src/fftw/fhb_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 13:55:59 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-backward 5 */
  27
  28 /*
  29  * This function contains 64 FP additions, 42 FP multiplications,
  30  * (or, 42 additions, 20 multiplications, 22 fused multiply/add),
  31  * 25 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 K951056516 = FFTW_KONST(+0.951056516295153572116439333379382143405698634);
  36 static const fftw_real K587785252 = FFTW_KONST(+0.587785252292473129168705954639072768597652438);
  37 static const fftw_real K2_000000000 = FFTW_KONST(+2.000000000000000000000000000000000000000000000);
  38 static const fftw_real K500000000 = FFTW_KONST(+0.500000000000000000000000000000000000000000000);
  39 static const fftw_real K1_118033988 = FFTW_KONST(+1.118033988749894848204586834365638117720309180);
  40 static const fftw_real K1_175570504 = FFTW_KONST(+1.175570504584946258337411909278145537195304875);
  41 static const fftw_real K1_902113032 = FFTW_KONST(+1.902113032590307144232878666758764286811397268);
  42
  43 /*
  44  * Generator Id's :
  45  * $Id$
  46  * $Id$
  47  * $Id$
  48  */
  49
  50 void fftw_hc2hc_backward_5(fftw_real *A, const fftw_complex *W, int iostride, int m, int dist)
  51 {
  52      int i;
  53      fftw_real *X;
  54      fftw_real *Y;
  55      X = A;
  56      Y = A + (5 * iostride);
  57      {
  58           fftw_real tmp70;
  59           fftw_real tmp72;
  60           fftw_real tmp61;
  61           fftw_real tmp64;
  62           fftw_real tmp65;
  63           fftw_real tmp66;
  64           fftw_real tmp71;
  65           fftw_real tmp67;
  66           ASSERT_ALIGNED_DOUBLE();
  67           {
  68                fftw_real tmp68;
  69                fftw_real tmp69;
  70                fftw_real tmp62;
  71                fftw_real tmp63;
  72                ASSERT_ALIGNED_DOUBLE();
  73                tmp68 = Y[-iostride];
  74                tmp69 = Y[-2 * iostride];
  75                tmp70 = (K1_902113032 * tmp68) + (K1_175570504 * tmp69);
  76                tmp72 = (K1_902113032 * tmp69) - (K1_175570504 * tmp68);
  77                tmp61 = X[0];
  78                tmp62 = X[iostride];
  79                tmp63 = X[2 * iostride];
  80                tmp64 = tmp62 + tmp63;
  81                tmp65 = K1_118033988 * (tmp62 - tmp63);
  82                tmp66 = tmp61 - (K500000000 * tmp64);
  83           }
  84           X[0] = tmp61 + (K2_000000000 * tmp64);
  85           tmp71 = tmp66 - tmp65;
  86           X[3 * iostride] = tmp71 - tmp72;
  87           X[2 * iostride] = tmp71 + tmp72;
  88           tmp67 = tmp65 + tmp66;
  89           X[4 * iostride] = tmp67 + tmp70;
  90           X[iostride] = tmp67 - tmp70;
  91      }
  92      X = X + dist;
  93      Y = Y - dist;
  94      for (i = 2; i < m; i = i + 2, X = X + dist, Y = Y - dist, W = W + 4) {
  95           fftw_real tmp13;
  96           fftw_real tmp32;
  97           fftw_real tmp50;
  98           fftw_real tmp39;
  99           fftw_real tmp20;
 100           fftw_real tmp38;
 101           fftw_real tmp21;
 102           fftw_real tmp34;
 103           fftw_real tmp28;
 104           fftw_real tmp33;
 105           fftw_real tmp43;
 106           fftw_real tmp55;
 107           ASSERT_ALIGNED_DOUBLE();
 108           {
 109                fftw_real tmp19;
 110                fftw_real tmp31;
 111                fftw_real tmp16;
 112                fftw_real tmp30;
 113                ASSERT_ALIGNED_DOUBLE();
 114                tmp13 = X[0];
 115                {
 116                     fftw_real tmp17;
 117                     fftw_real tmp18;
 118                     fftw_real tmp14;
 119                     fftw_real tmp15;
 120                     ASSERT_ALIGNED_DOUBLE();
 121                     tmp17 = X[2 * iostride];
 122                     tmp18 = Y[-3 * iostride];
 123                     tmp19 = tmp17 + tmp18;
 124                     tmp31 = tmp17 - tmp18;
 125                     tmp14 = X[iostride];
 126                     tmp15 = Y[-4 * iostride];
 127                     tmp16 = tmp14 + tmp15;
 128                     tmp30 = tmp14 - tmp15;
 129                }
 130                tmp32 = (K587785252 * tmp30) - (K951056516 * tmp31);
 131                tmp50 = (K951056516 * tmp30) + (K587785252 * tmp31);
 132                tmp39 = K559016994 * (tmp16 - tmp19);
 133                tmp20 = tmp16 + tmp19;
 134                tmp38 = tmp13 - (K250000000 * tmp20);
 135           }
 136           {
 137                fftw_real tmp27;
 138                fftw_real tmp42;
 139                fftw_real tmp24;
 140                fftw_real tmp41;
 141                ASSERT_ALIGNED_DOUBLE();
 142                tmp21 = Y[0];
 143                {
 144                     fftw_real tmp25;
 145                     fftw_real tmp26;
 146                     fftw_real tmp22;
 147                     fftw_real tmp23;
 148                     ASSERT_ALIGNED_DOUBLE();
 149                     tmp25 = Y[-2 * iostride];
 150                     tmp26 = X[3 * iostride];
 151                     tmp27 = tmp25 - tmp26;
 152                     tmp42 = tmp25 + tmp26;
 153                     tmp22 = Y[-iostride];
 154                     tmp23 = X[4 * iostride];
 155                     tmp24 = tmp22 - tmp23;
 156                     tmp41 = tmp22 + tmp23;
 157                }
 158                tmp34 = K559016994 * (tmp24 - tmp27);
 159                tmp28 = tmp24 + tmp27;
 160                tmp33 = tmp21 - (K250000000 * tmp28);
 161                tmp43 = (K587785252 * tmp41) - (K951056516 * tmp42);
 162                tmp55 = (K951056516 * tmp41) + (K587785252 * tmp42);
 163           }
 164           X[0] = tmp13 + tmp20;
 165           {
 166                fftw_real tmp52;
 167                fftw_real tmp58;
 168                fftw_real tmp56;
 169                fftw_real tmp60;
 170                fftw_real tmp51;
 171                fftw_real tmp54;
 172                ASSERT_ALIGNED_DOUBLE();
 173                tmp51 = tmp34 + tmp33;
 174                tmp52 = tmp50 + tmp51;
 175                tmp58 = tmp51 - tmp50;
 176                tmp54 = tmp39 + tmp38;
 177                tmp56 = tmp54 - tmp55;
 178                tmp60 = tmp54 + tmp55;
 179                {
 180                     fftw_real tmp49;
 181                     fftw_real tmp53;
 182                     fftw_real tmp57;
 183                     fftw_real tmp59;
 184                     ASSERT_ALIGNED_DOUBLE();
 185                     tmp49 = c_re(W[0]);
 186                     tmp53 = c_im(W[0]);
 187                     Y[-3 * iostride] = (tmp49 * tmp52) - (tmp53 * tmp56);
 188                     X[iostride] = (tmp53 * tmp52) + (tmp49 * tmp56);
 189                     tmp57 = c_re(W[3]);
 190                     tmp59 = c_im(W[3]);
 191                     Y[0] = (tmp57 * tmp58) - (tmp59 * tmp60);
 192                     X[4 * iostride] = (tmp59 * tmp58) + (tmp57 * tmp60);
 193                }
 194           }
 195           Y[-4 * iostride] = tmp21 + tmp28;
 196           {
 197                fftw_real tmp36;
 198                fftw_real tmp46;
 199                fftw_real tmp44;
 200                fftw_real tmp48;
 201                fftw_real tmp35;
 202                fftw_real tmp40;
 203                ASSERT_ALIGNED_DOUBLE();
 204                tmp35 = tmp33 - tmp34;
 205                tmp36 = tmp32 + tmp35;
 206                tmp46 = tmp35 - tmp32;
 207                tmp40 = tmp38 - tmp39;
 208                tmp44 = tmp40 - tmp43;
 209                tmp48 = tmp40 + tmp43;
 210                {
 211                     fftw_real tmp29;
 212                     fftw_real tmp37;
 213                     fftw_real tmp45;
 214                     fftw_real tmp47;
 215                     ASSERT_ALIGNED_DOUBLE();
 216                     tmp29 = c_re(W[1]);
 217                     tmp37 = c_im(W[1]);
 218                     Y[-2 * iostride] = (tmp29 * tmp36) - (tmp37 * tmp44);
 219                     X[2 * iostride] = (tmp37 * tmp36) + (tmp29 * tmp44);
 220                     tmp45 = c_re(W[2]);
 221                     tmp47 = c_im(W[2]);
 222                     Y[-iostride] = (tmp45 * tmp46) - (tmp47 * tmp48);
 223                     X[3 * iostride] = (tmp47 * tmp46) + (tmp45 * tmp48);
 224                }
 225           }
 226      }
 227      if (i == m) {
 228           fftw_real tmp10;
 229           fftw_real tmp12;
 230           fftw_real tmp1;
 231           fftw_real tmp4;
 232           fftw_real tmp5;
 233           fftw_real tmp6;
 234           fftw_real tmp11;
 235           fftw_real tmp7;
 236           ASSERT_ALIGNED_DOUBLE();
 237           {
 238                fftw_real tmp8;
 239                fftw_real tmp9;
 240                fftw_real tmp2;
 241                fftw_real tmp3;
 242                ASSERT_ALIGNED_DOUBLE();
 243                tmp8 = Y[-iostride];
 244                tmp9 = Y[0];
 245                tmp10 = (K1_902113032 * tmp8) + (K1_175570504 * tmp9);
 246                tmp12 = (K1_175570504 * tmp8) - (K1_902113032 * tmp9);
 247                tmp1 = X[2 * iostride];
 248                tmp2 = X[iostride];
 249                tmp3 = X[0];
 250                tmp4 = tmp2 + tmp3;
 251                tmp5 = (K500000000 * tmp4) - tmp1;
 252                tmp6 = K1_118033988 * (tmp3 - tmp2);
 253           }
 254           X[0] = tmp1 + (K2_000000000 * tmp4);
 255           tmp11 = tmp6 - tmp5;
 256           X[2 * iostride] = tmp11 + tmp12;
 257           X[3 * iostride] = tmp12 - tmp11;
 258           tmp7 = tmp5 + tmp6;
 259           X[iostride] = tmp7 - tmp10;
 260           X[4 * iostride] = -(tmp7 + tmp10);
 261      }
 262 }
 263
 264 static const int twiddle_order[] =
 265 {1, 2, 3, 4};
 266 fftw_codelet_desc fftw_hc2hc_backward_5_desc =
 267 {
 268      "fftw_hc2hc_backward_5",
 269      (void (*)()) fftw_hc2hc_backward_5,
 270      5,
 271      FFTW_BACKWARD,
 272      FFTW_HC2HC,
 273      124,
 274      4,
 275      twiddle_order,
 276 };