Home

Context Navigation

source: palm/trunk/SOURCE/pres.f90 @ 1117

Last change on this file since 1117 was 1117, checked in by suehring, 11 years ago
Bugfix in OpenMP parallelization.
Property svn:keywords set to `Id`
File size: 24.0 KB

Line
1	SUBROUTINE pres
2
3	!--------------------------------------------------------------------------------!
4	! This file is part of PALM.
5	!
6	! PALM is free software: you can redistribute it and/or modify it under the terms
7	! of the GNU General Public License as published by the Free Software Foundation,
8	! either version 3 of the License, or (at your option) any later 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 1997-2012 Leibniz University Hannover
18	!--------------------------------------------------------------------------------!
19	!
20	! Current revisions:
21	! -----------------
22	! Bugfix in OpenMP parallelization.
23	!
24	! Former revisions:
25	! -----------------
26	! $Id: pres.f90 1117 2013-03-27 11:15:36Z suehring $
27	!
28	! 1113 2013-03-10 02:48:14Z raasch
29	! GPU-porting of several loops, some loops rearranged
30	!
31	! 1111 2013-03-08 23:54:10Z
32	! openACC statements added,
33	! ibc_p_b = 2 removed
34	!
35	! 1092 2013-02-02 11:24:22Z raasch
36	! unused variables removed
37	!
38	! 1036 2012-10-22 13:43:42Z raasch
39	! code put under GPL (PALM 3.9)
40	!
41	! 1003 2012-09-14 14:35:53Z raasch
42	! adjustment of array tend for cases with unequal subdomain sizes removed
43	!
44	! 778 2011-11-07 14:18:25Z fricke
45	! New allocation of tend when multigrid is used and the collected field on PE0
46	! has more grid points than the subdomain of an PE.
47	!
48	! 719 2011-04-06 13:05:23Z gryschka
49	! Bugfix in volume flow control for double cyclic boundary conditions
50	!
51	! 709 2011-03-30 09:31:40Z raasch
52	! formatting adjustments
53	!
54	! 707 2011-03-29 11:39:40Z raasch
55	! Calculation of weighted average of p is now handled in the same way
56	! regardless of the number of ghost layers (advection scheme),
57	! multigrid and sor method are using p_loc for iterative advancements of
58	! pressure,
59	! localsum calculation modified for proper OpenMP reduction,
60	! bc_lr/ns replaced by bc_lr/ns_cyc
61	!
62	! 693 2011-03-08 09:..:..Z raasch
63	! bugfix: weighting coefficient added to ibm branch
64	!
65	! 680 2011-02-04 23:16:06Z gryschka
66	! bugfix: collective_wait
67	!
68	! 675 2011-01-19 10:56:55Z suehring
69	! Removed bugfix while copying tend.
70	!
71	! 673 2011-01-18 16:19:48Z suehring
72	! Weighting coefficients added for right computation of the pressure during
73	! Runge-Kutta substeps.
74	!
75	! 667 2010-12-23 12:06:00Z suehring/gryschka
76	! New allocation of tend when ws-scheme and multigrid is used. This is due to
77	! reasons of perforance of the data_exchange. The same is done with p after
78	! poismg is called.
79	! nxl-1, nxr+1, nys-1, nyn+1 replaced by nxlg, nxrg, nysg, nyng when no
80	! multigrid is used. Calls of exchange_horiz are modified.
81	! bugfix: After pressure correction no volume flow correction in case of
82	! non-cyclic boundary conditions
83	! (has to be done only before pressure correction)
84	! Call of SOR routine is referenced with ddzu_pres.
85	!
86	! 622 2010-12-10 08:08:13Z raasch
87	! optional barriers included in order to speed up collective operations
88	!
89	! 151 2008-03-07 13:42:18Z raasch
90	! Bugfix in volume flow control for non-cyclic boundary conditions
91	!
92	! 106 2007-08-16 14:30:26Z raasch
93	! Volume flow conservation added for the remaining three outflow boundaries
94	!
95	! 85 2007-05-11 09:35:14Z raasch
96	! Division through dt_3d replaced by multiplication of the inverse.
97	! For performance optimisation, this is done in the loop calculating the
98	! divergence instead of using a seperate loop.
99	!
100	! 75 2007-03-22 09:54:05Z raasch
101	! Volume flow control for non-cyclic boundary conditions added (currently only
102	! for the north boundary!!), 2nd+3rd argument removed from exchange horiz,
103	! mean vertical velocity is removed in case of Neumann boundary conditions
104	! both at the bottom and the top
105	!
106	! RCS Log replace by Id keyword, revision history cleaned up
107	!
108	! Revision 1.25 2006/04/26 13:26:12 raasch
109	! OpenMP optimization (+localsum, threadsum)
110	!
111	! Revision 1.1 1997/07/24 11:24:44 raasch
112	! Initial revision
113	!
114	!
115	! Description:
116	! ------------
117	! Compute the divergence of the provisional velocity field. Solve the Poisson
118	! equation for the perturbation pressure. Compute the final velocities using
119	! this perturbation pressure. Compute the remaining divergence.
120	!------------------------------------------------------------------------------!
121
122	USE arrays_3d
123	USE constants
124	USE control_parameters
125	USE cpulog
126	USE grid_variables
127	USE indices
128	USE interfaces
129	USE pegrid
130	USE poisfft_mod
131	USE poisfft_hybrid_mod
132	USE statistics
133
134	IMPLICIT NONE
135
136	INTEGER :: i, j, k
137
138	REAL :: ddt_3d, localsum, threadsum, d_weight_pres
139
140	REAL, DIMENSION(1:2) :: volume_flow_l, volume_flow_offset
141	REAL, DIMENSION(1:nzt) :: w_l, w_l_l
142
143
144	CALL cpu_log( log_point(8), 'pres', 'start' )
145
146
147	ddt_3d = 1.0 / dt_3d
148	d_weight_pres = 1.0 / weight_pres(intermediate_timestep_count)
149
150	!
151	!-- Multigrid method expects array d to have one ghost layer.
152	!--
153	IF ( psolver == 'multigrid' ) THEN
154
155	DEALLOCATE( d )
156	ALLOCATE( d(nzb:nzt+1,nys-1:nyn+1,nxl-1:nxr+1) )
157
158	!
159	!-- Since p is later used to hold the weighted average of the substeps, it
160	!-- cannot be used in the iterative solver. Therefore, its initial value is
161	!-- stored on p_loc, which is then iteratively advanced in every substep.
162	IF ( intermediate_timestep_count == 1 ) THEN
163	DO i = nxl-1, nxr+1
164	DO j = nys-1, nyn+1
165	DO k = nzb, nzt+1
166	p_loc(k,j,i) = p(k,j,i)
167	ENDDO
168	ENDDO
169	ENDDO
170	ENDIF
171
172	ELSEIF ( psolver == 'sor' .AND. intermediate_timestep_count == 1 ) THEN
173
174	!
175	!-- Since p is later used to hold the weighted average of the substeps, it
176	!-- cannot be used in the iterative solver. Therefore, its initial value is
177	!-- stored on p_loc, which is then iteratively advanced in every substep.
178	p_loc = p
179
180	ENDIF
181
182	!
183	!-- Conserve the volume flow at the outflow in case of non-cyclic lateral
184	!-- boundary conditions
185	!-- WARNING: so far, this conservation does not work at the left/south
186	!-- boundary if the topography at the inflow differs from that at the
187	!-- outflow! For this case, volume_flow_area needs adjustment!
188	!
189	!-- Left/right
190	IF ( conserve_volume_flow .AND. ( outflow_l .OR. outflow_r ) ) THEN
191
192	volume_flow(1) = 0.0
193	volume_flow_l(1) = 0.0
194
195	IF ( outflow_l ) THEN
196	i = 0
197	ELSEIF ( outflow_r ) THEN
198	i = nx+1
199	ENDIF
200
201	DO j = nys, nyn
202	!
203	!-- Sum up the volume flow through the south/north boundary
204	DO k = nzb_2d(j,i)+1, nzt
205	volume_flow_l(1) = volume_flow_l(1) + u(k,j,i) * dzw(k)
206	ENDDO
207	ENDDO
208
209	#if defined( __parallel )
210	IF ( collective_wait ) CALL MPI_BARRIER( comm1dy, ierr )
211	CALL MPI_ALLREDUCE( volume_flow_l(1), volume_flow(1), 1, MPI_REAL, &
212	MPI_SUM, comm1dy, ierr )
213	#else
214	volume_flow = volume_flow_l
215	#endif
216	volume_flow_offset(1) = ( volume_flow_initial(1) - volume_flow(1) ) &
217	/ volume_flow_area(1)
218
219	DO j = nysg, nyng
220	DO k = nzb_2d(j,i)+1, nzt
221	u(k,j,i) = u(k,j,i) + volume_flow_offset(1)
222	ENDDO
223	ENDDO
224
225	ENDIF
226
227	!
228	!-- South/north
229	IF ( conserve_volume_flow .AND. ( outflow_n .OR. outflow_s ) ) THEN
230
231	volume_flow(2) = 0.0
232	volume_flow_l(2) = 0.0
233
234	IF ( outflow_s ) THEN
235	j = 0
236	ELSEIF ( outflow_n ) THEN
237	j = ny+1
238	ENDIF
239
240	DO i = nxl, nxr
241	!
242	!-- Sum up the volume flow through the south/north boundary
243	DO k = nzb_2d(j,i)+1, nzt
244	volume_flow_l(2) = volume_flow_l(2) + v(k,j,i) * dzw(k)
245	ENDDO
246	ENDDO
247
248	#if defined( __parallel )
249	IF ( collective_wait ) CALL MPI_BARRIER( comm1dx, ierr )
250	CALL MPI_ALLREDUCE( volume_flow_l(2), volume_flow(2), 1, MPI_REAL, &
251	MPI_SUM, comm1dx, ierr )
252	#else
253	volume_flow = volume_flow_l
254	#endif
255	volume_flow_offset(2) = ( volume_flow_initial(2) - volume_flow(2) ) &
256	/ volume_flow_area(2)
257
258	DO i = nxlg, nxrg
259	DO k = nzb_v_inner(j,i)+1, nzt
260	v(k,j,i) = v(k,j,i) + volume_flow_offset(2)
261	ENDDO
262	ENDDO
263
264	ENDIF
265
266	!
267	!-- Remove mean vertical velocity
268	IF ( ibc_p_b == 1 .AND. ibc_p_t == 1 ) THEN
269	IF ( simulated_time > 0.0 ) THEN ! otherwise nzb_w_inner not yet known
270	w_l = 0.0; w_l_l = 0.0
271	DO i = nxl, nxr
272	DO j = nys, nyn
273	DO k = nzb_w_inner(j,i)+1, nzt
274	w_l_l(k) = w_l_l(k) + w(k,j,i)
275	ENDDO
276	ENDDO
277	ENDDO
278	#if defined( __parallel )
279	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
280	CALL MPI_ALLREDUCE( w_l_l(1), w_l(1), nzt, MPI_REAL, MPI_SUM, &
281	comm2d, ierr )
282	#else
283	w_l = w_l_l
284	#endif
285	DO k = 1, nzt
286	w_l(k) = w_l(k) / ngp_2dh_outer(k,0)
287	ENDDO
288	DO i = nxlg, nxrg
289	DO j = nysg, nyng
290	DO k = nzb_w_inner(j,i)+1, nzt
291	w(k,j,i) = w(k,j,i) - w_l(k)
292	ENDDO
293	ENDDO
294	ENDDO
295	ENDIF
296	ENDIF
297
298	!
299	!-- Compute the divergence of the provisional velocity field.
300	CALL cpu_log( log_point_s(1), 'divergence', 'start' )
301
302	IF ( psolver == 'multigrid' ) THEN
303	!$OMP PARALLEL DO SCHEDULE( STATIC )
304	DO i = nxl-1, nxr+1
305	DO j = nys-1, nyn+1
306	DO k = nzb, nzt+1
307	d(k,j,i) = 0.0
308	ENDDO
309	ENDDO
310	ENDDO
311	ELSE
312	!$OMP PARALLEL DO SCHEDULE( STATIC )
313	!$acc kernels present( d )
314	!$acc loop
315	DO i = nxl, nxr
316	DO j = nys, nyn
317	!$acc loop vector(32)
318	DO k = nzb+1, nzt
319	d(k,j,i) = 0.0
320	ENDDO
321	ENDDO
322	ENDDO
323	!$acc end kernels
324	ENDIF
325
326	localsum = 0.0
327	threadsum = 0.0
328
329	#if defined( __ibm )
330	!$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum)
331	!$OMP DO SCHEDULE( STATIC )
332	DO i = nxl, nxr
333	DO j = nys, nyn
334	DO k = nzb_s_inner(j,i)+1, nzt
335	d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * ddx + &
336	( v(k,j+1,i) - v(k,j,i) ) * ddy + &
337	( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) ) * ddt_3d &
338	* d_weight_pres
339	ENDDO
340	!
341	!-- Compute possible PE-sum of divergences for flow_statistics
342	DO k = nzb_s_inner(j,i)+1, nzt
343	threadsum = threadsum + ABS( d(k,j,i) )
344	ENDDO
345
346	ENDDO
347	ENDDO
348
349	localsum = localsum + threadsum * dt_3d * &
350	weight_pres(intermediate_timestep_count)
351
352	!$OMP END PARALLEL
353	#else
354
355	!$OMP PARALLEL PRIVATE (i,j,k)
356	!$OMP DO SCHEDULE( STATIC )
357	!$acc kernels present( d, ddzw, nzb_s_inner, u, v, w )
358	!$acc loop
359	DO i = nxl, nxr
360	DO j = nys, nyn
361	!$acc loop vector(32)
362	DO k = 1, nzt
363	IF ( k > nzb_s_inner(j,i) ) THEN
364	d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * ddx + &
365	( v(k,j+1,i) - v(k,j,i) ) * ddy + &
366	( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) ) * ddt_3d &
367	* d_weight_pres
368	ENDIF
369	ENDDO
370	ENDDO
371	ENDDO
372	!$acc end kernels
373	!$OMP END PARALLEL
374
375	!
376	!-- Compute possible PE-sum of divergences for flow_statistics
377	!$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum)
378	!$OMP DO SCHEDULE( STATIC )
379	DO i = nxl, nxr
380	DO j = nys, nyn
381	DO k = nzb+1, nzt
382	threadsum = threadsum + ABS( d(k,j,i) )
383	ENDDO
384	ENDDO
385	ENDDO
386	localsum = localsum + threadsum * dt_3d * &
387	weight_pres(intermediate_timestep_count)
388	!$OMP END PARALLEL
389	#endif
390
391	!
392	!-- For completeness, set the divergence sum of all statistic regions to those
393	!-- of the total domain
394	sums_divold_l(0:statistic_regions) = localsum
395
396	CALL cpu_log( log_point_s(1), 'divergence', 'stop' )
397
398	!
399	!-- Compute the pressure perturbation solving the Poisson equation
400	IF ( psolver(1:7) == 'poisfft' ) THEN
401
402	!
403	!-- Solve Poisson equation via FFT and solution of tridiagonal matrices
404	IF ( psolver == 'poisfft' ) THEN
405	!
406	!-- Solver for 2d-decomposition
407	CALL poisfft( d, tend )
408
409	ELSEIF ( psolver == 'poisfft_hybrid' ) THEN
410	!
411	!-- Solver for 1d-decomposition (using MPI and OpenMP).
412	!-- The old hybrid-solver is still included here, as long as there
413	!-- are some optimization problems in poisfft
414	CALL poisfft_hybrid( d )
415
416	ENDIF
417
418	!
419	!-- Store computed perturbation pressure and set boundary condition in
420	!-- z-direction
421	!$OMP PARALLEL DO
422	!$acc kernels present( d, tend )
423	!$acc loop
424	DO i = nxl, nxr
425	DO j = nys, nyn
426	!$acc loop vector( 32 )
427	DO k = nzb+1, nzt
428	tend(k,j,i) = d(k,j,i)
429	ENDDO
430	ENDDO
431	ENDDO
432	!$acc end kernels
433
434	!
435	!-- Bottom boundary:
436	!-- This condition is only required for internal output. The pressure
437	!-- gradient (dp(nzb+1)-dp(nzb))/dz is not used anywhere else.
438	IF ( ibc_p_b == 1 ) THEN
439	!
440	!-- Neumann (dp/dz = 0)
441	!$OMP PARALLEL DO
442	!$acc kernels present( nzb_s_inner, tend )
443	DO i = nxlg, nxrg
444	DO j = nysg, nyng
445	tend(nzb_s_inner(j,i),j,i) = tend(nzb_s_inner(j,i)+1,j,i)
446	ENDDO
447	ENDDO
448	!$acc end kernels
449
450	ELSE
451	!
452	!-- Dirichlet
453	!$OMP PARALLEL DO
454	!$acc kernels present( tend )
455	DO i = nxlg, nxrg
456	DO j = nysg, nyng
457	tend(nzb_s_inner(j,i),j,i) = 0.0
458	ENDDO
459	ENDDO
460	!$acc end kernels
461
462	ENDIF
463
464	!
465	!-- Top boundary
466	IF ( ibc_p_t == 1 ) THEN
467	!
468	!-- Neumann
469	!$OMP PARALLEL DO
470	!$acc kernels present( tend )
471	DO i = nxlg, nxrg
472	DO j = nysg, nyng
473	tend(nzt+1,j,i) = tend(nzt,j,i)
474	ENDDO
475	ENDDO
476	!$acc end kernels
477
478	ELSE
479	!
480	!-- Dirichlet
481	!$OMP PARALLEL DO
482	!$acc kernels present( tend )
483	DO i = nxlg, nxrg
484	DO j = nysg, nyng
485	tend(nzt+1,j,i) = 0.0
486	ENDDO
487	ENDDO
488	!$acc end kernels
489
490	ENDIF
491
492	!
493	!-- Exchange boundaries for p
494	IF ( numprocs == 1 ) THEN ! workaround for single-core GPU runs
495	on_device = .TRUE. ! to be removed after complete porting
496	ELSE ! of ghost point exchange
497	!$acc update host( tend )
498	ENDIF
499	CALL exchange_horiz( tend, nbgp )
500	IF ( numprocs == 1 ) THEN ! workaround for single-core GPU runs
501	on_device = .FALSE. ! to be removed after complete porting
502	ELSE ! of ghost point exchange
503	!$acc update device( tend )
504	ENDIF
505
506	ELSEIF ( psolver == 'sor' ) THEN
507
508	!
509	!-- Solve Poisson equation for perturbation pressure using SOR-Red/Black
510	!-- scheme
511	CALL sor( d, ddzu_pres, ddzw, p_loc )
512	tend = p_loc
513
514	ELSEIF ( psolver == 'multigrid' ) THEN
515
516	!
517	!-- Solve Poisson equation for perturbation pressure using Multigrid scheme,
518	!-- array tend is used to store the residuals, logical exchange_mg is used
519	!-- to discern data exchange in multigrid ( 1 ghostpoint ) and normal grid
520	!-- ( nbgp ghost points ).
521
522	!-- If the number of grid points of the gathered grid, which is collected
523	!-- on PE0, is larger than the number of grid points of an PE, than array
524	!-- tend will be enlarged.
525	IF ( gathered_size > subdomain_size ) THEN
526	DEALLOCATE( tend )
527	ALLOCATE( tend(nzb:nzt_mg(mg_switch_to_pe0_level)+1,nys_mg( &
528	mg_switch_to_pe0_level)-1:nyn_mg(mg_switch_to_pe0_level)+1,&
529	nxl_mg(mg_switch_to_pe0_level)-1:nxr_mg( &
530	mg_switch_to_pe0_level)+1) )
531	ENDIF
532
533	CALL poismg( tend )
534
535	IF ( gathered_size > subdomain_size ) THEN
536	DEALLOCATE( tend )
537	ALLOCATE( tend(nzb:nzt+1,nysg:nyng,nxlg:nxrg) )
538	ENDIF
539
540	!
541	!-- Restore perturbation pressure on tend because this array is used
542	!-- further below to correct the velocity fields
543	DO i = nxl-1, nxr+1
544	DO j = nys-1, nyn+1
545	DO k = nzb, nzt+1
546	tend(k,j,i) = p_loc(k,j,i)
547	ENDDO
548	ENDDO
549	ENDDO
550
551	ENDIF
552
553	!
554	!-- Store perturbation pressure on array p, used for pressure data output.
555	!-- Ghost layers are added in the output routines (except sor-method: see below)
556	IF ( intermediate_timestep_count == 1 ) THEN
557	!$OMP PARALLEL PRIVATE (i,j,k)
558	!$OMP DO
559	!$acc kernels present( p, tend, weight_substep )
560	!$acc loop
561	DO i = nxl-1, nxr+1
562	DO j = nys-1, nyn+1
563	!$acc loop vector( 32 )
564	DO k = nzb, nzt+1
565	p(k,j,i) = tend(k,j,i) * &
566	weight_substep(intermediate_timestep_count)
567	ENDDO
568	ENDDO
569	ENDDO
570	!$acc end kernels
571	!$OMP END PARALLEL
572
573	ELSE
574	!$OMP PARALLEL PRIVATE (i,j,k)
575	!$OMP DO
576	!$acc kernels present( p, tend, weight_substep )
577	!$acc loop
578	DO i = nxl-1, nxr+1
579	DO j = nys-1, nyn+1
580	!$acc loop vector( 32 )
581	DO k = nzb, nzt+1
582	p(k,j,i) = p(k,j,i) + tend(k,j,i) * &
583	weight_substep(intermediate_timestep_count)
584	ENDDO
585	ENDDO
586	ENDDO
587	!$acc end kernels
588	!$OMP END PARALLEL
589
590	ENDIF
591
592	!
593	!-- SOR-method needs ghost layers for the next timestep
594	IF ( psolver == 'sor' ) CALL exchange_horiz( p, nbgp )
595
596	!
597	!-- Correction of the provisional velocities with the current perturbation
598	!-- pressure just computed
599	!$acc update host( u, v, w )
600	IF ( conserve_volume_flow .AND. ( bc_lr_cyc .OR. bc_ns_cyc ) ) THEN
601	volume_flow_l(1) = 0.0
602	volume_flow_l(2) = 0.0
603	ENDIF
604
605	!$OMP PARALLEL PRIVATE (i,j,k)
606	!$OMP DO
607	!$acc kernels present( ddzu, nzb_u_inner, nzb_v_inner, nzb_w_inner, tend, u, v, w, weight_pres )
608	!$acc loop
609	DO i = nxl, nxr
610	DO j = nys, nyn
611	!$acc loop vector( 32 )
612	DO k = 1, nzt
613	IF ( k > nzb_w_inner(j,i) ) THEN
614	w(k,j,i) = w(k,j,i) - dt_3d * &
615	( tend(k+1,j,i) - tend(k,j,i) ) * ddzu(k+1) * &
616	weight_pres(intermediate_timestep_count)
617	ENDIF
618	ENDDO
619	!$acc loop vector( 32 )
620	DO k = 1, nzt
621	IF ( k > nzb_u_inner(j,i) ) THEN
622	u(k,j,i) = u(k,j,i) - dt_3d * &
623	( tend(k,j,i) - tend(k,j,i-1) ) * ddx * &
624	weight_pres(intermediate_timestep_count)
625	ENDIF
626	ENDDO
627	!$acc loop vector( 32 )
628	DO k = 1, nzt
629	IF ( k > nzb_v_inner(j,i) ) THEN
630	v(k,j,i) = v(k,j,i) - dt_3d * &
631	( tend(k,j,i) - tend(k,j-1,i) ) * ddy * &
632	weight_pres(intermediate_timestep_count)
633	ENDIF
634	ENDDO
635
636	ENDDO
637	ENDDO
638	!$acc end kernels
639	!$OMP END PARALLEL
640
641	!
642	!-- Sum up the volume flow through the right and north boundary
643	IF ( conserve_volume_flow .AND. bc_lr_cyc .AND. bc_ns_cyc .AND. &
644	nxr == nx ) THEN
645
646	!$OMP PARALLEL PRIVATE (j,k)
647	!$OMP DO
648	DO j = nys, nyn
649	!$OMP CRITICAL
650	DO k = nzb_2d(j,nx) + 1, nzt
651	volume_flow_l(1) = volume_flow_l(1) + u(k,j,nx) * dzw(k)
652	ENDDO
653	!$OMP END CRITICAL
654	ENDDO
655	!$OMP END PARALLEL
656
657	ENDIF
658
659	IF ( conserve_volume_flow .AND. bc_ns_cyc .AND. bc_lr_cyc .AND. &
660	nyn == ny ) THEN
661
662	!$OMP PARALLEL PRIVATE (i,k)
663	!$OMP DO
664	DO i = nxl, nxr
665	!$OMP CRITICAL
666	DO k = nzb_2d(ny,i) + 1, nzt
667	volume_flow_l(2) = volume_flow_l(2) + v(k,ny,i) * dzw(k)
668	ENDDO
669	!$OMP END CRITICAL
670	ENDDO
671	!$OMP END PARALLEL
672
673	ENDIF
674
675	!
676	!-- Conserve the volume flow
677	IF ( conserve_volume_flow .AND. ( bc_lr_cyc .AND. bc_ns_cyc ) ) THEN
678
679	#if defined( __parallel )
680	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
681	CALL MPI_ALLREDUCE( volume_flow_l(1), volume_flow(1), 2, MPI_REAL, &
682	MPI_SUM, comm2d, ierr )
683	#else
684	volume_flow = volume_flow_l
685	#endif
686
687	volume_flow_offset = ( volume_flow_initial - volume_flow ) / &
688	volume_flow_area
689
690	!$OMP PARALLEL PRIVATE (i,j,k)
691	!$OMP DO
692	DO i = nxl, nxr
693	DO j = nys, nyn
694	DO k = nzb_u_inner(j,i) + 1, nzt
695	u(k,j,i) = u(k,j,i) + volume_flow_offset(1)
696	ENDDO
697	DO k = nzb_v_inner(j,i) + 1, nzt
698	v(k,j,i) = v(k,j,i) + volume_flow_offset(2)
699	ENDDO
700	ENDDO
701	ENDDO
702
703	!$OMP END PARALLEL
704
705	ENDIF
706
707	!
708	!-- Exchange of boundaries for the velocities
709	IF ( numprocs == 1 ) THEN ! workaround for single-core GPU runs
710	on_device = .TRUE. ! to be removed after complete porting
711	ELSE ! of ghost point exchange
712	!$acc update host( u, v, w )
713	ENDIF
714	CALL exchange_horiz( u, nbgp )
715	CALL exchange_horiz( v, nbgp )
716	CALL exchange_horiz( w, nbgp )
717	IF ( numprocs == 1 ) THEN ! workaround for single-core GPU runs
718	on_device = .FALSE. ! to be removed after complete porting
719	ELSE ! of ghost point exchange
720	!$acc update device( u, v, w )
721	ENDIF
722
723	!
724	!-- Compute the divergence of the corrected velocity field,
725	!-- a possible PE-sum is computed in flow_statistics
726	CALL cpu_log( log_point_s(1), 'divergence', 'start' )
727	sums_divnew_l = 0.0
728
729	!
730	!-- d must be reset to zero because it can contain nonzero values below the
731	!-- topography
732	IF ( topography /= 'flat' ) d = 0.0
733
734	localsum = 0.0
735	threadsum = 0.0
736
737	!$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum)
738	!$OMP DO SCHEDULE( STATIC )
739	#if defined( __ibm )
740	DO i = nxl, nxr
741	DO j = nys, nyn
742	DO k = nzb_s_inner(j,i)+1, nzt
743	d(k,j,i) = ( u(k,j,i+1) - u(k,j,i) ) * ddx + &
744	( v(k,j+1,i) - v(k,j,i) ) * ddy + &
745	( w(k,j,i) - w(k-1,j,i) ) * ddzw(k)
746	ENDDO
747	DO k = nzb+1, nzt
748	threadsum = threadsum + ABS( d(k,j,i) )
749	ENDDO
750	ENDDO
751	ENDDO
752	#else
753	!$acc kernels present( d, ddzw, nzb_s_inner, u, v, w )
754	!$acc loop
755	DO i = nxl, nxr
756	DO j = nys, nyn
757	!$acc loop vector( 32 )
758	DO k = 1, nzt
759	IF ( k > nzb_s_inner(j,i) ) THEN
760	d(k,j,i) = ( u(k,j,i+1) - u(k,j,i) ) * ddx + &
761	( v(k,j+1,i) - v(k,j,i) ) * ddy + &
762	( w(k,j,i) - w(k-1,j,i) ) * ddzw(k)
763	ENDIF
764	ENDDO
765	ENDDO
766	ENDDO
767	!$acc end kernels
768	!
769	!-- Compute possible PE-sum of divergences for flow_statistics
770	DO i = nxl, nxr
771	DO j = nys, nyn
772	DO k = nzb_s_inner(j,i)+1, nzt
773	threadsum = threadsum + ABS( d(k,j,i) )
774	ENDDO
775	ENDDO
776	ENDDO
777	#endif
778
779	localsum = localsum + threadsum
780	!$OMP END PARALLEL
781
782	!
783	!-- For completeness, set the divergence sum of all statistic regions to those
784	!-- of the total domain
785	sums_divnew_l(0:statistic_regions) = localsum
786
787	CALL cpu_log( log_point_s(1), 'divergence', 'stop' )
788
789	CALL cpu_log( log_point(8), 'pres', 'stop' )
790
791
792	END SUBROUTINE pres

Note: See TracBrowser for help on using the repository browser.

Download in other formats:

| Impressum | ©Leibniz Universität Hannover |