apps/codecs/liba52/imdct.c

   1 /*
   2  * imdct.c
   3  * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org>
   4  * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
   5  *
   6  * The ifft algorithms in this file have been largely inspired by Dan
   7  * Bernstein's work, djbfft, available at http://cr.yp.to/djbfft.html
   8  *
   9  * This file is part of a52dec, a free ATSC A-52 stream decoder.
  10  * See http://liba52.sourceforge.net/ for updates.
  11  *
  12  * a52dec is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License as published by
  14  * the Free Software Foundation; either version 2 of the License, or
  15  * (at your option) any later version.
  16  *
  17  * a52dec is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  * GNU General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU General Public License
  23  * along with this program; if not, write to the Free Software
  24  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  25  */
  26
  27 #include "config-a52.h"
  28
  29 #include <math.h>
  30 #include <stdio.h>
  31 #ifdef LIBA52_DJBFFT
  32 #include <fftc4.h>
  33 #include <fftc8.h>
  34 #endif
  35 #ifndef M_PI
  36 #define M_PI 3.1415926535897932384626433832795029
  37 #endif
  38 #include <inttypes.h>
  39
  40 #include "a52.h"
  41 #include "a52_internal.h"
  42 #include "mm_accel.h"
  43
  44 typedef struct complex_s {
  45     sample_t real;
  46     sample_t imag;
  47 } complex_t;
  48
  49 static const uint8_t fftorder[] = {
  50       0,128, 64,192, 32,160,224, 96, 16,144, 80,208,240,112, 48,176,
  51       8,136, 72,200, 40,168,232,104,248,120, 56,184, 24,152,216, 88,
  52       4,132, 68,196, 36,164,228,100, 20,148, 84,212,244,116, 52,180,
  53     252,124, 60,188, 28,156,220, 92, 12,140, 76,204,236,108, 44,172,
  54       2,130, 66,194, 34,162,226, 98, 18,146, 82,210,242,114, 50,178,
  55      10,138, 74,202, 42,170,234,106,250,122, 58,186, 26,154,218, 90,
  56     254,126, 62,190, 30,158,222, 94, 14,142, 78,206,238,110, 46,174,
  57       6,134, 70,198, 38,166,230,102,246,118, 54,182, 22,150,214, 86
  58 };
  59
  60 /* Root values for IFFT */
  61 //static sample_t roots16[3];
  62 //static sample_t roots32[7];
  63 //static sample_t roots64[15];
  64 //static sample_t roots128[31];
  65
  66 /* Twiddle factors for IMDCT */
  67 //static complex_t pre1[128];
  68 //static complex_t post1[64];
  69 //static complex_t pre2[64];
  70 //static complex_t post2[32];
  71
  72 //static sample_t a52_imdct_window[256];
  73 #include "imdct_lookups.h"
  74
  75 static void (* ifft128) (complex_t * buf);
  76 static void (* ifft64) (complex_t * buf);
  77
  78 static inline void ifft2 (complex_t * buf)
  79 {
  80     sample_t r, i;
  81
  82     r = buf[0].real;
  83     i = buf[0].imag;
  84     buf[0].real += buf[1].real;
  85     buf[0].imag += buf[1].imag;
  86     buf[1].real = r - buf[1].real;
  87     buf[1].imag = i - buf[1].imag;
  88 }
  89
  90 static inline void ifft4 (complex_t * buf)
  91 {
  92     sample_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
  93
  94     tmp1 = buf[0].real + buf[1].real;
  95     tmp2 = buf[3].real + buf[2].real;
  96     tmp3 = buf[0].imag + buf[1].imag;
  97     tmp4 = buf[2].imag + buf[3].imag;
  98     tmp5 = buf[0].real - buf[1].real;
  99     tmp6 = buf[0].imag - buf[1].imag;
 100     tmp7 = buf[2].imag - buf[3].imag;
 101     tmp8 = buf[3].real - buf[2].real;
 102
 103     buf[0].real = tmp1 + tmp2;
 104     buf[0].imag = tmp3 + tmp4;
 105     buf[2].real = tmp1 - tmp2;
 106     buf[2].imag = tmp3 - tmp4;
 107     buf[1].real = tmp5 + tmp7;
 108     buf[1].imag = tmp6 + tmp8;
 109     buf[3].real = tmp5 - tmp7;
 110     buf[3].imag = tmp6 - tmp8;
 111 }
 112
 113 /* basic radix-2 ifft butterfly */
 114
 115 #define BUTTERFLY_0(t0,t1,W0,W1,d0,d1) do {     \
 116     t0 = MUL (W1, d1) + MUL (W0, d0);           \
 117     t1 = MUL (W0, d1) - MUL (W1, d0);           \
 118 } while (0)
 119
 120 /* radix-2 ifft butterfly with bias */
 121
 122 #define BUTTERFLY_B(t0,t1,W0,W1,d0,d1) do {     \
 123     t0 = BIAS (MUL (d1, W1) + MUL (d0, W0));    \
 124     t1 = BIAS (MUL (d1, W0) - MUL (d0, W1));    \
 125 } while (0)
 126
 127 /* the basic split-radix ifft butterfly */
 128
 129 #define BUTTERFLY(a0,a1,a2,a3,wr,wi) do {               \
 130     BUTTERFLY_0 (tmp5, tmp6, wr, wi, a2.real, a2.imag); \
 131     BUTTERFLY_0 (tmp8, tmp7, wr, wi, a3.imag, a3.real); \
 132     tmp1 = tmp5 + tmp7;                                 \
 133     tmp2 = tmp6 + tmp8;                                 \
 134     tmp3 = tmp6 - tmp8;                                 \
 135     tmp4 = tmp7 - tmp5;                                 \
 136     a2.real = a0.real - tmp1;                           \
 137     a2.imag = a0.imag - tmp2;                           \
 138     a3.real = a1.real - tmp3;                           \
 139     a3.imag = a1.imag - tmp4;                           \
 140     a0.real += tmp1;                                    \
 141     a0.imag += tmp2;                                    \
 142     a1.real += tmp3;                                    \
 143     a1.imag += tmp4;                                    \
 144 } while (0)
 145
 146 /* split-radix ifft butterfly, specialized for wr=1 wi=0 */
 147
 148 #define BUTTERFLY_ZERO(a0,a1,a2,a3) do {        \
 149     tmp1 = a2.real + a3.real;                   \
 150     tmp2 = a2.imag + a3.imag;                   \
 151     tmp3 = a2.imag - a3.imag;                   \
 152     tmp4 = a3.real - a2.real;                   \
 153     a2.real = a0.real - tmp1;                   \
 154     a2.imag = a0.imag - tmp2;                   \
 155     a3.real = a1.real - tmp3;                   \
 156     a3.imag = a1.imag - tmp4;                   \
 157     a0.real += tmp1;                            \
 158     a0.imag += tmp2;                            \
 159     a1.real += tmp3;                            \
 160     a1.imag += tmp4;                            \
 161 } while (0)
 162
 163 /* split-radix ifft butterfly, specialized for wr=wi */
 164
 165 #define BUTTERFLY_HALF(a0,a1,a2,a3,w) do {      \
 166     tmp5 = MUL (a2.real + a2.imag, w);          \
 167     tmp6 = MUL (a2.imag - a2.real, w);          \
 168     tmp7 = MUL (a3.real - a3.imag, w);          \
 169     tmp8 = MUL (a3.imag + a3.real, w);          \
 170     tmp1 = tmp5 + tmp7;                         \
 171     tmp2 = tmp6 + tmp8;                         \
 172     tmp3 = tmp6 - tmp8;                         \
 173     tmp4 = tmp7 - tmp5;                         \
 174     a2.real = a0.real - tmp1;                   \
 175     a2.imag = a0.imag - tmp2;                   \
 176     a3.real = a1.real - tmp3;                   \
 177     a3.imag = a1.imag - tmp4;                   \
 178     a0.real += tmp1;                            \
 179     a0.imag += tmp2;                            \
 180     a1.real += tmp3;                            \
 181     a1.imag += tmp4;                            \
 182 } while (0)
 183
 184 static inline void ifft8 (complex_t * buf)
 185 {
 186     sample_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
 187
 188     ifft4 (buf);
 189     ifft2 (buf + 4);
 190     ifft2 (buf + 6);
 191     BUTTERFLY_ZERO (buf[0], buf[2], buf[4], buf[6]);
 192     BUTTERFLY_HALF (buf[1], buf[3], buf[5], buf[7], roots16[1]);
 193 }
 194
 195 static void ifft_pass (complex_t * buf, const sample_t * weight, int n)
 196 {
 197     complex_t * buf1;
 198     complex_t * buf2;
 199     complex_t * buf3;
 200     sample_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
 201     int i;
 202
 203     buf++;
 204     buf1 = buf + n;
 205     buf2 = buf + 2 * n;
 206     buf3 = buf + 3 * n;
 207
 208     BUTTERFLY_ZERO (buf[-1], buf1[-1], buf2[-1], buf3[-1]);
 209
 210     i = n - 1;
 211
 212     do {
 213         BUTTERFLY (buf[0], buf1[0], buf2[0], buf3[0],
 214                    weight[0], weight[2*i-n]);
 215         buf++;
 216         buf1++;
 217         buf2++;
 218         buf3++;
 219         weight++;
 220     } while (--i);
 221 }
 222
 223 static void ifft16 (complex_t * buf)
 224 {
 225     ifft8 (buf);
 226     ifft4 (buf + 8);
 227     ifft4 (buf + 12);
 228     ifft_pass (buf, roots16, 4);
 229 }
 230
 231 static void ifft32 (complex_t * buf)
 232 {
 233     ifft16 (buf);
 234     ifft8 (buf + 16);
 235     ifft8 (buf + 24);
 236     ifft_pass (buf, roots32, 8);
 237 }
 238
 239 static void ifft64_c (complex_t * buf)
 240 {
 241     ifft32 (buf);
 242     ifft16 (buf + 32);
 243     ifft16 (buf + 48);
 244     ifft_pass (buf, roots64, 16);
 245 }
 246
 247 static void ifft128_c (complex_t * buf)
 248 {
 249     ifft32 (buf);
 250     ifft16 (buf + 32);
 251     ifft16 (buf + 48);
 252     ifft_pass (buf, roots64, 16);
 253
 254     ifft32 (buf + 64);
 255     ifft32 (buf + 96);
 256     ifft_pass (buf, roots128, 32);
 257 }
 258
 259 void a52_imdct_512 (sample_t * data, sample_t * delay, sample_t bias)
 260 {
 261     int i, k;
 262     sample_t t_r, t_i, a_r, a_i, b_r, b_i, w_1, w_2;
 263     const sample_t * window = a52_imdct_window;
 264     complex_t buf[128];
 265
 266     for (i = 0; i < 128; i++) {
 267         k = fftorder[i];
 268         t_r = pre1[i].real;
 269         t_i = pre1[i].imag;
 270         BUTTERFLY_0 (buf[i].real, buf[i].imag, t_r, t_i, data[k], data[255-k]);
 271     }
 272
 273     ifft128 (buf);
 274
 275     /* Post IFFT complex multiply plus IFFT complex conjugate*/
 276     /* Window and convert to real valued signal */
 277     for (i = 0; i < 64; i++) {
 278         /* y[n] = z[n] * (xcos1[n] + j * xsin1[n]) ; */
 279         t_r = post1[i].real;
 280         t_i = post1[i].imag;
 281         BUTTERFLY_0 (a_r, a_i, t_i, t_r, buf[i].imag, buf[i].real);
 282         BUTTERFLY_0 (b_r, b_i, t_r, t_i, buf[127-i].imag, buf[127-i].real);
 283
 284         w_1 = window[2*i];
 285         w_2 = window[255-2*i];
 286         BUTTERFLY_B (data[255-2*i], data[2*i], w_2, w_1, a_r, delay[2*i]);
 287         delay[2*i] = a_i;
 288
 289         w_1 = window[2*i+1];
 290         w_2 = window[254-2*i];
 291         BUTTERFLY_B (data[2*i+1], data[254-2*i], w_1, w_2, b_r, delay[2*i+1]);
 292         delay[2*i+1] = b_i;
 293     }
 294 }
 295
 296 void a52_imdct_256 (sample_t * data, sample_t * delay, sample_t bias)
 297 {
 298     int i, k;
 299     sample_t t_r, t_i, a_r, a_i, b_r, b_i, c_r, c_i, d_r, d_i, w_1, w_2;
 300     const sample_t * window = a52_imdct_window;
 301     complex_t buf1[64], buf2[64];
 302
 303     /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
 304     for (i = 0; i < 64; i++) {
 305         k = fftorder[i];
 306         t_r = pre2[i].real;
 307         t_i = pre2[i].imag;
 308         BUTTERFLY_0 (buf1[i].real, buf1[i].imag, t_r, t_i, data[k], data[254-k]);
 309         BUTTERFLY_0 (buf2[i].real, buf2[i].imag, t_r, t_i, data[k+1], data[255-k]);
 310     }
 311
 312     ifft64 (buf1);
 313     ifft64 (buf2);
 314
 315     /* Post IFFT complex multiply */
 316     /* Window and convert to real valued signal */
 317     for (i = 0; i < 32; i++) {
 318         /* y1[n] = z1[n] * (xcos2[n] + j * xs in2[n]) ; */
 319         t_r = post2[i].real;
 320         t_i = post2[i].imag;
 321         BUTTERFLY_0 (a_r, a_i, t_i, t_r, buf1[i].imag, buf1[i].real);
 322         BUTTERFLY_0 (b_r, b_i, t_r, t_i, buf1[63-i].imag, buf1[63-i].real);
 323         BUTTERFLY_0 (c_r, c_i, t_i, t_r, buf2[i].imag, buf2[i].real);
 324         BUTTERFLY_0 (d_r, d_i, t_r, t_i, buf2[63-i].imag, buf2[63-i].real);
 325
 326         w_1 = window[2*i];
 327         w_2 = window[255-2*i];
 328         BUTTERFLY_B (data[255-2*i], data[2*i], w_2, w_1, a_r, delay[2*i]);
 329         delay[2*i] = c_i;
 330
 331         w_1 = window[128+2*i];
 332         w_2 = window[127-2*i];
 333         BUTTERFLY_B (data[128+2*i], data[127-2*i], w_1, w_2, a_i, delay[127-2*i]);
 334         delay[127-2*i] = c_r;
 335
 336         w_1 = window[2*i+1];
 337         w_2 = window[254-2*i];
 338         BUTTERFLY_B (data[254-2*i], data[2*i+1], w_2, w_1, b_i, delay[2*i+1]);
 339         delay[2*i+1] = d_r;
 340
 341         w_1 = window[129+2*i];
 342         w_2 = window[126-2*i];
 343         BUTTERFLY_B (data[129+2*i], data[126-2*i], w_1, w_2, b_r, delay[126-2*i]);
 344         delay[126-2*i] = d_i;
 345     }
 346 }
 347
 348 /*
 349 static double besselI0 (double x)
 350 {
 351     double bessel = 1;
 352     int i = 100;
 353
 354     do
 355         bessel = bessel * x / (i * i) + 1;
 356     while (--i);
 357     return bessel;
 358 }
 359 */
 360
 361 void a52_imdct_init (uint32_t mm_accel)
 362 {
 363   (void)mm_accel;
 364 /*    int i, k;
 365     double sum;
 366     double local_imdct_window[256];*/
 367
 368     /* compute imdct window - kaiser-bessel derived window, alpha = 5.0 */
 369     /*    sum = 0;
 370     for (i = 0; i < 256; i++) {
 371         sum += besselI0 (i * (256 - i) * (5 * M_PI / 256) * (5 * M_PI / 256));
 372         local_imdct_window[i] = sum;
 373     }
 374     sum++;
 375     */
 376     /*    for (i = 0; i < 256; i++)
 377         a52_imdct_window[i] = SAMPLE (sqrt (local_imdct_window[i] / sum));
 378
 379     printf("static sample_t a52_imdct_window[256]={");
 380     for (i=0;i<256;i++) {
 381       if ((i % 16)==0) { printf("\n"); }
 382       printf("%d,",a52_imdct_window[i]);
 383     }
 384     printf("\n}\n");
 385     */
 386
 387     /*    for (i = 0; i < 3; i++)
 388         roots16[i] = SAMPLE (cos ((M_PI / 8) * (i + 1)));
 389
 390     printf("static sample_t roots16[3]={%d,%d,%d};\n\n",roots16[0],roots16[1],roots16[2]);
 391
 392     for (i = 0; i < 7; i++)
 393         roots32[i] = SAMPLE (cos ((M_PI / 16) * (i + 1)));
 394
 395     printf("static sample_t roots32[7]={");
 396     for (i=0;i<7;i++) { printf("%d%s",roots32[i],(i < 6 ? "," : "")); }
 397     printf("};\n");
 398
 399     for (i = 0; i < 15; i++)
 400         roots64[i] = SAMPLE (cos ((M_PI / 32) * (i + 1)));
 401
 402     printf("static sample_t roots64[15]={");
 403     for (i=0;i<15;i++) { printf("%d%s",roots64[i],(i < 14 ? "," : "")); }
 404     printf("};\n");
 405
 406     for (i = 0; i < 31; i++)
 407         roots128[i] = SAMPLE (cos ((M_PI / 64) * (i + 1)));
 408
 409     printf("static sample_t roots128[31]={");
 410     for (i=0;i<31;i++) { printf("%d%s",roots128[i],(i < 30 ? "," : "")); }
 411     printf("};\n");
 412     */
 413     /*
 414     for (i = 0; i < 64; i++) {
 415         k = fftorder[i] / 2 + 64;
 416         pre1[i].real = SAMPLE (cos ((M_PI / 256) * (k - 0.25)));
 417         pre1[i].imag = SAMPLE (sin ((M_PI / 256) * (k - 0.25)));
 418     }
 419
 420     for (i = 64; i < 128; i++) {
 421         k = fftorder[i] / 2 + 64;
 422         pre1[i].real = SAMPLE (-cos ((M_PI / 256) * (k - 0.25)));
 423         pre1[i].imag = SAMPLE (-sin ((M_PI / 256) * (k - 0.25)));
 424     }
 425
 426     printf("static complex_t pre1[128]={");
 427     for (i=0;i<128;i++) { printf("{%d,%d}%s",pre1[i].real,pre1[i].imag,(i < 127 ? "," : "")); }
 428     printf("};\n");
 429     */
 430     /*
 431     for (i = 0; i < 64; i++) {
 432         post1[i].real = SAMPLE (cos ((M_PI / 256) * (i + 0.5)));
 433         post1[i].imag = SAMPLE (sin ((M_PI / 256) * (i + 0.5)));
 434     }
 435
 436     printf("static complex_t post1[64]={");
 437     for (i=0;i<64;i++) { printf("{%d,%d}%s",post1[i].real,post1[i].imag,(i < 63 ? "," : "")); }
 438     printf("};\n");
 439     */
 440
 441     /*
 442     for (i = 0; i < 64; i++) {
 443         k = fftorder[i] / 4;
 444         pre2[i].real = SAMPLE (cos ((M_PI / 128) * (k - 0.25)));
 445         pre2[i].imag = SAMPLE (sin ((M_PI / 128) * (k - 0.25)));
 446     }
 447
 448     printf("static complex_t pre2[64]={");
 449     for (i=0;i<64;i++) { printf("{%d,%d}%s",pre2[i].real,pre2[i].imag,(i < 63 ? "," : "")); }
 450     printf("};\n");
 451
 452     for (i = 0; i < 32; i++) {
 453         post2[i].real = SAMPLE (cos ((M_PI / 128) * (i + 0.5)));
 454         post2[i].imag = SAMPLE (sin ((M_PI / 128) * (i + 0.5)));
 455     }
 456
 457     printf("static complex_t post2[32]={");
 458     for (i=0;i<32;i++) { printf("{%d,%d}%s",post2[i].real,post2[i].imag,(i < 31 ? "," : "")); }
 459     printf("};\n");
 460     */
 461
 462 #ifdef LIBA52_DJBFFT
 463     if (mm_accel & MM_ACCEL_DJBFFT) {
 464 #ifndef LIBA52_DOUBLE
 465         ifft128 = (void (*) (complex_t *)) fftc4_un128;
 466         ifft64 = (void (*) (complex_t *)) fftc4_un64;
 467 #else
 468         ifft128 = (void (*) (complex_t *)) fftc8_un128;
 469         ifft64 = (void (*) (complex_t *)) fftc8_un64;
 470 #endif
 471     } else
 472 #endif
 473     {
 474         ifft128 = ifft128_c;
 475         ifft64 = ifft64_c;
 476     }
 477 }