1 | SUBROUTINE transpose_xy( f_in, work, f_out ) |
---|
2 | |
---|
3 | !------------------------------------------------------------------------------! |
---|
4 | ! Actual revisions: |
---|
5 | ! ----------------- |
---|
6 | ! |
---|
7 | ! |
---|
8 | ! Former revisions: |
---|
9 | ! ----------------- |
---|
10 | ! $Id: transpose.f90 198 2008-09-17 08:55:28Z heinze $ |
---|
11 | ! |
---|
12 | ! 164 2008-05-15 08:46:15Z raasch |
---|
13 | ! f_inv changed from subroutine argument to automatic array in order to do |
---|
14 | ! re-ordering from f_in to f_inv in one step, one array work is needed instead |
---|
15 | ! of work1 and work2 |
---|
16 | ! |
---|
17 | ! February 2007 |
---|
18 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
19 | ! |
---|
20 | ! Revision 1.2 2004/04/30 13:12:17 raasch |
---|
21 | ! Switched from mpi_alltoallv to the simpler mpi_alltoall, |
---|
22 | ! all former transpose-routine files collected in this file, enlarged |
---|
23 | ! transposition arrays introduced |
---|
24 | ! |
---|
25 | ! Revision 1.1 2004/04/30 13:08:16 raasch |
---|
26 | ! Initial revision (collection of former routines transpose_xy, transpose_xz, |
---|
27 | ! transpose_yx, transpose_yz, transpose_zx, transpose_zy) |
---|
28 | ! |
---|
29 | ! Revision 1.1 1997/07/24 11:25:18 raasch |
---|
30 | ! Initial revision |
---|
31 | ! |
---|
32 | ! |
---|
33 | ! Description: |
---|
34 | ! ------------ |
---|
35 | ! Transposition of input array (f_in) from x to y. For the input array, all |
---|
36 | ! elements along x reside on the same PE, while after transposition, all |
---|
37 | ! elements along y reside on the same PE. |
---|
38 | !------------------------------------------------------------------------------! |
---|
39 | |
---|
40 | USE cpulog |
---|
41 | USE indices |
---|
42 | USE interfaces |
---|
43 | USE pegrid |
---|
44 | USE transpose_indices |
---|
45 | |
---|
46 | IMPLICIT NONE |
---|
47 | |
---|
48 | INTEGER :: i, j, k, l, m, ys |
---|
49 | |
---|
50 | REAL :: f_in(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
---|
51 | f_inv(nys_x:nyn_xa,nzb_x:nzt_xa,0:nxa), & |
---|
52 | f_out(0:nya,nxl_y:nxr_ya,nzb_y:nzt_ya), & |
---|
53 | work(nnx*nny*nnz) |
---|
54 | |
---|
55 | #if defined( __parallel ) |
---|
56 | |
---|
57 | ! |
---|
58 | !-- Rearrange indices of input array in order to make data to be send |
---|
59 | !-- by MPI contiguous |
---|
60 | DO i = 0, nxa |
---|
61 | DO k = nzb_x, nzt_xa |
---|
62 | DO j = nys_x, nyn_xa |
---|
63 | f_inv(j,k,i) = f_in(i,j,k) |
---|
64 | ENDDO |
---|
65 | ENDDO |
---|
66 | ENDDO |
---|
67 | |
---|
68 | ! |
---|
69 | !-- Transpose array |
---|
70 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
71 | CALL MPI_ALLTOALL( f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, & |
---|
72 | work(1), sendrecvcount_xy, MPI_REAL, & |
---|
73 | comm1dy, ierr ) |
---|
74 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
75 | |
---|
76 | ! |
---|
77 | !-- Reorder transposed array |
---|
78 | m = 0 |
---|
79 | DO l = 0, pdims(2) - 1 |
---|
80 | ys = 0 + l * ( nyn_xa - nys_x + 1 ) |
---|
81 | DO i = nxl_y, nxr_ya |
---|
82 | DO k = nzb_y, nzt_ya |
---|
83 | DO j = ys, ys + nyn_xa - nys_x |
---|
84 | m = m + 1 |
---|
85 | f_out(j,i,k) = work(m) |
---|
86 | ENDDO |
---|
87 | ENDDO |
---|
88 | ENDDO |
---|
89 | ENDDO |
---|
90 | |
---|
91 | #endif |
---|
92 | |
---|
93 | END SUBROUTINE transpose_xy |
---|
94 | |
---|
95 | |
---|
96 | SUBROUTINE transpose_xz( f_in, work, f_out ) |
---|
97 | |
---|
98 | !------------------------------------------------------------------------------! |
---|
99 | ! Description: |
---|
100 | ! ------------ |
---|
101 | ! Transposition of input array (f_in) from x to z. For the input array, all |
---|
102 | ! elements along x reside on the same PE, while after transposition, all |
---|
103 | ! elements along z reside on the same PE. |
---|
104 | !------------------------------------------------------------------------------! |
---|
105 | |
---|
106 | USE cpulog |
---|
107 | USE indices |
---|
108 | USE interfaces |
---|
109 | USE pegrid |
---|
110 | USE transpose_indices |
---|
111 | |
---|
112 | IMPLICIT NONE |
---|
113 | |
---|
114 | INTEGER :: i, j, k, l, m, xs |
---|
115 | |
---|
116 | REAL :: f_in(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
---|
117 | f_inv(nys:nyna,nxl:nxra,1:nza), & |
---|
118 | f_out(1:nza,nys:nyna,nxl:nxra), & |
---|
119 | work(nnx*nny*nnz) |
---|
120 | |
---|
121 | #if defined( __parallel ) |
---|
122 | |
---|
123 | ! |
---|
124 | !-- If the PE grid is one-dimensional along y, the array has only to be |
---|
125 | !-- reordered locally and therefore no transposition has to be done. |
---|
126 | IF ( pdims(1) /= 1 ) THEN |
---|
127 | ! |
---|
128 | !-- Reorder input array for transposition |
---|
129 | m = 0 |
---|
130 | DO l = 0, pdims(1) - 1 |
---|
131 | xs = 0 + l * nnx |
---|
132 | DO k = nzb_x, nzt_xa |
---|
133 | DO i = xs, xs + nnx - 1 |
---|
134 | DO j = nys_x, nyn_xa |
---|
135 | m = m + 1 |
---|
136 | work(m) = f_in(i,j,k) |
---|
137 | ENDDO |
---|
138 | ENDDO |
---|
139 | ENDDO |
---|
140 | ENDDO |
---|
141 | |
---|
142 | ! |
---|
143 | !-- Transpose array |
---|
144 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
145 | CALL MPI_ALLTOALL( work(1), sendrecvcount_zx, MPI_REAL, & |
---|
146 | f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, & |
---|
147 | comm1dx, ierr ) |
---|
148 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
149 | |
---|
150 | ! |
---|
151 | !-- Reorder transposed array in a way that the z index is in first position |
---|
152 | DO k = 1, nza |
---|
153 | DO i = nxl, nxra |
---|
154 | DO j = nys, nyna |
---|
155 | f_out(k,j,i) = f_inv(j,i,k) |
---|
156 | ENDDO |
---|
157 | ENDDO |
---|
158 | ENDDO |
---|
159 | ELSE |
---|
160 | ! |
---|
161 | !-- Reorder the array in a way that the z index is in first position |
---|
162 | DO i = nxl, nxra |
---|
163 | DO j = nys, nyna |
---|
164 | DO k = 1, nza |
---|
165 | f_inv(j,i,k) = f_in(i,j,k) |
---|
166 | ENDDO |
---|
167 | ENDDO |
---|
168 | ENDDO |
---|
169 | |
---|
170 | DO k = 1, nza |
---|
171 | DO i = nxl, nxra |
---|
172 | DO j = nys, nyna |
---|
173 | f_out(k,j,i) = f_inv(j,i,k) |
---|
174 | ENDDO |
---|
175 | ENDDO |
---|
176 | ENDDO |
---|
177 | |
---|
178 | ENDIF |
---|
179 | |
---|
180 | |
---|
181 | #endif |
---|
182 | |
---|
183 | END SUBROUTINE transpose_xz |
---|
184 | |
---|
185 | |
---|
186 | SUBROUTINE transpose_yx( f_in, work, f_out ) |
---|
187 | |
---|
188 | !------------------------------------------------------------------------------! |
---|
189 | ! Description: |
---|
190 | ! ------------ |
---|
191 | ! Transposition of input array (f_in) from y to x. For the input array, all |
---|
192 | ! elements along y reside on the same PE, while after transposition, all |
---|
193 | ! elements along x reside on the same PE. |
---|
194 | !------------------------------------------------------------------------------! |
---|
195 | |
---|
196 | USE cpulog |
---|
197 | USE indices |
---|
198 | USE interfaces |
---|
199 | USE pegrid |
---|
200 | USE transpose_indices |
---|
201 | |
---|
202 | IMPLICIT NONE |
---|
203 | |
---|
204 | INTEGER :: i, j, k, l, m, ys |
---|
205 | |
---|
206 | REAL :: f_in(0:nya,nxl_y:nxr_ya,nzb_y:nzt_ya), & |
---|
207 | f_inv(nys_x:nyn_xa,nzb_x:nzt_xa,0:nxa), & |
---|
208 | f_out(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
---|
209 | work(nnx*nny*nnz) |
---|
210 | |
---|
211 | #if defined( __parallel ) |
---|
212 | |
---|
213 | ! |
---|
214 | !-- Reorder input array for transposition |
---|
215 | m = 0 |
---|
216 | DO l = 0, pdims(2) - 1 |
---|
217 | ys = 0 + l * ( nyn_xa - nys_x + 1 ) |
---|
218 | DO i = nxl_y, nxr_ya |
---|
219 | DO k = nzb_y, nzt_ya |
---|
220 | DO j = ys, ys + nyn_xa - nys_x |
---|
221 | m = m + 1 |
---|
222 | work(m) = f_in(j,i,k) |
---|
223 | ENDDO |
---|
224 | ENDDO |
---|
225 | ENDDO |
---|
226 | ENDDO |
---|
227 | |
---|
228 | ! |
---|
229 | !-- Transpose array |
---|
230 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
231 | CALL MPI_ALLTOALL( work(1), sendrecvcount_xy, MPI_REAL, & |
---|
232 | f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, & |
---|
233 | comm1dy, ierr ) |
---|
234 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
235 | |
---|
236 | ! |
---|
237 | !-- Reorder transposed array in a way that the x index is in first position |
---|
238 | DO i = 0, nxa |
---|
239 | DO k = nzb_x, nzt_xa |
---|
240 | DO j = nys_x, nyn_xa |
---|
241 | f_out(i,j,k) = f_inv(j,k,i) |
---|
242 | ENDDO |
---|
243 | ENDDO |
---|
244 | ENDDO |
---|
245 | |
---|
246 | #endif |
---|
247 | |
---|
248 | END SUBROUTINE transpose_yx |
---|
249 | |
---|
250 | |
---|
251 | SUBROUTINE transpose_yxd( f_in, work, f_out ) |
---|
252 | |
---|
253 | !------------------------------------------------------------------------------! |
---|
254 | ! Description: |
---|
255 | ! ------------ |
---|
256 | ! Transposition of input array (f_in) from y to x. For the input array, all |
---|
257 | ! elements along y reside on the same PE, while after transposition, all |
---|
258 | ! elements along x reside on the same PE. |
---|
259 | ! This is a direct transposition for arrays with indices in regular order |
---|
260 | ! (k,j,i) (cf. transpose_yx). |
---|
261 | !------------------------------------------------------------------------------! |
---|
262 | |
---|
263 | USE cpulog |
---|
264 | USE indices |
---|
265 | USE interfaces |
---|
266 | USE pegrid |
---|
267 | USE transpose_indices |
---|
268 | |
---|
269 | IMPLICIT NONE |
---|
270 | |
---|
271 | INTEGER :: i, j, k, l, m, recvcount_yx, sendcount_yx, xs |
---|
272 | |
---|
273 | REAL :: f_in(1:nza,nys:nyna,nxl:nxra), f_inv(nxl:nxra,1:nza,nys:nyna), & |
---|
274 | f_out(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
---|
275 | work(nnx*nny*nnz) |
---|
276 | |
---|
277 | #if defined( __parallel ) |
---|
278 | |
---|
279 | ! |
---|
280 | !-- Rearrange indices of input array in order to make data to be send |
---|
281 | !-- by MPI contiguous |
---|
282 | DO k = 1, nza |
---|
283 | DO j = nys, nyna |
---|
284 | DO i = nxl, nxra |
---|
285 | f_inv(i,k,j) = f_in(k,j,i) |
---|
286 | ENDDO |
---|
287 | ENDDO |
---|
288 | ENDDO |
---|
289 | |
---|
290 | ! |
---|
291 | !-- Transpose array |
---|
292 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
293 | CALL MPI_ALLTOALL( f_inv(nxl,1,nys), sendrecvcount_xy, MPI_REAL, & |
---|
294 | work(1), sendrecvcount_xy, MPI_REAL, & |
---|
295 | comm1dx, ierr ) |
---|
296 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
297 | |
---|
298 | ! |
---|
299 | !-- Reorder transposed array |
---|
300 | m = 0 |
---|
301 | DO l = 0, pdims(1) - 1 |
---|
302 | xs = 0 + l * nnx |
---|
303 | DO j = nys_x, nyn_xa |
---|
304 | DO k = 1, nza |
---|
305 | DO i = xs, xs + nnx - 1 |
---|
306 | m = m + 1 |
---|
307 | f_out(i,j,k) = work(m) |
---|
308 | ENDDO |
---|
309 | ENDDO |
---|
310 | ENDDO |
---|
311 | ENDDO |
---|
312 | |
---|
313 | #endif |
---|
314 | |
---|
315 | END SUBROUTINE transpose_yxd |
---|
316 | |
---|
317 | |
---|
318 | SUBROUTINE transpose_yz( f_in, work, f_out ) |
---|
319 | |
---|
320 | !------------------------------------------------------------------------------! |
---|
321 | ! Description: |
---|
322 | ! ------------ |
---|
323 | ! Transposition of input array (f_in) from y to z. For the input array, all |
---|
324 | ! elements along y reside on the same PE, while after transposition, all |
---|
325 | ! elements along z reside on the same PE. |
---|
326 | !------------------------------------------------------------------------------! |
---|
327 | |
---|
328 | USE cpulog |
---|
329 | USE indices |
---|
330 | USE interfaces |
---|
331 | USE pegrid |
---|
332 | USE transpose_indices |
---|
333 | |
---|
334 | IMPLICIT NONE |
---|
335 | |
---|
336 | INTEGER :: i, j, k, l, m, zs |
---|
337 | |
---|
338 | REAL :: f_in(0:nya,nxl_y:nxr_ya,nzb_y:nzt_ya), & |
---|
339 | f_inv(nxl_y:nxr_ya,nzb_y:nzt_ya,0:nya), & |
---|
340 | f_out(nxl_z:nxr_za,nys_z:nyn_za,1:nza), & |
---|
341 | work(nnx*nny*nnz) |
---|
342 | |
---|
343 | #if defined( __parallel ) |
---|
344 | |
---|
345 | ! |
---|
346 | !-- Rearrange indices of input array in order to make data to be send |
---|
347 | !-- by MPI contiguous |
---|
348 | DO j = 0, nya |
---|
349 | DO k = nzb_y, nzt_ya |
---|
350 | DO i = nxl_y, nxr_ya |
---|
351 | f_inv(i,k,j) = f_in(j,i,k) |
---|
352 | ENDDO |
---|
353 | ENDDO |
---|
354 | ENDDO |
---|
355 | |
---|
356 | ! |
---|
357 | !-- Move data to different array, because memory location of work1 is |
---|
358 | !-- needed further below (work1 = work2). |
---|
359 | !-- If the PE grid is one-dimensional along y, only local reordering |
---|
360 | !-- of the data is necessary and no transposition has to be done. |
---|
361 | IF ( pdims(1) == 1 ) THEN |
---|
362 | DO j = 0, nya |
---|
363 | DO k = nzb_y, nzt_ya |
---|
364 | DO i = nxl_y, nxr_ya |
---|
365 | f_out(i,j,k) = f_inv(i,k,j) |
---|
366 | ENDDO |
---|
367 | ENDDO |
---|
368 | ENDDO |
---|
369 | RETURN |
---|
370 | ENDIF |
---|
371 | |
---|
372 | ! |
---|
373 | !-- Transpose array |
---|
374 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
375 | CALL MPI_ALLTOALL( f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, & |
---|
376 | work(1), sendrecvcount_yz, MPI_REAL, & |
---|
377 | comm1dx, ierr ) |
---|
378 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
379 | |
---|
380 | ! |
---|
381 | !-- Reorder transposed array |
---|
382 | m = 0 |
---|
383 | DO l = 0, pdims(1) - 1 |
---|
384 | zs = 1 + l * ( nzt_ya - nzb_y + 1 ) |
---|
385 | DO j = nys_z, nyn_za |
---|
386 | DO k = zs, zs + nzt_ya - nzb_y |
---|
387 | DO i = nxl_z, nxr_za |
---|
388 | m = m + 1 |
---|
389 | f_out(i,j,k) = work(m) |
---|
390 | ENDDO |
---|
391 | ENDDO |
---|
392 | ENDDO |
---|
393 | ENDDO |
---|
394 | |
---|
395 | #endif |
---|
396 | |
---|
397 | END SUBROUTINE transpose_yz |
---|
398 | |
---|
399 | |
---|
400 | SUBROUTINE transpose_zx( f_in, work, f_out ) |
---|
401 | |
---|
402 | !------------------------------------------------------------------------------! |
---|
403 | ! Description: |
---|
404 | ! ------------ |
---|
405 | ! Transposition of input array (f_in) from z to x. For the input array, all |
---|
406 | ! elements along z reside on the same PE, while after transposition, all |
---|
407 | ! elements along x reside on the same PE. |
---|
408 | !------------------------------------------------------------------------------! |
---|
409 | |
---|
410 | USE cpulog |
---|
411 | USE indices |
---|
412 | USE interfaces |
---|
413 | USE pegrid |
---|
414 | USE transpose_indices |
---|
415 | |
---|
416 | IMPLICIT NONE |
---|
417 | |
---|
418 | INTEGER :: i, j, k, l, m, xs |
---|
419 | |
---|
420 | REAL :: f_in(1:nza,nys:nyna,nxl:nxra), f_inv(nys:nyna,nxl:nxra,1:nza), & |
---|
421 | f_out(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
---|
422 | work(nnx*nny*nnz) |
---|
423 | |
---|
424 | #if defined( __parallel ) |
---|
425 | |
---|
426 | ! |
---|
427 | !-- Rearrange indices of input array in order to make data to be send |
---|
428 | !-- by MPI contiguous |
---|
429 | DO k = 1,nza |
---|
430 | DO i = nxl, nxra |
---|
431 | DO j = nys, nyna |
---|
432 | f_inv(j,i,k) = f_in(k,j,i) |
---|
433 | ENDDO |
---|
434 | ENDDO |
---|
435 | ENDDO |
---|
436 | |
---|
437 | ! |
---|
438 | !-- Move data to different array, because memory location of work1 is |
---|
439 | !-- needed further below (work1 = work2). |
---|
440 | !-- If the PE grid is one-dimensional along y, only local reordering |
---|
441 | !-- of the data is necessary and no transposition has to be done. |
---|
442 | IF ( pdims(1) == 1 ) THEN |
---|
443 | DO k = 1, nza |
---|
444 | DO i = nxl, nxra |
---|
445 | DO j = nys, nyna |
---|
446 | f_out(i,j,k) = f_inv(j,i,k) |
---|
447 | ENDDO |
---|
448 | ENDDO |
---|
449 | ENDDO |
---|
450 | RETURN |
---|
451 | ENDIF |
---|
452 | |
---|
453 | ! |
---|
454 | !-- Transpose array |
---|
455 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
456 | CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, & |
---|
457 | work(1), sendrecvcount_zx, MPI_REAL, & |
---|
458 | comm1dx, ierr ) |
---|
459 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
460 | |
---|
461 | ! |
---|
462 | !-- Reorder transposed array |
---|
463 | m = 0 |
---|
464 | DO l = 0, pdims(1) - 1 |
---|
465 | xs = 0 + l * nnx |
---|
466 | DO k = nzb_x, nzt_xa |
---|
467 | DO i = xs, xs + nnx - 1 |
---|
468 | DO j = nys_x, nyn_xa |
---|
469 | m = m + 1 |
---|
470 | f_out(i,j,k) = work(m) |
---|
471 | ENDDO |
---|
472 | ENDDO |
---|
473 | ENDDO |
---|
474 | ENDDO |
---|
475 | |
---|
476 | #endif |
---|
477 | |
---|
478 | END SUBROUTINE transpose_zx |
---|
479 | |
---|
480 | |
---|
481 | SUBROUTINE transpose_zy( f_in, work, f_out ) |
---|
482 | |
---|
483 | !------------------------------------------------------------------------------! |
---|
484 | ! Description: |
---|
485 | ! ------------ |
---|
486 | ! Transposition of input array (f_in) from z to y. For the input array, all |
---|
487 | ! elements along z reside on the same PE, while after transposition, all |
---|
488 | ! elements along y reside on the same PE. |
---|
489 | !------------------------------------------------------------------------------! |
---|
490 | |
---|
491 | USE cpulog |
---|
492 | USE indices |
---|
493 | USE interfaces |
---|
494 | USE pegrid |
---|
495 | USE transpose_indices |
---|
496 | |
---|
497 | IMPLICIT NONE |
---|
498 | |
---|
499 | INTEGER :: i, j, k, l, m, zs |
---|
500 | |
---|
501 | REAL :: f_in(nxl_z:nxr_za,nys_z:nyn_za,1:nza), & |
---|
502 | f_inv(nxl_y:nxr_ya,nzb_y:nzt_ya,0:nya), & |
---|
503 | f_out(0:nya,nxl_y:nxr_ya,nzb_y:nzt_ya), & |
---|
504 | work(nnx*nny*nnz) |
---|
505 | |
---|
506 | #if defined( __parallel ) |
---|
507 | |
---|
508 | ! |
---|
509 | !-- If the PE grid is one-dimensional along y, the array has only to be |
---|
510 | !-- reordered locally and therefore no transposition has to be done. |
---|
511 | IF ( pdims(1) /= 1 ) THEN |
---|
512 | ! |
---|
513 | !-- Reorder input array for transposition |
---|
514 | m = 0 |
---|
515 | DO l = 0, pdims(1) - 1 |
---|
516 | zs = 1 + l * ( nzt_ya - nzb_y + 1 ) |
---|
517 | DO j = nys_z, nyn_za |
---|
518 | DO k = zs, zs + nzt_ya - nzb_y |
---|
519 | DO i = nxl_z, nxr_za |
---|
520 | m = m + 1 |
---|
521 | work(m) = f_in(i,j,k) |
---|
522 | ENDDO |
---|
523 | ENDDO |
---|
524 | ENDDO |
---|
525 | ENDDO |
---|
526 | |
---|
527 | ! |
---|
528 | !-- Transpose array |
---|
529 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
530 | CALL MPI_ALLTOALL( work(1), sendrecvcount_yz, MPI_REAL, & |
---|
531 | f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, & |
---|
532 | comm1dx, ierr ) |
---|
533 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
534 | |
---|
535 | ! |
---|
536 | !-- Reorder transposed array in a way that the y index is in first position |
---|
537 | DO j = 0, nya |
---|
538 | DO k = nzb_y, nzt_ya |
---|
539 | DO i = nxl_y, nxr_ya |
---|
540 | f_out(j,i,k) = f_inv(i,k,j) |
---|
541 | ENDDO |
---|
542 | ENDDO |
---|
543 | ENDDO |
---|
544 | ELSE |
---|
545 | ! |
---|
546 | !-- Reorder the array in a way that the y index is in first position |
---|
547 | DO k = nzb_y, nzt_ya |
---|
548 | DO j = 0, nya |
---|
549 | DO i = nxl_y, nxr_ya |
---|
550 | f_inv(i,k,j) = f_in(i,j,k) |
---|
551 | ENDDO |
---|
552 | ENDDO |
---|
553 | ENDDO |
---|
554 | ! |
---|
555 | !-- Move data to output array |
---|
556 | DO k = nzb_y, nzt_ya |
---|
557 | DO i = nxl_y, nxr_ya |
---|
558 | DO j = 0, nya |
---|
559 | f_out(j,i,k) = f_inv(i,k,j) |
---|
560 | ENDDO |
---|
561 | ENDDO |
---|
562 | ENDDO |
---|
563 | |
---|
564 | ENDIF |
---|
565 | |
---|
566 | #endif |
---|
567 | |
---|
568 | END SUBROUTINE transpose_zy |
---|
569 | |
---|
570 | |
---|
571 | SUBROUTINE transpose_zyd( f_in, work, f_out ) |
---|
572 | |
---|
573 | !------------------------------------------------------------------------------! |
---|
574 | ! Description: |
---|
575 | ! ------------ |
---|
576 | ! Transposition of input array (f_in) from z to y. For the input array, all |
---|
577 | ! elements along z reside on the same PE, while after transposition, all |
---|
578 | ! elements along y reside on the same PE. |
---|
579 | ! This is a direct transposition for arrays with indices in regular order |
---|
580 | ! (k,j,i) (cf. transpose_zy). |
---|
581 | !------------------------------------------------------------------------------! |
---|
582 | |
---|
583 | USE cpulog |
---|
584 | USE indices |
---|
585 | USE interfaces |
---|
586 | USE pegrid |
---|
587 | USE transpose_indices |
---|
588 | |
---|
589 | IMPLICIT NONE |
---|
590 | |
---|
591 | INTEGER :: i, j, k, l, m, ys |
---|
592 | |
---|
593 | REAL :: f_in(1:nza,nys:nyna,nxl:nxra), f_inv(nys:nyna,nxl:nxra,1:nza), & |
---|
594 | f_out(0:nya,nxl_yd:nxr_yda,nzb_yd:nzt_yda), & |
---|
595 | work(nnx*nny*nnz) |
---|
596 | |
---|
597 | #if defined( __parallel ) |
---|
598 | |
---|
599 | ! |
---|
600 | !-- Rearrange indices of input array in order to make data to be send |
---|
601 | !-- by MPI contiguous |
---|
602 | DO i = nxl, nxra |
---|
603 | DO j = nys, nyna |
---|
604 | DO k = 1, nza |
---|
605 | f_inv(j,i,k) = f_in(k,j,i) |
---|
606 | ENDDO |
---|
607 | ENDDO |
---|
608 | ENDDO |
---|
609 | |
---|
610 | ! |
---|
611 | !-- Move data to different array, because memory location of work1 is |
---|
612 | !-- needed further below (work1 = work2). |
---|
613 | !-- If the PE grid is one-dimensional along x, only local reordering |
---|
614 | !-- of the data is necessary and no transposition has to be done. |
---|
615 | IF ( pdims(2) == 1 ) THEN |
---|
616 | DO k = 1, nza |
---|
617 | DO i = nxl, nxra |
---|
618 | DO j = nys, nyna |
---|
619 | f_out(j,i,k) = f_inv(j,i,k) |
---|
620 | ENDDO |
---|
621 | ENDDO |
---|
622 | ENDDO |
---|
623 | RETURN |
---|
624 | ENDIF |
---|
625 | |
---|
626 | ! |
---|
627 | !-- Transpose array |
---|
628 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
629 | CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zyd, MPI_REAL, & |
---|
630 | work(1), sendrecvcount_zyd, MPI_REAL, & |
---|
631 | comm1dy, ierr ) |
---|
632 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
633 | |
---|
634 | ! |
---|
635 | !-- Reorder transposed array |
---|
636 | m = 0 |
---|
637 | DO l = 0, pdims(2) - 1 |
---|
638 | ys = 0 + l * nny |
---|
639 | DO k = nzb_yd, nzt_yda |
---|
640 | DO i = nxl_yd, nxr_yda |
---|
641 | DO j = ys, ys + nny - 1 |
---|
642 | m = m + 1 |
---|
643 | f_out(j,i,k) = work(m) |
---|
644 | ENDDO |
---|
645 | ENDDO |
---|
646 | ENDDO |
---|
647 | ENDDO |
---|
648 | |
---|
649 | #endif |
---|
650 | |
---|
651 | END SUBROUTINE transpose_zyd |
---|