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 libavcodec/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, 1.0);
 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     avctx->sample_fmt = SAMPLE_FMT_S16;
 130     return 0;
 131 }
 132
 133 /**
 134  * compute x^-0.25 with an exponent and mantissa table. We use linear
 135  * interpolation to reduce the mantissa table size at a small speed
 136  * expense (linear interpolation approximately doubles the number of
 137  * bits of precision).
 138  */
 139 static inline float pow_m1_4(WMACodecContext *s, float x)
 140 {
 141     union {
 142         float f;
 143         unsigned int v;
 144     } u, t;
 145     unsigned int e, m;
 146     float a, b;
 147
 148     u.f = x;
 149     e = u.v >> 23;
 150     m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1);
 151     /* build interpolation scale: 1 <= t < 2. */
 152     t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23);
 153     a = s->lsp_pow_m_table1[m];
 154     b = s->lsp_pow_m_table2[m];
 155     return s->lsp_pow_e_table[e] * (a + b * t.f);
 156 }
 157
 158 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
 159 {
 160     float wdel, a, b;
 161     int i, e, m;
 162
 163     wdel = M_PI / frame_len;
 164     for(i=0;i<frame_len;i++)
 165         s->lsp_cos_table[i] = 2.0f * cos(wdel * i);
 166
 167     /* tables for x^-0.25 computation */
 168     for(i=0;i<256;i++) {
 169         e = i - 126;
 170         s->lsp_pow_e_table[i] = pow(2.0, e * -0.25);
 171     }
 172
 173     /* NOTE: these two tables are needed to avoid two operations in
 174        pow_m1_4 */
 175     b = 1.0;
 176     for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) {
 177         m = (1 << LSP_POW_BITS) + i;
 178         a = (float)m * (0.5 / (1 << LSP_POW_BITS));
 179         a = pow(a, -0.25);
 180         s->lsp_pow_m_table1[i] = 2 * a - b;
 181         s->lsp_pow_m_table2[i] = b - a;
 182         b = a;
 183     }
 184 #if 0
 185     for(i=1;i<20;i++) {
 186         float v, r1, r2;
 187         v = 5.0 / i;
 188         r1 = pow_m1_4(s, v);
 189         r2 = pow(v,-0.25);
 190         printf("%f^-0.25=%f e=%f\n", v, r1, r2 - r1);
 191     }
 192 #endif
 193 }
 194
 195 /**
 196  * NOTE: We use the same code as Vorbis here
 197  * @todo optimize it further with SSE/3Dnow
 198  */
 199 static void wma_lsp_to_curve(WMACodecContext *s,
 200                              float *out, float *val_max_ptr,
 201                              int n, float *lsp)
 202 {
 203     int i, j;
 204     float p, q, w, v, val_max;
 205
 206     val_max = 0;
 207     for(i=0;i<n;i++) {
 208         p = 0.5f;
 209         q = 0.5f;
 210         w = s->lsp_cos_table[i];
 211         for(j=1;j<NB_LSP_COEFS;j+=2){
 212             q *= w - lsp[j - 1];
 213             p *= w - lsp[j];
 214         }
 215         p *= p * (2.0f - w);
 216         q *= q * (2.0f + w);
 217         v = p + q;
 218         v = pow_m1_4(s, v);
 219         if (v > val_max)
 220             val_max = v;
 221         out[i] = v;
 222     }
 223     *val_max_ptr = val_max;
 224 }
 225
 226 /**
 227  * decode exponents coded with LSP coefficients (same idea as Vorbis)
 228  */
 229 static void decode_exp_lsp(WMACodecContext *s, int ch)
 230 {
 231     float lsp_coefs[NB_LSP_COEFS];
 232     int val, i;
 233
 234     for(i = 0; i < NB_LSP_COEFS; i++) {
 235         if (i == 0 || i >= 8)
 236             val = get_bits(&s->gb, 3);
 237         else
 238             val = get_bits(&s->gb, 4);
 239         lsp_coefs[i] = ff_wma_lsp_codebook[i][val];
 240     }
 241
 242     wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch],
 243                      s->block_len, lsp_coefs);
 244 }
 245
 246 /**
 247  * decode exponents coded with VLC codes
 248  */
 249 static int decode_exp_vlc(WMACodecContext *s, int ch)
 250 {
 251     int last_exp, n, code;
 252     const uint16_t *ptr, *band_ptr;
 253     float v, *q, max_scale, *q_end;
 254
 255     band_ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
 256     ptr = band_ptr;
 257     q = s->exponents[ch];
 258     q_end = q + s->block_len;
 259     max_scale = 0;
 260     if (s->version == 1) {
 261         last_exp = get_bits(&s->gb, 5) + 10;
 262         /* XXX: use a table */
 263         v = pow(10, last_exp * (1.0 / 16.0));
 264         max_scale = v;
 265         n = *ptr++;
 266         do {
 267             *q++ = v;
 268         } while (--n);
 269     }else
 270         last_exp = 36;
 271
 272     while (q < q_end) {
 273         code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX);
 274         if (code < 0)
 275             return -1;
 276         /* NOTE: this offset is the same as MPEG4 AAC ! */
 277         last_exp += code - 60;
 278         /* XXX: use a table */
 279         v = pow(10, last_exp * (1.0 / 16.0));
 280         if (v > max_scale)
 281             max_scale = v;
 282         n = *ptr++;
 283         do {
 284             *q++ = v;
 285         } while (--n);
 286     }
 287     s->max_exponent[ch] = max_scale;
 288     return 0;
 289 }
 290
 291
 292 /**
 293  * Apply MDCT window and add into output.
 294  *
 295  * We ensure that when the windows overlap their squared sum
 296  * is always 1 (MDCT reconstruction rule).
 297  */
 298 static void wma_window(WMACodecContext *s, float *out)
 299 {
 300     float *in = s->output;
 301     int block_len, bsize, n;
 302
 303     /* left part */
 304     if (s->block_len_bits <= s->prev_block_len_bits) {
 305         block_len = s->block_len;
 306         bsize = s->frame_len_bits - s->block_len_bits;
 307
 308         s->dsp.vector_fmul_add_add(out, in, s->windows[bsize],
 309                                    out, 0, block_len, 1);
 310
 311     } else {
 312         block_len = 1 << s->prev_block_len_bits;
 313         n = (s->block_len - block_len) / 2;
 314         bsize = s->frame_len_bits - s->prev_block_len_bits;
 315
 316         s->dsp.vector_fmul_add_add(out+n, in+n, s->windows[bsize],
 317                                    out+n, 0, block_len, 1);
 318
 319         memcpy(out+n+block_len, in+n+block_len, n*sizeof(float));
 320     }
 321
 322     out += s->block_len;
 323     in += s->block_len;
 324
 325     /* right part */
 326     if (s->block_len_bits <= s->next_block_len_bits) {
 327         block_len = s->block_len;
 328         bsize = s->frame_len_bits - s->block_len_bits;
 329
 330         s->dsp.vector_fmul_reverse(out, in, s->windows[bsize], block_len);
 331
 332     } else {
 333         block_len = 1 << s->next_block_len_bits;
 334         n = (s->block_len - block_len) / 2;
 335         bsize = s->frame_len_bits - s->next_block_len_bits;
 336
 337         memcpy(out, in, n*sizeof(float));
 338
 339         s->dsp.vector_fmul_reverse(out+n, in+n, s->windows[bsize], block_len);
 340
 341         memset(out+n+block_len, 0, n*sizeof(float));
 342     }
 343 }
 344
 345
 346 /**
 347  * @return 0 if OK. 1 if last block of frame. return -1 if
 348  * unrecorrable error.
 349  */
 350 static int wma_decode_block(WMACodecContext *s)
 351 {
 352     int n, v, a, ch, bsize;
 353     int coef_nb_bits, total_gain;
 354     int nb_coefs[MAX_CHANNELS];
 355     float mdct_norm;
 356
 357 #ifdef TRACE
 358     tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num);
 359 #endif
 360
 361     /* compute current block length */
 362     if (s->use_variable_block_len) {
 363         n = av_log2(s->nb_block_sizes - 1) + 1;
 364
 365         if (s->reset_block_lengths) {
 366             s->reset_block_lengths = 0;
 367             v = get_bits(&s->gb, n);
 368             if (v >= s->nb_block_sizes)
 369                 return -1;
 370             s->prev_block_len_bits = s->frame_len_bits - v;
 371             v = get_bits(&s->gb, n);
 372             if (v >= s->nb_block_sizes)
 373                 return -1;
 374             s->block_len_bits = s->frame_len_bits - v;
 375         } else {
 376             /* update block lengths */
 377             s->prev_block_len_bits = s->block_len_bits;
 378             s->block_len_bits = s->next_block_len_bits;
 379         }
 380         v = get_bits(&s->gb, n);
 381         if (v >= s->nb_block_sizes)
 382             return -1;
 383         s->next_block_len_bits = s->frame_len_bits - v;
 384     } else {
 385         /* fixed block len */
 386         s->next_block_len_bits = s->frame_len_bits;
 387         s->prev_block_len_bits = s->frame_len_bits;
 388         s->block_len_bits = s->frame_len_bits;
 389     }
 390
 391     /* now check if the block length is coherent with the frame length */
 392     s->block_len = 1 << s->block_len_bits;
 393     if ((s->block_pos + s->block_len) > s->frame_len)
 394         return -1;
 395
 396     if (s->nb_channels == 2) {
 397         s->ms_stereo = get_bits1(&s->gb);
 398     }
 399     v = 0;
 400     for(ch = 0; ch < s->nb_channels; ch++) {
 401         a = get_bits1(&s->gb);
 402         s->channel_coded[ch] = a;
 403         v |= a;
 404     }
 405
 406     bsize = s->frame_len_bits - s->block_len_bits;
 407
 408     /* if no channel coded, no need to go further */
 409     /* XXX: fix potential framing problems */
 410     if (!v)
 411         goto next;
 412
 413     /* read total gain and extract corresponding number of bits for
 414        coef escape coding */
 415     total_gain = 1;
 416     for(;;) {
 417         a = get_bits(&s->gb, 7);
 418         total_gain += a;
 419         if (a != 127)
 420             break;
 421     }
 422
 423     coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
 424
 425     /* compute number of coefficients */
 426     n = s->coefs_end[bsize] - s->coefs_start;
 427     for(ch = 0; ch < s->nb_channels; ch++)
 428         nb_coefs[ch] = n;
 429
 430     /* complex coding */
 431     if (s->use_noise_coding) {
 432
 433         for(ch = 0; ch < s->nb_channels; ch++) {
 434             if (s->channel_coded[ch]) {
 435                 int i, n, a;
 436                 n = s->exponent_high_sizes[bsize];
 437                 for(i=0;i<n;i++) {
 438                     a = get_bits1(&s->gb);
 439                     s->high_band_coded[ch][i] = a;
 440                     /* if noise coding, the coefficients are not transmitted */
 441                     if (a)
 442                         nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
 443                 }
 444             }
 445         }
 446         for(ch = 0; ch < s->nb_channels; ch++) {
 447             if (s->channel_coded[ch]) {
 448                 int i, n, val, code;
 449
 450                 n = s->exponent_high_sizes[bsize];
 451                 val = (int)0x80000000;
 452                 for(i=0;i<n;i++) {
 453                     if (s->high_band_coded[ch][i]) {
 454                         if (val == (int)0x80000000) {
 455                             val = get_bits(&s->gb, 7) - 19;
 456                         } else {
 457                             code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX);
 458                             if (code < 0)
 459                                 return -1;
 460                             val += code - 18;
 461                         }
 462                         s->high_band_values[ch][i] = val;
 463                     }
 464                 }
 465             }
 466         }
 467     }
 468
 469     /* exponents can be reused in short blocks. */
 470     if ((s->block_len_bits == s->frame_len_bits) ||
 471         get_bits1(&s->gb)) {
 472         for(ch = 0; ch < s->nb_channels; ch++) {
 473             if (s->channel_coded[ch]) {
 474                 if (s->use_exp_vlc) {
 475                     if (decode_exp_vlc(s, ch) < 0)
 476                         return -1;
 477                 } else {
 478                     decode_exp_lsp(s, ch);
 479                 }
 480                 s->exponents_bsize[ch] = bsize;
 481             }
 482         }
 483     }
 484
 485     /* parse spectral coefficients : just RLE encoding */
 486     for(ch = 0; ch < s->nb_channels; ch++) {
 487         if (s->channel_coded[ch]) {
 488             int tindex;
 489             WMACoef* ptr = &s->coefs1[ch][0];
 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             memset(ptr, 0, s->block_len * sizeof(WMACoef));
 495             ff_wma_run_level_decode(s->avctx, &s->gb, &s->coef_vlc[tindex],
 496                   s->level_table[tindex], s->run_table[tindex],
 497                   0, ptr, 0, nb_coefs[ch],
 498                   s->block_len, s->frame_len_bits, coef_nb_bits);
 499         }
 500         if (s->version == 1 && s->nb_channels >= 2) {
 501             align_get_bits(&s->gb);
 502         }
 503     }
 504
 505     /* normalize */
 506     {
 507         int n4 = s->block_len / 2;
 508         mdct_norm = 1.0 / (float)n4;
 509         if (s->version == 1) {
 510             mdct_norm *= sqrt(n4);
 511         }
 512     }
 513
 514     /* finally compute the MDCT coefficients */
 515     for(ch = 0; ch < s->nb_channels; ch++) {
 516         if (s->channel_coded[ch]) {
 517             WMACoef *coefs1;
 518             float *coefs, *exponents, mult, mult1, noise;
 519             int i, j, n, n1, last_high_band, esize;
 520             float exp_power[HIGH_BAND_MAX_SIZE];
 521
 522             coefs1 = s->coefs1[ch];
 523             exponents = s->exponents[ch];
 524             esize = s->exponents_bsize[ch];
 525             mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
 526             mult *= mdct_norm;
 527             coefs = s->coefs[ch];
 528             if (s->use_noise_coding) {
 529                 mult1 = mult;
 530                 /* very low freqs : noise */
 531                 for(i = 0;i < s->coefs_start; i++) {
 532                     *coefs++ = s->noise_table[s->noise_index] *
 533                       exponents[i<<bsize>>esize] * mult1;
 534                     s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 535                 }
 536
 537                 n1 = s->exponent_high_sizes[bsize];
 538
 539                 /* compute power of high bands */
 540                 exponents = s->exponents[ch] +
 541                     (s->high_band_start[bsize]<<bsize);
 542                 last_high_band = 0; /* avoid warning */
 543                 for(j=0;j<n1;j++) {
 544                     n = s->exponent_high_bands[s->frame_len_bits -
 545                                               s->block_len_bits][j];
 546                     if (s->high_band_coded[ch][j]) {
 547                         float e2, v;
 548                         e2 = 0;
 549                         for(i = 0;i < n; i++) {
 550                             v = exponents[i<<bsize>>esize];
 551                             e2 += v * v;
 552                         }
 553                         exp_power[j] = e2 / n;
 554                         last_high_band = j;
 555                         tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n);
 556                     }
 557                     exponents += n<<bsize;
 558                 }
 559
 560                 /* main freqs and high freqs */
 561                 exponents = s->exponents[ch] + (s->coefs_start<<bsize);
 562                 for(j=-1;j<n1;j++) {
 563                     if (j < 0) {
 564                         n = s->high_band_start[bsize] -
 565                             s->coefs_start;
 566                     } else {
 567                         n = s->exponent_high_bands[s->frame_len_bits -
 568                                                   s->block_len_bits][j];
 569                     }
 570                     if (j >= 0 && s->high_band_coded[ch][j]) {
 571                         /* use noise with specified power */
 572                         mult1 = sqrt(exp_power[j] / exp_power[last_high_band]);
 573                         /* XXX: use a table */
 574                         mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05);
 575                         mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult);
 576                         mult1 *= mdct_norm;
 577                         for(i = 0;i < n; i++) {
 578                             noise = s->noise_table[s->noise_index];
 579                             s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 580                             *coefs++ =  noise *
 581                                 exponents[i<<bsize>>esize] * mult1;
 582                         }
 583                         exponents += n<<bsize;
 584                     } else {
 585                         /* coded values + small noise */
 586                         for(i = 0;i < n; i++) {
 587                             noise = s->noise_table[s->noise_index];
 588                             s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 589                             *coefs++ = ((*coefs1++) + noise) *
 590                                 exponents[i<<bsize>>esize] * mult;
 591                         }
 592                         exponents += n<<bsize;
 593                     }
 594                 }
 595
 596                 /* very high freqs : noise */
 597                 n = s->block_len - s->coefs_end[bsize];
 598                 mult1 = mult * exponents[((-1<<bsize))>>esize];
 599                 for(i = 0; i < n; i++) {
 600                     *coefs++ = s->noise_table[s->noise_index] * mult1;
 601                     s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 602                 }
 603             } else {
 604                 /* XXX: optimize more */
 605                 for(i = 0;i < s->coefs_start; i++)
 606                     *coefs++ = 0.0;
 607                 n = nb_coefs[ch];
 608                 for(i = 0;i < n; i++) {
 609                     *coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult;
 610                 }
 611                 n = s->block_len - s->coefs_end[bsize];
 612                 for(i = 0;i < n; i++)
 613                     *coefs++ = 0.0;
 614             }
 615         }
 616     }
 617
 618 #ifdef TRACE
 619     for(ch = 0; ch < s->nb_channels; ch++) {
 620         if (s->channel_coded[ch]) {
 621             dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len);
 622             dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len);
 623         }
 624     }
 625 #endif
 626
 627     if (s->ms_stereo && s->channel_coded[1]) {
 628         float a, b;
 629         int i;
 630
 631         /* nominal case for ms stereo: we do it before mdct */
 632         /* no need to optimize this case because it should almost
 633            never happen */
 634         if (!s->channel_coded[0]) {
 635             tprintf(s->avctx, "rare ms-stereo case happened\n");
 636             memset(s->coefs[0], 0, sizeof(float) * s->block_len);
 637             s->channel_coded[0] = 1;
 638         }
 639
 640         for(i = 0; i < s->block_len; i++) {
 641             a = s->coefs[0][i];
 642             b = s->coefs[1][i];
 643             s->coefs[0][i] = a + b;
 644             s->coefs[1][i] = a - b;
 645         }
 646     }
 647
 648 next:
 649     for(ch = 0; ch < s->nb_channels; ch++) {
 650         int n4, index;
 651
 652         n4 = s->block_len / 2;
 653         if(s->channel_coded[ch]){
 654             ff_imdct_calc(&s->mdct_ctx[bsize], s->output, s->coefs[ch]);
 655         }else if(!(s->ms_stereo && ch==1))
 656             memset(s->output, 0, sizeof(s->output));
 657
 658         /* multiply by the window and add in the frame */
 659         index = (s->frame_len / 2) + s->block_pos - n4;
 660         wma_window(s, &s->frame_out[ch][index]);
 661     }
 662
 663     /* update block number */
 664     s->block_num++;
 665     s->block_pos += s->block_len;
 666     if (s->block_pos >= s->frame_len)
 667         return 1;
 668     else
 669         return 0;
 670 }
 671
 672 /* decode a frame of frame_len samples */
 673 static int wma_decode_frame(WMACodecContext *s, int16_t *samples)
 674 {
 675     int ret, i, n, ch, incr;
 676     int16_t *ptr;
 677     float *iptr;
 678
 679 #ifdef TRACE
 680     tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len);
 681 #endif
 682
 683     /* read each block */
 684     s->block_num = 0;
 685     s->block_pos = 0;
 686     for(;;) {
 687         ret = wma_decode_block(s);
 688         if (ret < 0)
 689             return -1;
 690         if (ret)
 691             break;
 692     }
 693
 694     /* convert frame to integer */
 695     n = s->frame_len;
 696     incr = s->nb_channels;
 697     for(ch = 0; ch < s->nb_channels; ch++) {
 698         ptr = samples + ch;
 699         iptr = s->frame_out[ch];
 700
 701         for(i=0;i<n;i++) {
 702             *ptr = av_clip_int16(lrintf(*iptr++));
 703             ptr += incr;
 704         }
 705         /* prepare for next block */
 706         memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len],
 707                 s->frame_len * sizeof(float));
 708     }
 709
 710 #ifdef TRACE
 711     dump_shorts(s, "samples", samples, n * s->nb_channels);
 712 #endif
 713     return 0;
 714 }
 715
 716 static int wma_decode_superframe(AVCodecContext *avctx,
 717                                  void *data, int *data_size,
 718                                  AVPacket *avpkt)
 719 {
 720     const uint8_t *buf = avpkt->data;
 721     int buf_size = avpkt->size;
 722     WMACodecContext *s = avctx->priv_data;
 723     int nb_frames, bit_offset, i, pos, len;
 724     uint8_t *q;
 725     int16_t *samples;
 726
 727     tprintf(avctx, "***decode_superframe:\n");
 728
 729     if(buf_size==0){
 730         s->last_superframe_len = 0;
 731         return 0;
 732     }
 733     if (buf_size < s->block_align)
 734         return 0;
 735     buf_size = s->block_align;
 736
 737     samples = data;
 738
 739     init_get_bits(&s->gb, buf, buf_size*8);
 740
 741     if (s->use_bit_reservoir) {
 742         /* read super frame header */
 743         skip_bits(&s->gb, 4); /* super frame index */
 744         nb_frames = get_bits(&s->gb, 4) - 1;
 745
 746         if((nb_frames+1) * s->nb_channels * s->frame_len * sizeof(int16_t) > *data_size){
 747             av_log(s->avctx, AV_LOG_ERROR, "Insufficient output space\n");
 748             goto fail;
 749         }
 750
 751         bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
 752
 753         if (s->last_superframe_len > 0) {
 754             //        printf("skip=%d\n", s->last_bitoffset);
 755             /* add bit_offset bits to last frame */
 756             if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) >
 757                 MAX_CODED_SUPERFRAME_SIZE)
 758                 goto fail;
 759             q = s->last_superframe + s->last_superframe_len;
 760             len = bit_offset;
 761             while (len > 7) {
 762                 *q++ = (get_bits)(&s->gb, 8);
 763                 len -= 8;
 764             }
 765             if (len > 0) {
 766                 *q++ = (get_bits)(&s->gb, len) << (8 - len);
 767             }
 768
 769             /* XXX: bit_offset bits into last frame */
 770             init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8);
 771             /* skip unused bits */
 772             if (s->last_bitoffset > 0)
 773                 skip_bits(&s->gb, s->last_bitoffset);
 774             /* this frame is stored in the last superframe and in the
 775                current one */
 776             if (wma_decode_frame(s, samples) < 0)
 777                 goto fail;
 778             samples += s->nb_channels * s->frame_len;
 779         }
 780
 781         /* read each frame starting from bit_offset */
 782         pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
 783         init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8);
 784         len = pos & 7;
 785         if (len > 0)
 786             skip_bits(&s->gb, len);
 787
 788         s->reset_block_lengths = 1;
 789         for(i=0;i<nb_frames;i++) {
 790             if (wma_decode_frame(s, samples) < 0)
 791                 goto fail;
 792             samples += s->nb_channels * s->frame_len;
 793         }
 794
 795         /* we copy the end of the frame in the last frame buffer */
 796         pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
 797         s->last_bitoffset = pos & 7;
 798         pos >>= 3;
 799         len = buf_size - pos;
 800         if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) {
 801             goto fail;
 802         }
 803         s->last_superframe_len = len;
 804         memcpy(s->last_superframe, buf + pos, len);
 805     } else {
 806         if(s->nb_channels * s->frame_len * sizeof(int16_t) > *data_size){
 807             av_log(s->avctx, AV_LOG_ERROR, "Insufficient output space\n");
 808             goto fail;
 809         }
 810         /* single frame decode */
 811         if (wma_decode_frame(s, samples) < 0)
 812             goto fail;
 813         samples += s->nb_channels * s->frame_len;
 814     }
 815
 816 //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);
 817
 818     *data_size = (int8_t *)samples - (int8_t *)data;
 819     return s->block_align;
 820  fail:
 821     /* when error, we reset the bit reservoir */
 822     s->last_superframe_len = 0;
 823     return -1;
 824 }
 825
 826 AVCodec wmav1_decoder =
 827 {
 828     "wmav1",
 829     CODEC_TYPE_AUDIO,
 830     CODEC_ID_WMAV1,
 831     sizeof(WMACodecContext),
 832     wma_decode_init,
 833     NULL,
 834     ff_wma_end,
 835     wma_decode_superframe,
 836     .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
 837 };
 838
 839 AVCodec wmav2_decoder =
 840 {
 841     "wmav2",
 842     CODEC_TYPE_AUDIO,
 843     CODEC_ID_WMAV2,
 844     sizeof(WMACodecContext),
 845     wma_decode_init,
 846     NULL,
 847     ff_wma_end,
 848     wma_decode_superframe,
 849     .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),
 850 };