1 | SUBROUTINE transpose_xy( f_in, work, f_out ) |
---|
2 | |
---|
3 | !------------------------------------------------------------------------------! |
---|
4 | ! Current revisions: |
---|
5 | ! ----------------- |
---|
6 | ! |
---|
7 | ! |
---|
8 | ! Former revisions: |
---|
9 | ! ----------------- |
---|
10 | ! $Id: transpose.f90 1004 2012-09-14 14:56:50Z raasch $ |
---|
11 | ! |
---|
12 | ! 1003 2012-09-14 14:35:53Z raasch |
---|
13 | ! indices nxa, nya, etc. replaced by nx, ny, etc. |
---|
14 | ! |
---|
15 | ! 683 2011-02-09 14:25:15Z raasch |
---|
16 | ! openMP parallelization of transpositions for 2d-domain-decomposition |
---|
17 | ! |
---|
18 | ! 622 2010-12-10 08:08:13Z raasch |
---|
19 | ! optional barriers included in order to speed up collective operations |
---|
20 | ! |
---|
21 | ! 164 2008-05-15 08:46:15Z raasch |
---|
22 | ! f_inv changed from subroutine argument to automatic array in order to do |
---|
23 | ! re-ordering from f_in to f_inv in one step, one array work is needed instead |
---|
24 | ! of work1 and work2 |
---|
25 | ! |
---|
26 | ! February 2007 |
---|
27 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
28 | ! |
---|
29 | ! Revision 1.2 2004/04/30 13:12:17 raasch |
---|
30 | ! Switched from mpi_alltoallv to the simpler mpi_alltoall, |
---|
31 | ! all former transpose-routine files collected in this file, enlarged |
---|
32 | ! transposition arrays introduced |
---|
33 | ! |
---|
34 | ! Revision 1.1 2004/04/30 13:08:16 raasch |
---|
35 | ! Initial revision (collection of former routines transpose_xy, transpose_xz, |
---|
36 | ! transpose_yx, transpose_yz, transpose_zx, transpose_zy) |
---|
37 | ! |
---|
38 | ! Revision 1.1 1997/07/24 11:25:18 raasch |
---|
39 | ! Initial revision |
---|
40 | ! |
---|
41 | ! |
---|
42 | ! Description: |
---|
43 | ! ------------ |
---|
44 | ! Transposition of input array (f_in) from x to y. For the input array, all |
---|
45 | ! elements along x reside on the same PE, while after transposition, all |
---|
46 | ! elements along y reside on the same PE. |
---|
47 | !------------------------------------------------------------------------------! |
---|
48 | |
---|
49 | USE cpulog |
---|
50 | USE indices |
---|
51 | USE interfaces |
---|
52 | USE pegrid |
---|
53 | USE transpose_indices |
---|
54 | |
---|
55 | IMPLICIT NONE |
---|
56 | |
---|
57 | INTEGER :: i, j, k, l, m, ys |
---|
58 | |
---|
59 | REAL :: f_in(0:nx,nys_x:nyn_x,nzb_x:nzt_x), & |
---|
60 | f_inv(nys_x:nyn_x,nzb_x:nzt_x,0:nx), & |
---|
61 | f_out(0:ny,nxl_y:nxr_y,nzb_y:nzt_y), & |
---|
62 | work(nnx*nny*nnz) |
---|
63 | |
---|
64 | #if defined( __parallel ) |
---|
65 | |
---|
66 | ! |
---|
67 | !-- Rearrange indices of input array in order to make data to be send |
---|
68 | !-- by MPI contiguous |
---|
69 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
70 | !$OMP DO |
---|
71 | DO i = 0, nx |
---|
72 | DO k = nzb_x, nzt_x |
---|
73 | DO j = nys_x, nyn_x |
---|
74 | f_inv(j,k,i) = f_in(i,j,k) |
---|
75 | ENDDO |
---|
76 | ENDDO |
---|
77 | ENDDO |
---|
78 | !$OMP END PARALLEL |
---|
79 | |
---|
80 | ! |
---|
81 | !-- Transpose array |
---|
82 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
83 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
84 | CALL MPI_ALLTOALL( f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, & |
---|
85 | work(1), sendrecvcount_xy, MPI_REAL, & |
---|
86 | comm1dy, ierr ) |
---|
87 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
88 | |
---|
89 | ! |
---|
90 | !-- Reorder transposed array |
---|
91 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, ys ) |
---|
92 | !$OMP DO |
---|
93 | DO l = 0, pdims(2) - 1 |
---|
94 | m = l * ( nxr_y - nxl_y + 1 ) * ( nzt_y - nzb_y + 1 ) * & |
---|
95 | ( nyn_x - nys_x + 1 ) |
---|
96 | ys = 0 + l * ( nyn_x - nys_x + 1 ) |
---|
97 | DO i = nxl_y, nxr_y |
---|
98 | DO k = nzb_y, nzt_y |
---|
99 | DO j = ys, ys + nyn_x - nys_x |
---|
100 | m = m + 1 |
---|
101 | f_out(j,i,k) = work(m) |
---|
102 | ENDDO |
---|
103 | ENDDO |
---|
104 | ENDDO |
---|
105 | ENDDO |
---|
106 | !$OMP END PARALLEL |
---|
107 | |
---|
108 | #endif |
---|
109 | |
---|
110 | END SUBROUTINE transpose_xy |
---|
111 | |
---|
112 | |
---|
113 | SUBROUTINE transpose_xz( f_in, work, f_out ) |
---|
114 | |
---|
115 | !------------------------------------------------------------------------------! |
---|
116 | ! Description: |
---|
117 | ! ------------ |
---|
118 | ! Transposition of input array (f_in) from x to z. For the input array, all |
---|
119 | ! elements along x reside on the same PE, while after transposition, all |
---|
120 | ! elements along z reside on the same PE. |
---|
121 | !------------------------------------------------------------------------------! |
---|
122 | |
---|
123 | USE cpulog |
---|
124 | USE indices |
---|
125 | USE interfaces |
---|
126 | USE pegrid |
---|
127 | USE transpose_indices |
---|
128 | |
---|
129 | IMPLICIT NONE |
---|
130 | |
---|
131 | INTEGER :: i, j, k, l, m, xs |
---|
132 | |
---|
133 | REAL :: f_in(0:nx,nys_x:nyn_x,nzb_x:nzt_x), & |
---|
134 | f_inv(nys:nyn,nxl:nxr,1:nz), & |
---|
135 | f_out(1:nz,nys:nyn,nxl:nxr), & |
---|
136 | work(nnx*nny*nnz) |
---|
137 | |
---|
138 | #if defined( __parallel ) |
---|
139 | |
---|
140 | ! |
---|
141 | !-- If the PE grid is one-dimensional along y, the array has only to be |
---|
142 | !-- reordered locally and therefore no transposition has to be done. |
---|
143 | IF ( pdims(1) /= 1 ) THEN |
---|
144 | ! |
---|
145 | !-- Reorder input array for transposition |
---|
146 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, xs ) |
---|
147 | !$OMP DO |
---|
148 | DO l = 0, pdims(1) - 1 |
---|
149 | m = l * ( nzt_x - nzb_x + 1 ) * nnx * ( nyn_x - nys_x + 1 ) |
---|
150 | xs = 0 + l * nnx |
---|
151 | DO k = nzb_x, nzt_x |
---|
152 | DO i = xs, xs + nnx - 1 |
---|
153 | DO j = nys_x, nyn_x |
---|
154 | m = m + 1 |
---|
155 | work(m) = f_in(i,j,k) |
---|
156 | ENDDO |
---|
157 | ENDDO |
---|
158 | ENDDO |
---|
159 | ENDDO |
---|
160 | !$OMP END PARALLEL |
---|
161 | |
---|
162 | ! |
---|
163 | !-- Transpose array |
---|
164 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
165 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
166 | CALL MPI_ALLTOALL( work(1), sendrecvcount_zx, MPI_REAL, & |
---|
167 | f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, & |
---|
168 | comm1dx, ierr ) |
---|
169 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
170 | |
---|
171 | ! |
---|
172 | !-- Reorder transposed array in a way that the z index is in first position |
---|
173 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
174 | !$OMP DO |
---|
175 | DO k = 1, nz |
---|
176 | DO i = nxl, nxr |
---|
177 | DO j = nys, nyn |
---|
178 | f_out(k,j,i) = f_inv(j,i,k) |
---|
179 | ENDDO |
---|
180 | ENDDO |
---|
181 | ENDDO |
---|
182 | !$OMP END PARALLEL |
---|
183 | ELSE |
---|
184 | ! |
---|
185 | !-- Reorder the array in a way that the z index is in first position |
---|
186 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
187 | !$OMP DO |
---|
188 | DO i = nxl, nxr |
---|
189 | DO j = nys, nyn |
---|
190 | DO k = 1, nz |
---|
191 | f_inv(j,i,k) = f_in(i,j,k) |
---|
192 | ENDDO |
---|
193 | ENDDO |
---|
194 | ENDDO |
---|
195 | !$OMP END PARALLEL |
---|
196 | |
---|
197 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
198 | !$OMP DO |
---|
199 | DO k = 1, nz |
---|
200 | DO i = nxl, nxr |
---|
201 | DO j = nys, nyn |
---|
202 | f_out(k,j,i) = f_inv(j,i,k) |
---|
203 | ENDDO |
---|
204 | ENDDO |
---|
205 | ENDDO |
---|
206 | !$OMP END PARALLEL |
---|
207 | |
---|
208 | ENDIF |
---|
209 | |
---|
210 | |
---|
211 | #endif |
---|
212 | |
---|
213 | END SUBROUTINE transpose_xz |
---|
214 | |
---|
215 | |
---|
216 | SUBROUTINE transpose_yx( f_in, work, f_out ) |
---|
217 | |
---|
218 | !------------------------------------------------------------------------------! |
---|
219 | ! Description: |
---|
220 | ! ------------ |
---|
221 | ! Transposition of input array (f_in) from y to x. For the input array, all |
---|
222 | ! elements along y reside on the same PE, while after transposition, all |
---|
223 | ! elements along x reside on the same PE. |
---|
224 | !------------------------------------------------------------------------------! |
---|
225 | |
---|
226 | USE cpulog |
---|
227 | USE indices |
---|
228 | USE interfaces |
---|
229 | USE pegrid |
---|
230 | USE transpose_indices |
---|
231 | |
---|
232 | IMPLICIT NONE |
---|
233 | |
---|
234 | INTEGER :: i, j, k, l, m, ys |
---|
235 | |
---|
236 | REAL :: f_in(0:ny,nxl_y:nxr_y,nzb_y:nzt_y), & |
---|
237 | f_inv(nys_x:nyn_x,nzb_x:nzt_x,0:nx), & |
---|
238 | f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x), & |
---|
239 | work(nnx*nny*nnz) |
---|
240 | |
---|
241 | #if defined( __parallel ) |
---|
242 | |
---|
243 | ! |
---|
244 | !-- Reorder input array for transposition |
---|
245 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, ys ) |
---|
246 | !$OMP DO |
---|
247 | DO l = 0, pdims(2) - 1 |
---|
248 | m = l * ( nxr_y - nxl_y + 1 ) * ( nzt_y - nzb_y + 1 ) * & |
---|
249 | ( nyn_x - nys_x + 1 ) |
---|
250 | ys = 0 + l * ( nyn_x - nys_x + 1 ) |
---|
251 | DO i = nxl_y, nxr_y |
---|
252 | DO k = nzb_y, nzt_y |
---|
253 | DO j = ys, ys + nyn_x - nys_x |
---|
254 | m = m + 1 |
---|
255 | work(m) = f_in(j,i,k) |
---|
256 | ENDDO |
---|
257 | ENDDO |
---|
258 | ENDDO |
---|
259 | ENDDO |
---|
260 | !$OMP END PARALLEL |
---|
261 | |
---|
262 | ! |
---|
263 | !-- Transpose array |
---|
264 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
265 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
266 | CALL MPI_ALLTOALL( work(1), sendrecvcount_xy, MPI_REAL, & |
---|
267 | f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, & |
---|
268 | comm1dy, ierr ) |
---|
269 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
270 | |
---|
271 | ! |
---|
272 | !-- Reorder transposed array in a way that the x index is in first position |
---|
273 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
274 | !$OMP DO |
---|
275 | DO i = 0, nx |
---|
276 | DO k = nzb_x, nzt_x |
---|
277 | DO j = nys_x, nyn_x |
---|
278 | f_out(i,j,k) = f_inv(j,k,i) |
---|
279 | ENDDO |
---|
280 | ENDDO |
---|
281 | ENDDO |
---|
282 | !$OMP END PARALLEL |
---|
283 | |
---|
284 | #endif |
---|
285 | |
---|
286 | END SUBROUTINE transpose_yx |
---|
287 | |
---|
288 | |
---|
289 | SUBROUTINE transpose_yxd( f_in, work, f_out ) |
---|
290 | |
---|
291 | !------------------------------------------------------------------------------! |
---|
292 | ! Description: |
---|
293 | ! ------------ |
---|
294 | ! Transposition of input array (f_in) from y to x. For the input array, all |
---|
295 | ! elements along y reside on the same PE, while after transposition, all |
---|
296 | ! elements along x reside on the same PE. |
---|
297 | ! This is a direct transposition for arrays with indices in regular order |
---|
298 | ! (k,j,i) (cf. transpose_yx). |
---|
299 | !------------------------------------------------------------------------------! |
---|
300 | |
---|
301 | USE cpulog |
---|
302 | USE indices |
---|
303 | USE interfaces |
---|
304 | USE pegrid |
---|
305 | USE transpose_indices |
---|
306 | |
---|
307 | IMPLICIT NONE |
---|
308 | |
---|
309 | INTEGER :: i, j, k, l, m, recvcount_yx, sendcount_yx, xs |
---|
310 | |
---|
311 | REAL :: f_in(1:nz,nys:nyn,nxl:nxr), f_inv(nxl:nxr,1:nz,nys:nyn), & |
---|
312 | f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x), & |
---|
313 | work(nnx*nny*nnz) |
---|
314 | |
---|
315 | #if defined( __parallel ) |
---|
316 | |
---|
317 | ! |
---|
318 | !-- Rearrange indices of input array in order to make data to be send |
---|
319 | !-- by MPI contiguous |
---|
320 | DO k = 1, nz |
---|
321 | DO j = nys, nyn |
---|
322 | DO i = nxl, nxr |
---|
323 | f_inv(i,k,j) = f_in(k,j,i) |
---|
324 | ENDDO |
---|
325 | ENDDO |
---|
326 | ENDDO |
---|
327 | |
---|
328 | ! |
---|
329 | !-- Transpose array |
---|
330 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
331 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
332 | CALL MPI_ALLTOALL( f_inv(nxl,1,nys), sendrecvcount_xy, MPI_REAL, & |
---|
333 | work(1), sendrecvcount_xy, MPI_REAL, & |
---|
334 | comm1dx, ierr ) |
---|
335 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
336 | |
---|
337 | ! |
---|
338 | !-- Reorder transposed array |
---|
339 | m = 0 |
---|
340 | DO l = 0, pdims(1) - 1 |
---|
341 | xs = 0 + l * nnx |
---|
342 | DO j = nys_x, nyn_x |
---|
343 | DO k = 1, nz |
---|
344 | DO i = xs, xs + nnx - 1 |
---|
345 | m = m + 1 |
---|
346 | f_out(i,j,k) = work(m) |
---|
347 | ENDDO |
---|
348 | ENDDO |
---|
349 | ENDDO |
---|
350 | ENDDO |
---|
351 | |
---|
352 | #endif |
---|
353 | |
---|
354 | END SUBROUTINE transpose_yxd |
---|
355 | |
---|
356 | |
---|
357 | SUBROUTINE transpose_yz( f_in, work, f_out ) |
---|
358 | |
---|
359 | !------------------------------------------------------------------------------! |
---|
360 | ! Description: |
---|
361 | ! ------------ |
---|
362 | ! Transposition of input array (f_in) from y to z. For the input array, all |
---|
363 | ! elements along y reside on the same PE, while after transposition, all |
---|
364 | ! elements along z reside on the same PE. |
---|
365 | !------------------------------------------------------------------------------! |
---|
366 | |
---|
367 | USE cpulog |
---|
368 | USE indices |
---|
369 | USE interfaces |
---|
370 | USE pegrid |
---|
371 | USE transpose_indices |
---|
372 | |
---|
373 | IMPLICIT NONE |
---|
374 | |
---|
375 | INTEGER :: i, j, k, l, m, zs |
---|
376 | |
---|
377 | REAL :: f_in(0:ny,nxl_y:nxr_y,nzb_y:nzt_y), & |
---|
378 | f_inv(nxl_y:nxr_y,nzb_y:nzt_y,0:ny), & |
---|
379 | f_out(nxl_z:nxr_z,nys_z:nyn_z,1:nz), & |
---|
380 | work(nnx*nny*nnz) |
---|
381 | |
---|
382 | #if defined( __parallel ) |
---|
383 | |
---|
384 | ! |
---|
385 | !-- Rearrange indices of input array in order to make data to be send |
---|
386 | !-- by MPI contiguous |
---|
387 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
388 | !$OMP DO |
---|
389 | DO j = 0, ny |
---|
390 | DO k = nzb_y, nzt_y |
---|
391 | DO i = nxl_y, nxr_y |
---|
392 | f_inv(i,k,j) = f_in(j,i,k) |
---|
393 | ENDDO |
---|
394 | ENDDO |
---|
395 | ENDDO |
---|
396 | !$OMP END PARALLEL |
---|
397 | |
---|
398 | ! |
---|
399 | !-- Move data to different array, because memory location of work1 is |
---|
400 | !-- needed further below (work1 = work2). |
---|
401 | !-- If the PE grid is one-dimensional along y, only local reordering |
---|
402 | !-- of the data is necessary and no transposition has to be done. |
---|
403 | IF ( pdims(1) == 1 ) THEN |
---|
404 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
405 | !$OMP DO |
---|
406 | DO j = 0, ny |
---|
407 | DO k = nzb_y, nzt_y |
---|
408 | DO i = nxl_y, nxr_y |
---|
409 | f_out(i,j,k) = f_inv(i,k,j) |
---|
410 | ENDDO |
---|
411 | ENDDO |
---|
412 | ENDDO |
---|
413 | !$OMP END PARALLEL |
---|
414 | RETURN |
---|
415 | ENDIF |
---|
416 | |
---|
417 | ! |
---|
418 | !-- Transpose array |
---|
419 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
420 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
421 | CALL MPI_ALLTOALL( f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, & |
---|
422 | work(1), sendrecvcount_yz, MPI_REAL, & |
---|
423 | comm1dx, ierr ) |
---|
424 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
425 | |
---|
426 | ! |
---|
427 | !-- Reorder transposed array |
---|
428 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, zs ) |
---|
429 | !$OMP DO |
---|
430 | DO l = 0, pdims(1) - 1 |
---|
431 | m = l * ( nyn_z - nys_z + 1 ) * ( nzt_y - nzb_y + 1 ) * & |
---|
432 | ( nxr_z - nxl_z + 1 ) |
---|
433 | zs = 1 + l * ( nzt_y - nzb_y + 1 ) |
---|
434 | DO j = nys_z, nyn_z |
---|
435 | DO k = zs, zs + nzt_y - nzb_y |
---|
436 | DO i = nxl_z, nxr_z |
---|
437 | m = m + 1 |
---|
438 | f_out(i,j,k) = work(m) |
---|
439 | ENDDO |
---|
440 | ENDDO |
---|
441 | ENDDO |
---|
442 | ENDDO |
---|
443 | !$OMP END PARALLEL |
---|
444 | |
---|
445 | #endif |
---|
446 | |
---|
447 | END SUBROUTINE transpose_yz |
---|
448 | |
---|
449 | |
---|
450 | SUBROUTINE transpose_zx( f_in, work, f_out ) |
---|
451 | |
---|
452 | !------------------------------------------------------------------------------! |
---|
453 | ! Description: |
---|
454 | ! ------------ |
---|
455 | ! Transposition of input array (f_in) from z to x. For the input array, all |
---|
456 | ! elements along z reside on the same PE, while after transposition, all |
---|
457 | ! elements along x reside on the same PE. |
---|
458 | !------------------------------------------------------------------------------! |
---|
459 | |
---|
460 | USE cpulog |
---|
461 | USE indices |
---|
462 | USE interfaces |
---|
463 | USE pegrid |
---|
464 | USE transpose_indices |
---|
465 | |
---|
466 | IMPLICIT NONE |
---|
467 | |
---|
468 | INTEGER :: i, j, k, l, m, xs |
---|
469 | |
---|
470 | REAL :: f_in(1:nz,nys:nyn,nxl:nxr), f_inv(nys:nyn,nxl:nxr,1:nz), & |
---|
471 | f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x), & |
---|
472 | work(nnx*nny*nnz) |
---|
473 | |
---|
474 | #if defined( __parallel ) |
---|
475 | |
---|
476 | ! |
---|
477 | !-- Rearrange indices of input array in order to make data to be send |
---|
478 | !-- by MPI contiguous |
---|
479 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
480 | !$OMP DO |
---|
481 | DO k = 1,nz |
---|
482 | DO i = nxl, nxr |
---|
483 | DO j = nys, nyn |
---|
484 | f_inv(j,i,k) = f_in(k,j,i) |
---|
485 | ENDDO |
---|
486 | ENDDO |
---|
487 | ENDDO |
---|
488 | !$OMP END PARALLEL |
---|
489 | |
---|
490 | ! |
---|
491 | !-- Move data to different array, because memory location of work1 is |
---|
492 | !-- needed further below (work1 = work2). |
---|
493 | !-- If the PE grid is one-dimensional along y, only local reordering |
---|
494 | !-- of the data is necessary and no transposition has to be done. |
---|
495 | IF ( pdims(1) == 1 ) THEN |
---|
496 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
497 | !$OMP DO |
---|
498 | DO k = 1, nz |
---|
499 | DO i = nxl, nxr |
---|
500 | DO j = nys, nyn |
---|
501 | f_out(i,j,k) = f_inv(j,i,k) |
---|
502 | ENDDO |
---|
503 | ENDDO |
---|
504 | ENDDO |
---|
505 | !$OMP END PARALLEL |
---|
506 | RETURN |
---|
507 | ENDIF |
---|
508 | |
---|
509 | ! |
---|
510 | !-- Transpose array |
---|
511 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
512 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
513 | CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, & |
---|
514 | work(1), sendrecvcount_zx, MPI_REAL, & |
---|
515 | comm1dx, ierr ) |
---|
516 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
517 | |
---|
518 | ! |
---|
519 | !-- Reorder transposed array |
---|
520 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, xs ) |
---|
521 | !$OMP DO |
---|
522 | DO l = 0, pdims(1) - 1 |
---|
523 | m = l * ( nzt_x - nzb_x + 1 ) * nnx * ( nyn_x - nys_x + 1 ) |
---|
524 | xs = 0 + l * nnx |
---|
525 | DO k = nzb_x, nzt_x |
---|
526 | DO i = xs, xs + nnx - 1 |
---|
527 | DO j = nys_x, nyn_x |
---|
528 | m = m + 1 |
---|
529 | f_out(i,j,k) = work(m) |
---|
530 | ENDDO |
---|
531 | ENDDO |
---|
532 | ENDDO |
---|
533 | ENDDO |
---|
534 | !$OMP END PARALLEL |
---|
535 | |
---|
536 | #endif |
---|
537 | |
---|
538 | END SUBROUTINE transpose_zx |
---|
539 | |
---|
540 | |
---|
541 | SUBROUTINE transpose_zy( f_in, work, f_out ) |
---|
542 | |
---|
543 | !------------------------------------------------------------------------------! |
---|
544 | ! Description: |
---|
545 | ! ------------ |
---|
546 | ! Transposition of input array (f_in) from z to y. For the input array, all |
---|
547 | ! elements along z reside on the same PE, while after transposition, all |
---|
548 | ! elements along y reside on the same PE. |
---|
549 | !------------------------------------------------------------------------------! |
---|
550 | |
---|
551 | USE cpulog |
---|
552 | USE indices |
---|
553 | USE interfaces |
---|
554 | USE pegrid |
---|
555 | USE transpose_indices |
---|
556 | |
---|
557 | IMPLICIT NONE |
---|
558 | |
---|
559 | INTEGER :: i, j, k, l, m, zs |
---|
560 | |
---|
561 | REAL :: f_in(nxl_z:nxr_z,nys_z:nyn_z,1:nz), & |
---|
562 | f_inv(nxl_y:nxr_y,nzb_y:nzt_y,0:ny), & |
---|
563 | f_out(0:ny,nxl_y:nxr_y,nzb_y:nzt_y), & |
---|
564 | work(nnx*nny*nnz) |
---|
565 | |
---|
566 | #if defined( __parallel ) |
---|
567 | |
---|
568 | ! |
---|
569 | !-- If the PE grid is one-dimensional along y, the array has only to be |
---|
570 | !-- reordered locally and therefore no transposition has to be done. |
---|
571 | IF ( pdims(1) /= 1 ) THEN |
---|
572 | ! |
---|
573 | !-- Reorder input array for transposition |
---|
574 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, zs ) |
---|
575 | !$OMP DO |
---|
576 | DO l = 0, pdims(1) - 1 |
---|
577 | m = l * ( nyn_z - nys_z + 1 ) * ( nzt_y - nzb_y + 1 ) * & |
---|
578 | ( nxr_z - nxl_z + 1 ) |
---|
579 | zs = 1 + l * ( nzt_y - nzb_y + 1 ) |
---|
580 | DO j = nys_z, nyn_z |
---|
581 | DO k = zs, zs + nzt_y - nzb_y |
---|
582 | DO i = nxl_z, nxr_z |
---|
583 | m = m + 1 |
---|
584 | work(m) = f_in(i,j,k) |
---|
585 | ENDDO |
---|
586 | ENDDO |
---|
587 | ENDDO |
---|
588 | ENDDO |
---|
589 | !$OMP END PARALLEL |
---|
590 | |
---|
591 | ! |
---|
592 | !-- Transpose array |
---|
593 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
594 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
595 | CALL MPI_ALLTOALL( work(1), sendrecvcount_yz, MPI_REAL, & |
---|
596 | f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, & |
---|
597 | comm1dx, ierr ) |
---|
598 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
599 | |
---|
600 | ! |
---|
601 | !-- Reorder transposed array in a way that the y index is in first position |
---|
602 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
603 | !$OMP DO |
---|
604 | DO j = 0, ny |
---|
605 | DO k = nzb_y, nzt_y |
---|
606 | DO i = nxl_y, nxr_y |
---|
607 | f_out(j,i,k) = f_inv(i,k,j) |
---|
608 | ENDDO |
---|
609 | ENDDO |
---|
610 | ENDDO |
---|
611 | !$OMP END PARALLEL |
---|
612 | ELSE |
---|
613 | ! |
---|
614 | !-- Reorder the array in a way that the y index is in first position |
---|
615 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
616 | !$OMP DO |
---|
617 | DO k = nzb_y, nzt_y |
---|
618 | DO j = 0, ny |
---|
619 | DO i = nxl_y, nxr_y |
---|
620 | f_inv(i,k,j) = f_in(i,j,k) |
---|
621 | ENDDO |
---|
622 | ENDDO |
---|
623 | ENDDO |
---|
624 | !$OMP END PARALLEL |
---|
625 | ! |
---|
626 | !-- Move data to output array |
---|
627 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
628 | !$OMP DO |
---|
629 | DO k = nzb_y, nzt_y |
---|
630 | DO i = nxl_y, nxr_y |
---|
631 | DO j = 0, ny |
---|
632 | f_out(j,i,k) = f_inv(i,k,j) |
---|
633 | ENDDO |
---|
634 | ENDDO |
---|
635 | ENDDO |
---|
636 | !$OMP END PARALLEL |
---|
637 | |
---|
638 | ENDIF |
---|
639 | |
---|
640 | #endif |
---|
641 | |
---|
642 | END SUBROUTINE transpose_zy |
---|
643 | |
---|
644 | |
---|
645 | SUBROUTINE transpose_zyd( f_in, work, f_out ) |
---|
646 | |
---|
647 | !------------------------------------------------------------------------------! |
---|
648 | ! Description: |
---|
649 | ! ------------ |
---|
650 | ! Transposition of input array (f_in) from z to y. For the input array, all |
---|
651 | ! elements along z reside on the same PE, while after transposition, all |
---|
652 | ! elements along y reside on the same PE. |
---|
653 | ! This is a direct transposition for arrays with indices in regular order |
---|
654 | ! (k,j,i) (cf. transpose_zy). |
---|
655 | !------------------------------------------------------------------------------! |
---|
656 | |
---|
657 | USE cpulog |
---|
658 | USE indices |
---|
659 | USE interfaces |
---|
660 | USE pegrid |
---|
661 | USE transpose_indices |
---|
662 | |
---|
663 | IMPLICIT NONE |
---|
664 | |
---|
665 | INTEGER :: i, j, k, l, m, ys |
---|
666 | |
---|
667 | REAL :: f_in(1:nz,nys:nyn,nxl:nxr), f_inv(nys:nyn,nxl:nxr,1:nz), & |
---|
668 | f_out(0:ny,nxl_yd:nxr_yd,nzb_yd:nzt_yd), & |
---|
669 | work(nnx*nny*nnz) |
---|
670 | |
---|
671 | #if defined( __parallel ) |
---|
672 | |
---|
673 | ! |
---|
674 | !-- Rearrange indices of input array in order to make data to be send |
---|
675 | !-- by MPI contiguous |
---|
676 | DO i = nxl, nxr |
---|
677 | DO j = nys, nyn |
---|
678 | DO k = 1, nz |
---|
679 | f_inv(j,i,k) = f_in(k,j,i) |
---|
680 | ENDDO |
---|
681 | ENDDO |
---|
682 | ENDDO |
---|
683 | |
---|
684 | ! |
---|
685 | !-- Move data to different array, because memory location of work1 is |
---|
686 | !-- needed further below (work1 = work2). |
---|
687 | !-- If the PE grid is one-dimensional along x, only local reordering |
---|
688 | !-- of the data is necessary and no transposition has to be done. |
---|
689 | IF ( pdims(2) == 1 ) THEN |
---|
690 | DO k = 1, nz |
---|
691 | DO i = nxl, nxr |
---|
692 | DO j = nys, nyn |
---|
693 | f_out(j,i,k) = f_inv(j,i,k) |
---|
694 | ENDDO |
---|
695 | ENDDO |
---|
696 | ENDDO |
---|
697 | RETURN |
---|
698 | ENDIF |
---|
699 | |
---|
700 | ! |
---|
701 | !-- Transpose array |
---|
702 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
703 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
704 | CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zyd, MPI_REAL, & |
---|
705 | work(1), sendrecvcount_zyd, MPI_REAL, & |
---|
706 | comm1dy, ierr ) |
---|
707 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
708 | |
---|
709 | ! |
---|
710 | !-- Reorder transposed array |
---|
711 | m = 0 |
---|
712 | DO l = 0, pdims(2) - 1 |
---|
713 | ys = 0 + l * nny |
---|
714 | DO k = nzb_yd, nzt_yd |
---|
715 | DO i = nxl_yd, nxr_yd |
---|
716 | DO j = ys, ys + nny - 1 |
---|
717 | m = m + 1 |
---|
718 | f_out(j,i,k) = work(m) |
---|
719 | ENDDO |
---|
720 | ENDDO |
---|
721 | ENDDO |
---|
722 | ENDDO |
---|
723 | |
---|
724 | #endif |
---|
725 | |
---|
726 | END SUBROUTINE transpose_zyd |
---|