|
3828
|
1 /* |
|
|
2 |
|
7017
|
3 Copyright (C) 2001, 2002, 2004, 2005, 2006, 2007 John W. Eaton |
|
|
4 |
|
3828
|
5 This file is part of Octave. |
|
|
6 |
|
|
7 Octave is free software; you can redistribute it and/or modify it |
|
|
8 under the terms of the GNU General Public License as published by the |
|
7016
|
9 Free Software Foundation; either version 3 of the License, or (at your |
|
|
10 option) any later version. |
|
3828
|
11 |
|
|
12 Octave is distributed in the hope that it will be useful, but WITHOUT |
|
|
13 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
|
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
|
15 for more details. |
|
|
16 |
|
|
17 You should have received a copy of the GNU General Public License |
|
7016
|
18 along with Octave; see the file COPYING. If not, see |
|
|
19 <http://www.gnu.org/licenses/>. |
|
3828
|
20 |
|
|
21 */ |
|
|
22 |
|
|
23 #ifdef HAVE_CONFIG_H |
|
|
24 #include <config.h> |
|
|
25 #endif |
|
|
26 |
|
4773
|
27 #if defined (HAVE_FFTW3) |
|
3828
|
28 |
|
4775
|
29 #include <iostream> |
|
|
30 #include <vector> |
|
|
31 |
|
4786
|
32 #include "lo-error.h" |
|
3828
|
33 #include "oct-fftw.h" |
|
4786
|
34 #include "quit.h" |
|
3828
|
35 |
|
4809
|
36 // Helper class to create and cache fftw plans for both 1d and |
|
6228
|
37 // 2d. This implementation defaults to using FFTW_ESTIMATE to create |
|
|
38 // the plans, which in theory is suboptimal, but provides quit |
|
|
39 // reasonable performance. |
|
4773
|
40 |
|
|
41 // Also note that if FFTW_ESTIMATE is not used the planner in FFTW3 |
|
6228
|
42 // destroys the input and output arrays. We must therefore create a |
|
|
43 // temporary input array with the same size and 16-byte alignment as |
|
|
44 // the original array and use that for the planner. Note that we also |
|
|
45 // use any wisdom that is available, either in a FFTW3 system wide file |
|
|
46 // or as supplied by the user. |
|
4773
|
47 |
|
5775
|
48 // FIXME -- if we can ensure 16 byte alignment in Array<T> |
|
4809
|
49 // (<T> *data) the FFTW3 can use SIMD instructions for further |
|
|
50 // acceleration. |
|
4773
|
51 |
|
4809
|
52 // Note that it is profitable to store the FFTW3 plans, for small |
|
|
53 // ffts. |
|
3828
|
54 |
|
4808
|
55 octave_fftw_planner::octave_fftw_planner (void) |
|
3828
|
56 { |
|
6228
|
57 meth = ESTIMATE; |
|
3828
|
58 |
|
|
59 plan[0] = plan[1] = 0; |
|
4773
|
60 d[0] = d[1] = s[0] = s[1] = r[0] = r[1] = h[0] = h[1] = 0; |
|
4808
|
61 simd_align[0] = simd_align[1] = false; |
|
5044
|
62 inplace[0] = inplace[1] = false; |
|
4808
|
63 n[0] = n[1] = dim_vector (); |
|
4773
|
64 |
|
|
65 rplan = 0; |
|
|
66 rd = rs = rr = rh = 0; |
|
4808
|
67 rsimd_align = false; |
|
4773
|
68 rn = dim_vector (); |
|
5044
|
69 |
|
4809
|
70 // If we have a system wide wisdom file, import it. |
|
4808
|
71 fftw_import_system_wisdom (); |
|
3828
|
72 } |
|
|
73 |
|
6228
|
74 octave_fftw_planner::FftwMethod |
|
|
75 octave_fftw_planner::method (void) |
|
|
76 { |
|
|
77 return meth; |
|
|
78 } |
|
|
79 |
|
|
80 octave_fftw_planner::FftwMethod |
|
|
81 octave_fftw_planner::method (FftwMethod _meth) |
|
|
82 { |
|
|
83 FftwMethod ret = meth; |
|
|
84 if (_meth == ESTIMATE || _meth == MEASURE || |
|
|
85 _meth == PATIENT || _meth == EXHAUSTIVE || |
|
|
86 _meth == HYBRID) |
|
|
87 { |
|
|
88 if (meth != _meth) |
|
|
89 { |
|
|
90 meth = _meth; |
|
|
91 if (rplan) |
|
|
92 fftw_destroy_plan (rplan); |
|
|
93 if (plan[0]) |
|
|
94 fftw_destroy_plan (plan[0]); |
|
|
95 if (plan[1]) |
|
|
96 fftw_destroy_plan (plan[1]); |
|
|
97 rplan = plan[0] = plan[1] = 0; |
|
|
98 } |
|
|
99 } |
|
|
100 else |
|
|
101 ret = UNKNOWN; |
|
|
102 return ret; |
|
|
103 } |
|
|
104 |
|
4808
|
105 #define CHECK_SIMD_ALIGNMENT(x) \ |
|
6482
|
106 (((reinterpret_cast<ptrdiff_t> (x)) & 0xF) == 0) |
|
4808
|
107 |
|
3828
|
108 fftw_plan |
|
4773
|
109 octave_fftw_planner::create_plan (int dir, const int rank, |
|
5275
|
110 const dim_vector dims, octave_idx_type howmany, |
|
|
111 octave_idx_type stride, octave_idx_type dist, |
|
4773
|
112 const Complex *in, Complex *out) |
|
3828
|
113 { |
|
4773
|
114 int which = (dir == FFTW_FORWARD) ? 0 : 1; |
|
3828
|
115 fftw_plan *cur_plan_p = &plan[which]; |
|
|
116 bool create_new_plan = false; |
|
4808
|
117 bool ioalign = CHECK_SIMD_ALIGNMENT (in) && CHECK_SIMD_ALIGNMENT (out); |
|
5044
|
118 bool ioinplace = (in == out); |
|
3828
|
119 |
|
4809
|
120 // Don't create a new plan if we have a non SIMD plan already but |
|
|
121 // can do SIMD. This prevents endlessly recreating plans if we |
|
|
122 // change the alignment. |
|
|
123 |
|
4783
|
124 if (plan[which] == 0 || d[which] != dist || s[which] != stride |
|
5044
|
125 || r[which] != rank || h[which] != howmany |
|
|
126 || ioinplace != inplace[which] |
|
4808
|
127 || ((ioalign != simd_align[which]) ? !ioalign : false)) |
|
4773
|
128 create_new_plan = true; |
|
|
129 else |
|
4809
|
130 { |
|
|
131 // We still might not have the same shape of array. |
|
|
132 |
|
|
133 for (int i = 0; i < rank; i++) |
|
|
134 if (dims(i) != n[which](i)) |
|
|
135 { |
|
|
136 create_new_plan = true; |
|
|
137 break; |
|
|
138 } |
|
|
139 } |
|
3828
|
140 |
|
|
141 if (create_new_plan) |
|
|
142 { |
|
4773
|
143 d[which] = dist; |
|
|
144 s[which] = stride; |
|
|
145 r[which] = rank; |
|
|
146 h[which] = howmany; |
|
4808
|
147 simd_align[which] = ioalign; |
|
5044
|
148 inplace[which] = ioinplace; |
|
4773
|
149 n[which] = dims; |
|
|
150 |
|
6228
|
151 // Note reversal of dimensions for column major storage in FFTW. |
|
|
152 octave_idx_type nn = 1; |
|
|
153 OCTAVE_LOCAL_BUFFER (int, tmp, rank); |
|
|
154 |
|
|
155 for (int i = 0, j = rank-1; i < rank; i++, j--) |
|
|
156 { |
|
|
157 tmp[i] = dims(j); |
|
|
158 nn *= dims(j); |
|
|
159 } |
|
|
160 |
|
|
161 int plan_flags = 0; |
|
|
162 bool plan_destroys_in = true; |
|
|
163 |
|
|
164 switch (meth) |
|
|
165 { |
|
|
166 case UNKNOWN: |
|
|
167 case ESTIMATE: |
|
|
168 plan_flags |= FFTW_ESTIMATE; |
|
|
169 plan_destroys_in = false; |
|
|
170 break; |
|
|
171 case MEASURE: |
|
|
172 plan_flags |= FFTW_MEASURE; |
|
|
173 break; |
|
|
174 case PATIENT: |
|
|
175 plan_flags |= FFTW_PATIENT; |
|
|
176 break; |
|
|
177 case EXHAUSTIVE: |
|
|
178 plan_flags |= FFTW_EXHAUSTIVE; |
|
|
179 break; |
|
|
180 case HYBRID: |
|
|
181 if (nn < 8193) |
|
|
182 plan_flags |= FFTW_MEASURE; |
|
|
183 else |
|
|
184 { |
|
|
185 plan_flags |= FFTW_ESTIMATE; |
|
|
186 plan_destroys_in = false; |
|
|
187 } |
|
|
188 break; |
|
|
189 } |
|
|
190 |
|
4808
|
191 if (ioalign) |
|
|
192 plan_flags &= ~FFTW_UNALIGNED; |
|
|
193 else |
|
|
194 plan_flags |= FFTW_UNALIGNED; |
|
|
195 |
|
3828
|
196 if (*cur_plan_p) |
|
|
197 fftw_destroy_plan (*cur_plan_p); |
|
|
198 |
|
6228
|
199 if (plan_destroys_in) |
|
|
200 { |
|
|
201 // Create matrix with the same size and 16-byte alignment as input |
|
|
202 OCTAVE_LOCAL_BUFFER (Complex, itmp, nn * howmany + 32); |
|
|
203 itmp = reinterpret_cast<Complex *> |
|
|
204 (((reinterpret_cast<ptrdiff_t>(itmp) + 15) & ~ 0xF) + |
|
|
205 ((reinterpret_cast<ptrdiff_t> (in)) & 0xF)); |
|
4809
|
206 |
|
6228
|
207 *cur_plan_p = |
|
|
208 fftw_plan_many_dft (rank, tmp, howmany, |
|
|
209 reinterpret_cast<fftw_complex *> (itmp), |
|
|
210 0, stride, dist, reinterpret_cast<fftw_complex *> (out), |
|
|
211 0, stride, dist, dir, plan_flags); |
|
|
212 } |
|
|
213 else |
|
|
214 { |
|
|
215 *cur_plan_p = |
|
|
216 fftw_plan_many_dft (rank, tmp, howmany, |
|
4773
|
217 reinterpret_cast<fftw_complex *> (const_cast<Complex *> (in)), |
|
4774
|
218 0, stride, dist, reinterpret_cast<fftw_complex *> (out), |
|
|
219 0, stride, dist, dir, plan_flags); |
|
6228
|
220 } |
|
3828
|
221 |
|
|
222 if (*cur_plan_p == 0) |
|
|
223 (*current_liboctave_error_handler) ("Error creating fftw plan"); |
|
|
224 } |
|
|
225 |
|
|
226 return *cur_plan_p; |
|
|
227 } |
|
|
228 |
|
4773
|
229 fftw_plan |
|
|
230 octave_fftw_planner::create_plan (const int rank, const dim_vector dims, |
|
5275
|
231 octave_idx_type howmany, octave_idx_type stride, octave_idx_type dist, |
|
4773
|
232 const double *in, Complex *out) |
|
3828
|
233 { |
|
4773
|
234 fftw_plan *cur_plan_p = &rplan; |
|
3828
|
235 bool create_new_plan = false; |
|
4808
|
236 bool ioalign = CHECK_SIMD_ALIGNMENT (in) && CHECK_SIMD_ALIGNMENT (out); |
|
3828
|
237 |
|
4809
|
238 // Don't create a new plan if we have a non SIMD plan already but |
|
|
239 // can do SIMD. This prevents endlessly recreating plans if we |
|
|
240 // change the alignment. |
|
|
241 |
|
4783
|
242 if (rplan == 0 || rd != dist || rs != stride || rr != rank |
|
4808
|
243 || rh != howmany || ((ioalign != rsimd_align) ? !ioalign : false)) |
|
4773
|
244 create_new_plan = true; |
|
|
245 else |
|
4809
|
246 { |
|
|
247 // We still might not have the same shape of array. |
|
|
248 |
|
|
249 for (int i = 0; i < rank; i++) |
|
|
250 if (dims(i) != rn(i)) |
|
|
251 { |
|
|
252 create_new_plan = true; |
|
|
253 break; |
|
|
254 } |
|
|
255 } |
|
3828
|
256 |
|
|
257 if (create_new_plan) |
|
|
258 { |
|
4773
|
259 rd = dist; |
|
|
260 rs = stride; |
|
|
261 rr = rank; |
|
|
262 rh = howmany; |
|
4808
|
263 rsimd_align = ioalign; |
|
4773
|
264 rn = dims; |
|
|
265 |
|
6228
|
266 // Note reversal of dimensions for column major storage in FFTW. |
|
|
267 octave_idx_type nn = 1; |
|
|
268 OCTAVE_LOCAL_BUFFER (int, tmp, rank); |
|
|
269 |
|
|
270 for (int i = 0, j = rank-1; i < rank; i++, j--) |
|
|
271 { |
|
|
272 tmp[i] = dims(j); |
|
|
273 nn *= dims(j); |
|
|
274 } |
|
|
275 |
|
|
276 int plan_flags = 0; |
|
|
277 bool plan_destroys_in = true; |
|
|
278 |
|
|
279 switch (meth) |
|
|
280 { |
|
|
281 case UNKNOWN: |
|
|
282 case ESTIMATE: |
|
|
283 plan_flags |= FFTW_ESTIMATE; |
|
|
284 plan_destroys_in = false; |
|
|
285 break; |
|
|
286 case MEASURE: |
|
|
287 plan_flags |= FFTW_MEASURE; |
|
|
288 break; |
|
|
289 case PATIENT: |
|
|
290 plan_flags |= FFTW_PATIENT; |
|
|
291 break; |
|
|
292 case EXHAUSTIVE: |
|
|
293 plan_flags |= FFTW_EXHAUSTIVE; |
|
|
294 break; |
|
|
295 case HYBRID: |
|
|
296 if (nn < 8193) |
|
|
297 plan_flags |= FFTW_MEASURE; |
|
|
298 else |
|
|
299 { |
|
|
300 plan_flags |= FFTW_ESTIMATE; |
|
|
301 plan_destroys_in = false; |
|
|
302 } |
|
|
303 break; |
|
|
304 } |
|
|
305 |
|
4808
|
306 if (ioalign) |
|
|
307 plan_flags &= ~FFTW_UNALIGNED; |
|
|
308 else |
|
|
309 plan_flags |= FFTW_UNALIGNED; |
|
|
310 |
|
3828
|
311 if (*cur_plan_p) |
|
4773
|
312 fftw_destroy_plan (*cur_plan_p); |
|
3828
|
313 |
|
6228
|
314 if (plan_destroys_in) |
|
|
315 { |
|
|
316 // Create matrix with the same size and 16-byte alignment as input |
|
|
317 OCTAVE_LOCAL_BUFFER (double, itmp, nn + 32); |
|
|
318 itmp = reinterpret_cast<double *> |
|
|
319 (((reinterpret_cast<ptrdiff_t>(itmp) + 15) & ~ 0xF) + |
|
|
320 ((reinterpret_cast<ptrdiff_t> (in)) & 0xF)); |
|
4809
|
321 |
|
6228
|
322 *cur_plan_p = |
|
|
323 fftw_plan_many_dft_r2c (rank, tmp, howmany, itmp, |
|
|
324 0, stride, dist, reinterpret_cast<fftw_complex *> (out), |
|
|
325 0, stride, dist, plan_flags); |
|
|
326 } |
|
|
327 else |
|
|
328 { |
|
|
329 *cur_plan_p = |
|
|
330 fftw_plan_many_dft_r2c (rank, tmp, howmany, |
|
4773
|
331 (const_cast<double *> (in)), |
|
4774
|
332 0, stride, dist, reinterpret_cast<fftw_complex *> (out), |
|
|
333 0, stride, dist, plan_flags); |
|
6228
|
334 } |
|
3828
|
335 |
|
|
336 if (*cur_plan_p == 0) |
|
4773
|
337 (*current_liboctave_error_handler) ("Error creating fftw plan"); |
|
3828
|
338 } |
|
|
339 |
|
|
340 return *cur_plan_p; |
|
|
341 } |
|
|
342 |
|
6228
|
343 octave_fftw_planner fftw_planner; |
|
3828
|
344 |
|
4775
|
345 static inline void |
|
|
346 convert_packcomplex_1d (Complex *out, size_t nr, size_t nc, |
|
5275
|
347 octave_idx_type stride, octave_idx_type dist) |
|
4773
|
348 { |
|
4785
|
349 OCTAVE_QUIT; |
|
|
350 |
|
|
351 // Fill in the missing data. |
|
|
352 |
|
4773
|
353 for (size_t i = 0; i < nr; i++) |
|
|
354 for (size_t j = nc/2+1; j < nc; j++) |
|
|
355 out[j*stride + i*dist] = conj(out[(nc - j)*stride + i*dist]); |
|
4785
|
356 |
|
|
357 OCTAVE_QUIT; |
|
4773
|
358 } |
|
|
359 |
|
4775
|
360 static inline void |
|
|
361 convert_packcomplex_Nd (Complex *out, const dim_vector &dv) |
|
3828
|
362 { |
|
4773
|
363 size_t nc = dv(0); |
|
|
364 size_t nr = dv(1); |
|
4808
|
365 size_t np = (dv.length () > 2 ? dv.numel () / nc / nr : 1); |
|
4773
|
366 size_t nrp = nr * np; |
|
|
367 Complex *ptr1, *ptr2; |
|
|
368 |
|
4785
|
369 OCTAVE_QUIT; |
|
|
370 |
|
|
371 // Create space for the missing elements. |
|
|
372 |
|
4773
|
373 for (size_t i = 0; i < nrp; i++) |
|
|
374 { |
|
|
375 ptr1 = out + i * (nc/2 + 1) + nrp*((nc-1)/2); |
|
|
376 ptr2 = out + i * nc; |
|
|
377 for (size_t j = 0; j < nc/2+1; j++) |
|
|
378 *ptr2++ = *ptr1++; |
|
|
379 } |
|
|
380 |
|
4785
|
381 OCTAVE_QUIT; |
|
|
382 |
|
|
383 // Fill in the missing data for the rank = 2 case directly for speed. |
|
|
384 |
|
4773
|
385 for (size_t i = 0; i < np; i++) |
|
|
386 { |
|
|
387 for (size_t j = 1; j < nr; j++) |
|
|
388 for (size_t k = nc/2+1; k < nc; k++) |
|
|
389 out[k + (j + i*nr)*nc] = conj(out[nc - k + ((i+1)*nr - j)*nc]); |
|
|
390 |
|
|
391 for (size_t j = nc/2+1; j < nc; j++) |
|
|
392 out[j + i*nr*nc] = conj(out[(i*nr+1)*nc - j]); |
|
|
393 } |
|
|
394 |
|
4785
|
395 OCTAVE_QUIT; |
|
|
396 |
|
|
397 // Now do the permutations needed for rank > 2 cases. |
|
|
398 |
|
4773
|
399 size_t jstart = dv(0) * dv(1); |
|
|
400 size_t kstep = dv(0); |
|
|
401 size_t nel = dv.numel (); |
|
4785
|
402 |
|
4808
|
403 for (int inner = 2; inner < dv.length (); inner++) |
|
4773
|
404 { |
|
|
405 size_t jmax = jstart * dv(inner); |
|
|
406 for (size_t i = 0; i < nel; i+=jmax) |
|
|
407 for (size_t j = jstart, jj = jmax-jstart; j < jj; |
|
|
408 j+=jstart, jj-=jstart) |
|
|
409 for (size_t k = 0; k < jstart; k+= kstep) |
|
|
410 for (size_t l = nc/2+1; l < nc; l++) |
|
|
411 { |
|
|
412 Complex tmp = out[i+ j + k + l]; |
|
|
413 out[i + j + k + l] = out[i + jj + k + l]; |
|
|
414 out[i + jj + k + l] = tmp; |
|
|
415 } |
|
|
416 jstart = jmax; |
|
|
417 } |
|
4785
|
418 |
|
|
419 OCTAVE_QUIT; |
|
4773
|
420 } |
|
|
421 |
|
|
422 int |
|
|
423 octave_fftw::fft (const double *in, Complex *out, size_t npts, |
|
5275
|
424 size_t nsamples, octave_idx_type stride, octave_idx_type dist) |
|
4773
|
425 { |
|
|
426 dist = (dist < 0 ? npts : dist); |
|
|
427 |
|
|
428 dim_vector dv (npts); |
|
|
429 fftw_plan plan = fftw_planner.create_plan (1, dv, nsamples, stride, dist, |
|
|
430 in, out); |
|
|
431 |
|
|
432 fftw_execute_dft_r2c (plan, (const_cast<double *>(in)), |
|
|
433 reinterpret_cast<fftw_complex *> (out)); |
|
|
434 |
|
4809
|
435 // Need to create other half of the transform. |
|
|
436 |
|
4773
|
437 convert_packcomplex_1d (out, nsamples, npts, stride, dist); |
|
3828
|
438 |
|
|
439 return 0; |
|
|
440 } |
|
|
441 |
|
|
442 int |
|
4773
|
443 octave_fftw::fft (const Complex *in, Complex *out, size_t npts, |
|
5275
|
444 size_t nsamples, octave_idx_type stride, octave_idx_type dist) |
|
3828
|
445 { |
|
4773
|
446 dist = (dist < 0 ? npts : dist); |
|
|
447 |
|
|
448 dim_vector dv (npts); |
|
|
449 fftw_plan plan = fftw_planner.create_plan (FFTW_FORWARD, 1, dv, nsamples, |
|
|
450 stride, dist, in, out); |
|
|
451 |
|
|
452 fftw_execute_dft (plan, |
|
|
453 reinterpret_cast<fftw_complex *> (const_cast<Complex *>(in)), |
|
|
454 reinterpret_cast<fftw_complex *> (out)); |
|
|
455 |
|
|
456 return 0; |
|
|
457 } |
|
|
458 |
|
|
459 int |
|
|
460 octave_fftw::ifft (const Complex *in, Complex *out, size_t npts, |
|
5275
|
461 size_t nsamples, octave_idx_type stride, octave_idx_type dist) |
|
4773
|
462 { |
|
|
463 dist = (dist < 0 ? npts : dist); |
|
|
464 |
|
|
465 dim_vector dv (npts); |
|
|
466 fftw_plan plan = fftw_planner.create_plan (FFTW_BACKWARD, 1, dv, nsamples, |
|
|
467 stride, dist, in, out); |
|
|
468 |
|
|
469 fftw_execute_dft (plan, |
|
|
470 reinterpret_cast<fftw_complex *> (const_cast<Complex *>(in)), |
|
|
471 reinterpret_cast<fftw_complex *> (out)); |
|
3828
|
472 |
|
|
473 const Complex scale = npts; |
|
4773
|
474 for (size_t j = 0; j < nsamples; j++) |
|
|
475 for (size_t i = 0; i < npts; i++) |
|
|
476 out[i*stride + j*dist] /= scale; |
|
3828
|
477 |
|
|
478 return 0; |
|
|
479 } |
|
|
480 |
|
|
481 int |
|
4773
|
482 octave_fftw::fftNd (const double *in, Complex *out, const int rank, |
|
|
483 const dim_vector &dv) |
|
3828
|
484 { |
|
5275
|
485 octave_idx_type dist = 1; |
|
4773
|
486 for (int i = 0; i < rank; i++) |
|
|
487 dist *= dv(i); |
|
|
488 |
|
|
489 // Fool with the position of the start of the output matrix, so that |
|
4809
|
490 // creating other half of the matrix won't cause cache problems. |
|
|
491 |
|
5275
|
492 octave_idx_type offset = (dv.numel () / dv(0)) * ((dv(0) - 1) / 2); |
|
4773
|
493 |
|
|
494 fftw_plan plan = fftw_planner.create_plan (rank, dv, 1, 1, dist, |
|
|
495 in, out + offset); |
|
|
496 |
|
|
497 fftw_execute_dft_r2c (plan, (const_cast<double *>(in)), |
|
|
498 reinterpret_cast<fftw_complex *> (out+ offset)); |
|
|
499 |
|
4809
|
500 // Need to create other half of the transform. |
|
|
501 |
|
4773
|
502 convert_packcomplex_Nd (out, dv); |
|
3828
|
503 |
|
|
504 return 0; |
|
|
505 } |
|
|
506 |
|
|
507 int |
|
4773
|
508 octave_fftw::fftNd (const Complex *in, Complex *out, const int rank, |
|
|
509 const dim_vector &dv) |
|
3828
|
510 { |
|
5275
|
511 octave_idx_type dist = 1; |
|
4773
|
512 for (int i = 0; i < rank; i++) |
|
|
513 dist *= dv(i); |
|
|
514 |
|
|
515 fftw_plan plan = fftw_planner.create_plan (FFTW_FORWARD, rank, dv, 1, 1, |
|
|
516 dist, in, out); |
|
|
517 |
|
|
518 fftw_execute_dft (plan, |
|
|
519 reinterpret_cast<fftw_complex *> (const_cast<Complex *>(in)), |
|
|
520 reinterpret_cast<fftw_complex *> (out)); |
|
|
521 |
|
|
522 return 0; |
|
|
523 } |
|
3828
|
524 |
|
4773
|
525 int |
|
|
526 octave_fftw::ifftNd (const Complex *in, Complex *out, const int rank, |
|
4784
|
527 const dim_vector &dv) |
|
4773
|
528 { |
|
5275
|
529 octave_idx_type dist = 1; |
|
4773
|
530 for (int i = 0; i < rank; i++) |
|
|
531 dist *= dv(i); |
|
|
532 |
|
|
533 fftw_plan plan = fftw_planner.create_plan (FFTW_BACKWARD, rank, dv, 1, 1, |
|
|
534 dist, in, out); |
|
|
535 |
|
|
536 fftw_execute_dft (plan, |
|
|
537 reinterpret_cast<fftw_complex *> (const_cast<Complex *>(in)), |
|
|
538 reinterpret_cast<fftw_complex *> (out)); |
|
|
539 |
|
|
540 const size_t npts = dv.numel (); |
|
3828
|
541 const Complex scale = npts; |
|
|
542 for (size_t i = 0; i < npts; i++) |
|
4773
|
543 out[i] /= scale; |
|
3828
|
544 |
|
|
545 return 0; |
|
|
546 } |
|
|
547 |
|
|
548 #endif |
|
|
549 |
|
|
550 /* |
|
|
551 ;;; Local Variables: *** |
|
|
552 ;;; mode: C++ *** |
|
|
553 ;;; End: *** |
|
|
554 */ |
|
|
555 |
