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, flags2;
  86     uint8_t *extradata;
  87
  88     s->avctx = avctx;
  89
  90     /* extract flag infos */
  91     flags2 = 0;
  92     extradata = avctx->extradata;
  93     if (avctx->codec->id == CODEC_ID_WMAV1 && avctx->extradata_size >= 4) {
  94         flags2 = AV_RL16(extradata+2);
  95     } else if (avctx->codec->id == CODEC_ID_WMAV2 && avctx->extradata_size >= 6) {
  96         flags2 = AV_RL16(extradata+4);
  97     }
  98 // for(i=0; i<avctx->extradata_size; i++)
  99 //     av_log(NULL, AV_LOG_ERROR, "%02X ", extradata[i]);
 100
 101     s->use_exp_vlc = flags2 & 0x0001;
 102     s->use_bit_reservoir = flags2 & 0x0002;
 103     s->use_variable_block_len = flags2 & 0x0004;
 104
 105     if(ff_wma_init(avctx, flags2)<0)
 106         return -1;
 107
 108     /* init MDCT */
 109     for(i = 0; i < s->nb_block_sizes; i++)
 110         ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1, 1.0);
 111
 112     if (s->use_noise_coding) {
 113         init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(ff_wma_hgain_huffbits),
 114                  ff_wma_hgain_huffbits, 1, 1,
 115                  ff_wma_hgain_huffcodes, 2, 2, 0);
 116     }
 117
 118     if (s->use_exp_vlc) {
 119         init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_wma_scale_huffbits), //FIXME move out of context
 120                  ff_wma_scale_huffbits, 1, 1,
 121                  ff_wma_scale_huffcodes, 4, 4, 0);
 122     } else {
 123         wma_lsp_to_curve_init(s, s->frame_len);
 124     }
 125
 126     avctx->sample_fmt = SAMPLE_FMT_S16;
 127     return 0;
 128 }
 129
 130 /**
 131  * compute x^-0.25 with an exponent and mantissa table. We use linear
 132  * interpolation to reduce the mantissa table size at a small speed
 133  * expense (linear interpolation approximately doubles the number of
 134  * bits of precision).
 135  */
 136 static inline float pow_m1_4(WMACodecContext *s, float x)
 137 {
 138     union {
 139         float f;
 140         unsigned int v;
 141     } u, t;
 142     unsigned int e, m;
 143     float a, b;
 144
 145     u.f = x;
 146     e = u.v >> 23;
 147     m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1);
 148     /* build interpolation scale: 1 <= t < 2. */
 149     t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23);
 150     a = s->lsp_pow_m_table1[m];
 151     b = s->lsp_pow_m_table2[m];
 152     return s->lsp_pow_e_table[e] * (a + b * t.f);
 153 }
 154
 155 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
 156 {
 157     float wdel, a, b;
 158     int i, e, m;
 159
 160     wdel = M_PI / frame_len;
 161     for(i=0;i<frame_len;i++)
 162         s->lsp_cos_table[i] = 2.0f * cos(wdel * i);
 163
 164     /* tables for x^-0.25 computation */
 165     for(i=0;i<256;i++) {
 166         e = i - 126;
 167         s->lsp_pow_e_table[i] = pow(2.0, e * -0.25);
 168     }
 169
 170     /* NOTE: these two tables are needed to avoid two operations in
 171        pow_m1_4 */
 172     b = 1.0;
 173     for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) {
 174         m = (1 << LSP_POW_BITS) + i;
 175         a = (float)m * (0.5 / (1 << LSP_POW_BITS));
 176         a = pow(a, -0.25);
 177         s->lsp_pow_m_table1[i] = 2 * a - b;
 178         s->lsp_pow_m_table2[i] = b - a;
 179         b = a;
 180     }
 181 #if 0
 182     for(i=1;i<20;i++) {
 183         float v, r1, r2;
 184         v = 5.0 / i;
 185         r1 = pow_m1_4(s, v);
 186         r2 = pow(v,-0.25);
 187         printf("%f^-0.25=%f e=%f\n", v, r1, r2 - r1);
 188     }
 189 #endif
 190 }
 191
 192 /**
 193  * NOTE: We use the same code as Vorbis here
 194  * @todo optimize it further with SSE/3Dnow
 195  */
 196 static void wma_lsp_to_curve(WMACodecContext *s,
 197                              float *out, float *val_max_ptr,
 198                              int n, float *lsp)
 199 {
 200     int i, j;
 201     float p, q, w, v, val_max;
 202
 203     val_max = 0;
 204     for(i=0;i<n;i++) {
 205         p = 0.5f;
 206         q = 0.5f;
 207         w = s->lsp_cos_table[i];
 208         for(j=1;j<NB_LSP_COEFS;j+=2){
 209             q *= w - lsp[j - 1];
 210             p *= w - lsp[j];
 211         }
 212         p *= p * (2.0f - w);
 213         q *= q * (2.0f + w);
 214         v = p + q;
 215         v = pow_m1_4(s, v);
 216         if (v > val_max)
 217             val_max = v;
 218         out[i] = v;
 219     }
 220     *val_max_ptr = val_max;
 221 }
 222
 223 /**
 224  * decode exponents coded with LSP coefficients (same idea as Vorbis)
 225  */
 226 static void decode_exp_lsp(WMACodecContext *s, int ch)
 227 {
 228     float lsp_coefs[NB_LSP_COEFS];
 229     int val, i;
 230
 231     for(i = 0; i < NB_LSP_COEFS; i++) {
 232         if (i == 0 || i >= 8)
 233             val = get_bits(&s->gb, 3);
 234         else
 235             val = get_bits(&s->gb, 4);
 236         lsp_coefs[i] = ff_wma_lsp_codebook[i][val];
 237     }
 238
 239     wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch],
 240                      s->block_len, lsp_coefs);
 241 }
 242
 243 /** pow(10, i / 16.0) for i in -60..67 */
 244 static const float pow_tab[128] = {
 245     1.7782794100389e-04, 2.0535250264571e-04,
 246     2.3713737056617e-04, 2.7384196342644e-04,
 247     3.1622776601684e-04, 3.6517412725484e-04,
 248     4.2169650342858e-04, 4.8696752516586e-04,
 249     5.6234132519035e-04, 6.4938163157621e-04,
 250     7.4989420933246e-04, 8.6596432336006e-04,
 251     1.0000000000000e-03, 1.1547819846895e-03,
 252     1.3335214321633e-03, 1.5399265260595e-03,
 253     1.7782794100389e-03, 2.0535250264571e-03,
 254     2.3713737056617e-03, 2.7384196342644e-03,
 255     3.1622776601684e-03, 3.6517412725484e-03,
 256     4.2169650342858e-03, 4.8696752516586e-03,
 257     5.6234132519035e-03, 6.4938163157621e-03,
 258     7.4989420933246e-03, 8.6596432336006e-03,
 259     1.0000000000000e-02, 1.1547819846895e-02,
 260     1.3335214321633e-02, 1.5399265260595e-02,
 261     1.7782794100389e-02, 2.0535250264571e-02,
 262     2.3713737056617e-02, 2.7384196342644e-02,
 263     3.1622776601684e-02, 3.6517412725484e-02,
 264     4.2169650342858e-02, 4.8696752516586e-02,
 265     5.6234132519035e-02, 6.4938163157621e-02,
 266     7.4989420933246e-02, 8.6596432336007e-02,
 267     1.0000000000000e-01, 1.1547819846895e-01,
 268     1.3335214321633e-01, 1.5399265260595e-01,
 269     1.7782794100389e-01, 2.0535250264571e-01,
 270     2.3713737056617e-01, 2.7384196342644e-01,
 271     3.1622776601684e-01, 3.6517412725484e-01,
 272     4.2169650342858e-01, 4.8696752516586e-01,
 273     5.6234132519035e-01, 6.4938163157621e-01,
 274     7.4989420933246e-01, 8.6596432336007e-01,
 275     1.0000000000000e+00, 1.1547819846895e+00,
 276     1.3335214321633e+00, 1.5399265260595e+00,
 277     1.7782794100389e+00, 2.0535250264571e+00,
 278     2.3713737056617e+00, 2.7384196342644e+00,
 279     3.1622776601684e+00, 3.6517412725484e+00,
 280     4.2169650342858e+00, 4.8696752516586e+00,
 281     5.6234132519035e+00, 6.4938163157621e+00,
 282     7.4989420933246e+00, 8.6596432336007e+00,
 283     1.0000000000000e+01, 1.1547819846895e+01,
 284     1.3335214321633e+01, 1.5399265260595e+01,
 285     1.7782794100389e+01, 2.0535250264571e+01,
 286     2.3713737056617e+01, 2.7384196342644e+01,
 287     3.1622776601684e+01, 3.6517412725484e+01,
 288     4.2169650342858e+01, 4.8696752516586e+01,
 289     5.6234132519035e+01, 6.4938163157621e+01,
 290     7.4989420933246e+01, 8.6596432336007e+01,
 291     1.0000000000000e+02, 1.1547819846895e+02,
 292     1.3335214321633e+02, 1.5399265260595e+02,
 293     1.7782794100389e+02, 2.0535250264571e+02,
 294     2.3713737056617e+02, 2.7384196342644e+02,
 295     3.1622776601684e+02, 3.6517412725484e+02,
 296     4.2169650342858e+02, 4.8696752516586e+02,
 297     5.6234132519035e+02, 6.4938163157621e+02,
 298     7.4989420933246e+02, 8.6596432336007e+02,
 299     1.0000000000000e+03, 1.1547819846895e+03,
 300     1.3335214321633e+03, 1.5399265260595e+03,
 301     1.7782794100389e+03, 2.0535250264571e+03,
 302     2.3713737056617e+03, 2.7384196342644e+03,
 303     3.1622776601684e+03, 3.6517412725484e+03,
 304     4.2169650342858e+03, 4.8696752516586e+03,
 305     5.6234132519035e+03, 6.4938163157621e+03,
 306     7.4989420933246e+03, 8.6596432336007e+03,
 307     1.0000000000000e+04, 1.1547819846895e+04,
 308     1.3335214321633e+04, 1.5399265260595e+04,
 309 };
 310
 311 /**
 312  * decode exponents coded with VLC codes
 313  */
 314 static int decode_exp_vlc(WMACodecContext *s, int ch)
 315 {
 316     int last_exp, n, code;
 317     const uint16_t *ptr;
 318     float v, max_scale;
 319     uint32_t *q, *q_end, iv;
 320     const float *ptab = pow_tab + 60;
 321     const uint32_t *iptab = (const uint32_t*)ptab;
 322
 323     ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
 324     q = (uint32_t *)s->exponents[ch];
 325     q_end = q + s->block_len;
 326     max_scale = 0;
 327     if (s->version == 1) {
 328         last_exp = get_bits(&s->gb, 5) + 10;
 329         v = ptab[last_exp];
 330         iv = iptab[last_exp];
 331         max_scale = v;
 332         n = *ptr++;
 333         switch (n & 3) do {
 334         case 0: *q++ = iv;
 335         case 3: *q++ = iv;
 336         case 2: *q++ = iv;
 337         case 1: *q++ = iv;
 338         } while ((n -= 4) > 0);
 339     }else
 340         last_exp = 36;
 341
 342     while (q < q_end) {
 343         code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX);
 344         if (code < 0)
 345             return -1;
 346         /* NOTE: this offset is the same as MPEG4 AAC ! */
 347         last_exp += code - 60;
 348         if ((unsigned)last_exp + 60 > FF_ARRAY_ELEMS(pow_tab))
 349             return -1;
 350         v = ptab[last_exp];
 351         iv = iptab[last_exp];
 352         if (v > max_scale)
 353             max_scale = v;
 354         n = *ptr++;
 355         switch (n & 3) do {
 356         case 0: *q++ = iv;
 357         case 3: *q++ = iv;
 358         case 2: *q++ = iv;
 359         case 1: *q++ = iv;
 360         } while ((n -= 4) > 0);
 361     }
 362     s->max_exponent[ch] = max_scale;
 363     return 0;
 364 }
 365
 366
 367 /**
 368  * Apply MDCT window and add into output.
 369  *
 370  * We ensure that when the windows overlap their squared sum
 371  * is always 1 (MDCT reconstruction rule).
 372  */
 373 static void wma_window(WMACodecContext *s, float *out)
 374 {
 375     float *in = s->output;
 376     int block_len, bsize, n;
 377
 378     /* left part */
 379     if (s->block_len_bits <= s->prev_block_len_bits) {
 380         block_len = s->block_len;
 381         bsize = s->frame_len_bits - s->block_len_bits;
 382
 383         s->dsp.vector_fmul_add(out, in, s->windows[bsize],
 384                                out, block_len);
 385
 386     } else {
 387         block_len = 1 << s->prev_block_len_bits;
 388         n = (s->block_len - block_len) / 2;
 389         bsize = s->frame_len_bits - s->prev_block_len_bits;
 390
 391         s->dsp.vector_fmul_add(out+n, in+n, s->windows[bsize],
 392                                out+n, block_len);
 393
 394         memcpy(out+n+block_len, in+n+block_len, n*sizeof(float));
 395     }
 396
 397     out += s->block_len;
 398     in += s->block_len;
 399
 400     /* right part */
 401     if (s->block_len_bits <= s->next_block_len_bits) {
 402         block_len = s->block_len;
 403         bsize = s->frame_len_bits - s->block_len_bits;
 404
 405         s->dsp.vector_fmul_reverse(out, in, s->windows[bsize], block_len);
 406
 407     } else {
 408         block_len = 1 << s->next_block_len_bits;
 409         n = (s->block_len - block_len) / 2;
 410         bsize = s->frame_len_bits - s->next_block_len_bits;
 411
 412         memcpy(out, in, n*sizeof(float));
 413
 414         s->dsp.vector_fmul_reverse(out+n, in+n, s->windows[bsize], block_len);
 415
 416         memset(out+n+block_len, 0, n*sizeof(float));
 417     }
 418 }
 419
 420
 421 /**
 422  * @return 0 if OK. 1 if last block of frame. return -1 if
 423  * unrecorrable error.
 424  */
 425 static int wma_decode_block(WMACodecContext *s)
 426 {
 427     int n, v, a, ch, bsize;
 428     int coef_nb_bits, total_gain;
 429     int nb_coefs[MAX_CHANNELS];
 430     float mdct_norm;
 431
 432 #ifdef TRACE
 433     tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num);
 434 #endif
 435
 436     /* compute current block length */
 437     if (s->use_variable_block_len) {
 438         n = av_log2(s->nb_block_sizes - 1) + 1;
 439
 440         if (s->reset_block_lengths) {
 441             s->reset_block_lengths = 0;
 442             v = get_bits(&s->gb, n);
 443             if (v >= s->nb_block_sizes)
 444                 return -1;
 445             s->prev_block_len_bits = s->frame_len_bits - v;
 446             v = get_bits(&s->gb, n);
 447             if (v >= s->nb_block_sizes)
 448                 return -1;
 449             s->block_len_bits = s->frame_len_bits - v;
 450         } else {
 451             /* update block lengths */
 452             s->prev_block_len_bits = s->block_len_bits;
 453             s->block_len_bits = s->next_block_len_bits;
 454         }
 455         v = get_bits(&s->gb, n);
 456         if (v >= s->nb_block_sizes)
 457             return -1;
 458         s->next_block_len_bits = s->frame_len_bits - v;
 459     } else {
 460         /* fixed block len */
 461         s->next_block_len_bits = s->frame_len_bits;
 462         s->prev_block_len_bits = s->frame_len_bits;
 463         s->block_len_bits = s->frame_len_bits;
 464     }
 465
 466     /* now check if the block length is coherent with the frame length */
 467     s->block_len = 1 << s->block_len_bits;
 468     if ((s->block_pos + s->block_len) > s->frame_len)
 469         return -1;
 470
 471     if (s->nb_channels == 2) {
 472         s->ms_stereo = get_bits1(&s->gb);
 473     }
 474     v = 0;
 475     for(ch = 0; ch < s->nb_channels; ch++) {
 476         a = get_bits1(&s->gb);
 477         s->channel_coded[ch] = a;
 478         v |= a;
 479     }
 480
 481     bsize = s->frame_len_bits - s->block_len_bits;
 482
 483     /* if no channel coded, no need to go further */
 484     /* XXX: fix potential framing problems */
 485     if (!v)
 486         goto next;
 487
 488     /* read total gain and extract corresponding number of bits for
 489        coef escape coding */
 490     total_gain = 1;
 491     for(;;) {
 492         a = get_bits(&s->gb, 7);
 493         total_gain += a;
 494         if (a != 127)
 495             break;
 496     }
 497
 498     coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
 499
 500     /* compute number of coefficients */
 501     n = s->coefs_end[bsize] - s->coefs_start;
 502     for(ch = 0; ch < s->nb_channels; ch++)
 503         nb_coefs[ch] = n;
 504
 505     /* complex coding */
 506     if (s->use_noise_coding) {
 507
 508         for(ch = 0; ch < s->nb_channels; ch++) {
 509             if (s->channel_coded[ch]) {
 510                 int i, n, a;
 511                 n = s->exponent_high_sizes[bsize];
 512                 for(i=0;i<n;i++) {
 513                     a = get_bits1(&s->gb);
 514                     s->high_band_coded[ch][i] = a;
 515                     /* if noise coding, the coefficients are not transmitted */
 516                     if (a)
 517                         nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
 518                 }
 519             }
 520         }
 521         for(ch = 0; ch < s->nb_channels; ch++) {
 522             if (s->channel_coded[ch]) {
 523                 int i, n, val, code;
 524
 525                 n = s->exponent_high_sizes[bsize];
 526                 val = (int)0x80000000;
 527                 for(i=0;i<n;i++) {
 528                     if (s->high_band_coded[ch][i]) {
 529                         if (val == (int)0x80000000) {
 530                             val = get_bits(&s->gb, 7) - 19;
 531                         } else {
 532                             code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX);
 533                             if (code < 0)
 534                                 return -1;
 535                             val += code - 18;
 536                         }
 537                         s->high_band_values[ch][i] = val;
 538                     }
 539                 }
 540             }
 541         }
 542     }
 543
 544     /* exponents can be reused in short blocks. */
 545     if ((s->block_len_bits == s->frame_len_bits) ||
 546         get_bits1(&s->gb)) {
 547         for(ch = 0; ch < s->nb_channels; ch++) {
 548             if (s->channel_coded[ch]) {
 549                 if (s->use_exp_vlc) {
 550                     if (decode_exp_vlc(s, ch) < 0)
 551                         return -1;
 552                 } else {
 553                     decode_exp_lsp(s, ch);
 554                 }
 555                 s->exponents_bsize[ch] = bsize;
 556             }
 557         }
 558     }
 559
 560     /* parse spectral coefficients : just RLE encoding */
 561     for(ch = 0; ch < s->nb_channels; ch++) {
 562         if (s->channel_coded[ch]) {
 563             int tindex;
 564             WMACoef* ptr = &s->coefs1[ch][0];
 565
 566             /* special VLC tables are used for ms stereo because
 567                there is potentially less energy there */
 568             tindex = (ch == 1 && s->ms_stereo);
 569             memset(ptr, 0, s->block_len * sizeof(WMACoef));
 570             ff_wma_run_level_decode(s->avctx, &s->gb, &s->coef_vlc[tindex],
 571                   s->level_table[tindex], s->run_table[tindex],
 572                   0, ptr, 0, nb_coefs[ch],
 573                   s->block_len, s->frame_len_bits, coef_nb_bits);
 574         }
 575         if (s->version == 1 && s->nb_channels >= 2) {
 576             align_get_bits(&s->gb);
 577         }
 578     }
 579
 580     /* normalize */
 581     {
 582         int n4 = s->block_len / 2;
 583         mdct_norm = 1.0 / (float)n4;
 584         if (s->version == 1) {
 585             mdct_norm *= sqrt(n4);
 586         }
 587     }
 588
 589     /* finally compute the MDCT coefficients */
 590     for(ch = 0; ch < s->nb_channels; ch++) {
 591         if (s->channel_coded[ch]) {
 592             WMACoef *coefs1;
 593             float *coefs, *exponents, mult, mult1, noise;
 594             int i, j, n, n1, last_high_band, esize;
 595             float exp_power[HIGH_BAND_MAX_SIZE];
 596
 597             coefs1 = s->coefs1[ch];
 598             exponents = s->exponents[ch];
 599             esize = s->exponents_bsize[ch];
 600             mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
 601             mult *= mdct_norm;
 602             coefs = s->coefs[ch];
 603             if (s->use_noise_coding) {
 604                 mult1 = mult;
 605                 /* very low freqs : noise */
 606                 for(i = 0;i < s->coefs_start; i++) {
 607                     *coefs++ = s->noise_table[s->noise_index] *
 608                       exponents[i<<bsize>>esize] * mult1;
 609                     s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 610                 }
 611
 612                 n1 = s->exponent_high_sizes[bsize];
 613
 614                 /* compute power of high bands */
 615                 exponents = s->exponents[ch] +
 616                     (s->high_band_start[bsize]<<bsize);
 617                 last_high_band = 0; /* avoid warning */
 618                 for(j=0;j<n1;j++) {
 619                     n = s->exponent_high_bands[s->frame_len_bits -
 620                                               s->block_len_bits][j];
 621                     if (s->high_band_coded[ch][j]) {
 622                         float e2, v;
 623                         e2 = 0;
 624                         for(i = 0;i < n; i++) {
 625                             v = exponents[i<<bsize>>esize];
 626                             e2 += v * v;
 627                         }
 628                         exp_power[j] = e2 / n;
 629                         last_high_band = j;
 630                         tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n);
 631                     }
 632                     exponents += n<<bsize;
 633                 }
 634
 635                 /* main freqs and high freqs */
 636                 exponents = s->exponents[ch] + (s->coefs_start<<bsize);
 637                 for(j=-1;j<n1;j++) {
 638                     if (j < 0) {
 639                         n = s->high_band_start[bsize] -
 640                             s->coefs_start;
 641                     } else {
 642                         n = s->exponent_high_bands[s->frame_len_bits -
 643                                                   s->block_len_bits][j];
 644                     }
 645                     if (j >= 0 && s->high_band_coded[ch][j]) {
 646                         /* use noise with specified power */
 647                         mult1 = sqrt(exp_power[j] / exp_power[last_high_band]);
 648                         /* XXX: use a table */
 649                         mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05);
 650                         mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult);
 651                         mult1 *= mdct_norm;
 652                         for(i = 0;i < n; i++) {
 653                             noise = s->noise_table[s->noise_index];
 654                             s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 655                             *coefs++ =  noise *
 656                                 exponents[i<<bsize>>esize] * mult1;
 657                         }
 658                         exponents += n<<bsize;
 659                     } else {
 660                         /* coded values + small noise */
 661                         for(i = 0;i < n; i++) {
 662                             noise = s->noise_table[s->noise_index];
 663                             s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 664                             *coefs++ = ((*coefs1++) + noise) *
 665                                 exponents[i<<bsize>>esize] * mult;
 666                         }
 667                         exponents += n<<bsize;
 668                     }
 669                 }
 670
 671                 /* very high freqs : noise */
 672                 n = s->block_len - s->coefs_end[bsize];
 673                 mult1 = mult * exponents[((-1<<bsize))>>esize];
 674                 for(i = 0; i < n; i++) {
 675                     *coefs++ = s->noise_table[s->noise_index] * mult1;
 676                     s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
 677                 }
 678             } else {
 679                 /* XXX: optimize more */
 680                 for(i = 0;i < s->coefs_start; i++)
 681                     *coefs++ = 0.0;
 682                 n = nb_coefs[ch];
 683                 for(i = 0;i < n; i++) {
 684                     *coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult;
 685                 }
 686                 n = s->block_len - s->coefs_end[bsize];
 687                 for(i = 0;i < n; i++)
 688                     *coefs++ = 0.0;
 689             }
 690         }
 691     }
 692
 693 #ifdef TRACE
 694     for(ch = 0; ch < s->nb_channels; ch++) {
 695         if (s->channel_coded[ch]) {
 696             dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len);
 697             dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len);
 698         }
 699     }
 700 #endif
 701
 702     if (s->ms_stereo && s->channel_coded[1]) {
 703         /* nominal case for ms stereo: we do it before mdct */
 704         /* no need to optimize this case because it should almost
 705            never happen */
 706         if (!s->channel_coded[0]) {
 707             tprintf(s->avctx, "rare ms-stereo case happened\n");
 708             memset(s->coefs[0], 0, sizeof(float) * s->block_len);
 709             s->channel_coded[0] = 1;
 710         }
 711
 712         s->dsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len);
 713     }
 714
 715 next:
 716     for(ch = 0; ch < s->nb_channels; ch++) {
 717         int n4, index;
 718
 719         n4 = s->block_len / 2;
 720         if(s->channel_coded[ch]){
 721             ff_imdct_calc(&s->mdct_ctx[bsize], s->output, s->coefs[ch]);
 722         }else if(!(s->ms_stereo && ch==1))
 723             memset(s->output, 0, sizeof(s->output));
 724
 725         /* multiply by the window and add in the frame */
 726         index = (s->frame_len / 2) + s->block_pos - n4;
 727         wma_window(s, &s->frame_out[ch][index]);
 728     }
 729
 730     /* update block number */
 731     s->block_num++;
 732     s->block_pos += s->block_len;
 733     if (s->block_pos >= s->frame_len)
 734         return 1;
 735     else
 736         return 0;
 737 }
 738
 739 /* decode a frame of frame_len samples */
 740 static int wma_decode_frame(WMACodecContext *s, int16_t *samples)
 741 {
 742     int ret, i, n, ch, incr;
 743     int16_t *ptr;
 744     float *iptr;
 745
 746 #ifdef TRACE
 747     tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len);
 748 #endif
 749
 750     /* read each block */
 751     s->block_num = 0;
 752     s->block_pos = 0;
 753     for(;;) {
 754         ret = wma_decode_block(s);
 755         if (ret < 0)
 756             return -1;
 757         if (ret)
 758             break;
 759     }
 760
 761     /* convert frame to integer */
 762     n = s->frame_len;
 763     incr = s->nb_channels;
 764     for(ch = 0; ch < s->nb_channels; ch++) {
 765         ptr = samples + ch;
 766         iptr = s->frame_out[ch];
 767
 768         for(i=0;i<n;i++) {
 769             *ptr = av_clip_int16(lrintf(*iptr++));
 770             ptr += incr;
 771         }
 772         /* prepare for next block */
 773         memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len],
 774                 s->frame_len * sizeof(float));
 775     }
 776
 777 #ifdef TRACE
 778     dump_shorts(s, "samples", samples, n * s->nb_channels);
 779 #endif
 780     return 0;
 781 }
 782
 783 static int wma_decode_superframe(AVCodecContext *avctx,
 784                                  void *data, int *data_size,
 785                                  AVPacket *avpkt)
 786 {
 787     const uint8_t *buf = avpkt->data;
 788     int buf_size = avpkt->size;
 789     WMACodecContext *s = avctx->priv_data;
 790     int nb_frames, bit_offset, i, pos, len;
 791     uint8_t *q;
 792     int16_t *samples;
 793
 794     tprintf(avctx, "***decode_superframe:\n");
 795
 796     if(buf_size==0){
 797         s->last_superframe_len = 0;
 798         return 0;
 799     }
 800     if (buf_size < s->block_align)
 801         return 0;
 802     buf_size = s->block_align;
 803
 804     samples = data;
 805
 806     init_get_bits(&s->gb, buf, buf_size*8);
 807
 808     if (s->use_bit_reservoir) {
 809         /* read super frame header */
 810         skip_bits(&s->gb, 4); /* super frame index */
 811         nb_frames = get_bits(&s->gb, 4) - 1;
 812
 813         if((nb_frames+1) * s->nb_channels * s->frame_len * sizeof(int16_t) > *data_size){
 814             av_log(s->avctx, AV_LOG_ERROR, "Insufficient output space\n");
 815             goto fail;
 816         }
 817
 818         bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
 819
 820         if (s->last_superframe_len > 0) {
 821             //        printf("skip=%d\n", s->last_bitoffset);
 822             /* add bit_offset bits to last frame */
 823             if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) >
 824                 MAX_CODED_SUPERFRAME_SIZE)
 825                 goto fail;
 826             q = s->last_superframe + s->last_superframe_len;
 827             len = bit_offset;
 828             while (len > 7) {
 829                 *q++ = (get_bits)(&s->gb, 8);
 830                 len -= 8;
 831             }
 832             if (len > 0) {
 833                 *q++ = (get_bits)(&s->gb, len) << (8 - len);
 834             }
 835
 836             /* XXX: bit_offset bits into last frame */
 837             init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8);
 838             /* skip unused bits */
 839             if (s->last_bitoffset > 0)
 840                 skip_bits(&s->gb, s->last_bitoffset);
 841             /* this frame is stored in the last superframe and in the
 842                current one */
 843             if (wma_decode_frame(s, samples) < 0)
 844                 goto fail;
 845             samples += s->nb_channels * s->frame_len;
 846         }
 847
 848         /* read each frame starting from bit_offset */
 849         pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
 850         init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8);
 851         len = pos & 7;
 852         if (len > 0)
 853             skip_bits(&s->gb, len);
 854
 855         s->reset_block_lengths = 1;
 856         for(i=0;i<nb_frames;i++) {
 857             if (wma_decode_frame(s, samples) < 0)
 858                 goto fail;
 859             samples += s->nb_channels * s->frame_len;
 860         }
 861
 862         /* we copy the end of the frame in the last frame buffer */
 863         pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
 864         s->last_bitoffset = pos & 7;
 865         pos >>= 3;
 866         len = buf_size - pos;
 867         if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) {
 868             goto fail;
 869         }
 870         s->last_superframe_len = len;
 871         memcpy(s->last_superframe, buf + pos, len);
 872     } else {
 873         if(s->nb_channels * s->frame_len * sizeof(int16_t) > *data_size){
 874             av_log(s->avctx, AV_LOG_ERROR, "Insufficient output space\n");
 875             goto fail;
 876         }
 877         /* single frame decode */
 878         if (wma_decode_frame(s, samples) < 0)
 879             goto fail;
 880         samples += s->nb_channels * s->frame_len;
 881     }
 882
 883 //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);
 884
 885     *data_size = (int8_t *)samples - (int8_t *)data;
 886     return s->block_align;
 887  fail:
 888     /* when error, we reset the bit reservoir */
 889     s->last_superframe_len = 0;
 890     return -1;
 891 }
 892
 893 AVCodec wmav1_decoder =
 894 {
 895     "wmav1",
 896     CODEC_TYPE_AUDIO,
 897     CODEC_ID_WMAV1,
 898     sizeof(WMACodecContext),
 899     wma_decode_init,
 900     NULL,
 901     ff_wma_end,
 902     wma_decode_superframe,
 903     .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
 904 };
 905
 906 AVCodec wmav2_decoder =
 907 {
 908     "wmav2",
 909     CODEC_TYPE_AUDIO,
 910     CODEC_ID_WMAV2,
 911     sizeof(WMACodecContext),
 912     wma_decode_init,
 913     NULL,
 914     ff_wma_end,
 915     wma_decode_superframe,
 916     .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),
 917 };