libfaad2/sbr_qmf.c

   1 /*
   2 ** FAAD2 - Freeware Advanced Audio (AAC) Decoder including SBR decoding
   3 ** Copyright (C) 2003-2004 M. Bakker, Ahead Software AG, http://www.nero.com
   4 **
   5 ** This program is free software; you can redistribute it and/or modify
   6 ** it under the terms of the GNU General Public License as published by
   7 ** the Free Software Foundation; either version 2 of the License, or
   8 ** (at your option) any later version.
   9 **
  10 ** This program is distributed in the hope that it will be useful,
  11 ** but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 ** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 ** GNU General Public License for more details.
  14 **
  15 ** You should have received a copy of the GNU General Public License
  16 ** along with this program; if not, write to the Free Software
  17 ** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18 **
  19 ** Any non-GPL usage of this software or parts of this software is strictly
  20 ** forbidden.
  21 **
  22 ** Commercial non-GPL licensing of this software is possible.
  23 ** For more info contact Ahead Software through Mpeg4AAClicense@nero.com.
  24 **
  25 ** $Id: sbr_qmf.c,v 1.27 2004/09/04 14:56:28 menno Exp $
  26 **/
  27
  28 #include "common.h"
  29 #include "structs.h"
  30
  31 #ifdef SBR_DEC
  32
  33
  34 #include <stdlib.h>
  35 #include <string.h>
  36 #include "sbr_dct.h"
  37 #include "sbr_qmf.h"
  38 #include "sbr_qmf_c.h"
  39 #include "sbr_syntax.h"
  40
  41 qmfa_info *qmfa_init(uint8_t channels)
  42 {
  43     qmfa_info *qmfa = (qmfa_info*)faad_malloc(sizeof(qmfa_info));
  44
  45         /* x is implemented as double ringbuffer */
  46     qmfa->x = (real_t*)faad_malloc(2 * channels * 10 * sizeof(real_t));
  47     memset(qmfa->x, 0, 2 * channels * 10 * sizeof(real_t));
  48
  49         /* ringbuffer index */
  50         qmfa->x_index = 0;
  51
  52     qmfa->channels = channels;
  53
  54     return qmfa;
  55 }
  56
  57 void qmfa_end(qmfa_info *qmfa)
  58 {
  59     if (qmfa)
  60     {
  61         if (qmfa->x) faad_free(qmfa->x);
  62         faad_free(qmfa);
  63     }
  64 }
  65
  66 void sbr_qmf_analysis_32(sbr_info *sbr, qmfa_info *qmfa, const real_t *input,
  67                          qmf_t X[MAX_NTSRHFG][64], uint8_t offset, uint8_t kx)
  68 {
  69     ALIGN real_t u[64];
  70 #ifndef SBR_LOW_POWER
  71     ALIGN real_t in_real[32], in_imag[32], out_real[32], out_imag[32];
  72 #else
  73     ALIGN real_t y[32];
  74 #endif
  75     uint16_t in = 0;
  76     uint8_t l;
  77
  78     /* qmf subsample l */
  79     for (l = 0; l < sbr->numTimeSlotsRate; l++)
  80     {
  81         int16_t n;
  82
  83         /* shift input buffer x */
  84                 /* input buffer is not shifted anymore, x is implemented as double ringbuffer */
  85         //memmove(qmfa->x + 32, qmfa->x, (320-32)*sizeof(real_t));
  86
  87         /* add new samples to input buffer x */
  88         for (n = 32 - 1; n >= 0; n--)
  89         {
  90 #ifdef FIXED_POINT
  91             qmfa->x[qmfa->x_index + n] = qmfa->x[qmfa->x_index + n + 320] = (input[in++]) >> 4;
  92 #else
  93             qmfa->x[qmfa->x_index + n] = qmfa->x[qmfa->x_index + n + 320] = input[in++];
  94 #endif
  95         }
  96
  97         /* window and summation to create array u */
  98         for (n = 0; n < 64; n++)
  99         {
 100             u[n] = MUL_F(qmfa->x[qmfa->x_index + n], qmf_c[2*n]) +
 101                 MUL_F(qmfa->x[qmfa->x_index + n + 64], qmf_c[2*(n + 64)]) +
 102                 MUL_F(qmfa->x[qmfa->x_index + n + 128], qmf_c[2*(n + 128)]) +
 103                 MUL_F(qmfa->x[qmfa->x_index + n + 192], qmf_c[2*(n + 192)]) +
 104                 MUL_F(qmfa->x[qmfa->x_index + n + 256], qmf_c[2*(n + 256)]);
 105         }
 106
 107                 /* update ringbuffer index */
 108                 qmfa->x_index -= 32;
 109                 if (qmfa->x_index < 0)
 110                         qmfa->x_index = (320-32);
 111
 112         /* calculate 32 subband samples by introducing X */
 113 #ifdef SBR_LOW_POWER
 114         y[0] = u[48];
 115         for (n = 1; n < 16; n++)
 116             y[n] = u[n+48] + u[48-n];
 117         for (n = 16; n < 32; n++)
 118             y[n] = -u[n-16] + u[48-n];
 119
 120         DCT3_32_unscaled(u, y);
 121
 122         for (n = 0; n < 32; n++)
 123         {
 124             if (n < kx)
 125             {
 126 #ifdef FIXED_POINT
 127                 QMF_RE(X[l + offset][n]) = u[n] /*<< 1*/;
 128 #else
 129                 QMF_RE(X[l + offset][n]) = 2. * u[n];
 130 #endif
 131             } else {
 132                 QMF_RE(X[l + offset][n]) = 0;
 133             }
 134         }
 135 #else
 136
 137         // Reordering of data moved from DCT_IV to here
 138         in_imag[31] = u[1];
 139         in_real[0] = u[0];
 140         for (n = 1; n < 31; n++)
 141         {
 142             in_imag[31 - n] = u[n+1];
 143             in_real[n] = -u[64-n];
 144         }
 145         in_imag[0] = u[32];
 146         in_real[31] = -u[33];
 147
 148         // dct4_kernel is DCT_IV without reordering which is done before and after FFT
 149         dct4_kernel(in_real, in_imag, out_real, out_imag);
 150
 151         // Reordering of data moved from DCT_IV to here
 152         for (n = 0; n < 16; n++) {
 153             if (2*n+1 < kx) {
 154 #ifdef FIXED_POINT
 155                 QMF_RE(X[l + offset][2*n])   = out_real[n];
 156                 QMF_IM(X[l + offset][2*n])   = out_imag[n];
 157                 QMF_RE(X[l + offset][2*n+1]) = -out_imag[31-n];
 158                 QMF_IM(X[l + offset][2*n+1]) = -out_real[31-n];
 159 #else
 160                 QMF_RE(X[l + offset][2*n])   = 2. * out_real[n];
 161                 QMF_IM(X[l + offset][2*n])   = 2. * out_imag[n];
 162                 QMF_RE(X[l + offset][2*n+1]) = -2. * out_imag[31-n];
 163                 QMF_IM(X[l + offset][2*n+1]) = -2. * out_real[31-n];
 164 #endif
 165             } else {
 166                 if (2*n < kx) {
 167 #ifdef FIXED_POINT
 168                     QMF_RE(X[l + offset][2*n])   = out_real[n];
 169                     QMF_IM(X[l + offset][2*n])   = out_imag[n];
 170 #else
 171                     QMF_RE(X[l + offset][2*n])   = 2. * out_real[n];
 172                     QMF_IM(X[l + offset][2*n])   = 2. * out_imag[n];
 173 #endif
 174                 }
 175                 else {
 176                     QMF_RE(X[l + offset][2*n]) = 0;
 177                     QMF_IM(X[l + offset][2*n]) = 0;
 178                 }
 179                 QMF_RE(X[l + offset][2*n+1]) = 0;
 180                 QMF_IM(X[l + offset][2*n+1]) = 0;
 181             }
 182         }
 183 #endif
 184     }
 185 }
 186
 187 static const complex_t qmf32_pre_twiddle[] =
 188 {
 189     { FRAC_CONST(0.999924701839145), FRAC_CONST(-0.012271538285720) },
 190     { FRAC_CONST(0.999322384588350), FRAC_CONST(-0.036807222941359) },
 191     { FRAC_CONST(0.998118112900149), FRAC_CONST(-0.061320736302209) },
 192     { FRAC_CONST(0.996312612182778), FRAC_CONST(-0.085797312344440) },
 193     { FRAC_CONST(0.993906970002356), FRAC_CONST(-0.110222207293883) },
 194     { FRAC_CONST(0.990902635427780), FRAC_CONST(-0.134580708507126) },
 195     { FRAC_CONST(0.987301418157858), FRAC_CONST(-0.158858143333861) },
 196     { FRAC_CONST(0.983105487431216), FRAC_CONST(-0.183039887955141) },
 197     { FRAC_CONST(0.978317370719628), FRAC_CONST(-0.207111376192219) },
 198     { FRAC_CONST(0.972939952205560), FRAC_CONST(-0.231058108280671) },
 199     { FRAC_CONST(0.966976471044852), FRAC_CONST(-0.254865659604515) },
 200     { FRAC_CONST(0.960430519415566), FRAC_CONST(-0.278519689385053) },
 201     { FRAC_CONST(0.953306040354194), FRAC_CONST(-0.302005949319228) },
 202     { FRAC_CONST(0.945607325380521), FRAC_CONST(-0.325310292162263) },
 203     { FRAC_CONST(0.937339011912575), FRAC_CONST(-0.348418680249435) },
 204     { FRAC_CONST(0.928506080473216), FRAC_CONST(-0.371317193951838) },
 205     { FRAC_CONST(0.919113851690058), FRAC_CONST(-0.393992040061048) },
 206     { FRAC_CONST(0.909167983090522), FRAC_CONST(-0.416429560097637) },
 207     { FRAC_CONST(0.898674465693954), FRAC_CONST(-0.438616238538528) },
 208     { FRAC_CONST(0.887639620402854), FRAC_CONST(-0.460538710958240) },
 209     { FRAC_CONST(0.876070094195407), FRAC_CONST(-0.482183772079123) },
 210     { FRAC_CONST(0.863972856121587), FRAC_CONST(-0.503538383725718) },
 211     { FRAC_CONST(0.851355193105265), FRAC_CONST(-0.524589682678469) },
 212     { FRAC_CONST(0.838224705554838), FRAC_CONST(-0.545324988422046) },
 213     { FRAC_CONST(0.824589302785025), FRAC_CONST(-0.565731810783613) },
 214     { FRAC_CONST(0.810457198252595), FRAC_CONST(-0.585797857456439) },
 215     { FRAC_CONST(0.795836904608884), FRAC_CONST(-0.605511041404326) },
 216     { FRAC_CONST(0.780737228572094), FRAC_CONST(-0.624859488142386) },
 217     { FRAC_CONST(0.765167265622459), FRAC_CONST(-0.643831542889791) },
 218     { FRAC_CONST(0.749136394523459), FRAC_CONST(-0.662415777590172) },
 219     { FRAC_CONST(0.732654271672413), FRAC_CONST(-0.680600997795453) },
 220     { FRAC_CONST(0.715730825283819), FRAC_CONST(-0.698376249408973) }
 221 };
 222
 223 qmfs_info *qmfs_init(uint8_t channels)
 224 {
 225     qmfs_info *qmfs = (qmfs_info*)faad_malloc(sizeof(qmfs_info));
 226
 227         /* v is a double ringbuffer */
 228     qmfs->v = (real_t*)faad_malloc(2 * channels * 20 * sizeof(real_t));
 229     memset(qmfs->v, 0, 2 * channels * 20 * sizeof(real_t));
 230
 231     qmfs->v_index = 0;
 232
 233     qmfs->channels = channels;
 234
 235     return qmfs;
 236 }
 237
 238 void qmfs_end(qmfs_info *qmfs)
 239 {
 240     if (qmfs)
 241     {
 242         if (qmfs->v) faad_free(qmfs->v);
 243         faad_free(qmfs);
 244     }
 245 }
 246
 247 #ifdef SBR_LOW_POWER
 248
 249 void sbr_qmf_synthesis_32(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSRHFG][64],
 250                           real_t *output)
 251 {
 252     ALIGN real_t x[16];
 253     ALIGN real_t y[16];
 254     int16_t n, k, out = 0;
 255     uint8_t l;
 256
 257     /* qmf subsample l */
 258     for (l = 0; l < sbr->numTimeSlotsRate; l++)
 259     {
 260         /* shift buffers */
 261         /* we are not shifting v, it is a double ringbuffer */
 262         //memmove(qmfs->v + 64, qmfs->v, (640-64)*sizeof(real_t));
 263
 264         /* calculate 64 samples */
 265         for (k = 0; k < 16; k++)
 266         {
 267 #ifdef FIXED_POINT
 268             y[k] = (QMF_RE(X[l][k]) - QMF_RE(X[l][31 - k]));
 269             x[k] = (QMF_RE(X[l][k]) + QMF_RE(X[l][31 - k]));
 270 #else
 271             y[k] = (QMF_RE(X[l][k]) - QMF_RE(X[l][31 - k])) / 32.0;
 272             x[k] = (QMF_RE(X[l][k]) + QMF_RE(X[l][31 - k])) / 32.0;
 273 #endif
 274         }
 275
 276         /* even n samples */
 277         DCT2_16_unscaled(x, x);
 278         /* odd n samples */
 279         DCT4_16(y, y);
 280
 281         for (n = 8; n < 24; n++)
 282         {
 283             qmfs->v[qmfs->v_index + n*2] = qmfs->v[qmfs->v_index + 640 + n*2] = x[n-8];
 284             qmfs->v[qmfs->v_index + n*2+1] = qmfs->v[qmfs->v_index + 640 + n*2+1] = y[n-8];
 285         }
 286         for (n = 0; n < 16; n++)
 287         {
 288             qmfs->v[qmfs->v_index + n] = qmfs->v[qmfs->v_index + 640 + n] = qmfs->v[qmfs->v_index + 32-n];
 289         }
 290         qmfs->v[qmfs->v_index + 48] = qmfs->v[qmfs->v_index + 640 + 48] = 0;
 291         for (n = 1; n < 16; n++)
 292         {
 293             qmfs->v[qmfs->v_index + 48+n] = qmfs->v[qmfs->v_index + 640 + 48+n] = -qmfs->v[qmfs->v_index + 48-n];
 294         }
 295
 296         /* calculate 32 output samples and window */
 297         for (k = 0; k < 32; k++)
 298         {
 299             output[out++] = MUL_F(qmfs->v[qmfs->v_index + k], qmf_c[2*k]) +
 300                 MUL_F(qmfs->v[qmfs->v_index + 96 + k], qmf_c[64 + 2*k]) +
 301                 MUL_F(qmfs->v[qmfs->v_index + 128 + k], qmf_c[128 + 2*k]) +
 302                 MUL_F(qmfs->v[qmfs->v_index + 224 + k], qmf_c[192 + 2*k]) +
 303                 MUL_F(qmfs->v[qmfs->v_index + 256 + k], qmf_c[256 + 2*k]) +
 304                 MUL_F(qmfs->v[qmfs->v_index + 352 + k], qmf_c[320 + 2*k]) +
 305                 MUL_F(qmfs->v[qmfs->v_index + 384 + k], qmf_c[384 + 2*k]) +
 306                 MUL_F(qmfs->v[qmfs->v_index + 480 + k], qmf_c[448 + 2*k]) +
 307                 MUL_F(qmfs->v[qmfs->v_index + 512 + k], qmf_c[512 + 2*k]) +
 308                 MUL_F(qmfs->v[qmfs->v_index + 608 + k], qmf_c[576 + 2*k]);
 309         }
 310
 311         /* update the ringbuffer index */
 312         qmfs->v_index -= 64;
 313         if (qmfs->v_index < 0)
 314             qmfs->v_index = (640-64);
 315     }
 316 }
 317
 318 void sbr_qmf_synthesis_64(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSRHFG][64],
 319                           real_t *output)
 320 {
 321     ALIGN real_t x[64];
 322     ALIGN real_t y[64];
 323     int16_t n, k, out = 0;
 324     uint8_t l;
 325
 326
 327     /* qmf subsample l */
 328     for (l = 0; l < sbr->numTimeSlotsRate; l++)
 329     {
 330         /* shift buffers */
 331         /* we are not shifting v, it is a double ringbuffer */
 332         //memmove(qmfs->v + 128, qmfs->v, (1280-128)*sizeof(real_t));
 333
 334         /* calculate 128 samples */
 335         for (k = 0; k < 32; k++)
 336         {
 337 #ifdef FIXED_POINT
 338             y[k] = (QMF_RE(X[l][k]) - QMF_RE(X[l][63 - k]));
 339             x[k] = (QMF_RE(X[l][k]) + QMF_RE(X[l][63 - k]));
 340 #else
 341             y[k] = (QMF_RE(X[l][k]) - QMF_RE(X[l][63 - k])) / 32.0;
 342             x[k] = (QMF_RE(X[l][k]) + QMF_RE(X[l][63 - k])) / 32.0;
 343 #endif
 344         }
 345
 346         /* even n samples */
 347         DCT2_32_unscaled(x, x);
 348         /* odd n samples */
 349         DCT4_32(y, y);
 350
 351         for (n = 16; n < 48; n++)
 352         {
 353             qmfs->v[qmfs->v_index + n*2]   = qmfs->v[qmfs->v_index + 1280 + n*2]   = x[n-16];
 354             qmfs->v[qmfs->v_index + n*2+1] = qmfs->v[qmfs->v_index + 1280 + n*2+1] = y[n-16];
 355         }
 356         for (n = 0; n < 32; n++)
 357         {
 358             qmfs->v[qmfs->v_index + n] = qmfs->v[qmfs->v_index + 1280 + n] = qmfs->v[qmfs->v_index + 64-n];
 359         }
 360         qmfs->v[qmfs->v_index + 96] = qmfs->v[qmfs->v_index + 1280 + 96] = 0;
 361         for (n = 1; n < 32; n++)
 362         {
 363             qmfs->v[qmfs->v_index + 96+n] = qmfs->v[qmfs->v_index + 1280 + 96+n] = -qmfs->v[qmfs->v_index + 96-n];
 364         }
 365
 366         /* calculate 64 output samples and window */
 367         for (k = 0; k < 64; k++)
 368         {
 369             output[out++] = MUL_F(qmfs->v[qmfs->v_index + k], qmf_c[k]) +
 370                 MUL_F(qmfs->v[qmfs->v_index + 192 + k], qmf_c[64 + k]) +
 371                 MUL_F(qmfs->v[qmfs->v_index + 256 + k], qmf_c[128 + k]) +
 372                 MUL_F(qmfs->v[qmfs->v_index + 256 + 192 + k], qmf_c[128 + 64 + k]) +
 373                 MUL_F(qmfs->v[qmfs->v_index + 512 + k], qmf_c[256 + k]) +
 374                 MUL_F(qmfs->v[qmfs->v_index + 512 + 192 + k], qmf_c[256 + 64 + k]) +
 375                 MUL_F(qmfs->v[qmfs->v_index + 768 + k], qmf_c[384 + k]) +
 376                 MUL_F(qmfs->v[qmfs->v_index + 768 + 192 + k], qmf_c[384 + 64 + k]) +
 377                 MUL_F(qmfs->v[qmfs->v_index + 1024 + k], qmf_c[512 + k]) +
 378                 MUL_F(qmfs->v[qmfs->v_index + 1024 + 192 + k], qmf_c[512 + 64 + k]);
 379         }
 380
 381         /* update the ringbuffer index */
 382         qmfs->v_index -= 128;
 383         if (qmfs->v_index < 0)
 384             qmfs->v_index = (1280-128);
 385     }
 386 }
 387 #else
 388 void sbr_qmf_synthesis_32(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSRHFG][64],
 389                           real_t *output)
 390 {
 391     ALIGN real_t x1[32], x2[32];
 392 #ifndef FIXED_POINT
 393     real_t scale = 1.f/64.f;
 394 #endif
 395     int16_t n, k, out = 0;
 396     uint8_t l;
 397
 398
 399     /* qmf subsample l */
 400     for (l = 0; l < sbr->numTimeSlotsRate; l++)
 401     {
 402         /* shift buffer v */
 403         /* buffer is not shifted, we are using a ringbuffer */
 404         //memmove(qmfs->v + 64, qmfs->v, (640-64)*sizeof(real_t));
 405
 406         /* calculate 64 samples */
 407         /* complex pre-twiddle */
 408         for (k = 0; k < 32; k++)
 409         {
 410             x1[k] = MUL_F(QMF_RE(X[l][k]), RE(qmf32_pre_twiddle[k])) - MUL_F(QMF_IM(X[l][k]), IM(qmf32_pre_twiddle[k]));
 411             x2[k] = MUL_F(QMF_IM(X[l][k]), RE(qmf32_pre_twiddle[k])) + MUL_F(QMF_RE(X[l][k]), IM(qmf32_pre_twiddle[k]));
 412
 413 #ifndef FIXED_POINT
 414             x1[k] *= scale;
 415             x2[k] *= scale;
 416 #else
 417             x1[k] >>= 1;
 418             x2[k] >>= 1;
 419 #endif
 420         }
 421
 422         /* transform */
 423         DCT4_32(x1, x1);
 424         DST4_32(x2, x2);
 425
 426         for (n = 0; n < 32; n++)
 427         {
 428             qmfs->v[qmfs->v_index + n]      = qmfs->v[qmfs->v_index + 640 + n]      = -x1[n] + x2[n];
 429             qmfs->v[qmfs->v_index + 63 - n] = qmfs->v[qmfs->v_index + 640 + 63 - n] =  x1[n] + x2[n];
 430         }
 431
 432         /* calculate 32 output samples and window */
 433         for (k = 0; k < 32; k++)
 434         {
 435             output[out++] = MUL_F(qmfs->v[qmfs->v_index + k], qmf_c[2*k]) +
 436                 MUL_F(qmfs->v[qmfs->v_index + 96 + k], qmf_c[64 + 2*k]) +
 437                 MUL_F(qmfs->v[qmfs->v_index + 128 + k], qmf_c[128 + 2*k]) +
 438                 MUL_F(qmfs->v[qmfs->v_index + 224 + k], qmf_c[192 + 2*k]) +
 439                 MUL_F(qmfs->v[qmfs->v_index + 256 + k], qmf_c[256 + 2*k]) +
 440                 MUL_F(qmfs->v[qmfs->v_index + 352 + k], qmf_c[320 + 2*k]) +
 441                 MUL_F(qmfs->v[qmfs->v_index + 384 + k], qmf_c[384 + 2*k]) +
 442                 MUL_F(qmfs->v[qmfs->v_index + 480 + k], qmf_c[448 + 2*k]) +
 443                 MUL_F(qmfs->v[qmfs->v_index + 512 + k], qmf_c[512 + 2*k]) +
 444                 MUL_F(qmfs->v[qmfs->v_index + 608 + k], qmf_c[576 + 2*k]);
 445         }
 446
 447         /* update ringbuffer index */
 448         qmfs->v_index -= 64;
 449         if (qmfs->v_index < 0)
 450             qmfs->v_index = (640 - 64);
 451     }
 452 }
 453
 454 void sbr_qmf_synthesis_64(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSRHFG][64],
 455                           real_t *output)
 456 {
 457 //    ALIGN real_t x1[64], x2[64];
 458 #ifndef SBR_LOW_POWER
 459     ALIGN real_t in_real1[32], in_imag1[32], out_real1[32], out_imag1[32];
 460     ALIGN real_t in_real2[32], in_imag2[32], out_real2[32], out_imag2[32];
 461 #endif
 462     qmf_t * pX;
 463     real_t * pring_buffer_1, * pring_buffer_3;
 464 //    real_t * ptemp_1, * ptemp_2;
 465 #ifdef PREFER_POINTERS
 466     // These pointers are used if target platform has autoinc address generators
 467     real_t * pring_buffer_2, * pring_buffer_4;
 468     real_t * pring_buffer_5, * pring_buffer_6;
 469     real_t * pring_buffer_7, * pring_buffer_8;
 470     real_t * pring_buffer_9, * pring_buffer_10;
 471     const real_t * pqmf_c_1, * pqmf_c_2, * pqmf_c_3, * pqmf_c_4;
 472     const real_t * pqmf_c_5, * pqmf_c_6, * pqmf_c_7, * pqmf_c_8;
 473     const real_t * pqmf_c_9, * pqmf_c_10;
 474 #endif // #ifdef PREFER_POINTERS
 475 #ifndef FIXED_POINT
 476     real_t scale = 1.f/64.f;
 477 #endif
 478     int16_t n, k, out = 0;
 479     uint8_t l;
 480
 481
 482     /* qmf subsample l */
 483     for (l = 0; l < sbr->numTimeSlotsRate; l++)
 484     {
 485         /* shift buffer v */
 486                 /* buffer is not shifted, we use double ringbuffer */
 487                 //memmove(qmfs->v + 128, qmfs->v, (1280-128)*sizeof(real_t));
 488
 489         /* calculate 128 samples */
 490 #ifndef FIXED_POINT
 491
 492         pX = X[l];
 493
 494         in_imag1[31] = scale*QMF_RE(pX[1]);
 495         in_real1[0]  = scale*QMF_RE(pX[0]);
 496         in_imag2[31] = scale*QMF_IM(pX[63-1]);
 497         in_real2[0]  = scale*QMF_IM(pX[63-0]);
 498         for (k = 1; k < 31; k++)
 499         {
 500             in_imag1[31 - k] = scale*QMF_RE(pX[2*k + 1]);
 501             in_real1[     k] = scale*QMF_RE(pX[2*k    ]);
 502             in_imag2[31 - k] = scale*QMF_IM(pX[63 - (2*k + 1)]);
 503             in_real2[     k] = scale*QMF_IM(pX[63 - (2*k    )]);
 504         }
 505         in_imag1[0]  = scale*QMF_RE(pX[63]);
 506         in_real1[31] = scale*QMF_RE(pX[62]);
 507         in_imag2[0]  = scale*QMF_IM(pX[63-63]);
 508         in_real2[31] = scale*QMF_IM(pX[63-62]);
 509
 510 #else
 511
 512         pX = X[l];
 513
 514         in_imag1[31] = QMF_RE(pX[1]) >> 1;
 515         in_real1[0]  = QMF_RE(pX[0]) >> 1;
 516         in_imag2[31] = QMF_IM(pX[62]) >> 1;
 517         in_real2[0]  = QMF_IM(pX[63]) >> 1;
 518         for (k = 1; k < 31; k++)
 519         {
 520             in_imag1[31 - k] = QMF_RE(pX[2*k + 1]) >> 1;
 521             in_real1[     k] = QMF_RE(pX[2*k    ]) >> 1;
 522             in_imag2[31 - k] = QMF_IM(pX[63 - (2*k + 1)]) >> 1;
 523             in_real2[     k] = QMF_IM(pX[63 - (2*k    )]) >> 1;
 524         }
 525         in_imag1[0]  = QMF_RE(pX[63]) >> 1;
 526         in_real1[31] = QMF_RE(pX[62]) >> 1;
 527         in_imag2[0]  = QMF_IM(pX[0]) >> 1;
 528         in_real2[31] = QMF_IM(pX[1]) >> 1;
 529
 530 #endif
 531
 532
 533         // dct4_kernel is DCT_IV without reordering which is done before and after FFT
 534         dct4_kernel(in_real1, in_imag1, out_real1, out_imag1);
 535         dct4_kernel(in_real2, in_imag2, out_real2, out_imag2);
 536
 537
 538         pring_buffer_1 = qmfs->v + qmfs->v_index;
 539         pring_buffer_3 = pring_buffer_1 + 1280;
 540 #ifdef PREFER_POINTERS
 541         pring_buffer_2 = pring_buffer_1 + 127;
 542         pring_buffer_4 = pring_buffer_1 + (1280 + 127);
 543 #endif // #ifdef PREFER_POINTERS
 544 //        ptemp_1 = x1;
 545 //        ptemp_2 = x2;
 546 #ifdef PREFER_POINTERS
 547         for (n = 0; n < 32; n ++)
 548         {
 549             //real_t x1 = *ptemp_1++;
 550             //real_t x2 = *ptemp_2++;
 551             // pring_buffer_3 and pring_buffer_4 are needed only for double ring buffer
 552             *pring_buffer_1++ = *pring_buffer_3++ = out_real2[n] - out_real1[n];
 553             *pring_buffer_2-- = *pring_buffer_4-- = out_real2[n] + out_real1[n];
 554             //x1 = *ptemp_1++;
 555             //x2 = *ptemp_2++;
 556             *pring_buffer_1++ = *pring_buffer_3++ = out_imag2[31-n] + out_imag1[31-n];
 557             *pring_buffer_2-- = *pring_buffer_4-- = out_imag2[31-n] - out_imag1[31-n];
 558         }
 559 #else // #ifdef PREFER_POINTERS
 560
 561         for (n = 0; n < 32; n++)
 562         {
 563             // pring_buffer_3 and pring_buffer_4 are needed only for double ring buffer
 564             pring_buffer_1[2*n]         = pring_buffer_3[2*n]         = out_real2[n] - out_real1[n];
 565             pring_buffer_1[127-2*n]     = pring_buffer_3[127-2*n]     = out_real2[n] + out_real1[n];
 566             pring_buffer_1[2*n+1]       = pring_buffer_3[2*n+1]       = out_imag2[31-n] + out_imag1[31-n];
 567             pring_buffer_1[127-(2*n+1)] = pring_buffer_3[127-(2*n+1)] = out_imag2[31-n] - out_imag1[31-n];
 568         }
 569
 570 #endif // #ifdef PREFER_POINTERS
 571
 572         pring_buffer_1 = qmfs->v + qmfs->v_index;
 573 #ifdef PREFER_POINTERS
 574         pring_buffer_2 = pring_buffer_1 + 192;
 575         pring_buffer_3 = pring_buffer_1 + 256;
 576         pring_buffer_4 = pring_buffer_1 + (256 + 192);
 577         pring_buffer_5 = pring_buffer_1 + 512;
 578         pring_buffer_6 = pring_buffer_1 + (512 + 192);
 579         pring_buffer_7 = pring_buffer_1 + 768;
 580         pring_buffer_8 = pring_buffer_1 + (768 + 192);
 581         pring_buffer_9 = pring_buffer_1 + 1024;
 582         pring_buffer_10 = pring_buffer_1 + (1024 + 192);
 583         pqmf_c_1 = qmf_c;
 584         pqmf_c_2 = qmf_c + 64;
 585         pqmf_c_3 = qmf_c + 128;
 586         pqmf_c_4 = qmf_c + 192;
 587         pqmf_c_5 = qmf_c + 256;
 588         pqmf_c_6 = qmf_c + 320;
 589         pqmf_c_7 = qmf_c + 384;
 590         pqmf_c_8 = qmf_c + 448;
 591         pqmf_c_9 = qmf_c + 512;
 592         pqmf_c_10 = qmf_c + 576;
 593 #endif // #ifdef PREFER_POINTERS
 594
 595         /* calculate 64 output samples and window */
 596         for (k = 0; k < 64; k++)
 597         {
 598 #ifdef PREFER_POINTERS
 599             output[out++] =
 600                 MUL_F(*pring_buffer_1++,  *pqmf_c_1++) +
 601                 MUL_F(*pring_buffer_2++,  *pqmf_c_2++) +
 602                 MUL_F(*pring_buffer_3++,  *pqmf_c_3++) +
 603                 MUL_F(*pring_buffer_4++,  *pqmf_c_4++) +
 604                 MUL_F(*pring_buffer_5++,  *pqmf_c_5++) +
 605                 MUL_F(*pring_buffer_6++,  *pqmf_c_6++) +
 606                 MUL_F(*pring_buffer_7++,  *pqmf_c_7++) +
 607                 MUL_F(*pring_buffer_8++,  *pqmf_c_8++) +
 608                 MUL_F(*pring_buffer_9++,  *pqmf_c_9++) +
 609                 MUL_F(*pring_buffer_10++, *pqmf_c_10++);
 610 #else // #ifdef PREFER_POINTERS
 611             output[out++] =
 612                 MUL_F(pring_buffer_1[k+0],          qmf_c[k+0])   +
 613                 MUL_F(pring_buffer_1[k+192],        qmf_c[k+64])  +
 614                 MUL_F(pring_buffer_1[k+256],        qmf_c[k+128]) +
 615                 MUL_F(pring_buffer_1[k+(256+192)],  qmf_c[k+192]) +
 616                 MUL_F(pring_buffer_1[k+512],        qmf_c[k+256]) +
 617                 MUL_F(pring_buffer_1[k+(512+192)],  qmf_c[k+320]) +
 618                 MUL_F(pring_buffer_1[k+768],        qmf_c[k+384]) +
 619                 MUL_F(pring_buffer_1[k+(768+192)],  qmf_c[k+448]) +
 620                 MUL_F(pring_buffer_1[k+1024],       qmf_c[k+512]) +
 621                 MUL_F(pring_buffer_1[k+(1024+192)], qmf_c[k+576]);
 622 #endif // #ifdef PREFER_POINTERS
 623         }
 624
 625         /* update ringbuffer index */
 626         qmfs->v_index -= 128;
 627         if (qmfs->v_index < 0)
 628             qmfs->v_index = (1280 - 128);
 629     }
 630 }
 631 #endif
 632
 633 #endif