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  */
   7
   8 #define real float
   9
  10 #include "mpg123.h"
  11
  12 #ifdef HAVE_ALTIVEC
  13
  14 #ifndef SYS_DARWIN
  15 #include <altivec.h>
  16 #endif
  17
  18 // used to build registers permutation vectors (vcprm)
  19 // the 's' are for words in the _s_econd vector
  20 #define WORD_0 0x00,0x01,0x02,0x03
  21 #define WORD_1 0x04,0x05,0x06,0x07
  22 #define WORD_2 0x08,0x09,0x0a,0x0b
  23 #define WORD_3 0x0c,0x0d,0x0e,0x0f
  24 #define WORD_s0 0x10,0x11,0x12,0x13
  25 #define WORD_s1 0x14,0x15,0x16,0x17
  26 #define WORD_s2 0x18,0x19,0x1a,0x1b
  27 #define WORD_s3 0x1c,0x1d,0x1e,0x1f
  28
  29 #ifdef SYS_DARWIN
  30 #define vcprm(a,b,c,d) (const vector unsigned char)(WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d)
  31 #else
  32 #define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
  33 #endif
  34
  35 // vcprmle is used to keep the same index as in the SSE version.
  36 // it's the same as vcprm, with the index inversed
  37 // ('le' is Little Endian)
  38 #define vcprmle(a,b,c,d) vcprm(d,c,b,a)
  39
  40 // used to build inverse/identity vectors (vcii)
  41 // n is _n_egative, p is _p_ositive
  42 #define FLOAT_n -1.
  43 #define FLOAT_p 1.
  44
  45 #ifdef SYS_DARWIN
  46 #define vcii(a,b,c,d) (const vector float)(FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d)
  47 #else
  48 #define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}
  49 #endif
  50
  51 #ifdef SYS_DARWIN
  52 #define FOUROF(a) (a)
  53 #else
  54 #define FOUROF(a) {a,a,a,a}
  55 #endif
  56
  57 void dct64_altivec(real *a,real *b,real *c)
  58 {
  59   real __attribute__ ((aligned(16))) b1[0x20];
  60   real __attribute__ ((aligned(16))) b2[0x20];
  61
  62   real *out0 = a;
  63   real *out1 = b;
  64   real *samples = c;
  65
  66   const vector float vczero = (const vector float)FOUROF(0.);
  67   const vector unsigned char reverse = (const vector unsigned char)vcprm(3,2,1,0);
  68
  69
  70   if (((unsigned long)b1 & 0x0000000F) ||
  71       ((unsigned long)b2 & 0x0000000F))
  72
  73   {
  74     printf("MISALIGNED:\t%p\t%p\t%p\t%p\t%p\n",
  75            b1, b2, a, b, samples);
  76   }
  77
  78
  79 #ifdef ALTIVEC_USE_REFERENCE_C_CODE
  80
  81   {
  82     register real *costab = mp3lib_pnts[0];
  83
  84     b1[0x00] = samples[0x00] + samples[0x1F];
  85     b1[0x01] = samples[0x01] + samples[0x1E];
  86     b1[0x02] = samples[0x02] + samples[0x1D];
  87     b1[0x03] = samples[0x03] + samples[0x1C];
  88     b1[0x04] = samples[0x04] + samples[0x1B];
  89     b1[0x05] = samples[0x05] + samples[0x1A];
  90     b1[0x06] = samples[0x06] + samples[0x19];
  91     b1[0x07] = samples[0x07] + samples[0x18];
  92     b1[0x08] = samples[0x08] + samples[0x17];
  93     b1[0x09] = samples[0x09] + samples[0x16];
  94     b1[0x0A] = samples[0x0A] + samples[0x15];
  95     b1[0x0B] = samples[0x0B] + samples[0x14];
  96     b1[0x0C] = samples[0x0C] + samples[0x13];
  97     b1[0x0D] = samples[0x0D] + samples[0x12];
  98     b1[0x0E] = samples[0x0E] + samples[0x11];
  99     b1[0x0F] = samples[0x0F] + samples[0x10];
 100     b1[0x10] = (samples[0x0F] - samples[0x10]) * costab[0xF];
 101     b1[0x11] = (samples[0x0E] - samples[0x11]) * costab[0xE];
 102     b1[0x12] = (samples[0x0D] - samples[0x12]) * costab[0xD];
 103     b1[0x13] = (samples[0x0C] - samples[0x13]) * costab[0xC];
 104     b1[0x14] = (samples[0x0B] - samples[0x14]) * costab[0xB];
 105     b1[0x15] = (samples[0x0A] - samples[0x15]) * costab[0xA];
 106     b1[0x16] = (samples[0x09] - samples[0x16]) * costab[0x9];
 107     b1[0x17] = (samples[0x08] - samples[0x17]) * costab[0x8];
 108     b1[0x18] = (samples[0x07] - samples[0x18]) * costab[0x7];
 109     b1[0x19] = (samples[0x06] - samples[0x19]) * costab[0x6];
 110     b1[0x1A] = (samples[0x05] - samples[0x1A]) * costab[0x5];
 111     b1[0x1B] = (samples[0x04] - samples[0x1B]) * costab[0x4];
 112     b1[0x1C] = (samples[0x03] - samples[0x1C]) * costab[0x3];
 113     b1[0x1D] = (samples[0x02] - samples[0x1D]) * costab[0x2];
 114     b1[0x1E] = (samples[0x01] - samples[0x1E]) * costab[0x1];
 115     b1[0x1F] = (samples[0x00] - samples[0x1F]) * costab[0x0];
 116
 117   }
 118   {
 119     register real *costab = mp3lib_pnts[1];
 120
 121     b2[0x00] = b1[0x00] + b1[0x0F];
 122     b2[0x01] = b1[0x01] + b1[0x0E];
 123     b2[0x02] = b1[0x02] + b1[0x0D];
 124     b2[0x03] = b1[0x03] + b1[0x0C];
 125     b2[0x04] = b1[0x04] + b1[0x0B];
 126     b2[0x05] = b1[0x05] + b1[0x0A];
 127     b2[0x06] = b1[0x06] + b1[0x09];
 128     b2[0x07] = b1[0x07] + b1[0x08];
 129     b2[0x08] = (b1[0x07] - b1[0x08]) * costab[7];
 130     b2[0x09] = (b1[0x06] - b1[0x09]) * costab[6];
 131     b2[0x0A] = (b1[0x05] - b1[0x0A]) * costab[5];
 132     b2[0x0B] = (b1[0x04] - b1[0x0B]) * costab[4];
 133     b2[0x0C] = (b1[0x03] - b1[0x0C]) * costab[3];
 134     b2[0x0D] = (b1[0x02] - b1[0x0D]) * costab[2];
 135     b2[0x0E] = (b1[0x01] - b1[0x0E]) * costab[1];
 136     b2[0x0F] = (b1[0x00] - b1[0x0F]) * costab[0];
 137     b2[0x10] = b1[0x10] + b1[0x1F];
 138     b2[0x11] = b1[0x11] + b1[0x1E];
 139     b2[0x12] = b1[0x12] + b1[0x1D];
 140     b2[0x13] = b1[0x13] + b1[0x1C];
 141     b2[0x14] = b1[0x14] + b1[0x1B];
 142     b2[0x15] = b1[0x15] + b1[0x1A];
 143     b2[0x16] = b1[0x16] + b1[0x19];
 144     b2[0x17] = b1[0x17] + b1[0x18];
 145     b2[0x18] = (b1[0x18] - b1[0x17]) * costab[7];
 146     b2[0x19] = (b1[0x19] - b1[0x16]) * costab[6];
 147     b2[0x1A] = (b1[0x1A] - b1[0x15]) * costab[5];
 148     b2[0x1B] = (b1[0x1B] - b1[0x14]) * costab[4];
 149     b2[0x1C] = (b1[0x1C] - b1[0x13]) * costab[3];
 150     b2[0x1D] = (b1[0x1D] - b1[0x12]) * costab[2];
 151     b2[0x1E] = (b1[0x1E] - b1[0x11]) * costab[1];
 152     b2[0x1F] = (b1[0x1F] - b1[0x10]) * costab[0];
 153
 154   }
 155
 156   {
 157     register real *costab = mp3lib_pnts[2];
 158
 159     b1[0x00] = b2[0x00] + b2[0x07];
 160     b1[0x01] = b2[0x01] + b2[0x06];
 161     b1[0x02] = b2[0x02] + b2[0x05];
 162     b1[0x03] = b2[0x03] + b2[0x04];
 163     b1[0x04] = (b2[0x03] - b2[0x04]) * costab[3];
 164     b1[0x05] = (b2[0x02] - b2[0x05]) * costab[2];
 165     b1[0x06] = (b2[0x01] - b2[0x06]) * costab[1];
 166     b1[0x07] = (b2[0x00] - b2[0x07]) * costab[0];
 167     b1[0x08] = b2[0x08] + b2[0x0F];
 168     b1[0x09] = b2[0x09] + b2[0x0E];
 169     b1[0x0A] = b2[0x0A] + b2[0x0D];
 170     b1[0x0B] = b2[0x0B] + b2[0x0C];
 171     b1[0x0C] = (b2[0x0C] - b2[0x0B]) * costab[3];
 172     b1[0x0D] = (b2[0x0D] - b2[0x0A]) * costab[2];
 173     b1[0x0E] = (b2[0x0E] - b2[0x09]) * costab[1];
 174     b1[0x0F] = (b2[0x0F] - b2[0x08]) * costab[0];
 175     b1[0x10] = b2[0x10] + b2[0x17];
 176     b1[0x11] = b2[0x11] + b2[0x16];
 177     b1[0x12] = b2[0x12] + b2[0x15];
 178     b1[0x13] = b2[0x13] + b2[0x14];
 179     b1[0x14] = (b2[0x13] - b2[0x14]) * costab[3];
 180     b1[0x15] = (b2[0x12] - b2[0x15]) * costab[2];
 181     b1[0x16] = (b2[0x11] - b2[0x16]) * costab[1];
 182     b1[0x17] = (b2[0x10] - b2[0x17]) * costab[0];
 183     b1[0x18] = b2[0x18] + b2[0x1F];
 184     b1[0x19] = b2[0x19] + b2[0x1E];
 185     b1[0x1A] = b2[0x1A] + b2[0x1D];
 186     b1[0x1B] = b2[0x1B] + b2[0x1C];
 187     b1[0x1C] = (b2[0x1C] - b2[0x1B]) * costab[3];
 188     b1[0x1D] = (b2[0x1D] - b2[0x1A]) * costab[2];
 189     b1[0x1E] = (b2[0x1E] - b2[0x19]) * costab[1];
 190     b1[0x1F] = (b2[0x1F] - b2[0x18]) * costab[0];
 191   }
 192
 193 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
 194
 195   // How does it work ?
 196   // the first three passes are reproducted in the three block below
 197   // all computations are done on a 4 elements vector
 198   // 'reverse' is a special perumtation vector used to reverse
 199   // the order of the elements inside a vector.
 200   // note that all loads/stores to b1 (b2) between passes 1 and 2 (2 and 3)
 201   // have been removed, all elements are stored inside b1vX (b2vX)
 202   {
 203     register vector float
 204       b1v0, b1v1, b1v2, b1v3,
 205       b1v4, b1v5, b1v6, b1v7;
 206     register vector float
 207       temp1, temp2;
 208
 209     {
 210       register real *costab = mp3lib_pnts[0];
 211
 212       register vector float
 213         samplesv1, samplesv2, samplesv3, samplesv4,
 214         samplesv5, samplesv6, samplesv7, samplesv8,
 215         samplesv9;
 216       register vector unsigned char samples_perm = vec_lvsl(0, samples);
 217       register vector float costabv1, costabv2, costabv3, costabv4, costabv5;
 218       register vector unsigned char costab_perm = vec_lvsl(0, costab);
 219
 220       samplesv1 = vec_ld(0, samples);
 221       samplesv2 = vec_ld(16, samples);
 222       samplesv1 = vec_perm(samplesv1, samplesv2, samples_perm);
 223       samplesv3 = vec_ld(32, samples);
 224       samplesv2 = vec_perm(samplesv2, samplesv3, samples_perm);
 225       samplesv4 = vec_ld(48, samples);
 226       samplesv3 = vec_perm(samplesv3, samplesv4, samples_perm);
 227       samplesv5 = vec_ld(64, samples);
 228       samplesv4 = vec_perm(samplesv4, samplesv5, samples_perm);
 229       samplesv6 = vec_ld(80, samples);
 230       samplesv5 = vec_perm(samplesv5, samplesv6, samples_perm);
 231       samplesv7 = vec_ld(96, samples);
 232       samplesv6 = vec_perm(samplesv6, samplesv7, samples_perm);
 233       samplesv8 = vec_ld(112, samples);
 234       samplesv7 = vec_perm(samplesv7, samplesv8, samples_perm);
 235       samplesv9 = vec_ld(128, samples);
 236       samplesv8 = vec_perm(samplesv8, samplesv9, samples_perm);
 237
 238       temp1 = vec_add(samplesv1,
 239                       vec_perm(samplesv8, samplesv8, reverse));
 240       //vec_st(temp1, 0, b1);
 241       b1v0 = temp1;
 242       temp1 = vec_add(samplesv2,
 243                       vec_perm(samplesv7, samplesv7, reverse));
 244       //vec_st(temp1, 16, b1);
 245       b1v1 = temp1;
 246       temp1 = vec_add(samplesv3,
 247                       vec_perm(samplesv6, samplesv6, reverse));
 248       //vec_st(temp1, 32, b1);
 249       b1v2 = temp1;
 250       temp1 = vec_add(samplesv4,
 251                       vec_perm(samplesv5, samplesv5, reverse));
 252       //vec_st(temp1, 48, b1);
 253       b1v3 = temp1;
 254
 255       costabv1 = vec_ld(0, costab);
 256       costabv2 = vec_ld(16, costab);
 257       costabv1 = vec_perm(costabv1, costabv2, costab_perm);
 258       costabv3 = vec_ld(32, costab);
 259       costabv2 = vec_perm(costabv2, costabv3, costab_perm);
 260       costabv4 = vec_ld(48, costab);
 261       costabv3 = vec_perm(costabv3, costabv4, costab_perm);
 262       costabv5 = vec_ld(64, costab);
 263       costabv4 = vec_perm(costabv4, costabv5, costab_perm);
 264
 265       temp1 = vec_sub(vec_perm(samplesv4, samplesv4, reverse),
 266                       samplesv5);
 267       temp2 = vec_madd(temp1,
 268                        vec_perm(costabv4, costabv4, reverse),
 269                        vczero);
 270       //vec_st(temp2, 64, b1);
 271       b1v4 = temp2;
 272
 273       temp1 = vec_sub(vec_perm(samplesv3, samplesv3, reverse),
 274                       samplesv6);
 275       temp2 = vec_madd(temp1,
 276                        vec_perm(costabv3, costabv3, reverse),
 277                        vczero);
 278       //vec_st(temp2, 80, b1);
 279       b1v5 = temp2;
 280       temp1 = vec_sub(vec_perm(samplesv2, samplesv2, reverse),
 281                       samplesv7);
 282       temp2 = vec_madd(temp1,
 283                        vec_perm(costabv2, costabv2, reverse),
 284                        vczero);
 285       //vec_st(temp2, 96, b1);
 286       b1v6 = temp2;
 287
 288       temp1 = vec_sub(vec_perm(samplesv1, samplesv1, reverse),
 289                       samplesv8);
 290       temp2 = vec_madd(temp1,
 291                        vec_perm(costabv1, costabv1, reverse),
 292                        vczero);
 293       //vec_st(temp2, 112, b1);
 294       b1v7 = temp2;
 295
 296     }
 297
 298     {
 299       register vector float
 300         b2v0, b2v1, b2v2, b2v3,
 301         b2v4, b2v5, b2v6, b2v7;
 302       {
 303         register real *costab = mp3lib_pnts[1];
 304         register vector float costabv1r, costabv2r, costabv1, costabv2, costabv3;
 305         register vector unsigned char costab_perm = vec_lvsl(0, costab);
 306
 307         costabv1 = vec_ld(0, costab);
 308         costabv2 = vec_ld(16, costab);
 309         costabv1 = vec_perm(costabv1, costabv2, costab_perm);
 310         costabv3  = vec_ld(32, costab);
 311         costabv2 = vec_perm(costabv2, costabv3 , costab_perm);
 312         costabv1r = vec_perm(costabv1, costabv1, reverse);
 313         costabv2r = vec_perm(costabv2, costabv2, reverse);
 314
 315         temp1 = vec_add(b1v0, vec_perm(b1v3, b1v3, reverse));
 316         //vec_st(temp1, 0, b2);
 317         b2v0 = temp1;
 318         temp1 = vec_add(b1v1, vec_perm(b1v2, b1v2, reverse));
 319         //vec_st(temp1, 16, b2);
 320         b2v1 = temp1;
 321         temp2 = vec_sub(vec_perm(b1v1, b1v1, reverse), b1v2);
 322         temp1 = vec_madd(temp2, costabv2r, vczero);
 323         //vec_st(temp1, 32, b2);
 324         b2v2 = temp1;
 325         temp2 = vec_sub(vec_perm(b1v0, b1v0, reverse), b1v3);
 326         temp1 = vec_madd(temp2, costabv1r, vczero);
 327         //vec_st(temp1, 48, b2);
 328         b2v3 = temp1;
 329         temp1 = vec_add(b1v4, vec_perm(b1v7, b1v7, reverse));
 330         //vec_st(temp1, 64, b2);
 331         b2v4 = temp1;
 332         temp1 = vec_add(b1v5, vec_perm(b1v6, b1v6, reverse));
 333         //vec_st(temp1, 80, b2);
 334         b2v5 = temp1;
 335         temp2 = vec_sub(b1v6, vec_perm(b1v5, b1v5, reverse));
 336         temp1 = vec_madd(temp2, costabv2r, vczero);
 337         //vec_st(temp1, 96, b2);
 338         b2v6 = temp1;
 339         temp2 = vec_sub(b1v7, vec_perm(b1v4, b1v4, reverse));
 340         temp1 = vec_madd(temp2, costabv1r, vczero);
 341         //vec_st(temp1, 112, b2);
 342         b2v7 = temp1;
 343       }
 344
 345       {
 346         register real *costab = mp3lib_pnts[2];
 347
 348
 349         vector float costabv1r, costabv1, costabv2;
 350         vector unsigned char costab_perm = vec_lvsl(0, costab);
 351
 352         costabv1 = vec_ld(0, costab);
 353         costabv2 = vec_ld(16, costab);
 354         costabv1 = vec_perm(costabv1, costabv2, costab_perm);
 355         costabv1r = vec_perm(costabv1, costabv1, reverse);
 356
 357         temp1 = vec_add(b2v0, vec_perm(b2v1, b2v1, reverse));
 358         vec_st(temp1, 0, b1);
 359         temp2 = vec_sub(vec_perm(b2v0, b2v0, reverse), b2v1);
 360         temp1 = vec_madd(temp2, costabv1r, vczero);
 361         vec_st(temp1, 16, b1);
 362
 363         temp1 = vec_add(b2v2, vec_perm(b2v3, b2v3, reverse));
 364         vec_st(temp1, 32, b1);
 365         temp2 = vec_sub(b2v3, vec_perm(b2v2, b2v2, reverse));
 366         temp1 = vec_madd(temp2, costabv1r, vczero);
 367         vec_st(temp1, 48, b1);
 368
 369         temp1 = vec_add(b2v4, vec_perm(b2v5, b2v5, reverse));
 370         vec_st(temp1, 64, b1);
 371         temp2 = vec_sub(vec_perm(b2v4, b2v4, reverse), b2v5);
 372         temp1 = vec_madd(temp2, costabv1r, vczero);
 373         vec_st(temp1, 80, b1);
 374
 375         temp1 = vec_add(b2v6, vec_perm(b2v7, b2v7, reverse));
 376         vec_st(temp1, 96, b1);
 377         temp2 = vec_sub(b2v7, vec_perm(b2v6, b2v6, reverse));
 378         temp1 = vec_madd(temp2, costabv1r, vczero);
 379         vec_st(temp1, 112, b1);
 380
 381       }
 382     }
 383   }
 384
 385 #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
 386
 387   {
 388     register real const cos0 = mp3lib_pnts[3][0];
 389     register real const cos1 = mp3lib_pnts[3][1];
 390
 391     b2[0x00] = b1[0x00] + b1[0x03];
 392     b2[0x01] = b1[0x01] + b1[0x02];
 393     b2[0x02] = (b1[0x01] - b1[0x02]) * cos1;
 394     b2[0x03] = (b1[0x00] - b1[0x03]) * cos0;
 395     b2[0x04] = b1[0x04] + b1[0x07];
 396     b2[0x05] = b1[0x05] + b1[0x06];
 397     b2[0x06] = (b1[0x06] - b1[0x05]) * cos1;
 398     b2[0x07] = (b1[0x07] - b1[0x04]) * cos0;
 399     b2[0x08] = b1[0x08] + b1[0x0B];
 400     b2[0x09] = b1[0x09] + b1[0x0A];
 401     b2[0x0A] = (b1[0x09] - b1[0x0A]) * cos1;
 402     b2[0x0B] = (b1[0x08] - b1[0x0B]) * cos0;
 403     b2[0x0C] = b1[0x0C] + b1[0x0F];
 404     b2[0x0D] = b1[0x0D] + b1[0x0E];
 405     b2[0x0E] = (b1[0x0E] - b1[0x0D]) * cos1;
 406     b2[0x0F] = (b1[0x0F] - b1[0x0C]) * cos0;
 407     b2[0x10] = b1[0x10] + b1[0x13];
 408     b2[0x11] = b1[0x11] + b1[0x12];
 409     b2[0x12] = (b1[0x11] - b1[0x12]) * cos1;
 410     b2[0x13] = (b1[0x10] - b1[0x13]) * cos0;
 411     b2[0x14] = b1[0x14] + b1[0x17];
 412     b2[0x15] = b1[0x15] + b1[0x16];
 413     b2[0x16] = (b1[0x16] - b1[0x15]) * cos1;
 414     b2[0x17] = (b1[0x17] - b1[0x14]) * cos0;
 415     b2[0x18] = b1[0x18] + b1[0x1B];
 416     b2[0x19] = b1[0x19] + b1[0x1A];
 417     b2[0x1A] = (b1[0x19] - b1[0x1A]) * cos1;
 418     b2[0x1B] = (b1[0x18] - b1[0x1B]) * cos0;
 419     b2[0x1C] = b1[0x1C] + b1[0x1F];
 420     b2[0x1D] = b1[0x1D] + b1[0x1E];
 421     b2[0x1E] = (b1[0x1E] - b1[0x1D]) * cos1;
 422     b2[0x1F] = (b1[0x1F] - b1[0x1C]) * cos0;
 423   }
 424
 425   {
 426     register real const cos0 = mp3lib_pnts[4][0];
 427
 428     b1[0x00] = b2[0x00] + b2[0x01];
 429     b1[0x01] = (b2[0x00] - b2[0x01]) * cos0;
 430     b1[0x02] = b2[0x02] + b2[0x03];
 431     b1[0x03] = (b2[0x03] - b2[0x02]) * cos0;
 432     b1[0x02] += b1[0x03];
 433
 434     b1[0x04] = b2[0x04] + b2[0x05];
 435     b1[0x05] = (b2[0x04] - b2[0x05]) * cos0;
 436     b1[0x06] = b2[0x06] + b2[0x07];
 437     b1[0x07] = (b2[0x07] - b2[0x06]) * cos0;
 438     b1[0x06] += b1[0x07];
 439     b1[0x04] += b1[0x06];
 440     b1[0x06] += b1[0x05];
 441     b1[0x05] += b1[0x07];
 442
 443     b1[0x08] = b2[0x08] + b2[0x09];
 444     b1[0x09] = (b2[0x08] - b2[0x09]) * cos0;
 445     b1[0x0A] = b2[0x0A] + b2[0x0B];
 446     b1[0x0B] = (b2[0x0B] - b2[0x0A]) * cos0;
 447     b1[0x0A] += b1[0x0B];
 448
 449     b1[0x0C] = b2[0x0C] + b2[0x0D];
 450     b1[0x0D] = (b2[0x0C] - b2[0x0D]) * cos0;
 451     b1[0x0E] = b2[0x0E] + b2[0x0F];
 452     b1[0x0F] = (b2[0x0F] - b2[0x0E]) * cos0;
 453     b1[0x0E] += b1[0x0F];
 454     b1[0x0C] += b1[0x0E];
 455     b1[0x0E] += b1[0x0D];
 456     b1[0x0D] += b1[0x0F];
 457
 458     b1[0x10] = b2[0x10] + b2[0x11];
 459     b1[0x11] = (b2[0x10] - b2[0x11]) * cos0;
 460     b1[0x12] = b2[0x12] + b2[0x13];
 461     b1[0x13] = (b2[0x13] - b2[0x12]) * cos0;
 462     b1[0x12] += b1[0x13];
 463
 464     b1[0x14] = b2[0x14] + b2[0x15];
 465     b1[0x15] = (b2[0x14] - b2[0x15]) * cos0;
 466     b1[0x16] = b2[0x16] + b2[0x17];
 467     b1[0x17] = (b2[0x17] - b2[0x16]) * cos0;
 468     b1[0x16] += b1[0x17];
 469     b1[0x14] += b1[0x16];
 470     b1[0x16] += b1[0x15];
 471     b1[0x15] += b1[0x17];
 472
 473     b1[0x18] = b2[0x18] + b2[0x19];
 474     b1[0x19] = (b2[0x18] - b2[0x19]) * cos0;
 475     b1[0x1A] = b2[0x1A] + b2[0x1B];
 476     b1[0x1B] = (b2[0x1B] - b2[0x1A]) * cos0;
 477     b1[0x1A] += b1[0x1B];
 478
 479     b1[0x1C] = b2[0x1C] + b2[0x1D];
 480     b1[0x1D] = (b2[0x1C] - b2[0x1D]) * cos0;
 481     b1[0x1E] = b2[0x1E] + b2[0x1F];
 482     b1[0x1F] = (b2[0x1F] - b2[0x1E]) * cos0;
 483     b1[0x1E] += b1[0x1F];
 484     b1[0x1C] += b1[0x1E];
 485     b1[0x1E] += b1[0x1D];
 486     b1[0x1D] += b1[0x1F];
 487   }
 488
 489   out0[0x10*16] = b1[0x00];
 490   out0[0x10*12] = b1[0x04];
 491   out0[0x10* 8] = b1[0x02];
 492   out0[0x10* 4] = b1[0x06];
 493   out0[0x10* 0] = b1[0x01];
 494   out1[0x10* 0] = b1[0x01];
 495   out1[0x10* 4] = b1[0x05];
 496   out1[0x10* 8] = b1[0x03];
 497   out1[0x10*12] = b1[0x07];
 498
 499   b1[0x08] += b1[0x0C];
 500   out0[0x10*14] = b1[0x08];
 501   b1[0x0C] += b1[0x0a];
 502   out0[0x10*10] = b1[0x0C];
 503   b1[0x0A] += b1[0x0E];
 504   out0[0x10* 6] = b1[0x0A];
 505   b1[0x0E] += b1[0x09];
 506   out0[0x10* 2] = b1[0x0E];
 507   b1[0x09] += b1[0x0D];
 508   out1[0x10* 2] = b1[0x09];
 509   b1[0x0D] += b1[0x0B];
 510   out1[0x10* 6] = b1[0x0D];
 511   b1[0x0B] += b1[0x0F];
 512   out1[0x10*10] = b1[0x0B];
 513   out1[0x10*14] = b1[0x0F];
 514
 515   b1[0x18] += b1[0x1C];
 516   out0[0x10*15] = b1[0x10] + b1[0x18];
 517   out0[0x10*13] = b1[0x18] + b1[0x14];
 518   b1[0x1C] += b1[0x1a];
 519   out0[0x10*11] = b1[0x14] + b1[0x1C];
 520   out0[0x10* 9] = b1[0x1C] + b1[0x12];
 521   b1[0x1A] += b1[0x1E];
 522   out0[0x10* 7] = b1[0x12] + b1[0x1A];
 523   out0[0x10* 5] = b1[0x1A] + b1[0x16];
 524   b1[0x1E] += b1[0x19];
 525   out0[0x10* 3] = b1[0x16] + b1[0x1E];
 526   out0[0x10* 1] = b1[0x1E] + b1[0x11];
 527   b1[0x19] += b1[0x1D];
 528   out1[0x10* 1] = b1[0x11] + b1[0x19];
 529   out1[0x10* 3] = b1[0x19] + b1[0x15];
 530   b1[0x1D] += b1[0x1B];
 531   out1[0x10* 5] = b1[0x15] + b1[0x1D];
 532   out1[0x10* 7] = b1[0x1D] + b1[0x13];
 533   b1[0x1B] += b1[0x1F];
 534   out1[0x10* 9] = b1[0x13] + b1[0x1B];
 535   out1[0x10*11] = b1[0x1B] + b1[0x17];
 536   out1[0x10*13] = b1[0x17] + b1[0x1F];
 537   out1[0x10*15] = b1[0x1F];
 538 }
 539
 540 #endif /* HAVE_ALTIVEC */
 541