302#if POW_FRAC_BITS > FRAC_BITS
59d3e367
FB
303 a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
304 /* correct overflow */
305 if (a >= 2 * (1 << FRAC_BITS)) {
306 a = a >> 1;
307 eq++;
308 }
309#endif
f9ed4f88 310 *exp_ptr = eq;
f9ed4f88 311 return a;
f9ed4f88 312}
de6d9b64
FB
313
314static int decode_init(AVCodecContext * avctx)
315{
316 MPADecodeContext *s = avctx->priv_data;
b587a7cb 317 static int init=0;
239c2f4c 318 int i, j, k;
de6d9b64 319
a3a5f4d6
MN
320#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
321 avctx->sample_fmt= SAMPLE_FMT_S32;
322#else
323 avctx->sample_fmt= SAMPLE_FMT_S16;
324#endif
325
ac806113
MN
326 if(avctx->antialias_algo == FF_AA_INT)
327 s->compute_antialias= compute_antialias_integer;
328 else
329 s->compute_antialias= compute_antialias_float;
330
8c5b5683 331 if (!init && !avctx->parse_only) {
239c2f4c
FB
332 /* scale factors table for layer 1/2 */
333 for(i=0;i<64;i++) {
334 int shift, mod;
335 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
81552334 336 shift = (i / 3);
239c2f4c 337 mod = i % 3;
239c2f4c
FB
338 scale_factor_modshift[i] = mod | (shift << 2);
339 }
340
341 /* scale factor multiply for layer 1 */
342 for(i=0;i<15;i++) {
343 int n, norm;
344 n = i + 2;
0c1a9eda 345 norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
81552334
FB
346 scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
347 scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
348 scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
239c2f4c
FB
349 dprintf("%d: norm=%x s=%x %x %x\n",
350 i, norm,
351 scale_factor_mult[i][0],
352 scale_factor_mult[i][1],
353 scale_factor_mult[i][2]);
354 }
355
bf1f4da0 356 ff_mpa_synth_init(window);
239c2f4c
FB
357
358 /* huffman decode tables */
359 huff_code_table[0] = NULL;
360 for(i=1;i<16;i++) {
361 const HuffTable *h = &mpa_huff_tables[i];
5c91a675
ZK
362 int xsize, x, y;
363 unsigned int n;
0c1a9eda 364 uint8_t *code_table;
239c2f4c
FB
365
366 xsize = h->xsize;
367 n = xsize * xsize;
368 /* XXX: fail test */
369 init_vlc(&huff_vlc[i], 8, n,
073c2593 370 h->bits, 1, 1, h->codes, 2, 2, 1);
239c2f4c
FB
371
372 code_table = av_mallocz(n);
373 j = 0;
374 for(x=0;x<xsize;x++) {
375 for(y=0;y<xsize;y++)
376 code_table[j++] = (x << 4) | y;
377 }
378 huff_code_table[i] = code_table;
379 }
380 for(i=0;i<2;i++) {
381 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
073c2593 382 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
239c2f4c
FB
383 }
384
385 for(i=0;i<9;i++) {
386 k = 0;
387 for(j=0;j<22;j++) {
388 band_index_long[i][j] = k;
389 k += band_size_long[i][j];
390 }
391 band_index_long[i][22] = k;
392 }
393
855ea723 394 /* compute n ^ (4/3) and store it in mantissa/exp format */
8d1f2ba5
MN
395 table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));
396 if(!table_4_3_exp)
855ea723 397 return -1;
8d1f2ba5
MN
398 table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));
399 if(!table_4_3_value)
239c2f4c 400 return -1;
239c2f4c 401
f9ed4f88 402 int_pow_init();
239c2f4c 403 for(i=1;i<TABLE_4_3_SIZE;i++) {
d04728bb 404 double f, fm;
239c2f4c 405 int e, m;
d04728bb
MN
406 f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
407 fm = frexp(f, &e);
408 m = FIXHR(fm*0.5);
409 e+= FRAC_BITS - 31;
410
239c2f4c
FB
411 /* normalized to FRAC_BITS */
412 table_4_3_value[i] = m;
d04728bb
MN
413// av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
414 table_4_3_exp[i] = -e;
239c2f4c 415 }
239c2f4c
FB
416
417 for(i=0;i<7;i++) {
418 float f;
419 int v;
420 if (i != 6) {
421 f = tan((double)i * M_PI / 12.0);
422 v = FIXR(f / (1.0 + f));
423 } else {
424 v = FIXR(1.0);
425 }
426 is_table[0][i] = v;
427 is_table[1][6 - i] = v;
428 }
429 /* invalid values */
430 for(i=7;i<16;i++)
431 is_table[0][i] = is_table[1][i] = 0.0;
432
433 for(i=0;i<16;i++) {
434 double f;
435 int e, k;
436
437 for(j=0;j<2;j++) {
438 e = -(j + 1) * ((i + 1) >> 1);
439 f = pow(2.0, e / 4.0);
440 k = i & 1;
441 is_table_lsf[j][k ^ 1][i] = FIXR(f);
442 is_table_lsf[j][k][i] = FIXR(1.0);
443 dprintf("is_table_lsf %d %d: %x %x\n",
444 i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
445 }
446 }
447
448 for(i=0;i<8;i++) {
449 float ci, cs, ca;
450 ci = ci_table[i];
451 cs = 1.0 / sqrt(1.0 + ci * ci);
452 ca = cs * ci;
453 csa_table[i][0] = FIX(cs);
454 csa_table[i][1] = FIX(ca);
a1e257b2
MN
455 csa_table[i][2] = FIX(ca) + FIX(cs);
456 csa_table[i][3] = FIX(ca) - FIX(cs);
457 csa_table_float[i][0] = cs;
458 csa_table_float[i][1] = ca;
459 csa_table_float[i][2] = ca + cs;
460 csa_table_float[i][3] = ca - cs;
461// printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
711ae726 462// av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
239c2f4c
FB
463 }
464
465 /* compute mdct windows */
466 for(i=0;i<36;i++) {
711ae726
MN
467 for(j=0; j<4; j++){
468 double d;
469 if(j==2) continue;
470
471 d= sin(M_PI * (i + 0.5) / 36.0);
472 if(j==1){
473 if (i>=30) d= 0;
474 else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
475 else if(i>=18) d= 1;
476 }else if(j==3){
477 if (i< 6) d= 0;
478 else if(i< 12) d= sin(M_PI * (i - 6 + 0.5) / 12.0);
479 else if(i< 18) d= 1;
480 }
481 //merge last stage of imdct into the window coefficients
482 if (i/9 == 0) d*= 0.5 / cos(M_PI*(2*( i) +19)/72);
483 else if(i/9 == 1) d*= 0.5 / cos(M_PI*(2*(17 - i) +19)/72);
484 else if(i/9 == 2) d*= 0.5 / cos(M_PI*(2*( i) +19)/72);
485 else d*=-0.5 / cos(M_PI*(2*(17 - i) +19)/72);
486 mdct_win[j][i] = FIXHR((d / (1<<5)));
487// av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
488 }
239c2f4c
FB
489 }
490
491 for(i=0;i<12;i++)
492 mdct_win[2][i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));
493
494 /* NOTE: we do frequency inversion adter the MDCT by changing
495 the sign of the right window coefs */
496 for(j=0;j<4;j++) {
497 for(i=0;i<36;i+=2) {
498 mdct_win[j + 4][i] = mdct_win[j][i];
499 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
500 }
501 }
502
503#if defined(DEBUG)
504 for(j=0;j<8;j++) {
505 printf("win%d=\n", j);
506 for(i=0;i<36;i++)
507 printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);
508 printf("\n");
509 }
510#endif
de6d9b64 511 init = 1;
de6d9b64
FB
512 }
513
514 s->inbuf_index = 0;
515 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
516 s->inbuf_ptr = s->inbuf;
239c2f4c
FB
517#ifdef DEBUG
518 s->frame_count = 0;
519#endif
1ede228a
RT
520 if (avctx->codec_id == CODEC_ID_MP3ADU)
521 s->adu_mode = 1;
de6d9b64
FB
522 return 0;
523}
524
ef9f7306 525/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
239c2f4c
FB
526
527/* cos(i*pi/64) */
528
529#define COS0_0 FIXR(0.50060299823519630134)
530#define COS0_1 FIXR(0.50547095989754365998)
531#define COS0_2 FIXR(0.51544730992262454697)
532#define COS0_3 FIXR(0.53104259108978417447)
533#define COS0_4 FIXR(0.55310389603444452782)
534#define COS0_5 FIXR(0.58293496820613387367)
535#define COS0_6 FIXR(0.62250412303566481615)
536#define COS0_7 FIXR(0.67480834145500574602)
537#define COS0_8 FIXR(0.74453627100229844977)
538#define COS0_9 FIXR(0.83934964541552703873)
539#define COS0_10 FIXR(0.97256823786196069369)
540#define COS0_11 FIXR(1.16943993343288495515)
541#define COS0_12 FIXR(1.48416461631416627724)
542#define COS0_13 FIXR(2.05778100995341155085)
543#define COS0_14 FIXR(3.40760841846871878570)
544#define COS0_15 FIXR(10.19000812354805681150)
545
546#define COS1_0 FIXR(0.50241928618815570551)
547#define COS1_1 FIXR(0.52249861493968888062)
548#define COS1_2 FIXR(0.56694403481635770368)
549#define COS1_3 FIXR(0.64682178335999012954)
550#define COS1_4 FIXR(0.78815462345125022473)
551#define COS1_5 FIXR(1.06067768599034747134)
552#define COS1_6 FIXR(1.72244709823833392782)
553#define COS1_7 FIXR(5.10114861868916385802)
554
555#define COS2_0 FIXR(0.50979557910415916894)
556#define COS2_1 FIXR(0.60134488693504528054)
557#define COS2_2 FIXR(0.89997622313641570463)
558#define COS2_3 FIXR(2.56291544774150617881)
559
560#define COS3_0 FIXR(0.54119610014619698439)
561#define COS3_1 FIXR(1.30656296487637652785)
562
563#define COS4_0 FIXR(0.70710678118654752439)
564
565/* butterfly operator */
566#define BF(a, b, c)\
567{\
568 tmp0 = tab[a] + tab[b];\
569 tmp1 = tab[a] - tab[b];\
570 tab[a] = tmp0;\
571 tab[b] = MULL(tmp1, c);\
572}
573
574#define BF1(a, b, c, d)\
575{\
576 BF(a, b, COS4_0);\
577 BF(c, d, -COS4_0);\
578 tab[c] += tab[d];\
579}
580
581#define BF2(a, b, c, d)\
582{\
583 BF(a, b, COS4_0);\
584 BF(c, d, -COS4_0);\
585 tab[c] += tab[d];\
586 tab[a] += tab[c];\
587 tab[c] += tab[b];\
588 tab[b] += tab[d];\
589}
590
591#define ADD(a, b) tab[a] += tab[b]
592
593/* DCT32 without 1/sqrt(2) coef zero scaling. */
0c1a9eda 594static void dct32(int32_t *out, int32_t *tab)
239c2f4c
FB
595{
596 int tmp0, tmp1;
597
598 /* pass 1 */
599 BF(0, 31, COS0_0);
600 BF(1, 30, COS0_1);
601 BF(2, 29, COS0_2);
602 BF(3, 28, COS0_3);
603 BF(4, 27, COS0_4);
604 BF(5, 26, COS0_5);
605 BF(6, 25, COS0_6);
606 BF(7, 24, COS0_7);
607 BF(8, 23, COS0_8);
608 BF(9, 22, COS0_9);
609 BF(10, 21, COS0_10);
610 BF(11, 20, COS0_11);
611 BF(12, 19, COS0_12);
612 BF(13, 18, COS0_13);
613 BF(14, 17, COS0_14);
614 BF(15, 16, COS0_15);
615
616 /* pass 2 */
617 BF(0, 15, COS1_0);
618 BF(1, 14, COS1_1);
619 BF(2, 13, COS1_2);
620 BF(3, 12, COS1_3);
621 BF(4, 11, COS1_4);
622 BF(5, 10, COS1_5);
623 BF(6, 9, COS1_6);
624 BF(7, 8, COS1_7);
625
626 BF(16, 31, -COS1_0);
627 BF(17, 30, -COS1_1);
628 BF(18, 29, -COS1_2);
629 BF(19, 28, -COS1_3);
630 BF(20, 27, -COS1_4);
631 BF(21, 26, -COS1_5);
632 BF(22, 25, -COS1_6);
633 BF(23, 24, -COS1_7);
634
635 /* pass 3 */
636 BF(0, 7, COS2_0);
637 BF(1, 6, COS2_1);
638 BF(2, 5, COS2_2);
639 BF(3, 4, COS2_3);
640
641 BF(8, 15, -COS2_0);
642 BF(9, 14, -COS2_1);
643 BF(10, 13, -COS2_2);
644 BF(11, 12, -COS2_3);
645
646 BF(16, 23, COS2_0);
647 BF(17, 22, COS2_1);
648 BF(18, 21, COS2_2);
649 BF(19, 20, COS2_3);
650
651 BF(24, 31, -COS2_0);
652 BF(25, 30, -COS2_1);
653 BF(26, 29, -COS2_2);
654 BF(27, 28, -COS2_3);
655
656 /* pass 4 */
657 BF(0, 3, COS3_0);
658 BF(1, 2, COS3_1);
659
660 BF(4, 7, -COS3_0);
661 BF(5, 6, -COS3_1);
662
663 BF(8, 11, COS3_0);
664 BF(9, 10, COS3_1);
665
666 BF(12, 15, -COS3_0);
667 BF(13, 14, -COS3_1);
668
669 BF(16, 19, COS3_0);
670 BF(17, 18, COS3_1);
671
672 BF(20, 23, -COS3_0);
673 BF(21, 22, -COS3_1);
674
675 BF(24, 27, COS3_0);
676 BF(25, 26, COS3_1);
677
678 BF(28, 31, -COS3_0);
679 BF(29, 30, -COS3_1);
680
681 /* pass 5 */
682 BF1(0, 1, 2, 3);
683 BF2(4, 5, 6, 7);
684 BF1(8, 9, 10, 11);
685 BF2(12, 13, 14, 15);
686 BF1(16, 17, 18, 19);
687 BF2(20, 21, 22, 23);
688 BF1(24, 25, 26, 27);
689 BF2(28, 29, 30, 31);
690
691 /* pass 6 */
692
693 ADD( 8, 12);
694 ADD(12, 10);
695 ADD(10, 14);
696 ADD(14, 9);
697 ADD( 9, 13);
698 ADD(13, 11);
699 ADD(11, 15);
700
701 out[ 0] = tab[0];
702 out[16] = tab[1];
703 out[ 8] = tab[2];
704 out[24] = tab[3];
705 out[ 4] = tab[4];
706 out[20] = tab[5];
707 out[12] = tab[6];
708 out[28] = tab[7];
709 out[ 2] = tab[8];
710 out[18] = tab[9];
711 out[10] = tab[10];
712 out[26] = tab[11];
713 out[ 6] = tab[12];
714 out[22] = tab[13];
715 out[14] = tab[14];
716 out[30] = tab[15];
717
718 ADD(24, 28);
719 ADD(28, 26);
720 ADD(26, 30);
721 ADD(30, 25);
722 ADD(25, 29);
723 ADD(29, 27);
724 ADD(27, 31);
725
726 out[ 1] = tab[16] + tab[24];
727 out[17] = tab[17] + tab[25];
728 out[ 9] = tab[18] + tab[26];
729 out[25] = tab[19] + tab[27];
730 out[ 5] = tab[20] + tab[28];
731 out[21] = tab[21] + tab[29];
732 out[13] = tab[22] + tab[30];
733 out[29] = tab[23] + tab[31];
734 out[ 3] = tab[24] + tab[20];
735 out[19] = tab[25] + tab[21];
736 out[11] = tab[26] + tab[22];
737 out[27] = tab[27] + tab[23];
738 out[ 7] = tab[28] + tab[18];
739 out[23] = tab[29] + tab[19];
740 out[15] = tab[30] + tab[17];
741 out[31] = tab[31];
742}
743
239c2f4c
FB
744#if FRAC_BITS <= 15
745
a7a85899 746static inline int round_sample(int *sum)
8c5b5683
FB
747{
748 int sum1;
a7a85899
MN
749 sum1 = (*sum) >> OUT_SHIFT;
750 *sum &= (1<<OUT_SHIFT)-1;
a3a5f4d6
MN
751 if (sum1 < OUT_MIN)
752 sum1 = OUT_MIN;
753 else if (sum1 > OUT_MAX)
754 sum1 = OUT_MAX;
8c5b5683 755 return sum1;
239c2f4c
FB
756}
757
8c5b5683
FB
758#if defined(ARCH_POWERPC_405)
759
760/* signed 16x16 -> 32 multiply add accumulate */
761#define MACS(rt, ra, rb) \
762 asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
763
764/* signed 16x16 -> 32 multiply */
765#define MULS(ra, rb) \
766 ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })
239c2f4c
FB
767
768#else
769
8c5b5683
FB
770/* signed 16x16 -> 32 multiply add accumulate */
771#define MACS(rt, ra, rb) rt += (ra) * (rb)
772
773/* signed 16x16 -> 32 multiply */
774#define MULS(ra, rb) ((ra) * (rb))
775
776#endif
777
778#else
779
a7a85899 780static inline int round_sample(int64_t *sum)
8c5b5683
FB
781{
782 int sum1;
a7a85899
MN
783 sum1 = (int)((*sum) >> OUT_SHIFT);
784 *sum &= (1<<OUT_SHIFT)-1;
a3a5f4d6
MN
785 if (sum1 < OUT_MIN)
786 sum1 = OUT_MIN;
787 else if (sum1 > OUT_MAX)
788 sum1 = OUT_MAX;
8c5b5683 789 return sum1;
239c2f4c
FB
790}
791
8c5b5683
FB
792#define MULS(ra, rb) MUL64(ra, rb)
793
794#endif
795
796#define SUM8(sum, op, w, p) \
239c2f4c 797{ \
8c5b5683
FB
798 sum op MULS((w)[0 * 64], p[0 * 64]);\
799 sum op MULS((w)[1 * 64], p[1 * 64]);\
800 sum op MULS((w)[2 * 64], p[2 * 64]);\
801 sum op MULS((w)[3 * 64], p[3 * 64]);\
802 sum op MULS((w)[4 * 64], p[4 * 64]);\
803 sum op MULS((w)[5 * 64], p[5 * 64]);\
804 sum op MULS((w)[6 * 64], p[6 * 64]);\
805 sum op MULS((w)[7 * 64], p[7 * 64]);\
806}
807
808#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
809{ \
810 int tmp;\
811 tmp = p[0 * 64];\
812 sum1 op1 MULS((w1)[0 * 64], tmp);\
813 sum2 op2 MULS((w2)[0 * 64], tmp);\
814 tmp = p[1 * 64];\
815 sum1 op1 MULS((w1)[1 * 64], tmp);\
816 sum2 op2 MULS((w2)[1 * 64], tmp);\
817 tmp = p[2 * 64];\
818 sum1 op1 MULS((w1)[2 * 64], tmp);\
819 sum2 op2 MULS((w2)[2 * 64], tmp);\
820 tmp = p[3 * 64];\
821 sum1 op1 MULS((w1)[3 * 64], tmp);\
822 sum2 op2 MULS((w2)[3 * 64], tmp);\
823 tmp = p[4 * 64];\
824 sum1 op1 MULS((w1)[4 * 64], tmp);\
825 sum2 op2 MULS((w2)[4 * 64], tmp);\
826 tmp = p[5 * 64];\
827 sum1 op1 MULS((w1)[5 * 64], tmp);\
828 sum2 op2 MULS((w2)[5 * 64], tmp);\
829 tmp = p[6 * 64];\
830 sum1 op1 MULS((w1)[6 * 64], tmp);\
831 sum2 op2 MULS((w2)[6 * 64], tmp);\
832 tmp = p[7 * 64];\
833 sum1 op1 MULS((w1)[7 * 64], tmp);\
834 sum2 op2 MULS((w2)[7 * 64], tmp);\
239c2f4c
FB
835}
836
bf1f4da0
AB
837void ff_mpa_synth_init(MPA_INT *window)
838{
839 int i;
840
841 /* max = 18760, max sum over all 16 coefs : 44736 */
842 for(i=0;i<257;i++) {
843 int v;
844 v = mpa_enwindow[i];
845#if WFRAC_BITS < 16
846 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
847#endif
848 window[i] = v;
849 if ((i & 63) != 0)
850 v = -v;
851 if (i != 0)
852 window[512 - i] = v;
853 }
854}
239c2f4c
FB
855
856/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
857 32 samples. */
858/* XXX: optimize by avoiding ring buffer usage */
bf1f4da0 859void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
093c6e50 860 MPA_INT *window, int *dither_state,
a3a5f4d6 861 OUT_INT *samples, int incr,
0c1a9eda 862 int32_t sb_samples[SBLIMIT])
239c2f4c 863{
0c1a9eda 864 int32_t tmp[32];
8c5b5683 865 register MPA_INT *synth_buf;
491c4a10 866 register const MPA_INT *w, *w2, *p;
239c2f4c 867 int j, offset, v;
a3a5f4d6 868 OUT_INT *samples2;
239c2f4c 869#if FRAC_BITS <= 15
8c5b5683 870 int sum, sum2;
239c2f4c 871#else
8c5b5683 872 int64_t sum, sum2;
239c2f4c 873#endif
bf1f4da0 874
239c2f4c
FB
875 dct32(tmp, sb_samples);
876
bf1f4da0
AB
877 offset = *synth_buf_offset;
878 synth_buf = synth_buf_ptr + offset;
239c2f4c
FB
879
880 for(j=0;j<32;j++) {
881 v = tmp[j];
882#if FRAC_BITS <= 15
81552334
FB
883 /* NOTE: can cause a loss in precision if very high amplitude
884 sound */
239c2f4c
FB
885 if (v > 32767)
886 v = 32767;
887 else if (v < -32768)
888 v = -32768;
889#endif
890 synth_buf[j] = v;
891 }
892 /* copy to avoid wrap */
893 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
894
8c5b5683 895 samples2 = samples + 31 * incr;
239c2f4c 896 w = window;
8c5b5683
FB
897 w2 = window + 31;
898
093c6e50 899 sum = *dither_state;
8c5b5683
FB
900 p = synth_buf + 16;
901 SUM8(sum, +=, w, p);
902 p = synth_buf + 48;
903 SUM8(sum, -=, w + 32, p);
a7a85899 904 *samples = round_sample(&sum);
8c5b5683 905 samples += incr;
239c2f4c
FB
906 w++;
907
8c5b5683
FB
908 /* we calculate two samples at the same time to avoid one memory
909 access per two sample */
910 for(j=1;j<16;j++) {
8c5b5683
FB
911 sum2 = 0;
912 p = synth_buf + 16 + j;
913 SUM8P2(sum, +=, sum2, -=, w, w2, p);
914 p = synth_buf + 48 - j;
915 SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
916
a7a85899 917 *samples = round_sample(&sum);
8c5b5683 918 samples += incr;
a7a85899
MN
919 sum += sum2;
920 *samples2 = round_sample(&sum);
8c5b5683 921 samples2 -= incr;
239c2f4c 922 w++;
8c5b5683 923 w2--;
239c2f4c 924 }
8c5b5683
FB
925
926 p = synth_buf + 32;
8c5b5683 927 SUM8(sum, -=, w + 32, p);
a7a85899 928 *samples = round_sample(&sum);
093c6e50 929 *dither_state= sum;
8c5b5683 930
239c2f4c 931 offset = (offset - 32) & 511;
bf1f4da0 932 *synth_buf_offset = offset;
239c2f4c
FB
933}
934
935/* cos(pi*i/24) */
936#define C1 FIXR(0.99144486137381041114)
937#define C3 FIXR(0.92387953251128675612)
938#define C5 FIXR(0.79335334029123516458)
939#define C7 FIXR(0.60876142900872063941)
940#define C9 FIXR(0.38268343236508977173)
941#define C11 FIXR(0.13052619222005159154)
942
943/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
944 cases. */
945static void imdct12(int *out, int *in)
946{
947 int tmp;
0c1a9eda 948 int64_t in1_3, in1_9, in4_3, in4_9;
239c2f4c
FB
949
950 in1_3 = MUL64(in[1], C3);
951 in1_9 = MUL64(in[1], C9);
952 in4_3 = MUL64(in[4], C3);
953 in4_9 = MUL64(in[4], C9);
954
955 tmp = FRAC_RND(MUL64(in[0], C7) - in1_3 - MUL64(in[2], C11) +
956 MUL64(in[3], C1) - in4_9 - MUL64(in[5], C5));
957 out[0] = tmp;
958 out[5] = -tmp;
959 tmp = FRAC_RND(MUL64(in[0] - in[3], C9) - in1_3 +
960 MUL64(in[2] + in[5], C3) - in4_9);
961 out[1] = tmp;
962 out[4] = -tmp;
963 tmp = FRAC_RND(MUL64(in[0], C11) - in1_9 + MUL64(in[2], C7) -
964 MUL64(in[3], C5) + in4_3 - MUL64(in[5], C1));
965 out[2] = tmp;
966 out[3] = -tmp;
967 tmp = FRAC_RND(MUL64(-in[0], C5) + in1_9 + MUL64(in[2], C1) +
968 MUL64(in[3], C11) - in4_3 - MUL64(in[5], C7));
969 out[6] = tmp;
970 out[11] = tmp;
971 tmp = FRAC_RND(MUL64(-in[0] + in[3], C3) - in1_9 +
972 MUL64(in[2] + in[5], C9) + in4_3);
973 out[7] = tmp;
974 out[10] = tmp;
975 tmp = FRAC_RND(-MUL64(in[0], C1) - in1_3 - MUL64(in[2], C5) -
976 MUL64(in[3], C7) - in4_9 - MUL64(in[5], C11));
977 out[8] = tmp;
978 out[9] = tmp;
979}
980
981#undef C1
982#undef C3
983#undef C5
984#undef C7
985#undef C9
986#undef C11
987
988/* cos(pi*i/18) */
711ae726
MN
989#define C1 FIXHR(0.98480775301220805936/2)
990#define C2 FIXHR(0.93969262078590838405/2)
991#define C3 FIXHR(0.86602540378443864676/2)
992#define C4 FIXHR(0.76604444311897803520/2)
993#define C5 FIXHR(0.64278760968653932632/2)
994#define C6 FIXHR(0.5/2)
995#define C7 FIXHR(0.34202014332566873304/2)
996#define C8 FIXHR(0.17364817766693034885/2)
997
239c2f4c
FB
998
999/* 0.5 / cos(pi*(2*i+1)/36) */
1000static const int icos36[9] = {
1001 FIXR(0.50190991877167369479),
1002 FIXR(0.51763809020504152469),
1003 FIXR(0.55168895948124587824),
1004 FIXR(0.61038729438072803416),
1005 FIXR(0.70710678118654752439),
1006 FIXR(0.87172339781054900991),
1007 FIXR(1.18310079157624925896),
1008 FIXR(1.93185165257813657349),
1009 FIXR(5.73685662283492756461),
1010};
239c2f4c 1011/* using Lee like decomposition followed by hand coded 9 points DCT */
711ae726 1012static void imdct36(int *out, int *buf, int *in, int *win)
239c2f4c
FB
1013{
1014 int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1015 int tmp[18], *tmp1, *in1;
239c2f4c
FB
1016
1017 for(i=17;i>=1;i--)
1018 in[i] += in[i-1];
1019 for(i=17;i>=3;i-=2)
1020 in[i] += in[i-2];
1021
1022 for(j=0;j<2;j++) {
1023 tmp1 = tmp + j;
1024 in1 = in + j;
711ae726
MN
1025#if 0
1026//more accurate but slower
1027 int64_t t0, t1, t2, t3;
1028 t2 = in1[2*4] + in1[2*8] - in1[2*2];
1029
1030 t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1031 t1 = in1[2*0] - in1[2*6];
1032 tmp1[ 6] = t1 - (t2>>1);
1033 tmp1[16] = t1 + t2;
1034
1035 t0 = MUL64(2*(in1[2*2] + in1[2*4]), C2);
1036 t1 = MUL64( in1[2*4] - in1[2*8] , -2*C8);
1037 t2 = MUL64(2*(in1[2*2] + in1[2*8]), -C4);
1038
1039 tmp1[10] = (t3 - t0 - t2) >> 32;
1040 tmp1[ 2] = (t3 + t0 + t1) >> 32;
1041 tmp1[14] = (t3 + t2 - t1) >> 32;
1042
1043 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1044 t2 = MUL64(2*(in1[2*1] + in1[2*5]), C1);
1045 t3 = MUL64( in1[2*5] - in1[2*7] , -2*C7);
1046 t0 = MUL64(2*in1[2*3], C3);
1047
1048 t1 = MUL64(2*(in1[2*1] + in1[2*7]), -C5);
1049
1050 tmp1[ 0] = (t2 + t3 + t0) >> 32;
1051 tmp1[12] = (t2 + t1 - t0) >> 32;
1052 tmp1[ 8] = (t3 - t1 - t0) >> 32;
1053#else
1054 t2 = in1[2*4] + in1[2*8] - in1[2*2];
1055
1056 t3 = in1[2*0] + (in1[2*6]>>1);
1057 t1 = in1[2*0] - in1[2*6];
1058 tmp1[ 6] = t1 - (t2>>1);
1059 tmp1[16] = t1 + t2;
1060
1061 t0 = MULH(2*(in1[2*2] + in1[2*4]), C2);
1062 t1 = MULH( in1[2*4] - in1[2*8] , -2*C8);
1063 t2 = MULH(2*(in1[2*2] + in1[2*8]), -C4);
1064
1065 tmp1[10] = t3 - t0 - t2;
1066 tmp1[ 2] = t3 + t0 + t1;
1067 tmp1[14] = t3 + t2 - t1;
1068
1069 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1070 t2 = MULH(2*(in1[2*1] + in1[2*5]), C1);
1071 t3 = MULH( in1[2*5] - in1[2*7] , -2*C7);
1072 t0 = MULH(2*in1[2*3], C3);
239c2f4c 1073
711ae726
MN
1074 t1 = MULH(2*(in1[2*1] + in1[2*7]), -C5);
1075
1076 tmp1[ 0] = t2 + t3 + t0;
1077 tmp1[12] = t2 + t1 - t0;
1078 tmp1[ 8] = t3 - t1 - t0;
1079#endif
239c2f4c
FB
1080 }
1081
1082 i = 0;
1083 for(j=0;j<4;j++) {
1084 t0 = tmp[i];
1085 t1 = tmp[i + 2];
1086 s0 = t1 + t0;
1087 s2 = t1 - t0;
1088
1089 t2 = tmp[i + 1];
1090 t3 = tmp[i + 3];
1091 s1 = MULL(t3 + t2, icos36[j]);
1092 s3 = MULL(t3 - t2, icos36[8 - j]);
1093
711ae726
MN
1094 t0 = (s0 + s1) << 5;
1095 t1 = (s0 - s1) << 5;
1096 out[(9 + j)*SBLIMIT] = -MULH(t1, win[9 + j]) + buf[9 + j];
1097 out[(8 - j)*SBLIMIT] = MULH(t1, win[8 - j]) + buf[8 - j];
1098 buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1099 buf[8 - j] = MULH(t0, win[18 + 8 - j]);
239c2f4c 1100
711ae726
MN
1101 t0 = (s2 + s3) << 5;
1102 t1 = (s2 - s3) << 5;
1103 out[(9 + 8 - j)*SBLIMIT] = -MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
1104 out[( j)*SBLIMIT] = MULH(t1, win[ j]) + buf[ j];
1105 buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1106 buf[ + j] = MULH(t0, win[18 + j]);
239c2f4c
FB
1107 i += 4;
1108 }
1109
1110 s0 = tmp[16];
1111 s1 = MULL(tmp[17], icos36[4]);
711ae726
MN
1112 t0 = (s0 + s1) << 5;
1113 t1 = (s0 - s1) << 5;
1114 out[(9 + 4)*SBLIMIT] = -MULH(t1, win[9 + 4]) + buf[9 + 4];
1115 out[(8 - 4)*SBLIMIT] = MULH(t1, win[8 - 4]) + buf[8 - 4];
1116 buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1117 buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
239c2f4c
FB
1118}
1119
de6d9b64 1120/* header decoding. MUST check the header before because no
239c2f4c
FB
1121 consistency check is done there. Return 1 if free format found and
1122 that the frame size must be computed externally */
0c1a9eda 1123static int decode_header(MPADecodeContext *s, uint32_t header)
de6d9b64 1124{
239c2f4c
FB
1125 int sample_rate, frame_size, mpeg25, padding;
1126 int sample_rate_index, bitrate_index;
de6d9b64 1127 if (header & (1<<20)) {
239c2f4c
FB
1128 s->lsf = (header & (1<<19)) ? 0 : 1;
1129 mpeg25 = 0;
de6d9b64 1130 } else {
239c2f4c
FB
1131 s->lsf = 1;
1132 mpeg25 = 1;
de6d9b64
FB
1133 }
1134
1135 s->layer = 4 - ((header >> 17) & 3);
1136 /* extract frequency */
239c2f4c
FB
1137 sample_rate_index = (header >> 10) & 3;
1138 sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1139 sample_rate_index += 3 * (s->lsf + mpeg25);
1140 s->sample_rate_index = sample_rate_index;
1141 s->error_protection = ((header >> 16) & 1) ^ 1;
81552334 1142 s->sample_rate = sample_rate;
de6d9b64 1143
239c2f4c
FB
1144 bitrate_index = (header >> 12) & 0xf;
1145 padding = (header >> 9) & 1;
1146 //extension = (header >> 8) & 1;
1147 s->mode = (header >> 6) & 3;
1148 s->mode_ext = (header >> 4) & 3;
1149 //copyright = (header >> 3) & 1;
1150 //original = (header >> 2) & 1;
1151 //emphasis = header & 3;
de6d9b64 1152
239c2f4c
FB
1153 if (s->mode == MPA_MONO)
1154 s->nb_channels = 1;
1155 else
1156 s->nb_channels = 2;
de6d9b64 1157
239c2f4c
FB
1158 if (bitrate_index != 0) {
1159 frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1160 s->bit_rate = frame_size * 1000;
1161 switch(s->layer) {
1162 case 1:
1163 frame_size = (frame_size * 12000) / sample_rate;
1164 frame_size = (frame_size + padding) * 4;
1165 break;
1166 case 2:
1167 frame_size = (frame_size * 144000) / sample_rate;
1168 frame_size += padding;
1169 break;
1170 default:
1171 case 3:
1172 frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1173 frame_size += padding;
1174 break;
1175 }
1176 s->frame_size = frame_size;
1177 } else {
1178 /* if no frame size computed, signal it */
1179 if (!s->free_format_frame_size)
1180 return 1;
1181 /* free format: compute bitrate and real frame size from the
1182 frame size we extracted by reading the bitstream */
1183 s->frame_size = s->free_format_frame_size;
1184 switch(s->layer) {
1185 case 1:
1186 s->frame_size += padding * 4;
1187 s->bit_rate = (s->frame_size * sample_rate) / 48000;
1188 break;
1189 case 2:
1190 s->frame_size += padding;
1191 s->bit_rate = (s->frame_size * sample_rate) / 144000;
1192 break;
1193 default:
1194 case 3:
1195 s->frame_size += padding;
1196 s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1197 break;
1198 }
de6d9b64 1199 }
239c2f4c 1200
fad9f495 1201#if defined(DEBUG)
239c2f4c
FB
1202 printf("layer%d, %d Hz, %d kbits/s, ",
1203 s->layer, s->sample_rate, s->bit_rate);
1204 if (s->nb_channels == 2) {
1205 if (s->layer == 3) {
1206 if (s->mode_ext & MODE_EXT_MS_STEREO)
1207 printf("ms-");
1208 if (s->mode_ext & MODE_EXT_I_STEREO)
1209 printf("i-");
1210 }
1211 printf("stereo");
1212 } else {
1213 printf("mono");
1214 }
1215 printf("\n");
de6d9b64 1216#endif
239c2f4c 1217 return 0;
de6d9b64
FB
1218}
1219
8c5b5683 1220/* useful helper to get mpeg audio stream infos. Return -1 if error in
962d6ae6
FB
1221 header, otherwise the coded frame size in bytes */
1222int mpa_decode_header(AVCodecContext *avctx, uint32_t head)
8c5b5683
FB
1223{
1224 MPADecodeContext s1, *s = &s1;
2caa92d9 1225 memset( s, 0, sizeof(MPADecodeContext) );
8c5b5683 1226
a7a85899 1227 if (ff_mpa_check_header(head) != 0)
8c5b5683
FB
1228 return -1;
1229
1230 if (decode_header(s, head) != 0) {
1231 return -1;
1232 }
1233
1234 switch(s->layer) {
1235 case 1:
962d6ae6 1236 avctx->frame_size = 384;
8c5b5683
FB
1237 break;
1238 case 2:
962d6ae6 1239 avctx->frame_size = 1152;
8c5b5683
FB
1240 break;
1241 default:
1242 case 3:
1243 if (s->lsf)
962d6ae6 1244 avctx->frame_size = 576;
8c5b5683 1245 else
962d6ae6 1246 avctx->frame_size = 1152;
8c5b5683
FB
1247 break;
1248 }
1249
962d6ae6
FB
1250 avctx->sample_rate = s->sample_rate;
1251 avctx->channels = s->nb_channels;
1252 avctx->bit_rate = s->bit_rate;
1253 avctx->sub_id = s->layer;
1254 return s->frame_size;
8c5b5683
FB
1255}
1256
239c2f4c
FB
1257/* return the number of decoded frames */
1258static int mp_decode_layer1(MPADecodeContext *s)
de6d9b64 1259{
239c2f4c 1260 int bound, i, v, n, ch, j, mant;
0c1a9eda
ZK
1261 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1262 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
239c2f4c
FB
1263
1264 if (s->mode == MPA_JSTEREO)
1265 bound = (s->mode_ext + 1) * 4;
1266 else
1267 bound = SBLIMIT;
1268
1269 /* allocation bits */
1270 for(i=0;i<bound;i++) {
1271 for(ch=0;ch<s->nb_channels;ch++) {
1272 allocation[ch][i] = get_bits(&s->gb, 4);
1273 }
1274 }
1275 for(i=bound;i<SBLIMIT;i++) {
1276 allocation[0][i] = get_bits(&s->gb, 4);
1277 }
1278
1279 /* scale factors */
1280 for(i=0;i<bound;i++) {
1281 for(ch=0;ch<s->nb_channels;ch++) {
1282 if (allocation[ch][i])
1283 scale_factors[ch][i] = get_bits(&s->gb, 6);
1284 }
1285 }
1286 for(i=bound;i<SBLIMIT;i++) {
1287 if (allocation[0][i]) {
1288 scale_factors[0][i] = get_bits(&s->gb, 6);
1289 scale_factors[1][i] = get_bits(&s->gb, 6);
1290 }
1291 }
de6d9b64 1292
239c2f4c
FB
1293 /* compute samples */
1294 for(j=0;j<12;j++) {
1295 for(i=0;i<bound;i++) {
1296 for(ch=0;ch<s->nb_channels;ch++) {
1297 n = allocation[ch][i];
1298 if (n) {
1299 mant = get_bits(&s->gb, n + 1);
1300 v = l1_unscale(n, mant, scale_factors[ch][i]);
1301 } else {
1302 v = 0;
1303 }
1304 s->sb_samples[ch][j][i] = v;
1305 }
1306 }
1307 for(i=bound;i<SBLIMIT;i++) {
1308 n = allocation[0][i];
1309 if (n) {
1310 mant = get_bits(&s->gb, n + 1);
1311 v = l1_unscale(n, mant, scale_factors[0][i]);
1312 s->sb_samples[0][j][i] = v;
1313 v = l1_unscale(n, mant, scale_factors[1][i]);
1314 s->sb_samples[1][j][i] = v;
1315 } else {
1316 s->sb_samples[0][j][i] = 0;
1317 s->sb_samples[1][j][i] = 0;
1318 }
1319 }
1320 }
1321 return 12;
1322}
1323
1324/* bitrate is in kb/s */
1325int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1326{
1327 int ch_bitrate, table;
de6d9b64 1328
239c2f4c
FB
1329 ch_bitrate = bitrate / nb_channels;
1330 if (!lsf) {
1331 if ((freq == 48000 && ch_bitrate >= 56) ||
1332 (ch_bitrate >= 56 && ch_bitrate <= 80))
1333 table = 0;
1334 else if (freq != 48000 && ch_bitrate >= 96)
1335 table = 1;
1336 else if (freq != 32000 && ch_bitrate <= 48)
1337 table = 2;
1338 else
1339 table = 3;
1340 } else {
1341 table = 4;
1342 }
1343 return table;
1344}
de6d9b64 1345
239c2f4c
FB
1346static int mp_decode_layer2(MPADecodeContext *s)
1347{
1348 int sblimit; /* number of used subbands */
1349 const unsigned char *alloc_table;
1350 int table, bit_alloc_bits, i, j, ch, bound, v;
1351 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1352 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1353 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1354 int scale, qindex, bits, steps, k, l, m, b;
de6d9b64 1355
239c2f4c
FB
1356 /* select decoding table */
1357 table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
1358 s->sample_rate, s->lsf);
1359 sblimit = sblimit_table[table];
1360 alloc_table = alloc_tables[table];
1361
1362 if (s->mode == MPA_JSTEREO)
1363 bound = (s->mode_ext + 1) * 4;
1364 else
1365 bound = sblimit;
1366
1367 dprintf("bound=%d sblimit=%d\n", bound, sblimit);
2caa92d9
MN
1368
1369 /* sanity check */
1370 if( bound > sblimit ) bound = sblimit;
1371
239c2f4c
FB
1372 /* parse bit allocation */
1373 j = 0;
1374 for(i=0;i<bound;i++) {
1375 bit_alloc_bits = alloc_table[j];
1376 for(ch=0;ch<s->nb_channels;ch++) {
1377 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1378 }
1379 j += 1 << bit_alloc_bits;
1380 }
1381 for(i=bound;i<sblimit;i++) {
1382 bit_alloc_bits = alloc_table[j];
1383 v = get_bits(&s->gb, bit_alloc_bits);
1384 bit_alloc[0][i] = v;
1385 bit_alloc[1][i] = v;
1386 j += 1 << bit_alloc_bits;
de6d9b64 1387 }
239c2f4c
FB
1388
1389#ifdef DEBUG
1390 {
1391 for(ch=0;ch<s->nb_channels;ch++) {
1392 for(i=0;i<sblimit;i++)
1393 printf(" %d", bit_alloc[ch][i]);
1394 printf("\n");
1395 }
1396 }
1397#endif
1398
1399 /* scale codes */
1400 for(i=0;i<sblimit;i++) {
1401 for(ch=0;ch<s->nb_channels;ch++) {
1402 if (bit_alloc[ch][i])
1403 scale_code[ch][i] = get_bits(&s->gb, 2);
1404 }
1405 }
1406
1407 /* scale factors */
1408 for(i=0;i<sblimit;i++) {
1409 for(ch=0;ch<s->nb_channels;ch++) {
1410 if (bit_alloc[ch][i]) {
1411 sf = scale_factors[ch][i];
1412 switch(scale_code[ch][i]) {
1413 default:
1414 case 0:
1415 sf[0] = get_bits(&s->gb, 6);
1416 sf[1] = get_bits(&s->gb, 6);
1417 sf[2] = get_bits(&s->gb, 6);
1418 break;
1419 case 2:
1420 sf[0] = get_bits(&s->gb, 6);
1421 sf[1] = sf[0];
1422 sf[2] = sf[0];
1423 break;
1424 case 1:
1425 sf[0] = get_bits(&s->gb, 6);
1426 sf[2] = get_bits(&s->gb, 6);
1427 sf[1] = sf[0];
1428 break;
1429 case 3:
1430 sf[0] = get_bits(&s->gb, 6);
1431 sf[2] = get_bits(&s->gb, 6);
1432 sf[1] = sf[2];
1433 break;
1434 }
1435 }
1436 }
1437 }
1438
1439#ifdef DEBUG
1440 for(ch=0;ch<s->nb_channels;ch++) {
1441 for(i=0;i<sblimit;i++) {
1442 if (bit_alloc[ch][i]) {
1443 sf = scale_factors[ch][i];
1444 printf(" %d %d %d", sf[0], sf[1], sf[2]);
1445 } else {
1446 printf(" -");
1447 }
1448 }
1449 printf("\n");
1450 }
1451#endif
1452
1453 /* samples */
1454 for(k=0;k<3;k++) {
1455 for(l=0;l<12;l+=3) {
1456 j = 0;
1457 for(i=0;i<bound;i++) {
1458 bit_alloc_bits = alloc_table[j];
1459 for(ch=0;ch<s->nb_channels;ch++) {
1460 b = bit_alloc[ch][i];
1461 if (b) {
1462 scale = scale_factors[ch][i][k];
1463 qindex = alloc_table[j+b];
1464 bits = quant_bits[qindex];
1465 if (bits < 0) {
1466 /* 3 values at the same time */
1467 v = get_bits(&s->gb, -bits);
1468 steps = quant_steps[qindex];
1469 s->sb_samples[ch][k * 12 + l + 0][i] =
1470 l2_unscale_group(steps, v % steps, scale);
1471 v = v / steps;
1472 s->sb_samples[ch][k * 12 + l + 1][i] =
1473 l2_unscale_group(steps, v % steps, scale);
1474 v = v / steps;
1475 s->sb_samples[ch][k * 12 + l + 2][i] =
1476 l2_unscale_group(steps, v, scale);
1477 } else {
1478 for(m=0;m<3;m++) {
1479 v = get_bits(&s->gb, bits);
1480 v = l1_unscale(bits - 1, v, scale);
1481 s->sb_samples[ch][k * 12 + l + m][i] = v;
1482 }
1483 }
1484 } else {
1485 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1486 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1487 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1488 }
1489 }
1490 /* next subband in alloc table */
1491 j += 1 << bit_alloc_bits;
1492 }
1493 /* XXX: find a way to avoid this duplication of code */
1494 for(i=bound;i<sblimit;i++) {
1495 bit_alloc_bits = alloc_table[j];
1496 b = bit_alloc[0][i];
1497 if (b) {
1498 int mant, scale0, scale1;
1499 scale0 = scale_factors[0][i][k];
1500 scale1 = scale_factors[1][i][k];
1501 qindex = alloc_table[j+b];
1502 bits = quant_bits[qindex];
1503 if (bits < 0) {
1504 /* 3 values at the same time */
1505 v = get_bits(&s->gb, -bits);
1506 steps = quant_steps[qindex];
1507 mant = v % steps;
1508 v = v / steps;
1509 s->sb_samples[0][k * 12 + l + 0][i] =
1510 l2_unscale_group(steps, mant, scale0);
1511 s->sb_samples[1][k * 12 + l + 0][i] =
1512 l2_unscale_group(steps, mant, scale1);
1513 mant = v % steps;
1514 v = v / steps;
1515 s->sb_samples[0][k * 12 + l + 1][i] =
1516 l2_unscale_group(steps, mant, scale0);
1517 s->sb_samples[1][k * 12 + l + 1][i] =
1518 l2_unscale_group(steps, mant, scale1);
1519 s->sb_samples[0][k * 12 + l + 2][i] =
1520 l2_unscale_group(steps, v, scale0);
1521 s->sb_samples[1][k * 12 + l + 2][i] =
1522 l2_unscale_group(steps, v, scale1);
1523 } else {
1524 for(m=0;m<3;m++) {
1525 mant = get_bits(&s->gb, bits);
1526 s->sb_samples[0][k * 12 + l + m][i] =
1527 l1_unscale(bits - 1, mant, scale0);
1528 s->sb_samples[1][k * 12 + l + m][i] =
1529 l1_unscale(bits - 1, mant, scale1);
1530 }
1531 }
1532 } else {
1533 s->sb_samples[0][k * 12 + l + 0][i] = 0;
1534 s->sb_samples[0][k * 12 + l + 1][i] = 0;
1535 s->sb_samples[0][k * 12 + l + 2][i] = 0;
1536 s->sb_samples[1][k * 12 + l + 0][i] = 0;
1537 s->sb_samples[1][k * 12 + l + 1][i] = 0;
1538 s->sb_samples[1][k * 12 + l + 2][i] = 0;
1539 }
1540 /* next subband in alloc table */
1541 j += 1 << bit_alloc_bits;
1542 }
1543 /* fill remaining samples to zero */
1544 for(i=sblimit;i<SBLIMIT;i++) {
1545 for(ch=0;ch<s->nb_channels;ch++) {
1546 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1547 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1548 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1549 }
1550 }
1551 }
1552 }
1553 return 3 * 12;
de6d9b64
FB
1554}
1555
1556/*
239c2f4c 1557 * Seek back in the stream for backstep bytes (at most 511 bytes)
de6d9b64 1558 */
5c91a675 1559static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)
de6d9b64 1560{
0c1a9eda 1561 uint8_t *ptr;
de6d9b64
FB
1562
1563 /* compute current position in stream */
228ef9dd 1564 ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));
8db1a1dd 1565
de6d9b64
FB
1566 /* copy old data before current one */
1567 ptr -= backstep;
239c2f4c
FB
1568 memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
1569 BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
de6d9b64 1570 /* init get bits again */
68f593b4 1571 init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);
de6d9b64 1572
239c2f4c
FB
1573 /* prepare next buffer */
1574 s->inbuf_index ^= 1;
1575 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1576 s->old_frame_size = s->frame_size;
1577}
1578
1579static inline void lsf_sf_expand(int *slen,
1580 int sf, int n1, int n2, int n3)
1581{
1582 if (n3) {
1583 slen[3] = sf % n3;
1584 sf /= n3;
1585 } else {
1586 slen[3] = 0;
1587 }
1588 if (n2) {
1589 slen[2] = sf % n2;
1590 sf /= n2;
1591 } else {
1592 slen[2] = 0;
1593 }
1594 slen[1] = sf % n1;
1595 sf /= n1;
1596 slen[0] = sf;
1597}
1598
1599static void exponents_from_scale_factors(MPADecodeContext *s,
1600 GranuleDef *g,
0c1a9eda 1601 int16_t *exponents)
239c2f4c 1602{
0c1a9eda 1603 const uint8_t *bstab, *pretab;
239c2f4c 1604 int len, i, j, k, l, v0, shift, gain, gains[3];
0c1a9eda 1605 int16_t *exp_ptr;
239c2f4c
FB
1606
1607 exp_ptr = exponents;
1608 gain = g->global_gain - 210;
1609 shift = g->scalefac_scale + 1;
1610
1611 bstab = band_size_long[s->sample_rate_index];
1612 pretab = mpa_pretab[g->preflag];
1613 for(i=0;i<g->long_end;i++) {
1614 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
1615 len = bstab[i];
1616 for(j=len;j>0;j--)
1617 *exp_ptr++ = v0;
1618 }
1619
1620 if (g->short_start < 13) {
1621 bstab = band_size_short[s->sample_rate_index];
1622 gains[0] = gain - (g->subblock_gain[0] << 3);
1623 gains[1] = gain - (g->subblock_gain[1] << 3);
1624 gains[2] = gain - (g->subblock_gain[2] << 3);
1625 k = g->long_end;
1626 for(i=g->short_start;i<13;i++) {
1627 len = bstab[i];
1628 for(l=0;l<3;l++) {
1629 v0 = gains[l] - (g->scale_factors[k++] << shift);
1630 for(j=len;j>0;j--)
1631 *exp_ptr++ = v0;
1632 }
1633 }
1634 }
1635}
1636
1637/* handle n = 0 too */
1638static inline int get_bitsz(GetBitContext *s, int n)
1639{
1640 if (n == 0)
1641 return 0;
1642 else
1643 return get_bits(s, n);
1644}
1645
1646static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
0c1a9eda 1647 int16_t *exponents, int end_pos)
239c2f4c
FB
1648{
1649 int s_index;
1650 int linbits, code, x, y, l, v, i, j, k, pos;
8db1a1dd 1651 GetBitContext last_gb;
239c2f4c 1652 VLC *vlc;
0c1a9eda 1653 uint8_t *code_table;
239c2f4c
FB
1654
1655 /* low frequencies (called big values) */
1656 s_index = 0;
1657 for(i=0;i<3;i++) {
1658 j = g->region_size[i];
1659 if (j == 0)
1660 continue;
1661 /* select vlc table */
1662 k = g->table_select[i];
1663 l = mpa_huff_data[k][0];
1664 linbits = mpa_huff_data[k][1];
1665 vlc = &huff_vlc[l];
1666 code_table = huff_code_table[l];
1667
1668 /* read huffcode and compute each couple */
1669 for(;j>0;j--) {
1670 if (get_bits_count(&s->gb) >= end_pos)
1671 break;
1672 if (code_table) {
1673 code = get_vlc(&s->gb, vlc);
1674 if (code < 0)
1675 return -1;
1676 y = code_table[code];
1677 x = y >> 4;
1678 y = y & 0x0f;
1679 } else {
1680 x = 0;
1681 y = 0;
1682 }
1683 dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
1684 i, g->region_size[i] - j, x, y, exponents[s_index]);
1685 if (x) {
1686 if (x == 15)
1687 x += get_bitsz(&s->gb, linbits);
1688 v = l3_unscale(x, exponents[s_index]);
1689 if (get_bits1(&s->gb))
1690 v = -v;
1691 } else {
1692 v = 0;
1693 }
1694 g->sb_hybrid[s_index++] = v;
1695 if (y) {
1696 if (y == 15)
1697 y += get_bitsz(&s->gb, linbits);
1698 v = l3_unscale(y, exponents[s_index]);
1699 if (get_bits1(&s->gb))
1700 v = -v;
1701 } else {
1702 v = 0;
1703 }
1704 g->sb_hybrid[s_index++] = v;
1705 }
1706 }
1707
1708 /* high frequencies */
1709 vlc = &huff_quad_vlc[g->count1table_select];
8db1a1dd 1710 last_gb.buffer = NULL;
239c2f4c
FB
1711 while (s_index <= 572) {
1712 pos = get_bits_count(&s->gb);
1713 if (pos >= end_pos) {
8db1a1dd 1714 if (pos > end_pos && last_gb.buffer != NULL) {
239c2f4c
FB
1715 /* some encoders generate an incorrect size for this
1716 part. We must go back into the data */
1717 s_index -= 4;
8db1a1dd 1718 s->gb = last_gb;
239c2f4c
FB
1719 }
1720 break;
1721 }
8db1a1dd
MN
1722 last_gb= s->gb;
1723
239c2f4c
FB
1724 code = get_vlc(&s->gb, vlc);
1725 dprintf("t=%d code=%d\n", g->count1table_select, code);
1726 if (code < 0)
1727 return -1;
1728 for(i=0;i<4;i++) {
1729 if (code & (8 >> i)) {
1730 /* non zero value. Could use a hand coded function for
1731 'one' value */
1732 v = l3_unscale(1, exponents[s_index]);
1733 if(get_bits1(&s->gb))
1734 v = -v;
1735 } else {
1736 v = 0;
1737 }
1738 g->sb_hybrid[s_index++] = v;
1739 }
1740 }
1741 while (s_index < 576)
1742 g->sb_hybrid[s_index++] = 0;
de6d9b64
FB
1743 return 0;
1744}
1745
239c2f4c
FB
1746/* Reorder short blocks from bitstream order to interleaved order. It
1747 would be faster to do it in parsing, but the code would be far more
1748 complicated */
1749static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1750{
1751 int i, j, k, len;
0c1a9eda
ZK
1752 int32_t *ptr, *dst, *ptr1;
1753 int32_t tmp[576];
239c2f4c
FB
1754
1755 if (g->block_type != 2)
1756 return;
1757
1758 if (g->switch_point) {
1759 if (s->sample_rate_index != 8) {
1760 ptr = g->sb_hybrid + 36;
1761 } else {
1762 ptr = g->sb_hybrid + 48;
1763 }
1764 } else {
1765 ptr = g->sb_hybrid;
1766 }
1767
1768 for(i=g->short_start;i<13;i++) {
1769 len = band_size_short[s->sample_rate_index][i];
1770 ptr1 = ptr;
1771 for(k=0;k<3;k++) {
1772 dst = tmp + k;
1773 for(j=len;j>0;j--) {
1774 *dst = *ptr++;
1775 dst += 3;
1776 }
1777 }
0c1a9eda 1778 memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));
239c2f4c
FB
1779 }
1780}
1781
1782#define ISQRT2 FIXR(0.70710678118654752440)
1783
1784static void compute_stereo(MPADecodeContext *s,
1785 GranuleDef *g0, GranuleDef *g1)
1786{
1787 int i, j, k, l;
0c1a9eda 1788 int32_t v1, v2;
239c2f4c 1789 int sf_max, tmp0, tmp1, sf, len, non_zero_found;
0c1a9eda
ZK
1790 int32_t (*is_tab)[16];
1791 int32_t *tab0, *tab1;
239c2f4c
FB
1792 int non_zero_found_short[3];
1793
1794 /* intensity stereo */
1795 if (s->mode_ext & MODE_EXT_I_STEREO) {
1796 if (!s->lsf) {
1797 is_tab = is_table;
1798 sf_max = 7;
1799 } else {
1800 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1801 sf_max = 16;
1802 }
1803
1804 tab0 = g0->sb_hybrid + 576;
1805 tab1 = g1->sb_hybrid + 576;
1806
1807 non_zero_found_short[0] = 0;
1808 non_zero_found_short[1] = 0;
1809 non_zero_found_short[2] = 0;
1810 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1811 for(i = 12;i >= g1->short_start;i--) {
1812 /* for last band, use previous scale factor */
1813 if (i != 11)
1814 k -= 3;
1815 len = band_size_short[s->sample_rate_index][i];
1816 for(l=2;l>=0;l--) {
1817 tab0 -= len;
1818 tab1 -= len;
1819 if (!non_zero_found_short[l]) {
1820 /* test if non zero band. if so, stop doing i-stereo */
1821 for(j=0;j<len;j++) {
1822 if (tab1[j] != 0) {
1823 non_zero_found_short[l] = 1;
1824 goto found1;
1825 }
1826 }
1827 sf = g1->scale_factors[k + l];
1828 if (sf >= sf_max)
1829 goto found1;
1830
1831 v1 = is_tab[0][sf];
1832 v2 = is_tab[1][sf];
1833 for(j=0;j<len;j++) {
1834 tmp0 = tab0[j];
1835 tab0[j] = MULL(tmp0, v1);
1836 tab1[j] = MULL(tmp0, v2);
1837 }
1838 } else {
1839 found1:
1840 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1841 /* lower part of the spectrum : do ms stereo
1842 if enabled */
1843 for(j=0;j<len;j++) {
1844 tmp0 = tab0[j];
1845 tmp1 = tab1[j];
1846 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1847 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1848 }
1849 }
1850 }
1851 }
1852 }
1853
1854 non_zero_found = non_zero_found_short[0] |
1855 non_zero_found_short[1] |
1856 non_zero_found_short[2];
1857
1858 for(i = g1->long_end - 1;i >= 0;i--) {
1859 len = band_size_long[s->sample_rate_index][i];
1860 tab0 -= len;
1861 tab1 -= len;
1862 /* test if non zero band. if so, stop doing i-stereo */
1863 if (!non_zero_found) {
1864 for(j=0;j<len;j++) {
1865 if (tab1[j] != 0) {
1866 non_zero_found = 1;
1867 goto found2;
1868 }
1869 }
1870 /* for last band, use previous scale factor */
1871 k = (i == 21) ? 20 : i;
1872 sf = g1->scale_factors[k];
1873 if (sf >= sf_max)
1874 goto found2;
1875 v1 = is_tab[0][sf];
1876 v2 = is_tab[1][sf];
1877 for(j=0;j<len;j++) {
1878 tmp0 = tab0[j];
1879 tab0[j] = MULL(tmp0, v1);
1880 tab1[j] = MULL(tmp0, v2);
1881 }
1882 } else {
1883 found2:
1884 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1885 /* lower part of the spectrum : do ms stereo
1886 if enabled */
1887 for(j=0;j<len;j++) {
1888 tmp0 = tab0[j];
1889 tmp1 = tab1[j];
1890 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1891 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1892 }
1893 }
1894 }
1895 }
1896 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1897 /* ms stereo ONLY */
1898 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1899 global gain */
1900 tab0 = g0->sb_hybrid;
1901 tab1 = g1->sb_hybrid;
1902 for(i=0;i<576;i++) {
1903 tmp0 = tab0[i];
1904 tmp1 = tab1[i];
1905 tab0[i] = tmp0 + tmp1;
1906 tab1[i] = tmp0 - tmp1;
1907 }
1908 }
1909}
1910
a1e257b2 1911static void compute_antialias_integer(MPADecodeContext *s,
239c2f4c
FB
1912 GranuleDef *g)
1913{
0c1a9eda 1914 int32_t *ptr, *p0, *p1, *csa;
a1e257b2 1915 int n, i, j;
239c2f4c
FB
1916
1917 /* we antialias only "long" bands */
1918 if (g->block_type == 2) {
1919 if (!g->switch_point)
1920 return;
1921 /* XXX: check this for 8000Hz case */
1922 n = 1;
1923 } else {
1924 n = SBLIMIT - 1;
1925 }
1926
1927 ptr = g->sb_hybrid + 18;
1928 for(i = n;i > 0;i--) {
1929 p0 = ptr - 1;
1930 p1 = ptr;
a1e257b2
MN
1931 csa = &csa_table[0][0];
1932 for(j=0;j<4;j++) {
1933 int tmp0 = *p0;
1934 int tmp1 = *p1;
1935#if 0
1936 *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));
1937 *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));
1938#else
1939 int64_t tmp2= MUL64(tmp0 + tmp1, csa[0]);
1940 *p0 = FRAC_RND(tmp2 - MUL64(tmp1, csa[2]));
1941 *p1 = FRAC_RND(tmp2 + MUL64(tmp0, csa[3]));
1942#endif
1943 p0--; p1++;
1944 csa += 4;
239c2f4c
FB
1945 tmp0 = *p0;
1946 tmp1 = *p1;
a1e257b2 1947#if 0
239c2f4c
FB
1948 *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));
1949 *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));
a1e257b2
MN
1950#else
1951 tmp2= MUL64(tmp0 + tmp1, csa[0]);
1952 *p0 = FRAC_RND(tmp2 - MUL64(tmp1, csa[2]));
1953 *p1 = FRAC_RND(tmp2 + MUL64(tmp0, csa[3]));
1954#endif
1955 p0--; p1++;
1956 csa += 4;
239c2f4c 1957 }
a1e257b2
MN
1958 ptr += 18;
1959 }
1960}
1961
1962static void compute_antialias_float(MPADecodeContext *s,
1963 GranuleDef *g)
1964{
1965 int32_t *ptr, *p0, *p1;
1966 int n, i, j;
1967
1968 /* we antialias only "long" bands */
1969 if (g->block_type == 2) {
1970 if (!g->switch_point)
1971 return;
1972 /* XXX: check this for 8000Hz case */
1973 n = 1;
1974 } else {
1975 n = SBLIMIT - 1;
1976 }
1977
1978 ptr = g->sb_hybrid + 18;
1979 for(i = n;i > 0;i--) {
1980 float *csa = &csa_table_float[0][0];
1981 p0 = ptr - 1;
1982 p1 = ptr;
1983 for(j=0;j<4;j++) {
1984 float tmp0 = *p0;
1985 float tmp1 = *p1;
1986#if 1
1987 *p0 = lrintf(tmp0 * csa[0] - tmp1 * csa[1]);
1988 *p1 = lrintf(tmp0 * csa[1] + tmp1 * csa[0]);
1989#else
1990 float tmp2= (tmp0 + tmp1) * csa[0];
1991 *p0 = lrintf(tmp2 - tmp1 * csa[2]);
1992 *p1 = lrintf(tmp2 + tmp0 * csa[3]);
1993#endif
1994 p0--; p1++;
1995 csa += 4;
1996 tmp0 = *p0;
1997 tmp1 = *p1;
1998#if 1
1999 *p0 = lrintf(tmp0 * csa[0] - tmp1 * csa[1]);
2000 *p1 = lrintf(tmp0 * csa[1] + tmp1 * csa[0]);
2001#else
2002 tmp2= (tmp0 + tmp1) * csa[0];
2003 *p0 = lrintf(tmp2 - tmp1 * csa[2]);
2004 *p1 = lrintf(tmp2 + tmp0 * csa[3]);
2005#endif
2006 p0--; p1++;
2007 csa += 4;
2008 }
2009 ptr += 18;
239c2f4c
FB
2010 }
2011}
2012
2013static void compute_imdct(MPADecodeContext *s,
2014 GranuleDef *g,
0c1a9eda
ZK
2015 int32_t *sb_samples,
2016 int32_t *mdct_buf)
239c2f4c 2017{
0c1a9eda
ZK
2018 int32_t *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1;
2019 int32_t in[6];
2020 int32_t out[36];
2021 int32_t out2[12];
239c2f4c
FB
2022 int i, j, k, mdct_long_end, v, sblimit;
2023
2024 /* find last non zero block */
2025 ptr = g->sb_hybrid + 576;
2026 ptr1 = g->sb_hybrid + 2 * 18;
2027 while (ptr >= ptr1) {
2028 ptr -= 6;
2029 v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
2030 if (v != 0)
2031 break;
2032 }
2033 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
2034
2035 if (g->block_type == 2) {
2036 /* XXX: check for 8000 Hz */
2037 if (g->switch_point)
2038 mdct_long_end = 2;
2039 else
2040 mdct_long_end = 0;
2041 } else {
2042 mdct_long_end = sblimit;
2043 }
2044
2045 buf = mdct_buf;
2046 ptr = g->sb_hybrid;
2047 for(j=0;j<mdct_long_end;j++) {
239c2f4c
FB
2048 /* apply window & overlap with previous buffer */
2049 out_ptr = sb_samples + j;
2050 /* select window */
2051 if (g->switch_point && j < 2)
2052 win1 = mdct_win[0];
2053 else
2054 win1 = mdct_win[g->block_type];
2055 /* select frequency inversion */
2056 win = win1 + ((4 * 36) & -(j & 1));
711ae726
MN
2057 imdct36(out_ptr, buf, ptr, win);
2058 out_ptr += 18*SBLIMIT;
239c2f4c
FB
2059 ptr += 18;
2060 buf += 18;
2061 }
2062 for(j=mdct_long_end;j<sblimit;j++) {
2063 for(i=0;i<6;i++) {
2064 out[i] = 0;
2065 out[6 + i] = 0;
2066 out[30+i] = 0;
2067 }
2068 /* select frequency inversion */
2069 win = mdct_win[2] + ((4 * 36) & -(j & 1));
2070 buf2 = out + 6;
2071 for(k=0;k<3;k++) {
2072 /* reorder input for short mdct */
2073 ptr1 = ptr + k;
2074 for(i=0;i<6;i++) {
2075 in[i] = *ptr1;
2076 ptr1 += 3;
2077 }
2078 imdct12(out2, in);
2079 /* apply 12 point window and do small overlap */
2080 for(i=0;i<6;i++) {
2081 buf2[i] = MULL(out2[i], win[i]) + buf2[i];
2082 buf2[i + 6] = MULL(out2[i + 6], win[i + 6]);
2083 }
2084 buf2 += 6;
2085 }
2086 /* overlap */
2087 out_ptr = sb_samples + j;
2088 for(i=0;i<18;i++) {
2089 *out_ptr = out[i] + buf[i];
2090 buf[i] = out[i + 18];
2091 out_ptr += SBLIMIT;
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
747a67fb 2109#if defined(DEBUG)
0c1a9eda 2110void sample_dump(int fnum, int32_t *tab, int n)
239c2f4c
FB
2111{
2112 static FILE *files[16], *f;
2113 char buf[512];
81552334 2114 int i;
0c1a9eda 2115 int32_t v;
81552334 2116
239c2f4c
FB
2117 f = files[fnum];
2118 if (!f) {
2fc8ea24 2119 snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
81552334
FB
2120 fnum,
2121#ifdef USE_HIGHPRECISION
2122 "hp"
2123#else
2124 "lp"
2125#endif
2126 );
239c2f4c
FB
2127 f = fopen(buf, "w");
2128 if (!f)
2129 return;
2130 files[fnum] = f;
2131 }
2132
2133 if (fnum == 0) {
239c2f4c 2134 static int pos = 0;
84af4a7e 2135 av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
239c2f4c 2136 for(i=0;i<n;i++) {
84af4a7e 2137 av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
239c2f4c 2138 if ((i % 18) == 17)
84af4a7e 2139 av_log(NULL, AV_LOG_DEBUG, "\n");
239c2f4c
FB
2140 }
2141 pos += n;
2142 }
81552334
FB
2143 for(i=0;i<n;i++) {
2144 /* normalize to 23 frac bits */
2145 v = tab[i] << (23 - FRAC_BITS);
0c1a9eda 2146 fwrite(&v, 1, sizeof(int32_t), f);
81552334 2147 }
239c2f4c
FB
2148}
2149#endif
2150
2151
2152/* main layer3 decoding function */
2153static 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;
0c1a9eda 2158 int16_t exponents[576];
239c2f4c
FB
2159
2160 /* read side info */
2161 if (s->lsf) {
2162 main_data_begin = get_bits(&s->gb, 8);
2163 if (s->nb_channels == 2)
2164 private_bits = get_bits(&s->gb, 2);
2165 else
2166 private_bits = get_bits(&s->gb, 1);
2167 nb_granules = 1;
2168 } else {
2169 main_data_begin = get_bits(&s->gb, 9);
2170 if (s->nb_channels == 2)
2171 private_bits = get_bits(&s->gb, 3);
2172 else
2173 private_bits = get_bits(&s->gb, 5);
2174 nb_granules = 2;
2175 for(ch=0;ch<s->nb_channels;ch++) {
2176 granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2177 granules[ch][1].scfsi = get_bits(&s->gb, 4);
2178 }
2179 }
2180
2181 for(gr=0;gr<nb_granules;gr++) {
2182 for(ch=0;ch<s->nb_channels;ch++) {
2183 dprintf("gr=%d ch=%d: side_info\n", gr, ch);
2184 g = &granules[ch][gr];
2185 g->part2_3_length = get_bits(&s->gb, 12);
2186 g->big_values = get_bits(&s->gb, 9);
2187 g->global_gain = get_bits(&s->gb, 8);
2188 /* if MS stereo only is selected, we precompute the
2189 1/sqrt(2) renormalization factor */
2190 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
2191 MODE_EXT_MS_STEREO)
2192 g->global_gain -= 2;
2193 if (s->lsf)
2194 g->scalefac_compress = get_bits(&s->gb, 9);
2195 else
2196 g->scalefac_compress = get_bits(&s->gb, 4);
2197 blocksplit_flag = get_bits(&s->gb, 1);
2198 if (blocksplit_flag) {
2199 g->block_type = get_bits(&s->gb, 2);
2200 if (g->block_type == 0)
2201 return -1;
2202 g->switch_point = get_bits(&s->gb, 1);
2203 for(i=0;i<2;i++)
2204 g->table_select[i] = get_bits(&s->gb, 5);
2205 for(i=0;i<3;i++)
2206 g->subblock_gain[i] = get_bits(&s->gb, 3);
2207 /* compute huffman coded region sizes */
2208 if (g->block_type == 2)
2209 g->region_size[0] = (36 / 2);
2210 else {
2211 if (s->sample_rate_index <= 2)
2212 g->region_size[0] = (36 / 2);
2213 else if (s->sample_rate_index != 8)
2214 g->region_size[0] = (54 / 2);
2215 else
2216 g->region_size[0] = (108 / 2);
2217 }
2218 g->region_size[1] = (576 / 2);
2219 } else {
2220 int region_address1, region_address2, l;
2221 g->block_type = 0;
2222 g->switch_point = 0;
2223 for(i=0;i<3;i++)
2224 g->table_select[i] = get_bits(&s->gb, 5);
2225 /* compute huffman coded region sizes */
2226 region_address1 = get_bits(&s->gb, 4);
2227 region_address2 = get_bits(&s->gb, 3);
2228 dprintf("region1=%d region2=%d\n",
2229 region_address1, region_address2);
2230 g->region_size[0] =
2231 band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2232 l = region_address1 + region_address2 + 2;
2233 /* should not overflow */
2234 if (l > 22)
2235 l = 22;
2236 g->region_size[1] =
2237 band_index_long[s->sample_rate_index][l] >> 1;
2238 }
2239 /* convert region offsets to region sizes and truncate
2240 size to big_values */
2241 g->region_size[2] = (576 / 2);
2242 j = 0;
2243 for(i=0;i<3;i++) {
2244 k = g->region_size[i];
2245 if (k > g->big_values)
2246 k = g->big_values;
2247 g->region_size[i] = k - j;
2248 j = k;
2249 }
2250
2251 /* compute band indexes */
2252 if (g->block_type == 2) {
2253 if (g->switch_point) {
2254 /* if switched mode, we handle the 36 first samples as
2255 long blocks. For 8000Hz, we handle the 48 first
2256 exponents as long blocks (XXX: check this!) */
2257 if (s->sample_rate_index <= 2)
2258 g->long_end = 8;
2259 else if (s->sample_rate_index != 8)
2260 g->long_end = 6;
2261 else
2262 g->long_end = 4; /* 8000 Hz */
2263
2264 if (s->sample_rate_index != 8)
2265 g->short_start = 3;
2266 else
2267 g->short_start = 2;
2268 } else {
2269 g->long_end = 0;
2270 g->short_start = 0;
2271 }
2272 } else {
2273 g->short_start = 13;
2274 g->long_end = 22;
2275 }
2276
2277 g->preflag = 0;
2278 if (!s->lsf)
2279 g->preflag = get_bits(&s->gb, 1);
2280 g->scalefac_scale = get_bits(&s->gb, 1);
2281 g->count1table_select = get_bits(&s->gb, 1);
2282 dprintf("block_type=%d switch_point=%d\n",
2283 g->block_type, g->switch_point);
2284 }
2285 }
2286
1ede228a 2287 if (!s->adu_mode) {
239c2f4c
FB
2288 /* now we get bits from the main_data_begin offset */
2289 dprintf("seekback: %d\n", main_data_begin);
2290 seek_to_maindata(s, main_data_begin);
1ede228a 2291 }
239c2f4c
FB
2292
2293 for(gr=0;gr<nb_granules;gr++) {
2294 for(ch=0;ch<s->nb_channels;ch++) {
2295 g = &granules[ch][gr];
2296
2297 bits_pos = get_bits_count(&s->gb);
2298
2299 if (!s->lsf) {
0c1a9eda 2300 uint8_t *sc;
239c2f4c
FB
2301 int slen, slen1, slen2;
2302
2303 /* MPEG1 scale factors */
2304 slen1 = slen_table[0][g->scalefac_compress];
2305 slen2 = slen_table[1][g->scalefac_compress];
2306 dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2307 if (g->block_type == 2) {
2308 n = g->switch_point ? 17 : 18;
2309 j = 0;
2310 for(i=0;i<n;i++)
2311 g->scale_factors[j++] = get_bitsz(&s->gb, slen1);
2312 for(i=0;i<18;i++)
2313 g->scale_factors[j++] = get_bitsz(&s->gb, slen2);
2314 for(i=0;i<3;i++)
2315 g->scale_factors[j++] = 0;
2316 } else {
2317 sc = granules[ch][0].scale_factors;
2318 j = 0;
2319 for(k=0;k<4;k++) {
2320 n = (k == 0 ? 6 : 5);
2321 if ((g->scfsi & (0x8 >> k)) == 0) {
2322 slen = (k < 2) ? slen1 : slen2;
2323 for(i=0;i<n;i++)
2324 g->scale_factors[j++] = get_bitsz(&s->gb, slen);
2325 } else {
2326 /* simply copy from last granule */
2327 for(i=0;i<n;i++) {
2328 g->scale_factors[j] = sc[j];
2329 j++;
2330 }
2331 }
2332 }
2333 g->scale_factors[j++] = 0;
2334 }
747a67fb 2335#if defined(DEBUG)
239c2f4c
FB
2336 {
2337 printf("scfsi=%x gr=%d ch=%d scale_factors:\n",
2338 g->scfsi, gr, ch);
2339 for(i=0;i<j;i++)
2340 printf(" %d", g->scale_factors[i]);
2341 printf("\n");
2342 }
2343#endif
2344 } else {
2345 int tindex, tindex2, slen[4], sl, sf;
2346
2347 /* LSF scale factors */
2348 if (g->block_type == 2) {
2349 tindex = g->switch_point ? 2 : 1;
2350 } else {
2351 tindex = 0;
2352 }
2353 sf = g->scalefac_compress;
2354 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2355 /* intensity stereo case */
2356 sf >>= 1;
2357 if (sf < 180) {
2358 lsf_sf_expand(slen, sf, 6, 6, 0);
2359 tindex2 = 3;
2360 } else if (sf < 244) {
2361 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2362 tindex2 = 4;
2363 } else {
2364 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2365 tindex2 = 5;
2366 }
2367 } else {
2368 /* normal case */
2369 if (sf < 400) {
2370 lsf_sf_expand(slen, sf, 5, 4, 4);
2371 tindex2 = 0;
2372 } else if (sf < 500) {
2373 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2374 tindex2 = 1;
2375 } else {
2376 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2377 tindex2 = 2;
2378 g->preflag = 1;
2379 }
2380 }
2381
2382 j = 0;
2383 for(k=0;k<4;k++) {
2384 n = lsf_nsf_table[tindex2][tindex][k];
2385 sl = slen[k];
2386 for(i=0;i<n;i++)
2387 g->scale_factors[j++] = get_bitsz(&s->gb, sl);
2388 }
2389 /* XXX: should compute exact size */
2390 for(;j<40;j++)
2391 g->scale_factors[j] = 0;
747a67fb 2392#if defined(DEBUG)
239c2f4c
FB
2393 {
2394 printf("gr=%d ch=%d scale_factors:\n",
2395 gr, ch);
2396 for(i=0;i<40;i++)
2397 printf(" %d", g->scale_factors[i]);
2398 printf("\n");
2399 }
2400#endif
2401 }
2402
2403 exponents_from_scale_factors(s, g, exponents);
2404
2405 /* read Huffman coded residue */
2406 if (huffman_decode(s, g, exponents,
2407 bits_pos + g->part2_3_length) < 0)
2408 return -1;
747a67fb
FB
2409#if defined(DEBUG)
2410 sample_dump(0, g->sb_hybrid, 576);
239c2f4c
FB
2411#endif
2412
2413 /* skip extension bits */
2414 bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2415 if (bits_left < 0) {
2416 dprintf("bits_left=%d\n", bits_left);
2417 return -1;
2418 }
2419 while (bits_left >= 16) {
2420 skip_bits(&s->gb, 16);
2421 bits_left -= 16;
2422 }
2423 if (bits_left > 0)
2424 skip_bits(&s->gb, bits_left);
2425 } /* ch */
2426
2427 if (s->nb_channels == 2)
2428 compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2429
2430 for(ch=0;ch<s->nb_channels;ch++) {
2431 g = &granules[ch][gr];
2432
2433 reorder_block(s, g);
747a67fb 2434#if defined(DEBUG)
239c2f4c
FB
2435 sample_dump(0, g->sb_hybrid, 576);
2436#endif
a1e257b2 2437 s->compute_antialias(s, g);
81552334 2438#if defined(DEBUG)
239c2f4c
FB
2439 sample_dump(1, g->sb_hybrid, 576);
2440#endif
2441 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
81552334 2442#if defined(DEBUG)
239c2f4c
FB
2443 sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2444#endif
2445 }
2446 } /* gr */
2447 return nb_granules * 18;
2448}
2449
2450static int mp_decode_frame(MPADecodeContext *s,
a3a5f4d6 2451 OUT_INT *samples)
239c2f4c
FB
2452{
2453 int i, nb_frames, ch;
a3a5f4d6 2454 OUT_INT *samples_ptr;
239c2f4c
FB
2455
2456 init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
68f593b4 2457 (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);
239c2f4c
FB
2458
2459 /* skip error protection field */
2460 if (s->error_protection)
2461 get_bits(&s->gb, 16);
2462
2463 dprintf("frame %d:\n", s->frame_count);
2464 switch(s->layer) {
2465 case 1:
2466 nb_frames = mp_decode_layer1(s);
2467 break;
2468 case 2:
2469 nb_frames = mp_decode_layer2(s);
2470 break;
2471 case 3:
2472 default:
2473 nb_frames = mp_decode_layer3(s);
2474 break;
2475 }
2476#if defined(DEBUG)
2477 for(i=0;i<nb_frames;i++) {
2478 for(ch=0;ch<s->nb_channels;ch++) {
2479 int j;
2480 printf("%d-%d:", i, ch);
2481 for(j=0;j<SBLIMIT;j++)
2482 printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2483 printf("\n");
2484 }
2485 }
2486#endif
2487 /* apply the synthesis filter */
2488 for(ch=0;ch<s->nb_channels;ch++) {
2489 samples_ptr = samples + ch;
2490 for(i=0;i<nb_frames;i++) {
bf1f4da0 2491 ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
093c6e50 2492 window, &s->dither_state,
bf1f4da0 2493 samples_ptr, s->nb_channels,
239c2f4c
FB
2494 s->sb_samples[ch][i]);
2495 samples_ptr += 32 * s->nb_channels;
2496 }
2497 }
2498#ifdef DEBUG
2499 s->frame_count++;
2500#endif
a3a5f4d6 2501 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
239c2f4c
FB
2502}
2503
de6d9b64
FB
2504static int decode_frame(AVCodecContext * avctx,
2505 void *data, int *data_size,
0c1a9eda 2506 uint8_t * buf, int buf_size)
de6d9b64
FB
2507{
2508 MPADecodeContext *s = avctx->priv_data;
0c1a9eda
ZK
2509 uint32_t header;
2510 uint8_t *buf_ptr;
de6d9b64 2511 int len, out_size;
a3a5f4d6 2512 OUT_INT *out_samples = data;
de6d9b64 2513
de6d9b64
FB
2514 buf_ptr = buf;
2515 while (buf_size > 0) {
2516 len = s->inbuf_ptr - s->inbuf;
2517 if (s->frame_size == 0) {
239c2f4c
FB
2518 /* special case for next header for first frame in free
2519 format case (XXX: find a simpler method) */
2520 if (s->free_format_next_header != 0) {
2521 s->inbuf[0] = s->free_format_next_header >> 24;
2522 s->inbuf[1] = s->free_format_next_header >> 16;
2523 s->inbuf[2] = s->free_format_next_header >> 8;
2524 s->inbuf[3] = s->free_format_next_header;
2525 s->inbuf_ptr = s->inbuf + 4;
2526 s->free_format_next_header = 0;
2527 goto got_header;
2528 }
2529 /* no header seen : find one. We need at least HEADER_SIZE
2530 bytes to parse it */
de6d9b64
FB
2531 len = HEADER_SIZE - len;
2532 if (len > buf_size)
2533 len = buf_size;
92d24f49 2534 if (len > 0) {
2d83f323
ZK
2535 memcpy(s->inbuf_ptr, buf_ptr, len);
2536 buf_ptr += len;
2537 buf_size -= len;
c152c983
ZK
2538 s->inbuf_ptr += len;
2539 }
2540 if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
239c2f4c 2541 got_header:
de6d9b64
FB
2542 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2543 (s->inbuf[2] << 8) | s->inbuf[3];
92d24f49 2544
a7a85899 2545 if (ff_mpa_check_header(header) < 0) {
de6d9b64 2546 /* no sync found : move by one byte (inefficient, but simple!) */
228ef9dd 2547 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
de6d9b64 2548 s->inbuf_ptr--;
239c2f4c
FB
2549 dprintf("skip %x\n", header);
2550 /* reset free format frame size to give a chance
2551 to get a new bitrate */
2552 s->free_format_frame_size = 0;
de6d9b64 2553 } else {
239c2f4c 2554 if (decode_header(s, header) == 1) {
81552334 2555 /* free format: prepare to compute frame size */
2d83f323 2556 s->frame_size = -1;
239c2f4c 2557 }
81552334
FB
2558 /* update codec info */
2559 avctx->sample_rate = s->sample_rate;
2560 avctx->channels = s->nb_channels;
2561 avctx->bit_rate = s->bit_rate;
98ce5991 2562 avctx->sub_id = s->layer;
8c5b5683
FB
2563 switch(s->layer) {
2564 case 1:
2565 avctx->frame_size = 384;
2566 break;
2567 case 2:
2568 avctx->frame_size = 1152;
2569 break;
2570 case 3:
2571 if (s->lsf)
2572 avctx->frame_size = 576;
2573 else
2574 avctx->frame_size = 1152;
2575 break;
2576 }
de6d9b64
FB
2577 }
2578 }
239c2f4c
FB
2579 } else if (s->frame_size == -1) {
2580 /* free format : find next sync to compute frame size */
2581 len = MPA_MAX_CODED_FRAME_SIZE - len;
2582 if (len > buf_size)
2583 len = buf_size;
2584 if (len == 0) {
3625e88a 2585 /* frame too long: resync */
239c2f4c 2586 s->frame_size = 0;
228ef9dd 2587 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
3625e88a 2588 s->inbuf_ptr--;
239c2f4c 2589 } else {
0c1a9eda
ZK
2590 uint8_t *p, *pend;
2591 uint32_t header1;
239c2f4c
FB
2592 int padding;
2593
2594 memcpy(s->inbuf_ptr, buf_ptr, len);
2595 /* check for header */
2596 p = s->inbuf_ptr - 3;
2597 pend = s->inbuf_ptr + len - 4;
2598 while (p <= pend) {
2599 header = (p[0] << 24) | (p[1] << 16) |
2600 (p[2] << 8) | p[3];
2601 header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2602 (s->inbuf[2] << 8) | s->inbuf[3];
2603 /* check with high probability that we have a
2604 valid header */
2605 if ((header & SAME_HEADER_MASK) ==
2606 (header1 & SAME_HEADER_MASK)) {
2607 /* header found: update pointers */
2608 len = (p + 4) - s->inbuf_ptr;
2609 buf_ptr += len;
2610 buf_size -= len;
2611 s->inbuf_ptr = p;
2612 /* compute frame size */
2613 s->free_format_next_header = header;
2614 s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2615 padding = (header1 >> 9) & 1;
2616 if (s->layer == 1)
2617 s->free_format_frame_size -= padding * 4;
2618 else
2619 s->free_format_frame_size -= padding;
2620 dprintf("free frame size=%d padding=%d\n",
2621 s->free_format_frame_size, padding);
2622 decode_header(s, header1);
2623 goto next_data;
2624 }
2625 p++;
2626 }
2627 /* not found: simply increase pointers */
2628 buf_ptr += len;
2629 s->inbuf_ptr += len;
2630 buf_size -= len;
2631 }
de6d9b64 2632 } else if (len < s->frame_size) {
de5123dc
ZK
2633 if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2634 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
de6d9b64
FB
2635 len = s->frame_size - len;
2636 if (len > buf_size)
2637 len = buf_size;
de6d9b64
FB
2638 memcpy(s->inbuf_ptr, buf_ptr, len);
2639 buf_ptr += len;
2640 s->inbuf_ptr += len;
2641 buf_size -= len;
8c5b5683
FB
2642 }
2643 next_data:
2644 if (s->frame_size > 0 &&
2645 (s->inbuf_ptr - s->inbuf) >= s->frame_size) {
2646 if (avctx->parse_only) {
2647 /* simply return the frame data */
2648 *(uint8_t **)data = s->inbuf;
2649 out_size = s->inbuf_ptr - s->inbuf;
2650 } else {
2651 out_size = mp_decode_frame(s, out_samples);
2652 }
de6d9b64
FB
2653 s->inbuf_ptr = s->inbuf;
2654 s->frame_size = 0;
2655 *data_size = out_size;
2656 break;
2657 }
2658 }
2659 return buf_ptr - buf;
2660}
2661
1ede228a
RT
2662
2663static int decode_frame_adu(AVCodecContext * avctx,
2664 void *data, int *data_size,
2665 uint8_t * buf, int buf_size)
2666{
2667 MPADecodeContext *s = avctx->priv_data;
2668 uint32_t header;
2669 int len, out_size;
a3a5f4d6 2670 OUT_INT *out_samples = data;
1ede228a
RT
2671
2672 len = buf_size;
2673
2674 // Discard too short frames
2675 if (buf_size < HEADER_SIZE) {
2676 *data_size = 0;
2677 return buf_size;
2678 }
2679
2680
2681 if (len > MPA_MAX_CODED_FRAME_SIZE)
2682 len = MPA_MAX_CODED_FRAME_SIZE;
2683
2684 memcpy(s->inbuf, buf, len);
2685 s->inbuf_ptr = s->inbuf + len;
2686
2687 // Get header and restore sync word
2688 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2689 (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;
2690
a7a85899 2691 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1ede228a
RT
2692 *data_size = 0;
2693 return buf_size;
2694 }
2695
2696 decode_header(s, header);
2697 /* update codec info */
2698 avctx->sample_rate = s->sample_rate;
2699 avctx->channels = s->nb_channels;
2700 avctx->bit_rate = s->bit_rate;
2701 avctx->sub_id = s->layer;
2702
2703 avctx->frame_size=s->frame_size = len;
2704
2705 if (avctx->parse_only) {
2706 /* simply return the frame data */
2707 *(uint8_t **)data = s->inbuf;
2708 out_size = s->inbuf_ptr - s->inbuf;
2709 } else {
2710 out_size = mp_decode_frame(s, out_samples);
2711 }
2712
2713 *data_size = out_size;
2714 return buf_size;
2715}
2716
2717
4b1f4f23 2718AVCodec mp2_decoder =
de6d9b64 2719{
4b1f4f23 2720 "mp2",
de6d9b64
FB
2721 CODEC_TYPE_AUDIO,
2722 CODEC_ID_MP2,
2723 sizeof(MPADecodeContext),
2724 decode_init,
2725 NULL,
2726 NULL,
2727 decode_frame,
8c5b5683 2728 CODEC_CAP_PARSE_ONLY,
de6d9b64 2729};
4b1f4f23
J
2730
2731AVCodec mp3_decoder =
2732{
2733 "mp3",
2734 CODEC_TYPE_AUDIO,
80783dc2 2735 CODEC_ID_MP3,
4b1f4f23
J
2736 sizeof(MPADecodeContext),
2737 decode_init,
2738 NULL,
2739 NULL,
2740 decode_frame,
8c5b5683 2741 CODEC_CAP_PARSE_ONLY,
4b1f4f23 2742};
1ede228a
RT
2743
2744AVCodec mp3adu_decoder =
2745{
2746 "mp3adu",
2747 CODEC_TYPE_AUDIO,
2748 CODEC_ID_MP3ADU,
2749 sizeof(MPADecodeContext),
2750 decode_init,
2751 NULL,
2752 NULL,
2753 decode_frame_adu,
2754 CODEC_CAP_PARSE_ONLY,
2755};