mpegaudiodec_template.c (64046B)
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 "config_components.h" 28 29 #include "libavutil/attributes.h" 30 #include "libavutil/avassert.h" 31 #include "libavutil/channel_layout.h" 32 #include "libavutil/crc.h" 33 #include "libavutil/float_dsp.h" 34 #include "libavutil/libm.h" 35 #include "libavutil/mem.h" 36 #include "libavutil/mem_internal.h" 37 #include "libavutil/thread.h" 38 39 #include "avcodec.h" 40 #include "decode.h" 41 #include "get_bits.h" 42 #include "mathops.h" 43 #include "mpegaudiodsp.h" 44 45 /* 46 * TODO: 47 * - test lsf / mpeg25 extensively. 48 */ 49 50 #include "mpegaudio.h" 51 #include "mpegaudiodecheader.h" 52 53 #define BACKSTEP_SIZE 512 54 #define EXTRABYTES 24 55 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES 56 57 /* layer 3 "granule" */ 58 typedef struct GranuleDef { 59 uint8_t scfsi; 60 int part2_3_length; 61 int big_values; 62 int global_gain; 63 int scalefac_compress; 64 uint8_t block_type; 65 uint8_t switch_point; 66 int table_select[3]; 67 int subblock_gain[3]; 68 uint8_t scalefac_scale; 69 uint8_t count1table_select; 70 int region_size[3]; /* number of huffman codes in each region */ 71 int preflag; 72 int short_start, long_end; /* long/short band indexes */ 73 uint8_t scale_factors[40]; 74 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */ 75 } GranuleDef; 76 77 typedef struct MPADecodeContext { 78 MPA_DECODE_HEADER 79 uint8_t last_buf[LAST_BUF_SIZE]; 80 int last_buf_size; 81 int extrasize; 82 /* next header (used in free format parsing) */ 83 uint32_t free_format_next_header; 84 GetBitContext gb; 85 GetBitContext in_gb; 86 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2]; 87 int synth_buf_offset[MPA_MAX_CHANNELS]; 88 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT]; 89 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */ 90 GranuleDef granules[2][2]; /* Used in Layer 3 */ 91 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3 92 int dither_state; 93 int err_recognition; 94 AVCodecContext* avctx; 95 MPADSPContext mpadsp; 96 void (*butterflies_float)(float *restrict v1, float *restrict v2, int len); 97 AVFrame *frame; 98 uint32_t crc; 99 } MPADecodeContext; 100 101 #define HEADER_SIZE 4 102 103 #include "mpegaudiodata.h" 104 105 #include "mpegaudio_tablegen.h" 106 /* intensity stereo coef table */ 107 static INTFLOAT is_table_lsf[2][2][16]; 108 109 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */ 110 static int32_t scale_factor_mult[15][3]; 111 /* mult table for layer 2 group quantization */ 112 113 #define SCALE_GEN(v) \ 114 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) } 115 116 static const int32_t scale_factor_mult2[3][3] = { 117 SCALE_GEN(4.0 / 3.0), /* 3 steps */ 118 SCALE_GEN(4.0 / 5.0), /* 5 steps */ 119 SCALE_GEN(4.0 / 9.0), /* 9 steps */ 120 }; 121 122 /** 123 * Convert region offsets to region sizes and truncate 124 * size to big_values. 125 */ 126 static void region_offset2size(GranuleDef *g) 127 { 128 int i, k, j = 0; 129 g->region_size[2] = 576 / 2; 130 for (i = 0; i < 3; i++) { 131 k = FFMIN(g->region_size[i], g->big_values); 132 g->region_size[i] = k - j; 133 j = k; 134 } 135 } 136 137 static void init_short_region(MPADecodeContext *s, GranuleDef *g) 138 { 139 if (g->block_type == 2) { 140 if (s->sample_rate_index != 8) 141 g->region_size[0] = (36 / 2); 142 else 143 g->region_size[0] = (72 / 2); 144 } else { 145 if (s->sample_rate_index <= 2) 146 g->region_size[0] = (36 / 2); 147 else if (s->sample_rate_index != 8) 148 g->region_size[0] = (54 / 2); 149 else 150 g->region_size[0] = (108 / 2); 151 } 152 g->region_size[1] = (576 / 2); 153 } 154 155 static void init_long_region(MPADecodeContext *s, GranuleDef *g, 156 int ra1, int ra2) 157 { 158 int l; 159 g->region_size[0] = ff_band_index_long[s->sample_rate_index][ra1 + 1]; 160 /* should not overflow */ 161 l = FFMIN(ra1 + ra2 + 2, 22); 162 g->region_size[1] = ff_band_index_long[s->sample_rate_index][ l]; 163 } 164 165 static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g) 166 { 167 if (g->block_type == 2) { 168 if (g->switch_point) { 169 if(s->sample_rate_index == 8) 170 avpriv_request_sample(s->avctx, "switch point in 8khz"); 171 /* if switched mode, we handle the 36 first samples as 172 long blocks. For 8000Hz, we handle the 72 first 173 exponents as long blocks */ 174 if (s->sample_rate_index <= 2) 175 g->long_end = 8; 176 else 177 g->long_end = 6; 178 179 g->short_start = 3; 180 } else { 181 g->long_end = 0; 182 g->short_start = 0; 183 } 184 } else { 185 g->short_start = 13; 186 g->long_end = 22; 187 } 188 } 189 190 /* layer 1 unscaling */ 191 /* n = number of bits of the mantissa minus 1 */ 192 static inline int l1_unscale(int n, int mant, int scale_factor) 193 { 194 int shift, mod; 195 int64_t val; 196 197 shift = ff_scale_factor_modshift[scale_factor]; 198 mod = shift & 3; 199 shift >>= 2; 200 val = MUL64((int)(mant + (-1U << n) + 1), scale_factor_mult[n-1][mod]); 201 shift += n; 202 /* NOTE: at this point, 1 <= shift >= 21 + 15 */ 203 return (int)((val + (1LL << (shift - 1))) >> shift); 204 } 205 206 static inline int l2_unscale_group(int steps, int mant, int scale_factor) 207 { 208 int shift, mod, val; 209 210 shift = ff_scale_factor_modshift[scale_factor]; 211 mod = shift & 3; 212 shift >>= 2; 213 214 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod]; 215 /* NOTE: at this point, 0 <= shift <= 21 */ 216 if (shift > 0) 217 val = (val + (1 << (shift - 1))) >> shift; 218 return val; 219 } 220 221 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */ 222 static inline int l3_unscale(int value, int exponent) 223 { 224 unsigned int m; 225 int e; 226 227 e = ff_table_4_3_exp [4 * value + (exponent & 3)]; 228 m = ff_table_4_3_value[4 * value + (exponent & 3)]; 229 e -= exponent >> 2; 230 #ifdef DEBUG 231 if(e < 1) 232 av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e); 233 #endif 234 if (e > (SUINT)31) 235 return 0; 236 m = (m + ((1U << e) >> 1)) >> e; 237 238 return m; 239 } 240 241 static av_cold void decode_init_static(void) 242 { 243 int i, j; 244 245 /* scale factor multiply for layer 1 */ 246 for (i = 0; i < 15; i++) { 247 int n, norm; 248 n = i + 2; 249 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1); 250 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS); 251 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS); 252 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS); 253 ff_dlog(NULL, "%d: norm=%x s=%"PRIx32" %"PRIx32" %"PRIx32"\n", i, 254 (unsigned)norm, 255 scale_factor_mult[i][0], 256 scale_factor_mult[i][1], 257 scale_factor_mult[i][2]); 258 } 259 260 /* compute n ^ (4/3) and store it in mantissa/exp format */ 261 262 mpegaudio_tableinit(); 263 264 for (i = 0; i < 16; i++) { 265 double f; 266 int e, k; 267 268 for (j = 0; j < 2; j++) { 269 e = -(j + 1) * ((i + 1) >> 1); 270 f = exp2(e / 4.0); 271 k = i & 1; 272 is_table_lsf[j][k ^ 1][i] = FIXR(f); 273 is_table_lsf[j][k ][i] = FIXR(1.0); 274 ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n", 275 i, j, (float) is_table_lsf[j][0][i], 276 (float) is_table_lsf[j][1][i]); 277 } 278 } 279 RENAME(ff_mpa_synth_init)(); 280 ff_mpegaudiodec_common_init_static(); 281 } 282 283 static av_cold int decode_init(AVCodecContext * avctx) 284 { 285 static AVOnce init_static_once = AV_ONCE_INIT; 286 MPADecodeContext *s = avctx->priv_data; 287 288 s->avctx = avctx; 289 290 #if USE_FLOATS 291 { 292 AVFloatDSPContext *fdsp; 293 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT); 294 if (!fdsp) 295 return AVERROR(ENOMEM); 296 s->butterflies_float = fdsp->butterflies_float; 297 av_free(fdsp); 298 } 299 #endif 300 301 ff_mpadsp_init(&s->mpadsp); 302 303 if (avctx->request_sample_fmt == OUT_FMT && 304 avctx->codec_id != AV_CODEC_ID_MP3ON4) 305 avctx->sample_fmt = OUT_FMT; 306 else 307 avctx->sample_fmt = OUT_FMT_P; 308 s->err_recognition = avctx->err_recognition; 309 310 if (avctx->codec_id == AV_CODEC_ID_MP3ADU) 311 s->adu_mode = 1; 312 313 ff_thread_once(&init_static_once, decode_init_static); 314 315 return 0; 316 } 317 318 #define C3 FIXHR(0.86602540378443864676/2) 319 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36) 320 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36) 321 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36) 322 323 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious 324 cases. */ 325 static void imdct12(INTFLOAT *out, SUINTFLOAT *in) 326 { 327 SUINTFLOAT in0, in1, in2, in3, in4, in5, t1, t2; 328 329 in0 = in[0*3]; 330 in1 = in[1*3] + in[0*3]; 331 in2 = in[2*3] + in[1*3]; 332 in3 = in[3*3] + in[2*3]; 333 in4 = in[4*3] + in[3*3]; 334 in5 = in[5*3] + in[4*3]; 335 in5 += in3; 336 in3 += in1; 337 338 in2 = MULH3(in2, C3, 2); 339 in3 = MULH3(in3, C3, 4); 340 341 t1 = in0 - in4; 342 t2 = MULH3(in1 - in5, C4, 2); 343 344 out[ 7] = 345 out[10] = t1 + t2; 346 out[ 1] = 347 out[ 4] = t1 - t2; 348 349 in0 += SHR(in4, 1); 350 in4 = in0 + in2; 351 in5 += 2*in1; 352 in1 = MULH3(in5 + in3, C5, 1); 353 out[ 8] = 354 out[ 9] = in4 + in1; 355 out[ 2] = 356 out[ 3] = in4 - in1; 357 358 in0 -= in2; 359 in5 = MULH3(in5 - in3, C6, 2); 360 out[ 0] = 361 out[ 5] = in0 - in5; 362 out[ 6] = 363 out[11] = in0 + in5; 364 } 365 366 static int handle_crc(MPADecodeContext *s, int sec_len) 367 { 368 if (s->error_protection && (s->err_recognition & AV_EF_CRCCHECK)) { 369 const uint8_t *buf = s->gb.buffer - HEADER_SIZE; 370 int sec_byte_len = sec_len >> 3; 371 int sec_rem_bits = sec_len & 7; 372 const AVCRC *crc_tab = av_crc_get_table(AV_CRC_16_ANSI); 373 uint8_t tmp_buf[4]; 374 uint32_t crc_val = av_crc(crc_tab, UINT16_MAX, &buf[2], 2); 375 crc_val = av_crc(crc_tab, crc_val, &buf[6], sec_byte_len); 376 377 AV_WB32(tmp_buf, 378 ((buf[6 + sec_byte_len] & (0xFF00U >> sec_rem_bits)) << 24) + 379 ((s->crc << 16) >> sec_rem_bits)); 380 381 crc_val = av_crc(crc_tab, crc_val, tmp_buf, 3); 382 383 if (crc_val) { 384 av_log(s->avctx, AV_LOG_ERROR, "CRC mismatch %X!\n", crc_val); 385 if (s->err_recognition & AV_EF_EXPLODE) 386 return AVERROR_INVALIDDATA; 387 } 388 } 389 return 0; 390 } 391 392 /* return the number of decoded frames */ 393 static int mp_decode_layer1(MPADecodeContext *s) 394 { 395 int bound, i, v, n, ch, j, mant; 396 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT]; 397 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT]; 398 int ret; 399 400 ret = handle_crc(s, (s->nb_channels == 1) ? 8*16 : 8*32); 401 if (ret < 0) 402 return ret; 403 404 if (s->mode == MPA_JSTEREO) 405 bound = (s->mode_ext + 1) * 4; 406 else 407 bound = SBLIMIT; 408 409 /* allocation bits */ 410 for (i = 0; i < bound; i++) { 411 for (ch = 0; ch < s->nb_channels; ch++) { 412 allocation[ch][i] = get_bits(&s->gb, 4); 413 } 414 } 415 for (i = bound; i < SBLIMIT; i++) 416 allocation[0][i] = get_bits(&s->gb, 4); 417 418 /* scale factors */ 419 for (i = 0; i < bound; i++) { 420 for (ch = 0; ch < s->nb_channels; ch++) { 421 if (allocation[ch][i]) 422 scale_factors[ch][i] = get_bits(&s->gb, 6); 423 } 424 } 425 for (i = bound; i < SBLIMIT; i++) { 426 if (allocation[0][i]) { 427 scale_factors[0][i] = get_bits(&s->gb, 6); 428 scale_factors[1][i] = get_bits(&s->gb, 6); 429 } 430 } 431 432 /* compute samples */ 433 for (j = 0; j < 12; j++) { 434 for (i = 0; i < bound; i++) { 435 for (ch = 0; ch < s->nb_channels; ch++) { 436 n = allocation[ch][i]; 437 if (n) { 438 mant = get_bits(&s->gb, n + 1); 439 v = l1_unscale(n, mant, scale_factors[ch][i]); 440 } else { 441 v = 0; 442 } 443 s->sb_samples[ch][j][i] = v; 444 } 445 } 446 for (i = bound; i < SBLIMIT; i++) { 447 n = allocation[0][i]; 448 if (n) { 449 mant = get_bits(&s->gb, n + 1); 450 v = l1_unscale(n, mant, scale_factors[0][i]); 451 s->sb_samples[0][j][i] = v; 452 v = l1_unscale(n, mant, scale_factors[1][i]); 453 s->sb_samples[1][j][i] = v; 454 } else { 455 s->sb_samples[0][j][i] = 0; 456 s->sb_samples[1][j][i] = 0; 457 } 458 } 459 } 460 return 12; 461 } 462 463 static int mp_decode_layer2(MPADecodeContext *s) 464 { 465 int sblimit; /* number of used subbands */ 466 const unsigned char *alloc_table; 467 int table, bit_alloc_bits, i, j, ch, bound, v; 468 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; 469 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; 470 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf; 471 int scale, qindex, bits, steps, k, l, m, b; 472 int ret; 473 474 /* select decoding table */ 475 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels, 476 s->sample_rate, s->lsf); 477 sblimit = ff_mpa_sblimit_table[table]; 478 alloc_table = ff_mpa_alloc_tables[table]; 479 480 if (s->mode == MPA_JSTEREO) 481 bound = (s->mode_ext + 1) * 4; 482 else 483 bound = sblimit; 484 485 ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit); 486 487 /* sanity check */ 488 if (bound > sblimit) 489 bound = sblimit; 490 491 /* parse bit allocation */ 492 j = 0; 493 for (i = 0; i < bound; i++) { 494 bit_alloc_bits = alloc_table[j]; 495 for (ch = 0; ch < s->nb_channels; ch++) 496 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits); 497 j += 1 << bit_alloc_bits; 498 } 499 for (i = bound; i < sblimit; i++) { 500 bit_alloc_bits = alloc_table[j]; 501 v = get_bits(&s->gb, bit_alloc_bits); 502 bit_alloc[0][i] = v; 503 bit_alloc[1][i] = v; 504 j += 1 << bit_alloc_bits; 505 } 506 507 /* scale codes */ 508 for (i = 0; i < sblimit; i++) { 509 for (ch = 0; ch < s->nb_channels; ch++) { 510 if (bit_alloc[ch][i]) 511 scale_code[ch][i] = get_bits(&s->gb, 2); 512 } 513 } 514 515 ret = handle_crc(s, get_bits_count(&s->gb) - 16); 516 if (ret < 0) 517 return ret; 518 519 /* scale factors */ 520 for (i = 0; i < sblimit; i++) { 521 for (ch = 0; ch < s->nb_channels; ch++) { 522 if (bit_alloc[ch][i]) { 523 sf = scale_factors[ch][i]; 524 switch (scale_code[ch][i]) { 525 default: 526 case 0: 527 sf[0] = get_bits(&s->gb, 6); 528 sf[1] = get_bits(&s->gb, 6); 529 sf[2] = get_bits(&s->gb, 6); 530 break; 531 case 2: 532 sf[0] = get_bits(&s->gb, 6); 533 sf[1] = sf[0]; 534 sf[2] = sf[0]; 535 break; 536 case 1: 537 sf[0] = get_bits(&s->gb, 6); 538 sf[2] = get_bits(&s->gb, 6); 539 sf[1] = sf[0]; 540 break; 541 case 3: 542 sf[0] = get_bits(&s->gb, 6); 543 sf[2] = get_bits(&s->gb, 6); 544 sf[1] = sf[2]; 545 break; 546 } 547 } 548 } 549 } 550 551 /* samples */ 552 for (k = 0; k < 3; k++) { 553 for (l = 0; l < 12; l += 3) { 554 j = 0; 555 for (i = 0; i < bound; i++) { 556 bit_alloc_bits = alloc_table[j]; 557 for (ch = 0; ch < s->nb_channels; ch++) { 558 b = bit_alloc[ch][i]; 559 if (b) { 560 scale = scale_factors[ch][i][k]; 561 qindex = alloc_table[j+b]; 562 bits = ff_mpa_quant_bits[qindex]; 563 if (bits < 0) { 564 int v2; 565 /* 3 values at the same time */ 566 v = get_bits(&s->gb, -bits); 567 v2 = ff_division_tabs[qindex][v]; 568 steps = ff_mpa_quant_steps[qindex]; 569 570 s->sb_samples[ch][k * 12 + l + 0][i] = 571 l2_unscale_group(steps, v2 & 15, scale); 572 s->sb_samples[ch][k * 12 + l + 1][i] = 573 l2_unscale_group(steps, (v2 >> 4) & 15, scale); 574 s->sb_samples[ch][k * 12 + l + 2][i] = 575 l2_unscale_group(steps, v2 >> 8 , scale); 576 } else { 577 for (m = 0; m < 3; m++) { 578 v = get_bits(&s->gb, bits); 579 v = l1_unscale(bits - 1, v, scale); 580 s->sb_samples[ch][k * 12 + l + m][i] = v; 581 } 582 } 583 } else { 584 s->sb_samples[ch][k * 12 + l + 0][i] = 0; 585 s->sb_samples[ch][k * 12 + l + 1][i] = 0; 586 s->sb_samples[ch][k * 12 + l + 2][i] = 0; 587 } 588 } 589 /* next subband in alloc table */ 590 j += 1 << bit_alloc_bits; 591 } 592 /* XXX: find a way to avoid this duplication of code */ 593 for (i = bound; i < sblimit; i++) { 594 bit_alloc_bits = alloc_table[j]; 595 b = bit_alloc[0][i]; 596 if (b) { 597 int mant, scale0, scale1; 598 scale0 = scale_factors[0][i][k]; 599 scale1 = scale_factors[1][i][k]; 600 qindex = alloc_table[j + b]; 601 bits = ff_mpa_quant_bits[qindex]; 602 if (bits < 0) { 603 /* 3 values at the same time */ 604 v = get_bits(&s->gb, -bits); 605 steps = ff_mpa_quant_steps[qindex]; 606 mant = v % steps; 607 v = v / steps; 608 s->sb_samples[0][k * 12 + l + 0][i] = 609 l2_unscale_group(steps, mant, scale0); 610 s->sb_samples[1][k * 12 + l + 0][i] = 611 l2_unscale_group(steps, mant, scale1); 612 mant = v % steps; 613 v = v / steps; 614 s->sb_samples[0][k * 12 + l + 1][i] = 615 l2_unscale_group(steps, mant, scale0); 616 s->sb_samples[1][k * 12 + l + 1][i] = 617 l2_unscale_group(steps, mant, scale1); 618 s->sb_samples[0][k * 12 + l + 2][i] = 619 l2_unscale_group(steps, v, scale0); 620 s->sb_samples[1][k * 12 + l + 2][i] = 621 l2_unscale_group(steps, v, scale1); 622 } else { 623 for (m = 0; m < 3; m++) { 624 mant = get_bits(&s->gb, bits); 625 s->sb_samples[0][k * 12 + l + m][i] = 626 l1_unscale(bits - 1, mant, scale0); 627 s->sb_samples[1][k * 12 + l + m][i] = 628 l1_unscale(bits - 1, mant, scale1); 629 } 630 } 631 } else { 632 s->sb_samples[0][k * 12 + l + 0][i] = 0; 633 s->sb_samples[0][k * 12 + l + 1][i] = 0; 634 s->sb_samples[0][k * 12 + l + 2][i] = 0; 635 s->sb_samples[1][k * 12 + l + 0][i] = 0; 636 s->sb_samples[1][k * 12 + l + 1][i] = 0; 637 s->sb_samples[1][k * 12 + l + 2][i] = 0; 638 } 639 /* next subband in alloc table */ 640 j += 1 << bit_alloc_bits; 641 } 642 /* fill remaining samples to zero */ 643 for (i = sblimit; i < SBLIMIT; i++) { 644 for (ch = 0; ch < s->nb_channels; ch++) { 645 s->sb_samples[ch][k * 12 + l + 0][i] = 0; 646 s->sb_samples[ch][k * 12 + l + 1][i] = 0; 647 s->sb_samples[ch][k * 12 + l + 2][i] = 0; 648 } 649 } 650 } 651 } 652 return 3 * 12; 653 } 654 655 #define SPLIT(dst,sf,n) \ 656 if (n == 3) { \ 657 int m = (sf * 171) >> 9; \ 658 dst = sf - 3 * m; \ 659 sf = m; \ 660 } else if (n == 4) { \ 661 dst = sf & 3; \ 662 sf >>= 2; \ 663 } else if (n == 5) { \ 664 int m = (sf * 205) >> 10; \ 665 dst = sf - 5 * m; \ 666 sf = m; \ 667 } else if (n == 6) { \ 668 int m = (sf * 171) >> 10; \ 669 dst = sf - 6 * m; \ 670 sf = m; \ 671 } else { \ 672 dst = 0; \ 673 } 674 675 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2, 676 int n3) 677 { 678 SPLIT(slen[3], sf, n3) 679 SPLIT(slen[2], sf, n2) 680 SPLIT(slen[1], sf, n1) 681 slen[0] = sf; 682 } 683 684 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g, 685 int16_t *exponents) 686 { 687 const uint8_t *bstab, *pretab; 688 int len, i, j, k, l, v0, shift, gain, gains[3]; 689 int16_t *exp_ptr; 690 691 exp_ptr = exponents; 692 gain = g->global_gain - 210; 693 shift = g->scalefac_scale + 1; 694 695 bstab = ff_band_size_long[s->sample_rate_index]; 696 pretab = ff_mpa_pretab[g->preflag]; 697 for (i = 0; i < g->long_end; i++) { 698 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400; 699 len = bstab[i]; 700 for (j = len; j > 0; j--) 701 *exp_ptr++ = v0; 702 } 703 704 if (g->short_start < 13) { 705 bstab = ff_band_size_short[s->sample_rate_index]; 706 gains[0] = gain - (g->subblock_gain[0] << 3); 707 gains[1] = gain - (g->subblock_gain[1] << 3); 708 gains[2] = gain - (g->subblock_gain[2] << 3); 709 k = g->long_end; 710 for (i = g->short_start; i < 13; i++) { 711 len = bstab[i]; 712 for (l = 0; l < 3; l++) { 713 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400; 714 for (j = len; j > 0; j--) 715 *exp_ptr++ = v0; 716 } 717 } 718 } 719 } 720 721 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, 722 int *end_pos2) 723 { 724 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits - s->extrasize * 8) { 725 s->gb = s->in_gb; 726 s->in_gb.buffer = NULL; 727 s->extrasize = 0; 728 av_assert2((get_bits_count(&s->gb) & 7) == 0); 729 skip_bits_long(&s->gb, *pos - *end_pos); 730 *end_pos2 = 731 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos; 732 *pos = get_bits_count(&s->gb); 733 } 734 } 735 736 /* Following is an optimized code for 737 INTFLOAT v = *src 738 if(get_bits1(&s->gb)) 739 v = -v; 740 *dst = v; 741 */ 742 #if USE_FLOATS 743 #define READ_FLIP_SIGN(dst,src) \ 744 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \ 745 AV_WN32A(dst, v); 746 #else 747 #define READ_FLIP_SIGN(dst,src) \ 748 v = -get_bits1(&s->gb); \ 749 *(dst) = (*(src) ^ v) - v; 750 #endif 751 752 static int huffman_decode(MPADecodeContext *s, GranuleDef *g, 753 int16_t *exponents, int end_pos2) 754 { 755 int s_index; 756 int i; 757 int last_pos, bits_left; 758 VLC *vlc; 759 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits - s->extrasize * 8); 760 761 /* low frequencies (called big values) */ 762 s_index = 0; 763 for (i = 0; i < 3; i++) { 764 const VLCElem *vlctab; 765 int j, k, l, linbits; 766 j = g->region_size[i]; 767 if (j == 0) 768 continue; 769 /* select vlc table */ 770 k = g->table_select[i]; 771 l = ff_mpa_huff_data[k][0]; 772 linbits = ff_mpa_huff_data[k][1]; 773 774 if (!l) { 775 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j); 776 s_index += 2 * j; 777 continue; 778 } 779 vlctab = ff_huff_vlc[l]; 780 781 /* read huffcode and compute each couple */ 782 for (; j > 0; j--) { 783 int exponent, x, y; 784 int v; 785 int pos = get_bits_count(&s->gb); 786 787 if (pos >= end_pos){ 788 switch_buffer(s, &pos, &end_pos, &end_pos2); 789 if (pos >= end_pos) 790 break; 791 } 792 y = get_vlc2(&s->gb, vlctab, 7, 3); 793 794 if (!y) { 795 g->sb_hybrid[s_index ] = 796 g->sb_hybrid[s_index + 1] = 0; 797 s_index += 2; 798 continue; 799 } 800 801 exponent= exponents[s_index]; 802 803 ff_dlog(s->avctx, "region=%d n=%d y=%d exp=%d\n", 804 i, g->region_size[i] - j, y, exponent); 805 if (y & 16) { 806 x = y >> 5; 807 y = y & 0x0f; 808 if (x < 15) { 809 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x) 810 } else { 811 x += get_bitsz(&s->gb, linbits); 812 v = l3_unscale(x, exponent); 813 if (get_bits1(&s->gb)) 814 v = -v; 815 g->sb_hybrid[s_index] = v; 816 } 817 if (y < 15) { 818 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y) 819 } else { 820 y += get_bitsz(&s->gb, linbits); 821 v = l3_unscale(y, exponent); 822 if (get_bits1(&s->gb)) 823 v = -v; 824 g->sb_hybrid[s_index + 1] = v; 825 } 826 } else { 827 x = y >> 5; 828 y = y & 0x0f; 829 x += y; 830 if (x < 15) { 831 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x) 832 } else { 833 x += get_bitsz(&s->gb, linbits); 834 v = l3_unscale(x, exponent); 835 if (get_bits1(&s->gb)) 836 v = -v; 837 g->sb_hybrid[s_index+!!y] = v; 838 } 839 g->sb_hybrid[s_index + !y] = 0; 840 } 841 s_index += 2; 842 } 843 } 844 845 /* high frequencies */ 846 vlc = &ff_huff_quad_vlc[g->count1table_select]; 847 last_pos = 0; 848 while (s_index <= 572) { 849 int pos, code; 850 pos = get_bits_count(&s->gb); 851 if (pos >= end_pos) { 852 if (pos > end_pos2 && last_pos) { 853 /* some encoders generate an incorrect size for this 854 part. We must go back into the data */ 855 s_index -= 4; 856 skip_bits_long(&s->gb, last_pos - pos); 857 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos); 858 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT)) 859 s_index=0; 860 break; 861 } 862 switch_buffer(s, &pos, &end_pos, &end_pos2); 863 if (pos >= end_pos) 864 break; 865 } 866 last_pos = pos; 867 868 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1); 869 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code); 870 g->sb_hybrid[s_index + 0] = 871 g->sb_hybrid[s_index + 1] = 872 g->sb_hybrid[s_index + 2] = 873 g->sb_hybrid[s_index + 3] = 0; 874 while (code) { 875 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 }; 876 int v; 877 int pos = s_index + idxtab[code]; 878 code ^= 8 >> idxtab[code]; 879 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos]) 880 } 881 s_index += 4; 882 } 883 /* skip extension bits */ 884 bits_left = end_pos2 - get_bits_count(&s->gb); 885 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) { 886 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left); 887 s_index=0; 888 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) { 889 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left); 890 s_index = 0; 891 } 892 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index)); 893 skip_bits_long(&s->gb, bits_left); 894 895 i = get_bits_count(&s->gb); 896 switch_buffer(s, &i, &end_pos, &end_pos2); 897 898 return 0; 899 } 900 901 /* Reorder short blocks from bitstream order to interleaved order. It 902 would be faster to do it in parsing, but the code would be far more 903 complicated */ 904 static void reorder_block(MPADecodeContext *s, GranuleDef *g) 905 { 906 int i, j, len; 907 INTFLOAT *ptr, *dst, *ptr1; 908 INTFLOAT tmp[576]; 909 910 if (g->block_type != 2) 911 return; 912 913 if (g->switch_point) { 914 if (s->sample_rate_index != 8) 915 ptr = g->sb_hybrid + 36; 916 else 917 ptr = g->sb_hybrid + 72; 918 } else { 919 ptr = g->sb_hybrid; 920 } 921 922 for (i = g->short_start; i < 13; i++) { 923 len = ff_band_size_short[s->sample_rate_index][i]; 924 ptr1 = ptr; 925 dst = tmp; 926 for (j = len; j > 0; j--) { 927 *dst++ = ptr[0*len]; 928 *dst++ = ptr[1*len]; 929 *dst++ = ptr[2*len]; 930 ptr++; 931 } 932 ptr += 2 * len; 933 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1)); 934 } 935 } 936 937 #define ISQRT2 FIXR(0.70710678118654752440) 938 939 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1) 940 { 941 int i, j, k, l; 942 int sf_max, sf, len, non_zero_found; 943 INTFLOAT *tab0, *tab1, v1, v2; 944 const INTFLOAT (*is_tab)[16]; 945 SUINTFLOAT tmp0, tmp1; 946 int non_zero_found_short[3]; 947 948 /* intensity stereo */ 949 if (s->mode_ext & MODE_EXT_I_STEREO) { 950 if (!s->lsf) { 951 is_tab = is_table; 952 sf_max = 7; 953 } else { 954 is_tab = is_table_lsf[g1->scalefac_compress & 1]; 955 sf_max = 16; 956 } 957 958 tab0 = g0->sb_hybrid + 576; 959 tab1 = g1->sb_hybrid + 576; 960 961 non_zero_found_short[0] = 0; 962 non_zero_found_short[1] = 0; 963 non_zero_found_short[2] = 0; 964 k = (13 - g1->short_start) * 3 + g1->long_end - 3; 965 for (i = 12; i >= g1->short_start; i--) { 966 /* for last band, use previous scale factor */ 967 if (i != 11) 968 k -= 3; 969 len = ff_band_size_short[s->sample_rate_index][i]; 970 for (l = 2; l >= 0; l--) { 971 tab0 -= len; 972 tab1 -= len; 973 if (!non_zero_found_short[l]) { 974 /* test if non zero band. if so, stop doing i-stereo */ 975 for (j = 0; j < len; j++) { 976 if (tab1[j] != 0) { 977 non_zero_found_short[l] = 1; 978 goto found1; 979 } 980 } 981 sf = g1->scale_factors[k + l]; 982 if (sf >= sf_max) 983 goto found1; 984 985 v1 = is_tab[0][sf]; 986 v2 = is_tab[1][sf]; 987 for (j = 0; j < len; j++) { 988 tmp0 = tab0[j]; 989 tab0[j] = MULLx(tmp0, v1, FRAC_BITS); 990 tab1[j] = MULLx(tmp0, v2, FRAC_BITS); 991 } 992 } else { 993 found1: 994 if (s->mode_ext & MODE_EXT_MS_STEREO) { 995 /* lower part of the spectrum : do ms stereo 996 if enabled */ 997 for (j = 0; j < len; j++) { 998 tmp0 = tab0[j]; 999 tmp1 = tab1[j]; 1000 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS); 1001 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS); 1002 } 1003 } 1004 } 1005 } 1006 } 1007 1008 non_zero_found = non_zero_found_short[0] | 1009 non_zero_found_short[1] | 1010 non_zero_found_short[2]; 1011 1012 for (i = g1->long_end - 1;i >= 0;i--) { 1013 len = ff_band_size_long[s->sample_rate_index][i]; 1014 tab0 -= len; 1015 tab1 -= len; 1016 /* test if non zero band. if so, stop doing i-stereo */ 1017 if (!non_zero_found) { 1018 for (j = 0; j < len; j++) { 1019 if (tab1[j] != 0) { 1020 non_zero_found = 1; 1021 goto found2; 1022 } 1023 } 1024 /* for last band, use previous scale factor */ 1025 k = (i == 21) ? 20 : i; 1026 sf = g1->scale_factors[k]; 1027 if (sf >= sf_max) 1028 goto found2; 1029 v1 = is_tab[0][sf]; 1030 v2 = is_tab[1][sf]; 1031 for (j = 0; j < len; j++) { 1032 tmp0 = tab0[j]; 1033 tab0[j] = MULLx(tmp0, v1, FRAC_BITS); 1034 tab1[j] = MULLx(tmp0, v2, FRAC_BITS); 1035 } 1036 } else { 1037 found2: 1038 if (s->mode_ext & MODE_EXT_MS_STEREO) { 1039 /* lower part of the spectrum : do ms stereo 1040 if enabled */ 1041 for (j = 0; j < len; j++) { 1042 tmp0 = tab0[j]; 1043 tmp1 = tab1[j]; 1044 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS); 1045 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS); 1046 } 1047 } 1048 } 1049 } 1050 } else if (s->mode_ext & MODE_EXT_MS_STEREO) { 1051 /* ms stereo ONLY */ 1052 /* NOTE: the 1/sqrt(2) normalization factor is included in the 1053 global gain */ 1054 #if USE_FLOATS 1055 s->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576); 1056 #else 1057 tab0 = g0->sb_hybrid; 1058 tab1 = g1->sb_hybrid; 1059 for (i = 0; i < 576; i++) { 1060 tmp0 = tab0[i]; 1061 tmp1 = tab1[i]; 1062 tab0[i] = tmp0 + tmp1; 1063 tab1[i] = tmp0 - tmp1; 1064 } 1065 #endif 1066 } 1067 } 1068 1069 #if USE_FLOATS 1070 #if HAVE_MIPSFPU 1071 # include "mips/compute_antialias_float.h" 1072 #endif /* HAVE_MIPSFPU */ 1073 #else 1074 #if HAVE_MIPSDSP 1075 # include "mips/compute_antialias_fixed.h" 1076 #endif /* HAVE_MIPSDSP */ 1077 #endif /* USE_FLOATS */ 1078 1079 #ifndef compute_antialias 1080 #if USE_FLOATS 1081 #define AA(j) do { \ 1082 float tmp0 = ptr[-1-j]; \ 1083 float tmp1 = ptr[ j]; \ 1084 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \ 1085 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \ 1086 } while (0) 1087 #else 1088 #define AA(j) do { \ 1089 SUINT tmp0 = ptr[-1-j]; \ 1090 SUINT tmp1 = ptr[ j]; \ 1091 SUINT tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \ 1092 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \ 1093 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \ 1094 } while (0) 1095 #endif 1096 1097 static void compute_antialias(MPADecodeContext *s, GranuleDef *g) 1098 { 1099 INTFLOAT *ptr; 1100 int n, i; 1101 1102 /* we antialias only "long" bands */ 1103 if (g->block_type == 2) { 1104 if (!g->switch_point) 1105 return; 1106 /* XXX: check this for 8000Hz case */ 1107 n = 1; 1108 } else { 1109 n = SBLIMIT - 1; 1110 } 1111 1112 ptr = g->sb_hybrid + 18; 1113 for (i = n; i > 0; i--) { 1114 AA(0); 1115 AA(1); 1116 AA(2); 1117 AA(3); 1118 AA(4); 1119 AA(5); 1120 AA(6); 1121 AA(7); 1122 1123 ptr += 18; 1124 } 1125 } 1126 #endif /* compute_antialias */ 1127 1128 static void compute_imdct(MPADecodeContext *s, GranuleDef *g, 1129 INTFLOAT *sb_samples, INTFLOAT *mdct_buf) 1130 { 1131 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1; 1132 INTFLOAT out2[12]; 1133 int i, j, mdct_long_end, sblimit; 1134 1135 /* find last non zero block */ 1136 ptr = g->sb_hybrid + 576; 1137 ptr1 = g->sb_hybrid + 2 * 18; 1138 while (ptr >= ptr1) { 1139 int32_t *p; 1140 ptr -= 6; 1141 p = (int32_t*)ptr; 1142 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5]) 1143 break; 1144 } 1145 sblimit = ((ptr - g->sb_hybrid) / 18) + 1; 1146 1147 if (g->block_type == 2) { 1148 /* XXX: check for 8000 Hz */ 1149 if (g->switch_point) 1150 mdct_long_end = 2; 1151 else 1152 mdct_long_end = 0; 1153 } else { 1154 mdct_long_end = sblimit; 1155 } 1156 1157 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid, 1158 mdct_long_end, g->switch_point, 1159 g->block_type); 1160 1161 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3); 1162 ptr = g->sb_hybrid + 18 * mdct_long_end; 1163 1164 for (j = mdct_long_end; j < sblimit; j++) { 1165 /* select frequency inversion */ 1166 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))]; 1167 out_ptr = sb_samples + j; 1168 1169 for (i = 0; i < 6; i++) { 1170 *out_ptr = buf[4*i]; 1171 out_ptr += SBLIMIT; 1172 } 1173 imdct12(out2, ptr + 0); 1174 for (i = 0; i < 6; i++) { 1175 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)]; 1176 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1); 1177 out_ptr += SBLIMIT; 1178 } 1179 imdct12(out2, ptr + 1); 1180 for (i = 0; i < 6; i++) { 1181 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)]; 1182 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1); 1183 out_ptr += SBLIMIT; 1184 } 1185 imdct12(out2, ptr + 2); 1186 for (i = 0; i < 6; i++) { 1187 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)]; 1188 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1); 1189 buf[4*(i + 6*2)] = 0; 1190 } 1191 ptr += 18; 1192 buf += (j&3) != 3 ? 1 : (4*18-3); 1193 } 1194 /* zero bands */ 1195 for (j = sblimit; j < SBLIMIT; j++) { 1196 /* overlap */ 1197 out_ptr = sb_samples + j; 1198 for (i = 0; i < 18; i++) { 1199 *out_ptr = buf[4*i]; 1200 buf[4*i] = 0; 1201 out_ptr += SBLIMIT; 1202 } 1203 buf += (j&3) != 3 ? 1 : (4*18-3); 1204 } 1205 } 1206 1207 /* main layer3 decoding function */ 1208 static int mp_decode_layer3(MPADecodeContext *s) 1209 { 1210 int nb_granules, main_data_begin; 1211 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos; 1212 GranuleDef *g; 1213 int16_t exponents[576]; //FIXME try INTFLOAT 1214 int ret; 1215 1216 /* read side info */ 1217 if (s->lsf) { 1218 ret = handle_crc(s, ((s->nb_channels == 1) ? 8*9 : 8*17)); 1219 main_data_begin = get_bits(&s->gb, 8); 1220 skip_bits(&s->gb, s->nb_channels); 1221 nb_granules = 1; 1222 } else { 1223 ret = handle_crc(s, ((s->nb_channels == 1) ? 8*17 : 8*32)); 1224 main_data_begin = get_bits(&s->gb, 9); 1225 if (s->nb_channels == 2) 1226 skip_bits(&s->gb, 3); 1227 else 1228 skip_bits(&s->gb, 5); 1229 nb_granules = 2; 1230 for (ch = 0; ch < s->nb_channels; ch++) { 1231 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */ 1232 s->granules[ch][1].scfsi = get_bits(&s->gb, 4); 1233 } 1234 } 1235 if (ret < 0) 1236 return ret; 1237 1238 for (gr = 0; gr < nb_granules; gr++) { 1239 for (ch = 0; ch < s->nb_channels; ch++) { 1240 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch); 1241 g = &s->granules[ch][gr]; 1242 g->part2_3_length = get_bits(&s->gb, 12); 1243 g->big_values = get_bits(&s->gb, 9); 1244 if (g->big_values > 288) { 1245 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n"); 1246 return AVERROR_INVALIDDATA; 1247 } 1248 1249 g->global_gain = get_bits(&s->gb, 8); 1250 /* if MS stereo only is selected, we precompute the 1251 1/sqrt(2) renormalization factor */ 1252 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == 1253 MODE_EXT_MS_STEREO) 1254 g->global_gain -= 2; 1255 if (s->lsf) 1256 g->scalefac_compress = get_bits(&s->gb, 9); 1257 else 1258 g->scalefac_compress = get_bits(&s->gb, 4); 1259 blocksplit_flag = get_bits1(&s->gb); 1260 if (blocksplit_flag) { 1261 g->block_type = get_bits(&s->gb, 2); 1262 if (g->block_type == 0) { 1263 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n"); 1264 return AVERROR_INVALIDDATA; 1265 } 1266 g->switch_point = get_bits1(&s->gb); 1267 for (i = 0; i < 2; i++) 1268 g->table_select[i] = get_bits(&s->gb, 5); 1269 for (i = 0; i < 3; i++) 1270 g->subblock_gain[i] = get_bits(&s->gb, 3); 1271 init_short_region(s, g); 1272 } else { 1273 int region_address1, region_address2; 1274 g->block_type = 0; 1275 g->switch_point = 0; 1276 for (i = 0; i < 3; i++) 1277 g->table_select[i] = get_bits(&s->gb, 5); 1278 /* compute huffman coded region sizes */ 1279 region_address1 = get_bits(&s->gb, 4); 1280 region_address2 = get_bits(&s->gb, 3); 1281 ff_dlog(s->avctx, "region1=%d region2=%d\n", 1282 region_address1, region_address2); 1283 init_long_region(s, g, region_address1, region_address2); 1284 } 1285 region_offset2size(g); 1286 compute_band_indexes(s, g); 1287 1288 g->preflag = 0; 1289 if (!s->lsf) 1290 g->preflag = get_bits1(&s->gb); 1291 g->scalefac_scale = get_bits1(&s->gb); 1292 g->count1table_select = get_bits1(&s->gb); 1293 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n", 1294 g->block_type, g->switch_point); 1295 } 1296 } 1297 1298 if (!s->adu_mode) { 1299 int skip; 1300 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb) >> 3); 1301 s->extrasize = av_clip((get_bits_left(&s->gb) >> 3) - s->extrasize, 0, 1302 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size)); 1303 av_assert1((get_bits_count(&s->gb) & 7) == 0); 1304 /* now we get bits from the main_data_begin offset */ 1305 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n", 1306 main_data_begin, s->last_buf_size); 1307 1308 memcpy(s->last_buf + s->last_buf_size, ptr, s->extrasize); 1309 s->in_gb = s->gb; 1310 init_get_bits(&s->gb, s->last_buf, (s->last_buf_size + s->extrasize) * 8); 1311 s->last_buf_size <<= 3; 1312 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) { 1313 for (ch = 0; ch < s->nb_channels; ch++) { 1314 g = &s->granules[ch][gr]; 1315 s->last_buf_size += g->part2_3_length; 1316 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid)); 1317 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); 1318 } 1319 } 1320 skip = s->last_buf_size - 8 * main_data_begin; 1321 if (skip >= s->gb.size_in_bits - s->extrasize * 8 && s->in_gb.buffer) { 1322 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits + s->extrasize * 8); 1323 s->gb = s->in_gb; 1324 s->in_gb.buffer = NULL; 1325 s->extrasize = 0; 1326 } else { 1327 skip_bits_long(&s->gb, skip); 1328 } 1329 } else { 1330 gr = 0; 1331 s->extrasize = 0; 1332 } 1333 1334 for (; gr < nb_granules; gr++) { 1335 for (ch = 0; ch < s->nb_channels; ch++) { 1336 g = &s->granules[ch][gr]; 1337 bits_pos = get_bits_count(&s->gb); 1338 1339 if (!s->lsf) { 1340 uint8_t *sc; 1341 int slen, slen1, slen2; 1342 1343 /* MPEG-1 scale factors */ 1344 slen1 = ff_slen_table[0][g->scalefac_compress]; 1345 slen2 = ff_slen_table[1][g->scalefac_compress]; 1346 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2); 1347 if (g->block_type == 2) { 1348 n = g->switch_point ? 17 : 18; 1349 j = 0; 1350 if (slen1) { 1351 for (i = 0; i < n; i++) 1352 g->scale_factors[j++] = get_bits(&s->gb, slen1); 1353 } else { 1354 for (i = 0; i < n; i++) 1355 g->scale_factors[j++] = 0; 1356 } 1357 if (slen2) { 1358 for (i = 0; i < 18; i++) 1359 g->scale_factors[j++] = get_bits(&s->gb, slen2); 1360 for (i = 0; i < 3; i++) 1361 g->scale_factors[j++] = 0; 1362 } else { 1363 for (i = 0; i < 21; i++) 1364 g->scale_factors[j++] = 0; 1365 } 1366 } else { 1367 sc = s->granules[ch][0].scale_factors; 1368 j = 0; 1369 for (k = 0; k < 4; k++) { 1370 n = k == 0 ? 6 : 5; 1371 if ((g->scfsi & (0x8 >> k)) == 0) { 1372 slen = (k < 2) ? slen1 : slen2; 1373 if (slen) { 1374 for (i = 0; i < n; i++) 1375 g->scale_factors[j++] = get_bits(&s->gb, slen); 1376 } else { 1377 for (i = 0; i < n; i++) 1378 g->scale_factors[j++] = 0; 1379 } 1380 } else { 1381 /* simply copy from last granule */ 1382 for (i = 0; i < n; i++) { 1383 g->scale_factors[j] = sc[j]; 1384 j++; 1385 } 1386 } 1387 } 1388 g->scale_factors[j++] = 0; 1389 } 1390 } else { 1391 int tindex, tindex2, slen[4], sl, sf; 1392 1393 /* LSF scale factors */ 1394 if (g->block_type == 2) 1395 tindex = g->switch_point ? 2 : 1; 1396 else 1397 tindex = 0; 1398 1399 sf = g->scalefac_compress; 1400 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) { 1401 /* intensity stereo case */ 1402 sf >>= 1; 1403 if (sf < 180) { 1404 lsf_sf_expand(slen, sf, 6, 6, 0); 1405 tindex2 = 3; 1406 } else if (sf < 244) { 1407 lsf_sf_expand(slen, sf - 180, 4, 4, 0); 1408 tindex2 = 4; 1409 } else { 1410 lsf_sf_expand(slen, sf - 244, 3, 0, 0); 1411 tindex2 = 5; 1412 } 1413 } else { 1414 /* normal case */ 1415 if (sf < 400) { 1416 lsf_sf_expand(slen, sf, 5, 4, 4); 1417 tindex2 = 0; 1418 } else if (sf < 500) { 1419 lsf_sf_expand(slen, sf - 400, 5, 4, 0); 1420 tindex2 = 1; 1421 } else { 1422 lsf_sf_expand(slen, sf - 500, 3, 0, 0); 1423 tindex2 = 2; 1424 g->preflag = 1; 1425 } 1426 } 1427 1428 j = 0; 1429 for (k = 0; k < 4; k++) { 1430 n = ff_lsf_nsf_table[tindex2][tindex][k]; 1431 sl = slen[k]; 1432 if (sl) { 1433 for (i = 0; i < n; i++) 1434 g->scale_factors[j++] = get_bits(&s->gb, sl); 1435 } else { 1436 for (i = 0; i < n; i++) 1437 g->scale_factors[j++] = 0; 1438 } 1439 } 1440 /* XXX: should compute exact size */ 1441 for (; j < 40; j++) 1442 g->scale_factors[j] = 0; 1443 } 1444 1445 exponents_from_scale_factors(s, g, exponents); 1446 1447 /* read Huffman coded residue */ 1448 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length); 1449 } /* ch */ 1450 1451 if (s->mode == MPA_JSTEREO) 1452 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]); 1453 1454 for (ch = 0; ch < s->nb_channels; ch++) { 1455 g = &s->granules[ch][gr]; 1456 1457 reorder_block(s, g); 1458 compute_antialias(s, g); 1459 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); 1460 } 1461 } /* gr */ 1462 if (get_bits_count(&s->gb) < 0) 1463 skip_bits_long(&s->gb, -get_bits_count(&s->gb)); 1464 return nb_granules * 18; 1465 } 1466 1467 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples, 1468 const uint8_t *buf, int buf_size) 1469 { 1470 int i, nb_frames, ch, ret; 1471 OUT_INT *samples_ptr; 1472 1473 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8); 1474 if (s->error_protection) 1475 s->crc = get_bits(&s->gb, 16); 1476 1477 switch(s->layer) { 1478 case 1: 1479 s->avctx->frame_size = 384; 1480 nb_frames = mp_decode_layer1(s); 1481 break; 1482 case 2: 1483 s->avctx->frame_size = 1152; 1484 nb_frames = mp_decode_layer2(s); 1485 break; 1486 case 3: 1487 s->avctx->frame_size = s->lsf ? 576 : 1152; 1488 default: 1489 nb_frames = mp_decode_layer3(s); 1490 1491 s->last_buf_size=0; 1492 if (s->in_gb.buffer) { 1493 align_get_bits(&s->gb); 1494 i = (get_bits_left(&s->gb) >> 3) - s->extrasize; 1495 if (i >= 0 && i <= BACKSTEP_SIZE) { 1496 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb) >> 3), i); 1497 s->last_buf_size=i; 1498 } else 1499 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i); 1500 s->gb = s->in_gb; 1501 s->in_gb.buffer = NULL; 1502 s->extrasize = 0; 1503 } 1504 1505 align_get_bits(&s->gb); 1506 av_assert1((get_bits_count(&s->gb) & 7) == 0); 1507 i = (get_bits_left(&s->gb) >> 3) - s->extrasize; 1508 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) { 1509 if (i < 0) 1510 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i); 1511 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE); 1512 } 1513 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0); 1514 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i); 1515 s->last_buf_size += i; 1516 } 1517 1518 if(nb_frames < 0) 1519 return nb_frames; 1520 1521 /* get output buffer */ 1522 if (!samples) { 1523 av_assert0(s->frame); 1524 s->frame->nb_samples = s->avctx->frame_size; 1525 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) 1526 return ret; 1527 samples = (OUT_INT **)s->frame->extended_data; 1528 } 1529 1530 /* apply the synthesis filter */ 1531 for (ch = 0; ch < s->nb_channels; ch++) { 1532 int sample_stride; 1533 if (s->avctx->sample_fmt == OUT_FMT_P) { 1534 samples_ptr = samples[ch]; 1535 sample_stride = 1; 1536 } else { 1537 samples_ptr = samples[0] + ch; 1538 sample_stride = s->nb_channels; 1539 } 1540 for (i = 0; i < nb_frames; i++) { 1541 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch], 1542 &(s->synth_buf_offset[ch]), 1543 RENAME(ff_mpa_synth_window), 1544 &s->dither_state, samples_ptr, 1545 sample_stride, s->sb_samples[ch][i]); 1546 samples_ptr += 32 * sample_stride; 1547 } 1548 } 1549 1550 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels; 1551 } 1552 1553 static int decode_frame(AVCodecContext *avctx, AVFrame *frame, 1554 int *got_frame_ptr, AVPacket *avpkt) 1555 { 1556 const uint8_t *buf = avpkt->data; 1557 int buf_size = avpkt->size; 1558 MPADecodeContext *s = avctx->priv_data; 1559 uint32_t header; 1560 int ret; 1561 1562 int skipped = 0; 1563 while(buf_size && !*buf){ 1564 buf++; 1565 buf_size--; 1566 skipped++; 1567 } 1568 1569 if (buf_size < HEADER_SIZE) 1570 return AVERROR_INVALIDDATA; 1571 1572 header = AV_RB32(buf); 1573 if (header >> 8 == AV_RB32("TAG") >> 8) { 1574 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n"); 1575 return buf_size + skipped; 1576 } 1577 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header); 1578 if (ret < 0) { 1579 av_log(avctx, AV_LOG_ERROR, "Header missing\n"); 1580 return AVERROR_INVALIDDATA; 1581 } else if (ret == 1) { 1582 /* free format: prepare to compute frame size */ 1583 s->frame_size = -1; 1584 return AVERROR_INVALIDDATA; 1585 } 1586 /* update codec info */ 1587 av_channel_layout_uninit(&avctx->ch_layout); 1588 avctx->ch_layout = s->nb_channels == 1 ? (AVChannelLayout)AV_CHANNEL_LAYOUT_MONO : 1589 (AVChannelLayout)AV_CHANNEL_LAYOUT_STEREO; 1590 if (!avctx->bit_rate) 1591 avctx->bit_rate = s->bit_rate; 1592 1593 if (s->frame_size <= 0) { 1594 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n"); 1595 return AVERROR_INVALIDDATA; 1596 } else if (s->frame_size < buf_size) { 1597 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n"); 1598 buf_size= s->frame_size; 1599 } 1600 1601 s->frame = frame; 1602 1603 ret = mp_decode_frame(s, NULL, buf, buf_size); 1604 if (ret >= 0) { 1605 s->frame->nb_samples = avctx->frame_size; 1606 *got_frame_ptr = 1; 1607 avctx->sample_rate = s->sample_rate; 1608 //FIXME maybe move the other codec info stuff from above here too 1609 } else { 1610 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n"); 1611 /* Only return an error if the bad frame makes up the whole packet or 1612 * the error is related to buffer management. 1613 * If there is more data in the packet, just consume the bad frame 1614 * instead of returning an error, which would discard the whole 1615 * packet. */ 1616 *got_frame_ptr = 0; 1617 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA) 1618 return ret; 1619 } 1620 s->frame_size = 0; 1621 return buf_size + skipped; 1622 } 1623 1624 static void mp_flush(MPADecodeContext *ctx) 1625 { 1626 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf)); 1627 memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf)); 1628 ctx->last_buf_size = 0; 1629 ctx->dither_state = 0; 1630 } 1631 1632 static void flush(AVCodecContext *avctx) 1633 { 1634 mp_flush(avctx->priv_data); 1635 } 1636 1637 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER 1638 static int decode_frame_adu(AVCodecContext *avctx, AVFrame *frame, 1639 int *got_frame_ptr, AVPacket *avpkt) 1640 { 1641 const uint8_t *buf = avpkt->data; 1642 int buf_size = avpkt->size; 1643 MPADecodeContext *s = avctx->priv_data; 1644 uint32_t header; 1645 int len, ret; 1646 1647 len = buf_size; 1648 1649 // Discard too short frames 1650 if (buf_size < HEADER_SIZE) { 1651 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n"); 1652 return AVERROR_INVALIDDATA; 1653 } 1654 1655 1656 if (len > MPA_MAX_CODED_FRAME_SIZE) 1657 len = MPA_MAX_CODED_FRAME_SIZE; 1658 1659 // Get header and restore sync word 1660 header = AV_RB32(buf) | 0xffe00000; 1661 1662 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header); 1663 if (ret < 0) { 1664 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n"); 1665 return ret; 1666 } 1667 /* update codec info */ 1668 avctx->sample_rate = s->sample_rate; 1669 av_channel_layout_uninit(&avctx->ch_layout); 1670 avctx->ch_layout = s->nb_channels == 1 ? (AVChannelLayout)AV_CHANNEL_LAYOUT_MONO : 1671 (AVChannelLayout)AV_CHANNEL_LAYOUT_STEREO; 1672 if (!avctx->bit_rate) 1673 avctx->bit_rate = s->bit_rate; 1674 1675 s->frame_size = len; 1676 1677 s->frame = frame; 1678 1679 ret = mp_decode_frame(s, NULL, buf, buf_size); 1680 if (ret < 0) { 1681 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n"); 1682 return ret; 1683 } 1684 1685 *got_frame_ptr = 1; 1686 1687 return buf_size; 1688 } 1689 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */ 1690 1691 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER 1692 1693 /** 1694 * Context for MP3On4 decoder 1695 */ 1696 typedef struct MP3On4DecodeContext { 1697 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances) 1698 int syncword; ///< syncword patch 1699 const uint8_t *coff; ///< channel offsets in output buffer 1700 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance 1701 } MP3On4DecodeContext; 1702 1703 #include "mpeg4audio.h" 1704 1705 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */ 1706 1707 /* number of mp3 decoder instances */ 1708 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 }; 1709 1710 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */ 1711 static const uint8_t chan_offset[8][5] = { 1712 { 0 }, 1713 { 0 }, // C 1714 { 0 }, // FLR 1715 { 2, 0 }, // C FLR 1716 { 2, 0, 3 }, // C FLR BS 1717 { 2, 0, 3 }, // C FLR BLRS 1718 { 2, 0, 4, 3 }, // C FLR BLRS LFE 1719 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE 1720 }; 1721 1722 /* mp3on4 channel layouts */ 1723 static const int16_t chan_layout[8] = { 1724 0, 1725 AV_CH_LAYOUT_MONO, 1726 AV_CH_LAYOUT_STEREO, 1727 AV_CH_LAYOUT_SURROUND, 1728 AV_CH_LAYOUT_4POINT0, 1729 AV_CH_LAYOUT_5POINT0, 1730 AV_CH_LAYOUT_5POINT1, 1731 AV_CH_LAYOUT_7POINT1 1732 }; 1733 1734 static av_cold int decode_close_mp3on4(AVCodecContext * avctx) 1735 { 1736 MP3On4DecodeContext *s = avctx->priv_data; 1737 int i; 1738 1739 for (i = 0; i < s->frames; i++) 1740 av_freep(&s->mp3decctx[i]); 1741 1742 return 0; 1743 } 1744 1745 1746 static av_cold int decode_init_mp3on4(AVCodecContext * avctx) 1747 { 1748 MP3On4DecodeContext *s = avctx->priv_data; 1749 MPEG4AudioConfig cfg; 1750 int i, ret; 1751 1752 if ((avctx->extradata_size < 2) || !avctx->extradata) { 1753 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n"); 1754 return AVERROR_INVALIDDATA; 1755 } 1756 1757 avpriv_mpeg4audio_get_config2(&cfg, avctx->extradata, 1758 avctx->extradata_size, 1, avctx); 1759 if (!cfg.chan_config || cfg.chan_config > 7) { 1760 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n"); 1761 return AVERROR_INVALIDDATA; 1762 } 1763 s->frames = mp3Frames[cfg.chan_config]; 1764 s->coff = chan_offset[cfg.chan_config]; 1765 av_channel_layout_uninit(&avctx->ch_layout); 1766 av_channel_layout_from_mask(&avctx->ch_layout, chan_layout[cfg.chan_config]); 1767 1768 if (cfg.sample_rate < 16000) 1769 s->syncword = 0xffe00000; 1770 else 1771 s->syncword = 0xfff00000; 1772 1773 /* Init the first mp3 decoder in standard way, so that all tables get builded 1774 * We replace avctx->priv_data with the context of the first decoder so that 1775 * decode_init() does not have to be changed. 1776 * Other decoders will be initialized here copying data from the first context 1777 */ 1778 // Allocate zeroed memory for the first decoder context 1779 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext)); 1780 if (!s->mp3decctx[0]) 1781 return AVERROR(ENOMEM); 1782 // Put decoder context in place to make init_decode() happy 1783 avctx->priv_data = s->mp3decctx[0]; 1784 ret = decode_init(avctx); 1785 // Restore mp3on4 context pointer 1786 avctx->priv_data = s; 1787 if (ret < 0) 1788 return ret; 1789 s->mp3decctx[0]->adu_mode = 1; // Set adu mode 1790 1791 /* Create a separate codec/context for each frame (first is already ok). 1792 * Each frame is 1 or 2 channels - up to 5 frames allowed 1793 */ 1794 for (i = 1; i < s->frames; i++) { 1795 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext)); 1796 if (!s->mp3decctx[i]) 1797 return AVERROR(ENOMEM); 1798 s->mp3decctx[i]->adu_mode = 1; 1799 s->mp3decctx[i]->avctx = avctx; 1800 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp; 1801 s->mp3decctx[i]->butterflies_float = s->mp3decctx[0]->butterflies_float; 1802 } 1803 1804 return 0; 1805 } 1806 1807 1808 static void flush_mp3on4(AVCodecContext *avctx) 1809 { 1810 int i; 1811 MP3On4DecodeContext *s = avctx->priv_data; 1812 1813 for (i = 0; i < s->frames; i++) 1814 mp_flush(s->mp3decctx[i]); 1815 } 1816 1817 1818 static int decode_frame_mp3on4(AVCodecContext *avctx, AVFrame *frame, 1819 int *got_frame_ptr, AVPacket *avpkt) 1820 { 1821 const uint8_t *buf = avpkt->data; 1822 int buf_size = avpkt->size; 1823 MP3On4DecodeContext *s = avctx->priv_data; 1824 MPADecodeContext *m; 1825 int fsize, len = buf_size, out_size = 0; 1826 uint32_t header; 1827 OUT_INT **out_samples; 1828 OUT_INT *outptr[2]; 1829 int fr, ch, ret; 1830 1831 /* get output buffer */ 1832 frame->nb_samples = MPA_FRAME_SIZE; 1833 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) 1834 return ret; 1835 out_samples = (OUT_INT **)frame->extended_data; 1836 1837 // Discard too short frames 1838 if (buf_size < HEADER_SIZE) 1839 return AVERROR_INVALIDDATA; 1840 1841 avctx->bit_rate = 0; 1842 1843 ch = 0; 1844 for (fr = 0; fr < s->frames; fr++) { 1845 fsize = AV_RB16(buf) >> 4; 1846 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE); 1847 m = s->mp3decctx[fr]; 1848 av_assert1(m); 1849 1850 if (fsize < HEADER_SIZE) { 1851 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n"); 1852 return AVERROR_INVALIDDATA; 1853 } 1854 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header 1855 1856 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header); 1857 if (ret < 0) { 1858 av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n"); 1859 return AVERROR_INVALIDDATA; 1860 } 1861 1862 if (ch + m->nb_channels > avctx->ch_layout.nb_channels || 1863 s->coff[fr] + m->nb_channels > avctx->ch_layout.nb_channels) { 1864 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec " 1865 "channel count\n"); 1866 return AVERROR_INVALIDDATA; 1867 } 1868 ch += m->nb_channels; 1869 1870 outptr[0] = out_samples[s->coff[fr]]; 1871 if (m->nb_channels > 1) 1872 outptr[1] = out_samples[s->coff[fr] + 1]; 1873 1874 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) { 1875 av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch); 1876 memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT)); 1877 if (m->nb_channels > 1) 1878 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT)); 1879 ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT); 1880 } 1881 1882 out_size += ret; 1883 buf += fsize; 1884 len -= fsize; 1885 1886 avctx->bit_rate += m->bit_rate; 1887 } 1888 if (ch != avctx->ch_layout.nb_channels) { 1889 av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n"); 1890 return AVERROR_INVALIDDATA; 1891 } 1892 1893 /* update codec info */ 1894 avctx->sample_rate = s->mp3decctx[0]->sample_rate; 1895 1896 frame->nb_samples = out_size / (avctx->ch_layout.nb_channels * sizeof(OUT_INT)); 1897 *got_frame_ptr = 1; 1898 1899 return buf_size; 1900 } 1901 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */