git.sesse.net Git - vlc/blob - src/ac3_decoder/ac3_imdct.c

   1 /*****************************************************************************
   2  * ac3_imdct.c: ac3 DCT
   3  *****************************************************************************
   4  * Copyright (C) 1999, 2000 VideoLAN
   5  * $Id: ac3_imdct.c,v 1.14 2001/03/21 13:42:34 sam Exp $
   6  *
   7  * Authors: Michel Kaempf <maxx@via.ecp.fr>
   8  *          Aaron Holtzman <aholtzma@engr.uvic.ca>
   9  *          Renaud Dartus <reno@videolan.org>
  10  *
  11  * This program is free software; you can redistribute it and/or modify
  12  * it under the terms of the GNU General Public License as published by
  13  * the Free Software Foundation; either version 2 of the License, or
  14  * (at your option) any later version.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  * GNU General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU General Public License
  22  * along with this program; if not, write to the Free Software
  23  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
  24  *****************************************************************************/
  25
  26 #include "defs.h"
  27
  28 #include <math.h>
  29 #include <stdio.h>
  30
  31 #include "int_types.h"
  32 #include "ac3_decoder.h"
  33 #include "ac3_internal.h"
  34
  35 #include "ac3_downmix.h"
  36
  37 void imdct_do_256(imdct_t * p_imdct, float x[],float y[], int id);
  38 void imdct_do_512(imdct_t * p_imdct, float x[],float y[], int id);
  39
  40 /* 128 point bit-reverse LUT */
  41 static const u8 bit_reverse_512[] = {
  42     0x00, 0x40, 0x20, 0x60, 0x10, 0x50, 0x30, 0x70,
  43     0x08, 0x48, 0x28, 0x68, 0x18, 0x58, 0x38, 0x78,
  44     0x04, 0x44, 0x24, 0x64, 0x14, 0x54, 0x34, 0x74,
  45     0x0c, 0x4c, 0x2c, 0x6c, 0x1c, 0x5c, 0x3c, 0x7c,
  46     0x02, 0x42, 0x22, 0x62, 0x12, 0x52, 0x32, 0x72,
  47     0x0a, 0x4a, 0x2a, 0x6a, 0x1a, 0x5a, 0x3a, 0x7a,
  48     0x06, 0x46, 0x26, 0x66, 0x16, 0x56, 0x36, 0x76,
  49     0x0e, 0x4e, 0x2e, 0x6e, 0x1e, 0x5e, 0x3e, 0x7e,
  50     0x01, 0x41, 0x21, 0x61, 0x11, 0x51, 0x31, 0x71,
  51     0x09, 0x49, 0x29, 0x69, 0x19, 0x59, 0x39, 0x79,
  52     0x05, 0x45, 0x25, 0x65, 0x15, 0x55, 0x35, 0x75,
  53     0x0d, 0x4d, 0x2d, 0x6d, 0x1d, 0x5d, 0x3d, 0x7d,
  54     0x03, 0x43, 0x23, 0x63, 0x13, 0x53, 0x33, 0x73,
  55     0x0b, 0x4b, 0x2b, 0x6b, 0x1b, 0x5b, 0x3b, 0x7b,
  56     0x07, 0x47, 0x27, 0x67, 0x17, 0x57, 0x37, 0x77,
  57     0x0f, 0x4f, 0x2f, 0x6f, 0x1f, 0x5f, 0x3f, 0x7f};
  58
  59 static const u8 bit_reverse_256[] = {
  60     0x00, 0x20, 0x10, 0x30, 0x08, 0x28, 0x18, 0x38,
  61     0x04, 0x24, 0x14, 0x34, 0x0c, 0x2c, 0x1c, 0x3c,
  62     0x02, 0x22, 0x12, 0x32, 0x0a, 0x2a, 0x1a, 0x3a,
  63     0x06, 0x26, 0x16, 0x36, 0x0e, 0x2e, 0x1e, 0x3e,
  64     0x01, 0x21, 0x11, 0x31, 0x09, 0x29, 0x19, 0x39,
  65     0x05, 0x25, 0x15, 0x35, 0x0d, 0x2d, 0x1d, 0x3d,
  66     0x03, 0x23, 0x13, 0x33, 0x0b, 0x2b, 0x1b, 0x3b,
  67     0x07, 0x27, 0x17, 0x37, 0x0f, 0x2f, 0x1f, 0x3f};
  68
  69 /* Windowing function for Modified DCT - Thank you acroread */
  70 static float window[] = {
  71     0.00014, 0.00024, 0.00037, 0.00051, 0.00067, 0.00086, 0.00107, 0.00130,
  72     0.00157, 0.00187, 0.00220, 0.00256, 0.00297, 0.00341, 0.00390, 0.00443,
  73     0.00501, 0.00564, 0.00632, 0.00706, 0.00785, 0.00871, 0.00962, 0.01061,
  74     0.01166, 0.01279, 0.01399, 0.01526, 0.01662, 0.01806, 0.01959, 0.02121,
  75     0.02292, 0.02472, 0.02662, 0.02863, 0.03073, 0.03294, 0.03527, 0.03770,
  76     0.04025, 0.04292, 0.04571, 0.04862, 0.05165, 0.05481, 0.05810, 0.06153,
  77     0.06508, 0.06878, 0.07261, 0.07658, 0.08069, 0.08495, 0.08935, 0.09389,
  78     0.09859, 0.10343, 0.10842, 0.11356, 0.11885, 0.12429, 0.12988, 0.13563,
  79     0.14152, 0.14757, 0.15376, 0.16011, 0.16661, 0.17325, 0.18005, 0.18699,
  80     0.19407, 0.20130, 0.20867, 0.21618, 0.22382, 0.23161, 0.23952, 0.24757,
  81     0.25574, 0.26404, 0.27246, 0.28100, 0.28965, 0.29841, 0.30729, 0.31626,
  82     0.32533, 0.33450, 0.34376, 0.35311, 0.36253, 0.37204, 0.38161, 0.39126,
  83     0.40096, 0.41072, 0.42054, 0.43040, 0.44030, 0.45023, 0.46020, 0.47019,
  84     0.48020, 0.49022, 0.50025, 0.51028, 0.52031, 0.53033, 0.54033, 0.55031,
  85     0.56026, 0.57019, 0.58007, 0.58991, 0.59970, 0.60944, 0.61912, 0.62873,
  86     0.63827, 0.64774, 0.65713, 0.66643, 0.67564, 0.68476, 0.69377, 0.70269,
  87     0.71150, 0.72019, 0.72877, 0.73723, 0.74557, 0.75378, 0.76186, 0.76981,
  88     0.77762, 0.78530, 0.79283, 0.80022, 0.80747, 0.81457, 0.82151, 0.82831,
  89     0.83496, 0.84145, 0.84779, 0.85398, 0.86001, 0.86588, 0.87160, 0.87716,
  90     0.88257, 0.88782, 0.89291, 0.89785, 0.90264, 0.90728, 0.91176, 0.91610,
  91     0.92028, 0.92432, 0.92822, 0.93197, 0.93558, 0.93906, 0.94240, 0.94560,
  92     0.94867, 0.95162, 0.95444, 0.95713, 0.95971, 0.96217, 0.96451, 0.96674,
  93     0.96887, 0.97089, 0.97281, 0.97463, 0.97635, 0.97799, 0.97953, 0.98099,
  94     0.98236, 0.98366, 0.98488, 0.98602, 0.98710, 0.98811, 0.98905, 0.98994,
  95     0.99076, 0.99153, 0.99225, 0.99291, 0.99353, 0.99411, 0.99464, 0.99513,
  96     0.99558, 0.99600, 0.99639, 0.99674, 0.99706, 0.99736, 0.99763, 0.99788,
  97     0.99811, 0.99831, 0.99850, 0.99867, 0.99882, 0.99895, 0.99908, 0.99919,
  98     0.99929, 0.99938, 0.99946, 0.99953, 0.99959, 0.99965, 0.99969, 0.99974,
  99     0.99978, 0.99981, 0.99984, 0.99986, 0.99988, 0.99990, 0.99992, 0.99993,
 100     0.99994, 0.99995, 0.99996, 0.99997, 0.99998, 0.99998, 0.99998, 0.99999,
 101     0.99999, 0.99999, 0.99999, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000,
 102     1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000, 1.00000 };
 103
 104 static __inline__ void swap_cmplx(complex_t *a, complex_t *b)
 105 {
 106     complex_t tmp;
 107
 108     tmp = *a;
 109     *a = *b;
 110     *b = tmp;
 111 }
 112
 113 static __inline__ complex_t cmplx_mult(complex_t a, complex_t b)
 114 {
 115     complex_t ret;
 116
 117     ret.real = a.real * b.real - a.imag * b.imag;
 118     ret.imag = a.real * b.imag + a.imag * b.real;
 119
 120     return ret;
 121 }
 122
 123 void imdct_init(imdct_t * p_imdct)
 124 {
 125     int i,k;
 126     complex_t angle_step;
 127     complex_t current_angle;
 128
 129     /* Twiddle factors to turn IFFT into IMDCT */
 130     for (i=0; i < N/4; i++) {
 131         p_imdct->xcos1[i] = -cos(2 * M_PI * (8*i+1)/(8*N)) ;
 132         p_imdct->xsin1[i] = -sin(2 * M_PI * (8*i+1)/(8*N)) ;
 133     }
 134
 135     /* More twiddle factors to turn IFFT into IMDCT */
 136     for (i=0; i < N/8; i++) {
 137         p_imdct->xcos2[i] = -cos(2 * M_PI * (8*i+1)/(4*N)) ;
 138         p_imdct->xsin2[i] = -sin(2 * M_PI * (8*i+1)/(4*N)) ;
 139     }
 140
 141     /* Canonical twiddle factors for FFT */
 142     p_imdct->w[0] = p_imdct->w_1;
 143     p_imdct->w[1] = p_imdct->w_2;
 144     p_imdct->w[2] = p_imdct->w_4;
 145     p_imdct->w[3] = p_imdct->w_8;
 146     p_imdct->w[4] = p_imdct->w_16;
 147     p_imdct->w[5] = p_imdct->w_32;
 148     p_imdct->w[6] = p_imdct->w_64;
 149
 150     for (i = 0; i < 7; i++) {
 151         angle_step.real = cos(-2.0f * M_PI / (1 << (i+1)));
 152         angle_step.imag = sin(-2.0f * M_PI / (1 << (i+1)));
 153
 154         current_angle.real = 1.0f;
 155         current_angle.imag = 0.0f;
 156
 157         for (k = 0; k < 1 << i; k++) {
 158             p_imdct->w[i][k] = current_angle;
 159             current_angle = cmplx_mult(current_angle,angle_step);
 160         }
 161     }
 162 }
 163
 164 void imdct (ac3dec_t * p_ac3dec, s16 * buffer)
 165 {
 166     int i, i_stream_done;
 167     int doable = 0;
 168     void (*do_imdct)(imdct_t * p_imdct, float x[],float y[], int id);
 169
 170     /* Test if dm in frequency is doable */
 171     if ( !(doable = p_ac3dec->audblk.blksw[0]) )
 172         do_imdct = imdct_do_512;
 173     else
 174         do_imdct = imdct_do_256;
 175
 176     /* Downmix in the frequency domain if all the channes use the same imdct */
 177     for (i=0; i < p_ac3dec->bsi.nfchans; i++)
 178     {
 179         if ( doable != p_ac3dec->audblk.blksw[i] )
 180         {
 181             do_imdct = NULL;
 182             break;
 183         }
 184     }
 185
 186     if (do_imdct)
 187     {
 188         i_stream_done = downmix(p_ac3dec, p_ac3dec->coeffs.fbw[0], buffer);
 189         do_imdct(&p_ac3dec->imdct,p_ac3dec->coeffs.fbw[0],p_ac3dec->samples.channel[0], 0);
 190         do_imdct(&p_ac3dec->imdct, p_ac3dec->coeffs.fbw[1],p_ac3dec->samples.channel[1], 1);
 191     } else {
 192         for (i=0; i<p_ac3dec->bsi.nfchans;i++) {
 193             if (p_ac3dec->audblk.blksw[i])
 194                 imdct_do_256(&p_ac3dec->imdct, p_ac3dec->coeffs.fbw[i],p_ac3dec->samples.channel[i], i);
 195             else
 196                 imdct_do_512(&p_ac3dec->imdct, p_ac3dec->coeffs.fbw[i],p_ac3dec->samples.channel[i], i);
 197         }
 198         i_stream_done = downmix(p_ac3dec, p_ac3dec->samples.channel[0], buffer);
 199     }
 200
 201     if ( !i_stream_done ) /* We have to stream sample */
 202     {
 203         stream_sample_2ch_to_s16_c(buffer, p_ac3dec->samples.channel[0],
 204                 p_ac3dec->samples.channel[1]);
 205     } else {
 206         stream_sample_1ch_to_s16_c(buffer, p_ac3dec->samples.channel[0]);
 207     }
 208
 209     /* XXX?? We don't bother with the IMDCT for the LFE as it's currently
 210      * unused. */
 211     //if (bsi->lfeon)
 212     //    imdct_do_512(coeffs->lfe,samples->channel[5],delay[5]);
 213 }
 214
 215 void imdct_do_512(imdct_t * p_imdct, float x[], float y[], int id)
 216 {
 217     int i,k;
 218     int p,q;
 219     int m;
 220     int two_m;
 221     int two_m_plus_one;
 222
 223     float tmp_a_i;
 224     float tmp_a_r;
 225     float tmp_b_i;
 226     float tmp_b_r;
 227
 228
 229     float *y_ptr;
 230     float *delay_ptr;
 231     float *window_ptr;
 232
 233     /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
 234     for (i=0; i < N/4; i++) {
 235         /* z[i] = (X[N/2-2*i-1] + j * X[2*i]) * (xcos1[i] + j * xsin1[i]) ; */
 236         p_imdct->buf[i].real =  (x[N/2-2*i-1] * p_imdct->xcos1[i])  - (x[2*i]       * p_imdct->xsin1[i]);
 237         p_imdct->buf[i].imag = -((x[2*i]       * p_imdct->xcos1[i])  + (x[N/2-2*i-1] * p_imdct->xsin1[i]));
 238     }
 239
 240     /* Bit reversed shuffling */
 241     for (i=0; i<N/4; i++) {
 242         k = bit_reverse_512[i];
 243         if (k < i)
 244             swap_cmplx(&p_imdct->buf[i],&p_imdct->buf[k]);
 245     }
 246
 247     /* FFT Merge */
 248     for (m=0; m < 7; m++) {
 249         two_m = (1 << m);
 250         two_m_plus_one = (1 << (m+1));
 251
 252         for (k = 0; k < two_m; k++) {
 253             for (i = 0; i < 128; i += two_m_plus_one) {
 254                 p = k + i;
 255                 q = p + two_m;
 256                 tmp_a_r = p_imdct->buf[p].real;
 257                 tmp_a_i = p_imdct->buf[p].imag;
 258                 tmp_b_r = p_imdct->buf[q].real * p_imdct->w[m][k].real - p_imdct->buf[q].imag * p_imdct->w[m][k].imag;
 259                 tmp_b_i = p_imdct->buf[q].imag * p_imdct->w[m][k].real + p_imdct->buf[q].real * p_imdct->w[m][k].imag;
 260                 p_imdct->buf[p].real = tmp_a_r + tmp_b_r;
 261                 p_imdct->buf[p].imag =  tmp_a_i + tmp_b_i;
 262                 p_imdct->buf[q].real = tmp_a_r - tmp_b_r;
 263                 p_imdct->buf[q].imag =  tmp_a_i - tmp_b_i;
 264             }
 265         }
 266     }
 267
 268     /* Post IFFT complex multiply  plus IFFT complex conjugate*/
 269     for (i=0; i < N/4; i++) {
 270         /* y[n] = z[n] * (xcos1[n] + j * xsin1[n]) ; */
 271         tmp_a_r =   p_imdct->buf[i].real;
 272         tmp_a_i = - p_imdct->buf[i].imag;
 273         p_imdct->buf[i].real =(tmp_a_r * p_imdct->xcos1[i])  - (tmp_a_i  * p_imdct->xsin1[i]);
 274         p_imdct->buf[i].imag =(tmp_a_r * p_imdct->xsin1[i])  + (tmp_a_i  * p_imdct->xcos1[i]);
 275     }
 276
 277     y_ptr = y;
 278     delay_ptr = p_imdct->delay[id];
 279     window_ptr = window;
 280     /* Window and convert to real valued signal */
 281     for (i=0; i<N/8; i++) {
 282         *y_ptr++   = 2.0f * (-p_imdct->buf[N/8+i].imag   * *window_ptr++ + *delay_ptr++);
 283         *y_ptr++   = 2.0f * (p_imdct->buf[N/8-i-1].real * *window_ptr++ + *delay_ptr++);
 284     }
 285
 286     for (i=0; i<N/8; i++) {
 287         *y_ptr++  = 2.0f * (-p_imdct->buf[i].real       * *window_ptr++ + *delay_ptr++);
 288         *y_ptr++  = 2.0f * (p_imdct->buf[N/4-i-1].imag * *window_ptr++ + *delay_ptr++);
 289     }
 290
 291     /* The trailing edge of the window goes into the delay line */
 292     delay_ptr = p_imdct->delay[id];
 293
 294     for (i=0; i<N/8; i++) {
 295         *delay_ptr++  = -p_imdct->buf[N/8+i].real   * *--window_ptr;
 296         *delay_ptr++  =  p_imdct->buf[N/8-i-1].imag * *--window_ptr;
 297     }
 298
 299     for (i=0; i<N/8; i++) {
 300         *delay_ptr++  =  p_imdct->buf[i].imag       * *--window_ptr;
 301         *delay_ptr++  = -p_imdct->buf[N/4-i-1].real * *--window_ptr;
 302     }
 303 }
 304
 305 void imdct_do_256(imdct_t * p_imdct,float x[],float y[], int id)
 306 {
 307     int i,k;
 308     int p,q;
 309     int m;
 310     int two_m;
 311     int two_m_plus_one;
 312
 313     float tmp_a_i;
 314     float tmp_a_r;
 315     float tmp_b_i;
 316     float tmp_b_r;
 317
 318     complex_t *buf_1, *buf_2;
 319
 320     buf_1 = &p_imdct->buf[0];
 321     buf_2 = &p_imdct->buf[64];
 322
 323     /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
 324     for (k=0; k<N/8; k++) {
 325         /* X1[k] = X[2*k]  */
 326         /* X2[k] = X[2*k+1]     */
 327
 328         p = 2 * (N/4-2*k-1);
 329         q = 2 * (2 * k);
 330
 331        /* Z1[k] = (X1[N/4-2*k-1] + j * X1[2*k]) * (xcos2[k] + j * xsin2[k]); */
 332         buf_1[k].real =    x[p] * p_imdct->xcos2[k] - x[q] * p_imdct->xsin2[k];
 333         buf_1[k].imag = - (x[q] * p_imdct->xcos2[k] + x[p] * p_imdct->xsin2[k]);
 334        /* Z2[k] = (X2[N/4-2*k-1] + j * X2[2*k]) * (xcos2[k] + j * xsin2[k]); */
 335         buf_2[k].real =    x[p + 1] * p_imdct->xcos2[k] - x[q + 1] * p_imdct->xsin2[k];
 336         buf_2[k].imag = - (x[q + 1] * p_imdct->xcos2[k] + x[p + 1] * p_imdct->xsin2[k]);
 337     }
 338
 339     /* IFFT Bit reversed shuffling */
 340     for (i=0; i<N/8; i++) {
 341         k = bit_reverse_256[i];
 342         if (k < i) {
 343             swap_cmplx(&buf_1[i],&buf_1[k]);
 344             swap_cmplx(&buf_2[i],&buf_2[k]);
 345         }
 346     }
 347
 348     /* FFT Merge */
 349     for (m=0; m < 6; m++) {
 350         two_m = (1 << m);
 351         two_m_plus_one = (1 << (m+1));
 352
 353         for (k = 0; k < two_m; k++) {
 354             for (i = 0; i < 64; i += two_m_plus_one) {
 355                 p = k + i;
 356                 q = p + two_m;
 357                 /* Do block 1 */
 358                 tmp_a_r = buf_1[p].real;
 359                 tmp_a_i = buf_1[p].imag;
 360                 tmp_b_r = buf_1[q].real * p_imdct->w[m][k].real - buf_1[q].imag * p_imdct->w[m][k].imag;
 361                 tmp_b_i = buf_1[q].imag * p_imdct->w[m][k].real + buf_1[q].real * p_imdct->w[m][k].imag;
 362                 buf_1[p].real = tmp_a_r + tmp_b_r;
 363                 buf_1[p].imag =  tmp_a_i + tmp_b_i;
 364                 buf_1[q].real = tmp_a_r - tmp_b_r;
 365                 buf_1[q].imag =  tmp_a_i - tmp_b_i;
 366
 367                 /* Do block 2 */
 368                 tmp_a_r = buf_2[p].real;
 369                 tmp_a_i = buf_2[p].imag;
 370                 tmp_b_r = buf_2[q].real * p_imdct->w[m][k].real - buf_2[q].imag * p_imdct->w[m][k].imag;
 371                 tmp_b_i = buf_2[q].imag * p_imdct->w[m][k].real + buf_2[q].real * p_imdct->w[m][k].imag;
 372                 buf_2[p].real = tmp_a_r + tmp_b_r;
 373                 buf_2[p].imag =  tmp_a_i + tmp_b_i;
 374                 buf_2[q].real = tmp_a_r - tmp_b_r;
 375                 buf_2[q].imag =  tmp_a_i - tmp_b_i;
 376             }
 377         }
 378     }
 379
 380     /* Post IFFT complex multiply */
 381     for (i=0; i < N/8; i++) {
 382         /* y1[n] = z1[n] * (xcos2[n] + j * xs in2[n]) ; */
 383         tmp_a_r =   buf_1[i].real;
 384         tmp_a_i = - buf_1[i].imag;
 385         buf_1[i].real =(tmp_a_r * p_imdct->xcos2[i])  - (tmp_a_i  * p_imdct->xsin2[i]);
 386         buf_1[i].imag =(tmp_a_r * p_imdct->xsin2[i])  + (tmp_a_i  * p_imdct->xcos2[i]);
 387         /* y2[n] = z2[n] * (xcos2[n] + j * xsin2[n]) ; */
 388         tmp_a_r =   buf_2[i].real;
 389         tmp_a_i = - buf_2[i].imag;
 390         buf_2[i].real =(tmp_a_r * p_imdct->xcos2[i])  - (tmp_a_i  * p_imdct->xsin2[i]);
 391         buf_2[i].imag =(tmp_a_r * p_imdct->xsin2[i])  + (tmp_a_i  * p_imdct->xcos2[i]);
 392     }
 393
 394     /* Window and convert to real valued signal */
 395     for (i=0; i<N/8; i++) {
 396         y[2*i]         = -buf_1[i].imag       * window[2*i];
 397         y[2*i+1]       =  buf_1[N/8-i-1].real * window[2*i+1];
 398         y[N/4+2*i]     = -buf_1[i].real       * window[N/4+2*i];
 399         y[N/4+2*i+1]   =  buf_1[N/8-i-1].imag * window[N/4+2*i+1];
 400         y[N/2+2*i]     = -buf_2[i].real       * window[N/2-2*i-1];
 401         y[N/2+2*i+1]   =  buf_2[N/8-i-1].imag * window[N/2-2*i-2];
 402         y[3*N/4+2*i]   =  buf_2[i].imag       * window[N/4-2*i-1];
 403         y[3*N/4+2*i+1] = -buf_2[N/8-i-1].real * window[N/4-2*i-2];
 404     }
 405
 406     /* Overlap and add */
 407     for (i=0; i<N/2; i++) {
 408         y[i] = 2 * (y[i] + p_imdct->delay[id][i]);
 409         p_imdct->delay[id][i] = y[N/2+i];
 410     }
 411 }