faster handling of zero coeffs in the low freq decode
[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
5509bffa 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
de6d9b64 18 */
983e3246
MN
19
20/**
21 * @file mpegaudiodec.c
22 * MPEG Audio decoder.
115329f1 23 */
983e3246 24
239c2f4c 25//#define DEBUG
de6d9b64 26#include "avcodec.h"
caa336b4 27#include "bitstream.h"
84f986c0 28#include "dsputil.h"
de6d9b64
FB
29
30/*
239c2f4c
FB
31 * TODO:
32 * - in low precision mode, use more 16 bit multiplies in synth filter
33 * - test lsf / mpeg25 extensively.
de6d9b64
FB
34 */
35
239c2f4c
FB
36/* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg
37 audio decoder */
81552334 38#ifdef CONFIG_MPEGAUDIO_HP
5ab1972b 39# define USE_HIGHPRECISION
81552334 40#endif
239c2f4c 41
d9b1c197 42#include "mpegaudio.h"
a3a5f4d6 43
239c2f4c
FB
44#define FRAC_ONE (1 << FRAC_BITS)
45
47a0cd74 46#ifdef ARCH_X86
f617aded
MN
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; })
47a0cd74
MN
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; })
69fd15f8
MN
58#elif defined(ARCH_ARMV4L)
59# define MULL(a, b) \
60 ({ int lo, hi;\
61 asm("smull %0, %1, %2, %3 \n\t"\
acba3238
AJ
62 "mov %0, %0, lsr %4\n\t"\
63 "add %1, %0, %1, lsl %5\n\t"\
64 : "=&r"(lo), "=&r"(hi)\
69fd15f8
MN
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))
acba3238 68# define MULH(a, b) ({ int lo, hi; asm ("smull %0, %1, %2, %3" : "=&r"(lo), "=&r"(hi) : "r"(b), "r"(a)); hi; })
47a0cd74
MN
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
73static always_inline int MULH(int a, int b){
74 return ((int64_t)(a) * (int64_t)(b))>>32;
75}
76#endif
239c2f4c
FB
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
711ae726 82#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
711ae726 83
239c2f4c
FB
84/****************/
85
de6d9b64
FB
86#define HEADER_SIZE 4
87#define BACKSTEP_SIZE 512
88
a1e257b2
MN
89struct GranuleDef;
90
de6d9b64 91typedef struct MPADecodeContext {
bb270c08 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) */
115329f1 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
d2a7718d
RT
123/**
124 * Context for MP3On4 decoder
125 */
126typedef 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
239c2f4c
FB
132/* layer 3 "granule" */
133typedef struct GranuleDef {
0c1a9eda 134 uint8_t scfsi;
239c2f4c
FB
135 int part2_3_length;
136 int big_values;
137 int global_gain;
138 int scalefac_compress;
0c1a9eda
ZK
139 uint8_t block_type;
140 uint8_t switch_point;
239c2f4c
FB
141 int table_select[3];
142 int subblock_gain[3];
0c1a9eda
ZK
143 uint8_t scalefac_scale;
144 uint8_t count1table_select;
239c2f4c
FB
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 */
0c1a9eda
ZK
148 uint8_t scale_factors[40];
149 int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */
239c2f4c 150} GranuleDef;
de6d9b64 151
239c2f4c
FB
152#define MODE_EXT_MS_STEREO 2
153#define MODE_EXT_I_STEREO 1
154
155/* layer 3 huffman tables */
156typedef struct HuffTable {
157 int xsize;
0c1a9eda
ZK
158 const uint8_t *bits;
159 const uint16_t *codes;
239c2f4c
FB
160} HuffTable;
161
162#include "mpegaudiodectab.h"
163
a1e257b2
MN
164static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);
165static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);
166
239c2f4c 167/* vlc structure for decoding layer 3 huffman tables */
115329f1 168static VLC huff_vlc[16];
239c2f4c
FB
169static VLC huff_quad_vlc[2];
170/* computed from band_size_long */
0c1a9eda 171static uint16_t band_index_long[9][23];
239c2f4c 172/* XXX: free when all decoders are closed */
d04728bb 173#define TABLE_4_3_SIZE (8191 + 16)*4
0c1a9eda 174static int8_t *table_4_3_exp;
0c1a9eda 175static uint32_t *table_4_3_value;
239c2f4c 176/* intensity stereo coef table */
0c1a9eda
ZK
177static int32_t is_table[2][16];
178static int32_t is_table_lsf[2][2][16];
a1e257b2
MN
179static int32_t csa_table[8][4];
180static float csa_table_float[8][4];
0c1a9eda 181static int32_t mdct_win[8][36];
239c2f4c
FB
182
183/* lower 2 bits: modulo 3, higher bits: shift */
0c1a9eda 184static uint16_t scale_factor_modshift[64];
239c2f4c 185/* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
0c1a9eda 186static int32_t scale_factor_mult[15][3];
239c2f4c
FB
187/* mult table for layer 2 group quantization */
188
189#define SCALE_GEN(v) \
190{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }
191
c26ae41d 192static const int32_t scale_factor_mult2[3][3] = {
81552334
FB
193 SCALE_GEN(4.0 / 3.0), /* 3 steps */
194 SCALE_GEN(4.0 / 5.0), /* 5 steps */
195 SCALE_GEN(4.0 / 9.0), /* 9 steps */
239c2f4c
FB
196};
197
bf1f4da0 198void ff_mpa_synth_init(MPA_INT *window);
a05c8d71 199static MPA_INT window[512] __attribute__((aligned(16)));
115329f1 200
239c2f4c
FB
201/* layer 1 unscaling */
202/* n = number of bits of the mantissa minus 1 */
203static inline int l1_unscale(int n, int mant, int scale_factor)
204{
205 int shift, mod;
0c1a9eda 206 int64_t val;
239c2f4c
FB
207
208 shift = scale_factor_modshift[scale_factor];
209 mod = shift & 3;
210 shift >>= 2;
211 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
212 shift += n;
81552334
FB
213 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
214 return (int)((val + (1LL << (shift - 1))) >> shift);
239c2f4c
FB
215}
216
217static inline int l2_unscale_group(int steps, int mant, int scale_factor)
218{
219 int shift, mod, val;
220
221 shift = scale_factor_modshift[scale_factor];
222 mod = shift & 3;
223 shift >>= 2;
81552334
FB
224
225 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
226 /* NOTE: at this point, 0 <= shift <= 21 */
227 if (shift > 0)
228 val = (val + (1 << (shift - 1))) >> shift;
229 return val;
239c2f4c
FB
230}
231
232/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
233static inline int l3_unscale(int value, int exponent)
234{
239c2f4c 235 unsigned int m;
239c2f4c
FB
236 int e;
237
d04728bb
MN
238 e = table_4_3_exp [4*value + (exponent&3)];
239 m = table_4_3_value[4*value + (exponent&3)];
240 e -= (exponent >> 2);
241 assert(e>=1);
239c2f4c 242 if (e > 31)
b696d2a6 243 return 0;
239c2f4c 244 m = (m + (1 << (e-1))) >> e;
d04728bb 245
239c2f4c 246 return m;
239c2f4c
FB
247}
248
f9ed4f88
FB
249/* all integer n^(4/3) computation code */
250#define DEV_ORDER 13
251
252#define POW_FRAC_BITS 24
253#define POW_FRAC_ONE (1 << POW_FRAC_BITS)
254#define POW_FIX(a) ((int)((a) * POW_FRAC_ONE))
0c1a9eda 255#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)
f9ed4f88
FB
256
257static int dev_4_3_coefs[DEV_ORDER];
258
88730be6 259#if 0 /* unused */
f9ed4f88
FB
260static int pow_mult3[3] = {
261 POW_FIX(1.0),
262 POW_FIX(1.25992104989487316476),
263 POW_FIX(1.58740105196819947474),
264};
88730be6 265#endif
f9ed4f88
FB
266
267static void int_pow_init(void)
268{
269 int i, a;
270
271 a = POW_FIX(1.0);
272 for(i=0;i<DEV_ORDER;i++) {
273 a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);
274 dev_4_3_coefs[i] = a;
275 }
276}
277
88730be6 278#if 0 /* unused, remove? */
f9ed4f88
FB
279/* return the mantissa and the binary exponent */
280static int int_pow(int i, int *exp_ptr)
281{
282 int e, er, eq, j;
283 int a, a1;
115329f1 284
f9ed4f88
FB
285 /* renormalize */
286 a = i;
287 e = POW_FRAC_BITS;
288 while (a < (1 << (POW_FRAC_BITS - 1))) {
289 a = a << 1;
290 e--;
291 }
292 a -= (1 << POW_FRAC_BITS);
293 a1 = 0;
294 for(j = DEV_ORDER - 1; j >= 0; j--)
295 a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);
296 a = (1 << POW_FRAC_BITS) + a1;
297 /* exponent compute (exact) */
298 e = e * 4;
299 er = e % 3;
300 eq = e / 3;
301 a = POW_MULL(a, pow_mult3[er]);
302 while (a >= 2 * POW_FRAC_ONE) {
303 a = a >> 1;
304 eq++;
305 }
306 /* convert to float */
307 while (a < POW_FRAC_ONE) {
308 a = a << 1;
309 eq--;
310 }
59d3e367 311 /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */
81552334 312#if POW_FRAC_BITS > FRAC_BITS
59d3e367
FB
313 a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);
314 /* correct overflow */
315 if (a >= 2 * (1 << FRAC_BITS)) {
316 a = a >> 1;
317 eq++;
318 }
319#endif
f9ed4f88 320 *exp_ptr = eq;
f9ed4f88 321 return a;
f9ed4f88 322}
88730be6 323#endif
de6d9b64
FB
324
325static int decode_init(AVCodecContext * avctx)
326{
327 MPADecodeContext *s = avctx->priv_data;
b587a7cb 328 static int init=0;
239c2f4c 329 int i, j, k;
de6d9b64 330
e6885654 331#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)
a3a5f4d6
MN
332 avctx->sample_fmt= SAMPLE_FMT_S32;
333#else
334 avctx->sample_fmt= SAMPLE_FMT_S16;
115329f1
DB
335#endif
336
ce4a29c0 337 if(avctx->antialias_algo != FF_AA_FLOAT)
ac806113
MN
338 s->compute_antialias= compute_antialias_integer;
339 else
340 s->compute_antialias= compute_antialias_float;
341
8c5b5683 342 if (!init && !avctx->parse_only) {
239c2f4c
FB
343 /* scale factors table for layer 1/2 */
344 for(i=0;i<64;i++) {
345 int shift, mod;
346 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
81552334 347 shift = (i / 3);
239c2f4c 348 mod = i % 3;
239c2f4c
FB
349 scale_factor_modshift[i] = mod | (shift << 2);
350 }
351
352 /* scale factor multiply for layer 1 */
353 for(i=0;i<15;i++) {
354 int n, norm;
355 n = i + 2;
0c1a9eda 356 norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
81552334
FB
357 scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);
358 scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);
359 scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);
239c2f4c 360 dprintf("%d: norm=%x s=%x %x %x\n",
115329f1 361 i, norm,
239c2f4c
FB
362 scale_factor_mult[i][0],
363 scale_factor_mult[i][1],
364 scale_factor_mult[i][2]);
365 }
115329f1 366
bb270c08 367 ff_mpa_synth_init(window);
115329f1 368
239c2f4c 369 /* huffman decode tables */
239c2f4c
FB
370 for(i=1;i<16;i++) {
371 const HuffTable *h = &mpa_huff_tables[i];
bb270c08 372 int xsize, x, y;
5c91a675 373 unsigned int n;
c7aa3696
MN
374 uint8_t tmp_bits [256];
375 uint16_t tmp_codes[256];
376
377 memset(tmp_bits , 0, sizeof(tmp_bits ));
378 memset(tmp_codes, 0, sizeof(tmp_codes));
239c2f4c
FB
379
380 xsize = h->xsize;
381 n = xsize * xsize;
115329f1 382
239c2f4c
FB
383 j = 0;
384 for(x=0;x<xsize;x++) {
c7aa3696
MN
385 for(y=0;y<xsize;y++){
386 tmp_bits [(x << 4) | y]= h->bits [j ];
387 tmp_codes[(x << 4) | y]= h->codes[j++];
388 }
239c2f4c 389 }
c7aa3696
MN
390
391 /* XXX: fail test */
392 init_vlc(&huff_vlc[i], 8, 256,
393 tmp_bits, 1, 1, tmp_codes, 2, 2, 1);
239c2f4c
FB
394 }
395 for(i=0;i<2;i++) {
115329f1 396 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
073c2593 397 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);
239c2f4c
FB
398 }
399
400 for(i=0;i<9;i++) {
401 k = 0;
402 for(j=0;j<22;j++) {
403 band_index_long[i][j] = k;
404 k += band_size_long[i][j];
405 }
406 band_index_long[i][22] = k;
407 }
408
bb270c08
DB
409 /* compute n ^ (4/3) and store it in mantissa/exp format */
410 table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));
8d1f2ba5 411 if(!table_4_3_exp)
bb270c08
DB
412 return -1;
413 table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));
8d1f2ba5 414 if(!table_4_3_value)
239c2f4c 415 return -1;
115329f1 416
f9ed4f88 417 int_pow_init();
239c2f4c 418 for(i=1;i<TABLE_4_3_SIZE;i++) {
d04728bb 419 double f, fm;
239c2f4c 420 int e, m;
d04728bb
MN
421 f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);
422 fm = frexp(f, &e);
f86f4481 423 m = (uint32_t)(fm*(1LL<<31) + 0.5);
44f1698a 424 e+= FRAC_BITS - 31 + 5;
d04728bb 425
239c2f4c
FB
426 /* normalized to FRAC_BITS */
427 table_4_3_value[i] = m;
d04728bb
MN
428// av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));
429 table_4_3_exp[i] = -e;
239c2f4c 430 }
115329f1 431
239c2f4c
FB
432 for(i=0;i<7;i++) {
433 float f;
434 int v;
435 if (i != 6) {
436 f = tan((double)i * M_PI / 12.0);
437 v = FIXR(f / (1.0 + f));
438 } else {
439 v = FIXR(1.0);
440 }
441 is_table[0][i] = v;
442 is_table[1][6 - i] = v;
443 }
444 /* invalid values */
445 for(i=7;i<16;i++)
446 is_table[0][i] = is_table[1][i] = 0.0;
447
448 for(i=0;i<16;i++) {
449 double f;
450 int e, k;
451
452 for(j=0;j<2;j++) {
453 e = -(j + 1) * ((i + 1) >> 1);
454 f = pow(2.0, e / 4.0);
455 k = i & 1;
456 is_table_lsf[j][k ^ 1][i] = FIXR(f);
457 is_table_lsf[j][k][i] = FIXR(1.0);
115329f1 458 dprintf("is_table_lsf %d %d: %x %x\n",
239c2f4c
FB
459 i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);
460 }
461 }
462
463 for(i=0;i<8;i++) {
464 float ci, cs, ca;
465 ci = ci_table[i];
466 cs = 1.0 / sqrt(1.0 + ci * ci);
467 ca = cs * ci;
ce4a29c0
MN
468 csa_table[i][0] = FIXHR(cs/4);
469 csa_table[i][1] = FIXHR(ca/4);
470 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
115329f1 471 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
a1e257b2
MN
472 csa_table_float[i][0] = cs;
473 csa_table_float[i][1] = ca;
474 csa_table_float[i][2] = ca + cs;
115329f1 475 csa_table_float[i][3] = ca - cs;
a1e257b2 476// printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));
711ae726 477// av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);
239c2f4c
FB
478 }
479
480 /* compute mdct windows */
481 for(i=0;i<36;i++) {
711ae726
MN
482 for(j=0; j<4; j++){
483 double d;
115329f1 484
125d6246
MN
485 if(j==2 && i%3 != 1)
486 continue;
115329f1 487
711ae726
MN
488 d= sin(M_PI * (i + 0.5) / 36.0);
489 if(j==1){
490 if (i>=30) d= 0;
491 else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
492 else if(i>=18) d= 1;
493 }else if(j==3){
494 if (i< 6) d= 0;
495 else if(i< 12) d= sin(M_PI * (i - 6 + 0.5) / 12.0);
496 else if(i< 18) d= 1;
497 }
498 //merge last stage of imdct into the window coefficients
125d6246
MN
499 d*= 0.5 / cos(M_PI*(2*i + 19)/72);
500
501 if(j==2)
502 mdct_win[j][i/3] = FIXHR((d / (1<<5)));
503 else
504 mdct_win[j][i ] = FIXHR((d / (1<<5)));
711ae726
MN
505// av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));
506 }
239c2f4c
FB
507 }
508
239c2f4c
FB
509 /* NOTE: we do frequency inversion adter the MDCT by changing
510 the sign of the right window coefs */
511 for(j=0;j<4;j++) {
512 for(i=0;i<36;i+=2) {
513 mdct_win[j + 4][i] = mdct_win[j][i];
514 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
515 }
516 }
517
518#if defined(DEBUG)
519 for(j=0;j<8;j++) {
267f7edc 520 av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);
239c2f4c 521 for(i=0;i<36;i++)
267f7edc
SH
522 av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);
523 av_log(avctx, AV_LOG_DEBUG, "\n");
239c2f4c
FB
524 }
525#endif
de6d9b64 526 init = 1;
de6d9b64
FB
527 }
528
529 s->inbuf_index = 0;
530 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
531 s->inbuf_ptr = s->inbuf;
239c2f4c
FB
532#ifdef DEBUG
533 s->frame_count = 0;
534#endif
1ede228a
RT
535 if (avctx->codec_id == CODEC_ID_MP3ADU)
536 s->adu_mode = 1;
de6d9b64
FB
537 return 0;
538}
539
ef9f7306 540/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */
239c2f4c
FB
541
542/* cos(i*pi/64) */
543
0bd2483a
MN
544#define COS0_0 FIXHR(0.50060299823519630134/2)
545#define COS0_1 FIXHR(0.50547095989754365998/2)
546#define COS0_2 FIXHR(0.51544730992262454697/2)
547#define COS0_3 FIXHR(0.53104259108978417447/2)
548#define COS0_4 FIXHR(0.55310389603444452782/2)
549#define COS0_5 FIXHR(0.58293496820613387367/2)
550#define COS0_6 FIXHR(0.62250412303566481615/2)
551#define COS0_7 FIXHR(0.67480834145500574602/2)
552#define COS0_8 FIXHR(0.74453627100229844977/2)
553#define COS0_9 FIXHR(0.83934964541552703873/2)
554#define COS0_10 FIXHR(0.97256823786196069369/2)
555#define COS0_11 FIXHR(1.16943993343288495515/4)
556#define COS0_12 FIXHR(1.48416461631416627724/4)
557#define COS0_13 FIXHR(2.05778100995341155085/8)
558#define COS0_14 FIXHR(3.40760841846871878570/8)
559#define COS0_15 FIXHR(10.19000812354805681150/32)
560
561#define COS1_0 FIXHR(0.50241928618815570551/2)
562#define COS1_1 FIXHR(0.52249861493968888062/2)
563#define COS1_2 FIXHR(0.56694403481635770368/2)
564#define COS1_3 FIXHR(0.64682178335999012954/2)
565#define COS1_4 FIXHR(0.78815462345125022473/2)
566#define COS1_5 FIXHR(1.06067768599034747134/4)
567#define COS1_6 FIXHR(1.72244709823833392782/4)
568#define COS1_7 FIXHR(5.10114861868916385802/16)
569
570#define COS2_0 FIXHR(0.50979557910415916894/2)
571#define COS2_1 FIXHR(0.60134488693504528054/2)
572#define COS2_2 FIXHR(0.89997622313641570463/2)
573#define COS2_3 FIXHR(2.56291544774150617881/8)
574
575#define COS3_0 FIXHR(0.54119610014619698439/2)
576#define COS3_1 FIXHR(1.30656296487637652785/4)
577
578#define COS4_0 FIXHR(0.70710678118654752439/2)
239c2f4c
FB
579
580/* butterfly operator */
0bd2483a 581#define BF(a, b, c, s)\
239c2f4c
FB
582{\
583 tmp0 = tab[a] + tab[b];\
584 tmp1 = tab[a] - tab[b];\
585 tab[a] = tmp0;\
0bd2483a 586 tab[b] = MULH(tmp1<<(s), c);\
239c2f4c
FB
587}
588
589#define BF1(a, b, c, d)\
590{\
0bd2483a
MN
591 BF(a, b, COS4_0, 1);\
592 BF(c, d,-COS4_0, 1);\
239c2f4c
FB
593 tab[c] += tab[d];\
594}
595
596#define BF2(a, b, c, d)\
597{\
0bd2483a
MN
598 BF(a, b, COS4_0, 1);\
599 BF(c, d,-COS4_0, 1);\
239c2f4c
FB
600 tab[c] += tab[d];\
601 tab[a] += tab[c];\
602 tab[c] += tab[b];\
603 tab[b] += tab[d];\
604}
605
606#define ADD(a, b) tab[a] += tab[b]
607
608/* DCT32 without 1/sqrt(2) coef zero scaling. */
0c1a9eda 609static void dct32(int32_t *out, int32_t *tab)
239c2f4c
FB
610{
611 int tmp0, tmp1;
612
613 /* pass 1 */
0bd2483a
MN
614 BF( 0, 31, COS0_0 , 1);
615 BF(15, 16, COS0_15, 5);
120aad7f 616 /* pass 2 */
0bd2483a
MN
617 BF( 0, 15, COS1_0 , 1);
618 BF(16, 31,-COS1_0 , 1);
120aad7f 619 /* pass 1 */
0bd2483a
MN
620 BF( 7, 24, COS0_7 , 1);
621 BF( 8, 23, COS0_8 , 1);
120aad7f 622 /* pass 2 */
0bd2483a
MN
623 BF( 7, 8, COS1_7 , 4);
624 BF(23, 24,-COS1_7 , 4);
120aad7f 625 /* pass 3 */
0bd2483a
MN
626 BF( 0, 7, COS2_0 , 1);
627 BF( 8, 15,-COS2_0 , 1);
628 BF(16, 23, COS2_0 , 1);
629 BF(24, 31,-COS2_0 , 1);
120aad7f 630 /* pass 1 */
0bd2483a
MN
631 BF( 3, 28, COS0_3 , 1);
632 BF(12, 19, COS0_12, 2);
239c2f4c 633 /* pass 2 */
0bd2483a
MN
634 BF( 3, 12, COS1_3 , 1);
635 BF(19, 28,-COS1_3 , 1);
120aad7f 636 /* pass 1 */
0bd2483a
MN
637 BF( 4, 27, COS0_4 , 1);
638 BF(11, 20, COS0_11, 2);
120aad7f 639 /* pass 2 */
0bd2483a
MN
640 BF( 4, 11, COS1_4 , 1);
641 BF(20, 27,-COS1_4 , 1);
120aad7f 642 /* pass 3 */
0bd2483a
MN
643 BF( 3, 4, COS2_3 , 3);
644 BF(11, 12,-COS2_3 , 3);
645 BF(19, 20, COS2_3 , 3);
646 BF(27, 28,-COS2_3 , 3);
120aad7f 647 /* pass 4 */
0bd2483a
MN
648 BF( 0, 3, COS3_0 , 1);
649 BF( 4, 7,-COS3_0 , 1);
650 BF( 8, 11, COS3_0 , 1);
651 BF(12, 15,-COS3_0 , 1);
652 BF(16, 19, COS3_0 , 1);
653 BF(20, 23,-COS3_0 , 1);
654 BF(24, 27, COS3_0 , 1);
655 BF(28, 31,-COS3_0 , 1);
115329f1 656
120aad7f
MN
657
658
659 /* pass 1 */
0bd2483a
MN
660 BF( 1, 30, COS0_1 , 1);
661 BF(14, 17, COS0_14, 3);
120aad7f 662 /* pass 2 */
0bd2483a
MN
663 BF( 1, 14, COS1_1 , 1);
664 BF(17, 30,-COS1_1 , 1);
120aad7f 665 /* pass 1 */
0bd2483a
MN
666 BF( 6, 25, COS0_6 , 1);
667 BF( 9, 22, COS0_9 , 1);
120aad7f 668 /* pass 2 */
0bd2483a
MN
669 BF( 6, 9, COS1_6 , 2);
670 BF(22, 25,-COS1_6 , 2);
239c2f4c 671 /* pass 3 */
0bd2483a
MN
672 BF( 1, 6, COS2_1 , 1);
673 BF( 9, 14,-COS2_1 , 1);
674 BF(17, 22, COS2_1 , 1);
675 BF(25, 30,-COS2_1 , 1);
239c2f4c 676
120aad7f 677 /* pass 1 */
0bd2483a
MN
678 BF( 2, 29, COS0_2 , 1);
679 BF(13, 18, COS0_13, 3);
120aad7f 680 /* pass 2 */
0bd2483a
MN
681 BF( 2, 13, COS1_2 , 1);
682 BF(18, 29,-COS1_2 , 1);
120aad7f 683 /* pass 1 */
0bd2483a
MN
684 BF( 5, 26, COS0_5 , 1);
685 BF(10, 21, COS0_10, 1);
120aad7f 686 /* pass 2 */
0bd2483a
MN
687 BF( 5, 10, COS1_5 , 2);
688 BF(21, 26,-COS1_5 , 2);
120aad7f 689 /* pass 3 */
0bd2483a
MN
690 BF( 2, 5, COS2_2 , 1);
691 BF(10, 13,-COS2_2 , 1);
692 BF(18, 21, COS2_2 , 1);
693 BF(26, 29,-COS2_2 , 1);
239c2f4c 694 /* pass 4 */
0bd2483a
MN
695 BF( 1, 2, COS3_1 , 2);
696 BF( 5, 6,-COS3_1 , 2);
697 BF( 9, 10, COS3_1 , 2);
698 BF(13, 14,-COS3_1 , 2);
699 BF(17, 18, COS3_1 , 2);
700 BF(21, 22,-COS3_1 , 2);
701 BF(25, 26, COS3_1 , 2);
702 BF(29, 30,-COS3_1 , 2);
115329f1 703
239c2f4c 704 /* pass 5 */
0bd2483a
MN
705 BF1( 0, 1, 2, 3);
706 BF2( 4, 5, 6, 7);
707 BF1( 8, 9, 10, 11);
239c2f4c
FB
708 BF2(12, 13, 14, 15);
709 BF1(16, 17, 18, 19);
710 BF2(20, 21, 22, 23);
711 BF1(24, 25, 26, 27);
712 BF2(28, 29, 30, 31);
115329f1 713
239c2f4c 714 /* pass 6 */
115329f1 715
239c2f4c
FB
716 ADD( 8, 12);
717 ADD(12, 10);
718 ADD(10, 14);
719 ADD(14, 9);
720 ADD( 9, 13);
721 ADD(13, 11);
722 ADD(11, 15);
723
724 out[ 0] = tab[0];
725 out[16] = tab[1];
726 out[ 8] = tab[2];
727 out[24] = tab[3];
728 out[ 4] = tab[4];
729 out[20] = tab[5];
730 out[12] = tab[6];
731 out[28] = tab[7];
732 out[ 2] = tab[8];
733 out[18] = tab[9];
734 out[10] = tab[10];
735 out[26] = tab[11];
736 out[ 6] = tab[12];
737 out[22] = tab[13];
738 out[14] = tab[14];
739 out[30] = tab[15];
115329f1 740
239c2f4c
FB
741 ADD(24, 28);
742 ADD(28, 26);
743 ADD(26, 30);
744 ADD(30, 25);
745 ADD(25, 29);
746 ADD(29, 27);
747 ADD(27, 31);
748
749 out[ 1] = tab[16] + tab[24];
750 out[17] = tab[17] + tab[25];
751 out[ 9] = tab[18] + tab[26];
752 out[25] = tab[19] + tab[27];
753 out[ 5] = tab[20] + tab[28];
754 out[21] = tab[21] + tab[29];
755 out[13] = tab[22] + tab[30];
756 out[29] = tab[23] + tab[31];
757 out[ 3] = tab[24] + tab[20];
758 out[19] = tab[25] + tab[21];
759 out[11] = tab[26] + tab[22];
760 out[27] = tab[27] + tab[23];
761 out[ 7] = tab[28] + tab[18];
762 out[23] = tab[29] + tab[19];
763 out[15] = tab[30] + tab[17];
764 out[31] = tab[31];
765}
766
239c2f4c
FB
767#if FRAC_BITS <= 15
768
a7a85899 769static inline int round_sample(int *sum)
8c5b5683
FB
770{
771 int sum1;
a7a85899
MN
772 sum1 = (*sum) >> OUT_SHIFT;
773 *sum &= (1<<OUT_SHIFT)-1;
a3a5f4d6
MN
774 if (sum1 < OUT_MIN)
775 sum1 = OUT_MIN;
776 else if (sum1 > OUT_MAX)
777 sum1 = OUT_MAX;
8c5b5683 778 return sum1;
239c2f4c
FB
779}
780
5ab1972b
MN
781# if defined(ARCH_POWERPC_405)
782 /* signed 16x16 -> 32 multiply add accumulate */
783# define MACS(rt, ra, rb) \
784 asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
785
786 /* signed 16x16 -> 32 multiply */
787# define MULS(ra, rb) \
788 ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })
789# else
790 /* signed 16x16 -> 32 multiply add accumulate */
791# define MACS(rt, ra, rb) rt += (ra) * (rb)
792
793 /* signed 16x16 -> 32 multiply */
794# define MULS(ra, rb) ((ra) * (rb))
795# endif
8c5b5683
FB
796#else
797
115329f1 798static inline int round_sample(int64_t *sum)
8c5b5683
FB
799{
800 int sum1;
a7a85899
MN
801 sum1 = (int)((*sum) >> OUT_SHIFT);
802 *sum &= (1<<OUT_SHIFT)-1;
a3a5f4d6
MN
803 if (sum1 < OUT_MIN)
804 sum1 = OUT_MIN;
805 else if (sum1 > OUT_MAX)
806 sum1 = OUT_MAX;
8c5b5683 807 return sum1;
239c2f4c
FB
808}
809
47a0cd74 810# define MULS(ra, rb) MUL64(ra, rb)
8c5b5683
FB
811#endif
812
813#define SUM8(sum, op, w, p) \
239c2f4c 814{ \
8c5b5683
FB
815 sum op MULS((w)[0 * 64], p[0 * 64]);\
816 sum op MULS((w)[1 * 64], p[1 * 64]);\
817 sum op MULS((w)[2 * 64], p[2 * 64]);\
818 sum op MULS((w)[3 * 64], p[3 * 64]);\
819 sum op MULS((w)[4 * 64], p[4 * 64]);\
820 sum op MULS((w)[5 * 64], p[5 * 64]);\
821 sum op MULS((w)[6 * 64], p[6 * 64]);\
822 sum op MULS((w)[7 * 64], p[7 * 64]);\
823}
824
825#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
826{ \
827 int tmp;\
828 tmp = p[0 * 64];\
829 sum1 op1 MULS((w1)[0 * 64], tmp);\
830 sum2 op2 MULS((w2)[0 * 64], tmp);\
831 tmp = p[1 * 64];\
832 sum1 op1 MULS((w1)[1 * 64], tmp);\
833 sum2 op2 MULS((w2)[1 * 64], tmp);\
834 tmp = p[2 * 64];\
835 sum1 op1 MULS((w1)[2 * 64], tmp);\
836 sum2 op2 MULS((w2)[2 * 64], tmp);\
837 tmp = p[3 * 64];\
838 sum1 op1 MULS((w1)[3 * 64], tmp);\
839 sum2 op2 MULS((w2)[3 * 64], tmp);\
840 tmp = p[4 * 64];\
841 sum1 op1 MULS((w1)[4 * 64], tmp);\
842 sum2 op2 MULS((w2)[4 * 64], tmp);\
843 tmp = p[5 * 64];\
844 sum1 op1 MULS((w1)[5 * 64], tmp);\
845 sum2 op2 MULS((w2)[5 * 64], tmp);\
846 tmp = p[6 * 64];\
847 sum1 op1 MULS((w1)[6 * 64], tmp);\
848 sum2 op2 MULS((w2)[6 * 64], tmp);\
849 tmp = p[7 * 64];\
850 sum1 op1 MULS((w1)[7 * 64], tmp);\
851 sum2 op2 MULS((w2)[7 * 64], tmp);\
239c2f4c
FB
852}
853
bf1f4da0
AB
854void ff_mpa_synth_init(MPA_INT *window)
855{
856 int i;
857
858 /* max = 18760, max sum over all 16 coefs : 44736 */
859 for(i=0;i<257;i++) {
860 int v;
861 v = mpa_enwindow[i];
862#if WFRAC_BITS < 16
863 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
864#endif
865 window[i] = v;
866 if ((i & 63) != 0)
867 v = -v;
868 if (i != 0)
869 window[512 - i] = v;
115329f1 870 }
bf1f4da0 871}
239c2f4c
FB
872
873/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
874 32 samples. */
875/* XXX: optimize by avoiding ring buffer usage */
bf1f4da0 876void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
bb270c08 877 MPA_INT *window, int *dither_state,
115329f1 878 OUT_INT *samples, int incr,
0c1a9eda 879 int32_t sb_samples[SBLIMIT])
239c2f4c 880{
0c1a9eda 881 int32_t tmp[32];
8c5b5683 882 register MPA_INT *synth_buf;
491c4a10 883 register const MPA_INT *w, *w2, *p;
239c2f4c 884 int j, offset, v;
a3a5f4d6 885 OUT_INT *samples2;
239c2f4c 886#if FRAC_BITS <= 15
8c5b5683 887 int sum, sum2;
239c2f4c 888#else
8c5b5683 889 int64_t sum, sum2;
239c2f4c 890#endif
bf1f4da0 891
239c2f4c 892 dct32(tmp, sb_samples);
115329f1 893
bf1f4da0
AB
894 offset = *synth_buf_offset;
895 synth_buf = synth_buf_ptr + offset;
239c2f4c
FB
896
897 for(j=0;j<32;j++) {
898 v = tmp[j];
899#if FRAC_BITS <= 15
81552334
FB
900 /* NOTE: can cause a loss in precision if very high amplitude
901 sound */
239c2f4c
FB
902 if (v > 32767)
903 v = 32767;
904 else if (v < -32768)
905 v = -32768;
906#endif
907 synth_buf[j] = v;
908 }
909 /* copy to avoid wrap */
910 memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));
911
8c5b5683 912 samples2 = samples + 31 * incr;
239c2f4c 913 w = window;
8c5b5683
FB
914 w2 = window + 31;
915
093c6e50 916 sum = *dither_state;
8c5b5683
FB
917 p = synth_buf + 16;
918 SUM8(sum, +=, w, p);
919 p = synth_buf + 48;
920 SUM8(sum, -=, w + 32, p);
a7a85899 921 *samples = round_sample(&sum);
8c5b5683 922 samples += incr;
239c2f4c
FB
923 w++;
924
8c5b5683
FB
925 /* we calculate two samples at the same time to avoid one memory
926 access per two sample */
927 for(j=1;j<16;j++) {
8c5b5683
FB
928 sum2 = 0;
929 p = synth_buf + 16 + j;
930 SUM8P2(sum, +=, sum2, -=, w, w2, p);
931 p = synth_buf + 48 - j;
932 SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);
933
a7a85899 934 *samples = round_sample(&sum);
8c5b5683 935 samples += incr;
a7a85899
MN
936 sum += sum2;
937 *samples2 = round_sample(&sum);
8c5b5683 938 samples2 -= incr;
239c2f4c 939 w++;
8c5b5683 940 w2--;
239c2f4c 941 }
115329f1 942
8c5b5683 943 p = synth_buf + 32;
8c5b5683 944 SUM8(sum, -=, w + 32, p);
a7a85899 945 *samples = round_sample(&sum);
093c6e50 946 *dither_state= sum;
8c5b5683 947
239c2f4c 948 offset = (offset - 32) & 511;
bf1f4da0 949 *synth_buf_offset = offset;
239c2f4c
FB
950}
951
125d6246
MN
952#define C3 FIXHR(0.86602540378443864676/2)
953
954/* 0.5 / cos(pi*(2*i+1)/36) */
955static const int icos36[9] = {
956 FIXR(0.50190991877167369479),
957 FIXR(0.51763809020504152469), //0
958 FIXR(0.55168895948124587824),
959 FIXR(0.61038729438072803416),
960 FIXR(0.70710678118654752439), //1
961 FIXR(0.87172339781054900991),
962 FIXR(1.18310079157624925896),
963 FIXR(1.93185165257813657349), //2
964 FIXR(5.73685662283492756461),
965};
239c2f4c 966
eb644776
MN
967/* 0.5 / cos(pi*(2*i+1)/36) */
968static const int icos36h[9] = {
969 FIXHR(0.50190991877167369479/2),
970 FIXHR(0.51763809020504152469/2), //0
971 FIXHR(0.55168895948124587824/2),
972 FIXHR(0.61038729438072803416/2),
973 FIXHR(0.70710678118654752439/2), //1
974 FIXHR(0.87172339781054900991/2),
975 FIXHR(1.18310079157624925896/4),
976 FIXHR(1.93185165257813657349/4), //2
977// FIXHR(5.73685662283492756461),
978};
979
239c2f4c
FB
980/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
981 cases. */
982static void imdct12(int *out, int *in)
983{
125d6246 984 int in0, in1, in2, in3, in4, in5, t1, t2;
44f1698a
MN
985
986 in0= in[0*3];
987 in1= in[1*3] + in[0*3];
988 in2= in[2*3] + in[1*3];
989 in3= in[3*3] + in[2*3];
990 in4= in[4*3] + in[3*3];
991 in5= in[5*3] + in[4*3];
125d6246
MN
992 in5 += in3;
993 in3 += in1;
994
995 in2= MULH(2*in2, C3);
eb644776 996 in3= MULH(4*in3, C3);
115329f1 997
125d6246 998 t1 = in0 - in4;
eb644776 999 t2 = MULH(2*(in1 - in5), icos36h[4]);
125d6246 1000
115329f1 1001 out[ 7]=
125d6246
MN
1002 out[10]= t1 + t2;
1003 out[ 1]=
1004 out[ 4]= t1 - t2;
1005
1006 in0 += in4>>1;
1007 in4 = in0 + in2;
eb644776
MN
1008 in5 += 2*in1;
1009 in1 = MULH(in5 + in3, icos36h[1]);
115329f1 1010 out[ 8]=
eb644776 1011 out[ 9]= in4 + in1;
125d6246 1012 out[ 2]=
eb644776 1013 out[ 3]= in4 - in1;
115329f1 1014
125d6246 1015 in0 -= in2;
eb644776 1016 in5 = MULH(2*(in5 - in3), icos36h[7]);
125d6246 1017 out[ 0]=
eb644776 1018 out[ 5]= in0 - in5;
125d6246 1019 out[ 6]=
eb644776 1020 out[11]= in0 + in5;
239c2f4c
FB
1021}
1022
239c2f4c 1023/* cos(pi*i/18) */
711ae726
MN
1024#define C1 FIXHR(0.98480775301220805936/2)
1025#define C2 FIXHR(0.93969262078590838405/2)
1026#define C3 FIXHR(0.86602540378443864676/2)
1027#define C4 FIXHR(0.76604444311897803520/2)
1028#define C5 FIXHR(0.64278760968653932632/2)
1029#define C6 FIXHR(0.5/2)
1030#define C7 FIXHR(0.34202014332566873304/2)
1031#define C8 FIXHR(0.17364817766693034885/2)
1032
239c2f4c 1033
239c2f4c 1034/* using Lee like decomposition followed by hand coded 9 points DCT */
711ae726 1035static void imdct36(int *out, int *buf, int *in, int *win)
239c2f4c
FB
1036{
1037 int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
1038 int tmp[18], *tmp1, *in1;
239c2f4c
FB
1039
1040 for(i=17;i>=1;i--)
1041 in[i] += in[i-1];
1042 for(i=17;i>=3;i-=2)
1043 in[i] += in[i-2];
1044
1045 for(j=0;j<2;j++) {
1046 tmp1 = tmp + j;
1047 in1 = in + j;
711ae726
MN
1048#if 0
1049//more accurate but slower
1050 int64_t t0, t1, t2, t3;
1051 t2 = in1[2*4] + in1[2*8] - in1[2*2];
115329f1 1052
711ae726
MN
1053 t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
1054 t1 = in1[2*0] - in1[2*6];
1055 tmp1[ 6] = t1 - (t2>>1);
1056 tmp1[16] = t1 + t2;
1057
1058 t0 = MUL64(2*(in1[2*2] + in1[2*4]), C2);
1059 t1 = MUL64( in1[2*4] - in1[2*8] , -2*C8);
1060 t2 = MUL64(2*(in1[2*2] + in1[2*8]), -C4);
115329f1 1061
711ae726
MN
1062 tmp1[10] = (t3 - t0 - t2) >> 32;
1063 tmp1[ 2] = (t3 + t0 + t1) >> 32;
1064 tmp1[14] = (t3 + t2 - t1) >> 32;
115329f1 1065
711ae726
MN
1066 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1067 t2 = MUL64(2*(in1[2*1] + in1[2*5]), C1);
1068 t3 = MUL64( in1[2*5] - in1[2*7] , -2*C7);
1069 t0 = MUL64(2*in1[2*3], C3);
1070
1071 t1 = MUL64(2*(in1[2*1] + in1[2*7]), -C5);
1072
1073 tmp1[ 0] = (t2 + t3 + t0) >> 32;
1074 tmp1[12] = (t2 + t1 - t0) >> 32;
1075 tmp1[ 8] = (t3 - t1 - t0) >> 32;
1076#else
1077 t2 = in1[2*4] + in1[2*8] - in1[2*2];
115329f1 1078
711ae726
MN
1079 t3 = in1[2*0] + (in1[2*6]>>1);
1080 t1 = in1[2*0] - in1[2*6];
1081 tmp1[ 6] = t1 - (t2>>1);
1082 tmp1[16] = t1 + t2;
1083
1084 t0 = MULH(2*(in1[2*2] + in1[2*4]), C2);
1085 t1 = MULH( in1[2*4] - in1[2*8] , -2*C8);
1086 t2 = MULH(2*(in1[2*2] + in1[2*8]), -C4);
115329f1 1087
711ae726
MN
1088 tmp1[10] = t3 - t0 - t2;
1089 tmp1[ 2] = t3 + t0 + t1;
1090 tmp1[14] = t3 + t2 - t1;
115329f1 1091
711ae726
MN
1092 tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
1093 t2 = MULH(2*(in1[2*1] + in1[2*5]), C1);
1094 t3 = MULH( in1[2*5] - in1[2*7] , -2*C7);
1095 t0 = MULH(2*in1[2*3], C3);
239c2f4c 1096
711ae726
MN
1097 t1 = MULH(2*(in1[2*1] + in1[2*7]), -C5);
1098
1099 tmp1[ 0] = t2 + t3 + t0;
1100 tmp1[12] = t2 + t1 - t0;
1101 tmp1[ 8] = t3 - t1 - t0;
1102#endif
239c2f4c
FB
1103 }
1104
1105 i = 0;
1106 for(j=0;j<4;j++) {
1107 t0 = tmp[i];
1108 t1 = tmp[i + 2];
1109 s0 = t1 + t0;
1110 s2 = t1 - t0;
1111
1112 t2 = tmp[i + 1];
1113 t3 = tmp[i + 3];
eb644776 1114 s1 = MULH(2*(t3 + t2), icos36h[j]);
239c2f4c 1115 s3 = MULL(t3 - t2, icos36[8 - j]);
115329f1 1116
44f1698a
MN
1117 t0 = s0 + s1;
1118 t1 = s0 - s1;
125d6246 1119 out[(9 + j)*SBLIMIT] = MULH(t1, win[9 + j]) + buf[9 + j];
711ae726
MN
1120 out[(8 - j)*SBLIMIT] = MULH(t1, win[8 - j]) + buf[8 - j];
1121 buf[9 + j] = MULH(t0, win[18 + 9 + j]);
1122 buf[8 - j] = MULH(t0, win[18 + 8 - j]);
115329f1 1123
44f1698a
MN
1124 t0 = s2 + s3;
1125 t1 = s2 - s3;
125d6246 1126 out[(9 + 8 - j)*SBLIMIT] = MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
711ae726
MN
1127 out[( j)*SBLIMIT] = MULH(t1, win[ j]) + buf[ j];
1128 buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
1129 buf[ + j] = MULH(t0, win[18 + j]);
239c2f4c
FB
1130 i += 4;
1131 }
1132
1133 s0 = tmp[16];
eb644776 1134 s1 = MULH(2*tmp[17], icos36h[4]);
44f1698a
MN
1135 t0 = s0 + s1;
1136 t1 = s0 - s1;
125d6246 1137 out[(9 + 4)*SBLIMIT] = MULH(t1, win[9 + 4]) + buf[9 + 4];
711ae726
MN
1138 out[(8 - 4)*SBLIMIT] = MULH(t1, win[8 - 4]) + buf[8 - 4];
1139 buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
1140 buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
239c2f4c
FB
1141}
1142
de6d9b64 1143/* header decoding. MUST check the header before because no
239c2f4c
FB
1144 consistency check is done there. Return 1 if free format found and
1145 that the frame size must be computed externally */
0c1a9eda 1146static int decode_header(MPADecodeContext *s, uint32_t header)
de6d9b64 1147{
239c2f4c
FB
1148 int sample_rate, frame_size, mpeg25, padding;
1149 int sample_rate_index, bitrate_index;
de6d9b64 1150 if (header & (1<<20)) {
239c2f4c
FB
1151 s->lsf = (header & (1<<19)) ? 0 : 1;
1152 mpeg25 = 0;
de6d9b64 1153 } else {
239c2f4c
FB
1154 s->lsf = 1;
1155 mpeg25 = 1;
de6d9b64 1156 }
115329f1 1157
de6d9b64
FB
1158 s->layer = 4 - ((header >> 17) & 3);
1159 /* extract frequency */
239c2f4c
FB
1160 sample_rate_index = (header >> 10) & 3;
1161 sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);
1162 sample_rate_index += 3 * (s->lsf + mpeg25);
1163 s->sample_rate_index = sample_rate_index;
1164 s->error_protection = ((header >> 16) & 1) ^ 1;
81552334 1165 s->sample_rate = sample_rate;
de6d9b64 1166
239c2f4c
FB
1167 bitrate_index = (header >> 12) & 0xf;
1168 padding = (header >> 9) & 1;
1169 //extension = (header >> 8) & 1;
1170 s->mode = (header >> 6) & 3;
1171 s->mode_ext = (header >> 4) & 3;
1172 //copyright = (header >> 3) & 1;
1173 //original = (header >> 2) & 1;
1174 //emphasis = header & 3;
de6d9b64 1175
239c2f4c
FB
1176 if (s->mode == MPA_MONO)
1177 s->nb_channels = 1;
1178 else
1179 s->nb_channels = 2;
115329f1 1180
239c2f4c
FB
1181 if (bitrate_index != 0) {
1182 frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];
1183 s->bit_rate = frame_size * 1000;
1184 switch(s->layer) {
1185 case 1:
1186 frame_size = (frame_size * 12000) / sample_rate;
1187 frame_size = (frame_size + padding) * 4;
1188 break;
1189 case 2:
1190 frame_size = (frame_size * 144000) / sample_rate;
1191 frame_size += padding;
1192 break;
1193 default:
1194 case 3:
1195 frame_size = (frame_size * 144000) / (sample_rate << s->lsf);
1196 frame_size += padding;
1197 break;
1198 }
1199 s->frame_size = frame_size;
1200 } else {
1201 /* if no frame size computed, signal it */
1202 if (!s->free_format_frame_size)
1203 return 1;
1204 /* free format: compute bitrate and real frame size from the
1205 frame size we extracted by reading the bitstream */
1206 s->frame_size = s->free_format_frame_size;
1207 switch(s->layer) {
1208 case 1:
1209 s->frame_size += padding * 4;
1210 s->bit_rate = (s->frame_size * sample_rate) / 48000;
1211 break;
1212 case 2:
1213 s->frame_size += padding;
1214 s->bit_rate = (s->frame_size * sample_rate) / 144000;
1215 break;
1216 default:
1217 case 3:
1218 s->frame_size += padding;
1219 s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;
1220 break;
1221 }
de6d9b64 1222 }
115329f1 1223
fad9f495 1224#if defined(DEBUG)
267f7edc 1225 dprintf("layer%d, %d Hz, %d kbits/s, ",
239c2f4c
FB
1226 s->layer, s->sample_rate, s->bit_rate);
1227 if (s->nb_channels == 2) {
1228 if (s->layer == 3) {
1229 if (s->mode_ext & MODE_EXT_MS_STEREO)
267f7edc 1230 dprintf("ms-");
239c2f4c 1231 if (s->mode_ext & MODE_EXT_I_STEREO)
267f7edc 1232 dprintf("i-");
239c2f4c 1233 }
267f7edc 1234 dprintf("stereo");
239c2f4c 1235 } else {
267f7edc 1236 dprintf("mono");
239c2f4c 1237 }
267f7edc 1238 dprintf("\n");
de6d9b64 1239#endif
239c2f4c 1240 return 0;
de6d9b64
FB
1241}
1242
8c5b5683 1243/* useful helper to get mpeg audio stream infos. Return -1 if error in
962d6ae6
FB
1244 header, otherwise the coded frame size in bytes */
1245int mpa_decode_header(AVCodecContext *avctx, uint32_t head)
8c5b5683
FB
1246{
1247 MPADecodeContext s1, *s = &s1;
2caa92d9 1248 memset( s, 0, sizeof(MPADecodeContext) );
8c5b5683 1249
a7a85899 1250 if (ff_mpa_check_header(head) != 0)
8c5b5683
FB
1251 return -1;
1252
1253 if (decode_header(s, head) != 0) {
1254 return -1;
1255 }
1256
1257 switch(s->layer) {
1258 case 1:
962d6ae6 1259 avctx->frame_size = 384;
8c5b5683
FB
1260 break;
1261 case 2:
962d6ae6 1262 avctx->frame_size = 1152;
8c5b5683
FB
1263 break;
1264 default:
1265 case 3:
1266 if (s->lsf)
962d6ae6 1267 avctx->frame_size = 576;
8c5b5683 1268 else
962d6ae6 1269 avctx->frame_size = 1152;
8c5b5683
FB
1270 break;
1271 }
1272
962d6ae6
FB
1273 avctx->sample_rate = s->sample_rate;
1274 avctx->channels = s->nb_channels;
1275 avctx->bit_rate = s->bit_rate;
1276 avctx->sub_id = s->layer;
1277 return s->frame_size;
8c5b5683
FB
1278}
1279
239c2f4c
FB
1280/* return the number of decoded frames */
1281static int mp_decode_layer1(MPADecodeContext *s)
de6d9b64 1282{
239c2f4c 1283 int bound, i, v, n, ch, j, mant;
0c1a9eda
ZK
1284 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
1285 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
239c2f4c 1286
115329f1 1287 if (s->mode == MPA_JSTEREO)
239c2f4c
FB
1288 bound = (s->mode_ext + 1) * 4;
1289 else
1290 bound = SBLIMIT;
1291
1292 /* allocation bits */
1293 for(i=0;i<bound;i++) {
1294 for(ch=0;ch<s->nb_channels;ch++) {
1295 allocation[ch][i] = get_bits(&s->gb, 4);
1296 }
1297 }
1298 for(i=bound;i<SBLIMIT;i++) {
1299 allocation[0][i] = get_bits(&s->gb, 4);
1300 }
1301
1302 /* scale factors */
1303 for(i=0;i<bound;i++) {
1304 for(ch=0;ch<s->nb_channels;ch++) {
1305 if (allocation[ch][i])
1306 scale_factors[ch][i] = get_bits(&s->gb, 6);
1307 }
1308 }
1309 for(i=bound;i<SBLIMIT;i++) {
1310 if (allocation[0][i]) {
1311 scale_factors[0][i] = get_bits(&s->gb, 6);
1312 scale_factors[1][i] = get_bits(&s->gb, 6);
1313 }
1314 }
115329f1 1315
239c2f4c
FB
1316 /* compute samples */
1317 for(j=0;j<12;j++) {
1318 for(i=0;i<bound;i++) {
1319 for(ch=0;ch<s->nb_channels;ch++) {
1320 n = allocation[ch][i];
1321 if (n) {
1322 mant = get_bits(&s->gb, n + 1);
1323 v = l1_unscale(n, mant, scale_factors[ch][i]);
1324 } else {
1325 v = 0;
1326 }
1327 s->sb_samples[ch][j][i] = v;
1328 }
1329 }
1330 for(i=bound;i<SBLIMIT;i++) {
1331 n = allocation[0][i];
1332 if (n) {
1333 mant = get_bits(&s->gb, n + 1);
1334 v = l1_unscale(n, mant, scale_factors[0][i]);
1335 s->sb_samples[0][j][i] = v;
1336 v = l1_unscale(n, mant, scale_factors[1][i]);
1337 s->sb_samples[1][j][i] = v;
1338 } else {
1339 s->sb_samples[0][j][i] = 0;
1340 s->sb_samples[1][j][i] = 0;
1341 }
1342 }
1343 }
1344 return 12;
1345}
1346
1347/* bitrate is in kb/s */
1348int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)
1349{
1350 int ch_bitrate, table;
115329f1 1351
239c2f4c
FB
1352 ch_bitrate = bitrate / nb_channels;
1353 if (!lsf) {
1354 if ((freq == 48000 && ch_bitrate >= 56) ||
115329f1 1355 (ch_bitrate >= 56 && ch_bitrate <= 80))
239c2f4c 1356 table = 0;
115329f1 1357 else if (freq != 48000 && ch_bitrate >= 96)
239c2f4c 1358 table = 1;
115329f1 1359 else if (freq != 32000 && ch_bitrate <= 48)
239c2f4c 1360 table = 2;
115329f1 1361 else
239c2f4c
FB
1362 table = 3;
1363 } else {
1364 table = 4;
1365 }
1366 return table;
1367}
de6d9b64 1368
239c2f4c
FB
1369static int mp_decode_layer2(MPADecodeContext *s)
1370{
1371 int sblimit; /* number of used subbands */
1372 const unsigned char *alloc_table;
1373 int table, bit_alloc_bits, i, j, ch, bound, v;
1374 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
1375 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
1376 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
1377 int scale, qindex, bits, steps, k, l, m, b;
de6d9b64 1378
239c2f4c 1379 /* select decoding table */
115329f1 1380 table = l2_select_table(s->bit_rate / 1000, s->nb_channels,
239c2f4c
FB
1381 s->sample_rate, s->lsf);
1382 sblimit = sblimit_table[table];
1383 alloc_table = alloc_tables[table];
1384
115329f1 1385 if (s->mode == MPA_JSTEREO)
239c2f4c
FB
1386 bound = (s->mode_ext + 1) * 4;
1387 else
1388 bound = sblimit;
1389
1390 dprintf("bound=%d sblimit=%d\n", bound, sblimit);
2caa92d9
MN
1391
1392 /* sanity check */
1393 if( bound > sblimit ) bound = sblimit;
1394
239c2f4c
FB
1395 /* parse bit allocation */
1396 j = 0;
1397 for(i=0;i<bound;i++) {
1398 bit_alloc_bits = alloc_table[j];
1399 for(ch=0;ch<s->nb_channels;ch++) {
1400 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
1401 }
1402 j += 1 << bit_alloc_bits;
1403 }
1404 for(i=bound;i<sblimit;i++) {
1405 bit_alloc_bits = alloc_table[j];
1406 v = get_bits(&s->gb, bit_alloc_bits);
1407 bit_alloc[0][i] = v;
1408 bit_alloc[1][i] = v;
1409 j += 1 << bit_alloc_bits;
de6d9b64 1410 }
239c2f4c
FB
1411
1412#ifdef DEBUG
1413 {
1414 for(ch=0;ch<s->nb_channels;ch++) {
1415 for(i=0;i<sblimit;i++)
267f7edc
SH
1416 dprintf(" %d", bit_alloc[ch][i]);
1417 dprintf("\n");
239c2f4c
FB
1418 }
1419 }
1420#endif
1421
1422 /* scale codes */
1423 for(i=0;i<sblimit;i++) {
1424 for(ch=0;ch<s->nb_channels;ch++) {
115329f1 1425 if (bit_alloc[ch][i])
239c2f4c
FB
1426 scale_code[ch][i] = get_bits(&s->gb, 2);
1427 }
1428 }
115329f1 1429
239c2f4c
FB
1430 /* scale factors */
1431 for(i=0;i<sblimit;i++) {
1432 for(ch=0;ch<s->nb_channels;ch++) {
1433 if (bit_alloc[ch][i]) {
1434 sf = scale_factors[ch][i];
1435 switch(scale_code[ch][i]) {
1436 default:
1437 case 0:
1438 sf[0] = get_bits(&s->gb, 6);
1439 sf[1] = get_bits(&s->gb, 6);
1440 sf[2] = get_bits(&s->gb, 6);
1441 break;
1442 case 2:
1443 sf[0] = get_bits(&s->gb, 6);
1444 sf[1] = sf[0];
1445 sf[2] = sf[0];
1446 break;
1447 case 1:
1448 sf[0] = get_bits(&s->gb, 6);
1449 sf[2] = get_bits(&s->gb, 6);
1450 sf[1] = sf[0];
1451 break;
1452 case 3:
1453 sf[0] = get_bits(&s->gb, 6);
1454 sf[2] = get_bits(&s->gb, 6);
1455 sf[1] = sf[2];
1456 break;
1457 }
1458 }
1459 }
1460 }
1461
1462#ifdef DEBUG
1463 for(ch=0;ch<s->nb_channels;ch++) {
1464 for(i=0;i<sblimit;i++) {
1465 if (bit_alloc[ch][i]) {
1466 sf = scale_factors[ch][i];
267f7edc 1467 dprintf(" %d %d %d", sf[0], sf[1], sf[2]);
239c2f4c 1468 } else {
267f7edc 1469 dprintf(" -");
239c2f4c
FB
1470 }
1471 }
267f7edc 1472 dprintf("\n");
239c2f4c
FB
1473 }
1474#endif
1475
1476 /* samples */
1477 for(k=0;k<3;k++) {
1478 for(l=0;l<12;l+=3) {
1479 j = 0;
1480 for(i=0;i<bound;i++) {
1481 bit_alloc_bits = alloc_table[j];
1482 for(ch=0;ch<s->nb_channels;ch++) {
1483 b = bit_alloc[ch][i];
1484 if (b) {
1485 scale = scale_factors[ch][i][k];
1486 qindex = alloc_table[j+b];
1487 bits = quant_bits[qindex];
1488 if (bits < 0) {
1489 /* 3 values at the same time */
1490 v = get_bits(&s->gb, -bits);
1491 steps = quant_steps[qindex];
115329f1 1492 s->sb_samples[ch][k * 12 + l + 0][i] =
239c2f4c
FB
1493 l2_unscale_group(steps, v % steps, scale);
1494 v = v / steps;
115329f1 1495 s->sb_samples[ch][k * 12 + l + 1][i] =
239c2f4c
FB
1496 l2_unscale_group(steps, v % steps, scale);
1497 v = v / steps;
115329f1 1498 s->sb_samples[ch][k * 12 + l + 2][i] =
239c2f4c
FB
1499 l2_unscale_group(steps, v, scale);
1500 } else {
1501 for(m=0;m<3;m++) {
1502 v = get_bits(&s->gb, bits);
1503 v = l1_unscale(bits - 1, v, scale);
1504 s->sb_samples[ch][k * 12 + l + m][i] = v;
1505 }
1506 }
1507 } else {
1508 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1509 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1510 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1511 }
1512 }
1513 /* next subband in alloc table */
115329f1 1514 j += 1 << bit_alloc_bits;
239c2f4c
FB
1515 }
1516 /* XXX: find a way to avoid this duplication of code */
1517 for(i=bound;i<sblimit;i++) {
1518 bit_alloc_bits = alloc_table[j];
1519 b = bit_alloc[0][i];
1520 if (b) {
1521 int mant, scale0, scale1;
1522 scale0 = scale_factors[0][i][k];
1523 scale1 = scale_factors[1][i][k];
1524 qindex = alloc_table[j+b];
1525 bits = quant_bits[qindex];
1526 if (bits < 0) {
1527 /* 3 values at the same time */
1528 v = get_bits(&s->gb, -bits);
1529 steps = quant_steps[qindex];
1530 mant = v % steps;
1531 v = v / steps;
115329f1 1532 s->sb_samples[0][k * 12 + l + 0][i] =
239c2f4c 1533 l2_unscale_group(steps, mant, scale0);
115329f1 1534 s->sb_samples[1][k * 12 + l + 0][i] =
239c2f4c
FB
1535 l2_unscale_group(steps, mant, scale1);
1536 mant = v % steps;
1537 v = v / steps;
115329f1 1538 s->sb_samples[0][k * 12 + l + 1][i] =
239c2f4c 1539 l2_unscale_group(steps, mant, scale0);
115329f1 1540 s->sb_samples[1][k * 12 + l + 1][i] =
239c2f4c 1541 l2_unscale_group(steps, mant, scale1);
115329f1 1542 s->sb_samples[0][k * 12 + l + 2][i] =
239c2f4c 1543 l2_unscale_group(steps, v, scale0);
115329f1 1544 s->sb_samples[1][k * 12 + l + 2][i] =
239c2f4c
FB
1545 l2_unscale_group(steps, v, scale1);
1546 } else {
1547 for(m=0;m<3;m++) {
1548 mant = get_bits(&s->gb, bits);
115329f1 1549 s->sb_samples[0][k * 12 + l + m][i] =
239c2f4c 1550 l1_unscale(bits - 1, mant, scale0);
115329f1 1551 s->sb_samples[1][k * 12 + l + m][i] =
239c2f4c
FB
1552 l1_unscale(bits - 1, mant, scale1);
1553 }
1554 }
1555 } else {
1556 s->sb_samples[0][k * 12 + l + 0][i] = 0;
1557 s->sb_samples[0][k * 12 + l + 1][i] = 0;
1558 s->sb_samples[0][k * 12 + l + 2][i] = 0;
1559 s->sb_samples[1][k * 12 + l + 0][i] = 0;
1560 s->sb_samples[1][k * 12 + l + 1][i] = 0;
1561 s->sb_samples[1][k * 12 + l + 2][i] = 0;
1562 }
1563 /* next subband in alloc table */
115329f1 1564 j += 1 << bit_alloc_bits;
239c2f4c
FB
1565 }
1566 /* fill remaining samples to zero */
1567 for(i=sblimit;i<SBLIMIT;i++) {
1568 for(ch=0;ch<s->nb_channels;ch++) {
1569 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
1570 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
1571 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
1572 }
1573 }
1574 }
1575 }
1576 return 3 * 12;
de6d9b64
FB
1577}
1578
1579/*
239c2f4c 1580 * Seek back in the stream for backstep bytes (at most 511 bytes)
de6d9b64 1581 */
5c91a675 1582static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)
de6d9b64 1583{
0c1a9eda 1584 uint8_t *ptr;
de6d9b64
FB
1585
1586 /* compute current position in stream */
228ef9dd 1587 ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));
8db1a1dd 1588
de6d9b64
FB
1589 /* copy old data before current one */
1590 ptr -= backstep;
115329f1 1591 memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] +
239c2f4c 1592 BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);
de6d9b64 1593 /* init get bits again */
68f593b4 1594 init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);
de6d9b64 1595
239c2f4c
FB
1596 /* prepare next buffer */
1597 s->inbuf_index ^= 1;
1598 s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];
1599 s->old_frame_size = s->frame_size;
1600}
1601
1602static inline void lsf_sf_expand(int *slen,
1603 int sf, int n1, int n2, int n3)
1604{
1605 if (n3) {
1606 slen[3] = sf % n3;
1607 sf /= n3;
1608 } else {
1609 slen[3] = 0;
1610 }
1611 if (n2) {
1612 slen[2] = sf % n2;
1613 sf /= n2;
1614 } else {
1615 slen[2] = 0;
1616 }
1617 slen[1] = sf % n1;
1618 sf /= n1;
1619 slen[0] = sf;
1620}
1621
115329f1 1622static void exponents_from_scale_factors(MPADecodeContext *s,
239c2f4c 1623 GranuleDef *g,
0c1a9eda 1624 int16_t *exponents)
239c2f4c 1625{
0c1a9eda 1626 const uint8_t *bstab, *pretab;
239c2f4c 1627 int len, i, j, k, l, v0, shift, gain, gains[3];
0c1a9eda 1628 int16_t *exp_ptr;
239c2f4c
FB
1629
1630 exp_ptr = exponents;
1631 gain = g->global_gain - 210;
1632 shift = g->scalefac_scale + 1;
1633
1634 bstab = band_size_long[s->sample_rate_index];
1635 pretab = mpa_pretab[g->preflag];
1636 for(i=0;i<g->long_end;i++) {
1637 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);
1638 len = bstab[i];
1639 for(j=len;j>0;j--)
1640 *exp_ptr++ = v0;
1641 }
1642
1643 if (g->short_start < 13) {
1644 bstab = band_size_short[s->sample_rate_index];
1645 gains[0] = gain - (g->subblock_gain[0] << 3);
1646 gains[1] = gain - (g->subblock_gain[1] << 3);
1647 gains[2] = gain - (g->subblock_gain[2] << 3);
1648 k = g->long_end;
1649 for(i=g->short_start;i<13;i++) {
1650 len = bstab[i];
1651 for(l=0;l<3;l++) {
1652 v0 = gains[l] - (g->scale_factors[k++] << shift);
1653 for(j=len;j>0;j--)
1654 *exp_ptr++ = v0;
1655 }
1656 }
1657 }
1658}
1659
1660/* handle n = 0 too */
1661static inline int get_bitsz(GetBitContext *s, int n)
1662{
1663 if (n == 0)
1664 return 0;
1665 else
1666 return get_bits(s, n);
1667}
1668
1669static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
0c1a9eda 1670 int16_t *exponents, int end_pos)
239c2f4c
FB
1671{
1672 int s_index;
1673 int linbits, code, x, y, l, v, i, j, k, pos;
8db1a1dd 1674 GetBitContext last_gb;
239c2f4c 1675 VLC *vlc;
239c2f4c
FB
1676
1677 /* low frequencies (called big values) */
1678 s_index = 0;
1679 for(i=0;i<3;i++) {
1680 j = g->region_size[i];
1681 if (j == 0)
1682 continue;
1683 /* select vlc table */
1684 k = g->table_select[i];
1685 l = mpa_huff_data[k][0];
1686 linbits = mpa_huff_data[k][1];
1687 vlc = &huff_vlc[l];
239c2f4c 1688
daf4cd9a
MN
1689 if(!l){
1690 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*j);
1691 s_index += 2*j;
1692 continue;
1693 }
1694
239c2f4c
FB
1695 /* read huffcode and compute each couple */
1696 for(;j>0;j--) {
1697 if (get_bits_count(&s->gb) >= end_pos)
1698 break;
daf4cd9a 1699 y = get_vlc2(&s->gb, vlc->table, 8, 3);
3328ec63
MN
1700
1701 if(!y){
1702 g->sb_hybrid[s_index ] =
1703 g->sb_hybrid[s_index+1] = 0;
1704 s_index += 2;
1705 continue;
1706 }
1707
daf4cd9a
MN
1708 x = y >> 4;
1709 y = y & 0x0f;
1710
115329f1 1711 dprintf("region=%d n=%d x=%d y=%d exp=%d\n",
239c2f4c
FB
1712 i, g->region_size[i] - j, x, y, exponents[s_index]);
1713 if (x) {
1714 if (x == 15)
1715 x += get_bitsz(&s->gb, linbits);
1716 v = l3_unscale(x, exponents[s_index]);
1717 if (get_bits1(&s->gb))
1718 v = -v;
1719 } else {
1720 v = 0;
1721 }
1722 g->sb_hybrid[s_index++] = v;
1723 if (y) {
1724 if (y == 15)
1725 y += get_bitsz(&s->gb, linbits);
1726 v = l3_unscale(y, exponents[s_index]);
1727 if (get_bits1(&s->gb))
1728 v = -v;
1729 } else {
1730 v = 0;
1731 }
1732 g->sb_hybrid[s_index++] = v;
1733 }
1734 }
115329f1 1735
239c2f4c
FB
1736 /* high frequencies */
1737 vlc = &huff_quad_vlc[g->count1table_select];
8db1a1dd 1738 last_gb.buffer = NULL;
239c2f4c
FB
1739 while (s_index <= 572) {
1740 pos = get_bits_count(&s->gb);
1741 if (pos >= end_pos) {
8db1a1dd 1742 if (pos > end_pos && last_gb.buffer != NULL) {
239c2f4c
FB
1743 /* some encoders generate an incorrect size for this
1744 part. We must go back into the data */
1745 s_index -= 4;
8db1a1dd 1746 s->gb = last_gb;
239c2f4c
FB
1747 }
1748 break;
1749 }
8db1a1dd
MN
1750 last_gb= s->gb;
1751
3c693e77 1752 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
239c2f4c 1753 dprintf("t=%d code=%d\n", g->count1table_select, code);
3c693e77
MN
1754 g->sb_hybrid[s_index+0]=
1755 g->sb_hybrid[s_index+1]=
1756 g->sb_hybrid[s_index+2]=
1757 g->sb_hybrid[s_index+3]= 0;
1758 while(code){
1759 const static int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
1760 int pos= s_index+idxtab[code];
1761 code ^= 8>>idxtab[code];
1762 v = l3_unscale(1, exponents[pos]);
1763 if(get_bits1(&s->gb))
1764 v = -v;
1765 g->sb_hybrid[pos] = v;
239c2f4c 1766 }
3c693e77 1767 s_index+=4;
239c2f4c 1768 }
daf4cd9a 1769 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
de6d9b64
FB
1770 return 0;
1771}
1772
239c2f4c
FB
1773/* Reorder short blocks from bitstream order to interleaved order. It
1774 would be faster to do it in parsing, but the code would be far more
1775 complicated */
1776static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1777{
1778 int i, j, k, len;
0c1a9eda
ZK
1779 int32_t *ptr, *dst, *ptr1;
1780 int32_t tmp[576];
239c2f4c
FB
1781
1782 if (g->block_type != 2)
1783 return;
1784
1785 if (g->switch_point) {
1786 if (s->sample_rate_index != 8) {
1787 ptr = g->sb_hybrid + 36;
1788 } else {
1789 ptr = g->sb_hybrid + 48;
1790 }
1791 } else {
1792 ptr = g->sb_hybrid;
1793 }
115329f1 1794
239c2f4c
FB
1795 for(i=g->short_start;i<13;i++) {
1796 len = band_size_short[s->sample_rate_index][i];
1797 ptr1 = ptr;
1798 for(k=0;k<3;k++) {
1799 dst = tmp + k;
1800 for(j=len;j>0;j--) {
1801 *dst = *ptr++;
1802 dst += 3;
1803 }
1804 }
0c1a9eda 1805 memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));
239c2f4c
FB
1806 }
1807}
1808
1809#define ISQRT2 FIXR(0.70710678118654752440)
1810
1811static void compute_stereo(MPADecodeContext *s,
1812 GranuleDef *g0, GranuleDef *g1)
1813{
1814 int i, j, k, l;
0c1a9eda 1815 int32_t v1, v2;
239c2f4c 1816 int sf_max, tmp0, tmp1, sf, len, non_zero_found;
0c1a9eda
ZK
1817 int32_t (*is_tab)[16];
1818 int32_t *tab0, *tab1;
239c2f4c
FB
1819 int non_zero_found_short[3];
1820
1821 /* intensity stereo */
1822 if (s->mode_ext & MODE_EXT_I_STEREO) {
1823 if (!s->lsf) {
1824 is_tab = is_table;
1825 sf_max = 7;
1826 } else {
1827 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1828 sf_max = 16;
1829 }
115329f1 1830
239c2f4c
FB
1831 tab0 = g0->sb_hybrid + 576;
1832 tab1 = g1->sb_hybrid + 576;
1833
1834 non_zero_found_short[0] = 0;
1835 non_zero_found_short[1] = 0;
1836 non_zero_found_short[2] = 0;
1837 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1838 for(i = 12;i >= g1->short_start;i--) {
1839 /* for last band, use previous scale factor */
1840 if (i != 11)
1841 k -= 3;
1842 len = band_size_short[s->sample_rate_index][i];
1843 for(l=2;l>=0;l--) {
1844 tab0 -= len;
1845 tab1 -= len;
1846 if (!non_zero_found_short[l]) {
1847 /* test if non zero band. if so, stop doing i-stereo */
1848 for(j=0;j<len;j++) {
1849 if (tab1[j] != 0) {
1850 non_zero_found_short[l] = 1;
1851 goto found1;
1852 }
1853 }
1854 sf = g1->scale_factors[k + l];
1855 if (sf >= sf_max)
1856 goto found1;
1857
1858 v1 = is_tab[0][sf];
1859 v2 = is_tab[1][sf];
1860 for(j=0;j<len;j++) {
1861 tmp0 = tab0[j];
1862 tab0[j] = MULL(tmp0, v1);
1863 tab1[j] = MULL(tmp0, v2);
1864 }
1865 } else {
1866 found1:
1867 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1868 /* lower part of the spectrum : do ms stereo
1869 if enabled */
1870 for(j=0;j<len;j++) {
1871 tmp0 = tab0[j];
1872 tmp1 = tab1[j];
1873 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1874 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1875 }
1876 }
1877 }
1878 }
1879 }
1880
115329f1
DB
1881 non_zero_found = non_zero_found_short[0] |
1882 non_zero_found_short[1] |
239c2f4c
FB
1883 non_zero_found_short[2];
1884
1885 for(i = g1->long_end - 1;i >= 0;i--) {
1886 len = band_size_long[s->sample_rate_index][i];
1887 tab0 -= len;
1888 tab1 -= len;
1889 /* test if non zero band. if so, stop doing i-stereo */
1890 if (!non_zero_found) {
1891 for(j=0;j<len;j++) {
1892 if (tab1[j] != 0) {
1893 non_zero_found = 1;
1894 goto found2;
1895 }
1896 }
1897 /* for last band, use previous scale factor */
1898 k = (i == 21) ? 20 : i;
1899 sf = g1->scale_factors[k];
1900 if (sf >= sf_max)
1901 goto found2;
1902 v1 = is_tab[0][sf];
1903 v2 = is_tab[1][sf];
1904 for(j=0;j<len;j++) {
1905 tmp0 = tab0[j];
1906 tab0[j] = MULL(tmp0, v1);
1907 tab1[j] = MULL(tmp0, v2);
1908 }
1909 } else {
1910 found2:
1911 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1912 /* lower part of the spectrum : do ms stereo
1913 if enabled */
1914 for(j=0;j<len;j++) {
1915 tmp0 = tab0[j];
1916 tmp1 = tab1[j];
1917 tab0[j] = MULL(tmp0 + tmp1, ISQRT2);
1918 tab1[j] = MULL(tmp0 - tmp1, ISQRT2);
1919 }
1920 }
1921 }
1922 }
1923 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1924 /* ms stereo ONLY */
1925 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1926 global gain */
1927 tab0 = g0->sb_hybrid;
1928 tab1 = g1->sb_hybrid;
1929 for(i=0;i<576;i++) {
1930 tmp0 = tab0[i];
1931 tmp1 = tab1[i];
1932 tab0[i] = tmp0 + tmp1;
1933 tab1[i] = tmp0 - tmp1;
1934 }
1935 }
1936}
1937
a1e257b2 1938static void compute_antialias_integer(MPADecodeContext *s,
239c2f4c
FB
1939 GranuleDef *g)
1940{
ce4a29c0
MN
1941 int32_t *ptr, *csa;
1942 int n, i;
239c2f4c
FB
1943
1944 /* we antialias only "long" bands */
1945 if (g->block_type == 2) {
1946 if (!g->switch_point)
1947 return;
1948 /* XXX: check this for 8000Hz case */
1949 n = 1;
1950 } else {
1951 n = SBLIMIT - 1;
1952 }
115329f1 1953
239c2f4c
FB
1954 ptr = g->sb_hybrid + 18;
1955 for(i = n;i > 0;i--) {
ce4a29c0
MN
1956 int tmp0, tmp1, tmp2;
1957 csa = &csa_table[0][0];
1958#define INT_AA(j) \
44f1698a
MN
1959 tmp0 = ptr[-1-j];\
1960 tmp1 = ptr[ j];\
ce4a29c0 1961 tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\
44f1698a
MN
1962 ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\
1963 ptr[ j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));
ce4a29c0
MN
1964
1965 INT_AA(0)
1966 INT_AA(1)
1967 INT_AA(2)
1968 INT_AA(3)
1969 INT_AA(4)
1970 INT_AA(5)
1971 INT_AA(6)
1972 INT_AA(7)
115329f1
DB
1973
1974 ptr += 18;
a1e257b2
MN
1975 }
1976}
1977
1978static void compute_antialias_float(MPADecodeContext *s,
1979 GranuleDef *g)
1980{
ce4a29c0
MN
1981 int32_t *ptr;
1982 int n, i;
a1e257b2
MN
1983
1984 /* we antialias only "long" bands */
1985 if (g->block_type == 2) {
1986 if (!g->switch_point)
1987 return;
1988 /* XXX: check this for 8000Hz case */
1989 n = 1;
1990 } else {
1991 n = SBLIMIT - 1;
1992 }
115329f1 1993
a1e257b2
MN
1994 ptr = g->sb_hybrid + 18;
1995 for(i = n;i > 0;i--) {
ce4a29c0 1996 float tmp0, tmp1;
115329f1 1997 float *csa = &csa_table_float[0][0];
ce4a29c0
MN
1998#define FLOAT_AA(j)\
1999 tmp0= ptr[-1-j];\
2000 tmp1= ptr[ j];\
2001 ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\
2002 ptr[ j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);
115329f1 2003
ce4a29c0
MN
2004 FLOAT_AA(0)
2005 FLOAT_AA(1)
2006 FLOAT_AA(2)
2007 FLOAT_AA(3)
2008 FLOAT_AA(4)
2009 FLOAT_AA(5)
2010 FLOAT_AA(6)
2011 FLOAT_AA(7)
2012
115329f1 2013 ptr += 18;
239c2f4c
FB
2014 }
2015}
2016
2017static void compute_imdct(MPADecodeContext *s,
115329f1 2018 GranuleDef *g,
0c1a9eda
ZK
2019 int32_t *sb_samples,
2020 int32_t *mdct_buf)
239c2f4c 2021{
125d6246 2022 int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;
0c1a9eda 2023 int32_t out2[12];
125d6246 2024 int i, j, mdct_long_end, v, sblimit;
239c2f4c
FB
2025
2026 /* find last non zero block */
2027 ptr = g->sb_hybrid + 576;
2028 ptr1 = g->sb_hybrid + 2 * 18;
2029 while (ptr >= ptr1) {
2030 ptr -= 6;
2031 v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];
2032 if (v != 0)
2033 break;
2034 }
2035 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
2036
2037 if (g->block_type == 2) {
2038 /* XXX: check for 8000 Hz */
2039 if (g->switch_point)
2040 mdct_long_end = 2;
2041 else
2042 mdct_long_end = 0;
2043 } else {
2044 mdct_long_end = sblimit;
2045 }
2046
2047 buf = mdct_buf;
2048 ptr = g->sb_hybrid;
2049 for(j=0;j<mdct_long_end;j++) {
239c2f4c
FB
2050 /* apply window & overlap with previous buffer */
2051 out_ptr = sb_samples + j;
2052 /* select window */
2053 if (g->switch_point && j < 2)
2054 win1 = mdct_win[0];
2055 else
2056 win1 = mdct_win[g->block_type];
2057 /* select frequency inversion */
2058 win = win1 + ((4 * 36) & -(j & 1));
711ae726
MN
2059 imdct36(out_ptr, buf, ptr, win);
2060 out_ptr += 18*SBLIMIT;
239c2f4c
FB
2061 ptr += 18;
2062 buf += 18;
2063 }
2064 for(j=mdct_long_end;j<sblimit;j++) {
239c2f4c
FB
2065 /* select frequency inversion */
2066 win = mdct_win[2] + ((4 * 36) & -(j & 1));
239c2f4c 2067 out_ptr = sb_samples + j;
115329f1 2068
125d6246
MN
2069 for(i=0; i<6; i++){
2070 *out_ptr = buf[i];
2071 out_ptr += SBLIMIT;
2072 }
2073 imdct12(out2, ptr + 0);
2074 for(i=0;i<6;i++) {
2075 *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];
2076 buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);
239c2f4c
FB
2077 out_ptr += SBLIMIT;
2078 }
125d6246
MN
2079 imdct12(out2, ptr + 1);
2080 for(i=0;i<6;i++) {
2081 *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];
2082 buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);
2083 out_ptr += SBLIMIT;
2084 }
2085 imdct12(out2, ptr + 2);
2086 for(i=0;i<6;i++) {
2087 buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];
2088 buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);
2089 buf[i + 6*2] = 0;
2090 }
239c2f4c
FB
2091 ptr += 18;
2092 buf += 18;
2093 }
2094 /* zero bands */
2095 for(j=sblimit;j<SBLIMIT;j++) {
2096 /* overlap */
2097 out_ptr = sb_samples + j;
2098 for(i=0;i<18;i++) {
2099 *out_ptr = buf[i];
2100 buf[i] = 0;
2101 out_ptr += SBLIMIT;
2102 }
2103 buf += 18;
2104 }
2105}
2106
747a67fb 2107#if defined(DEBUG)
0c1a9eda 2108void sample_dump(int fnum, int32_t *tab, int n)
239c2f4c
FB
2109{
2110 static FILE *files[16], *f;
2111 char buf[512];
81552334 2112 int i;
0c1a9eda 2113 int32_t v;
115329f1 2114
239c2f4c
FB
2115 f = files[fnum];
2116 if (!f) {
115329f1
DB
2117 snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",
2118 fnum,
81552334
FB
2119#ifdef USE_HIGHPRECISION
2120 "hp"
2121#else
2122 "lp"
2123#endif
2124 );
239c2f4c
FB
2125 f = fopen(buf, "w");
2126 if (!f)
2127 return;
2128 files[fnum] = f;
2129 }
115329f1 2130
239c2f4c 2131 if (fnum == 0) {
239c2f4c 2132 static int pos = 0;
84af4a7e 2133 av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);
239c2f4c 2134 for(i=0;i<n;i++) {
84af4a7e 2135 av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);
239c2f4c 2136 if ((i % 18) == 17)
84af4a7e 2137 av_log(NULL, AV_LOG_DEBUG, "\n");
239c2f4c
FB
2138 }
2139 pos += n;
2140 }
81552334
FB
2141 for(i=0;i<n;i++) {
2142 /* normalize to 23 frac bits */
2143 v = tab[i] << (23 - FRAC_BITS);
0c1a9eda 2144 fwrite(&v, 1, sizeof(int32_t), f);
81552334 2145 }
239c2f4c
FB
2146}
2147#endif
2148
2149
2150/* main layer3 decoding function */
2151static int mp_decode_layer3(MPADecodeContext *s)
2152{
2153 int nb_granules, main_data_begin, private_bits;
2154 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;
2155 GranuleDef granules[2][2], *g;
0c1a9eda 2156 int16_t exponents[576];
239c2f4c
FB
2157
2158 /* read side info */
2159 if (s->lsf) {
2160 main_data_begin = get_bits(&s->gb, 8);
2161 if (s->nb_channels == 2)
2162 private_bits = get_bits(&s->gb, 2);
2163 else
2164 private_bits = get_bits(&s->gb, 1);
2165 nb_granules = 1;
2166 } else {
2167 main_data_begin = get_bits(&s->gb, 9);
2168 if (s->nb_channels == 2)
2169 private_bits = get_bits(&s->gb, 3);
2170 else
2171 private_bits = get_bits(&s->gb, 5);
2172 nb_granules = 2;
2173 for(ch=0;ch<s->nb_channels;ch++) {
2174 granules[ch][0].scfsi = 0; /* all scale factors are transmitted */
2175 granules[ch][1].scfsi = get_bits(&s->gb, 4);
2176 }
2177 }
115329f1 2178
239c2f4c
FB
2179 for(gr=0;gr<nb_granules;gr++) {
2180 for(ch=0;ch<s->nb_channels;ch++) {
2181 dprintf("gr=%d ch=%d: side_info\n", gr, ch);
2182 g = &granules[ch][gr];
2183 g->part2_3_length = get_bits(&s->gb, 12);
2184 g->big_values = get_bits(&s->gb, 9);
2185 g->global_gain = get_bits(&s->gb, 8);
2186 /* if MS stereo only is selected, we precompute the
2187 1/sqrt(2) renormalization factor */
115329f1 2188 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
239c2f4c
FB
2189 MODE_EXT_MS_STEREO)
2190 g->global_gain -= 2;
2191 if (s->lsf)
2192 g->scalefac_compress = get_bits(&s->gb, 9);
2193 else
2194 g->scalefac_compress = get_bits(&s->gb, 4);
2195 blocksplit_flag = get_bits(&s->gb, 1);
2196 if (blocksplit_flag) {
2197 g->block_type = get_bits(&s->gb, 2);
2198 if (g->block_type == 0)
2199 return -1;
2200 g->switch_point = get_bits(&s->gb, 1);
2201 for(i=0;i<2;i++)
2202 g->table_select[i] = get_bits(&s->gb, 5);
115329f1 2203 for(i=0;i<3;i++)
239c2f4c
FB
2204 g->subblock_gain[i] = get_bits(&s->gb, 3);
2205 /* compute huffman coded region sizes */
2206 if (g->block_type == 2)
2207 g->region_size[0] = (36 / 2);
2208 else {
115329f1 2209 if (s->sample_rate_index <= 2)
239c2f4c 2210 g->region_size[0] = (36 / 2);
115329f1 2211 else if (s->sample_rate_index != 8)
239c2f4c
FB
2212 g->region_size[0] = (54 / 2);
2213 else
2214 g->region_size[0] = (108 / 2);
2215 }
2216 g->region_size[1] = (576 / 2);
2217 } else {
2218 int region_address1, region_address2, l;
2219 g->block_type = 0;
2220 g->switch_point = 0;
2221 for(i=0;i<3;i++)
2222 g->table_select[i] = get_bits(&s->gb, 5);
2223 /* compute huffman coded region sizes */
2224 region_address1 = get_bits(&s->gb, 4);
2225 region_address2 = get_bits(&s->gb, 3);
115329f1 2226 dprintf("region1=%d region2=%d\n",
239c2f4c 2227 region_address1, region_address2);
115329f1 2228 g->region_size[0] =
239c2f4c
FB
2229 band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;
2230 l = region_address1 + region_address2 + 2;
2231 /* should not overflow */
2232 if (l > 22)
2233 l = 22;
115329f1 2234 g->region_size[1] =
239c2f4c
FB
2235 band_index_long[s->sample_rate_index][l] >> 1;
2236 }
2237 /* convert region offsets to region sizes and truncate
2238 size to big_values */
2239 g->region_size[2] = (576 / 2);
2240 j = 0;
2241 for(i=0;i<3;i++) {
c7aa3696 2242 k = FFMIN(g->region_size[i], g->big_values);
239c2f4c
FB
2243 g->region_size[i] = k - j;
2244 j = k;
2245 }
2246
2247 /* compute band indexes */
2248 if (g->block_type == 2) {
2249 if (g->switch_point) {
2250 /* if switched mode, we handle the 36 first samples as
2251 long blocks. For 8000Hz, we handle the 48 first
2252 exponents as long blocks (XXX: check this!) */
2253 if (s->sample_rate_index <= 2)
2254 g->long_end = 8;
2255 else if (s->sample_rate_index != 8)
2256 g->long_end = 6;
2257 else
2258 g->long_end = 4; /* 8000 Hz */
115329f1 2259
c7aa3696 2260 g->short_start = 2 + (s->sample_rate_index != 8);
239c2f4c
FB
2261 } else {
2262 g->long_end = 0;
2263 g->short_start = 0;
2264 }
2265 } else {
2266 g->short_start = 13;
2267 g->long_end = 22;
2268 }
115329f1 2269
239c2f4c
FB
2270 g->preflag = 0;
2271 if (!s->lsf)
2272 g->preflag = get_bits(&s->gb, 1);
2273 g->scalefac_scale = get_bits(&s->gb, 1);
2274 g->count1table_select = get_bits(&s->gb, 1);
2275 dprintf("block_type=%d switch_point=%d\n",
2276 g->block_type, g->switch_point);
2277 }
2278 }
2279
1ede228a 2280 if (!s->adu_mode) {
239c2f4c
FB
2281 /* now we get bits from the main_data_begin offset */
2282 dprintf("seekback: %d\n", main_data_begin);
2283 seek_to_maindata(s, main_data_begin);
1ede228a 2284 }
239c2f4c
FB
2285
2286 for(gr=0;gr<nb_granules;gr++) {
2287 for(ch=0;ch<s->nb_channels;ch++) {
2288 g = &granules[ch][gr];
115329f1 2289
239c2f4c 2290 bits_pos = get_bits_count(&s->gb);
115329f1 2291
239c2f4c 2292 if (!s->lsf) {
0c1a9eda 2293 uint8_t *sc;
239c2f4c
FB
2294 int slen, slen1, slen2;
2295
2296 /* MPEG1 scale factors */
2297 slen1 = slen_table[0][g->scalefac_compress];
2298 slen2 = slen_table[1][g->scalefac_compress];
2299 dprintf("slen1=%d slen2=%d\n", slen1, slen2);
2300 if (g->block_type == 2) {
2301 n = g->switch_point ? 17 : 18;
2302 j = 0;
2303 for(i=0;i<n;i++)
2304 g->scale_factors[j++] = get_bitsz(&s->gb, slen1);
2305 for(i=0;i<18;i++)
2306 g->scale_factors[j++] = get_bitsz(&s->gb, slen2);
2307 for(i=0;i<3;i++)
2308 g->scale_factors[j++] = 0;
2309 } else {
2310 sc = granules[ch][0].scale_factors;
2311 j = 0;
2312 for(k=0;k<4;k++) {
2313 n = (k == 0 ? 6 : 5);
2314 if ((g->scfsi & (0x8 >> k)) == 0) {
2315 slen = (k < 2) ? slen1 : slen2;
2316 for(i=0;i<n;i++)
2317 g->scale_factors[j++] = get_bitsz(&s->gb, slen);
2318 } else {
2319 /* simply copy from last granule */
2320 for(i=0;i<n;i++) {
2321 g->scale_factors[j] = sc[j];
2322 j++;
2323 }
2324 }
2325 }
2326 g->scale_factors[j++] = 0;
2327 }
747a67fb 2328#if defined(DEBUG)
239c2f4c 2329 {
267f7edc 2330 dprintf("scfsi=%x gr=%d ch=%d scale_factors:\n",
239c2f4c
FB
2331 g->scfsi, gr, ch);
2332 for(i=0;i<j;i++)
267f7edc
SH
2333 dprintf(" %d", g->scale_factors[i]);
2334 dprintf("\n");
239c2f4c
FB
2335 }
2336#endif
2337 } else {
2338 int tindex, tindex2, slen[4], sl, sf;
2339
2340 /* LSF scale factors */
2341 if (g->block_type == 2) {
2342 tindex = g->switch_point ? 2 : 1;
2343 } else {
2344 tindex = 0;
2345 }
2346 sf = g->scalefac_compress;
2347 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
2348 /* intensity stereo case */
2349 sf >>= 1;
2350 if (sf < 180) {
2351 lsf_sf_expand(slen, sf, 6, 6, 0);
2352 tindex2 = 3;
2353 } else if (sf < 244) {
2354 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
2355 tindex2 = 4;
2356 } else {
2357 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
2358 tindex2 = 5;
2359 }
2360 } else {
2361 /* normal case */
2362 if (sf < 400) {
2363 lsf_sf_expand(slen, sf, 5, 4, 4);
2364 tindex2 = 0;
2365 } else if (sf < 500) {
2366 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
2367 tindex2 = 1;
2368 } else {
2369 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
2370 tindex2 = 2;
2371 g->preflag = 1;
2372 }
2373 }
2374
2375 j = 0;
2376 for(k=0;k<4;k++) {
2377 n = lsf_nsf_table[tindex2][tindex][k];
2378 sl = slen[k];
2379 for(i=0;i<n;i++)
2380 g->scale_factors[j++] = get_bitsz(&s->gb, sl);
2381 }
2382 /* XXX: should compute exact size */
2383 for(;j<40;j++)
2384 g->scale_factors[j] = 0;
747a67fb 2385#if defined(DEBUG)
239c2f4c 2386 {
267f7edc 2387 dprintf("gr=%d ch=%d scale_factors:\n",
239c2f4c
FB
2388 gr, ch);
2389 for(i=0;i<40;i++)
267f7edc
SH
2390 dprintf(" %d", g->scale_factors[i]);
2391 dprintf("\n");
239c2f4c
FB
2392 }
2393#endif
2394 }
2395
2396 exponents_from_scale_factors(s, g, exponents);
2397
2398 /* read Huffman coded residue */
2399 if (huffman_decode(s, g, exponents,
2400 bits_pos + g->part2_3_length) < 0)
2401 return -1;
747a67fb
FB
2402#if defined(DEBUG)
2403 sample_dump(0, g->sb_hybrid, 576);
239c2f4c
FB
2404#endif
2405
2406 /* skip extension bits */
2407 bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);
2408 if (bits_left < 0) {
2409 dprintf("bits_left=%d\n", bits_left);
2410 return -1;
2411 }
2412 while (bits_left >= 16) {
2413 skip_bits(&s->gb, 16);
2414 bits_left -= 16;
2415 }
2416 if (bits_left > 0)
2417 skip_bits(&s->gb, bits_left);
2418 } /* ch */
2419
2420 if (s->nb_channels == 2)
2421 compute_stereo(s, &granules[0][gr], &granules[1][gr]);
2422
2423 for(ch=0;ch<s->nb_channels;ch++) {
2424 g = &granules[ch][gr];
2425
2426 reorder_block(s, g);
747a67fb 2427#if defined(DEBUG)
239c2f4c
FB
2428 sample_dump(0, g->sb_hybrid, 576);
2429#endif
a1e257b2 2430 s->compute_antialias(s, g);
81552334 2431#if defined(DEBUG)
239c2f4c
FB
2432 sample_dump(1, g->sb_hybrid, 576);
2433#endif
115329f1 2434 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
81552334 2435#if defined(DEBUG)
239c2f4c
FB
2436 sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);
2437#endif
2438 }
2439 } /* gr */
2440 return nb_granules * 18;
2441}
2442
115329f1 2443static int mp_decode_frame(MPADecodeContext *s,
a3a5f4d6 2444 OUT_INT *samples)
239c2f4c
FB
2445{
2446 int i, nb_frames, ch;
a3a5f4d6 2447 OUT_INT *samples_ptr;
239c2f4c 2448
115329f1 2449 init_get_bits(&s->gb, s->inbuf + HEADER_SIZE,
68f593b4 2450 (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);
115329f1 2451
239c2f4c
FB
2452 /* skip error protection field */
2453 if (s->error_protection)
2454 get_bits(&s->gb, 16);
2455
2456 dprintf("frame %d:\n", s->frame_count);
2457 switch(s->layer) {
2458 case 1:
2459 nb_frames = mp_decode_layer1(s);
2460 break;
2461 case 2:
2462 nb_frames = mp_decode_layer2(s);
2463 break;
2464 case 3:
2465 default:
2466 nb_frames = mp_decode_layer3(s);
2467 break;
2468 }
2469#if defined(DEBUG)
2470 for(i=0;i<nb_frames;i++) {
2471 for(ch=0;ch<s->nb_channels;ch++) {
2472 int j;
267f7edc 2473 dprintf("%d-%d:", i, ch);
239c2f4c 2474 for(j=0;j<SBLIMIT;j++)
267f7edc
SH
2475 dprintf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);
2476 dprintf("\n");
239c2f4c
FB
2477 }
2478 }
2479#endif
2480 /* apply the synthesis filter */
2481 for(ch=0;ch<s->nb_channels;ch++) {
2482 samples_ptr = samples + ch;
2483 for(i=0;i<nb_frames;i++) {
bf1f4da0 2484 ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),
bb270c08
DB
2485 window, &s->dither_state,
2486 samples_ptr, s->nb_channels,
239c2f4c
FB
2487 s->sb_samples[ch][i]);
2488 samples_ptr += 32 * s->nb_channels;
2489 }
2490 }
2491#ifdef DEBUG
115329f1 2492 s->frame_count++;
239c2f4c 2493#endif
a3a5f4d6 2494 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
239c2f4c
FB
2495}
2496
de6d9b64 2497static int decode_frame(AVCodecContext * avctx,
bb270c08
DB
2498 void *data, int *data_size,
2499 uint8_t * buf, int buf_size)
de6d9b64
FB
2500{
2501 MPADecodeContext *s = avctx->priv_data;
0c1a9eda
ZK
2502 uint32_t header;
2503 uint8_t *buf_ptr;
de6d9b64 2504 int len, out_size;
a3a5f4d6 2505 OUT_INT *out_samples = data;
de6d9b64 2506
de6d9b64
FB
2507 buf_ptr = buf;
2508 while (buf_size > 0) {
bb270c08
DB
2509 len = s->inbuf_ptr - s->inbuf;
2510 if (s->frame_size == 0) {
239c2f4c
FB
2511 /* special case for next header for first frame in free
2512 format case (XXX: find a simpler method) */
2513 if (s->free_format_next_header != 0) {
2514 s->inbuf[0] = s->free_format_next_header >> 24;
2515 s->inbuf[1] = s->free_format_next_header >> 16;
2516 s->inbuf[2] = s->free_format_next_header >> 8;
2517 s->inbuf[3] = s->free_format_next_header;
2518 s->inbuf_ptr = s->inbuf + 4;
2519 s->free_format_next_header = 0;
2520 goto got_header;
2521 }
bb270c08 2522 /* no header seen : find one. We need at least HEADER_SIZE
239c2f4c 2523 bytes to parse it */
bb270c08
DB
2524 len = HEADER_SIZE - len;
2525 if (len > buf_size)
2526 len = buf_size;
2527 if (len > 0) {
2528 memcpy(s->inbuf_ptr, buf_ptr, len);
2529 buf_ptr += len;
2530 buf_size -= len;
2531 s->inbuf_ptr += len;
2532 }
2533 if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {
239c2f4c 2534 got_header:
bb270c08
DB
2535 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2536 (s->inbuf[2] << 8) | s->inbuf[3];
92d24f49 2537
bb270c08
DB
2538 if (ff_mpa_check_header(header) < 0) {
2539 /* no sync found : move by one byte (inefficient, but simple!) */
2540 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2541 s->inbuf_ptr--;
239c2f4c
FB
2542 dprintf("skip %x\n", header);
2543 /* reset free format frame size to give a chance
2544 to get a new bitrate */
2545 s->free_format_frame_size = 0;
bb270c08
DB
2546 } else {
2547 if (decode_header(s, header) == 1) {
81552334 2548 /* free format: prepare to compute frame size */
bb270c08 2549 s->frame_size = -1;
239c2f4c 2550 }
81552334
FB
2551 /* update codec info */
2552 avctx->sample_rate = s->sample_rate;
2553 avctx->channels = s->nb_channels;
2554 avctx->bit_rate = s->bit_rate;
98ce5991 2555 avctx->sub_id = s->layer;
8c5b5683
FB
2556 switch(s->layer) {
2557 case 1:
2558 avctx->frame_size = 384;
2559 break;
2560 case 2:
2561 avctx->frame_size = 1152;
2562 break;
2563 case 3:
2564 if (s->lsf)
2565 avctx->frame_size = 576;
2566 else
2567 avctx->frame_size = 1152;
2568 break;
2569 }
bb270c08
DB
2570 }
2571 }
239c2f4c
FB
2572 } else if (s->frame_size == -1) {
2573 /* free format : find next sync to compute frame size */
bb270c08
DB
2574 len = MPA_MAX_CODED_FRAME_SIZE - len;
2575 if (len > buf_size)
2576 len = buf_size;
239c2f4c 2577 if (len == 0) {
bb270c08 2578 /* frame too long: resync */
239c2f4c 2579 s->frame_size = 0;
bb270c08
DB
2580 memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);
2581 s->inbuf_ptr--;
239c2f4c 2582 } else {
0c1a9eda
ZK
2583 uint8_t *p, *pend;
2584 uint32_t header1;
239c2f4c
FB
2585 int padding;
2586
2587 memcpy(s->inbuf_ptr, buf_ptr, len);
2588 /* check for header */
2589 p = s->inbuf_ptr - 3;
2590 pend = s->inbuf_ptr + len - 4;
2591 while (p <= pend) {
2592 header = (p[0] << 24) | (p[1] << 16) |
2593 (p[2] << 8) | p[3];
2594 header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2595 (s->inbuf[2] << 8) | s->inbuf[3];
2596 /* check with high probability that we have a
2597 valid header */
2598 if ((header & SAME_HEADER_MASK) ==
2599 (header1 & SAME_HEADER_MASK)) {
2600 /* header found: update pointers */
2601 len = (p + 4) - s->inbuf_ptr;
2602 buf_ptr += len;
2603 buf_size -= len;
2604 s->inbuf_ptr = p;
2605 /* compute frame size */
2606 s->free_format_next_header = header;
2607 s->free_format_frame_size = s->inbuf_ptr - s->inbuf;
2608 padding = (header1 >> 9) & 1;
2609 if (s->layer == 1)
2610 s->free_format_frame_size -= padding * 4;
2611 else
2612 s->free_format_frame_size -= padding;
115329f1 2613 dprintf("free frame size=%d padding=%d\n",
239c2f4c
FB
2614 s->free_format_frame_size, padding);
2615 decode_header(s, header1);
2616 goto next_data;
2617 }
2618 p++;
2619 }
2620 /* not found: simply increase pointers */
2621 buf_ptr += len;
2622 s->inbuf_ptr += len;
2623 buf_size -= len;
2624 }
bb270c08 2625 } else if (len < s->frame_size) {
de5123dc
ZK
2626 if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)
2627 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;
bb270c08
DB
2628 len = s->frame_size - len;
2629 if (len > buf_size)
2630 len = buf_size;
2631 memcpy(s->inbuf_ptr, buf_ptr, len);
2632 buf_ptr += len;
2633 s->inbuf_ptr += len;
2634 buf_size -= len;
2635 }
8c5b5683 2636 next_data:
115329f1 2637 if (s->frame_size > 0 &&
8c5b5683
FB
2638 (s->inbuf_ptr - s->inbuf) >= s->frame_size) {
2639 if (avctx->parse_only) {
2640 /* simply return the frame data */
2641 *(uint8_t **)data = s->inbuf;
2642 out_size = s->inbuf_ptr - s->inbuf;
2643 } else {
2644 out_size = mp_decode_frame(s, out_samples);
2645 }
bb270c08
DB
2646 s->inbuf_ptr = s->inbuf;
2647 s->frame_size = 0;
02af2269 2648 if(out_size>=0)
bb270c08 2649 *data_size = out_size;
02af2269
MN
2650 else
2651 av_log(avctx, AV_LOG_DEBUG, "Error while decoding mpeg audio frame\n"); //FIXME return -1 / but also return the number of bytes consumed
bb270c08
DB
2652 break;
2653 }
de6d9b64
FB
2654 }
2655 return buf_ptr - buf;
2656}
2657
1ede228a
RT
2658
2659static int decode_frame_adu(AVCodecContext * avctx,
bb270c08
DB
2660 void *data, int *data_size,
2661 uint8_t * buf, int buf_size)
1ede228a
RT
2662{
2663 MPADecodeContext *s = avctx->priv_data;
2664 uint32_t header;
2665 int len, out_size;
a3a5f4d6 2666 OUT_INT *out_samples = data;
1ede228a
RT
2667
2668 len = buf_size;
2669
2670 // Discard too short frames
2671 if (buf_size < HEADER_SIZE) {
2672 *data_size = 0;
2673 return buf_size;
2674 }
2675
2676
2677 if (len > MPA_MAX_CODED_FRAME_SIZE)
2678 len = MPA_MAX_CODED_FRAME_SIZE;
2679
2680 memcpy(s->inbuf, buf, len);
2681 s->inbuf_ptr = s->inbuf + len;
2682
2683 // Get header and restore sync word
2684 header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |
2685 (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;
2686
a7a85899 2687 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1ede228a
RT
2688 *data_size = 0;
2689 return buf_size;
2690 }
2691
2692 decode_header(s, header);
2693 /* update codec info */
2694 avctx->sample_rate = s->sample_rate;
2695 avctx->channels = s->nb_channels;
2696 avctx->bit_rate = s->bit_rate;
2697 avctx->sub_id = s->layer;
2698
2699 avctx->frame_size=s->frame_size = len;
2700
2701 if (avctx->parse_only) {
2702 /* simply return the frame data */
2703 *(uint8_t **)data = s->inbuf;
2704 out_size = s->inbuf_ptr - s->inbuf;
2705 } else {
2706 out_size = mp_decode_frame(s, out_samples);
2707 }
2708
2709 *data_size = out_size;
2710 return buf_size;
2711}
2712
2713
d2a7718d
RT
2714/* Next 3 arrays are indexed by channel config number (passed via codecdata) */
2715static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2}; /* number of mp3 decoder instances */
2716static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */
2717/* offsets into output buffer, assume output order is FL FR BL BR C LFE */
2718static int chan_offset[9][5] = {
2719 {0},
2720 {0}, // C
2721 {0}, // FLR
2722 {2,0}, // C FLR
2723 {2,0,3}, // C FLR BS
2724 {4,0,2}, // C FLR BLRS
2725 {4,0,2,5}, // C FLR BLRS LFE
2726 {4,0,2,6,5}, // C FLR BLRS BLR LFE
2727 {0,2} // FLR BLRS
2728};
2729
2730
2731static int decode_init_mp3on4(AVCodecContext * avctx)
2732{
2733 MP3On4DecodeContext *s = avctx->priv_data;
2734 int i;
2735
2736 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
2737 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
2738 return -1;
2739 }
2740
2741 s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f;
2742 s->frames = mp3Frames[s->chan_cfg];
2743 if(!s->frames) {
2744 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
2745 return -1;
2746 }
2747 avctx->channels = mp3Channels[s->chan_cfg];
2748
2749 /* Init the first mp3 decoder in standard way, so that all tables get builded
2750 * We replace avctx->priv_data with the context of the first decoder so that
2751 * decode_init() does not have to be changed.
2752 * Other decoders will be inited here copying data from the first context
2753 */
2754 // Allocate zeroed memory for the first decoder context
2755 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
2756 // Put decoder context in place to make init_decode() happy
2757 avctx->priv_data = s->mp3decctx[0];
2758 decode_init(avctx);
2759 // Restore mp3on4 context pointer
2760 avctx->priv_data = s;
2761 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
2762
2763 /* Create a separate codec/context for each frame (first is already ok).
2764 * Each frame is 1 or 2 channels - up to 5 frames allowed
2765 */
2766 for (i = 1; i < s->frames; i++) {
2767 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
2768 s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;
2769 s->mp3decctx[i]->inbuf = &s->mp3decctx[i]->inbuf1[0][BACKSTEP_SIZE];
2770 s->mp3decctx[i]->inbuf_ptr = s->mp3decctx[i]->inbuf;
2771 s->mp3decctx[i]->adu_mode = 1;
2772 }
2773
2774 return 0;
2775}
2776
2777
2778static int decode_close_mp3on4(AVCodecContext * avctx)
2779{
2780 MP3On4DecodeContext *s = avctx->priv_data;
2781 int i;
2782
2783 for (i = 0; i < s->frames; i++)
2784 if (s->mp3decctx[i])
2785 av_free(s->mp3decctx[i]);
2786
2787 return 0;
2788}
2789
2790
2791static int decode_frame_mp3on4(AVCodecContext * avctx,
bb270c08
DB
2792 void *data, int *data_size,
2793 uint8_t * buf, int buf_size)
d2a7718d
RT
2794{
2795 MP3On4DecodeContext *s = avctx->priv_data;
2796 MPADecodeContext *m;
2797 int len, out_size = 0;
2798 uint32_t header;
2799 OUT_INT *out_samples = data;
2800 OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
2801 OUT_INT *outptr, *bp;
2802 int fsize;
2803 unsigned char *start2 = buf, *start;
2804 int fr, i, j, n;
2805 int off = avctx->channels;
2806 int *coff = chan_offset[s->chan_cfg];
2807
2808 len = buf_size;
2809
2810 // Discard too short frames
2811 if (buf_size < HEADER_SIZE) {
2812 *data_size = 0;
2813 return buf_size;
2814 }
2815
2816 // If only one decoder interleave is not needed
2817 outptr = s->frames == 1 ? out_samples : decoded_buf;
2818
2819 for (fr = 0; fr < s->frames; fr++) {
2820 start = start2;
2821 fsize = (start[0] << 4) | (start[1] >> 4);
2822 start2 += fsize;
2823 if (fsize > len)
2824 fsize = len;
2825 len -= fsize;
2826 if (fsize > MPA_MAX_CODED_FRAME_SIZE)
2827 fsize = MPA_MAX_CODED_FRAME_SIZE;
2828 m = s->mp3decctx[fr];
2829 assert (m != NULL);
2830 /* copy original to new */
2831 m->inbuf_ptr = m->inbuf + fsize;
2832 memcpy(m->inbuf, start, fsize);
2833
2834 // Get header
2835 header = (m->inbuf[0] << 24) | (m->inbuf[1] << 16) |
2836 (m->inbuf[2] << 8) | m->inbuf[3] | 0xfff00000;
2837
2838 if (ff_mpa_check_header(header) < 0) { // Bad header, discard block
2839 *data_size = 0;
2840 return buf_size;
2841 }
2842
2843 decode_header(m, header);
2844 mp_decode_frame(m, decoded_buf);
2845
2846 n = MPA_FRAME_SIZE * m->nb_channels;
2847 out_size += n * sizeof(OUT_INT);
2848 if(s->frames > 1) {
2849 /* interleave output data */
2850 bp = out_samples + coff[fr];
2851 if(m->nb_channels == 1) {
2852 for(j = 0; j < n; j++) {
2853 *bp = decoded_buf[j];
2854 bp += off;
2855 }
2856 } else {
2857 for(j = 0; j < n; j++) {
2858 bp[0] = decoded_buf[j++];
2859 bp[1] = decoded_buf[j];
2860 bp += off;
2861 }
2862 }
2863 }
2864 }
2865
2866 /* update codec info */
2867 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2868 avctx->frame_size= buf_size;
2869 avctx->bit_rate = 0;
2870 for (i = 0; i < s->frames; i++)
2871 avctx->bit_rate += s->mp3decctx[i]->bit_rate;
2872
2873 *data_size = out_size;
2874 return buf_size;
2875}
2876
2877
4b1f4f23 2878AVCodec mp2_decoder =
de6d9b64 2879{
4b1f4f23 2880 "mp2",
de6d9b64
FB
2881 CODEC_TYPE_AUDIO,
2882 CODEC_ID_MP2,
2883 sizeof(MPADecodeContext),
2884 decode_init,
2885 NULL,
2886 NULL,
2887 decode_frame,
8c5b5683 2888 CODEC_CAP_PARSE_ONLY,
de6d9b64 2889};
4b1f4f23
J
2890
2891AVCodec mp3_decoder =
2892{
2893 "mp3",
2894 CODEC_TYPE_AUDIO,
80783dc2 2895 CODEC_ID_MP3,
4b1f4f23
J
2896 sizeof(MPADecodeContext),
2897 decode_init,
2898 NULL,
2899 NULL,
2900 decode_frame,
8c5b5683 2901 CODEC_CAP_PARSE_ONLY,
4b1f4f23 2902};
1ede228a
RT
2903
2904AVCodec mp3adu_decoder =
2905{
2906 "mp3adu",
2907 CODEC_TYPE_AUDIO,
2908 CODEC_ID_MP3ADU,
2909 sizeof(MPADecodeContext),
2910 decode_init,
2911 NULL,
2912 NULL,
2913 decode_frame_adu,
2914 CODEC_CAP_PARSE_ONLY,
2915};
d2a7718d
RT
2916
2917AVCodec mp3on4_decoder =
2918{
2919 "mp3on4",
2920 CODEC_TYPE_AUDIO,
2921 CODEC_ID_MP3ON4,
2922 sizeof(MP3On4DecodeContext),
2923 decode_init_mp3on4,
2924 NULL,
2925 decode_close_mp3on4,
2926 decode_frame_mp3on4,
2927 0
2928};