src/fftw/fhf_5.c

   1 /*
   2  * Copyright (c) 1997-1999 Massachusetts Institute of Technology
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  17  *
  18  */
  19
  20 /* This file was automatically generated --- DO NOT EDIT */
  21 /* Generated on Tue May 18 13:55:30 EDT 1999 */
  22
  23 #include <fftw-int.h>
  24 #include <fftw.h>
  25
  26 /* Generated by: ./genfft -magic-alignment-check -magic-twiddle-load-all -magic-variables 4 -magic-loopi -hc2hc-forward 5 */
  27
  28 /*
  29  * This function contains 64 FP additions, 40 FP multiplications,
  30  * (or, 44 additions, 20 multiplications, 20 fused multiply/add),
  31  * 27 stack variables, and 40 memory accesses
  32  */
  33 static const fftw_real K250000000 = FFTW_KONST(+0.250000000000000000000000000000000000000000000);
  34 static const fftw_real K559016994 = FFTW_KONST(+0.559016994374947424102293417182819058860154590);
  35 static const fftw_real K587785252 = FFTW_KONST(+0.587785252292473129168705954639072768597652438);
  36 static const fftw_real K951056516 = FFTW_KONST(+0.951056516295153572116439333379382143405698634);
  37
  38 /*
  39  * Generator Id's :
  40  * $Id$
  41  * $Id$
  42  * $Id$
  43  */
  44
  45 void fftw_hc2hc_forward_5(fftw_real *A, const fftw_complex *W, int iostride, int m, int dist)
  46 {
  47      int i;
  48      fftw_real *X;
  49      fftw_real *Y;
  50      X = A;
  51      Y = A + (5 * iostride);
  52      {
  53           fftw_real tmp70;
  54           fftw_real tmp67;
  55           fftw_real tmp68;
  56           fftw_real tmp63;
  57           fftw_real tmp71;
  58           fftw_real tmp66;
  59           fftw_real tmp69;
  60           fftw_real tmp72;
  61           ASSERT_ALIGNED_DOUBLE();
  62           tmp70 = X[0];
  63           {
  64                fftw_real tmp61;
  65                fftw_real tmp62;
  66                fftw_real tmp64;
  67                fftw_real tmp65;
  68                ASSERT_ALIGNED_DOUBLE();
  69                tmp61 = X[4 * iostride];
  70                tmp62 = X[iostride];
  71                tmp67 = tmp62 + tmp61;
  72                tmp64 = X[2 * iostride];
  73                tmp65 = X[3 * iostride];
  74                tmp68 = tmp64 + tmp65;
  75                tmp63 = tmp61 - tmp62;
  76                tmp71 = tmp67 + tmp68;
  77                tmp66 = tmp64 - tmp65;
  78           }
  79           Y[-iostride] = (K951056516 * tmp63) - (K587785252 * tmp66);
  80           Y[-2 * iostride] = (K587785252 * tmp63) + (K951056516 * tmp66);
  81           X[0] = tmp70 + tmp71;
  82           tmp69 = K559016994 * (tmp67 - tmp68);
  83           tmp72 = tmp70 - (K250000000 * tmp71);
  84           X[iostride] = tmp69 + tmp72;
  85           X[2 * iostride] = tmp72 - tmp69;
  86      }
  87      X = X + dist;
  88      Y = Y - dist;
  89      for (i = 2; i < m; i = i + 2, X = X + dist, Y = Y - dist, W = W + 4) {
  90           fftw_real tmp13;
  91           fftw_real tmp52;
  92           fftw_real tmp42;
  93           fftw_real tmp45;
  94           fftw_real tmp49;
  95           fftw_real tmp50;
  96           fftw_real tmp51;
  97           fftw_real tmp54;
  98           fftw_real tmp53;
  99           fftw_real tmp24;
 100           fftw_real tmp35;
 101           fftw_real tmp36;
 102           ASSERT_ALIGNED_DOUBLE();
 103           tmp13 = X[0];
 104           tmp52 = Y[-4 * iostride];
 105           {
 106                fftw_real tmp18;
 107                fftw_real tmp40;
 108                fftw_real tmp34;
 109                fftw_real tmp44;
 110                fftw_real tmp23;
 111                fftw_real tmp41;
 112                fftw_real tmp29;
 113                fftw_real tmp43;
 114                ASSERT_ALIGNED_DOUBLE();
 115                {
 116                     fftw_real tmp15;
 117                     fftw_real tmp17;
 118                     fftw_real tmp14;
 119                     fftw_real tmp16;
 120                     ASSERT_ALIGNED_DOUBLE();
 121                     tmp15 = X[iostride];
 122                     tmp17 = Y[-3 * iostride];
 123                     tmp14 = c_re(W[0]);
 124                     tmp16 = c_im(W[0]);
 125                     tmp18 = (tmp14 * tmp15) - (tmp16 * tmp17);
 126                     tmp40 = (tmp16 * tmp15) + (tmp14 * tmp17);
 127                }
 128                {
 129                     fftw_real tmp31;
 130                     fftw_real tmp33;
 131                     fftw_real tmp30;
 132                     fftw_real tmp32;
 133                     ASSERT_ALIGNED_DOUBLE();
 134                     tmp31 = X[3 * iostride];
 135                     tmp33 = Y[-iostride];
 136                     tmp30 = c_re(W[2]);
 137                     tmp32 = c_im(W[2]);
 138                     tmp34 = (tmp30 * tmp31) - (tmp32 * tmp33);
 139                     tmp44 = (tmp32 * tmp31) + (tmp30 * tmp33);
 140                }
 141                {
 142                     fftw_real tmp20;
 143                     fftw_real tmp22;
 144                     fftw_real tmp19;
 145                     fftw_real tmp21;
 146                     ASSERT_ALIGNED_DOUBLE();
 147                     tmp20 = X[4 * iostride];
 148                     tmp22 = Y[0];
 149                     tmp19 = c_re(W[3]);
 150                     tmp21 = c_im(W[3]);
 151                     tmp23 = (tmp19 * tmp20) - (tmp21 * tmp22);
 152                     tmp41 = (tmp21 * tmp20) + (tmp19 * tmp22);
 153                }
 154                {
 155                     fftw_real tmp26;
 156                     fftw_real tmp28;
 157                     fftw_real tmp25;
 158                     fftw_real tmp27;
 159                     ASSERT_ALIGNED_DOUBLE();
 160                     tmp26 = X[2 * iostride];
 161                     tmp28 = Y[-2 * iostride];
 162                     tmp25 = c_re(W[1]);
 163                     tmp27 = c_im(W[1]);
 164                     tmp29 = (tmp25 * tmp26) - (tmp27 * tmp28);
 165                     tmp43 = (tmp27 * tmp26) + (tmp25 * tmp28);
 166                }
 167                tmp42 = tmp40 - tmp41;
 168                tmp45 = tmp43 - tmp44;
 169                tmp49 = tmp40 + tmp41;
 170                tmp50 = tmp43 + tmp44;
 171                tmp51 = tmp49 + tmp50;
 172                tmp54 = tmp29 - tmp34;
 173                tmp53 = tmp18 - tmp23;
 174                tmp24 = tmp18 + tmp23;
 175                tmp35 = tmp29 + tmp34;
 176                tmp36 = tmp24 + tmp35;
 177           }
 178           X[0] = tmp13 + tmp36;
 179           {
 180                fftw_real tmp46;
 181                fftw_real tmp48;
 182                fftw_real tmp39;
 183                fftw_real tmp47;
 184                fftw_real tmp37;
 185                fftw_real tmp38;
 186                ASSERT_ALIGNED_DOUBLE();
 187                tmp46 = (K951056516 * tmp42) + (K587785252 * tmp45);
 188                tmp48 = (K951056516 * tmp45) - (K587785252 * tmp42);
 189                tmp37 = K559016994 * (tmp24 - tmp35);
 190                tmp38 = tmp13 - (K250000000 * tmp36);
 191                tmp39 = tmp37 + tmp38;
 192                tmp47 = tmp38 - tmp37;
 193                Y[-4 * iostride] = tmp39 - tmp46;
 194                X[iostride] = tmp39 + tmp46;
 195                X[2 * iostride] = tmp47 - tmp48;
 196                Y[-3 * iostride] = tmp47 + tmp48;
 197           }
 198           Y[0] = tmp51 + tmp52;
 199           {
 200                fftw_real tmp55;
 201                fftw_real tmp60;
 202                fftw_real tmp58;
 203                fftw_real tmp59;
 204                fftw_real tmp56;
 205                fftw_real tmp57;
 206                ASSERT_ALIGNED_DOUBLE();
 207                tmp55 = (K951056516 * tmp53) + (K587785252 * tmp54);
 208                tmp60 = (K951056516 * tmp54) - (K587785252 * tmp53);
 209                tmp56 = K559016994 * (tmp49 - tmp50);
 210                tmp57 = tmp52 - (K250000000 * tmp51);
 211                tmp58 = tmp56 + tmp57;
 212                tmp59 = tmp57 - tmp56;
 213                X[4 * iostride] = -(tmp55 + tmp58);
 214                Y[-iostride] = tmp58 - tmp55;
 215                X[3 * iostride] = -(tmp59 - tmp60);
 216                Y[-2 * iostride] = tmp60 + tmp59;
 217           }
 218      }
 219      if (i == m) {
 220           fftw_real tmp8;
 221           fftw_real tmp3;
 222           fftw_real tmp6;
 223           fftw_real tmp9;
 224           fftw_real tmp12;
 225           fftw_real tmp11;
 226           fftw_real tmp7;
 227           fftw_real tmp10;
 228           ASSERT_ALIGNED_DOUBLE();
 229           tmp8 = X[0];
 230           {
 231                fftw_real tmp1;
 232                fftw_real tmp2;
 233                fftw_real tmp4;
 234                fftw_real tmp5;
 235                ASSERT_ALIGNED_DOUBLE();
 236                tmp1 = X[2 * iostride];
 237                tmp2 = X[3 * iostride];
 238                tmp3 = tmp1 - tmp2;
 239                tmp4 = X[4 * iostride];
 240                tmp5 = X[iostride];
 241                tmp6 = tmp4 - tmp5;
 242                tmp9 = tmp3 + tmp6;
 243                tmp12 = tmp4 + tmp5;
 244                tmp11 = tmp1 + tmp2;
 245           }
 246           X[2 * iostride] = tmp8 + tmp9;
 247           tmp7 = K559016994 * (tmp3 - tmp6);
 248           tmp10 = tmp8 - (K250000000 * tmp9);
 249           X[0] = tmp7 + tmp10;
 250           X[iostride] = tmp10 - tmp7;
 251           Y[0] = -((K951056516 * tmp11) + (K587785252 * tmp12));
 252           Y[-iostride] = -((K951056516 * tmp12) - (K587785252 * tmp11));
 253      }
 254 }
 255
 256 static const int twiddle_order[] =
 257 {1, 2, 3, 4};
 258 fftw_codelet_desc fftw_hc2hc_forward_5_desc =
 259 {
 260      "fftw_hc2hc_forward_5",
 261      (void (*)()) fftw_hc2hc_forward_5,
 262      5,
 263      FFTW_FORWARD,
 264      FFTW_HC2HC,
 265      113,
 266      4,
 267      twiddle_order,
 268 };