1 | MODULE poisfft_hybrid_mod |
---|
2 | !------------------------------------------------------------------------------ |
---|
3 | ! |
---|
4 | ! Current revisions: |
---|
5 | ! ----------------- |
---|
6 | ! New cpp directive "__check" implemented which is used by check_namelist_files |
---|
7 | ! (most of the code is unneeded by check_namelist_files). |
---|
8 | ! |
---|
9 | ! Former revisions: |
---|
10 | ! ----------------- |
---|
11 | ! $Id: poisfft_hybrid.f90 807 2012-01-25 11:53:51Z maronga $ |
---|
12 | ! |
---|
13 | ! 667 2010-12-23 12:06:00Z suehring/gryschka |
---|
14 | ! ddzu replaced by ddzu_pres due to changes in zu(0) |
---|
15 | ! |
---|
16 | ! 415 2009-12-15 10:26:23Z raasch |
---|
17 | ! Dimension of array stat in cascade change to prevent type problems with___ |
---|
18 | ! mpi2 libraries |
---|
19 | ! |
---|
20 | ! 274 2009-03-26 15:11:21Z heinze |
---|
21 | ! Output of messages replaced by message handling routine. |
---|
22 | ! |
---|
23 | ! Feb. 2007 |
---|
24 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
25 | ! |
---|
26 | ! Revision 1.11 2004/04/30 12:43:14 raasch |
---|
27 | ! Renaming of fft routines, additional argument in calls of fft_y_m |
---|
28 | ! |
---|
29 | ! Revision 1.2 2002/12/19 16:08:31 raasch |
---|
30 | ! Preprocessor directive KKMP introduced (OMP does NOT work), |
---|
31 | ! array tri will be a shared array in OpenMP loop, to get better cache |
---|
32 | ! utilization, the i index (x-direction) will be executed in stride |
---|
33 | ! "istride" as outer loop and in a shorter inner loop, |
---|
34 | ! overlapping of computation and communication realized by new routine |
---|
35 | ! poisfft_hybrid_nodes, name of old routine poisfft_hybrid changed to |
---|
36 | ! poisfft_hybrid_omp, STOP statement replaced by call of subroutine local_stop |
---|
37 | ! |
---|
38 | ! |
---|
39 | ! Description: |
---|
40 | ! ------------ |
---|
41 | ! Solution of the Poisson equation with a 2D spectral method. |
---|
42 | ! Hybrid version for parallel computers using a 1D domain decomposition, |
---|
43 | ! realized with MPI, along x and parallelization with OPEN-MP along y |
---|
44 | ! (routine poisfft_hybrid_omp). In a second version (poisfft_hybrid_nodes), |
---|
45 | ! optimization is realized by overlapping of computation and communication |
---|
46 | ! and by simultaneously executing as many communication calls as switches |
---|
47 | ! per logical partition (LPAR) are available. This version comes into |
---|
48 | ! effect if more than one node is used and if the environment variable |
---|
49 | ! tasks_per_node is set in a way that it can be devided by switch_per_lpar |
---|
50 | ! without any rest. |
---|
51 | ! |
---|
52 | ! WARNING: In case of OpenMP, there are problems with allocating large |
---|
53 | ! arrays in parallel regions. |
---|
54 | ! |
---|
55 | ! Copyright Klaus Ketelsen / Siegfried Raasch May 2002 |
---|
56 | !------------------------------------------------------------------------------! |
---|
57 | |
---|
58 | USE fft_xy |
---|
59 | USE indices |
---|
60 | USE pegrid |
---|
61 | |
---|
62 | IMPLICIT NONE |
---|
63 | |
---|
64 | INTEGER, PARAMETER :: switch_per_lpar = 2 |
---|
65 | |
---|
66 | INTEGER, SAVE :: nxl_a, nxr_a, & ! total x dimension |
---|
67 | nxl_p, nxr_p, & ! partial x dimension |
---|
68 | nys_a, nyn_a, & ! total y dimension |
---|
69 | nys_p, nyn_p, & ! partial y dimension |
---|
70 | |
---|
71 | npe_s, & ! total number of PEs for solver |
---|
72 | nwords, & ! number of points to be exchanged |
---|
73 | ! with MPI_ALLTOALL |
---|
74 | n_omp_threads ! number of OpenMP threads |
---|
75 | |
---|
76 | ! |
---|
77 | !-- Variables for multi node version (cluster version) using routine |
---|
78 | !-- poisfft_hybrid_nodes |
---|
79 | INTEGER, SAVE :: comm_nodes, & ! communicater nodes |
---|
80 | comm_node_all, & ! communicater all PEs node version |
---|
81 | comm_tasks, & ! communicater tasks |
---|
82 | me, me_node, me_task,& ! identity of this PE |
---|
83 | nodes, & ! number of nodes |
---|
84 | tasks_per_logical_node = -1 ! default no cluster |
---|
85 | |
---|
86 | |
---|
87 | PRIVATE |
---|
88 | |
---|
89 | |
---|
90 | #if .NOT. defined ( __check ) |
---|
91 | PUBLIC poisfft_hybrid, poisfft_hybrid_ini |
---|
92 | |
---|
93 | |
---|
94 | ! |
---|
95 | !-- Public interfaces |
---|
96 | INTERFACE poisfft_hybrid_ini |
---|
97 | MODULE PROCEDURE poisfft_hybrid_ini |
---|
98 | END INTERFACE poisfft_hybrid_ini |
---|
99 | |
---|
100 | INTERFACE poisfft_hybrid |
---|
101 | MODULE PROCEDURE poisfft_hybrid |
---|
102 | END INTERFACE poisfft_hybrid |
---|
103 | |
---|
104 | ! |
---|
105 | !-- Private interfaces |
---|
106 | INTERFACE poisfft_hybrid_omp |
---|
107 | MODULE PROCEDURE poisfft_hybrid_omp |
---|
108 | END INTERFACE poisfft_hybrid_omp |
---|
109 | |
---|
110 | INTERFACE poisfft_hybrid_omp_vec |
---|
111 | MODULE PROCEDURE poisfft_hybrid_omp_vec |
---|
112 | END INTERFACE poisfft_hybrid_omp_vec |
---|
113 | |
---|
114 | INTERFACE poisfft_hybrid_nodes |
---|
115 | MODULE PROCEDURE poisfft_hybrid_nodes |
---|
116 | END INTERFACE poisfft_hybrid_nodes |
---|
117 | |
---|
118 | INTERFACE tridia_hybrid |
---|
119 | MODULE PROCEDURE tridia_hybrid |
---|
120 | END INTERFACE tridia_hybrid |
---|
121 | |
---|
122 | INTERFACE cascade |
---|
123 | MODULE PROCEDURE cascade |
---|
124 | END INTERFACE cascade |
---|
125 | #else |
---|
126 | PUBLIC poisfft_hybrid_ini |
---|
127 | |
---|
128 | ! |
---|
129 | !-- Public interfaces |
---|
130 | INTERFACE poisfft_hybrid_ini |
---|
131 | MODULE PROCEDURE poisfft_hybrid_ini |
---|
132 | END INTERFACE poisfft_hybrid_ini |
---|
133 | #endif |
---|
134 | |
---|
135 | CONTAINS |
---|
136 | |
---|
137 | |
---|
138 | SUBROUTINE poisfft_hybrid_ini |
---|
139 | |
---|
140 | USE control_parameters |
---|
141 | USE pegrid |
---|
142 | |
---|
143 | IMPLICIT NONE |
---|
144 | |
---|
145 | CHARACTER(LEN=8) :: cdummy |
---|
146 | INTEGER :: idummy, istat |
---|
147 | INTEGER, DIMENSION(2) :: coords, dims |
---|
148 | |
---|
149 | LOGICAL, DIMENSION(2) :: period = .false., re_dims |
---|
150 | |
---|
151 | |
---|
152 | ! |
---|
153 | !-- Set the internal index values for the hybrid solver |
---|
154 | #if defined( __parallel ) |
---|
155 | npe_s = pdims(1) |
---|
156 | #else |
---|
157 | npe_s = 1 |
---|
158 | #endif |
---|
159 | nxl_a = 0 |
---|
160 | nxr_a = nx |
---|
161 | nxl_p = 0 |
---|
162 | nxr_p = ( ( nx+1 ) / npe_s ) - 1 |
---|
163 | nys_a = nys |
---|
164 | nyn_a = nyn |
---|
165 | nys_p = 0 |
---|
166 | nyn_p = ( ( ny+1 ) / npe_s ) - 1 |
---|
167 | |
---|
168 | nwords = ( nxr_p-nxl_p+1 ) * nz * ( nyn_p-nys_p+1 ) |
---|
169 | |
---|
170 | #if defined( __KKMP ) .AND. .NOT. defined ( __check ) |
---|
171 | CALL LOCAL_GETENV( 'OMP_NUM_THREADS', 15, cdummy, idummy ) |
---|
172 | READ ( cdummy, '(I8)' ) n_omp_threads |
---|
173 | IF ( n_omp_threads > 1 ) THEN |
---|
174 | WRITE( message_string, * ) 'Number of OpenMP threads = ', & |
---|
175 | n_omp_threads |
---|
176 | CALL message( 'poisfft_hybrid_ini', 'PA0280', 0, 0, 0, 6, 0 ) |
---|
177 | ENDIF |
---|
178 | #else |
---|
179 | n_omp_threads = 1 |
---|
180 | #endif |
---|
181 | ! |
---|
182 | !-- Initialize the one-dimensional FFT routines |
---|
183 | CALL fft_init |
---|
184 | |
---|
185 | ! |
---|
186 | !-- Setup for multi node version (poisfft_hybrid_nodes) |
---|
187 | IF ( n_omp_threads == 1 .AND. & |
---|
188 | ( host(1:4) == 'ibmh' .OR. host(1:4) == 'ibmb' ) ) THEN |
---|
189 | |
---|
190 | IF ( tasks_per_node /= -9999 ) THEN |
---|
191 | ! |
---|
192 | !-- Multi node version requires that the available number of |
---|
193 | !-- switches per logical partition must be an integral divisor |
---|
194 | !-- of the chosen number of tasks per node |
---|
195 | IF ( MOD( tasks_per_node, switch_per_lpar ) == 0 ) THEN |
---|
196 | ! |
---|
197 | !-- Set the switch which decides about usage of the multi node |
---|
198 | !-- version |
---|
199 | IF ( tasks_per_node / switch_per_lpar > 1 .AND. & |
---|
200 | numprocs > tasks_per_node ) THEN |
---|
201 | tasks_per_logical_node = tasks_per_node / switch_per_lpar |
---|
202 | ENDIF |
---|
203 | |
---|
204 | IF ( tasks_per_logical_node > -1 ) THEN |
---|
205 | |
---|
206 | WRITE( message_string, * ) 'running optimized ', & |
---|
207 | 'multinode version', & |
---|
208 | '&switch_per_lpar = ', & |
---|
209 | switch_per_lpar, & |
---|
210 | '&tasks_per_lpar = ', & |
---|
211 | tasks_per_node, & |
---|
212 | 'tasks_per_logical_node = ', & |
---|
213 | tasks_per_logical_node |
---|
214 | CALL message( 'poisfft_hybrid_ini', 'PA0281', 0, 0, 0, 6, 0 ) |
---|
215 | |
---|
216 | ENDIF |
---|
217 | |
---|
218 | ENDIF |
---|
219 | ENDIF |
---|
220 | ENDIF |
---|
221 | |
---|
222 | ! |
---|
223 | !-- Determine sub-topologies for multi node version |
---|
224 | IF ( tasks_per_logical_node >= 2 ) THEN |
---|
225 | |
---|
226 | #if defined( __parallel ) .AND. .NOT. defined ( __check ) |
---|
227 | nodes = ( numprocs + tasks_per_logical_node - 1 ) / & |
---|
228 | tasks_per_logical_node |
---|
229 | dims(1) = nodes |
---|
230 | dims(2) = tasks_per_logical_node |
---|
231 | |
---|
232 | CALL MPI_CART_CREATE( comm2d, 2, dims, period, .FALSE., & |
---|
233 | comm_node_all, istat ) |
---|
234 | CALL MPI_COMM_RANK( comm_node_all, me, istat ) |
---|
235 | |
---|
236 | re_dims(1) = .TRUE. |
---|
237 | re_dims(2) = .FALSE. |
---|
238 | CALL MPI_CART_SUB( comm_node_all, re_dims, comm_nodes, istat ) |
---|
239 | CALL MPI_COMM_RANK( comm_nodes, me_node, istat ) |
---|
240 | |
---|
241 | re_dims(1) = .FALSE. |
---|
242 | re_dims(2) = .TRUE. |
---|
243 | CALL MPI_CART_SUB( comm_node_all, re_dims, comm_tasks, istat ) |
---|
244 | CALL MPI_COMM_RANK( comm_tasks, me_task, istat ) |
---|
245 | |
---|
246 | ! write(0,*) 'who am i',myid,me,me_node,me_task,nodes,& |
---|
247 | ! tasks_per_logical_node |
---|
248 | #elif .NOT. defined( __parallel ) |
---|
249 | message_string = 'parallel environment (MPI) required' |
---|
250 | CALL message( 'poisfft_hybrid_ini', 'PA0282', 1, 2, 0, 6, 0 ) |
---|
251 | #endif |
---|
252 | ENDIF |
---|
253 | |
---|
254 | END SUBROUTINE poisfft_hybrid_ini |
---|
255 | |
---|
256 | #if .NOT. defined ( __check ) |
---|
257 | SUBROUTINE poisfft_hybrid( ar ) |
---|
258 | |
---|
259 | USE control_parameters |
---|
260 | USE interfaces |
---|
261 | |
---|
262 | IMPLICIT NONE |
---|
263 | |
---|
264 | REAL, DIMENSION(1:nz,nys:nyn,nxl:nxr) :: ar |
---|
265 | |
---|
266 | IF ( host(1:3) == 'nec' ) THEN |
---|
267 | CALL poisfft_hybrid_omp_vec( ar ) |
---|
268 | ELSE |
---|
269 | IF ( tasks_per_logical_node == -1 ) THEN |
---|
270 | CALL poisfft_hybrid_omp( ar ) |
---|
271 | ELSE |
---|
272 | CALL poisfft_hybrid_nodes( ar ) |
---|
273 | ENDIF |
---|
274 | ENDIF |
---|
275 | |
---|
276 | END SUBROUTINE poisfft_hybrid |
---|
277 | |
---|
278 | |
---|
279 | SUBROUTINE poisfft_hybrid_omp ( ar ) |
---|
280 | |
---|
281 | USE cpulog |
---|
282 | USE interfaces |
---|
283 | |
---|
284 | IMPLICIT NONE |
---|
285 | |
---|
286 | INTEGER, PARAMETER :: istride = 4 ! stride of i loop |
---|
287 | INTEGER :: i, ii, ir, iei, iouter, istat, j, jj, k, m, n, jthread |
---|
288 | |
---|
289 | REAL, DIMENSION(1:nz,nys:nyn,nxl:nxr) :: ar |
---|
290 | |
---|
291 | REAL, DIMENSION(0:nx) :: fftx_ar |
---|
292 | REAL, DIMENSION(0:ny,istride) :: ffty_ar |
---|
293 | |
---|
294 | REAL, DIMENSION(0:nx,nz) :: tri_ar |
---|
295 | |
---|
296 | REAL, DIMENSION(nxl_p:nxr_p,nz,nys_p:nyn_p,npe_s) :: work1, work2 |
---|
297 | #if defined( __KKMP ) |
---|
298 | INTEGER :: omp_get_thread_num |
---|
299 | REAL, DIMENSION(:,:,:,:), ALLOCATABLE :: tri |
---|
300 | ALLOCATE( tri(5,0:nx,0:nz-1,n_omp_threads ) ) |
---|
301 | #else |
---|
302 | REAL, DIMENSION(5,0:nx,0:nz-1,1) :: tri |
---|
303 | #endif |
---|
304 | |
---|
305 | |
---|
306 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_omp', 'start' ) |
---|
307 | |
---|
308 | CALL cpu_log( log_point_s(7), 'fft_y', 'start' ) |
---|
309 | |
---|
310 | !$OMP PARALLEL PRIVATE (i,iouter,ii,ir,iei,j,k,m,n,ffty_ar) |
---|
311 | !$OMP DO |
---|
312 | ! |
---|
313 | !-- Store grid points to be transformed on a 1d-array, do the fft |
---|
314 | !-- and sample the results on a 4d-array |
---|
315 | DO iouter = nxl_p, nxr_p, istride ! stride loop, better cache |
---|
316 | iei = MIN( iouter+istride-1, nxr_p ) |
---|
317 | DO k = 1, nz |
---|
318 | |
---|
319 | DO i = iouter, iei |
---|
320 | ii = nxl + i |
---|
321 | ir = i - iouter + 1 |
---|
322 | |
---|
323 | DO j = nys_a, nyn_a |
---|
324 | ffty_ar(j,ir) = ar(k,j,ii) |
---|
325 | ENDDO |
---|
326 | |
---|
327 | CALL fft_y( ffty_ar(:,ir), 'forward' ) |
---|
328 | ENDDO |
---|
329 | |
---|
330 | m = nys_a |
---|
331 | DO n = 1, npe_s |
---|
332 | DO j = nys_p, nyn_p |
---|
333 | DO i = iouter, iei |
---|
334 | ir = i - iouter + 1 |
---|
335 | work1(i,k,j,n) = ffty_ar(m,ir) |
---|
336 | ENDDO |
---|
337 | m = m+1 |
---|
338 | ENDDO |
---|
339 | ENDDO |
---|
340 | |
---|
341 | ENDDO |
---|
342 | ENDDO |
---|
343 | !$OMP END PARALLEL |
---|
344 | |
---|
345 | CALL cpu_log( log_point_s(7), 'fft_y', 'pause' ) |
---|
346 | |
---|
347 | #if defined( __parallel ) |
---|
348 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
349 | |
---|
350 | CALL MPI_ALLTOALL( work1(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
351 | work2(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
352 | comm2d, istat ) |
---|
353 | |
---|
354 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
355 | #else |
---|
356 | work2 = work1 |
---|
357 | #endif |
---|
358 | |
---|
359 | CALL cpu_log( log_point_s(33), 'fft_x + tridia', 'start' ) |
---|
360 | |
---|
361 | #if defined( __KKMP ) |
---|
362 | !$OMP PARALLEL PRIVATE (i,j,jj,k,m,n,fftx_ar,tri_ar,jthread) |
---|
363 | !$OMP DO |
---|
364 | DO j = nys_p, nyn_p |
---|
365 | jthread = omp_get_thread_num() + 1 |
---|
366 | #else |
---|
367 | DO j = nys_p, nyn_p |
---|
368 | jthread = 1 |
---|
369 | #endif |
---|
370 | DO k = 1, nz |
---|
371 | |
---|
372 | m = nxl_a |
---|
373 | DO n = 1, npe_s |
---|
374 | DO i = nxl_p, nxr_p |
---|
375 | fftx_ar(m) = work2(i,k,j,n) |
---|
376 | m = m+1 |
---|
377 | ENDDO |
---|
378 | ENDDO |
---|
379 | |
---|
380 | CALL fft_x( fftx_ar, 'forward' ) |
---|
381 | |
---|
382 | DO i = nxl_a, nxr_a |
---|
383 | tri_ar(i,k) = fftx_ar(i) |
---|
384 | ENDDO |
---|
385 | |
---|
386 | ENDDO |
---|
387 | |
---|
388 | jj = myid * (nyn_p-nys_p+1) + j |
---|
389 | CALL tridia_hybrid( jj, tri_ar, tri(:,:,:,jthread)) |
---|
390 | |
---|
391 | DO k = 1, nz |
---|
392 | DO i = nxl_a, nxr_a |
---|
393 | fftx_ar(i) = tri_ar (i,k) |
---|
394 | ENDDO |
---|
395 | |
---|
396 | CALL fft_x( fftx_ar, 'backward' ) |
---|
397 | |
---|
398 | m = nxl_a |
---|
399 | DO n = 1, npe_s |
---|
400 | DO i = nxl_p, nxr_p |
---|
401 | work2(i,k,j,n) = fftx_ar(m) |
---|
402 | m = m+1 |
---|
403 | ENDDO |
---|
404 | ENDDO |
---|
405 | |
---|
406 | ENDDO |
---|
407 | ENDDO |
---|
408 | #if defined( __KKMP ) |
---|
409 | !$OMP END PARALLEL |
---|
410 | #endif |
---|
411 | |
---|
412 | CALL cpu_log( log_point_s(33), 'fft_x + tridia', 'stop' ) |
---|
413 | |
---|
414 | #if defined( __parallel ) |
---|
415 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
416 | nwords = (nxr_p-nxl_p+1) * nz * (nyn_p-nys_p+1) |
---|
417 | |
---|
418 | CALL MPI_ALLTOALL( work2(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
419 | work1(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
420 | comm2d, istat ) |
---|
421 | |
---|
422 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
423 | #else |
---|
424 | work1 = work2 |
---|
425 | #endif |
---|
426 | |
---|
427 | CALL cpu_log( log_point_s(7), 'fft_y', 'continue' ) |
---|
428 | |
---|
429 | !$OMP PARALLEL PRIVATE (i,iouter,ii,ir,iei,j,k,m,n,ffty_ar) |
---|
430 | !$OMP DO |
---|
431 | DO iouter = nxl_p, nxr_p, istride |
---|
432 | iei = MIN( iouter+istride-1, nxr_p ) |
---|
433 | DO k = 1, nz |
---|
434 | |
---|
435 | m = nys_a |
---|
436 | DO n = 1, npe_s |
---|
437 | DO j = nys_p, nyn_p |
---|
438 | DO i = iouter, iei |
---|
439 | ir = i - iouter + 1 |
---|
440 | ffty_ar(m,ir) = work1 (i,k,j,n) |
---|
441 | ENDDO |
---|
442 | m = m+1 |
---|
443 | ENDDO |
---|
444 | ENDDO |
---|
445 | |
---|
446 | DO i = iouter, iei |
---|
447 | ii = nxl + i |
---|
448 | ir = i - iouter + 1 |
---|
449 | CALL fft_y( ffty_ar(:,ir), 'backward' ) |
---|
450 | |
---|
451 | DO j = nys_a, nyn_a |
---|
452 | ar(k,j,ii) = ffty_ar(j,ir) |
---|
453 | ENDDO |
---|
454 | ENDDO |
---|
455 | |
---|
456 | ENDDO |
---|
457 | ENDDO |
---|
458 | !$OMP END PARALLEL |
---|
459 | |
---|
460 | CALL cpu_log( log_point_s(7), 'fft_y', 'stop' ) |
---|
461 | |
---|
462 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_omp', 'stop' ) |
---|
463 | |
---|
464 | #if defined( __KKMP ) |
---|
465 | DEALLOCATE( tri ) |
---|
466 | #endif |
---|
467 | |
---|
468 | END SUBROUTINE poisfft_hybrid_omp |
---|
469 | |
---|
470 | |
---|
471 | SUBROUTINE poisfft_hybrid_omp_vec ( ar ) |
---|
472 | |
---|
473 | USE cpulog |
---|
474 | USE interfaces |
---|
475 | |
---|
476 | IMPLICIT NONE |
---|
477 | |
---|
478 | INTEGER, PARAMETER :: istride = 4 ! stride of i loop |
---|
479 | INTEGER :: i, ii, ir, iei, iouter, istat, j, jj, k, m, n, jthread |
---|
480 | |
---|
481 | REAL, DIMENSION(0:nx,nz) :: tri_ar |
---|
482 | |
---|
483 | REAL, DIMENSION(1:nz,nys:nyn,nxl:nxr) :: ar |
---|
484 | |
---|
485 | REAL, DIMENSION(0:ny+3,nz,nxl_p:nxr_p) :: ffty_ar3 |
---|
486 | REAL, DIMENSION(0:nx+3,nz,nys_p:nyn_p) :: fftx_ar3 |
---|
487 | |
---|
488 | REAL, DIMENSION(nxl_p:nxr_p,nz,nys_p:nyn_p,npe_s) :: work1, work2 |
---|
489 | #if defined( __KKMP ) |
---|
490 | INTEGER :: omp_get_thread_num |
---|
491 | REAL, DIMENSION(:,:,:,:), ALLOCATABLE :: tri |
---|
492 | ALLOCATE( tri(5,0:nx,0:nz-1,n_omp_threads ) ) |
---|
493 | #else |
---|
494 | REAL, DIMENSION(5,0:nx,0:nz-1,1) :: tri |
---|
495 | #endif |
---|
496 | |
---|
497 | |
---|
498 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_vec', 'start' ) |
---|
499 | |
---|
500 | CALL cpu_log( log_point_s(7), 'fft_y_m', 'start' ) |
---|
501 | |
---|
502 | !$OMP PARALLEL PRIVATE (i,j,k,m,n) |
---|
503 | !$OMP DO |
---|
504 | ! |
---|
505 | !-- Store grid points to be transformed on a 1d-array, do the fft |
---|
506 | !-- and sample the results on a 4d-array |
---|
507 | DO i = nxl_p, nxr_p |
---|
508 | |
---|
509 | DO j = nys_a, nyn_a |
---|
510 | DO k = 1, nz |
---|
511 | ffty_ar3(j,k,i) = ar(k,j,i+nxl) |
---|
512 | ENDDO |
---|
513 | ENDDO |
---|
514 | |
---|
515 | CALL fft_y_m( ffty_ar3(:,:,i), ny+3, 'forward' ) |
---|
516 | ENDDO |
---|
517 | |
---|
518 | !$OMP DO |
---|
519 | DO k = 1, nz |
---|
520 | m = nys_a |
---|
521 | DO n = 1, npe_s |
---|
522 | DO j = nys_p, nyn_p |
---|
523 | DO i = nxl_p, nxr_p |
---|
524 | work1(i,k,j,n) = ffty_ar3(m,k,i) |
---|
525 | ENDDO |
---|
526 | m = m+1 |
---|
527 | ENDDO |
---|
528 | ENDDO |
---|
529 | ENDDO |
---|
530 | !$OMP END PARALLEL |
---|
531 | |
---|
532 | CALL cpu_log( log_point_s(7), 'fft_y_m', 'pause' ) |
---|
533 | |
---|
534 | #if defined( __parallel ) |
---|
535 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
536 | CALL MPI_ALLTOALL( work1(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
537 | work2(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
538 | comm2d, istat ) |
---|
539 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
540 | #else |
---|
541 | work2 = work1 |
---|
542 | #endif |
---|
543 | |
---|
544 | CALL cpu_log( log_point_s(33), 'fft_x_m + tridia', 'start' ) |
---|
545 | |
---|
546 | #if defined( __KKMP ) |
---|
547 | !$OMP PARALLEL PRIVATE (i,j,jj,k,m,n,tri_ar,jthread) |
---|
548 | !$OMP DO |
---|
549 | DO j = nys_p, nyn_p |
---|
550 | jthread = omp_get_thread_num() + 1 |
---|
551 | #else |
---|
552 | DO j = nys_p, nyn_p |
---|
553 | jthread = 1 |
---|
554 | #endif |
---|
555 | DO k = 1, nz |
---|
556 | |
---|
557 | m = nxl_a |
---|
558 | DO n = 1, npe_s |
---|
559 | DO i = nxl_p, nxr_p |
---|
560 | fftx_ar3(m,k,j) = work2(i,k,j,n) |
---|
561 | m = m+1 |
---|
562 | ENDDO |
---|
563 | ENDDO |
---|
564 | ENDDO |
---|
565 | |
---|
566 | CALL fft_x_m( fftx_ar3(:,:,j), 'forward' ) |
---|
567 | |
---|
568 | DO k = 1, nz |
---|
569 | DO i = nxl_a, nxr_a |
---|
570 | tri_ar(i,k) = fftx_ar3(i,k,j) |
---|
571 | ENDDO |
---|
572 | ENDDO |
---|
573 | |
---|
574 | jj = myid * (nyn_p-nys_p+1) + j |
---|
575 | CALL tridia_hybrid( jj, tri_ar, tri(:,:,:,jthread)) |
---|
576 | |
---|
577 | DO k = 1, nz |
---|
578 | DO i = nxl_a, nxr_a |
---|
579 | fftx_ar3(i,k,j) = tri_ar (i,k) |
---|
580 | ENDDO |
---|
581 | ENDDO |
---|
582 | |
---|
583 | CALL fft_x_m( fftx_ar3(:,:,j), 'backward' ) |
---|
584 | |
---|
585 | DO k = 1, nz |
---|
586 | m = nxl_a |
---|
587 | DO n = 1, npe_s |
---|
588 | DO i = nxl_p, nxr_p |
---|
589 | work2(i,k,j,n) = fftx_ar3(m,k,j) |
---|
590 | m = m+1 |
---|
591 | ENDDO |
---|
592 | ENDDO |
---|
593 | ENDDO |
---|
594 | |
---|
595 | ENDDO |
---|
596 | #if defined( __KKMP ) |
---|
597 | !$OMP END PARALLEL |
---|
598 | #endif |
---|
599 | |
---|
600 | CALL cpu_log( log_point_s(33), 'fft_x_m + tridia', 'stop' ) |
---|
601 | |
---|
602 | #if defined( __parallel ) |
---|
603 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
604 | nwords = (nxr_p-nxl_p+1) * nz * (nyn_p-nys_p+1) |
---|
605 | CALL MPI_ALLTOALL( work2(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
606 | work1(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
607 | comm2d, istat ) |
---|
608 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
609 | #else |
---|
610 | work1 = work2 |
---|
611 | #endif |
---|
612 | |
---|
613 | CALL cpu_log( log_point_s(7), 'fft_y_m', 'continue' ) |
---|
614 | |
---|
615 | !$OMP PARALLEL PRIVATE (i,iouter,ii,ir,iei,j,k,m,n) |
---|
616 | !$OMP DO |
---|
617 | DO k = 1, nz |
---|
618 | m = nys_a |
---|
619 | DO n = 1, npe_s |
---|
620 | DO j = nys_p, nyn_p |
---|
621 | DO i = nxl_p, nxr_p |
---|
622 | ffty_ar3(m,k,i) = work1(i,k,j,n) |
---|
623 | ENDDO |
---|
624 | m = m+1 |
---|
625 | ENDDO |
---|
626 | ENDDO |
---|
627 | ENDDO |
---|
628 | |
---|
629 | !$OMP DO |
---|
630 | DO i = nxl_p, nxr_p |
---|
631 | CALL fft_y_m( ffty_ar3(:,:,i), ny+3, 'backward' ) |
---|
632 | DO j = nys_a, nyn_a |
---|
633 | DO k = 1, nz |
---|
634 | ar(k,j,i+nxl) = ffty_ar3(j,k,i) |
---|
635 | ENDDO |
---|
636 | ENDDO |
---|
637 | ENDDO |
---|
638 | !$OMP END PARALLEL |
---|
639 | |
---|
640 | CALL cpu_log( log_point_s(7), 'fft_y_m', 'stop' ) |
---|
641 | |
---|
642 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_vec', 'stop' ) |
---|
643 | |
---|
644 | #if defined( __KKMP ) |
---|
645 | DEALLOCATE( tri ) |
---|
646 | #endif |
---|
647 | |
---|
648 | END SUBROUTINE poisfft_hybrid_omp_vec |
---|
649 | |
---|
650 | |
---|
651 | SUBROUTINE poisfft_hybrid_nodes ( ar ) |
---|
652 | |
---|
653 | USE cpulog |
---|
654 | USE interfaces |
---|
655 | |
---|
656 | IMPLICIT NONE |
---|
657 | |
---|
658 | INTEGER, PARAMETER :: istride = 4 ! stride of i loop |
---|
659 | INTEGER :: i, iei, ii, iouter, ir, istat, j, jj, k, m, & |
---|
660 | n, nn, nt, nw1, nw2 |
---|
661 | |
---|
662 | REAL, DIMENSION(1:nz,nys:nyn,nxl:nxr) :: ar |
---|
663 | |
---|
664 | REAL, DIMENSION(0:nx) :: fftx_ar |
---|
665 | REAL, DIMENSION(0:ny,istride) :: ffty_ar |
---|
666 | |
---|
667 | REAL, DIMENSION(0:nx,nz) :: tri_ar |
---|
668 | |
---|
669 | REAL, DIMENSION(nxl_p:nxr_p,nz,tasks_per_logical_node, & |
---|
670 | nodes,nys_p:nyn_p) :: work1,work2 |
---|
671 | REAL, DIMENSION(5,0:nx,0:nz-1) :: tri |
---|
672 | |
---|
673 | |
---|
674 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_nodes', 'start' ) |
---|
675 | |
---|
676 | CALL cpu_log( log_point_s(7), 'fft_y', 'start' ) |
---|
677 | |
---|
678 | ! |
---|
679 | !-- Store grid points to be transformed on a 1d-array, do the fft |
---|
680 | !-- and sample the results on a 4d-array |
---|
681 | DO iouter = nxl_p, nxr_p, istride ! stride loop, better cache |
---|
682 | iei = MIN( iouter+istride-1, nxr_p ) |
---|
683 | DO k = 1, nz |
---|
684 | |
---|
685 | DO i = iouter, iei |
---|
686 | ii = nxl + i |
---|
687 | ir = i - iouter + 1 |
---|
688 | |
---|
689 | DO j = nys_a, nyn_a |
---|
690 | ffty_ar(j,ir) = ar(k,j,ii) |
---|
691 | ENDDO |
---|
692 | |
---|
693 | CALL fft_y( ffty_ar(:,ir), 'forward' ) |
---|
694 | ENDDO |
---|
695 | |
---|
696 | m = nys_a |
---|
697 | DO nn = 1, nodes |
---|
698 | DO nt = 1, tasks_per_logical_node |
---|
699 | DO j = nys_p, nyn_p |
---|
700 | DO i = iouter, iei |
---|
701 | ir = i - iouter + 1 |
---|
702 | work1(i,k,nt,nn,j) = ffty_ar(m,ir) |
---|
703 | ENDDO |
---|
704 | m = m+1 |
---|
705 | ENDDO |
---|
706 | ENDDO |
---|
707 | ENDDO |
---|
708 | |
---|
709 | ENDDO |
---|
710 | ENDDO |
---|
711 | |
---|
712 | CALL cpu_log( log_point_s(7), 'fft_y', 'pause' ) |
---|
713 | |
---|
714 | CALL cpu_log( log_point_s(32), 'alltoall_task', 'start' ) |
---|
715 | nw1 = SIZE( work1, 1 ) * SIZE( work1, 2 ) |
---|
716 | DO nn = 1, nodes |
---|
717 | DO j = nys_p, nyn_p |
---|
718 | #if defined( __parallel ) |
---|
719 | CALL MPI_ALLTOALL( work1(nxl_p,1,1,nn,j), nw1, MPI_REAL, & |
---|
720 | work2(nxl_p,1,1,nn,j), nw1, MPI_REAL, & |
---|
721 | comm_tasks, istat ) |
---|
722 | #endif |
---|
723 | ENDDO |
---|
724 | ENDDO |
---|
725 | CALL cpu_log( log_point_s(32), 'alltoall_task', 'stop' ) |
---|
726 | |
---|
727 | |
---|
728 | DO j = nys_p, nyn_p |
---|
729 | |
---|
730 | CALL cascade( 1, j, nys_p, nyn_p ) |
---|
731 | nw2 = nw1 * SIZE( work1, 3 ) |
---|
732 | CALL cpu_log( log_point_s(37), 'alltoall_node', 'start' ) |
---|
733 | #if defined( __parallel ) |
---|
734 | CALL MPI_ALLTOALL( work2(nxl_p,1,1,1,j), nw2, MPI_REAL, & |
---|
735 | work1(nxl_p,1,1,1,j), nw2, MPI_REAL, & |
---|
736 | comm_nodes, istat ) |
---|
737 | #endif |
---|
738 | CALL cpu_log( log_point_s(37), 'alltoall_node', 'pause' ) |
---|
739 | CALL cascade( 2, j, nys_p, nyn_p ) |
---|
740 | |
---|
741 | CALL cpu_log( log_point_s(33), 'fft_x + tridia', 'start' ) |
---|
742 | DO k = 1, nz |
---|
743 | |
---|
744 | m = nxl_a |
---|
745 | DO nn = 1, nodes |
---|
746 | DO nt = 1, tasks_per_logical_node |
---|
747 | DO i = nxl_p, nxr_p |
---|
748 | fftx_ar(m) = work1(i,k,nt,nn,j) |
---|
749 | m = m+1 |
---|
750 | ENDDO |
---|
751 | ENDDO |
---|
752 | ENDDO |
---|
753 | |
---|
754 | CALL fft_x( fftx_ar, 'forward' ) |
---|
755 | |
---|
756 | DO i = nxl_a, nxr_a |
---|
757 | tri_ar(i,k) = fftx_ar(i) |
---|
758 | ENDDO |
---|
759 | |
---|
760 | ENDDO |
---|
761 | |
---|
762 | jj = myid * (nyn_p-nys_p+1) + j |
---|
763 | CALL tridia_hybrid( jj, tri_ar, tri(:,:,:) ) |
---|
764 | |
---|
765 | DO k = 1, nz |
---|
766 | DO i = nxl_a, nxr_a |
---|
767 | fftx_ar(i) = tri_ar(i,k) |
---|
768 | ENDDO |
---|
769 | |
---|
770 | CALL fft_x( fftx_ar, 'backward' ) |
---|
771 | |
---|
772 | m = nxl_a |
---|
773 | DO nn = 1, nodes |
---|
774 | DO nt = 1, tasks_per_logical_node |
---|
775 | DO i = nxl_p, nxr_p |
---|
776 | work1(i,k,nt,nn,j) = fftx_ar(m) |
---|
777 | m = m+1 |
---|
778 | ENDDO |
---|
779 | ENDDO |
---|
780 | ENDDO |
---|
781 | ENDDO |
---|
782 | |
---|
783 | CALL cpu_log( log_point_s(33), 'fft_x + tridia', 'stop' ) |
---|
784 | nw2 = nw1 * SIZE( work1, 3 ) |
---|
785 | CALL cpu_log( log_point_s(37), 'alltoall_node', 'continue' ) |
---|
786 | #if defined( __parallel ) |
---|
787 | CALL MPI_ALLTOALL( work1(nxl_p,1,1,1,j), nw2, MPI_REAL, & |
---|
788 | work2(nxl_p,1,1,1,j), nw2, MPI_REAL, & |
---|
789 | comm_nodes, istat ) |
---|
790 | #endif |
---|
791 | CALL cpu_log( log_point_s(37), 'alltoall_node', 'stop' ) |
---|
792 | |
---|
793 | ENDDO |
---|
794 | |
---|
795 | CALL cpu_log( log_point_s(32), 'alltoall_task', 'start' ) |
---|
796 | DO nn = 1, nodes |
---|
797 | DO j = nys_p, nyn_p |
---|
798 | #if defined( __parallel ) |
---|
799 | CALL MPI_ALLTOALL( work2(nxl_p,1,1,nn,j), nw1, MPI_REAL, & |
---|
800 | work1(nxl_p,1,1,nn,j), nw1, MPI_REAL, & |
---|
801 | comm_tasks, istat ) |
---|
802 | #endif |
---|
803 | ENDDO |
---|
804 | ENDDO |
---|
805 | CALL cpu_log( log_point_s(32), 'alltoall_task', 'stop' ) |
---|
806 | |
---|
807 | CALL cpu_log( log_point_s(7), 'fft_y', 'continue' ) |
---|
808 | |
---|
809 | DO iouter = nxl_p, nxr_p, istride |
---|
810 | iei = MIN( iouter+istride-1, nxr_p ) |
---|
811 | DO k = 1, nz |
---|
812 | |
---|
813 | m = nys_a |
---|
814 | DO nn = 1, nodes |
---|
815 | DO nt = 1, tasks_per_logical_node |
---|
816 | DO j = nys_p, nyn_p |
---|
817 | DO i = iouter, iei |
---|
818 | ir = i - iouter + 1 |
---|
819 | ffty_ar(m,ir) = work1(i,k,nt,nn,j) |
---|
820 | ENDDO |
---|
821 | m = m+1 |
---|
822 | ENDDO |
---|
823 | ENDDO |
---|
824 | ENDDO |
---|
825 | |
---|
826 | DO i = iouter, iei |
---|
827 | ii = nxl + i |
---|
828 | ir = i - iouter + 1 |
---|
829 | CALL fft_y( ffty_ar(:,ir), 'backward' ) |
---|
830 | |
---|
831 | DO j = nys_a, nyn_a |
---|
832 | ar(k,j,ii) = ffty_ar(j,ir) |
---|
833 | ENDDO |
---|
834 | ENDDO |
---|
835 | |
---|
836 | ENDDO |
---|
837 | ENDDO |
---|
838 | |
---|
839 | CALL cpu_log( log_point_s(7), 'fft_y', 'stop' ) |
---|
840 | |
---|
841 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_nodes', 'stop' ) |
---|
842 | |
---|
843 | END SUBROUTINE poisfft_hybrid_nodes |
---|
844 | |
---|
845 | |
---|
846 | |
---|
847 | SUBROUTINE tridia_hybrid( j, ar, tri ) |
---|
848 | |
---|
849 | USE arrays_3d |
---|
850 | USE control_parameters |
---|
851 | USE grid_variables |
---|
852 | |
---|
853 | IMPLICIT NONE |
---|
854 | |
---|
855 | INTEGER :: i, j, k, nnyh |
---|
856 | REAL, DIMENSION(0:nx,nz) :: ar |
---|
857 | REAL, DIMENSION(0:nx,0:nz-1) :: ar1 |
---|
858 | REAL, DIMENSION(5,0:nx,0:nz-1) :: tri |
---|
859 | |
---|
860 | nnyh = (ny+1) / 2 |
---|
861 | |
---|
862 | tri = 0.0 |
---|
863 | ! |
---|
864 | !-- Define constant elements of the tridiagonal matrix. |
---|
865 | DO k = 0, nz-1 |
---|
866 | DO i = 0,nx |
---|
867 | tri(2,i,k) = ddzu_pres(k+1) * ddzw(k+1) |
---|
868 | tri(3,i,k) = ddzu_pres(k+2) * ddzw(k+1) |
---|
869 | ENDDO |
---|
870 | ENDDO |
---|
871 | |
---|
872 | IF ( j <= nnyh ) THEN |
---|
873 | CALL maketri_hybrid( j ) |
---|
874 | ELSE |
---|
875 | CALL maketri_hybrid( ny+1-j) |
---|
876 | ENDIF |
---|
877 | CALL zerleg_hybrid |
---|
878 | CALL substi_hybrid( ar, tri ) |
---|
879 | |
---|
880 | CONTAINS |
---|
881 | |
---|
882 | SUBROUTINE maketri_hybrid( j ) |
---|
883 | |
---|
884 | !----------------------------------------------------------------------! |
---|
885 | ! maketri ! |
---|
886 | ! ! |
---|
887 | ! computes the i- and j-dependent component of the matrix ! |
---|
888 | !----------------------------------------------------------------------! |
---|
889 | |
---|
890 | USE constants |
---|
891 | |
---|
892 | IMPLICIT NONE |
---|
893 | |
---|
894 | INTEGER :: i, j, k, nnxh |
---|
895 | REAL :: a, c |
---|
896 | |
---|
897 | REAL, DIMENSION(0:nx) :: l |
---|
898 | |
---|
899 | |
---|
900 | nnxh = (nx+1) / 2 |
---|
901 | ! |
---|
902 | !-- Provide the tridiagonal matrix for solution of the Poisson equation |
---|
903 | !-- in Fourier space. The coefficients are computed following the method |
---|
904 | !-- of Schmidt et al. (DFVLR-Mitteilung 84-15) --> departs from Stephan |
---|
905 | !-- Siano's original version. |
---|
906 | DO i = 0,nx |
---|
907 | IF ( i >= 0 .AND. i < nnxh ) THEN |
---|
908 | l(i) = 2.0 * ( 1.0 - COS( ( 2.0 * pi * i ) / & |
---|
909 | FLOAT( nx+1 ) ) ) / ( dx * dx ) + & |
---|
910 | 2.0 * ( 1.0 - COS( ( 2.0 * pi * j ) / & |
---|
911 | FLOAT( ny+1 ) ) ) / ( dy * dy ) |
---|
912 | ELSEIF ( i == nnxh ) THEN |
---|
913 | l(i) = 2.0 * ( 1.0 - COS( ( 2.0 * pi * ( nx+1-i ) ) / & |
---|
914 | FLOAT( nx+1 ) ) ) / ( dx * dx ) + & |
---|
915 | 2.0 * ( 1.0 - COS( ( 2.0 * pi * j ) / & |
---|
916 | FLOAT(ny+1) ) ) / ( dy * dy ) |
---|
917 | ELSE |
---|
918 | l(i) = 2.0 * ( 1.0 - COS( ( 2.0 * pi * ( nx+1-i ) ) / & |
---|
919 | FLOAT( nx+1 ) ) ) / ( dx * dx ) + & |
---|
920 | 2.0 * ( 1.0 - COS( ( 2.0 * pi * j ) / & |
---|
921 | FLOAT( ny+1 ) ) ) / ( dy * dy ) |
---|
922 | ENDIF |
---|
923 | ENDDO |
---|
924 | |
---|
925 | DO k = 0,nz-1 |
---|
926 | DO i = 0, nx |
---|
927 | a = -1.0 * ddzu_pres(k+2) * ddzw(k+1) |
---|
928 | c = -1.0 * ddzu_pres(k+1) * ddzw(k+1) |
---|
929 | tri(1,i,k) = a + c - l(i) |
---|
930 | ENDDO |
---|
931 | ENDDO |
---|
932 | IF ( ibc_p_b == 1 .OR. ibc_p_b == 2 ) THEN |
---|
933 | DO i = 0,nx |
---|
934 | tri(1,i,0) = tri(1,i,0) + tri(2,i,0) |
---|
935 | ENDDO |
---|
936 | ENDIF |
---|
937 | IF ( ibc_p_t == 1 ) THEN |
---|
938 | DO i = 0,nx |
---|
939 | tri(1,i,nz-1) = tri(1,i,nz-1) + tri(3,i,nz-1) |
---|
940 | ENDDO |
---|
941 | ENDIF |
---|
942 | |
---|
943 | END SUBROUTINE maketri_hybrid |
---|
944 | |
---|
945 | |
---|
946 | SUBROUTINE zerleg_hybrid |
---|
947 | |
---|
948 | !----------------------------------------------------------------------! |
---|
949 | ! zerleg ! |
---|
950 | ! ! |
---|
951 | ! Splitting of the tridiagonal matrix (Thomas algorithm) ! |
---|
952 | !----------------------------------------------------------------------! |
---|
953 | |
---|
954 | USE indices |
---|
955 | |
---|
956 | IMPLICIT NONE |
---|
957 | |
---|
958 | INTEGER :: i, k |
---|
959 | |
---|
960 | ! |
---|
961 | !-- Splitting |
---|
962 | DO i = 0, nx |
---|
963 | tri(4,i,0) = tri(1,i,0) |
---|
964 | ENDDO |
---|
965 | DO k = 1, nz-1 |
---|
966 | DO i = 0,nx |
---|
967 | tri(5,i,k) = tri(2,i,k) / tri(4,i,k-1) |
---|
968 | tri(4,i,k) = tri(1,i,k) - tri(3,i,k-1) * tri(5,i,k) |
---|
969 | ENDDO |
---|
970 | ENDDO |
---|
971 | |
---|
972 | END SUBROUTINE zerleg_hybrid |
---|
973 | |
---|
974 | SUBROUTINE substi_hybrid( ar, tri ) |
---|
975 | |
---|
976 | !----------------------------------------------------------------------! |
---|
977 | ! substi ! |
---|
978 | ! ! |
---|
979 | ! Substitution (Forward and Backward) (Thomas algorithm) ! |
---|
980 | !----------------------------------------------------------------------! |
---|
981 | |
---|
982 | IMPLICIT NONE |
---|
983 | |
---|
984 | INTEGER :: i, j, k |
---|
985 | REAL, DIMENSION(0:nx,nz) :: ar |
---|
986 | REAL, DIMENSION(0:nx,0:nz-1) :: ar1 |
---|
987 | REAL, DIMENSION(5,0:nx,0:nz-1) :: tri |
---|
988 | |
---|
989 | ! |
---|
990 | !-- Forward substitution |
---|
991 | DO i = 0, nx |
---|
992 | ar1(i,0) = ar(i,1) |
---|
993 | ENDDO |
---|
994 | DO k = 1, nz - 1 |
---|
995 | DO i = 0,nx |
---|
996 | ar1(i,k) = ar(i,k+1) - tri(5,i,k) * ar1(i,k-1) |
---|
997 | ENDDO |
---|
998 | ENDDO |
---|
999 | |
---|
1000 | ! |
---|
1001 | !-- Backward substitution |
---|
1002 | DO i = 0,nx |
---|
1003 | ar(i,nz) = ar1(i,nz-1) / tri(4,i,nz-1) |
---|
1004 | ENDDO |
---|
1005 | DO k = nz-2, 0, -1 |
---|
1006 | DO i = 0,nx |
---|
1007 | ar(i,k+1) = ( ar1(i,k) - tri(3,i,k) * ar(i,k+2) ) & |
---|
1008 | / tri(4,i,k) |
---|
1009 | ENDDO |
---|
1010 | ENDDO |
---|
1011 | |
---|
1012 | END SUBROUTINE substi_hybrid |
---|
1013 | |
---|
1014 | END SUBROUTINE tridia_hybrid |
---|
1015 | |
---|
1016 | |
---|
1017 | SUBROUTINE cascade( loca, j, nys_p, nyn_p ) |
---|
1018 | |
---|
1019 | USE cpulog |
---|
1020 | |
---|
1021 | IMPLICIT NONE |
---|
1022 | |
---|
1023 | INTEGER :: ier, j, loca, nyn_p, nys_p, req, reqa(1) |
---|
1024 | INTEGER, SAVE :: tag = 10 |
---|
1025 | #if defined( __parallel ) |
---|
1026 | INTEGER, DIMENSION(MPI_STATUS_SIZE) :: stat |
---|
1027 | INTEGER, DIMENSION(MPI_STATUS_SIZE,1) :: stata |
---|
1028 | #endif |
---|
1029 | |
---|
1030 | REAL :: buf, buf1 |
---|
1031 | |
---|
1032 | |
---|
1033 | buf = 1.0 |
---|
1034 | buf1 = 1.1 |
---|
1035 | IF ( me_node == 0 ) THEN ! first node only |
---|
1036 | |
---|
1037 | SELECT CASE ( loca ) |
---|
1038 | |
---|
1039 | CASE ( 1 ) ! before alltoall |
---|
1040 | |
---|
1041 | IF( me_task > 0 ) THEN ! first task does not wait |
---|
1042 | #if defined( __parallel ) |
---|
1043 | CALL MPI_SENDRECV( buf, 1, MPI_REAL, me_task-1, 0, & |
---|
1044 | buf1, 1, MPI_REAL, me_task-1, 0, & |
---|
1045 | comm_tasks, stat, ierr ) |
---|
1046 | #endif |
---|
1047 | ELSEIF ( j > nys_p ) THEN |
---|
1048 | req = 0 |
---|
1049 | tag = MOD( tag-10, 10 ) + 10 |
---|
1050 | #if defined( __parallel ) |
---|
1051 | CALL MPI_IRECV( buf, 1, MPI_REAL, tasks_per_logical_node-1,& |
---|
1052 | tag, comm_tasks, req, ierr ) |
---|
1053 | reqa = req |
---|
1054 | CALL MPI_WAITALL( 1, reqa, stata, ierr ) |
---|
1055 | #endif |
---|
1056 | ENDIF |
---|
1057 | |
---|
1058 | CASE ( 2 ) ! after alltoall |
---|
1059 | |
---|
1060 | IF ( me_task < tasks_per_logical_node-1 ) THEN ! last task |
---|
1061 | #if defined( __parallel ) |
---|
1062 | CALL MPI_SENDRECV( buf, 1, MPI_REAL, me_task+1, 0, & |
---|
1063 | buf1, 1, MPI_REAL, me_task+1, 0, & |
---|
1064 | comm_tasks, stat, ierr) |
---|
1065 | #endif |
---|
1066 | ELSEIF ( j < nyn_p ) THEN |
---|
1067 | req = 0 |
---|
1068 | tag = MOD( tag-10, 10 ) + 10 |
---|
1069 | #if defined( __parallel ) |
---|
1070 | CALL MPI_ISEND( buf, 1, MPI_REAL, 0, tag, comm_tasks, req, & |
---|
1071 | ierr ) |
---|
1072 | #endif |
---|
1073 | ENDIF |
---|
1074 | |
---|
1075 | END SELECT |
---|
1076 | |
---|
1077 | ENDIF |
---|
1078 | |
---|
1079 | END SUBROUTINE cascade |
---|
1080 | #endif |
---|
1081 | END MODULE poisfft_hybrid_mod |
---|