mp3lib/dct64_altivec.c

   1
   2 /*
   3  * Discrete Cosine Tansform (DCT) for Altivec
   4  * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
   5  * based upon code from "mp3lib/dct64.c"
   6  * This file is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public License
   8  */
   9
  10 #include <stdio.h>
  11 #include "mpg123.h"
  12
  13 #ifdef HAVE_ALTIVEC_H
  14 #include <altivec.h>
  15 #endif
  16
  17 // used to build registers permutation vectors (vcprm)
  18 // the 's' are for words in the _s_econd vector
  19 #define WORD_0 0x00,0x01,0x02,0x03
  20 #define WORD_1 0x04,0x05,0x06,0x07
  21 #define WORD_2 0x08,0x09,0x0a,0x0b
  22 #define WORD_3 0x0c,0x0d,0x0e,0x0f
  23 #define WORD_s0 0x10,0x11,0x12,0x13
  24 #define WORD_s1 0x14,0x15,0x16,0x17
  25 #define WORD_s2 0x18,0x19,0x1a,0x1b
  26 #define WORD_s3 0x1c,0x1d,0x1e,0x1f
  27
  28 #define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
  29 #define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}
  30
  31 #define FOUROF(a) {a,a,a,a}
  32
  33 // vcprmle is used to keep the same index as in the SSE version.
  34 // it's the same as vcprm, with the index inversed
  35 // ('le' is Little Endian)
  36 #define vcprmle(a,b,c,d) vcprm(d,c,b,a)
  37
  38 // used to build inverse/identity vectors (vcii)
  39 // n is _n_egative, p is _p_ositive
  40 #define FLOAT_n -1.
  41 #define FLOAT_p 1.
  42
  43 void dct64_altivec(real *a,real *b,real *c)
  44 {
  45   real __attribute__ ((aligned(16))) b1[0x20];
  46   real __attribute__ ((aligned(16))) b2[0x20];
  47
  48   real *out0 = a;
  49   real *out1 = b;
  50   real *samples = c;
  51
  52   const vector float vczero = (const vector float)FOUROF(0.);
  53   const vector unsigned char reverse = (const vector unsigned char)vcprm(3,2,1,0);
  54
  55
  56   if (((unsigned long)b1 & 0x0000000F) ||
  57       ((unsigned long)b2 & 0x0000000F))
  58
  59   {
  60     printf("MISALIGNED:\t%p\t%p\t%p\t%p\t%p\n",
  61            b1, b2, a, b, samples);
  62   }
  63
  64
  65 #ifdef ALTIVEC_USE_REFERENCE_C_CODE
  66
  67   {
  68     register real *costab = mp3lib_pnts[0];
  69
  70     b1[0x00] = samples[0x00] + samples[0x1F];
  71     b1[0x01] = samples[0x01] + samples[0x1E];
  72     b1[0x02] = samples[0x02] + samples[0x1D];
  73     b1[0x03] = samples[0x03] + samples[0x1C];
  74     b1[0x04] = samples[0x04] + samples[0x1B];
  75     b1[0x05] = samples[0x05] + samples[0x1A];
  76     b1[0x06] = samples[0x06] + samples[0x19];
  77     b1[0x07] = samples[0x07] + samples[0x18];
  78     b1[0x08] = samples[0x08] + samples[0x17];
  79     b1[0x09] = samples[0x09] + samples[0x16];
  80     b1[0x0A] = samples[0x0A] + samples[0x15];
  81     b1[0x0B] = samples[0x0B] + samples[0x14];
  82     b1[0x0C] = samples[0x0C] + samples[0x13];
  83     b1[0x0D] = samples[0x0D] + samples[0x12];
  84     b1[0x0E] = samples[0x0E] + samples[0x11];
  85     b1[0x0F] = samples[0x0F] + samples[0x10];
  86     b1[0x10] = (samples[0x0F] - samples[0x10]) * costab[0xF];
  87     b1[0x11] = (samples[0x0E] - samples[0x11]) * costab[0xE];
  88     b1[0x12] = (samples[0x0D] - samples[0x12]) * costab[0xD];
  89     b1[0x13] = (samples[0x0C] - samples[0x13]) * costab[0xC];
  90     b1[0x14] = (samples[0x0B] - samples[0x14]) * costab[0xB];
  91     b1[0x15] = (samples[0x0A] - samples[0x15]) * costab[0xA];
  92     b1[0x16] = (samples[0x09] - samples[0x16]) * costab[0x9];
  93     b1[0x17] = (samples[0x08] - samples[0x17]) * costab[0x8];
  94     b1[0x18] = (samples[0x07] - samples[0x18]) * costab[0x7];
  95     b1[0x19] = (samples[0x06] - samples[0x19]) * costab[0x6];
  96     b1[0x1A] = (samples[0x05] - samples[0x1A]) * costab[0x5];
  97     b1[0x1B] = (samples[0x04] - samples[0x1B]) * costab[0x4];
  98     b1[0x1C] = (samples[0x03] - samples[0x1C]) * costab[0x3];
  99     b1[0x1D] = (samples[0x02] - samples[0x1D]) * costab[0x2];
 100     b1[0x1E] = (samples[0x01] - samples[0x1E]) * costab[0x1];
 101     b1[0x1F] = (samples[0x00] - samples[0x1F]) * costab[0x0];
 102
 103   }
 104   {
 105     register real *costab = mp3lib_pnts[1];
 106
 107     b2[0x00] = b1[0x00] + b1[0x0F];
 108     b2[0x01] = b1[0x01] + b1[0x0E];
 109     b2[0x02] = b1[0x02] + b1[0x0D];
 110     b2[0x03] = b1[0x03] + b1[0x0C];
 111     b2[0x04] = b1[0x04] + b1[0x0B];
 112     b2[0x05] = b1[0x05] + b1[0x0A];
 113     b2[0x06] = b1[0x06] + b1[0x09];
 114     b2[0x07] = b1[0x07] + b1[0x08];
 115     b2[0x08] = (b1[0x07] - b1[0x08]) * costab[7];
 116     b2[0x09] = (b1[0x06] - b1[0x09]) * costab[6];
 117     b2[0x0A] = (b1[0x05] - b1[0x0A]) * costab[5];
 118     b2[0x0B] = (b1[0x04] - b1[0x0B]) * costab[4];
 119     b2[0x0C] = (b1[0x03] - b1[0x0C]) * costab[3];
 120     b2[0x0D] = (b1[0x02] - b1[0x0D]) * costab[2];
 121     b2[0x0E] = (b1[0x01] - b1[0x0E]) * costab[1];
 122     b2[0x0F] = (b1[0x00] - b1[0x0F]) * costab[0];
 123     b2[0x10] = b1[0x10] + b1[0x1F];
 124     b2[0x11] = b1[0x11] + b1[0x1E];
 125     b2[0x12] = b1[0x12] + b1[0x1D];
 126     b2[0x13] = b1[0x13] + b1[0x1C];
 127     b2[0x14] = b1[0x14] + b1[0x1B];
 128     b2[0x15] = b1[0x15] + b1[0x1A];
 129     b2[0x16] = b1[0x16] + b1[0x19];
 130     b2[0x17] = b1[0x17] + b1[0x18];
 131     b2[0x18] = (b1[0x18] - b1[0x17]) * costab[7];
 132     b2[0x19] = (b1[0x19] - b1[0x16]) * costab[6];
 133     b2[0x1A] = (b1[0x1A] - b1[0x15]) * costab[5];
 134     b2[0x1B] = (b1[0x1B] - b1[0x14]) * costab[4];
 135     b2[0x1C] = (b1[0x1C] - b1[0x13]) * costab[3];
 136     b2[0x1D] = (b1[0x1D] - b1[0x12]) * costab[2];
 137     b2[0x1E] = (b1[0x1E] - b1[0x11]) * costab[1];
 138     b2[0x1F] = (b1[0x1F] - b1[0x10]) * costab[0];
 139
 140   }
 141
 142   {
 143     register real *costab = mp3lib_pnts[2];
 144
 145     b1[0x00] = b2[0x00] + b2[0x07];
 146     b1[0x01] = b2[0x01] + b2[0x06];
 147     b1[0x02] = b2[0x02] + b2[0x05];
 148     b1[0x03] = b2[0x03] + b2[0x04];
 149     b1[0x04] = (b2[0x03] - b2[0x04]) * costab[3];
 150     b1[0x05] = (b2[0x02] - b2[0x05]) * costab[2];
 151     b1[0x06] = (b2[0x01] - b2[0x06]) * costab[1];
 152     b1[0x07] = (b2[0x00] - b2[0x07]) * costab[0];
 153     b1[0x08] = b2[0x08] + b2[0x0F];
 154     b1[0x09] = b2[0x09] + b2[0x0E];
 155     b1[0x0A] = b2[0x0A] + b2[0x0D];
 156     b1[0x0B] = b2[0x0B] + b2[0x0C];
 157     b1[0x0C] = (b2[0x0C] - b2[0x0B]) * costab[3];
 158     b1[0x0D] = (b2[0x0D] - b2[0x0A]) * costab[2];
 159     b1[0x0E] = (b2[0x0E] - b2[0x09]) * costab[1];
 160     b1[0x0F] = (b2[0x0F] - b2[0x08]) * costab[0];
 161     b1[0x10] = b2[0x10] + b2[0x17];
 162     b1[0x11] = b2[0x11] + b2[0x16];
 163     b1[0x12] = b2[0x12] + b2[0x15];
 164     b1[0x13] = b2[0x13] + b2[0x14];
 165     b1[0x14] = (b2[0x13] - b2[0x14]) * costab[3];
 166     b1[0x15] = (b2[0x12] - b2[0x15]) * costab[2];
 167     b1[0x16] = (b2[0x11] - b2[0x16]) * costab[1];
 168     b1[0x17] = (b2[0x10] - b2[0x17]) * costab[0];
 169     b1[0x18] = b2[0x18] + b2[0x1F];
 170     b1[0x19] = b2[0x19] + b2[0x1E];
 171     b1[0x1A] = b2[0x1A] + b2[0x1D];
 172     b1[0x1B] = b2[0x1B] + b2[0x1C];
 173     b1[0x1C] = (b2[0x1C] - b2[0x1B]) * costab[3];
 174     b1[0x1D] = (b2[0x1D] - b2[0x1A]) * costab[2];
 175     b1[0x1E] = (b2[0x1E] - b2[0x19]) * costab[1];
 176     b1[0x1F] = (b2[0x1F] - b2[0x18]) * costab[0];
 177   }
 178
 179 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
 180
 181   // How does it work ?
 182   // the first three passes are reproducted in the three block below
 183   // all computations are done on a 4 elements vector
 184   // 'reverse' is a special perumtation vector used to reverse
 185   // the order of the elements inside a vector.
 186   // note that all loads/stores to b1 (b2) between passes 1 and 2 (2 and 3)
 187   // have been removed, all elements are stored inside b1vX (b2vX)
 188   {
 189     register vector float
 190       b1v0, b1v1, b1v2, b1v3,
 191       b1v4, b1v5, b1v6, b1v7;
 192     register vector float
 193       temp1, temp2;
 194
 195     {
 196       register real *costab = mp3lib_pnts[0];
 197
 198       register vector float
 199         samplesv1, samplesv2, samplesv3, samplesv4,
 200         samplesv5, samplesv6, samplesv7, samplesv8,
 201         samplesv9;
 202       register vector unsigned char samples_perm = vec_lvsl(0, samples);
 203       register vector float costabv1, costabv2, costabv3, costabv4, costabv5;
 204       register vector unsigned char costab_perm = vec_lvsl(0, costab);
 205
 206       samplesv1 = vec_ld(0, samples);
 207       samplesv2 = vec_ld(16, samples);
 208       samplesv1 = vec_perm(samplesv1, samplesv2, samples_perm);
 209       samplesv3 = vec_ld(32, samples);
 210       samplesv2 = vec_perm(samplesv2, samplesv3, samples_perm);
 211       samplesv4 = vec_ld(48, samples);
 212       samplesv3 = vec_perm(samplesv3, samplesv4, samples_perm);
 213       samplesv5 = vec_ld(64, samples);
 214       samplesv4 = vec_perm(samplesv4, samplesv5, samples_perm);
 215       samplesv6 = vec_ld(80, samples);
 216       samplesv5 = vec_perm(samplesv5, samplesv6, samples_perm);
 217       samplesv7 = vec_ld(96, samples);
 218       samplesv6 = vec_perm(samplesv6, samplesv7, samples_perm);
 219       samplesv8 = vec_ld(112, samples);
 220       samplesv7 = vec_perm(samplesv7, samplesv8, samples_perm);
 221       samplesv9 = vec_ld(128, samples);
 222       samplesv8 = vec_perm(samplesv8, samplesv9, samples_perm);
 223
 224       temp1 = vec_add(samplesv1,
 225                       vec_perm(samplesv8, samplesv8, reverse));
 226       //vec_st(temp1, 0, b1);
 227       b1v0 = temp1;
 228       temp1 = vec_add(samplesv2,
 229                       vec_perm(samplesv7, samplesv7, reverse));
 230       //vec_st(temp1, 16, b1);
 231       b1v1 = temp1;
 232       temp1 = vec_add(samplesv3,
 233                       vec_perm(samplesv6, samplesv6, reverse));
 234       //vec_st(temp1, 32, b1);
 235       b1v2 = temp1;
 236       temp1 = vec_add(samplesv4,
 237                       vec_perm(samplesv5, samplesv5, reverse));
 238       //vec_st(temp1, 48, b1);
 239       b1v3 = temp1;
 240
 241       costabv1 = vec_ld(0, costab);
 242       costabv2 = vec_ld(16, costab);
 243       costabv1 = vec_perm(costabv1, costabv2, costab_perm);
 244       costabv3 = vec_ld(32, costab);
 245       costabv2 = vec_perm(costabv2, costabv3, costab_perm);
 246       costabv4 = vec_ld(48, costab);
 247       costabv3 = vec_perm(costabv3, costabv4, costab_perm);
 248       costabv5 = vec_ld(64, costab);
 249       costabv4 = vec_perm(costabv4, costabv5, costab_perm);
 250
 251       temp1 = vec_sub(vec_perm(samplesv4, samplesv4, reverse),
 252                       samplesv5);
 253       temp2 = vec_madd(temp1,
 254                        vec_perm(costabv4, costabv4, reverse),
 255                        vczero);
 256       //vec_st(temp2, 64, b1);
 257       b1v4 = temp2;
 258
 259       temp1 = vec_sub(vec_perm(samplesv3, samplesv3, reverse),
 260                       samplesv6);
 261       temp2 = vec_madd(temp1,
 262                        vec_perm(costabv3, costabv3, reverse),
 263                        vczero);
 264       //vec_st(temp2, 80, b1);
 265       b1v5 = temp2;
 266       temp1 = vec_sub(vec_perm(samplesv2, samplesv2, reverse),
 267                       samplesv7);
 268       temp2 = vec_madd(temp1,
 269                        vec_perm(costabv2, costabv2, reverse),
 270                        vczero);
 271       //vec_st(temp2, 96, b1);
 272       b1v6 = temp2;
 273
 274       temp1 = vec_sub(vec_perm(samplesv1, samplesv1, reverse),
 275                       samplesv8);
 276       temp2 = vec_madd(temp1,
 277                        vec_perm(costabv1, costabv1, reverse),
 278                        vczero);
 279       //vec_st(temp2, 112, b1);
 280       b1v7 = temp2;
 281
 282     }
 283
 284     {
 285       register vector float
 286         b2v0, b2v1, b2v2, b2v3,
 287         b2v4, b2v5, b2v6, b2v7;
 288       {
 289         register real *costab = mp3lib_pnts[1];
 290         register vector float costabv1r, costabv2r, costabv1, costabv2, costabv3;
 291         register vector unsigned char costab_perm = vec_lvsl(0, costab);
 292
 293         costabv1 = vec_ld(0, costab);
 294         costabv2 = vec_ld(16, costab);
 295         costabv1 = vec_perm(costabv1, costabv2, costab_perm);
 296         costabv3  = vec_ld(32, costab);
 297         costabv2 = vec_perm(costabv2, costabv3 , costab_perm);
 298         costabv1r = vec_perm(costabv1, costabv1, reverse);
 299         costabv2r = vec_perm(costabv2, costabv2, reverse);
 300
 301         temp1 = vec_add(b1v0, vec_perm(b1v3, b1v3, reverse));
 302         //vec_st(temp1, 0, b2);
 303         b2v0 = temp1;
 304         temp1 = vec_add(b1v1, vec_perm(b1v2, b1v2, reverse));
 305         //vec_st(temp1, 16, b2);
 306         b2v1 = temp1;
 307         temp2 = vec_sub(vec_perm(b1v1, b1v1, reverse), b1v2);
 308         temp1 = vec_madd(temp2, costabv2r, vczero);
 309         //vec_st(temp1, 32, b2);
 310         b2v2 = temp1;
 311         temp2 = vec_sub(vec_perm(b1v0, b1v0, reverse), b1v3);
 312         temp1 = vec_madd(temp2, costabv1r, vczero);
 313         //vec_st(temp1, 48, b2);
 314         b2v3 = temp1;
 315         temp1 = vec_add(b1v4, vec_perm(b1v7, b1v7, reverse));
 316         //vec_st(temp1, 64, b2);
 317         b2v4 = temp1;
 318         temp1 = vec_add(b1v5, vec_perm(b1v6, b1v6, reverse));
 319         //vec_st(temp1, 80, b2);
 320         b2v5 = temp1;
 321         temp2 = vec_sub(b1v6, vec_perm(b1v5, b1v5, reverse));
 322         temp1 = vec_madd(temp2, costabv2r, vczero);
 323         //vec_st(temp1, 96, b2);
 324         b2v6 = temp1;
 325         temp2 = vec_sub(b1v7, vec_perm(b1v4, b1v4, reverse));
 326         temp1 = vec_madd(temp2, costabv1r, vczero);
 327         //vec_st(temp1, 112, b2);
 328         b2v7 = temp1;
 329       }
 330
 331       {
 332         register real *costab = mp3lib_pnts[2];
 333
 334
 335         vector float costabv1r, costabv1, costabv2;
 336         vector unsigned char costab_perm = vec_lvsl(0, costab);
 337
 338         costabv1 = vec_ld(0, costab);
 339         costabv2 = vec_ld(16, costab);
 340         costabv1 = vec_perm(costabv1, costabv2, costab_perm);
 341         costabv1r = vec_perm(costabv1, costabv1, reverse);
 342
 343         temp1 = vec_add(b2v0, vec_perm(b2v1, b2v1, reverse));
 344         vec_st(temp1, 0, b1);
 345         temp2 = vec_sub(vec_perm(b2v0, b2v0, reverse), b2v1);
 346         temp1 = vec_madd(temp2, costabv1r, vczero);
 347         vec_st(temp1, 16, b1);
 348
 349         temp1 = vec_add(b2v2, vec_perm(b2v3, b2v3, reverse));
 350         vec_st(temp1, 32, b1);
 351         temp2 = vec_sub(b2v3, vec_perm(b2v2, b2v2, reverse));
 352         temp1 = vec_madd(temp2, costabv1r, vczero);
 353         vec_st(temp1, 48, b1);
 354
 355         temp1 = vec_add(b2v4, vec_perm(b2v5, b2v5, reverse));
 356         vec_st(temp1, 64, b1);
 357         temp2 = vec_sub(vec_perm(b2v4, b2v4, reverse), b2v5);
 358         temp1 = vec_madd(temp2, costabv1r, vczero);
 359         vec_st(temp1, 80, b1);
 360
 361         temp1 = vec_add(b2v6, vec_perm(b2v7, b2v7, reverse));
 362         vec_st(temp1, 96, b1);
 363         temp2 = vec_sub(b2v7, vec_perm(b2v6, b2v6, reverse));
 364         temp1 = vec_madd(temp2, costabv1r, vczero);
 365         vec_st(temp1, 112, b1);
 366
 367       }
 368     }
 369   }
 370
 371 #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
 372
 373   {
 374     register real const cos0 = mp3lib_pnts[3][0];
 375     register real const cos1 = mp3lib_pnts[3][1];
 376
 377     b2[0x00] = b1[0x00] + b1[0x03];
 378     b2[0x01] = b1[0x01] + b1[0x02];
 379     b2[0x02] = (b1[0x01] - b1[0x02]) * cos1;
 380     b2[0x03] = (b1[0x00] - b1[0x03]) * cos0;
 381     b2[0x04] = b1[0x04] + b1[0x07];
 382     b2[0x05] = b1[0x05] + b1[0x06];
 383     b2[0x06] = (b1[0x06] - b1[0x05]) * cos1;
 384     b2[0x07] = (b1[0x07] - b1[0x04]) * cos0;
 385     b2[0x08] = b1[0x08] + b1[0x0B];
 386     b2[0x09] = b1[0x09] + b1[0x0A];
 387     b2[0x0A] = (b1[0x09] - b1[0x0A]) * cos1;
 388     b2[0x0B] = (b1[0x08] - b1[0x0B]) * cos0;
 389     b2[0x0C] = b1[0x0C] + b1[0x0F];
 390     b2[0x0D] = b1[0x0D] + b1[0x0E];
 391     b2[0x0E] = (b1[0x0E] - b1[0x0D]) * cos1;
 392     b2[0x0F] = (b1[0x0F] - b1[0x0C]) * cos0;
 393     b2[0x10] = b1[0x10] + b1[0x13];
 394     b2[0x11] = b1[0x11] + b1[0x12];
 395     b2[0x12] = (b1[0x11] - b1[0x12]) * cos1;
 396     b2[0x13] = (b1[0x10] - b1[0x13]) * cos0;
 397     b2[0x14] = b1[0x14] + b1[0x17];
 398     b2[0x15] = b1[0x15] + b1[0x16];
 399     b2[0x16] = (b1[0x16] - b1[0x15]) * cos1;
 400     b2[0x17] = (b1[0x17] - b1[0x14]) * cos0;
 401     b2[0x18] = b1[0x18] + b1[0x1B];
 402     b2[0x19] = b1[0x19] + b1[0x1A];
 403     b2[0x1A] = (b1[0x19] - b1[0x1A]) * cos1;
 404     b2[0x1B] = (b1[0x18] - b1[0x1B]) * cos0;
 405     b2[0x1C] = b1[0x1C] + b1[0x1F];
 406     b2[0x1D] = b1[0x1D] + b1[0x1E];
 407     b2[0x1E] = (b1[0x1E] - b1[0x1D]) * cos1;
 408     b2[0x1F] = (b1[0x1F] - b1[0x1C]) * cos0;
 409   }
 410
 411   {
 412     register real const cos0 = mp3lib_pnts[4][0];
 413
 414     b1[0x00] = b2[0x00] + b2[0x01];
 415     b1[0x01] = (b2[0x00] - b2[0x01]) * cos0;
 416     b1[0x02] = b2[0x02] + b2[0x03];
 417     b1[0x03] = (b2[0x03] - b2[0x02]) * cos0;
 418     b1[0x02] += b1[0x03];
 419
 420     b1[0x04] = b2[0x04] + b2[0x05];
 421     b1[0x05] = (b2[0x04] - b2[0x05]) * cos0;
 422     b1[0x06] = b2[0x06] + b2[0x07];
 423     b1[0x07] = (b2[0x07] - b2[0x06]) * cos0;
 424     b1[0x06] += b1[0x07];
 425     b1[0x04] += b1[0x06];
 426     b1[0x06] += b1[0x05];
 427     b1[0x05] += b1[0x07];
 428
 429     b1[0x08] = b2[0x08] + b2[0x09];
 430     b1[0x09] = (b2[0x08] - b2[0x09]) * cos0;
 431     b1[0x0A] = b2[0x0A] + b2[0x0B];
 432     b1[0x0B] = (b2[0x0B] - b2[0x0A]) * cos0;
 433     b1[0x0A] += b1[0x0B];
 434
 435     b1[0x0C] = b2[0x0C] + b2[0x0D];
 436     b1[0x0D] = (b2[0x0C] - b2[0x0D]) * cos0;
 437     b1[0x0E] = b2[0x0E] + b2[0x0F];
 438     b1[0x0F] = (b2[0x0F] - b2[0x0E]) * cos0;
 439     b1[0x0E] += b1[0x0F];
 440     b1[0x0C] += b1[0x0E];
 441     b1[0x0E] += b1[0x0D];
 442     b1[0x0D] += b1[0x0F];
 443
 444     b1[0x10] = b2[0x10] + b2[0x11];
 445     b1[0x11] = (b2[0x10] - b2[0x11]) * cos0;
 446     b1[0x12] = b2[0x12] + b2[0x13];
 447     b1[0x13] = (b2[0x13] - b2[0x12]) * cos0;
 448     b1[0x12] += b1[0x13];
 449
 450     b1[0x14] = b2[0x14] + b2[0x15];
 451     b1[0x15] = (b2[0x14] - b2[0x15]) * cos0;
 452     b1[0x16] = b2[0x16] + b2[0x17];
 453     b1[0x17] = (b2[0x17] - b2[0x16]) * cos0;
 454     b1[0x16] += b1[0x17];
 455     b1[0x14] += b1[0x16];
 456     b1[0x16] += b1[0x15];
 457     b1[0x15] += b1[0x17];
 458
 459     b1[0x18] = b2[0x18] + b2[0x19];
 460     b1[0x19] = (b2[0x18] - b2[0x19]) * cos0;
 461     b1[0x1A] = b2[0x1A] + b2[0x1B];
 462     b1[0x1B] = (b2[0x1B] - b2[0x1A]) * cos0;
 463     b1[0x1A] += b1[0x1B];
 464
 465     b1[0x1C] = b2[0x1C] + b2[0x1D];
 466     b1[0x1D] = (b2[0x1C] - b2[0x1D]) * cos0;
 467     b1[0x1E] = b2[0x1E] + b2[0x1F];
 468     b1[0x1F] = (b2[0x1F] - b2[0x1E]) * cos0;
 469     b1[0x1E] += b1[0x1F];
 470     b1[0x1C] += b1[0x1E];
 471     b1[0x1E] += b1[0x1D];
 472     b1[0x1D] += b1[0x1F];
 473   }
 474
 475   out0[0x10*16] = b1[0x00];
 476   out0[0x10*12] = b1[0x04];
 477   out0[0x10* 8] = b1[0x02];
 478   out0[0x10* 4] = b1[0x06];
 479   out0[0x10* 0] = b1[0x01];
 480   out1[0x10* 0] = b1[0x01];
 481   out1[0x10* 4] = b1[0x05];
 482   out1[0x10* 8] = b1[0x03];
 483   out1[0x10*12] = b1[0x07];
 484
 485   b1[0x08] += b1[0x0C];
 486   out0[0x10*14] = b1[0x08];
 487   b1[0x0C] += b1[0x0a];
 488   out0[0x10*10] = b1[0x0C];
 489   b1[0x0A] += b1[0x0E];
 490   out0[0x10* 6] = b1[0x0A];
 491   b1[0x0E] += b1[0x09];
 492   out0[0x10* 2] = b1[0x0E];
 493   b1[0x09] += b1[0x0D];
 494   out1[0x10* 2] = b1[0x09];
 495   b1[0x0D] += b1[0x0B];
 496   out1[0x10* 6] = b1[0x0D];
 497   b1[0x0B] += b1[0x0F];
 498   out1[0x10*10] = b1[0x0B];
 499   out1[0x10*14] = b1[0x0F];
 500
 501   b1[0x18] += b1[0x1C];
 502   out0[0x10*15] = b1[0x10] + b1[0x18];
 503   out0[0x10*13] = b1[0x18] + b1[0x14];
 504   b1[0x1C] += b1[0x1a];
 505   out0[0x10*11] = b1[0x14] + b1[0x1C];
 506   out0[0x10* 9] = b1[0x1C] + b1[0x12];
 507   b1[0x1A] += b1[0x1E];
 508   out0[0x10* 7] = b1[0x12] + b1[0x1A];
 509   out0[0x10* 5] = b1[0x1A] + b1[0x16];
 510   b1[0x1E] += b1[0x19];
 511   out0[0x10* 3] = b1[0x16] + b1[0x1E];
 512   out0[0x10* 1] = b1[0x1E] + b1[0x11];
 513   b1[0x19] += b1[0x1D];
 514   out1[0x10* 1] = b1[0x11] + b1[0x19];
 515   out1[0x10* 3] = b1[0x19] + b1[0x15];
 516   b1[0x1D] += b1[0x1B];
 517   out1[0x10* 5] = b1[0x15] + b1[0x1D];
 518   out1[0x10* 7] = b1[0x1D] + b1[0x13];
 519   b1[0x1B] += b1[0x1F];
 520   out1[0x10* 9] = b1[0x13] + b1[0x1B];
 521   out1[0x10*11] = b1[0x1B] + b1[0x17];
 522   out1[0x10*13] = b1[0x17] + b1[0x1F];
 523   out1[0x10*15] = b1[0x1F];
 524 }