1 /*
2 * MPEG Audio decoder
3 * Copyright (c) 2001, 2002 Fabrice Bellard
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * MPEG Audio decoder
25 */
26
27 #include "libavutil/attributes.h"
28 #include "libavutil/avassert.h"
29 #include "libavutil/channel_layout.h"
30 #include "libavutil/float_dsp.h"
31 #include "libavutil/libm.h"
32 #include "avcodec.h"
33 #include "get_bits.h"
34 #include "internal.h"
35 #include "mathops.h"
36 #include "mpegaudiodsp.h"
37
38 /*
39 * TODO:
40 * - test lsf / mpeg25 extensively.
41 */
42
43 #include "mpegaudio.h"
44 #include "mpegaudiodecheader.h"
45
46 #define BACKSTEP_SIZE 512
47 #define EXTRABYTES 24
48 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES
49
50 /* layer 3 "granule" */
51 typedef struct GranuleDef {
52 uint8_t scfsi;
53 int part2_3_length;
54 int big_values;
55 int global_gain;
56 int scalefac_compress;
57 uint8_t block_type;
58 uint8_t switch_point;
59 int table_select[3];
60 int subblock_gain[3];
61 uint8_t scalefac_scale;
62 uint8_t count1table_select;
63 int region_size[3]; /* number of huffman codes in each region */
64 int preflag;
65 int short_start, long_end; /* long/short band indexes */
66 uint8_t scale_factors[40];
67 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
68 } GranuleDef;
69
70 typedef struct MPADecodeContext {
71 MPA_DECODE_HEADER
72 uint8_t last_buf[LAST_BUF_SIZE];
73 int last_buf_size;
74 /* next header (used in free format parsing) */
75 uint32_t free_format_next_header;
76 GetBitContext gb;
77 GetBitContext in_gb;
78 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
79 int synth_buf_offset[MPA_MAX_CHANNELS];
80 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
81 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
82 GranuleDef granules[2][2]; /* Used in Layer 3 */
83 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
84 int dither_state;
85 int err_recognition;
86 AVCodecContext* avctx;
87 MPADSPContext mpadsp;
88 AVFloatDSPContext fdsp;
89 AVFrame *frame;
90 } MPADecodeContext;
91
92 #define HEADER_SIZE 4
93
94 #include "mpegaudiodata.h"
95 #include "mpegaudiodectab.h"
96
97 /* vlc structure for decoding layer 3 huffman tables */
98 static VLC huff_vlc[16];
99 static VLC_TYPE huff_vlc_tables[
100 0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +
101 142 + 204 + 190 + 170 + 542 + 460 + 662 + 414
102 ][2];
103 static const int huff_vlc_tables_sizes[16] = {
104 0, 128, 128, 128, 130, 128, 154, 166,
105 142, 204, 190, 170, 542, 460, 662, 414
106 };
107 static VLC huff_quad_vlc[2];
108 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
109 static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };
110 /* computed from band_size_long */
111 static uint16_t band_index_long[9][23];
112 #include "mpegaudio_tablegen.h"
113 /* intensity stereo coef table */
114 static INTFLOAT is_table[2][16];
115 static INTFLOAT is_table_lsf[2][2][16];
116 static INTFLOAT csa_table[8][4];
117
118 static int16_t division_tab3[1<<6 ];
119 static int16_t division_tab5[1<<8 ];
120 static int16_t division_tab9[1<<11];
121
122 static int16_t * const division_tabs[4] = {
123 division_tab3, division_tab5, NULL, division_tab9
124 };
125
126 /* lower 2 bits: modulo 3, higher bits: shift */
127 static uint16_t scale_factor_modshift[64];
128 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
129 static int32_t scale_factor_mult[15][3];
130 /* mult table for layer 2 group quantization */
131
132 #define SCALE_GEN(v) \
133 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
134
135 static const int32_t scale_factor_mult2[3][3] = {
136 SCALE_GEN(4.0 / 3.0), /* 3 steps */
137 SCALE_GEN(4.0 / 5.0), /* 5 steps */
138 SCALE_GEN(4.0 / 9.0), /* 9 steps */
139 };
140
141 /**
142 * Convert region offsets to region sizes and truncate
143 * size to big_values.
144 */
region_offset2size(GranuleDef * g)145 static void region_offset2size(GranuleDef *g)
146 {
147 int i, k, j = 0;
148 g->region_size[2] = 576 / 2;
149 for (i = 0; i < 3; i++) {
150 k = FFMIN(g->region_size[i], g->big_values);
151 g->region_size[i] = k - j;
152 j = k;
153 }
154 }
155
init_short_region(MPADecodeContext * s,GranuleDef * g)156 static void init_short_region(MPADecodeContext *s, GranuleDef *g)
157 {
158 if (g->block_type == 2) {
159 if (s->sample_rate_index != 8)
160 g->region_size[0] = (36 / 2);
161 else
162 g->region_size[0] = (72 / 2);
163 } else {
164 if (s->sample_rate_index <= 2)
165 g->region_size[0] = (36 / 2);
166 else if (s->sample_rate_index != 8)
167 g->region_size[0] = (54 / 2);
168 else
169 g->region_size[0] = (108 / 2);
170 }
171 g->region_size[1] = (576 / 2);
172 }
173
init_long_region(MPADecodeContext * s,GranuleDef * g,int ra1,int ra2)174 static void init_long_region(MPADecodeContext *s, GranuleDef *g,
175 int ra1, int ra2)
176 {
177 int l;
178 g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
179 /* should not overflow */
180 l = FFMIN(ra1 + ra2 + 2, 22);
181 g->region_size[1] = band_index_long[s->sample_rate_index][ l] >> 1;
182 }
183
compute_band_indexes(MPADecodeContext * s,GranuleDef * g)184 static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
185 {
186 if (g->block_type == 2) {
187 if (g->switch_point) {
188 if(s->sample_rate_index == 8)
189 avpriv_request_sample(s->avctx, "switch point in 8khz");
190 /* if switched mode, we handle the 36 first samples as
191 long blocks. For 8000Hz, we handle the 72 first
192 exponents as long blocks */
193 if (s->sample_rate_index <= 2)
194 g->long_end = 8;
195 else
196 g->long_end = 6;
197
198 g->short_start = 3;
199 } else {
200 g->long_end = 0;
201 g->short_start = 0;
202 }
203 } else {
204 g->short_start = 13;
205 g->long_end = 22;
206 }
207 }
208
209 /* layer 1 unscaling */
210 /* n = number of bits of the mantissa minus 1 */
l1_unscale(int n,int mant,int scale_factor)211 static inline int l1_unscale(int n, int mant, int scale_factor)
212 {
213 int shift, mod;
214 int64_t val;
215
216 shift = scale_factor_modshift[scale_factor];
217 mod = shift & 3;
218 shift >>= 2;
219 val = MUL64((int)(mant + (-1U << n) + 1), scale_factor_mult[n-1][mod]);
220 shift += n;
221 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
222 return (int)((val + (LLN(1) << (shift - 1))) >> shift);
223 }
224
l2_unscale_group(int steps,int mant,int scale_factor)225 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
226 {
227 int shift, mod, val;
228
229 shift = scale_factor_modshift[scale_factor];
230 mod = shift & 3;
231 shift >>= 2;
232
233 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
234 /* NOTE: at this point, 0 <= shift <= 21 */
235 if (shift > 0)
236 val = (val + (1 << (shift - 1))) >> shift;
237 return val;
238 }
239
240 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
l3_unscale(int value,int exponent)241 static inline int l3_unscale(int value, int exponent)
242 {
243 unsigned int m;
244 int e;
245
246 e = table_4_3_exp [4 * value + (exponent & 3)];
247 m = table_4_3_value[4 * value + (exponent & 3)];
248 e -= exponent >> 2;
249 #ifdef DEBUG
250 if(e < 1)
251 av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e);
252 #endif
253 if (e > 31)
254 return 0;
255 m = (m + (1 << (e - 1))) >> e;
256
257 return m;
258 }
259
decode_init_static(void)260 static av_cold void decode_init_static(void)
261 {
262 int i, j, k;
263 int offset;
264
265 /* scale factors table for layer 1/2 */
266 for (i = 0; i < 64; i++) {
267 int shift, mod;
268 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
269 shift = i / 3;
270 mod = i % 3;
271 scale_factor_modshift[i] = mod | (shift << 2);
272 }
273
274 /* scale factor multiply for layer 1 */
275 for (i = 0; i < 15; i++) {
276 int n, norm;
277 n = i + 2;
278 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
279 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
280 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
281 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
282 av_dlog(NULL, "%d: norm=%x s=%x %x %x\n", i, norm,
283 scale_factor_mult[i][0],
284 scale_factor_mult[i][1],
285 scale_factor_mult[i][2]);
286 }
287
288 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
289
290 /* huffman decode tables */
291 offset = 0;
292 for (i = 1; i < 16; i++) {
293 const HuffTable *h = &mpa_huff_tables[i];
294 int xsize, x, y;
295 uint8_t tmp_bits [512] = { 0 };
296 uint16_t tmp_codes[512] = { 0 };
297
298 xsize = h->xsize;
299
300 j = 0;
301 for (x = 0; x < xsize; x++) {
302 for (y = 0; y < xsize; y++) {
303 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
304 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
305 }
306 }
307
308 /* XXX: fail test */
309 huff_vlc[i].table = huff_vlc_tables+offset;
310 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
311 init_vlc(&huff_vlc[i], 7, 512,
312 tmp_bits, 1, 1, tmp_codes, 2, 2,
313 INIT_VLC_USE_NEW_STATIC);
314 offset += huff_vlc_tables_sizes[i];
315 }
316 av_assert0(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
317
318 offset = 0;
319 for (i = 0; i < 2; i++) {
320 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
321 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
322 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
323 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
324 INIT_VLC_USE_NEW_STATIC);
325 offset += huff_quad_vlc_tables_sizes[i];
326 }
327 av_assert0(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
328
329 for (i = 0; i < 9; i++) {
330 k = 0;
331 for (j = 0; j < 22; j++) {
332 band_index_long[i][j] = k;
333 k += band_size_long[i][j];
334 }
335 band_index_long[i][22] = k;
336 }
337
338 /* compute n ^ (4/3) and store it in mantissa/exp format */
339
340 mpegaudio_tableinit();
341
342 for (i = 0; i < 4; i++) {
343 if (ff_mpa_quant_bits[i] < 0) {
344 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
345 int val1, val2, val3, steps;
346 int val = j;
347 steps = ff_mpa_quant_steps[i];
348 val1 = val % steps;
349 val /= steps;
350 val2 = val % steps;
351 val3 = val / steps;
352 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
353 }
354 }
355 }
356
357
358 for (i = 0; i < 7; i++) {
359 float f;
360 INTFLOAT v;
361 if (i != 6) {
362 f = tan((double)i * M_PI / 12.0);
363 v = FIXR(f / (1.0 + f));
364 } else {
365 v = FIXR(1.0);
366 }
367 is_table[0][ i] = v;
368 is_table[1][6 - i] = v;
369 }
370 /* invalid values */
371 for (i = 7; i < 16; i++)
372 is_table[0][i] = is_table[1][i] = 0.0;
373
374 for (i = 0; i < 16; i++) {
375 double f;
376 int e, k;
377
378 for (j = 0; j < 2; j++) {
379 e = -(j + 1) * ((i + 1) >> 1);
380 f = exp2(e / 4.0);
381 k = i & 1;
382 is_table_lsf[j][k ^ 1][i] = FIXR(f);
383 is_table_lsf[j][k ][i] = FIXR(1.0);
384 av_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
385 i, j, (float) is_table_lsf[j][0][i],
386 (float) is_table_lsf[j][1][i]);
387 }
388 }
389
390 for (i = 0; i < 8; i++) {
391 float ci, cs, ca;
392 ci = ci_table[i];
393 cs = 1.0 / sqrt(1.0 + ci * ci);
394 ca = cs * ci;
395 #if !USE_FLOATS
396 csa_table[i][0] = FIXHR(cs/4);
397 csa_table[i][1] = FIXHR(ca/4);
398 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
399 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
400 #else
401 csa_table[i][0] = cs;
402 csa_table[i][1] = ca;
403 csa_table[i][2] = ca + cs;
404 csa_table[i][3] = ca - cs;
405 #endif
406 }
407 }
408
decode_init(AVCodecContext * avctx)409 static av_cold int decode_init(AVCodecContext * avctx)
410 {
411 static int initialized_tables = 0;
412 MPADecodeContext *s = avctx->priv_data;
413
414 if (!initialized_tables) {
415 decode_init_static();
416 initialized_tables = 1;
417 }
418
419 s->avctx = avctx;
420
421 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
422 ff_mpadsp_init(&s->mpadsp);
423
424 if (avctx->request_sample_fmt == OUT_FMT &&
425 avctx->codec_id != AV_CODEC_ID_MP3ON4)
426 avctx->sample_fmt = OUT_FMT;
427 else
428 avctx->sample_fmt = OUT_FMT_P;
429 s->err_recognition = avctx->err_recognition;
430
431 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
432 s->adu_mode = 1;
433
434 return 0;
435 }
436
437 #define C3 FIXHR(0.86602540378443864676/2)
438 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
439 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
440 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
441
442 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
443 cases. */
imdct12(INTFLOAT * out,INTFLOAT * in)444 static void imdct12(INTFLOAT *out, INTFLOAT *in)
445 {
446 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
447
448 in0 = in[0*3];
449 in1 = in[1*3] + in[0*3];
450 in2 = in[2*3] + in[1*3];
451 in3 = in[3*3] + in[2*3];
452 in4 = in[4*3] + in[3*3];
453 in5 = in[5*3] + in[4*3];
454 in5 += in3;
455 in3 += in1;
456
457 in2 = MULH3(in2, C3, 2);
458 in3 = MULH3(in3, C3, 4);
459
460 t1 = in0 - in4;
461 t2 = MULH3(in1 - in5, C4, 2);
462
463 out[ 7] =
464 out[10] = t1 + t2;
465 out[ 1] =
466 out[ 4] = t1 - t2;
467
468 in0 += SHR(in4, 1);
469 in4 = in0 + in2;
470 in5 += 2*in1;
471 in1 = MULH3(in5 + in3, C5, 1);
472 out[ 8] =
473 out[ 9] = in4 + in1;
474 out[ 2] =
475 out[ 3] = in4 - in1;
476
477 in0 -= in2;
478 in5 = MULH3(in5 - in3, C6, 2);
479 out[ 0] =
480 out[ 5] = in0 - in5;
481 out[ 6] =
482 out[11] = in0 + in5;
483 }
484
485 /* return the number of decoded frames */
mp_decode_layer1(MPADecodeContext * s)486 static int mp_decode_layer1(MPADecodeContext *s)
487 {
488 int bound, i, v, n, ch, j, mant;
489 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
490 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
491
492 if (s->mode == MPA_JSTEREO)
493 bound = (s->mode_ext + 1) * 4;
494 else
495 bound = SBLIMIT;
496
497 /* allocation bits */
498 for (i = 0; i < bound; i++) {
499 for (ch = 0; ch < s->nb_channels; ch++) {
500 allocation[ch][i] = get_bits(&s->gb, 4);
501 }
502 }
503 for (i = bound; i < SBLIMIT; i++)
504 allocation[0][i] = get_bits(&s->gb, 4);
505
506 /* scale factors */
507 for (i = 0; i < bound; i++) {
508 for (ch = 0; ch < s->nb_channels; ch++) {
509 if (allocation[ch][i])
510 scale_factors[ch][i] = get_bits(&s->gb, 6);
511 }
512 }
513 for (i = bound; i < SBLIMIT; i++) {
514 if (allocation[0][i]) {
515 scale_factors[0][i] = get_bits(&s->gb, 6);
516 scale_factors[1][i] = get_bits(&s->gb, 6);
517 }
518 }
519
520 /* compute samples */
521 for (j = 0; j < 12; j++) {
522 for (i = 0; i < bound; i++) {
523 for (ch = 0; ch < s->nb_channels; ch++) {
524 n = allocation[ch][i];
525 if (n) {
526 mant = get_bits(&s->gb, n + 1);
527 v = l1_unscale(n, mant, scale_factors[ch][i]);
528 } else {
529 v = 0;
530 }
531 s->sb_samples[ch][j][i] = v;
532 }
533 }
534 for (i = bound; i < SBLIMIT; i++) {
535 n = allocation[0][i];
536 if (n) {
537 mant = get_bits(&s->gb, n + 1);
538 v = l1_unscale(n, mant, scale_factors[0][i]);
539 s->sb_samples[0][j][i] = v;
540 v = l1_unscale(n, mant, scale_factors[1][i]);
541 s->sb_samples[1][j][i] = v;
542 } else {
543 s->sb_samples[0][j][i] = 0;
544 s->sb_samples[1][j][i] = 0;
545 }
546 }
547 }
548 return 12;
549 }
550
mp_decode_layer2(MPADecodeContext * s)551 static int mp_decode_layer2(MPADecodeContext *s)
552 {
553 int sblimit; /* number of used subbands */
554 const unsigned char *alloc_table;
555 int table, bit_alloc_bits, i, j, ch, bound, v;
556 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
557 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
558 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
559 int scale, qindex, bits, steps, k, l, m, b;
560
561 /* select decoding table */
562 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
563 s->sample_rate, s->lsf);
564 sblimit = ff_mpa_sblimit_table[table];
565 alloc_table = ff_mpa_alloc_tables[table];
566
567 if (s->mode == MPA_JSTEREO)
568 bound = (s->mode_ext + 1) * 4;
569 else
570 bound = sblimit;
571
572 av_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
573
574 /* sanity check */
575 if (bound > sblimit)
576 bound = sblimit;
577
578 /* parse bit allocation */
579 j = 0;
580 for (i = 0; i < bound; i++) {
581 bit_alloc_bits = alloc_table[j];
582 for (ch = 0; ch < s->nb_channels; ch++)
583 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
584 j += 1 << bit_alloc_bits;
585 }
586 for (i = bound; i < sblimit; i++) {
587 bit_alloc_bits = alloc_table[j];
588 v = get_bits(&s->gb, bit_alloc_bits);
589 bit_alloc[0][i] = v;
590 bit_alloc[1][i] = v;
591 j += 1 << bit_alloc_bits;
592 }
593
594 /* scale codes */
595 for (i = 0; i < sblimit; i++) {
596 for (ch = 0; ch < s->nb_channels; ch++) {
597 if (bit_alloc[ch][i])
598 scale_code[ch][i] = get_bits(&s->gb, 2);
599 }
600 }
601
602 /* scale factors */
603 for (i = 0; i < sblimit; i++) {
604 for (ch = 0; ch < s->nb_channels; ch++) {
605 if (bit_alloc[ch][i]) {
606 sf = scale_factors[ch][i];
607 switch (scale_code[ch][i]) {
608 default:
609 case 0:
610 sf[0] = get_bits(&s->gb, 6);
611 sf[1] = get_bits(&s->gb, 6);
612 sf[2] = get_bits(&s->gb, 6);
613 break;
614 case 2:
615 sf[0] = get_bits(&s->gb, 6);
616 sf[1] = sf[0];
617 sf[2] = sf[0];
618 break;
619 case 1:
620 sf[0] = get_bits(&s->gb, 6);
621 sf[2] = get_bits(&s->gb, 6);
622 sf[1] = sf[0];
623 break;
624 case 3:
625 sf[0] = get_bits(&s->gb, 6);
626 sf[2] = get_bits(&s->gb, 6);
627 sf[1] = sf[2];
628 break;
629 }
630 }
631 }
632 }
633
634 /* samples */
635 for (k = 0; k < 3; k++) {
636 for (l = 0; l < 12; l += 3) {
637 j = 0;
638 for (i = 0; i < bound; i++) {
639 bit_alloc_bits = alloc_table[j];
640 for (ch = 0; ch < s->nb_channels; ch++) {
641 b = bit_alloc[ch][i];
642 if (b) {
643 scale = scale_factors[ch][i][k];
644 qindex = alloc_table[j+b];
645 bits = ff_mpa_quant_bits[qindex];
646 if (bits < 0) {
647 int v2;
648 /* 3 values at the same time */
649 v = get_bits(&s->gb, -bits);
650 v2 = division_tabs[qindex][v];
651 steps = ff_mpa_quant_steps[qindex];
652
653 s->sb_samples[ch][k * 12 + l + 0][i] =
654 l2_unscale_group(steps, v2 & 15, scale);
655 s->sb_samples[ch][k * 12 + l + 1][i] =
656 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
657 s->sb_samples[ch][k * 12 + l + 2][i] =
658 l2_unscale_group(steps, v2 >> 8 , scale);
659 } else {
660 for (m = 0; m < 3; m++) {
661 v = get_bits(&s->gb, bits);
662 v = l1_unscale(bits - 1, v, scale);
663 s->sb_samples[ch][k * 12 + l + m][i] = v;
664 }
665 }
666 } else {
667 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
668 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
669 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
670 }
671 }
672 /* next subband in alloc table */
673 j += 1 << bit_alloc_bits;
674 }
675 /* XXX: find a way to avoid this duplication of code */
676 for (i = bound; i < sblimit; i++) {
677 bit_alloc_bits = alloc_table[j];
678 b = bit_alloc[0][i];
679 if (b) {
680 int mant, scale0, scale1;
681 scale0 = scale_factors[0][i][k];
682 scale1 = scale_factors[1][i][k];
683 qindex = alloc_table[j+b];
684 bits = ff_mpa_quant_bits[qindex];
685 if (bits < 0) {
686 /* 3 values at the same time */
687 v = get_bits(&s->gb, -bits);
688 steps = ff_mpa_quant_steps[qindex];
689 mant = v % steps;
690 v = v / steps;
691 s->sb_samples[0][k * 12 + l + 0][i] =
692 l2_unscale_group(steps, mant, scale0);
693 s->sb_samples[1][k * 12 + l + 0][i] =
694 l2_unscale_group(steps, mant, scale1);
695 mant = v % steps;
696 v = v / steps;
697 s->sb_samples[0][k * 12 + l + 1][i] =
698 l2_unscale_group(steps, mant, scale0);
699 s->sb_samples[1][k * 12 + l + 1][i] =
700 l2_unscale_group(steps, mant, scale1);
701 s->sb_samples[0][k * 12 + l + 2][i] =
702 l2_unscale_group(steps, v, scale0);
703 s->sb_samples[1][k * 12 + l + 2][i] =
704 l2_unscale_group(steps, v, scale1);
705 } else {
706 for (m = 0; m < 3; m++) {
707 mant = get_bits(&s->gb, bits);
708 s->sb_samples[0][k * 12 + l + m][i] =
709 l1_unscale(bits - 1, mant, scale0);
710 s->sb_samples[1][k * 12 + l + m][i] =
711 l1_unscale(bits - 1, mant, scale1);
712 }
713 }
714 } else {
715 s->sb_samples[0][k * 12 + l + 0][i] = 0;
716 s->sb_samples[0][k * 12 + l + 1][i] = 0;
717 s->sb_samples[0][k * 12 + l + 2][i] = 0;
718 s->sb_samples[1][k * 12 + l + 0][i] = 0;
719 s->sb_samples[1][k * 12 + l + 1][i] = 0;
720 s->sb_samples[1][k * 12 + l + 2][i] = 0;
721 }
722 /* next subband in alloc table */
723 j += 1 << bit_alloc_bits;
724 }
725 /* fill remaining samples to zero */
726 for (i = sblimit; i < SBLIMIT; i++) {
727 for (ch = 0; ch < s->nb_channels; ch++) {
728 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
729 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
730 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
731 }
732 }
733 }
734 }
735 return 3 * 12;
736 }
737
738 #define SPLIT(dst,sf,n) \
739 if (n == 3) { \
740 int m = (sf * 171) >> 9; \
741 dst = sf - 3 * m; \
742 sf = m; \
743 } else if (n == 4) { \
744 dst = sf & 3; \
745 sf >>= 2; \
746 } else if (n == 5) { \
747 int m = (sf * 205) >> 10; \
748 dst = sf - 5 * m; \
749 sf = m; \
750 } else if (n == 6) { \
751 int m = (sf * 171) >> 10; \
752 dst = sf - 6 * m; \
753 sf = m; \
754 } else { \
755 dst = 0; \
756 }
757
lsf_sf_expand(int * slen,int sf,int n1,int n2,int n3)758 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
759 int n3)
760 {
761 SPLIT(slen[3], sf, n3)
762 SPLIT(slen[2], sf, n2)
763 SPLIT(slen[1], sf, n1)
764 slen[0] = sf;
765 }
766
exponents_from_scale_factors(MPADecodeContext * s,GranuleDef * g,int16_t * exponents)767 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
768 int16_t *exponents)
769 {
770 const uint8_t *bstab, *pretab;
771 int len, i, j, k, l, v0, shift, gain, gains[3];
772 int16_t *exp_ptr;
773
774 exp_ptr = exponents;
775 gain = g->global_gain - 210;
776 shift = g->scalefac_scale + 1;
777
778 bstab = band_size_long[s->sample_rate_index];
779 pretab = mpa_pretab[g->preflag];
780 for (i = 0; i < g->long_end; i++) {
781 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
782 len = bstab[i];
783 for (j = len; j > 0; j--)
784 *exp_ptr++ = v0;
785 }
786
787 if (g->short_start < 13) {
788 bstab = band_size_short[s->sample_rate_index];
789 gains[0] = gain - (g->subblock_gain[0] << 3);
790 gains[1] = gain - (g->subblock_gain[1] << 3);
791 gains[2] = gain - (g->subblock_gain[2] << 3);
792 k = g->long_end;
793 for (i = g->short_start; i < 13; i++) {
794 len = bstab[i];
795 for (l = 0; l < 3; l++) {
796 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
797 for (j = len; j > 0; j--)
798 *exp_ptr++ = v0;
799 }
800 }
801 }
802 }
803
804 /* handle n = 0 too */
get_bitsz(GetBitContext * s,int n)805 static inline int get_bitsz(GetBitContext *s, int n)
806 {
807 return n ? get_bits(s, n) : 0;
808 }
809
810
switch_buffer(MPADecodeContext * s,int * pos,int * end_pos,int * end_pos2)811 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
812 int *end_pos2)
813 {
814 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits) {
815 s->gb = s->in_gb;
816 s->in_gb.buffer = NULL;
817 av_assert2((get_bits_count(&s->gb) & 7) == 0);
818 skip_bits_long(&s->gb, *pos - *end_pos);
819 *end_pos2 =
820 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
821 *pos = get_bits_count(&s->gb);
822 }
823 }
824
825 /* Following is a optimized code for
826 INTFLOAT v = *src
827 if(get_bits1(&s->gb))
828 v = -v;
829 *dst = v;
830 */
831 #if USE_FLOATS
832 #define READ_FLIP_SIGN(dst,src) \
833 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
834 AV_WN32A(dst, v);
835 #else
836 #define READ_FLIP_SIGN(dst,src) \
837 v = -get_bits1(&s->gb); \
838 *(dst) = (*(src) ^ v) - v;
839 #endif
840
huffman_decode(MPADecodeContext * s,GranuleDef * g,int16_t * exponents,int end_pos2)841 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
842 int16_t *exponents, int end_pos2)
843 {
844 int s_index;
845 int i;
846 int last_pos, bits_left;
847 VLC *vlc;
848 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits);
849
850 /* low frequencies (called big values) */
851 s_index = 0;
852 for (i = 0; i < 3; i++) {
853 int j, k, l, linbits;
854 j = g->region_size[i];
855 if (j == 0)
856 continue;
857 /* select vlc table */
858 k = g->table_select[i];
859 l = mpa_huff_data[k][0];
860 linbits = mpa_huff_data[k][1];
861 vlc = &huff_vlc[l];
862
863 if (!l) {
864 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
865 s_index += 2 * j;
866 continue;
867 }
868
869 /* read huffcode and compute each couple */
870 for (; j > 0; j--) {
871 int exponent, x, y;
872 int v;
873 int pos = get_bits_count(&s->gb);
874
875 if (pos >= end_pos){
876 switch_buffer(s, &pos, &end_pos, &end_pos2);
877 if (pos >= end_pos)
878 break;
879 }
880 y = get_vlc2(&s->gb, vlc->table, 7, 3);
881
882 if (!y) {
883 g->sb_hybrid[s_index ] =
884 g->sb_hybrid[s_index+1] = 0;
885 s_index += 2;
886 continue;
887 }
888
889 exponent= exponents[s_index];
890
891 av_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
892 i, g->region_size[i] - j, x, y, exponent);
893 if (y & 16) {
894 x = y >> 5;
895 y = y & 0x0f;
896 if (x < 15) {
897 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
898 } else {
899 x += get_bitsz(&s->gb, linbits);
900 v = l3_unscale(x, exponent);
901 if (get_bits1(&s->gb))
902 v = -v;
903 g->sb_hybrid[s_index] = v;
904 }
905 if (y < 15) {
906 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
907 } else {
908 y += get_bitsz(&s->gb, linbits);
909 v = l3_unscale(y, exponent);
910 if (get_bits1(&s->gb))
911 v = -v;
912 g->sb_hybrid[s_index+1] = v;
913 }
914 } else {
915 x = y >> 5;
916 y = y & 0x0f;
917 x += y;
918 if (x < 15) {
919 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
920 } else {
921 x += get_bitsz(&s->gb, linbits);
922 v = l3_unscale(x, exponent);
923 if (get_bits1(&s->gb))
924 v = -v;
925 g->sb_hybrid[s_index+!!y] = v;
926 }
927 g->sb_hybrid[s_index + !y] = 0;
928 }
929 s_index += 2;
930 }
931 }
932
933 /* high frequencies */
934 vlc = &huff_quad_vlc[g->count1table_select];
935 last_pos = 0;
936 while (s_index <= 572) {
937 int pos, code;
938 pos = get_bits_count(&s->gb);
939 if (pos >= end_pos) {
940 if (pos > end_pos2 && last_pos) {
941 /* some encoders generate an incorrect size for this
942 part. We must go back into the data */
943 s_index -= 4;
944 skip_bits_long(&s->gb, last_pos - pos);
945 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
946 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
947 s_index=0;
948 break;
949 }
950 switch_buffer(s, &pos, &end_pos, &end_pos2);
951 if (pos >= end_pos)
952 break;
953 }
954 last_pos = pos;
955
956 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
957 av_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
958 g->sb_hybrid[s_index+0] =
959 g->sb_hybrid[s_index+1] =
960 g->sb_hybrid[s_index+2] =
961 g->sb_hybrid[s_index+3] = 0;
962 while (code) {
963 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
964 int v;
965 int pos = s_index + idxtab[code];
966 code ^= 8 >> idxtab[code];
967 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
968 }
969 s_index += 4;
970 }
971 /* skip extension bits */
972 bits_left = end_pos2 - get_bits_count(&s->gb);
973 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
974 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
975 s_index=0;
976 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
977 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
978 s_index = 0;
979 }
980 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
981 skip_bits_long(&s->gb, bits_left);
982
983 i = get_bits_count(&s->gb);
984 switch_buffer(s, &i, &end_pos, &end_pos2);
985
986 return 0;
987 }
988
989 /* Reorder short blocks from bitstream order to interleaved order. It
990 would be faster to do it in parsing, but the code would be far more
991 complicated */
reorder_block(MPADecodeContext * s,GranuleDef * g)992 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
993 {
994 int i, j, len;
995 INTFLOAT *ptr, *dst, *ptr1;
996 INTFLOAT tmp[576];
997
998 if (g->block_type != 2)
999 return;
1000
1001 if (g->switch_point) {
1002 if (s->sample_rate_index != 8)
1003 ptr = g->sb_hybrid + 36;
1004 else
1005 ptr = g->sb_hybrid + 72;
1006 } else {
1007 ptr = g->sb_hybrid;
1008 }
1009
1010 for (i = g->short_start; i < 13; i++) {
1011 len = band_size_short[s->sample_rate_index][i];
1012 ptr1 = ptr;
1013 dst = tmp;
1014 for (j = len; j > 0; j--) {
1015 *dst++ = ptr[0*len];
1016 *dst++ = ptr[1*len];
1017 *dst++ = ptr[2*len];
1018 ptr++;
1019 }
1020 ptr += 2 * len;
1021 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1022 }
1023 }
1024
1025 #define ISQRT2 FIXR(0.70710678118654752440)
1026
compute_stereo(MPADecodeContext * s,GranuleDef * g0,GranuleDef * g1)1027 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1028 {
1029 int i, j, k, l;
1030 int sf_max, sf, len, non_zero_found;
1031 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1032 int non_zero_found_short[3];
1033
1034 /* intensity stereo */
1035 if (s->mode_ext & MODE_EXT_I_STEREO) {
1036 if (!s->lsf) {
1037 is_tab = is_table;
1038 sf_max = 7;
1039 } else {
1040 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1041 sf_max = 16;
1042 }
1043
1044 tab0 = g0->sb_hybrid + 576;
1045 tab1 = g1->sb_hybrid + 576;
1046
1047 non_zero_found_short[0] = 0;
1048 non_zero_found_short[1] = 0;
1049 non_zero_found_short[2] = 0;
1050 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1051 for (i = 12; i >= g1->short_start; i--) {
1052 /* for last band, use previous scale factor */
1053 if (i != 11)
1054 k -= 3;
1055 len = band_size_short[s->sample_rate_index][i];
1056 for (l = 2; l >= 0; l--) {
1057 tab0 -= len;
1058 tab1 -= len;
1059 if (!non_zero_found_short[l]) {
1060 /* test if non zero band. if so, stop doing i-stereo */
1061 for (j = 0; j < len; j++) {
1062 if (tab1[j] != 0) {
1063 non_zero_found_short[l] = 1;
1064 goto found1;
1065 }
1066 }
1067 sf = g1->scale_factors[k + l];
1068 if (sf >= sf_max)
1069 goto found1;
1070
1071 v1 = is_tab[0][sf];
1072 v2 = is_tab[1][sf];
1073 for (j = 0; j < len; j++) {
1074 tmp0 = tab0[j];
1075 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1076 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1077 }
1078 } else {
1079 found1:
1080 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1081 /* lower part of the spectrum : do ms stereo
1082 if enabled */
1083 for (j = 0; j < len; j++) {
1084 tmp0 = tab0[j];
1085 tmp1 = tab1[j];
1086 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1087 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1088 }
1089 }
1090 }
1091 }
1092 }
1093
1094 non_zero_found = non_zero_found_short[0] |
1095 non_zero_found_short[1] |
1096 non_zero_found_short[2];
1097
1098 for (i = g1->long_end - 1;i >= 0;i--) {
1099 len = band_size_long[s->sample_rate_index][i];
1100 tab0 -= len;
1101 tab1 -= len;
1102 /* test if non zero band. if so, stop doing i-stereo */
1103 if (!non_zero_found) {
1104 for (j = 0; j < len; j++) {
1105 if (tab1[j] != 0) {
1106 non_zero_found = 1;
1107 goto found2;
1108 }
1109 }
1110 /* for last band, use previous scale factor */
1111 k = (i == 21) ? 20 : i;
1112 sf = g1->scale_factors[k];
1113 if (sf >= sf_max)
1114 goto found2;
1115 v1 = is_tab[0][sf];
1116 v2 = is_tab[1][sf];
1117 for (j = 0; j < len; j++) {
1118 tmp0 = tab0[j];
1119 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1120 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1121 }
1122 } else {
1123 found2:
1124 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1125 /* lower part of the spectrum : do ms stereo
1126 if enabled */
1127 for (j = 0; j < len; j++) {
1128 tmp0 = tab0[j];
1129 tmp1 = tab1[j];
1130 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1131 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1132 }
1133 }
1134 }
1135 }
1136 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1137 /* ms stereo ONLY */
1138 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1139 global gain */
1140 #if USE_FLOATS
1141 s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1142 #else
1143 tab0 = g0->sb_hybrid;
1144 tab1 = g1->sb_hybrid;
1145 for (i = 0; i < 576; i++) {
1146 tmp0 = tab0[i];
1147 tmp1 = tab1[i];
1148 tab0[i] = tmp0 + tmp1;
1149 tab1[i] = tmp0 - tmp1;
1150 }
1151 #endif
1152 }
1153 }
1154
1155 #if USE_FLOATS
1156 #if HAVE_MIPSFPU
1157 # include "mips/compute_antialias_float.h"
1158 #endif /* HAVE_MIPSFPU */
1159 #else
1160 #if HAVE_MIPSDSPR1
1161 # include "mips/compute_antialias_fixed.h"
1162 #endif /* HAVE_MIPSDSPR1 */
1163 #endif /* USE_FLOATS */
1164
1165 #ifndef compute_antialias
1166 #if USE_FLOATS
1167 #define AA(j) do { \
1168 float tmp0 = ptr[-1-j]; \
1169 float tmp1 = ptr[ j]; \
1170 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1171 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1172 } while (0)
1173 #else
1174 #define AA(j) do { \
1175 int tmp0 = ptr[-1-j]; \
1176 int tmp1 = ptr[ j]; \
1177 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1178 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1179 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1180 } while (0)
1181 #endif
1182
compute_antialias(MPADecodeContext * s,GranuleDef * g)1183 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1184 {
1185 INTFLOAT *ptr;
1186 int n, i;
1187
1188 /* we antialias only "long" bands */
1189 if (g->block_type == 2) {
1190 if (!g->switch_point)
1191 return;
1192 /* XXX: check this for 8000Hz case */
1193 n = 1;
1194 } else {
1195 n = SBLIMIT - 1;
1196 }
1197
1198 ptr = g->sb_hybrid + 18;
1199 for (i = n; i > 0; i--) {
1200 AA(0);
1201 AA(1);
1202 AA(2);
1203 AA(3);
1204 AA(4);
1205 AA(5);
1206 AA(6);
1207 AA(7);
1208
1209 ptr += 18;
1210 }
1211 }
1212 #endif /* compute_antialias */
1213
compute_imdct(MPADecodeContext * s,GranuleDef * g,INTFLOAT * sb_samples,INTFLOAT * mdct_buf)1214 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1215 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1216 {
1217 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1218 INTFLOAT out2[12];
1219 int i, j, mdct_long_end, sblimit;
1220
1221 /* find last non zero block */
1222 ptr = g->sb_hybrid + 576;
1223 ptr1 = g->sb_hybrid + 2 * 18;
1224 while (ptr >= ptr1) {
1225 int32_t *p;
1226 ptr -= 6;
1227 p = (int32_t*)ptr;
1228 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1229 break;
1230 }
1231 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1232
1233 if (g->block_type == 2) {
1234 /* XXX: check for 8000 Hz */
1235 if (g->switch_point)
1236 mdct_long_end = 2;
1237 else
1238 mdct_long_end = 0;
1239 } else {
1240 mdct_long_end = sblimit;
1241 }
1242
1243 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1244 mdct_long_end, g->switch_point,
1245 g->block_type);
1246
1247 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1248 ptr = g->sb_hybrid + 18 * mdct_long_end;
1249
1250 for (j = mdct_long_end; j < sblimit; j++) {
1251 /* select frequency inversion */
1252 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1253 out_ptr = sb_samples + j;
1254
1255 for (i = 0; i < 6; i++) {
1256 *out_ptr = buf[4*i];
1257 out_ptr += SBLIMIT;
1258 }
1259 imdct12(out2, ptr + 0);
1260 for (i = 0; i < 6; i++) {
1261 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1262 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1263 out_ptr += SBLIMIT;
1264 }
1265 imdct12(out2, ptr + 1);
1266 for (i = 0; i < 6; i++) {
1267 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1268 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1269 out_ptr += SBLIMIT;
1270 }
1271 imdct12(out2, ptr + 2);
1272 for (i = 0; i < 6; i++) {
1273 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1274 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1275 buf[4*(i + 6*2)] = 0;
1276 }
1277 ptr += 18;
1278 buf += (j&3) != 3 ? 1 : (4*18-3);
1279 }
1280 /* zero bands */
1281 for (j = sblimit; j < SBLIMIT; j++) {
1282 /* overlap */
1283 out_ptr = sb_samples + j;
1284 for (i = 0; i < 18; i++) {
1285 *out_ptr = buf[4*i];
1286 buf[4*i] = 0;
1287 out_ptr += SBLIMIT;
1288 }
1289 buf += (j&3) != 3 ? 1 : (4*18-3);
1290 }
1291 }
1292
1293 /* main layer3 decoding function */
mp_decode_layer3(MPADecodeContext * s)1294 static int mp_decode_layer3(MPADecodeContext *s)
1295 {
1296 int nb_granules, main_data_begin;
1297 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1298 GranuleDef *g;
1299 int16_t exponents[576]; //FIXME try INTFLOAT
1300
1301 /* read side info */
1302 if (s->lsf) {
1303 main_data_begin = get_bits(&s->gb, 8);
1304 skip_bits(&s->gb, s->nb_channels);
1305 nb_granules = 1;
1306 } else {
1307 main_data_begin = get_bits(&s->gb, 9);
1308 if (s->nb_channels == 2)
1309 skip_bits(&s->gb, 3);
1310 else
1311 skip_bits(&s->gb, 5);
1312 nb_granules = 2;
1313 for (ch = 0; ch < s->nb_channels; ch++) {
1314 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1315 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1316 }
1317 }
1318
1319 for (gr = 0; gr < nb_granules; gr++) {
1320 for (ch = 0; ch < s->nb_channels; ch++) {
1321 av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1322 g = &s->granules[ch][gr];
1323 g->part2_3_length = get_bits(&s->gb, 12);
1324 g->big_values = get_bits(&s->gb, 9);
1325 if (g->big_values > 288) {
1326 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1327 return AVERROR_INVALIDDATA;
1328 }
1329
1330 g->global_gain = get_bits(&s->gb, 8);
1331 /* if MS stereo only is selected, we precompute the
1332 1/sqrt(2) renormalization factor */
1333 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1334 MODE_EXT_MS_STEREO)
1335 g->global_gain -= 2;
1336 if (s->lsf)
1337 g->scalefac_compress = get_bits(&s->gb, 9);
1338 else
1339 g->scalefac_compress = get_bits(&s->gb, 4);
1340 blocksplit_flag = get_bits1(&s->gb);
1341 if (blocksplit_flag) {
1342 g->block_type = get_bits(&s->gb, 2);
1343 if (g->block_type == 0) {
1344 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1345 return AVERROR_INVALIDDATA;
1346 }
1347 g->switch_point = get_bits1(&s->gb);
1348 for (i = 0; i < 2; i++)
1349 g->table_select[i] = get_bits(&s->gb, 5);
1350 for (i = 0; i < 3; i++)
1351 g->subblock_gain[i] = get_bits(&s->gb, 3);
1352 init_short_region(s, g);
1353 } else {
1354 int region_address1, region_address2;
1355 g->block_type = 0;
1356 g->switch_point = 0;
1357 for (i = 0; i < 3; i++)
1358 g->table_select[i] = get_bits(&s->gb, 5);
1359 /* compute huffman coded region sizes */
1360 region_address1 = get_bits(&s->gb, 4);
1361 region_address2 = get_bits(&s->gb, 3);
1362 av_dlog(s->avctx, "region1=%d region2=%d\n",
1363 region_address1, region_address2);
1364 init_long_region(s, g, region_address1, region_address2);
1365 }
1366 region_offset2size(g);
1367 compute_band_indexes(s, g);
1368
1369 g->preflag = 0;
1370 if (!s->lsf)
1371 g->preflag = get_bits1(&s->gb);
1372 g->scalefac_scale = get_bits1(&s->gb);
1373 g->count1table_select = get_bits1(&s->gb);
1374 av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1375 g->block_type, g->switch_point);
1376 }
1377 }
1378
1379 if (!s->adu_mode) {
1380 int skip;
1381 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1382 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0, EXTRABYTES);
1383 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1384 /* now we get bits from the main_data_begin offset */
1385 av_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1386 main_data_begin, s->last_buf_size);
1387
1388 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1389 s->in_gb = s->gb;
1390 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1391 #if !UNCHECKED_BITSTREAM_READER
1392 s->gb.size_in_bits_plus8 += FFMAX(extrasize, LAST_BUF_SIZE - s->last_buf_size) * 8;
1393 #endif
1394 s->last_buf_size <<= 3;
1395 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1396 for (ch = 0; ch < s->nb_channels; ch++) {
1397 g = &s->granules[ch][gr];
1398 s->last_buf_size += g->part2_3_length;
1399 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1400 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1401 }
1402 }
1403 skip = s->last_buf_size - 8 * main_data_begin;
1404 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1405 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1406 s->gb = s->in_gb;
1407 s->in_gb.buffer = NULL;
1408 } else {
1409 skip_bits_long(&s->gb, skip);
1410 }
1411 } else {
1412 gr = 0;
1413 }
1414
1415 for (; gr < nb_granules; gr++) {
1416 for (ch = 0; ch < s->nb_channels; ch++) {
1417 g = &s->granules[ch][gr];
1418 bits_pos = get_bits_count(&s->gb);
1419
1420 if (!s->lsf) {
1421 uint8_t *sc;
1422 int slen, slen1, slen2;
1423
1424 /* MPEG1 scale factors */
1425 slen1 = slen_table[0][g->scalefac_compress];
1426 slen2 = slen_table[1][g->scalefac_compress];
1427 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1428 if (g->block_type == 2) {
1429 n = g->switch_point ? 17 : 18;
1430 j = 0;
1431 if (slen1) {
1432 for (i = 0; i < n; i++)
1433 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1434 } else {
1435 for (i = 0; i < n; i++)
1436 g->scale_factors[j++] = 0;
1437 }
1438 if (slen2) {
1439 for (i = 0; i < 18; i++)
1440 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1441 for (i = 0; i < 3; i++)
1442 g->scale_factors[j++] = 0;
1443 } else {
1444 for (i = 0; i < 21; i++)
1445 g->scale_factors[j++] = 0;
1446 }
1447 } else {
1448 sc = s->granules[ch][0].scale_factors;
1449 j = 0;
1450 for (k = 0; k < 4; k++) {
1451 n = k == 0 ? 6 : 5;
1452 if ((g->scfsi & (0x8 >> k)) == 0) {
1453 slen = (k < 2) ? slen1 : slen2;
1454 if (slen) {
1455 for (i = 0; i < n; i++)
1456 g->scale_factors[j++] = get_bits(&s->gb, slen);
1457 } else {
1458 for (i = 0; i < n; i++)
1459 g->scale_factors[j++] = 0;
1460 }
1461 } else {
1462 /* simply copy from last granule */
1463 for (i = 0; i < n; i++) {
1464 g->scale_factors[j] = sc[j];
1465 j++;
1466 }
1467 }
1468 }
1469 g->scale_factors[j++] = 0;
1470 }
1471 } else {
1472 int tindex, tindex2, slen[4], sl, sf;
1473
1474 /* LSF scale factors */
1475 if (g->block_type == 2)
1476 tindex = g->switch_point ? 2 : 1;
1477 else
1478 tindex = 0;
1479
1480 sf = g->scalefac_compress;
1481 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1482 /* intensity stereo case */
1483 sf >>= 1;
1484 if (sf < 180) {
1485 lsf_sf_expand(slen, sf, 6, 6, 0);
1486 tindex2 = 3;
1487 } else if (sf < 244) {
1488 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1489 tindex2 = 4;
1490 } else {
1491 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1492 tindex2 = 5;
1493 }
1494 } else {
1495 /* normal case */
1496 if (sf < 400) {
1497 lsf_sf_expand(slen, sf, 5, 4, 4);
1498 tindex2 = 0;
1499 } else if (sf < 500) {
1500 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1501 tindex2 = 1;
1502 } else {
1503 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1504 tindex2 = 2;
1505 g->preflag = 1;
1506 }
1507 }
1508
1509 j = 0;
1510 for (k = 0; k < 4; k++) {
1511 n = lsf_nsf_table[tindex2][tindex][k];
1512 sl = slen[k];
1513 if (sl) {
1514 for (i = 0; i < n; i++)
1515 g->scale_factors[j++] = get_bits(&s->gb, sl);
1516 } else {
1517 for (i = 0; i < n; i++)
1518 g->scale_factors[j++] = 0;
1519 }
1520 }
1521 /* XXX: should compute exact size */
1522 for (; j < 40; j++)
1523 g->scale_factors[j] = 0;
1524 }
1525
1526 exponents_from_scale_factors(s, g, exponents);
1527
1528 /* read Huffman coded residue */
1529 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1530 } /* ch */
1531
1532 if (s->mode == MPA_JSTEREO)
1533 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1534
1535 for (ch = 0; ch < s->nb_channels; ch++) {
1536 g = &s->granules[ch][gr];
1537
1538 reorder_block(s, g);
1539 compute_antialias(s, g);
1540 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1541 }
1542 } /* gr */
1543 if (get_bits_count(&s->gb) < 0)
1544 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1545 return nb_granules * 18;
1546 }
1547
mp_decode_frame(MPADecodeContext * s,OUT_INT ** samples,const uint8_t * buf,int buf_size)1548 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1549 const uint8_t *buf, int buf_size)
1550 {
1551 int i, nb_frames, ch, ret;
1552 OUT_INT *samples_ptr;
1553
1554 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1555
1556 /* skip error protection field */
1557 if (s->error_protection)
1558 skip_bits(&s->gb, 16);
1559
1560 switch(s->layer) {
1561 case 1:
1562 s->avctx->frame_size = 384;
1563 nb_frames = mp_decode_layer1(s);
1564 break;
1565 case 2:
1566 s->avctx->frame_size = 1152;
1567 nb_frames = mp_decode_layer2(s);
1568 break;
1569 case 3:
1570 s->avctx->frame_size = s->lsf ? 576 : 1152;
1571 default:
1572 nb_frames = mp_decode_layer3(s);
1573
1574 s->last_buf_size=0;
1575 if (s->in_gb.buffer) {
1576 align_get_bits(&s->gb);
1577 i = get_bits_left(&s->gb)>>3;
1578 if (i >= 0 && i <= BACKSTEP_SIZE) {
1579 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1580 s->last_buf_size=i;
1581 } else
1582 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1583 s->gb = s->in_gb;
1584 s->in_gb.buffer = NULL;
1585 }
1586
1587 align_get_bits(&s->gb);
1588 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1589 i = get_bits_left(&s->gb) >> 3;
1590
1591 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1592 if (i < 0)
1593 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1594 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1595 }
1596 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1597 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1598 s->last_buf_size += i;
1599 }
1600
1601 if(nb_frames < 0)
1602 return nb_frames;
1603
1604 /* get output buffer */
1605 if (!samples) {
1606 av_assert0(s->frame);
1607 s->frame->nb_samples = s->avctx->frame_size;
1608 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)
1609 return ret;
1610 samples = (OUT_INT **)s->frame->extended_data;
1611 }
1612
1613 /* apply the synthesis filter */
1614 for (ch = 0; ch < s->nb_channels; ch++) {
1615 int sample_stride;
1616 if (s->avctx->sample_fmt == OUT_FMT_P) {
1617 samples_ptr = samples[ch];
1618 sample_stride = 1;
1619 } else {
1620 samples_ptr = samples[0] + ch;
1621 sample_stride = s->nb_channels;
1622 }
1623 for (i = 0; i < nb_frames; i++) {
1624 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1625 &(s->synth_buf_offset[ch]),
1626 RENAME(ff_mpa_synth_window),
1627 &s->dither_state, samples_ptr,
1628 sample_stride, s->sb_samples[ch][i]);
1629 samples_ptr += 32 * sample_stride;
1630 }
1631 }
1632
1633 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1634 }
1635
decode_frame(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)1636 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1637 AVPacket *avpkt)
1638 {
1639 const uint8_t *buf = avpkt->data;
1640 int buf_size = avpkt->size;
1641 MPADecodeContext *s = avctx->priv_data;
1642 uint32_t header;
1643 int ret;
1644
1645 while(buf_size && !*buf){
1646 buf++;
1647 buf_size--;
1648 }
1649
1650 if (buf_size < HEADER_SIZE)
1651 return AVERROR_INVALIDDATA;
1652
1653 header = AV_RB32(buf);
1654 if (header>>8 == AV_RB32("TAG")>>8) {
1655 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1656 return buf_size;
1657 }
1658 if (ff_mpa_check_header(header) < 0) {
1659 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1660 return AVERROR_INVALIDDATA;
1661 }
1662
1663 if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
1664 /* free format: prepare to compute frame size */
1665 s->frame_size = -1;
1666 return AVERROR_INVALIDDATA;
1667 }
1668 /* update codec info */
1669 avctx->channels = s->nb_channels;
1670 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1671 if (!avctx->bit_rate)
1672 avctx->bit_rate = s->bit_rate;
1673
1674 if (s->frame_size <= 0 || s->frame_size > buf_size) {
1675 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1676 return AVERROR_INVALIDDATA;
1677 } else if (s->frame_size < buf_size) {
1678 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1679 buf_size= s->frame_size;
1680 }
1681
1682 s->frame = data;
1683
1684 ret = mp_decode_frame(s, NULL, buf, buf_size);
1685 if (ret >= 0) {
1686 s->frame->nb_samples = avctx->frame_size;
1687 *got_frame_ptr = 1;
1688 avctx->sample_rate = s->sample_rate;
1689 //FIXME maybe move the other codec info stuff from above here too
1690 } else {
1691 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1692 /* Only return an error if the bad frame makes up the whole packet or
1693 * the error is related to buffer management.
1694 * If there is more data in the packet, just consume the bad frame
1695 * instead of returning an error, which would discard the whole
1696 * packet. */
1697 *got_frame_ptr = 0;
1698 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1699 return ret;
1700 }
1701 s->frame_size = 0;
1702 return buf_size;
1703 }
1704
mp_flush(MPADecodeContext * ctx)1705 static void mp_flush(MPADecodeContext *ctx)
1706 {
1707 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1708 memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));
1709 ctx->last_buf_size = 0;
1710 ctx->dither_state = 0;
1711 }
1712
flush(AVCodecContext * avctx)1713 static void flush(AVCodecContext *avctx)
1714 {
1715 mp_flush(avctx->priv_data);
1716 }
1717
1718 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
decode_frame_adu(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)1719 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1720 int *got_frame_ptr, AVPacket *avpkt)
1721 {
1722 const uint8_t *buf = avpkt->data;
1723 int buf_size = avpkt->size;
1724 MPADecodeContext *s = avctx->priv_data;
1725 uint32_t header;
1726 int len, ret;
1727 int av_unused out_size;
1728
1729 len = buf_size;
1730
1731 // Discard too short frames
1732 if (buf_size < HEADER_SIZE) {
1733 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1734 return AVERROR_INVALIDDATA;
1735 }
1736
1737
1738 if (len > MPA_MAX_CODED_FRAME_SIZE)
1739 len = MPA_MAX_CODED_FRAME_SIZE;
1740
1741 // Get header and restore sync word
1742 header = AV_RB32(buf) | 0xffe00000;
1743
1744 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1745 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1746 return AVERROR_INVALIDDATA;
1747 }
1748
1749 avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1750 /* update codec info */
1751 avctx->sample_rate = s->sample_rate;
1752 avctx->channels = s->nb_channels;
1753 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1754 if (!avctx->bit_rate)
1755 avctx->bit_rate = s->bit_rate;
1756
1757 s->frame_size = len;
1758
1759 s->frame = data;
1760
1761 ret = mp_decode_frame(s, NULL, buf, buf_size);
1762 if (ret < 0) {
1763 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1764 return ret;
1765 }
1766
1767 *got_frame_ptr = 1;
1768
1769 return buf_size;
1770 }
1771 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1772
1773 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1774
1775 /**
1776 * Context for MP3On4 decoder
1777 */
1778 typedef struct MP3On4DecodeContext {
1779 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1780 int syncword; ///< syncword patch
1781 const uint8_t *coff; ///< channel offsets in output buffer
1782 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1783 } MP3On4DecodeContext;
1784
1785 #include "mpeg4audio.h"
1786
1787 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1788
1789 /* number of mp3 decoder instances */
1790 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1791
1792 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1793 static const uint8_t chan_offset[8][5] = {
1794 { 0 },
1795 { 0 }, // C
1796 { 0 }, // FLR
1797 { 2, 0 }, // C FLR
1798 { 2, 0, 3 }, // C FLR BS
1799 { 2, 0, 3 }, // C FLR BLRS
1800 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1801 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1802 };
1803
1804 /* mp3on4 channel layouts */
1805 static const int16_t chan_layout[8] = {
1806 0,
1807 AV_CH_LAYOUT_MONO,
1808 AV_CH_LAYOUT_STEREO,
1809 AV_CH_LAYOUT_SURROUND,
1810 AV_CH_LAYOUT_4POINT0,
1811 AV_CH_LAYOUT_5POINT0,
1812 AV_CH_LAYOUT_5POINT1,
1813 AV_CH_LAYOUT_7POINT1
1814 };
1815
decode_close_mp3on4(AVCodecContext * avctx)1816 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1817 {
1818 MP3On4DecodeContext *s = avctx->priv_data;
1819 int i;
1820
1821 for (i = 0; i < s->frames; i++)
1822 av_free(s->mp3decctx[i]);
1823
1824 return 0;
1825 }
1826
1827
decode_init_mp3on4(AVCodecContext * avctx)1828 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1829 {
1830 MP3On4DecodeContext *s = avctx->priv_data;
1831 MPEG4AudioConfig cfg;
1832 int i;
1833
1834 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1835 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1836 return AVERROR_INVALIDDATA;
1837 }
1838
1839 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1840 avctx->extradata_size * 8, 1);
1841 if (!cfg.chan_config || cfg.chan_config > 7) {
1842 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1843 return AVERROR_INVALIDDATA;
1844 }
1845 s->frames = mp3Frames[cfg.chan_config];
1846 s->coff = chan_offset[cfg.chan_config];
1847 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1848 avctx->channel_layout = chan_layout[cfg.chan_config];
1849
1850 if (cfg.sample_rate < 16000)
1851 s->syncword = 0xffe00000;
1852 else
1853 s->syncword = 0xfff00000;
1854
1855 /* Init the first mp3 decoder in standard way, so that all tables get builded
1856 * We replace avctx->priv_data with the context of the first decoder so that
1857 * decode_init() does not have to be changed.
1858 * Other decoders will be initialized here copying data from the first context
1859 */
1860 // Allocate zeroed memory for the first decoder context
1861 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1862 if (!s->mp3decctx[0])
1863 goto alloc_fail;
1864 // Put decoder context in place to make init_decode() happy
1865 avctx->priv_data = s->mp3decctx[0];
1866 decode_init(avctx);
1867 // Restore mp3on4 context pointer
1868 avctx->priv_data = s;
1869 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1870
1871 /* Create a separate codec/context for each frame (first is already ok).
1872 * Each frame is 1 or 2 channels - up to 5 frames allowed
1873 */
1874 for (i = 1; i < s->frames; i++) {
1875 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1876 if (!s->mp3decctx[i])
1877 goto alloc_fail;
1878 s->mp3decctx[i]->adu_mode = 1;
1879 s->mp3decctx[i]->avctx = avctx;
1880 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1881 }
1882
1883 return 0;
1884 alloc_fail:
1885 decode_close_mp3on4(avctx);
1886 return AVERROR(ENOMEM);
1887 }
1888
1889
flush_mp3on4(AVCodecContext * avctx)1890 static void flush_mp3on4(AVCodecContext *avctx)
1891 {
1892 int i;
1893 MP3On4DecodeContext *s = avctx->priv_data;
1894
1895 for (i = 0; i < s->frames; i++)
1896 mp_flush(s->mp3decctx[i]);
1897 }
1898
1899
decode_frame_mp3on4(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)1900 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1901 int *got_frame_ptr, AVPacket *avpkt)
1902 {
1903 AVFrame *frame = data;
1904 const uint8_t *buf = avpkt->data;
1905 int buf_size = avpkt->size;
1906 MP3On4DecodeContext *s = avctx->priv_data;
1907 MPADecodeContext *m;
1908 int fsize, len = buf_size, out_size = 0;
1909 uint32_t header;
1910 OUT_INT **out_samples;
1911 OUT_INT *outptr[2];
1912 int fr, ch, ret;
1913
1914 /* get output buffer */
1915 frame->nb_samples = MPA_FRAME_SIZE;
1916 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1917 return ret;
1918 out_samples = (OUT_INT **)frame->extended_data;
1919
1920 // Discard too short frames
1921 if (buf_size < HEADER_SIZE)
1922 return AVERROR_INVALIDDATA;
1923
1924 avctx->bit_rate = 0;
1925
1926 ch = 0;
1927 for (fr = 0; fr < s->frames; fr++) {
1928 fsize = AV_RB16(buf) >> 4;
1929 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1930 m = s->mp3decctx[fr];
1931 av_assert1(m);
1932
1933 if (fsize < HEADER_SIZE) {
1934 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1935 return AVERROR_INVALIDDATA;
1936 }
1937 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1938
1939 if (ff_mpa_check_header(header) < 0) {
1940 av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n");
1941 return AVERROR_INVALIDDATA;
1942 }
1943
1944 avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1945
1946 if (ch + m->nb_channels > avctx->channels ||
1947 s->coff[fr] + m->nb_channels > avctx->channels) {
1948 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1949 "channel count\n");
1950 return AVERROR_INVALIDDATA;
1951 }
1952 ch += m->nb_channels;
1953
1954 outptr[0] = out_samples[s->coff[fr]];
1955 if (m->nb_channels > 1)
1956 outptr[1] = out_samples[s->coff[fr] + 1];
1957
1958 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) {
1959 av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch);
1960 memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1961 if (m->nb_channels > 1)
1962 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1963 ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT);
1964 }
1965
1966 out_size += ret;
1967 buf += fsize;
1968 len -= fsize;
1969
1970 avctx->bit_rate += m->bit_rate;
1971 }
1972 if (ch != avctx->channels) {
1973 av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n");
1974 return AVERROR_INVALIDDATA;
1975 }
1976
1977 /* update codec info */
1978 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1979
1980 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1981 *got_frame_ptr = 1;
1982
1983 return buf_size;
1984 }
1985 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */
1986