Rev 3 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 2 | pj | 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 13:55:34 EDT 1999 */ |
||
| 22 | |||
| 107 | pj | 23 | #include <fftw-int.h> |
| 24 | #include <fftw.h> |
||
| 2 | pj | 25 | |
| 26 | /* Generated by: ./genfft -magic-alignment-check -magic-twiddle-load-all -magic-variables 4 -magic-loopi -hc2hc-forward 8 */ |
||
| 27 | |||
| 28 | /* |
||
| 29 | * This function contains 108 FP additions, 44 FP multiplications, |
||
| 30 | * (or, 90 additions, 26 multiplications, 18 fused multiply/add), |
||
| 31 | * 29 stack variables, and 64 memory accesses |
||
| 32 | */ |
||
| 33 | static const fftw_real K382683432 = FFTW_KONST(+0.382683432365089771728459984030398866761344562); |
||
| 34 | static const fftw_real K923879532 = FFTW_KONST(+0.923879532511286756128183189396788286822416626); |
||
| 35 | static const fftw_real K707106781 = FFTW_KONST(+0.707106781186547524400844362104849039284835938); |
||
| 36 | |||
| 37 | /* |
||
| 38 | * Generator Id's : |
||
| 107 | pj | 39 | * $Id: fhf_8.c,v 1.2 2003-03-24 11:14:58 pj Exp $ |
| 40 | * $Id: fhf_8.c,v 1.2 2003-03-24 11:14:58 pj Exp $ |
||
| 41 | * $Id: fhf_8.c,v 1.2 2003-03-24 11:14:58 pj Exp $ |
||
| 2 | pj | 42 | */ |
| 43 | |||
| 44 | void fftw_hc2hc_forward_8(fftw_real *A, const fftw_complex *W, int iostride, int m, int dist) |
||
| 45 | { |
||
| 46 | int i; |
||
| 47 | fftw_real *X; |
||
| 48 | fftw_real *Y; |
||
| 49 | X = A; |
||
| 50 | Y = A + (8 * iostride); |
||
| 51 | { |
||
| 52 | fftw_real tmp105; |
||
| 53 | fftw_real tmp109; |
||
| 54 | fftw_real tmp115; |
||
| 55 | fftw_real tmp121; |
||
| 56 | fftw_real tmp108; |
||
| 57 | fftw_real tmp118; |
||
| 58 | fftw_real tmp112; |
||
| 59 | fftw_real tmp120; |
||
| 60 | ASSERT_ALIGNED_DOUBLE(); |
||
| 61 | { |
||
| 62 | fftw_real tmp103; |
||
| 63 | fftw_real tmp104; |
||
| 64 | fftw_real tmp113; |
||
| 65 | fftw_real tmp114; |
||
| 66 | ASSERT_ALIGNED_DOUBLE(); |
||
| 67 | tmp103 = X[0]; |
||
| 68 | tmp104 = X[4 * iostride]; |
||
| 69 | tmp105 = tmp103 + tmp104; |
||
| 70 | tmp109 = tmp103 - tmp104; |
||
| 71 | tmp113 = X[7 * iostride]; |
||
| 72 | tmp114 = X[3 * iostride]; |
||
| 73 | tmp115 = tmp113 - tmp114; |
||
| 74 | tmp121 = tmp113 + tmp114; |
||
| 75 | } |
||
| 76 | { |
||
| 77 | fftw_real tmp106; |
||
| 78 | fftw_real tmp107; |
||
| 79 | fftw_real tmp110; |
||
| 80 | fftw_real tmp111; |
||
| 81 | ASSERT_ALIGNED_DOUBLE(); |
||
| 82 | tmp106 = X[2 * iostride]; |
||
| 83 | tmp107 = X[6 * iostride]; |
||
| 84 | tmp108 = tmp106 + tmp107; |
||
| 85 | tmp118 = tmp106 - tmp107; |
||
| 86 | tmp110 = X[iostride]; |
||
| 87 | tmp111 = X[5 * iostride]; |
||
| 88 | tmp112 = tmp110 - tmp111; |
||
| 89 | tmp120 = tmp110 + tmp111; |
||
| 90 | } |
||
| 91 | { |
||
| 92 | fftw_real tmp119; |
||
| 93 | fftw_real tmp122; |
||
| 94 | fftw_real tmp116; |
||
| 95 | fftw_real tmp117; |
||
| 96 | ASSERT_ALIGNED_DOUBLE(); |
||
| 97 | X[2 * iostride] = tmp105 - tmp108; |
||
| 98 | tmp119 = tmp105 + tmp108; |
||
| 99 | tmp122 = tmp120 + tmp121; |
||
| 100 | X[4 * iostride] = tmp119 - tmp122; |
||
| 101 | X[0] = tmp119 + tmp122; |
||
| 102 | Y[-2 * iostride] = tmp121 - tmp120; |
||
| 103 | tmp116 = K707106781 * (tmp112 + tmp115); |
||
| 104 | X[3 * iostride] = tmp109 - tmp116; |
||
| 105 | X[iostride] = tmp109 + tmp116; |
||
| 106 | tmp117 = K707106781 * (tmp115 - tmp112); |
||
| 107 | Y[-iostride] = tmp117 - tmp118; |
||
| 108 | Y[-3 * iostride] = tmp118 + tmp117; |
||
| 109 | } |
||
| 110 | } |
||
| 111 | X = X + dist; |
||
| 112 | Y = Y - dist; |
||
| 113 | for (i = 2; i < m; i = i + 2, X = X + dist, Y = Y - dist, W = W + 7) { |
||
| 114 | fftw_real tmp29; |
||
| 115 | fftw_real tmp65; |
||
| 116 | fftw_real tmp92; |
||
| 117 | fftw_real tmp97; |
||
| 118 | fftw_real tmp63; |
||
| 119 | fftw_real tmp75; |
||
| 120 | fftw_real tmp78; |
||
| 121 | fftw_real tmp87; |
||
| 122 | fftw_real tmp40; |
||
| 123 | fftw_real tmp98; |
||
| 124 | fftw_real tmp68; |
||
| 125 | fftw_real tmp89; |
||
| 126 | fftw_real tmp52; |
||
| 127 | fftw_real tmp70; |
||
| 128 | fftw_real tmp73; |
||
| 129 | fftw_real tmp86; |
||
| 130 | ASSERT_ALIGNED_DOUBLE(); |
||
| 131 | { |
||
| 132 | fftw_real tmp23; |
||
| 133 | fftw_real tmp91; |
||
| 134 | fftw_real tmp28; |
||
| 135 | fftw_real tmp90; |
||
| 136 | ASSERT_ALIGNED_DOUBLE(); |
||
| 137 | tmp23 = X[0]; |
||
| 138 | tmp91 = Y[-7 * iostride]; |
||
| 139 | { |
||
| 140 | fftw_real tmp25; |
||
| 141 | fftw_real tmp27; |
||
| 142 | fftw_real tmp24; |
||
| 143 | fftw_real tmp26; |
||
| 144 | ASSERT_ALIGNED_DOUBLE(); |
||
| 145 | tmp25 = X[4 * iostride]; |
||
| 146 | tmp27 = Y[-3 * iostride]; |
||
| 147 | tmp24 = c_re(W[3]); |
||
| 148 | tmp26 = c_im(W[3]); |
||
| 149 | tmp28 = (tmp24 * tmp25) - (tmp26 * tmp27); |
||
| 150 | tmp90 = (tmp26 * tmp25) + (tmp24 * tmp27); |
||
| 151 | } |
||
| 152 | tmp29 = tmp23 + tmp28; |
||
| 153 | tmp65 = tmp23 - tmp28; |
||
| 154 | tmp92 = tmp90 + tmp91; |
||
| 155 | tmp97 = tmp91 - tmp90; |
||
| 156 | } |
||
| 157 | { |
||
| 158 | fftw_real tmp57; |
||
| 159 | fftw_real tmp76; |
||
| 160 | fftw_real tmp62; |
||
| 161 | fftw_real tmp77; |
||
| 162 | ASSERT_ALIGNED_DOUBLE(); |
||
| 163 | { |
||
| 164 | fftw_real tmp54; |
||
| 165 | fftw_real tmp56; |
||
| 166 | fftw_real tmp53; |
||
| 167 | fftw_real tmp55; |
||
| 168 | ASSERT_ALIGNED_DOUBLE(); |
||
| 169 | tmp54 = X[7 * iostride]; |
||
| 170 | tmp56 = Y[0]; |
||
| 171 | tmp53 = c_re(W[6]); |
||
| 172 | tmp55 = c_im(W[6]); |
||
| 173 | tmp57 = (tmp53 * tmp54) - (tmp55 * tmp56); |
||
| 174 | tmp76 = (tmp55 * tmp54) + (tmp53 * tmp56); |
||
| 175 | } |
||
| 176 | { |
||
| 177 | fftw_real tmp59; |
||
| 178 | fftw_real tmp61; |
||
| 179 | fftw_real tmp58; |
||
| 180 | fftw_real tmp60; |
||
| 181 | ASSERT_ALIGNED_DOUBLE(); |
||
| 182 | tmp59 = X[3 * iostride]; |
||
| 183 | tmp61 = Y[-4 * iostride]; |
||
| 184 | tmp58 = c_re(W[2]); |
||
| 185 | tmp60 = c_im(W[2]); |
||
| 186 | tmp62 = (tmp58 * tmp59) - (tmp60 * tmp61); |
||
| 187 | tmp77 = (tmp60 * tmp59) + (tmp58 * tmp61); |
||
| 188 | } |
||
| 189 | tmp63 = tmp57 + tmp62; |
||
| 190 | tmp75 = tmp57 - tmp62; |
||
| 191 | tmp78 = tmp76 - tmp77; |
||
| 192 | tmp87 = tmp76 + tmp77; |
||
| 193 | } |
||
| 194 | { |
||
| 195 | fftw_real tmp34; |
||
| 196 | fftw_real tmp66; |
||
| 197 | fftw_real tmp39; |
||
| 198 | fftw_real tmp67; |
||
| 199 | ASSERT_ALIGNED_DOUBLE(); |
||
| 200 | { |
||
| 201 | fftw_real tmp31; |
||
| 202 | fftw_real tmp33; |
||
| 203 | fftw_real tmp30; |
||
| 204 | fftw_real tmp32; |
||
| 205 | ASSERT_ALIGNED_DOUBLE(); |
||
| 206 | tmp31 = X[2 * iostride]; |
||
| 207 | tmp33 = Y[-5 * iostride]; |
||
| 208 | tmp30 = c_re(W[1]); |
||
| 209 | tmp32 = c_im(W[1]); |
||
| 210 | tmp34 = (tmp30 * tmp31) - (tmp32 * tmp33); |
||
| 211 | tmp66 = (tmp32 * tmp31) + (tmp30 * tmp33); |
||
| 212 | } |
||
| 213 | { |
||
| 214 | fftw_real tmp36; |
||
| 215 | fftw_real tmp38; |
||
| 216 | fftw_real tmp35; |
||
| 217 | fftw_real tmp37; |
||
| 218 | ASSERT_ALIGNED_DOUBLE(); |
||
| 219 | tmp36 = X[6 * iostride]; |
||
| 220 | tmp38 = Y[-iostride]; |
||
| 221 | tmp35 = c_re(W[5]); |
||
| 222 | tmp37 = c_im(W[5]); |
||
| 223 | tmp39 = (tmp35 * tmp36) - (tmp37 * tmp38); |
||
| 224 | tmp67 = (tmp37 * tmp36) + (tmp35 * tmp38); |
||
| 225 | } |
||
| 226 | tmp40 = tmp34 + tmp39; |
||
| 227 | tmp98 = tmp34 - tmp39; |
||
| 228 | tmp68 = tmp66 - tmp67; |
||
| 229 | tmp89 = tmp66 + tmp67; |
||
| 230 | } |
||
| 231 | { |
||
| 232 | fftw_real tmp46; |
||
| 233 | fftw_real tmp71; |
||
| 234 | fftw_real tmp51; |
||
| 235 | fftw_real tmp72; |
||
| 236 | ASSERT_ALIGNED_DOUBLE(); |
||
| 237 | { |
||
| 238 | fftw_real tmp43; |
||
| 239 | fftw_real tmp45; |
||
| 240 | fftw_real tmp42; |
||
| 241 | fftw_real tmp44; |
||
| 242 | ASSERT_ALIGNED_DOUBLE(); |
||
| 243 | tmp43 = X[iostride]; |
||
| 244 | tmp45 = Y[-6 * iostride]; |
||
| 245 | tmp42 = c_re(W[0]); |
||
| 246 | tmp44 = c_im(W[0]); |
||
| 247 | tmp46 = (tmp42 * tmp43) - (tmp44 * tmp45); |
||
| 248 | tmp71 = (tmp44 * tmp43) + (tmp42 * tmp45); |
||
| 249 | } |
||
| 250 | { |
||
| 251 | fftw_real tmp48; |
||
| 252 | fftw_real tmp50; |
||
| 253 | fftw_real tmp47; |
||
| 254 | fftw_real tmp49; |
||
| 255 | ASSERT_ALIGNED_DOUBLE(); |
||
| 256 | tmp48 = X[5 * iostride]; |
||
| 257 | tmp50 = Y[-2 * iostride]; |
||
| 258 | tmp47 = c_re(W[4]); |
||
| 259 | tmp49 = c_im(W[4]); |
||
| 260 | tmp51 = (tmp47 * tmp48) - (tmp49 * tmp50); |
||
| 261 | tmp72 = (tmp49 * tmp48) + (tmp47 * tmp50); |
||
| 262 | } |
||
| 263 | tmp52 = tmp46 + tmp51; |
||
| 264 | tmp70 = tmp46 - tmp51; |
||
| 265 | tmp73 = tmp71 - tmp72; |
||
| 266 | tmp86 = tmp71 + tmp72; |
||
| 267 | } |
||
| 268 | { |
||
| 269 | fftw_real tmp41; |
||
| 270 | fftw_real tmp64; |
||
| 271 | fftw_real tmp85; |
||
| 272 | fftw_real tmp88; |
||
| 273 | ASSERT_ALIGNED_DOUBLE(); |
||
| 274 | tmp41 = tmp29 + tmp40; |
||
| 275 | tmp64 = tmp52 + tmp63; |
||
| 276 | Y[-4 * iostride] = tmp41 - tmp64; |
||
| 277 | X[0] = tmp41 + tmp64; |
||
| 278 | { |
||
| 279 | fftw_real tmp95; |
||
| 280 | fftw_real tmp96; |
||
| 281 | fftw_real tmp93; |
||
| 282 | fftw_real tmp94; |
||
| 283 | ASSERT_ALIGNED_DOUBLE(); |
||
| 284 | tmp95 = tmp92 - tmp89; |
||
| 285 | tmp96 = tmp63 - tmp52; |
||
| 286 | X[6 * iostride] = -(tmp95 - tmp96); |
||
| 287 | Y[-2 * iostride] = tmp96 + tmp95; |
||
| 288 | tmp93 = tmp89 + tmp92; |
||
| 289 | tmp94 = tmp86 + tmp87; |
||
| 290 | X[4 * iostride] = -(tmp93 - tmp94); |
||
| 291 | Y[0] = tmp94 + tmp93; |
||
| 292 | } |
||
| 293 | tmp85 = tmp29 - tmp40; |
||
| 294 | tmp88 = tmp86 - tmp87; |
||
| 295 | Y[-6 * iostride] = tmp85 - tmp88; |
||
| 296 | X[2 * iostride] = tmp85 + tmp88; |
||
| 297 | { |
||
| 298 | fftw_real tmp81; |
||
| 299 | fftw_real tmp99; |
||
| 300 | fftw_real tmp84; |
||
| 301 | fftw_real tmp100; |
||
| 302 | fftw_real tmp82; |
||
| 303 | fftw_real tmp83; |
||
| 304 | ASSERT_ALIGNED_DOUBLE(); |
||
| 305 | tmp81 = tmp65 - tmp68; |
||
| 306 | tmp99 = tmp97 - tmp98; |
||
| 307 | tmp82 = tmp73 - tmp70; |
||
| 308 | tmp83 = tmp75 + tmp78; |
||
| 309 | tmp84 = K707106781 * (tmp82 - tmp83); |
||
| 310 | tmp100 = K707106781 * (tmp82 + tmp83); |
||
| 311 | Y[-7 * iostride] = tmp81 - tmp84; |
||
| 312 | X[3 * iostride] = tmp81 + tmp84; |
||
| 313 | X[5 * iostride] = -(tmp99 - tmp100); |
||
| 314 | Y[-iostride] = tmp100 + tmp99; |
||
| 315 | } |
||
| 316 | { |
||
| 317 | fftw_real tmp69; |
||
| 318 | fftw_real tmp101; |
||
| 319 | fftw_real tmp80; |
||
| 320 | fftw_real tmp102; |
||
| 321 | fftw_real tmp74; |
||
| 322 | fftw_real tmp79; |
||
| 323 | ASSERT_ALIGNED_DOUBLE(); |
||
| 324 | tmp69 = tmp65 + tmp68; |
||
| 325 | tmp101 = tmp98 + tmp97; |
||
| 326 | tmp74 = tmp70 + tmp73; |
||
| 327 | tmp79 = tmp75 - tmp78; |
||
| 328 | tmp80 = K707106781 * (tmp74 + tmp79); |
||
| 329 | tmp102 = K707106781 * (tmp79 - tmp74); |
||
| 330 | Y[-5 * iostride] = tmp69 - tmp80; |
||
| 331 | X[iostride] = tmp69 + tmp80; |
||
| 332 | X[7 * iostride] = -(tmp101 - tmp102); |
||
| 333 | Y[-3 * iostride] = tmp102 + tmp101; |
||
| 334 | } |
||
| 335 | } |
||
| 336 | } |
||
| 337 | if (i == m) { |
||
| 338 | fftw_real tmp1; |
||
| 339 | fftw_real tmp19; |
||
| 340 | fftw_real tmp4; |
||
| 341 | fftw_real tmp18; |
||
| 342 | fftw_real tmp8; |
||
| 343 | fftw_real tmp14; |
||
| 344 | fftw_real tmp11; |
||
| 345 | fftw_real tmp15; |
||
| 346 | fftw_real tmp2; |
||
| 347 | fftw_real tmp3; |
||
| 348 | ASSERT_ALIGNED_DOUBLE(); |
||
| 349 | tmp1 = X[0]; |
||
| 350 | tmp19 = X[4 * iostride]; |
||
| 351 | tmp2 = X[2 * iostride]; |
||
| 352 | tmp3 = X[6 * iostride]; |
||
| 353 | tmp4 = K707106781 * (tmp2 - tmp3); |
||
| 354 | tmp18 = K707106781 * (tmp2 + tmp3); |
||
| 355 | { |
||
| 356 | fftw_real tmp6; |
||
| 357 | fftw_real tmp7; |
||
| 358 | fftw_real tmp9; |
||
| 359 | fftw_real tmp10; |
||
| 360 | ASSERT_ALIGNED_DOUBLE(); |
||
| 361 | tmp6 = X[iostride]; |
||
| 362 | tmp7 = X[5 * iostride]; |
||
| 363 | tmp8 = (K923879532 * tmp6) - (K382683432 * tmp7); |
||
| 364 | tmp14 = (K382683432 * tmp6) + (K923879532 * tmp7); |
||
| 365 | tmp9 = X[3 * iostride]; |
||
| 366 | tmp10 = X[7 * iostride]; |
||
| 367 | tmp11 = (K382683432 * tmp9) - (K923879532 * tmp10); |
||
| 368 | tmp15 = (K923879532 * tmp9) + (K382683432 * tmp10); |
||
| 369 | } |
||
| 370 | { |
||
| 371 | fftw_real tmp5; |
||
| 372 | fftw_real tmp12; |
||
| 373 | fftw_real tmp21; |
||
| 374 | fftw_real tmp22; |
||
| 375 | ASSERT_ALIGNED_DOUBLE(); |
||
| 376 | tmp5 = tmp1 + tmp4; |
||
| 377 | tmp12 = tmp8 + tmp11; |
||
| 378 | X[3 * iostride] = tmp5 - tmp12; |
||
| 379 | X[0] = tmp5 + tmp12; |
||
| 380 | tmp21 = tmp11 - tmp8; |
||
| 381 | tmp22 = tmp19 - tmp18; |
||
| 382 | Y[-2 * iostride] = tmp21 - tmp22; |
||
| 383 | Y[-iostride] = tmp21 + tmp22; |
||
| 384 | } |
||
| 385 | { |
||
| 386 | fftw_real tmp17; |
||
| 387 | fftw_real tmp20; |
||
| 388 | fftw_real tmp13; |
||
| 389 | fftw_real tmp16; |
||
| 390 | ASSERT_ALIGNED_DOUBLE(); |
||
| 391 | tmp17 = tmp14 + tmp15; |
||
| 392 | tmp20 = tmp18 + tmp19; |
||
| 393 | Y[0] = -(tmp17 + tmp20); |
||
| 394 | Y[-3 * iostride] = tmp20 - tmp17; |
||
| 395 | tmp13 = tmp1 - tmp4; |
||
| 396 | tmp16 = tmp14 - tmp15; |
||
| 397 | X[2 * iostride] = tmp13 - tmp16; |
||
| 398 | X[iostride] = tmp13 + tmp16; |
||
| 399 | } |
||
| 400 | } |
||
| 401 | } |
||
| 402 | |||
| 403 | static const int twiddle_order[] = |
||
| 404 | {1, 2, 3, 4, 5, 6, 7}; |
||
| 405 | fftw_codelet_desc fftw_hc2hc_forward_8_desc = |
||
| 406 | { |
||
| 407 | "fftw_hc2hc_forward_8", |
||
| 408 | (void (*)()) fftw_hc2hc_forward_8, |
||
| 409 | 8, |
||
| 410 | FFTW_FORWARD, |
||
| 411 | FFTW_HC2HC, |
||
| 412 | 179, |
||
| 413 | 7, |
||
| 414 | twiddle_order, |
||
| 415 | }; |