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
  76 /*
  77 static void (* ifft128) (complex_t * buf);
  78 static void (* ifft64) (complex_t * buf);
  79
  80 static inline void ifft2 (complex_t * buf)
  81 {
  82     sample_t r, i;
  83
  84     r = buf[0].real;
  85     i = buf[0].imag;
  86     buf[0].real += buf[1].real;
  87     buf[0].imag += buf[1].imag;
  88     buf[1].real = r - buf[1].real;
  89     buf[1].imag = i - buf[1].imag;
  90 }
  91
  92 static inline void ifft4 (complex_t * buf)
  93 {
  94     sample_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
  95
  96     tmp1 = buf[0].real + buf[1].real;
  97     tmp2 = buf[3].real + buf[2].real;
  98     tmp3 = buf[0].imag + buf[1].imag;
  99     tmp4 = buf[2].imag + buf[3].imag;
 100     tmp5 = buf[0].real - buf[1].real;
 101     tmp6 = buf[0].imag - buf[1].imag;
 102     tmp7 = buf[2].imag - buf[3].imag;
 103     tmp8 = buf[3].real - buf[2].real;
 104
 105     buf[0].real = tmp1 + tmp2;
 106     buf[0].imag = tmp3 + tmp4;
 107     buf[2].real = tmp1 - tmp2;
 108     buf[2].imag = tmp3 - tmp4;
 109     buf[1].real = tmp5 + tmp7;
 110     buf[1].imag = tmp6 + tmp8;
 111     buf[3].real = tmp5 - tmp7;
 112     buf[3].imag = tmp6 - tmp8;
 113 }
 114 */
 115 /* basic radix-2 ifft butterfly */
 116
 117 #define BUTTERFLY_0(t0,t1,W0,W1,d0,d1) do {     \
 118     t0 = MUL (W1, d1) + MUL (W0, d0);           \
 119     t1 = MUL (W0, d1) - MUL (W1, d0);           \
 120 } while (0)
 121
 122 /* radix-2 ifft butterfly with bias */
 123
 124 #define BUTTERFLY_B(t0,t1,W0,W1,d0,d1) do {     \
 125     t0 = BIAS (MUL (d1, W1) + MUL (d0, W0));    \
 126     t1 = BIAS (MUL (d1, W0) - MUL (d0, W1));    \
 127 } while (0)
 128
 129 /* the basic split-radix ifft butterfly */
 130
 131 #define BUTTERFLY(a0,a1,a2,a3,wr,wi) do {               \
 132     BUTTERFLY_0 (tmp5, tmp6, wr, wi, a2.real, a2.imag); \
 133     BUTTERFLY_0 (tmp8, tmp7, wr, wi, a3.imag, a3.real); \
 134     tmp1 = tmp5 + tmp7;                                 \
 135     tmp2 = tmp6 + tmp8;                                 \
 136     tmp3 = tmp6 - tmp8;                                 \
 137     tmp4 = tmp7 - tmp5;                                 \
 138     a2.real = a0.real - tmp1;                           \
 139     a2.imag = a0.imag - tmp2;                           \
 140     a3.real = a1.real - tmp3;                           \
 141     a3.imag = a1.imag - tmp4;                           \
 142     a0.real += tmp1;                                    \
 143     a0.imag += tmp2;                                    \
 144     a1.real += tmp3;                                    \
 145     a1.imag += tmp4;                                    \
 146 } while (0)
 147
 148 /* split-radix ifft butterfly, specialized for wr=1 wi=0 */
 149
 150 #define BUTTERFLY_ZERO(a0,a1,a2,a3) do {        \
 151     tmp1 = a2.real + a3.real;                   \
 152     tmp2 = a2.imag + a3.imag;                   \
 153     tmp3 = a2.imag - a3.imag;                   \
 154     tmp4 = a3.real - a2.real;                   \
 155     a2.real = a0.real - tmp1;                   \
 156     a2.imag = a0.imag - tmp2;                   \
 157     a3.real = a1.real - tmp3;                   \
 158     a3.imag = a1.imag - tmp4;                   \
 159     a0.real += tmp1;                            \
 160     a0.imag += tmp2;                            \
 161     a1.real += tmp3;                            \
 162     a1.imag += tmp4;                            \
 163 } while (0)
 164
 165 /* split-radix ifft butterfly, specialized for wr=wi */
 166 /*
 167 #define BUTTERFLY_HALF(a0,a1,a2,a3,w) do {      \
 168     tmp5 = MUL (a2.real + a2.imag, w);          \
 169     tmp6 = MUL (a2.imag - a2.real, w);          \
 170     tmp7 = MUL (a3.real - a3.imag, w);          \
 171     tmp8 = MUL (a3.imag + a3.real, w);          \
 172     tmp1 = tmp5 + tmp7;                         \
 173     tmp2 = tmp6 + tmp8;                         \
 174     tmp3 = tmp6 - tmp8;                         \
 175     tmp4 = tmp7 - tmp5;                         \
 176     a2.real = a0.real - tmp1;                   \
 177     a2.imag = a0.imag - tmp2;                   \
 178     a3.real = a1.real - tmp3;                   \
 179     a3.imag = a1.imag - tmp4;                   \
 180     a0.real += tmp1;                            \
 181     a0.imag += tmp2;                            \
 182     a1.real += tmp3;                            \
 183     a1.imag += tmp4;                            \
 184 } while (0)
 185
 186 static inline void ifft8 (complex_t * buf)
 187 {
 188     sample_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
 189
 190     ifft4 (buf);
 191     ifft2 (buf + 4);
 192     ifft2 (buf + 6);
 193     BUTTERFLY_ZERO (buf[0], buf[2], buf[4], buf[6]);
 194     BUTTERFLY_HALF (buf[1], buf[3], buf[5], buf[7], roots16[1]);
 195 }
 196
 197 static void ifft_pass (complex_t * buf, const sample_t * weight, int n)
 198 {
 199     complex_t * buf1;
 200     complex_t * buf2;
 201     complex_t * buf3;
 202     sample_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
 203     int i;
 204
 205     buf++;
 206     buf1 = buf + n;
 207     buf2 = buf + 2 * n;
 208     buf3 = buf + 3 * n;
 209
 210     BUTTERFLY_ZERO (buf[-1], buf1[-1], buf2[-1], buf3[-1]);
 211
 212     i = n - 1;
 213
 214     do {
 215         BUTTERFLY (buf[0], buf1[0], buf2[0], buf3[0],
 216                    weight[0], weight[2*i-n]);
 217         buf++;
 218         buf1++;
 219         buf2++;
 220         buf3++;
 221         weight++;
 222     } while (--i);
 223 }
 224
 225 static void ifft16 (complex_t * buf)
 226 {
 227     ifft8 (buf);
 228     ifft4 (buf + 8);
 229     ifft4 (buf + 12);
 230     ifft_pass (buf, roots16, 4);
 231 }
 232
 233 static void ifft32 (complex_t * buf)
 234 {
 235     ifft16 (buf);
 236     ifft8 (buf + 16);
 237     ifft8 (buf + 24);
 238     ifft_pass (buf, roots32, 8);
 239 }
 240
 241 static void ifft64_c (complex_t * buf)
 242 {
 243     ifft32 (buf);
 244     ifft16 (buf + 32);
 245     ifft16 (buf + 48);
 246     ifft_pass (buf, roots64, 16);
 247 }
 248
 249 static void ifft128_c (complex_t * buf)
 250 {
 251     ifft32 (buf);
 252     ifft16 (buf + 32);
 253     ifft16 (buf + 48);
 254     ifft_pass (buf, roots64, 16);
 255
 256     ifft32 (buf + 64);
 257     ifft32 (buf + 96);
 258     ifft_pass (buf, roots128, 32);
 259 }
 260 */
 261 void a52_imdct_512 (sample_t * data, sample_t * delay)
 262 {
 263     int i, k;
 264     sample_t t_r, t_i, a_r, a_i, b_r, b_i, w_1, w_2;
 265     const sample_t * window = a52_imdct_window;
 266     FFTComplex buf[128];
 267
 268     for (i = 0; i < 128; i++) {
 269         k = fftorder[i];
 270         t_r = pre1[i].real;
 271         t_i = pre1[i].imag;
 272         BUTTERFLY_0 (buf[i].re, buf[i].im, t_r, t_i, data[k], data[255-k]);
 273     }
 274
 275     //ifft128 (buf);
 276     ff_fft_calc_c(7, (FFTComplex *)&buf);
 277
 278     /* Post IFFT complex multiply plus IFFT complex conjugate*/
 279     /* Window and convert to real valued signal */
 280     for (i = 0; i < 64; i++) {
 281         /* y[n] = z[n] * (xcos1[n] + j * xsin1[n]) ; */
 282         t_r = post1[i].real;
 283         t_i = post1[i].imag;
 284         BUTTERFLY_0 (a_r, a_i, t_i, t_r, buf[i].im, buf[i].re);
 285         BUTTERFLY_0 (b_r, b_i, t_r, t_i, buf[127-i].im, buf[127-i].re);
 286
 287         w_1 = window[2*i];
 288         w_2 = window[255-2*i];
 289         BUTTERFLY_B (data[255-2*i], data[2*i], w_2, w_1, a_r, delay[2*i]);
 290         delay[2*i] = a_i;
 291
 292         w_1 = window[2*i+1];
 293         w_2 = window[254-2*i];
 294         BUTTERFLY_B (data[2*i+1], data[254-2*i], w_1, w_2, b_r, delay[2*i+1]);
 295         delay[2*i+1] = b_i;
 296     }
 297 }
 298
 299 void a52_imdct_256 (sample_t * data, sample_t * delay)
 300 {
 301     int i, k;
 302     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;
 303     const sample_t * window = a52_imdct_window;
 304     FFTComplex buf1[64], buf2[64];
 305
 306     /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
 307     for (i = 0; i < 64; i++) {
 308         k = fftorder[i];
 309         t_r = pre2[i].real;
 310         t_i = pre2[i].imag;
 311         BUTTERFLY_0 (buf1[i].re, buf1[i].im, t_r, t_i, data[k], data[254-k]);
 312         BUTTERFLY_0 (buf2[i].re, buf2[i].im, t_r, t_i, data[k+1], data[255-k]);
 313     }
 314
 315     //ifft64 (buf1);
 316     //ifft64 (buf2);
 317     ff_fft_calc_c(6, (FFTComplex *)&buf1);
 318     ff_fft_calc_c(6, (FFTComplex *)&buf2);
 319
 320     /* Post IFFT complex multiply */
 321     /* Window and convert to real valued signal */
 322     for (i = 0; i < 32; i++) {
 323         /* y1[n] = z1[n] * (xcos2[n] + j * xs in2[n]) ; */
 324         t_r = post2[i].real;
 325         t_i = post2[i].imag;
 326         BUTTERFLY_0 (a_r, a_i, t_i, t_r, buf1[i].im, buf1[i].re);
 327         BUTTERFLY_0 (b_r, b_i, t_r, t_i, buf1[63-i].im, buf1[63-i].re);
 328         BUTTERFLY_0 (c_r, c_i, t_i, t_r, buf2[i].im, buf2[i].re);
 329         BUTTERFLY_0 (d_r, d_i, t_r, t_i, buf2[63-i].im, buf2[63-i].re);
 330
 331         w_1 = window[2*i];
 332         w_2 = window[255-2*i];
 333         BUTTERFLY_B (data[255-2*i], data[2*i], w_2, w_1, a_r, delay[2*i]);
 334         delay[2*i] = c_i;
 335
 336         w_1 = window[128+2*i];
 337         w_2 = window[127-2*i];
 338         BUTTERFLY_B (data[128+2*i], data[127-2*i], w_1, w_2, a_i, delay[127-2*i]);
 339         delay[127-2*i] = c_r;
 340
 341         w_1 = window[2*i+1];
 342         w_2 = window[254-2*i];
 343         BUTTERFLY_B (data[254-2*i], data[2*i+1], w_2, w_1, b_i, delay[2*i+1]);
 344         delay[2*i+1] = d_r;
 345
 346         w_1 = window[129+2*i];
 347         w_2 = window[126-2*i];
 348         BUTTERFLY_B (data[129+2*i], data[126-2*i], w_1, w_2, b_r, delay[126-2*i]);
 349         delay[126-2*i] = d_i;
 350     }
 351 }
 352
 353 /*
 354 static double besselI0 (double x)
 355 {
 356     double bessel = 1;
 357     int i = 100;
 358
 359     do
 360         bessel = bessel * x / (i * i) + 1;
 361     while (--i);
 362     return bessel;
 363 }
 364 */
 365
 366 void a52_imdct_init (uint32_t mm_accel)
 367 {
 368   (void)mm_accel;
 369   //ff_fft_init(&s128, 7, 1);
 370   //ff_fft_init(&s64, 6, 1);
 371
 372 /*    int i, k;
 373     double sum;
 374     double local_imdct_window[256];*/
 375
 376     /* compute imdct window - kaiser-bessel derived window, alpha = 5.0 */
 377     /*    sum = 0;
 378     for (i = 0; i < 256; i++) {
 379         sum += besselI0 (i * (256 - i) * (5 * M_PI / 256) * (5 * M_PI / 256));
 380         local_imdct_window[i] = sum;
 381     }
 382     sum++;
 383     */
 384     /*    for (i = 0; i < 256; i++)
 385         a52_imdct_window[i] = SAMPLE (sqrt (local_imdct_window[i] / sum));
 386
 387     printf("static sample_t a52_imdct_window[256]={");
 388     for (i=0;i<256;i++) {
 389       if ((i % 16)==0) { printf("\n"); }
 390       printf("%d,",a52_imdct_window[i]);
 391     }
 392     printf("\n}\n");
 393     */
 394
 395     /*    for (i = 0; i < 3; i++)
 396         roots16[i] = SAMPLE (cos ((M_PI / 8) * (i + 1)));
 397
 398     printf("static sample_t roots16[3]={%d,%d,%d};\n\n",roots16[0],roots16[1],roots16[2]);
 399
 400     for (i = 0; i < 7; i++)
 401         roots32[i] = SAMPLE (cos ((M_PI / 16) * (i + 1)));
 402
 403     printf("static sample_t roots32[7]={");
 404     for (i=0;i<7;i++) { printf("%d%s",roots32[i],(i < 6 ? "," : "")); }
 405     printf("};\n");
 406
 407     for (i = 0; i < 15; i++)
 408         roots64[i] = SAMPLE (cos ((M_PI / 32) * (i + 1)));
 409
 410     printf("static sample_t roots64[15]={");
 411     for (i=0;i<15;i++) { printf("%d%s",roots64[i],(i < 14 ? "," : "")); }
 412     printf("};\n");
 413
 414     for (i = 0; i < 31; i++)
 415         roots128[i] = SAMPLE (cos ((M_PI / 64) * (i + 1)));
 416
 417     printf("static sample_t roots128[31]={");
 418     for (i=0;i<31;i++) { printf("%d%s",roots128[i],(i < 30 ? "," : "")); }
 419     printf("};\n");
 420     */
 421     /*
 422     for (i = 0; i < 64; i++) {
 423         k = fftorder[i] / 2 + 64;
 424         pre1[i].real = SAMPLE (cos ((M_PI / 256) * (k - 0.25)));
 425         pre1[i].imag = SAMPLE (sin ((M_PI / 256) * (k - 0.25)));
 426     }
 427
 428     for (i = 64; i < 128; i++) {
 429         k = fftorder[i] / 2 + 64;
 430         pre1[i].real = SAMPLE (-cos ((M_PI / 256) * (k - 0.25)));
 431         pre1[i].imag = SAMPLE (-sin ((M_PI / 256) * (k - 0.25)));
 432     }
 433
 434     printf("static complex_t pre1[128]={");
 435     for (i=0;i<128;i++) { printf("{%d,%d}%s",pre1[i].real,pre1[i].imag,(i < 127 ? "," : "")); }
 436     printf("};\n");
 437     */
 438     /*
 439     for (i = 0; i < 64; i++) {
 440         post1[i].real = SAMPLE (cos ((M_PI / 256) * (i + 0.5)));
 441         post1[i].imag = SAMPLE (sin ((M_PI / 256) * (i + 0.5)));
 442     }
 443
 444     printf("static complex_t post1[64]={");
 445     for (i=0;i<64;i++) { printf("{%d,%d}%s",post1[i].real,post1[i].imag,(i < 63 ? "," : "")); }
 446     printf("};\n");
 447     */
 448
 449     /*
 450     for (i = 0; i < 64; i++) {
 451         k = fftorder[i] / 4;
 452         pre2[i].real = SAMPLE (cos ((M_PI / 128) * (k - 0.25)));
 453         pre2[i].imag = SAMPLE (sin ((M_PI / 128) * (k - 0.25)));
 454     }
 455
 456     printf("static complex_t pre2[64]={");
 457     for (i=0;i<64;i++) { printf("{%d,%d}%s",pre2[i].real,pre2[i].imag,(i < 63 ? "," : "")); }
 458     printf("};\n");
 459
 460     for (i = 0; i < 32; i++) {
 461         post2[i].real = SAMPLE (cos ((M_PI / 128) * (i + 0.5)));
 462         post2[i].imag = SAMPLE (sin ((M_PI / 128) * (i + 0.5)));
 463     }
 464
 465     printf("static complex_t post2[32]={");
 466     for (i=0;i<32;i++) { printf("{%d,%d}%s",post2[i].real,post2[i].imag,(i < 31 ? "," : "")); }
 467     printf("};\n");
 468
 469
 470 #ifdef LIBA52_DJBFFT
 471     if (mm_accel & MM_ACCEL_DJBFFT) {
 472 #ifndef LIBA52_DOUBLE
 473         ifft128 = (void (*) (complex_t *)) fftc4_un128;
 474         ifft64 = (void (*) (complex_t *)) fftc4_un64;
 475 #else
 476         ifft128 = (void (*) (complex_t *)) fftc8_un128;
 477         ifft64 = (void (*) (complex_t *)) fftc8_un64;
 478 #endif
 479     } else
 480 #endif
 481     {
 482         ifft128 = ifft128_c;
 483         ifft64 = ifft64_c;
 484     }
 485     */
 486 }