1 | !> @file transpose.f90 |
---|
2 | !------------------------------------------------------------------------------! |
---|
3 | ! This file is part of the PALM model system. |
---|
4 | ! |
---|
5 | ! PALM is free software: you can redistribute it and/or modify it under the |
---|
6 | ! terms of the GNU General Public License as published by the Free Software |
---|
7 | ! Foundation, either version 3 of the License, or (at your option) any later |
---|
8 | ! version. |
---|
9 | ! |
---|
10 | ! PALM is distributed in the hope that it will be useful, but WITHOUT ANY |
---|
11 | ! WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR |
---|
12 | ! A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
---|
13 | ! |
---|
14 | ! You should have received a copy of the GNU General Public License along with |
---|
15 | ! PALM. If not, see <http://www.gnu.org/licenses/>. |
---|
16 | ! |
---|
17 | ! Copyright 1997-2019 Leibniz Universitaet Hannover |
---|
18 | !------------------------------------------------------------------------------! |
---|
19 | ! |
---|
20 | ! Current revisions: |
---|
21 | ! ----------------- |
---|
22 | ! |
---|
23 | ! |
---|
24 | ! Former revisions: |
---|
25 | ! ----------------- |
---|
26 | ! $Id: transpose.f90 4180 2019-08-21 14:37:54Z scharf $ |
---|
27 | ! loop reordering for performance optimization |
---|
28 | ! |
---|
29 | ! 3832 2019-03-28 13:16:58Z raasch |
---|
30 | ! loop reordering for performance optimization |
---|
31 | ! |
---|
32 | ! 3694 2019-01-23 17:01:49Z knoop |
---|
33 | ! OpenACC port for SPEC |
---|
34 | ! |
---|
35 | !------------------------------------------------------------------------------! |
---|
36 | ! Description: |
---|
37 | ! ------------ |
---|
38 | !> Resorting data for the transposition from x to y. The transposition itself |
---|
39 | !> is carried out in transpose_xy |
---|
40 | !------------------------------------------------------------------------------! |
---|
41 | SUBROUTINE resort_for_xy( f_in, f_inv ) |
---|
42 | |
---|
43 | |
---|
44 | USE indices, & |
---|
45 | ONLY: nx |
---|
46 | |
---|
47 | USE kinds |
---|
48 | |
---|
49 | USE transpose_indices, & |
---|
50 | ONLY: nyn_x, nys_x, nzb_x, nzt_x |
---|
51 | |
---|
52 | IMPLICIT NONE |
---|
53 | |
---|
54 | REAL(wp) :: f_in(0:nx,nys_x:nyn_x,nzb_x:nzt_x) !< |
---|
55 | REAL(wp) :: f_inv(nys_x:nyn_x,nzb_x:nzt_x,0:nx) !< |
---|
56 | |
---|
57 | |
---|
58 | INTEGER(iwp) :: i !< |
---|
59 | INTEGER(iwp) :: j !< |
---|
60 | INTEGER(iwp) :: k !< |
---|
61 | ! |
---|
62 | !-- Rearrange indices of input array in order to make data to be send |
---|
63 | !-- by MPI contiguous |
---|
64 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
65 | !$OMP DO |
---|
66 | #if __acc_fft_device |
---|
67 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
68 | !$ACC PRESENT(f_inv, f_in) |
---|
69 | #endif |
---|
70 | DO k = nzb_x, nzt_x |
---|
71 | DO j = nys_x, nyn_x |
---|
72 | DO i = 0, nx |
---|
73 | f_inv(j,k,i) = f_in(i,j,k) |
---|
74 | ENDDO |
---|
75 | ENDDO |
---|
76 | ENDDO |
---|
77 | !$OMP END PARALLEL |
---|
78 | |
---|
79 | END SUBROUTINE resort_for_xy |
---|
80 | |
---|
81 | |
---|
82 | !------------------------------------------------------------------------------! |
---|
83 | ! Description: |
---|
84 | ! ------------ |
---|
85 | !> Transposition of input array (f_in) from x to y. For the input array, all |
---|
86 | !> elements along x reside on the same PE, while after transposition, all |
---|
87 | !> elements along y reside on the same PE. |
---|
88 | !------------------------------------------------------------------------------! |
---|
89 | SUBROUTINE transpose_xy( f_inv, f_out ) |
---|
90 | |
---|
91 | |
---|
92 | USE cpulog, & |
---|
93 | ONLY: cpu_log, cpu_log_nowait, log_point_s |
---|
94 | |
---|
95 | USE indices, & |
---|
96 | ONLY: nx, ny |
---|
97 | |
---|
98 | USE kinds |
---|
99 | |
---|
100 | USE pegrid |
---|
101 | |
---|
102 | USE transpose_indices, & |
---|
103 | ONLY: nxl_y, nxr_y, nyn_x, nys_x, nzb_x, nzb_y, nzt_x, nzt_y |
---|
104 | |
---|
105 | IMPLICIT NONE |
---|
106 | |
---|
107 | INTEGER(iwp) :: i !< |
---|
108 | INTEGER(iwp) :: j !< |
---|
109 | INTEGER(iwp) :: k !< |
---|
110 | INTEGER(iwp) :: l !< |
---|
111 | INTEGER(iwp) :: ys !< |
---|
112 | |
---|
113 | REAL(wp) :: f_inv(nys_x:nyn_x,nzb_x:nzt_x,0:nx) !< |
---|
114 | REAL(wp) :: f_out(0:ny,nxl_y:nxr_y,nzb_y:nzt_y) !< |
---|
115 | |
---|
116 | REAL(wp), DIMENSION(nyn_x-nys_x+1,nzb_y:nzt_y,nxl_y:nxr_y,0:pdims(2)-1) :: work !< |
---|
117 | #if __acc_fft_device |
---|
118 | !$ACC DECLARE CREATE(work) |
---|
119 | #endif |
---|
120 | |
---|
121 | |
---|
122 | IF ( numprocs /= 1 ) THEN |
---|
123 | |
---|
124 | #if defined( __parallel ) |
---|
125 | ! |
---|
126 | !-- Transpose array |
---|
127 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start', cpu_log_nowait ) |
---|
128 | |
---|
129 | #if __acc_fft_device |
---|
130 | #ifndef __cuda_aware_mpi |
---|
131 | !$ACC UPDATE HOST(f_inv) |
---|
132 | #else |
---|
133 | !$ACC HOST_DATA USE_DEVICE(work, f_inv) |
---|
134 | #endif |
---|
135 | #endif |
---|
136 | |
---|
137 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
138 | CALL MPI_ALLTOALL( f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, & |
---|
139 | work(1,nzb_y,nxl_y,0), sendrecvcount_xy, MPI_REAL, & |
---|
140 | comm1dy, ierr ) |
---|
141 | |
---|
142 | #if __acc_fft_device |
---|
143 | #ifndef __cuda_aware_mpi |
---|
144 | !$ACC UPDATE DEVICE(work) |
---|
145 | #else |
---|
146 | !$ACC END HOST_DATA |
---|
147 | #endif |
---|
148 | #endif |
---|
149 | |
---|
150 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
151 | |
---|
152 | ! |
---|
153 | !-- Reorder transposed array |
---|
154 | !$OMP PARALLEL PRIVATE ( i, j, k, l, ys ) |
---|
155 | !$OMP DO |
---|
156 | DO l = 0, pdims(2) - 1 |
---|
157 | ys = 0 + l * ( nyn_x - nys_x + 1 ) |
---|
158 | #if __acc_fft_device |
---|
159 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
160 | !$ACC PRESENT(f_out, work) |
---|
161 | #endif |
---|
162 | DO i = nxl_y, nxr_y |
---|
163 | DO k = nzb_y, nzt_y |
---|
164 | DO j = ys, ys + nyn_x - nys_x |
---|
165 | f_out(j,i,k) = work(j-ys+1,k,i,l) |
---|
166 | ENDDO |
---|
167 | ENDDO |
---|
168 | ENDDO |
---|
169 | ENDDO |
---|
170 | !$OMP END PARALLEL |
---|
171 | #endif |
---|
172 | |
---|
173 | ELSE |
---|
174 | |
---|
175 | ! |
---|
176 | !-- Reorder transposed array |
---|
177 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
178 | !$OMP DO |
---|
179 | #if __acc_fft_device |
---|
180 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
181 | !$ACC PRESENT(f_out, f_inv) |
---|
182 | #endif |
---|
183 | DO k = nzb_y, nzt_y |
---|
184 | DO i = nxl_y, nxr_y |
---|
185 | DO j = 0, ny |
---|
186 | f_out(j,i,k) = f_inv(j,k,i) |
---|
187 | ENDDO |
---|
188 | ENDDO |
---|
189 | ENDDO |
---|
190 | !$OMP END PARALLEL |
---|
191 | |
---|
192 | ENDIF |
---|
193 | |
---|
194 | END SUBROUTINE transpose_xy |
---|
195 | |
---|
196 | |
---|
197 | !------------------------------------------------------------------------------! |
---|
198 | ! Description: |
---|
199 | ! ------------ |
---|
200 | !> Resorting data after the transposition from x to z. The transposition itself |
---|
201 | !> is carried out in transpose_xz |
---|
202 | !------------------------------------------------------------------------------! |
---|
203 | SUBROUTINE resort_for_xz( f_inv, f_out ) |
---|
204 | |
---|
205 | |
---|
206 | USE indices, & |
---|
207 | ONLY: nxl, nxr, nyn, nys, nz |
---|
208 | |
---|
209 | USE kinds |
---|
210 | |
---|
211 | IMPLICIT NONE |
---|
212 | |
---|
213 | REAL(wp) :: f_inv(nys:nyn,nxl:nxr,1:nz) !< |
---|
214 | REAL(wp) :: f_out(1:nz,nys:nyn,nxl:nxr) !< |
---|
215 | |
---|
216 | INTEGER(iwp) :: i !< |
---|
217 | INTEGER(iwp) :: j !< |
---|
218 | INTEGER(iwp) :: k !< |
---|
219 | ! |
---|
220 | !-- Rearrange indices of input array in order to make data to be send |
---|
221 | !-- by MPI contiguous. |
---|
222 | !-- In case of parallel fft/transposition, scattered store is faster in |
---|
223 | !-- backward direction!!! |
---|
224 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
225 | !$OMP DO |
---|
226 | #if __acc_fft_device |
---|
227 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
228 | !$ACC PRESENT(f_out, f_inv) |
---|
229 | #endif |
---|
230 | DO i = nxl, nxr |
---|
231 | DO j = nys, nyn |
---|
232 | DO k = 1, nz |
---|
233 | f_out(k,j,i) = f_inv(j,i,k) |
---|
234 | ENDDO |
---|
235 | ENDDO |
---|
236 | ENDDO |
---|
237 | !$OMP END PARALLEL |
---|
238 | |
---|
239 | END SUBROUTINE resort_for_xz |
---|
240 | |
---|
241 | |
---|
242 | !------------------------------------------------------------------------------! |
---|
243 | ! Description: |
---|
244 | ! ------------ |
---|
245 | !> Transposition of input array (f_in) from x to z. For the input array, all |
---|
246 | !> elements along x reside on the same PE, while after transposition, all |
---|
247 | !> elements along z reside on the same PE. |
---|
248 | !------------------------------------------------------------------------------! |
---|
249 | SUBROUTINE transpose_xz( f_in, f_inv ) |
---|
250 | |
---|
251 | |
---|
252 | USE cpulog, & |
---|
253 | ONLY: cpu_log, cpu_log_nowait, log_point_s |
---|
254 | |
---|
255 | USE indices, & |
---|
256 | ONLY: nnx, nx, nxl, nxr, nyn, nys, nz |
---|
257 | |
---|
258 | USE kinds |
---|
259 | |
---|
260 | USE pegrid |
---|
261 | |
---|
262 | USE transpose_indices, & |
---|
263 | ONLY: nyn_x, nys_x, nzb_x, nzt_x |
---|
264 | |
---|
265 | IMPLICIT NONE |
---|
266 | |
---|
267 | INTEGER(iwp) :: i !< |
---|
268 | INTEGER(iwp) :: j !< |
---|
269 | INTEGER(iwp) :: k !< |
---|
270 | INTEGER(iwp) :: l !< |
---|
271 | INTEGER(iwp) :: xs !< |
---|
272 | |
---|
273 | REAL(wp) :: f_in(0:nx,nys_x:nyn_x,nzb_x:nzt_x) !< |
---|
274 | REAL(wp) :: f_inv(nys:nyn,nxl:nxr,1:nz) !< |
---|
275 | |
---|
276 | REAL(wp), DIMENSION(nys_x:nyn_x,nnx,nzb_x:nzt_x,0:pdims(1)-1) :: work !< |
---|
277 | #if __acc_fft_device |
---|
278 | !$ACC DECLARE CREATE(work) |
---|
279 | #endif |
---|
280 | |
---|
281 | |
---|
282 | ! |
---|
283 | !-- If the PE grid is one-dimensional along y, the array has only to be |
---|
284 | !-- reordered locally and therefore no transposition has to be done. |
---|
285 | IF ( pdims(1) /= 1 ) THEN |
---|
286 | |
---|
287 | #if defined( __parallel ) |
---|
288 | ! |
---|
289 | !-- Reorder input array for transposition |
---|
290 | !$OMP PARALLEL PRIVATE ( i, j, k, l, xs ) |
---|
291 | !$OMP DO |
---|
292 | DO l = 0, pdims(1) - 1 |
---|
293 | xs = 0 + l * nnx |
---|
294 | #if __acc_fft_device |
---|
295 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
296 | !$ACC PRESENT(work, f_in) |
---|
297 | #endif |
---|
298 | DO k = nzb_x, nzt_x |
---|
299 | DO i = xs, xs + nnx - 1 |
---|
300 | DO j = nys_x, nyn_x |
---|
301 | work(j,i-xs+1,k,l) = f_in(i,j,k) |
---|
302 | ENDDO |
---|
303 | ENDDO |
---|
304 | ENDDO |
---|
305 | ENDDO |
---|
306 | !$OMP END PARALLEL |
---|
307 | |
---|
308 | ! |
---|
309 | !-- Transpose array |
---|
310 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start', cpu_log_nowait ) |
---|
311 | |
---|
312 | #if __acc_fft_device |
---|
313 | #ifndef __cuda_aware_mpi |
---|
314 | !$ACC UPDATE HOST(work) |
---|
315 | #else |
---|
316 | !$ACC HOST_DATA USE_DEVICE(work, f_inv) |
---|
317 | #endif |
---|
318 | #endif |
---|
319 | |
---|
320 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
321 | CALL MPI_ALLTOALL( work(nys_x,1,nzb_x,0), sendrecvcount_zx, MPI_REAL, & |
---|
322 | f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, & |
---|
323 | comm1dx, ierr ) |
---|
324 | |
---|
325 | #if __acc_fft_device |
---|
326 | #ifndef __cuda_aware_mpi |
---|
327 | !$ACC UPDATE DEVICE(f_inv) |
---|
328 | #else |
---|
329 | !$ACC END HOST_DATA |
---|
330 | #endif |
---|
331 | #endif |
---|
332 | |
---|
333 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
334 | #endif |
---|
335 | |
---|
336 | ELSE |
---|
337 | |
---|
338 | ! |
---|
339 | !-- Reorder the array in a way that the z index is in first position |
---|
340 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
341 | !$OMP DO |
---|
342 | #if __acc_fft_device |
---|
343 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
344 | !$ACC PRESENT(f_inv, f_in) |
---|
345 | #endif |
---|
346 | DO i = nxl, nxr |
---|
347 | DO j = nys, nyn |
---|
348 | DO k = 1, nz |
---|
349 | f_inv(j,i,k) = f_in(i,j,k) |
---|
350 | ENDDO |
---|
351 | ENDDO |
---|
352 | ENDDO |
---|
353 | !$OMP END PARALLEL |
---|
354 | |
---|
355 | ENDIF |
---|
356 | |
---|
357 | END SUBROUTINE transpose_xz |
---|
358 | |
---|
359 | |
---|
360 | !------------------------------------------------------------------------------! |
---|
361 | ! Description: |
---|
362 | ! ------------ |
---|
363 | !> Resorting data after the transposition from y to x. The transposition itself |
---|
364 | !> is carried out in transpose_yx |
---|
365 | !------------------------------------------------------------------------------! |
---|
366 | SUBROUTINE resort_for_yx( f_inv, f_out ) |
---|
367 | |
---|
368 | |
---|
369 | USE indices, & |
---|
370 | ONLY: nx |
---|
371 | |
---|
372 | USE kinds |
---|
373 | |
---|
374 | USE transpose_indices, & |
---|
375 | ONLY: nyn_x, nys_x, nzb_x, nzt_x |
---|
376 | |
---|
377 | IMPLICIT NONE |
---|
378 | |
---|
379 | REAL(wp) :: f_inv(nys_x:nyn_x,nzb_x:nzt_x,0:nx) !< |
---|
380 | REAL(wp) :: f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x) !< |
---|
381 | |
---|
382 | |
---|
383 | INTEGER(iwp) :: i !< |
---|
384 | INTEGER(iwp) :: j !< |
---|
385 | INTEGER(iwp) :: k !< |
---|
386 | ! |
---|
387 | !-- Rearrange indices of input array in order to make data to be send |
---|
388 | !-- by MPI contiguous |
---|
389 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
390 | !$OMP DO |
---|
391 | #if __acc_fft_device |
---|
392 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
393 | !$ACC PRESENT(f_out, f_inv) |
---|
394 | #endif |
---|
395 | DO k = nzb_x, nzt_x |
---|
396 | DO j = nys_x, nyn_x |
---|
397 | DO i = 0, nx |
---|
398 | f_out(i,j,k) = f_inv(j,k,i) |
---|
399 | ENDDO |
---|
400 | ENDDO |
---|
401 | ENDDO |
---|
402 | !$OMP END PARALLEL |
---|
403 | |
---|
404 | END SUBROUTINE resort_for_yx |
---|
405 | |
---|
406 | |
---|
407 | !------------------------------------------------------------------------------! |
---|
408 | ! Description: |
---|
409 | ! ------------ |
---|
410 | !> Transposition of input array (f_in) from y to x. For the input array, all |
---|
411 | !> elements along y reside on the same PE, while after transposition, all |
---|
412 | !> elements along x reside on the same PE. |
---|
413 | !------------------------------------------------------------------------------! |
---|
414 | SUBROUTINE transpose_yx( f_in, f_inv ) |
---|
415 | |
---|
416 | |
---|
417 | USE cpulog, & |
---|
418 | ONLY: cpu_log, cpu_log_nowait, log_point_s |
---|
419 | |
---|
420 | USE indices, & |
---|
421 | ONLY: nx, ny |
---|
422 | |
---|
423 | USE kinds |
---|
424 | |
---|
425 | USE pegrid |
---|
426 | |
---|
427 | USE transpose_indices, & |
---|
428 | ONLY: nxl_y, nxr_y, nyn_x, nys_x, nzb_x, nzb_y, nzt_x, nzt_y |
---|
429 | |
---|
430 | IMPLICIT NONE |
---|
431 | |
---|
432 | INTEGER(iwp) :: i !< |
---|
433 | INTEGER(iwp) :: j !< |
---|
434 | INTEGER(iwp) :: k !< |
---|
435 | INTEGER(iwp) :: l !< |
---|
436 | INTEGER(iwp) :: ys !< |
---|
437 | |
---|
438 | REAL(wp) :: f_in(0:ny,nxl_y:nxr_y,nzb_y:nzt_y) !< |
---|
439 | REAL(wp) :: f_inv(nys_x:nyn_x,nzb_x:nzt_x,0:nx) !< |
---|
440 | |
---|
441 | REAL(wp), DIMENSION(nyn_x-nys_x+1,nzb_y:nzt_y,nxl_y:nxr_y,0:pdims(2)-1) :: work !< |
---|
442 | #if __acc_fft_device |
---|
443 | !$ACC DECLARE CREATE(work) |
---|
444 | #endif |
---|
445 | |
---|
446 | |
---|
447 | IF ( numprocs /= 1 ) THEN |
---|
448 | |
---|
449 | #if defined( __parallel ) |
---|
450 | ! |
---|
451 | !-- Reorder input array for transposition |
---|
452 | !$OMP PARALLEL PRIVATE ( i, j, k, l, ys ) |
---|
453 | !$OMP DO |
---|
454 | DO l = 0, pdims(2) - 1 |
---|
455 | ys = 0 + l * ( nyn_x - nys_x + 1 ) |
---|
456 | #if __acc_fft_device |
---|
457 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
458 | !$ACC PRESENT(work, f_in) |
---|
459 | #endif |
---|
460 | DO i = nxl_y, nxr_y |
---|
461 | DO k = nzb_y, nzt_y |
---|
462 | DO j = ys, ys + nyn_x - nys_x |
---|
463 | work(j-ys+1,k,i,l) = f_in(j,i,k) |
---|
464 | ENDDO |
---|
465 | ENDDO |
---|
466 | ENDDO |
---|
467 | ENDDO |
---|
468 | !$OMP END PARALLEL |
---|
469 | |
---|
470 | ! |
---|
471 | !-- Transpose array |
---|
472 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start', cpu_log_nowait ) |
---|
473 | |
---|
474 | #if __acc_fft_device |
---|
475 | #ifndef __cuda_aware_mpi |
---|
476 | !$ACC UPDATE HOST(work) |
---|
477 | #else |
---|
478 | !$ACC HOST_DATA USE_DEVICE(work, f_inv) |
---|
479 | #endif |
---|
480 | #endif |
---|
481 | |
---|
482 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
483 | CALL MPI_ALLTOALL( work(1,nzb_y,nxl_y,0), sendrecvcount_xy, MPI_REAL, & |
---|
484 | f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, & |
---|
485 | comm1dy, ierr ) |
---|
486 | |
---|
487 | #if __acc_fft_device |
---|
488 | #ifndef __cuda_aware_mpi |
---|
489 | !$ACC UPDATE DEVICE(f_inv) |
---|
490 | #else |
---|
491 | !$ACC END HOST_DATA |
---|
492 | #endif |
---|
493 | #endif |
---|
494 | |
---|
495 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
496 | #endif |
---|
497 | |
---|
498 | ELSE |
---|
499 | |
---|
500 | ! |
---|
501 | !-- Reorder array f_in the same way as ALLTOALL did it |
---|
502 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
503 | !$OMP DO |
---|
504 | #if __acc_fft_device |
---|
505 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
506 | !$ACC PRESENT(f_inv, f_in) |
---|
507 | #endif |
---|
508 | DO i = nxl_y, nxr_y |
---|
509 | DO k = nzb_y, nzt_y |
---|
510 | DO j = 0, ny |
---|
511 | f_inv(j,k,i) = f_in(j,i,k) |
---|
512 | ENDDO |
---|
513 | ENDDO |
---|
514 | ENDDO |
---|
515 | !$OMP END PARALLEL |
---|
516 | |
---|
517 | ENDIF |
---|
518 | |
---|
519 | END SUBROUTINE transpose_yx |
---|
520 | |
---|
521 | |
---|
522 | !------------------------------------------------------------------------------! |
---|
523 | ! Description: |
---|
524 | ! ------------ |
---|
525 | !> Transposition of input array (f_in) from y to x. For the input array, all |
---|
526 | !> elements along y reside on the same PE, while after transposition, all |
---|
527 | !> elements along x reside on the same PE. |
---|
528 | !> This is a direct transposition for arrays with indices in regular order |
---|
529 | !> (k,j,i) (cf. transpose_yx). |
---|
530 | !------------------------------------------------------------------------------! |
---|
531 | SUBROUTINE transpose_yxd( f_in, f_out ) |
---|
532 | |
---|
533 | |
---|
534 | USE cpulog, & |
---|
535 | ONLY: cpu_log, log_point_s |
---|
536 | |
---|
537 | USE indices, & |
---|
538 | ONLY: nnx, nny, nnz, nx, nxl, nxr, nyn, nys, nz |
---|
539 | |
---|
540 | USE kinds |
---|
541 | |
---|
542 | USE pegrid |
---|
543 | |
---|
544 | USE transpose_indices, & |
---|
545 | ONLY: nyn_x, nys_x, nzb_x, nzt_x |
---|
546 | |
---|
547 | IMPLICIT NONE |
---|
548 | |
---|
549 | INTEGER(iwp) :: i !< |
---|
550 | INTEGER(iwp) :: j !< |
---|
551 | INTEGER(iwp) :: k !< |
---|
552 | INTEGER(iwp) :: l !< |
---|
553 | INTEGER(iwp) :: m !< |
---|
554 | INTEGER(iwp) :: xs !< |
---|
555 | |
---|
556 | REAL(wp) :: f_in(1:nz,nys:nyn,nxl:nxr) !< |
---|
557 | REAL(wp) :: f_inv(nxl:nxr,1:nz,nys:nyn) !< |
---|
558 | REAL(wp) :: f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x) !< |
---|
559 | REAL(wp) :: work(nnx*nny*nnz) !< |
---|
560 | #if defined( __parallel ) |
---|
561 | |
---|
562 | ! |
---|
563 | !-- Rearrange indices of input array in order to make data to be send |
---|
564 | !-- by MPI contiguous |
---|
565 | DO k = 1, nz |
---|
566 | DO j = nys, nyn |
---|
567 | DO i = nxl, nxr |
---|
568 | f_inv(i,k,j) = f_in(k,j,i) |
---|
569 | ENDDO |
---|
570 | ENDDO |
---|
571 | ENDDO |
---|
572 | |
---|
573 | ! |
---|
574 | !-- Transpose array |
---|
575 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
576 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
577 | CALL MPI_ALLTOALL( f_inv(nxl,1,nys), sendrecvcount_xy, MPI_REAL, & |
---|
578 | work(1), sendrecvcount_xy, MPI_REAL, & |
---|
579 | comm1dx, ierr ) |
---|
580 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
581 | |
---|
582 | ! |
---|
583 | !-- Reorder transposed array |
---|
584 | m = 0 |
---|
585 | DO l = 0, pdims(1) - 1 |
---|
586 | xs = 0 + l * nnx |
---|
587 | DO j = nys_x, nyn_x |
---|
588 | DO k = 1, nz |
---|
589 | DO i = xs, xs + nnx - 1 |
---|
590 | m = m + 1 |
---|
591 | f_out(i,j,k) = work(m) |
---|
592 | ENDDO |
---|
593 | ENDDO |
---|
594 | ENDDO |
---|
595 | ENDDO |
---|
596 | |
---|
597 | #endif |
---|
598 | |
---|
599 | END SUBROUTINE transpose_yxd |
---|
600 | |
---|
601 | |
---|
602 | !------------------------------------------------------------------------------! |
---|
603 | ! Description: |
---|
604 | ! ------------ |
---|
605 | !> Resorting data for the transposition from y to z. The transposition itself |
---|
606 | !> is carried out in transpose_yz |
---|
607 | !------------------------------------------------------------------------------! |
---|
608 | SUBROUTINE resort_for_yz( f_in, f_inv ) |
---|
609 | |
---|
610 | |
---|
611 | USE indices, & |
---|
612 | ONLY: ny |
---|
613 | |
---|
614 | USE kinds |
---|
615 | |
---|
616 | USE transpose_indices, & |
---|
617 | ONLY: nxl_y, nxr_y, nzb_y, nzt_y |
---|
618 | |
---|
619 | IMPLICIT NONE |
---|
620 | |
---|
621 | REAL(wp) :: f_in(0:ny,nxl_y:nxr_y,nzb_y:nzt_y) !< |
---|
622 | REAL(wp) :: f_inv(nxl_y:nxr_y,nzb_y:nzt_y,0:ny) !< |
---|
623 | |
---|
624 | INTEGER(iwp) :: i !< |
---|
625 | INTEGER(iwp) :: j !< |
---|
626 | INTEGER(iwp) :: k !< |
---|
627 | |
---|
628 | ! |
---|
629 | !-- Rearrange indices of input array in order to make data to be send |
---|
630 | !-- by MPI contiguous |
---|
631 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
632 | !$OMP DO |
---|
633 | #if __acc_fft_device |
---|
634 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
635 | !$ACC PRESENT(f_inv, f_in) |
---|
636 | #endif |
---|
637 | DO k = nzb_y, nzt_y |
---|
638 | DO i = nxl_y, nxr_y |
---|
639 | DO j = 0, ny |
---|
640 | f_inv(i,k,j) = f_in(j,i,k) |
---|
641 | ENDDO |
---|
642 | ENDDO |
---|
643 | ENDDO |
---|
644 | !$OMP END PARALLEL |
---|
645 | |
---|
646 | END SUBROUTINE resort_for_yz |
---|
647 | |
---|
648 | |
---|
649 | !------------------------------------------------------------------------------! |
---|
650 | ! Description: |
---|
651 | ! ------------ |
---|
652 | !> Transposition of input array (f_in) from y to z. For the input array, all |
---|
653 | !> elements along y reside on the same PE, while after transposition, all |
---|
654 | !> elements along z reside on the same PE. |
---|
655 | !------------------------------------------------------------------------------! |
---|
656 | SUBROUTINE transpose_yz( f_inv, f_out ) |
---|
657 | |
---|
658 | |
---|
659 | USE cpulog, & |
---|
660 | ONLY: cpu_log, cpu_log_nowait, log_point_s |
---|
661 | |
---|
662 | USE indices, & |
---|
663 | ONLY: ny, nz |
---|
664 | |
---|
665 | USE kinds |
---|
666 | |
---|
667 | USE pegrid |
---|
668 | |
---|
669 | USE transpose_indices, & |
---|
670 | ONLY: nxl_y, nxl_z, nxr_y, nxr_z, nyn_z, nys_z, nzb_y, nzt_y |
---|
671 | |
---|
672 | IMPLICIT NONE |
---|
673 | |
---|
674 | INTEGER(iwp) :: i !< |
---|
675 | INTEGER(iwp) :: j !< |
---|
676 | INTEGER(iwp) :: k !< |
---|
677 | INTEGER(iwp) :: l !< |
---|
678 | INTEGER(iwp) :: zs !< |
---|
679 | |
---|
680 | REAL(wp) :: f_inv(nxl_y:nxr_y,nzb_y:nzt_y,0:ny) !< |
---|
681 | REAL(wp) :: f_out(nxl_z:nxr_z,nys_z:nyn_z,1:nz) !< |
---|
682 | |
---|
683 | REAL(wp), DIMENSION(nxl_z:nxr_z,nzt_y-nzb_y+1,nys_z:nyn_z,0:pdims(1)-1) :: work !< |
---|
684 | #if __acc_fft_device |
---|
685 | !$ACC DECLARE CREATE(work) |
---|
686 | #endif |
---|
687 | |
---|
688 | |
---|
689 | ! |
---|
690 | !-- If the PE grid is one-dimensional along y, only local reordering |
---|
691 | !-- of the data is necessary and no transposition has to be done. |
---|
692 | IF ( pdims(1) == 1 ) THEN |
---|
693 | |
---|
694 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
695 | !$OMP DO |
---|
696 | #if __acc_fft_device |
---|
697 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
698 | !$ACC PRESENT(f_out, f_inv) |
---|
699 | #endif |
---|
700 | DO j = 0, ny |
---|
701 | DO k = nzb_y, nzt_y |
---|
702 | DO i = nxl_y, nxr_y |
---|
703 | f_out(i,j,k) = f_inv(i,k,j) |
---|
704 | ENDDO |
---|
705 | ENDDO |
---|
706 | ENDDO |
---|
707 | !$OMP END PARALLEL |
---|
708 | |
---|
709 | ELSE |
---|
710 | |
---|
711 | #if defined( __parallel ) |
---|
712 | ! |
---|
713 | !-- Transpose array |
---|
714 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start', cpu_log_nowait ) |
---|
715 | |
---|
716 | #if __acc_fft_device |
---|
717 | #ifndef __cuda_aware_mpi |
---|
718 | !$ACC UPDATE HOST(f_inv) |
---|
719 | #else |
---|
720 | !$ACC HOST_DATA USE_DEVICE(work, f_inv) |
---|
721 | #endif |
---|
722 | #endif |
---|
723 | |
---|
724 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
725 | CALL MPI_ALLTOALL( f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, & |
---|
726 | work(nxl_z,1,nys_z,0), sendrecvcount_yz, MPI_REAL, & |
---|
727 | comm1dx, ierr ) |
---|
728 | |
---|
729 | #if __acc_fft_device |
---|
730 | #ifndef __cuda_aware_mpi |
---|
731 | !$ACC UPDATE DEVICE(work) |
---|
732 | #else |
---|
733 | !$ACC END HOST_DATA |
---|
734 | #endif |
---|
735 | #endif |
---|
736 | |
---|
737 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
738 | |
---|
739 | ! |
---|
740 | !-- Reorder transposed array |
---|
741 | !$OMP PARALLEL PRIVATE ( i, j, k, l, zs ) |
---|
742 | !$OMP DO |
---|
743 | DO l = 0, pdims(1) - 1 |
---|
744 | zs = 1 + l * ( nzt_y - nzb_y + 1 ) |
---|
745 | #if __acc_fft_device |
---|
746 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
747 | !$ACC PRESENT(f_out, work) |
---|
748 | #endif |
---|
749 | DO j = nys_z, nyn_z |
---|
750 | DO k = zs, zs + nzt_y - nzb_y |
---|
751 | DO i = nxl_z, nxr_z |
---|
752 | f_out(i,j,k) = work(i,k-zs+1,j,l) |
---|
753 | ENDDO |
---|
754 | ENDDO |
---|
755 | ENDDO |
---|
756 | ENDDO |
---|
757 | !$OMP END PARALLEL |
---|
758 | #endif |
---|
759 | |
---|
760 | ENDIF |
---|
761 | |
---|
762 | END SUBROUTINE transpose_yz |
---|
763 | |
---|
764 | |
---|
765 | !------------------------------------------------------------------------------! |
---|
766 | ! Description: |
---|
767 | ! ------------ |
---|
768 | !> Resorting data for the transposition from z to x. The transposition itself |
---|
769 | !> is carried out in transpose_zx |
---|
770 | !------------------------------------------------------------------------------! |
---|
771 | SUBROUTINE resort_for_zx( f_in, f_inv ) |
---|
772 | |
---|
773 | |
---|
774 | USE indices, & |
---|
775 | ONLY: nxl, nxr, nyn, nys, nz |
---|
776 | |
---|
777 | USE kinds |
---|
778 | |
---|
779 | IMPLICIT NONE |
---|
780 | |
---|
781 | REAL(wp) :: f_in(1:nz,nys:nyn,nxl:nxr) !< |
---|
782 | REAL(wp) :: f_inv(nys:nyn,nxl:nxr,1:nz) !< |
---|
783 | |
---|
784 | INTEGER(iwp) :: i !< |
---|
785 | INTEGER(iwp) :: j !< |
---|
786 | INTEGER(iwp) :: k !< |
---|
787 | |
---|
788 | ! |
---|
789 | !-- Rearrange indices of input array in order to make data to be send |
---|
790 | !-- by MPI contiguous |
---|
791 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
792 | !$OMP DO |
---|
793 | #if __acc_fft_device |
---|
794 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
795 | !$ACC PRESENT(f_in, f_inv) |
---|
796 | #endif |
---|
797 | DO i = nxl, nxr |
---|
798 | DO j = nys, nyn |
---|
799 | DO k = 1,nz |
---|
800 | f_inv(j,i,k) = f_in(k,j,i) |
---|
801 | ENDDO |
---|
802 | ENDDO |
---|
803 | ENDDO |
---|
804 | !$OMP END PARALLEL |
---|
805 | |
---|
806 | END SUBROUTINE resort_for_zx |
---|
807 | |
---|
808 | |
---|
809 | !------------------------------------------------------------------------------! |
---|
810 | ! Description: |
---|
811 | ! ------------ |
---|
812 | !> Transposition of input array (f_in) from z to x. For the input array, all |
---|
813 | !> elements along z reside on the same PE, while after transposition, all |
---|
814 | !> elements along x reside on the same PE. |
---|
815 | !------------------------------------------------------------------------------! |
---|
816 | SUBROUTINE transpose_zx( f_inv, f_out ) |
---|
817 | |
---|
818 | |
---|
819 | USE cpulog, & |
---|
820 | ONLY: cpu_log, cpu_log_nowait, log_point_s |
---|
821 | |
---|
822 | USE indices, & |
---|
823 | ONLY: nnx, nx, nxl, nxr, nyn, nys, nz |
---|
824 | |
---|
825 | USE kinds |
---|
826 | |
---|
827 | USE pegrid |
---|
828 | |
---|
829 | USE transpose_indices, & |
---|
830 | ONLY: nyn_x, nys_x, nzb_x, nzt_x |
---|
831 | |
---|
832 | IMPLICIT NONE |
---|
833 | |
---|
834 | INTEGER(iwp) :: i !< |
---|
835 | INTEGER(iwp) :: j !< |
---|
836 | INTEGER(iwp) :: k !< |
---|
837 | INTEGER(iwp) :: l !< |
---|
838 | INTEGER(iwp) :: xs !< |
---|
839 | |
---|
840 | REAL(wp) :: f_inv(nys:nyn,nxl:nxr,1:nz) !< |
---|
841 | REAL(wp) :: f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x) !< |
---|
842 | |
---|
843 | REAL(wp), DIMENSION(nys_x:nyn_x,nnx,nzb_x:nzt_x,0:pdims(1)-1) :: work !< |
---|
844 | #if __acc_fft_device |
---|
845 | !$ACC DECLARE CREATE(work) |
---|
846 | #endif |
---|
847 | |
---|
848 | |
---|
849 | ! |
---|
850 | !-- If the PE grid is one-dimensional along y, only local reordering |
---|
851 | !-- of the data is necessary and no transposition has to be done. |
---|
852 | IF ( pdims(1) == 1 ) THEN |
---|
853 | |
---|
854 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
855 | !$OMP DO |
---|
856 | #if __acc_fft_device |
---|
857 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
858 | !$ACC PRESENT(f_out, f_inv) |
---|
859 | #endif |
---|
860 | DO k = 1, nz |
---|
861 | DO i = nxl, nxr |
---|
862 | DO j = nys, nyn |
---|
863 | f_out(i,j,k) = f_inv(j,i,k) |
---|
864 | ENDDO |
---|
865 | ENDDO |
---|
866 | ENDDO |
---|
867 | !$OMP END PARALLEL |
---|
868 | |
---|
869 | ELSE |
---|
870 | |
---|
871 | #if defined( __parallel ) |
---|
872 | ! |
---|
873 | !-- Transpose array |
---|
874 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start', cpu_log_nowait ) |
---|
875 | |
---|
876 | #if __acc_fft_device |
---|
877 | #ifndef __cuda_aware_mpi |
---|
878 | !$ACC UPDATE HOST(f_inv) |
---|
879 | #else |
---|
880 | !$ACC HOST_DATA USE_DEVICE(work, f_inv) |
---|
881 | #endif |
---|
882 | #endif |
---|
883 | |
---|
884 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
885 | CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, & |
---|
886 | work(nys_x,1,nzb_x,0), sendrecvcount_zx, MPI_REAL, & |
---|
887 | comm1dx, ierr ) |
---|
888 | |
---|
889 | #if __acc_fft_device |
---|
890 | #ifndef __cuda_aware_mpi |
---|
891 | !$ACC UPDATE DEVICE(work) |
---|
892 | #else |
---|
893 | !$ACC END HOST_DATA |
---|
894 | #endif |
---|
895 | #endif |
---|
896 | |
---|
897 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
898 | |
---|
899 | ! |
---|
900 | !-- Reorder transposed array |
---|
901 | !$OMP PARALLEL PRIVATE ( i, j, k, l, xs ) |
---|
902 | !$OMP DO |
---|
903 | DO l = 0, pdims(1) - 1 |
---|
904 | xs = 0 + l * nnx |
---|
905 | #if __acc_fft_device |
---|
906 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
907 | !$ACC PRESENT(f_out, work) |
---|
908 | #endif |
---|
909 | DO k = nzb_x, nzt_x |
---|
910 | DO i = xs, xs + nnx - 1 |
---|
911 | DO j = nys_x, nyn_x |
---|
912 | f_out(i,j,k) = work(j,i-xs+1,k,l) |
---|
913 | ENDDO |
---|
914 | ENDDO |
---|
915 | ENDDO |
---|
916 | ENDDO |
---|
917 | !$OMP END PARALLEL |
---|
918 | #endif |
---|
919 | |
---|
920 | ENDIF |
---|
921 | |
---|
922 | END SUBROUTINE transpose_zx |
---|
923 | |
---|
924 | |
---|
925 | !------------------------------------------------------------------------------! |
---|
926 | ! Description: |
---|
927 | ! ------------ |
---|
928 | !> Resorting data after the transposition from z to y. The transposition itself |
---|
929 | !> is carried out in transpose_zy |
---|
930 | !------------------------------------------------------------------------------! |
---|
931 | SUBROUTINE resort_for_zy( f_inv, f_out ) |
---|
932 | |
---|
933 | |
---|
934 | USE indices, & |
---|
935 | ONLY: ny |
---|
936 | |
---|
937 | USE kinds |
---|
938 | |
---|
939 | USE transpose_indices, & |
---|
940 | ONLY: nxl_y, nxr_y, nzb_y, nzt_y |
---|
941 | |
---|
942 | IMPLICIT NONE |
---|
943 | |
---|
944 | REAL(wp) :: f_inv(nxl_y:nxr_y,nzb_y:nzt_y,0:ny) !< |
---|
945 | REAL(wp) :: f_out(0:ny,nxl_y:nxr_y,nzb_y:nzt_y) !< |
---|
946 | |
---|
947 | |
---|
948 | INTEGER(iwp) :: i !< |
---|
949 | INTEGER(iwp) :: j !< |
---|
950 | INTEGER(iwp) :: k !< |
---|
951 | |
---|
952 | ! |
---|
953 | !-- Rearrange indices of input array in order to make data to be send |
---|
954 | !-- by MPI contiguous |
---|
955 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
956 | !$OMP DO |
---|
957 | #if __acc_fft_device |
---|
958 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
959 | !$ACC PRESENT(f_out, f_inv) |
---|
960 | #endif |
---|
961 | DO k = nzb_y, nzt_y |
---|
962 | DO i = nxl_y, nxr_y |
---|
963 | DO j = 0, ny |
---|
964 | f_out(j,i,k) = f_inv(i,k,j) |
---|
965 | ENDDO |
---|
966 | ENDDO |
---|
967 | ENDDO |
---|
968 | !$OMP END PARALLEL |
---|
969 | |
---|
970 | END SUBROUTINE resort_for_zy |
---|
971 | |
---|
972 | |
---|
973 | !------------------------------------------------------------------------------! |
---|
974 | ! Description:cpu_log_nowait |
---|
975 | ! ------------ |
---|
976 | !> Transposition of input array (f_in) from z to y. For the input array, all |
---|
977 | !> elements along z reside on the same PE, while after transposition, all |
---|
978 | !> elements along y reside on the same PE. |
---|
979 | !------------------------------------------------------------------------------! |
---|
980 | SUBROUTINE transpose_zy( f_in, f_inv ) |
---|
981 | |
---|
982 | |
---|
983 | USE cpulog, & |
---|
984 | ONLY: cpu_log, cpu_log_nowait, log_point_s |
---|
985 | |
---|
986 | USE indices, & |
---|
987 | ONLY: ny, nz |
---|
988 | |
---|
989 | USE kinds |
---|
990 | |
---|
991 | USE pegrid |
---|
992 | |
---|
993 | USE transpose_indices, & |
---|
994 | ONLY: nxl_y, nxl_z, nxr_y, nxr_z, nyn_z, nys_z, nzb_y, nzt_y |
---|
995 | |
---|
996 | IMPLICIT NONE |
---|
997 | |
---|
998 | INTEGER(iwp) :: i !< |
---|
999 | INTEGER(iwp) :: j !< |
---|
1000 | INTEGER(iwp) :: k !< |
---|
1001 | INTEGER(iwp) :: l !< |
---|
1002 | INTEGER(iwp) :: zs !< |
---|
1003 | |
---|
1004 | REAL(wp) :: f_in(nxl_z:nxr_z,nys_z:nyn_z,1:nz) !< |
---|
1005 | REAL(wp) :: f_inv(nxl_y:nxr_y,nzb_y:nzt_y,0:ny) !< |
---|
1006 | |
---|
1007 | REAL(wp), DIMENSION(nxl_z:nxr_z,nzt_y-nzb_y+1,nys_z:nyn_z,0:pdims(1)-1) :: work !< |
---|
1008 | #if __acc_fft_device |
---|
1009 | !$ACC DECLARE CREATE(work) |
---|
1010 | #endif |
---|
1011 | |
---|
1012 | ! |
---|
1013 | !-- If the PE grid is one-dimensional along y, the array has only to be |
---|
1014 | !-- reordered locally and therefore no transposition has to be done. |
---|
1015 | IF ( pdims(1) /= 1 ) THEN |
---|
1016 | |
---|
1017 | #if defined( __parallel ) |
---|
1018 | ! |
---|
1019 | !-- Reorder input array for transposition |
---|
1020 | !$OMP PARALLEL PRIVATE ( i, j, k, l, zs ) |
---|
1021 | !$OMP DO |
---|
1022 | DO l = 0, pdims(1) - 1 |
---|
1023 | zs = 1 + l * ( nzt_y - nzb_y + 1 ) |
---|
1024 | #if __acc_fft_device |
---|
1025 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
1026 | !$ACC PRESENT(work, f_in) |
---|
1027 | #endif |
---|
1028 | DO j = nys_z, nyn_z |
---|
1029 | DO k = zs, zs + nzt_y - nzb_y |
---|
1030 | DO i = nxl_z, nxr_z |
---|
1031 | work(i,k-zs+1,j,l) = f_in(i,j,k) |
---|
1032 | ENDDO |
---|
1033 | ENDDO |
---|
1034 | ENDDO |
---|
1035 | ENDDO |
---|
1036 | !$OMP END PARALLEL |
---|
1037 | |
---|
1038 | ! |
---|
1039 | !-- Transpose array |
---|
1040 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start', cpu_log_nowait ) |
---|
1041 | |
---|
1042 | #if __acc_fft_device |
---|
1043 | #ifndef __cuda_aware_mpi |
---|
1044 | !$ACC UPDATE HOST(work) |
---|
1045 | #else |
---|
1046 | !$ACC HOST_DATA USE_DEVICE(work, f_inv) |
---|
1047 | #endif |
---|
1048 | #endif |
---|
1049 | |
---|
1050 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
1051 | CALL MPI_ALLTOALL( work(nxl_z,1,nys_z,0), sendrecvcount_yz, MPI_REAL, & |
---|
1052 | f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, & |
---|
1053 | comm1dx, ierr ) |
---|
1054 | |
---|
1055 | #if __acc_fft_device |
---|
1056 | #ifndef __cuda_aware_mpi |
---|
1057 | !$ACC UPDATE DEVICE(f_inv) |
---|
1058 | #else |
---|
1059 | !$ACC END HOST_DATA |
---|
1060 | #endif |
---|
1061 | #endif |
---|
1062 | |
---|
1063 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
1064 | #endif |
---|
1065 | |
---|
1066 | ELSE |
---|
1067 | ! |
---|
1068 | !-- Reorder the array in the same way like ALLTOALL did it |
---|
1069 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
---|
1070 | !$OMP DO |
---|
1071 | #if __acc_fft_device |
---|
1072 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
1073 | !$ACC PRESENT(f_inv, f_in) |
---|
1074 | #endif |
---|
1075 | DO k = nzb_y, nzt_y |
---|
1076 | DO j = 0, ny |
---|
1077 | DO i = nxl_y, nxr_y |
---|
1078 | f_inv(i,k,j) = f_in(i,j,k) |
---|
1079 | ENDDO |
---|
1080 | ENDDO |
---|
1081 | ENDDO |
---|
1082 | !$OMP END PARALLEL |
---|
1083 | |
---|
1084 | ENDIF |
---|
1085 | |
---|
1086 | END SUBROUTINE transpose_zy |
---|
1087 | |
---|
1088 | |
---|
1089 | !------------------------------------------------------------------------------! |
---|
1090 | ! Description: |
---|
1091 | ! ------------ |
---|
1092 | !> Transposition of input array (f_in) from z to y. For the input array, all |
---|
1093 | !> elements along z reside on the same PE, while after transposition, all |
---|
1094 | !> elements along y reside on the same PE. |
---|
1095 | !> This is a direct transposition for arrays with indices in regular order |
---|
1096 | !> (k,j,i) (cf. transpose_zy). |
---|
1097 | !------------------------------------------------------------------------------! |
---|
1098 | SUBROUTINE transpose_zyd( f_in, f_out ) |
---|
1099 | |
---|
1100 | |
---|
1101 | USE cpulog, & |
---|
1102 | ONLY: cpu_log, log_point_s |
---|
1103 | |
---|
1104 | USE indices, & |
---|
1105 | ONLY: nnx, nny, nnz, nxl, nxr, nyn, nys, ny, nz |
---|
1106 | |
---|
1107 | USE kinds |
---|
1108 | |
---|
1109 | USE pegrid |
---|
1110 | |
---|
1111 | USE transpose_indices, & |
---|
1112 | ONLY: nxl_yd, nxr_yd, nzb_yd, nzt_yd |
---|
1113 | |
---|
1114 | IMPLICIT NONE |
---|
1115 | |
---|
1116 | INTEGER(iwp) :: i !< |
---|
1117 | INTEGER(iwp) :: j !< |
---|
1118 | INTEGER(iwp) :: k !< |
---|
1119 | INTEGER(iwp) :: l !< |
---|
1120 | INTEGER(iwp) :: m !< |
---|
1121 | INTEGER(iwp) :: ys !< |
---|
1122 | |
---|
1123 | REAL(wp) :: f_in(1:nz,nys:nyn,nxl:nxr) !< |
---|
1124 | REAL(wp) :: f_inv(nys:nyn,nxl:nxr,1:nz) !< |
---|
1125 | REAL(wp) :: f_out(0:ny,nxl_yd:nxr_yd,nzb_yd:nzt_yd) !< |
---|
1126 | REAL(wp) :: work(nnx*nny*nnz) !< |
---|
1127 | |
---|
1128 | #if defined( __parallel ) |
---|
1129 | |
---|
1130 | ! |
---|
1131 | !-- Rearrange indices of input array in order to make data to be send |
---|
1132 | !-- by MPI contiguous |
---|
1133 | DO i = nxl, nxr |
---|
1134 | DO j = nys, nyn |
---|
1135 | DO k = 1, nz |
---|
1136 | f_inv(j,i,k) = f_in(k,j,i) |
---|
1137 | ENDDO |
---|
1138 | ENDDO |
---|
1139 | ENDDO |
---|
1140 | |
---|
1141 | ! |
---|
1142 | !-- Move data to different array, because memory location of work1 is |
---|
1143 | !-- needed further below (work1 = work2). |
---|
1144 | !-- If the PE grid is one-dimensional along x, only local reordering |
---|
1145 | !-- of the data is necessary and no transposition has to be done. |
---|
1146 | IF ( pdims(2) == 1 ) THEN |
---|
1147 | DO k = 1, nz |
---|
1148 | DO i = nxl, nxr |
---|
1149 | DO j = nys, nyn |
---|
1150 | f_out(j,i,k) = f_inv(j,i,k) |
---|
1151 | ENDDO |
---|
1152 | ENDDO |
---|
1153 | ENDDO |
---|
1154 | RETURN |
---|
1155 | ENDIF |
---|
1156 | |
---|
1157 | ! |
---|
1158 | !-- Transpose array |
---|
1159 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
1160 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
1161 | CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zyd, MPI_REAL, & |
---|
1162 | work(1), sendrecvcount_zyd, MPI_REAL, & |
---|
1163 | comm1dy, ierr ) |
---|
1164 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
1165 | |
---|
1166 | ! |
---|
1167 | !-- Reorder transposed array |
---|
1168 | m = 0 |
---|
1169 | DO l = 0, pdims(2) - 1 |
---|
1170 | ys = 0 + l * nny |
---|
1171 | DO k = nzb_yd, nzt_yd |
---|
1172 | DO i = nxl_yd, nxr_yd |
---|
1173 | DO j = ys, ys + nny - 1 |
---|
1174 | m = m + 1 |
---|
1175 | f_out(j,i,k) = work(m) |
---|
1176 | ENDDO |
---|
1177 | ENDDO |
---|
1178 | ENDDO |
---|
1179 | ENDDO |
---|
1180 | |
---|
1181 | #endif |
---|
1182 | |
---|
1183 | END SUBROUTINE transpose_zyd |
---|