libavcodec/wmadec.c

   1 /*
   2  * WMA compatible decoder
   3  * Copyright (c) 2002 The FFmpeg Project.
   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 wmadec.c
  24  * WMA compatible decoder.
  25  * This decoder handles Microsoft Windows Media Audio data, versions 1 & 2.
  26  * WMA v1 is identified by audio format 0x160 in Microsoft media files
  27  * (ASF/AVI/WAV). WMA v2 is identified by audio format 0x161.
  28  *
  29  * To use this decoder, a calling application must supply the extra data
  30  * bytes provided with the WMA data. These are the extra, codec-specific
  31  * bytes at the end of a WAVEFORMATEX data structure. Transmit these bytes
  32  * to the decoder using the extradata[_size] fields in AVCodecContext. There
  33  * should be 4 extra bytes for v1 data and 6 extra bytes for v2 data.
  34  */
  35
  36 #include "avcodec.h"
  37 #include "wma.h"
  38
  39 #undef NDEBUG
  40 #include <assert.h>
  41
  42 #define EXPVLCBITS 8
  43 #define EXPMAX ((19+EXPVLCBITS-1)/EXPVLCBITS)
  44
  45 #define HGAINVLCBITS 9
  46 #define HGAINMAX ((13+HGAINVLCBITS-1)/HGAINVLCBITS)
  47
  48 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len);
  49
  50 #ifdef TRACE
  51 static void dump_shorts(WMACodecContext *s, const char *name, const short *tab, int n)
  52 {
  53     int i;
  54
  55     tprintf(s->avctx, "%s[%d]:\n", name, n);
  56     for(i=0;i<n;i++) {
  57         if ((i & 7) == 0)
  58             tprintf(s->avctx, "%4d: ", i);
  59         tprintf(s->avctx, " %5d.0", tab[i]);
  60         if ((i & 7) == 7)
  61             tprintf(s->avctx, "\n");
  62     }
  63 }
  64
  65 static void dump_floats(WMACodecContext *s, const char *name, int prec, const float *tab, int n)
  66 {
  67     int i;
  68
  69     tprintf(s->avctx, "%s[%d]:\n", name, n);
  70     for(i=0;i<n;i++) {
  71         if ((i & 7) == 0)
  72             tprintf(s->avctx, "%4d: ", i);
  73         tprintf(s->avctx, " %8.*f", prec, tab[i]);
  74         if ((i & 7) == 7)
  75             tprintf(s->avctx, "\n");
  76     }
  77     if ((i & 7) != 0)
  78         tprintf(s->avctx, "\n");
  79 }
  80 #endif
  81
  82 static int wma_decode_init(AVCodecContext * avctx)
  83 {
  84     WMACodecContext *s = avctx->priv_data;
  85     int i, flags1, flags2;
  86     uint8_t *extradata;
  87
  88     s->avctx = avctx;
  89
  90     /* extract flag infos */
  91     flags1 = 0;
  92     flags2 = 0;
  93     extradata = avctx->extradata;
  94     if (avctx->codec->id == CODEC_ID_WMAV1 && avctx->extradata_size >= 4) {
  95         flags1 = AV_RL16(extradata);
  96         flags2 = AV_RL16(extradata+2);
  97     } else if (avctx->codec->id == CODEC_ID_WMAV2 && avctx->extradata_size >= 6) {
  98         flags1 = AV_RL32(extradata);
  99         flags2 = AV_RL16(extradata+4);
 100     }
 101 // for(i=0; i<avctx->extradata_size; i++)
 102 //     av_log(NULL, AV_LOG_ERROR, "%02X ", extradata[i]);
 103
 104     s->use_exp_vlc = flags2 & 0x0001;
 105     s->use_bit_reservoir = flags2 & 0x0002;
 106     s->use_variable_block_len = flags2 & 0x0004;
 107
 108     if(ff_wma_init(avctx, flags2)<0)
 109         return -1;
 110
 111     /* init MDCT */
 112     for(i = 0; i < s->nb_block_sizes; i++)
 113         ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1);
 114
 115     if (s->use_noise_coding) {
 116         init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(ff_wma_hgain_huffbits),
 117                  ff_wma_hgain_huffbits, 1, 1,
 118                  ff_wma_hgain_huffcodes, 2, 2, 0);
 119     }
 120
 121     if (s->use_exp_vlc) {
 122         init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_wma_scale_huffbits), //FIXME move out of context
 123                  ff_wma_scale_huffbits, 1, 1,
 124                  ff_wma_scale_huffcodes, 4, 4, 0);
 125     } else {
 126         wma_lsp_to_curve_init(s, s->frame_len);
 127     }
 128
 129     return 0;
 130 }
 131
 132 /**
 133  * compute x^-0.25 with an exponent and mantissa table. We use linear
 134  * interpolation to reduce the mantissa table size at a small speed
 135  * expense (linear interpolation approximately doubles the number of
 136  * bits of precision).
 137  */
 138 static inline float pow_m1_4(WMACodecContext *s, float x)
 139 {
 140     union {
 141         float f;
 142         unsigned int v;
 143     } u, t;
 144     unsigned int e, m;
 145     float a, b;
 146
 147     u.f = x;
 148     e = u.v >> 23;
 149     m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1);
 150     /* build interpolation scale: 1 <= t < 2. */
 151     t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23);
 152     a = s->lsp_pow_m_table1[m];
 153     b = s->lsp_pow_m_table2[m];
 154     return s->lsp_pow_e_table[e] * (a + b * t.f);
 155 }
 156
 157 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
 158 {
 159     float wdel, a, b;
 160     int i, e, m;
 161
 162     wdel = M_PI / frame_len;
 163     for(i=0;i<frame_len;i++)
 164         s->lsp_cos_table[i] = 2.0f * cos(wdel * i);
 165
 166     /* tables for x^-0.25 computation */
 167     for(i=0;i<256;i++) {
 168         e = i - 126;
 169         s->lsp_pow_e_table[i] = pow(2.0, e * -0.25);
 170     }
 171
 172     /* NOTE: these two tables are needed to avoid two operations in
 173        pow_m1_4 */
 174     b = 1.0;
 175     for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) {
 176         m = (1 << LSP_POW_BITS) + i;
 177         a = (float)m * (0.5 / (1 << LSP_POW_BITS));
 178         a = pow(a, -0.25);
 179         s->lsp_pow_m_table1[i] = 2 * a - b;
 180         s->lsp_pow_m_table2[i] = b - a;
 181         b = a;
 182     }
 183 #if 0
 184     for(i=1;i<20;i++) {
 185         float v, r1, r2;
 186         v = 5.0 / i;
 187         r1 = pow_m1_4(s, v);
 188         r2 = pow(v,-0.25);
 189         printf("%f^-0.25=%f e=%f\n", v, r1, r2 - r1);
 190     }
 191 #endif
 192 }
 193
 194 /**
 195  * NOTE: We use the same code as Vorbis here
 196  * @todo optimize it further with SSE/3Dnow
 197  */
 198 static void wma_lsp_to_curve(WMACodecContext *s,
 199                              float *out, float *val_max_ptr,
 200                              int n, float *lsp)
 201 {
 202     int i, j;
 203     float p, q, w, v, val_max;
 204
 205     val_max = 0;
 206     for(i=0;i<n;i++) {
 207         p = 0.5f;
 208         q = 0.5f;
 209         w = s->lsp_cos_table[i];
 210         for(j=1;j<NB_LSP_COEFS;j+=2){
 211             q *= w - lsp[j - 1];
 212             p *= w - lsp[j];
 213         }
 214         p *= p * (2.0f - w);
 215         q *= q * (2.0f + w);
 216         v = p + q;
 217         v = pow_m1_4(s, v);
 218         if (v > val_max)
 219             val_max = v;
 220         out[i] = v;
 221     }
 222     *val_max_ptr = val_max;
 223 }
 224
 225 /**
 226  * decode exponents coded with LSP coefficients (same idea as Vorbis)
 227  */
 228 static void decode_exp_lsp(WMACodecContext *s, int ch)
 229 {
 230     float lsp_coefs[NB_LSP_COEFS];
 231     int val, i;
 232
 233     for(i = 0; i < NB_LSP_COEFS; i++) {
 234         if (i == 0 || i >= 8)
 235             val = get_bits(&s->gb, 3);
 236         else
 237             val = get_bits(&s->gb, 4);
 238         lsp_coefs[i] = ff_wma_lsp_codebook[i][val];
 239     }
 240
 241     wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch],
 242                      s->block_len, lsp_coefs);
 243 }
 244
 245 /**
 246  * decode exponents coded with VLC codes
 247  */
 248 static int decode_exp_vlc(WMACodecContext *s, int ch)
 249 {
 250     int last_exp, n, code;
 251     const uint16_t *ptr, *band_ptr;
 252     float v, *q, max_scale, *q_end;
 253
 254     band_ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
 255     ptr = band_ptr;
 256     q = s->exponents[ch];
 257     q_end = q + s->block_len;
 258     max_scale = 0;
 259     if (s->version == 1) {
 260         last_exp = get_bits(&s->gb, 5) + 10;
 261         /* XXX: use a table */
 262         v = pow(10, last_exp * (1.0 / 16.0));
 263         max_scale = v;
 264         n = *ptr++;
 265         do {
 266             *q++ = v;
 267         } while (--n);
 268     }else
 269         last_exp = 36;
 270
 271     while (q < q_end) {
 272         code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX);
 273         if (code < 0)
 274             return -1;
 275         /* NOTE: this offset is the same as MPEG4 AAC ! */
 276         last_exp += code - 60;
 277         /* XXX: use a table */
 278         v = pow(10, last_exp * (1.0 / 16.0));
 279         if (v > max_scale)
 280             max_scale = v;
 281         n = *ptr++;
 282         do {
 283             *q++ = v;
 284         } while (--n);
 285     }
 286     s->max_exponent[ch] = max_scale;
 287     return 0;
 288 }
 289
 290
 291 /**
 292  * Apply MDCT window and add into output.
 293  *
 294  * We ensure that when the windows overlap their squared sum
 295  * is always 1 (MDCT reconstruction rule).
 296  */
 297 static void wma_window(WMACodecContext *s, float *out)
 298 {
 299     float *in = s->output;
 300     int block_len, bsize, n;
 301
 302     /* left part */
 303     if (s->block_len_bits <= s->prev_block_len_bits) {
 304         block_len = s->block_len;
 305         bsize = s->frame_len_bits - s->block_len_bits;
 306
 307         s->dsp.vector_fmul_add_add(out, in, s->windows[bsize],
 308                                    out, 0, block_len, 1);
 309
 310     } else {
 311         block_len = 1 << s->prev_block_len_bits;
 312         n = (s->block_len - block_len) / 2;
 313         bsize = s->frame_len_bits - s->prev_block_len_bits;
 314
 315         s->dsp.vector_fmul_add_add(out+n, in+n, s->windows[bsize],
 316                                    out+n, 0, block_len, 1);
 317
 318         memcpy(out+n+block_len, in+n+block_len, n*sizeof(float));
 319     }
 320
 321     out += s->block_len;
 322     in += s->block_len;
 323
 324     /* right part */
 325     if (s->block_len_bits <= s->next_block_len_bits) {
 326         block_len = s->block_len;
 327         bsize = s->frame_len_bits - s->block_len_bits;
 328
 329         s->dsp.vector_fmul_reverse(out, in, s->windows[bsize], block_len);
 330
 331     } else {
 332         block_len = 1 << s->next_block_len_bits;
 333         n = (s->block_len - block_len) / 2;
 334         bsize = s->frame_len_bits - s->next_block_len_bits;
 335
 336         memcpy(out, in, n*sizeof(float));
 337
 338         s->dsp.vector_fmul_reverse(out+n, in+n, s->windows[bsize], block_len);
 339
 340         memset(out+n+block_len, 0, n*sizeof(float));
 341     }
 342 }
 343
 344
 345 /**
 346  * @return 0 if OK. 1 if last block of frame. return -1 if
 347  * unrecorrable error.
 348  */
 349 static int wma_decode_block(WMACodecContext *s)
 350 {
 351     int n, v, a, ch, code, bsize;
 352     int coef_nb_bits, total_gain;
 353     int nb_coefs[MAX_CHANNELS];
 354     float mdct_norm;
 355
 356 #ifdef TRACE
 357     tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num);
 358 #endif
 359
 360     /* compute current block length */
 361     if (s->use_variable_block_len) {
 362         n = av_log2(s->nb_block_sizes - 1) + 1;
 363
 364         if (s->reset_block_lengths) {
 365             s->reset_block_lengths = 0;
 366             v = get_bits(&s->gb, n);
 367             if (v >= s->nb_block_sizes)
 368                 return -1;
 369             s->prev_block_len_bits = s->frame_len_bits - v;
 370             v = get_bits(&s->gb, n);
 371             if (v >= s->nb_block_sizes)
 372                 return -1;
 373             s->block_len_bits = s->frame_len_bits - v;
 374         } else {
 375             /* update block lengths */
 376             s->prev_block_len_bits = s->block_len_bits;
 377             s->block_len_bits = s->next_block_len_bits;
 378         }
 379         v = get_bits(&s->gb, n);
 380         if (v >= s->nb_block_sizes)
 381             return -1;
 382         s->next_block_len_bits = s->frame_len_bits - v;
 383     } else {
 384         /* fixed block len */
 385         s->next_block_len_bits = s->frame_len_bits;
 386         s->prev_block_len_bits = s->frame_len_bits;
 387         s->block_len_bits = s->frame_len_bits;
 388     }
 389
 390     /* now check if the block length is coherent with the frame length */
 391     s->block_len = 1 << s->block_len_bits;
 392     if ((s->block_pos + s->block_len) > s->frame_len)
 393         return -1;
 394
 395     if (s->nb_channels == 2) {
 396         s->ms_stereo = get_bits1(&s->gb);
 397     }
 398     v = 0;
 399     for(ch = 0; ch < s->nb_channels; ch++) {
 400         a = get_bits1(&s->gb);
 401         s->channel_coded[ch] = a;
 402         v |= a;
 403     }
 404     /* if no channel coded, no need to go further */
 405     /* XXX: fix potential framing problems */
 406     if (!v)
 407         goto next;
 408
 409     bsize = s->frame_len_bits - s->block_len_bits;
 410
 411     /* read total gain and extract corresponding number of bits for
 412        coef escape coding */
 413     total_gain = 1;
 414     for(;;) {
 415         a = get_bits(&s->gb, 7);
 416         total_gain += a;
 417         if (a != 127)
 418             break;
 419     }
 420
 421     coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
 422
 423     /* compute number of coefficients */
 424     n = s->coefs_end[bsize] - s->coefs_start;
 425     for(ch = 0; ch < s->nb_channels; ch++)
 426         nb_coefs[ch] = n;
 427
 428     /* complex coding */
 429     if (s->use_noise_coding) {
 430
 431         for(ch = 0; ch < s->nb_channels; ch++) {
 432             if (s->channel_coded[ch]) {
 433                 int i, n, a;
 434                 n = s->exponent_high_sizes[bsize];
 435                 for(i=0;i<n;i++) {
 436                     a = get_bits1(&s->gb);
 437                     s->high_band_coded[ch][i] = a;
 438                     /* if noise coding, the coefficients are not transmitted */
 439                     if (a)
 440                         nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
 441                 }
 442             }
 443         }
 444         for(ch = 0; ch < s->nb_channels; ch++) {
 445             if (s->channel_coded[ch]) {
 446                 int i, n, val, code;
 447
 448                 n = s->exponent_high_sizes[bsize];
 449                 val = (int)0x80000000;
 450                 for(i=0;i<n;i++) {
 451                     if (s->high_band_coded[ch][i]) {
 452                         if (val == (int)0x80000000) {
 453                             val = get_bits(&s->gb, 7) - 19;
 454                         } else {
 455                             code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX);
 456                             if (code < 0)
 457                                 return -1;
 458                             val += code - 18;
 459                         }
 460                         s->high_band_values[ch][i] = val;
 461                     }
 462                 }
 463             }
 464         }
 465     }
 466
 467     /* exponents can be reused in short blocks. */
 468     if ((s->block_len_bits == s->frame_len_bits) ||
 469         get_bits1(&s->gb)) {
 470         for(ch = 0; ch < s->nb_channels; ch++) {
 471             if (s->channel_coded[ch]) {
 472                 if (s->use_exp_vlc) {
 473                     if (decode_exp_vlc(s, ch) < 0)
 474                         return -1;
 475                 } else {
 476                     decode_exp_lsp(s, ch);
 477                 }
 478                 s->exponents_bsize[ch] = bsize;
 479             }
 480         }
 481     }
 482
 483     /* parse spectral coefficients : just RLE encoding */
 484     for(ch = 0; ch < s->nb_channels; ch++) {
 485         if (s->channel_coded[ch]) {
 486             VLC *coef_vlc;
 487             int level, run, sign, tindex;
 488             int16_t *ptr, *eptr;
 489             const uint16_t *level_table, *run_table;
 490
 491             /* special VLC tables are used for ms stereo because
 492                there is potentially less energy there */
 493             tindex = (ch == 1 && s->ms_stereo);
 494             coef_vlc = &s->coef_vlc[tindex];
 495             run_table = s->run_table[tindex];
 496             level_table = s->level_table[tindex];
 497             /* XXX: optimize */
 498             ptr = &s->coefs1[ch][0];
 499             eptr = ptr + nb_coefs[ch];
 500             memset(ptr, 0, s->block_len * sizeof(int16_t));
 501             for(;;) {
 502                 code = get_vlc2(&s->gb, coef_vlc->table, VLCBITS, VLCMAX);
 503                 if (code < 0)
 504                     return -1;
 505                 if (code == 1) {
 506                     /* EOB */
 507                     break;
 508                 } else if (code == 0) {
 509                     /* escape */
 510                     level = get_bits(&s->gb, coef_nb_bits);
 511                     /* NOTE: this is rather suboptimal. reading
 512                        block_len_bits would be better */
 513                     run = get_bits(&s->gb, s->frame_len_bits);
 514                 } else {
 515                     /* normal code */
 516                     run = run_table[code];
 517                     level = level_table[code];
 518                 }
 519                 sign = get_bits1(&s->gb);
 520                 if (!sign)
 521                     level = -level;
 522                 ptr += run;
 523                 if (ptr >= eptr)
 524                 {
 525                     av_log(NULL, AV_LOG_ERROR, "overflow in spectral RLE, ignoring\n");
 526                     break;
 527                 }
 528                 *ptr++ = level;
 529                 /* NOTE: EOB can be omitted */
 530                 if (ptr >= eptr)
 531                     break;
 532             }
 533         }
 534         if (s->version == 1 && s->nb_channels >= 2) {
 535             align_get_bits(&s->gb);
 536         }
 537     }
 538
 539     /* normalize */
 540     {
 541         int n4 = s->block_len / 2;
 542         mdct_norm = 1.0 / (float)n4;
 543         if (s->version == 1) {
 544             mdct_norm *= sqrt(n4);
 545         }
 546     }
 547
 548     /* finally compute the MDCT coefficients */
 549     for(ch = 0; ch < s->nb_channels; ch++) {
 550         if (s->channel_coded[ch]) {
 551             int16_t *coefs1;
 552             float *coefs, *exponents, mult, mult1, noise;
 553             int i, j, n, n1, last_high_band, esize;
 554             float exp_power[HIGH_BAND_MAX_SIZE];
 555
 556             coefs1 = s->coefs1[ch];
 557             exponents = s->exponents[ch];
 558             esize = s->exponents_bsize[ch];
 559             mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
 560             mult *= mdct_norm;
 561             coefs = s->coefs[ch];
 562             if (s->use_noise_coding) {
 563                 mult1 = mult;
 564                 /* very low freqs : noise */
 565                 for(i = 0;i < s->coefs_start; i++) {
 566                     *coefs++ = s->noise_table[s->noise_index] *
 567                       exponents[i<<bsize>>esize] * mult1;
 568                     s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 569                 }
 570
 571                 n1 = s->exponent_high_sizes[bsize];
 572
 573                 /* compute power of high bands */
 574                 exponents = s->exponents[ch] +
 575                     (s->high_band_start[bsize]<<bsize);
 576                 last_high_band = 0; /* avoid warning */
 577                 for(j=0;j<n1;j++) {
 578                     n = s->exponent_high_bands[s->frame_len_bits -
 579                                               s->block_len_bits][j];
 580                     if (s->high_band_coded[ch][j]) {
 581                         float e2, v;
 582                         e2 = 0;
 583                         for(i = 0;i < n; i++) {
 584                             v = exponents[i<<bsize>>esize];
 585                             e2 += v * v;
 586                         }
 587                         exp_power[j] = e2 / n;
 588                         last_high_band = j;
 589                         tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n);
 590                     }
 591                     exponents += n<<bsize;
 592                 }
 593
 594                 /* main freqs and high freqs */
 595                 exponents = s->exponents[ch] + (s->coefs_start<<bsize);
 596                 for(j=-1;j<n1;j++) {
 597                     if (j < 0) {
 598                         n = s->high_band_start[bsize] -
 599                             s->coefs_start;
 600                     } else {
 601                         n = s->exponent_high_bands[s->frame_len_bits -
 602                                                   s->block_len_bits][j];
 603                     }
 604                     if (j >= 0 && s->high_band_coded[ch][j]) {
 605                         /* use noise with specified power */
 606                         mult1 = sqrt(exp_power[j] / exp_power[last_high_band]);
 607                         /* XXX: use a table */
 608                         mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05);
 609                         mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult);
 610                         mult1 *= mdct_norm;
 611                         for(i = 0;i < n; i++) {
 612                             noise = s->noise_table[s->noise_index];
 613                             s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 614                             *coefs++ =  noise *
 615                                 exponents[i<<bsize>>esize] * mult1;
 616                         }
 617                         exponents += n<<bsize;
 618                     } else {
 619                         /* coded values + small noise */
 620                         for(i = 0;i < n; i++) {
 621                             noise = s->noise_table[s->noise_index];
 622                             s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 623                             *coefs++ = ((*coefs1++) + noise) *
 624                                 exponents[i<<bsize>>esize] * mult;
 625                         }
 626                         exponents += n<<bsize;
 627                     }
 628                 }
 629
 630                 /* very high freqs : noise */
 631                 n = s->block_len - s->coefs_end[bsize];
 632                 mult1 = mult * exponents[((-1<<bsize))>>esize];
 633                 for(i = 0; i < n; i++) {
 634                     *coefs++ = s->noise_table[s->noise_index] * mult1;
 635                     s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 636                 }
 637             } else {
 638                 /* XXX: optimize more */
 639                 for(i = 0;i < s->coefs_start; i++)
 640                     *coefs++ = 0.0;
 641                 n = nb_coefs[ch];
 642                 for(i = 0;i < n; i++) {
 643                     *coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult;
 644                 }
 645                 n = s->block_len - s->coefs_end[bsize];
 646                 for(i = 0;i < n; i++)
 647                     *coefs++ = 0.0;
 648             }
 649         }
 650     }
 651
 652 #ifdef TRACE
 653     for(ch = 0; ch < s->nb_channels; ch++) {
 654         if (s->channel_coded[ch]) {
 655             dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len);
 656             dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len);
 657         }
 658     }
 659 #endif
 660
 661     if (s->ms_stereo && s->channel_coded[1]) {
 662         float a, b;
 663         int i;
 664
 665         /* nominal case for ms stereo: we do it before mdct */
 666         /* no need to optimize this case because it should almost
 667            never happen */
 668         if (!s->channel_coded[0]) {
 669             tprintf(s->avctx, "rare ms-stereo case happened\n");
 670             memset(s->coefs[0], 0, sizeof(float) * s->block_len);
 671             s->channel_coded[0] = 1;
 672         }
 673
 674         for(i = 0; i < s->block_len; i++) {
 675             a = s->coefs[0][i];
 676             b = s->coefs[1][i];
 677             s->coefs[0][i] = a + b;
 678             s->coefs[1][i] = a - b;
 679         }
 680     }
 681
 682     for(ch = 0; ch < s->nb_channels; ch++) {
 683         if (s->channel_coded[ch]) {
 684             int n4, index, n;
 685
 686             n = s->block_len;
 687             n4 = s->block_len / 2;
 688             s->mdct_ctx[bsize].fft.imdct_calc(&s->mdct_ctx[bsize],
 689                           s->output, s->coefs[ch], s->mdct_tmp);
 690
 691             /* multiply by the window and add in the frame */
 692             index = (s->frame_len / 2) + s->block_pos - n4;
 693             wma_window(s, &s->frame_out[ch][index]);
 694
 695             /* specific fast case for ms-stereo : add to second
 696                channel if it is not coded */
 697             if (s->ms_stereo && !s->channel_coded[1]) {
 698                 wma_window(s, &s->frame_out[1][index]);
 699             }
 700         }
 701     }
 702  next:
 703     /* update block number */
 704     s->block_num++;
 705     s->block_pos += s->block_len;
 706     if (s->block_pos >= s->frame_len)
 707         return 1;
 708     else
 709         return 0;
 710 }
 711
 712 /* decode a frame of frame_len samples */
 713 static int wma_decode_frame(WMACodecContext *s, int16_t *samples)
 714 {
 715     int ret, i, n, ch, incr;
 716     int16_t *ptr;
 717     float *iptr;
 718
 719 #ifdef TRACE
 720     tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len);
 721 #endif
 722
 723     /* read each block */
 724     s->block_num = 0;
 725     s->block_pos = 0;
 726     for(;;) {
 727         ret = wma_decode_block(s);
 728         if (ret < 0)
 729             return -1;
 730         if (ret)
 731             break;
 732     }
 733
 734     /* convert frame to integer */
 735     n = s->frame_len;
 736     incr = s->nb_channels;
 737     for(ch = 0; ch < s->nb_channels; ch++) {
 738         ptr = samples + ch;
 739         iptr = s->frame_out[ch];
 740
 741         for(i=0;i<n;i++) {
 742             *ptr = av_clip_int16(lrintf(*iptr++));
 743             ptr += incr;
 744         }
 745         /* prepare for next block */
 746         memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len],
 747                 s->frame_len * sizeof(float));
 748     }
 749
 750 #ifdef TRACE
 751     dump_shorts(s, "samples", samples, n * s->nb_channels);
 752 #endif
 753     return 0;
 754 }
 755
 756 static int wma_decode_superframe(AVCodecContext *avctx,
 757                                  void *data, int *data_size,
 758                                  const uint8_t *buf, int buf_size)
 759 {
 760     WMACodecContext *s = avctx->priv_data;
 761     int nb_frames, bit_offset, i, pos, len;
 762     uint8_t *q;
 763     int16_t *samples;
 764
 765     tprintf(avctx, "***decode_superframe:\n");
 766
 767     if(buf_size==0){
 768         s->last_superframe_len = 0;
 769         return 0;
 770     }
 771     if (buf_size < s->block_align)
 772         return 0;
 773     buf_size = s->block_align;
 774
 775     samples = data;
 776
 777     init_get_bits(&s->gb, buf, buf_size*8);
 778
 779     if (s->use_bit_reservoir) {
 780         /* read super frame header */
 781         skip_bits(&s->gb, 4); /* super frame index */
 782         nb_frames = get_bits(&s->gb, 4) - 1;
 783
 784         bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
 785
 786         if (s->last_superframe_len > 0) {
 787             //        printf("skip=%d\n", s->last_bitoffset);
 788             /* add bit_offset bits to last frame */
 789             if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) >
 790                 MAX_CODED_SUPERFRAME_SIZE)
 791                 goto fail;
 792             q = s->last_superframe + s->last_superframe_len;
 793             len = bit_offset;
 794             while (len > 7) {
 795                 *q++ = (get_bits)(&s->gb, 8);
 796                 len -= 8;
 797             }
 798             if (len > 0) {
 799                 *q++ = (get_bits)(&s->gb, len) << (8 - len);
 800             }
 801
 802             /* XXX: bit_offset bits into last frame */
 803             init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8);
 804             /* skip unused bits */
 805             if (s->last_bitoffset > 0)
 806                 skip_bits(&s->gb, s->last_bitoffset);
 807             /* this frame is stored in the last superframe and in the
 808                current one */
 809             if (wma_decode_frame(s, samples) < 0)
 810                 goto fail;
 811             samples += s->nb_channels * s->frame_len;
 812         }
 813
 814         /* read each frame starting from bit_offset */
 815         pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
 816         init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8);
 817         len = pos & 7;
 818         if (len > 0)
 819             skip_bits(&s->gb, len);
 820
 821         s->reset_block_lengths = 1;
 822         for(i=0;i<nb_frames;i++) {
 823             if (wma_decode_frame(s, samples) < 0)
 824                 goto fail;
 825             samples += s->nb_channels * s->frame_len;
 826         }
 827
 828         /* we copy the end of the frame in the last frame buffer */
 829         pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
 830         s->last_bitoffset = pos & 7;
 831         pos >>= 3;
 832         len = buf_size - pos;
 833         if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) {
 834             goto fail;
 835         }
 836         s->last_superframe_len = len;
 837         memcpy(s->last_superframe, buf + pos, len);
 838     } else {
 839         /* single frame decode */
 840         if (wma_decode_frame(s, samples) < 0)
 841             goto fail;
 842         samples += s->nb_channels * s->frame_len;
 843     }
 844
 845 //av_log(NULL, AV_LOG_ERROR, "%d %d %d %d outbytes:%d eaten:%d\n", s->frame_len_bits, s->block_len_bits, s->frame_len, s->block_len,        (int8_t *)samples - (int8_t *)data, s->block_align);
 846
 847     *data_size = (int8_t *)samples - (int8_t *)data;
 848     return s->block_align;
 849  fail:
 850     /* when error, we reset the bit reservoir */
 851     s->last_superframe_len = 0;
 852     return -1;
 853 }
 854
 855 AVCodec wmav1_decoder =
 856 {
 857     "wmav1",
 858     CODEC_TYPE_AUDIO,
 859     CODEC_ID_WMAV1,
 860     sizeof(WMACodecContext),
 861     wma_decode_init,
 862     NULL,
 863     ff_wma_end,
 864     wma_decode_superframe,
 865 };
 866
 867 AVCodec wmav2_decoder =
 868 {
 869     "wmav2",
 870     CODEC_TYPE_AUDIO,
 871     CODEC_ID_WMAV2,
 872     sizeof(WMACodecContext),
 873     wma_decode_init,
 874     NULL,
 875     ff_wma_end,
 876     wma_decode_superframe,
 877 };