Home

Context Navigation

source: palm/trunk/SOURCE/pres.f90 @ 1113

Last change on this file since 1113 was 1113, checked in by raasch, 11 years ago
GPU porting of boundary conditions and routine pres; index bug removec from radiation boundary condition
Property svn:keywords set to `Id`
File size: 24.1 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	! GPU-porting of several loops, some loops rearranged
23	!
24	! Former revisions:
25	! -----------------
26	! $Id: pres.f90 1113 2013-03-10 02:48:14Z raasch $
27	!
28	! 1111 2013-03-08 23:54:10Z
29	! openACC statements added,
30	! ibc_p_b = 2 removed
31	!
32	! 1092 2013-02-02 11:24:22Z raasch
33	! unused variables removed
34	!
35	! 1036 2012-10-22 13:43:42Z raasch
36	! code put under GPL (PALM 3.9)
37	!
38	! 1003 2012-09-14 14:35:53Z raasch
39	! adjustment of array tend for cases with unequal subdomain sizes removed
40	!
41	! 778 2011-11-07 14:18:25Z fricke
42	! New allocation of tend when multigrid is used and the collected field on PE0
43	! has more grid points than the subdomain of an PE.
44	!
45	! 719 2011-04-06 13:05:23Z gryschka
46	! Bugfix in volume flow control for double cyclic boundary conditions
47	!
48	! 709 2011-03-30 09:31:40Z raasch
49	! formatting adjustments
50	!
51	! 707 2011-03-29 11:39:40Z raasch
52	! Calculation of weighted average of p is now handled in the same way
53	! regardless of the number of ghost layers (advection scheme),
54	! multigrid and sor method are using p_loc for iterative advancements of
55	! pressure,
56	! localsum calculation modified for proper OpenMP reduction,
57	! bc_lr/ns replaced by bc_lr/ns_cyc
58	!
59	! 693 2011-03-08 09:..:..Z raasch
60	! bugfix: weighting coefficient added to ibm branch
61	!
62	! 680 2011-02-04 23:16:06Z gryschka
63	! bugfix: collective_wait
64	!
65	! 675 2011-01-19 10:56:55Z suehring
66	! Removed bugfix while copying tend.
67	!
68	! 673 2011-01-18 16:19:48Z suehring
69	! Weighting coefficients added for right computation of the pressure during
70	! Runge-Kutta substeps.
71	!
72	! 667 2010-12-23 12:06:00Z suehring/gryschka
73	! New allocation of tend when ws-scheme and multigrid is used. This is due to
74	! reasons of perforance of the data_exchange. The same is done with p after
75	! poismg is called.
76	! nxl-1, nxr+1, nys-1, nyn+1 replaced by nxlg, nxrg, nysg, nyng when no
77	! multigrid is used. Calls of exchange_horiz are modified.
78	! bugfix: After pressure correction no volume flow correction in case of
79	! non-cyclic boundary conditions
80	! (has to be done only before pressure correction)
81	! Call of SOR routine is referenced with ddzu_pres.
82	!
83	! 622 2010-12-10 08:08:13Z raasch
84	! optional barriers included in order to speed up collective operations
85	!
86	! 151 2008-03-07 13:42:18Z raasch
87	! Bugfix in volume flow control for non-cyclic boundary conditions
88	!
89	! 106 2007-08-16 14:30:26Z raasch
90	! Volume flow conservation added for the remaining three outflow boundaries
91	!
92	! 85 2007-05-11 09:35:14Z raasch
93	! Division through dt_3d replaced by multiplication of the inverse.
94	! For performance optimisation, this is done in the loop calculating the
95	! divergence instead of using a seperate loop.
96	!
97	! 75 2007-03-22 09:54:05Z raasch
98	! Volume flow control for non-cyclic boundary conditions added (currently only
99	! for the north boundary!!), 2nd+3rd argument removed from exchange horiz,
100	! mean vertical velocity is removed in case of Neumann boundary conditions
101	! both at the bottom and the top
102	!
103	! RCS Log replace by Id keyword, revision history cleaned up
104	!
105	! Revision 1.25 2006/04/26 13:26:12 raasch
106	! OpenMP optimization (+localsum, threadsum)
107	!
108	! Revision 1.1 1997/07/24 11:24:44 raasch
109	! Initial revision
110	!
111	!
112	! Description:
113	! ------------
114	! Compute the divergence of the provisional velocity field. Solve the Poisson
115	! equation for the perturbation pressure. Compute the final velocities using
116	! this perturbation pressure. Compute the remaining divergence.
117	!------------------------------------------------------------------------------!
118
119	USE arrays_3d
120	USE constants
121	USE control_parameters
122	USE cpulog
123	USE grid_variables
124	USE indices
125	USE interfaces
126	USE pegrid
127	USE poisfft_mod
128	USE poisfft_hybrid_mod
129	USE statistics
130
131	IMPLICIT NONE
132
133	INTEGER :: i, j, k
134
135	REAL :: ddt_3d, localsum, threadsum, d_weight_pres
136
137	REAL, DIMENSION(1:2) :: volume_flow_l, volume_flow_offset
138	REAL, DIMENSION(1:nzt) :: w_l, w_l_l
139
140
141	CALL cpu_log( log_point(8), 'pres', 'start' )
142
143
144	ddt_3d = 1.0 / dt_3d
145	d_weight_pres = 1.0 / weight_pres(intermediate_timestep_count)
146
147	!
148	!-- Multigrid method expects array d to have one ghost layer.
149	!--
150	IF ( psolver == 'multigrid' ) THEN
151
152	DEALLOCATE( d )
153	ALLOCATE( d(nzb:nzt+1,nys-1:nyn+1,nxl-1:nxr+1) )
154
155	!
156	!-- Since p is later used to hold the weighted average of the substeps, it
157	!-- cannot be used in the iterative solver. Therefore, its initial value is
158	!-- stored on p_loc, which is then iteratively advanced in every substep.
159	IF ( intermediate_timestep_count == 1 ) THEN
160	DO i = nxl-1, nxr+1
161	DO j = nys-1, nyn+1
162	DO k = nzb, nzt+1
163	p_loc(k,j,i) = p(k,j,i)
164	ENDDO
165	ENDDO
166	ENDDO
167	ENDIF
168
169	ELSEIF ( psolver == 'sor' .AND. intermediate_timestep_count == 1 ) THEN
170
171	!
172	!-- Since p is later used to hold the weighted average of the substeps, it
173	!-- cannot be used in the iterative solver. Therefore, its initial value is
174	!-- stored on p_loc, which is then iteratively advanced in every substep.
175	p_loc = p
176
177	ENDIF
178
179	!
180	!-- Conserve the volume flow at the outflow in case of non-cyclic lateral
181	!-- boundary conditions
182	!-- WARNING: so far, this conservation does not work at the left/south
183	!-- boundary if the topography at the inflow differs from that at the
184	!-- outflow! For this case, volume_flow_area needs adjustment!
185	!
186	!-- Left/right
187	IF ( conserve_volume_flow .AND. ( outflow_l .OR. outflow_r ) ) THEN
188
189	volume_flow(1) = 0.0
190	volume_flow_l(1) = 0.0
191
192	IF ( outflow_l ) THEN
193	i = 0
194	ELSEIF ( outflow_r ) THEN
195	i = nx+1
196	ENDIF
197
198	DO j = nys, nyn
199	!
200	!-- Sum up the volume flow through the south/north boundary
201	DO k = nzb_2d(j,i)+1, nzt
202	volume_flow_l(1) = volume_flow_l(1) + u(k,j,i) * dzw(k)
203	ENDDO
204	ENDDO
205
206	#if defined( __parallel )
207	IF ( collective_wait ) CALL MPI_BARRIER( comm1dy, ierr )
208	CALL MPI_ALLREDUCE( volume_flow_l(1), volume_flow(1), 1, MPI_REAL, &
209	MPI_SUM, comm1dy, ierr )
210	#else
211	volume_flow = volume_flow_l
212	#endif
213	volume_flow_offset(1) = ( volume_flow_initial(1) - volume_flow(1) ) &
214	/ volume_flow_area(1)
215
216	DO j = nysg, nyng
217	DO k = nzb_2d(j,i)+1, nzt
218	u(k,j,i) = u(k,j,i) + volume_flow_offset(1)
219	ENDDO
220	ENDDO
221
222	ENDIF
223
224	!
225	!-- South/north
226	IF ( conserve_volume_flow .AND. ( outflow_n .OR. outflow_s ) ) THEN
227
228	volume_flow(2) = 0.0
229	volume_flow_l(2) = 0.0
230
231	IF ( outflow_s ) THEN
232	j = 0
233	ELSEIF ( outflow_n ) THEN
234	j = ny+1
235	ENDIF
236
237	DO i = nxl, nxr
238	!
239	!-- Sum up the volume flow through the south/north boundary
240	DO k = nzb_2d(j,i)+1, nzt
241	volume_flow_l(2) = volume_flow_l(2) + v(k,j,i) * dzw(k)
242	ENDDO
243	ENDDO
244
245	#if defined( __parallel )
246	IF ( collective_wait ) CALL MPI_BARRIER( comm1dx, ierr )
247	CALL MPI_ALLREDUCE( volume_flow_l(2), volume_flow(2), 1, MPI_REAL, &
248	MPI_SUM, comm1dx, ierr )
249	#else
250	volume_flow = volume_flow_l
251	#endif
252	volume_flow_offset(2) = ( volume_flow_initial(2) - volume_flow(2) ) &
253	/ volume_flow_area(2)
254
255	DO i = nxlg, nxrg
256	DO k = nzb_v_inner(j,i)+1, nzt
257	v(k,j,i) = v(k,j,i) + volume_flow_offset(2)
258	ENDDO
259	ENDDO
260
261	ENDIF
262
263	!
264	!-- Remove mean vertical velocity
265	IF ( ibc_p_b == 1 .AND. ibc_p_t == 1 ) THEN
266	IF ( simulated_time > 0.0 ) THEN ! otherwise nzb_w_inner not yet known
267	w_l = 0.0; w_l_l = 0.0
268	DO i = nxl, nxr
269	DO j = nys, nyn
270	DO k = nzb_w_inner(j,i)+1, nzt
271	w_l_l(k) = w_l_l(k) + w(k,j,i)
272	ENDDO
273	ENDDO
274	ENDDO
275	#if defined( __parallel )
276	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
277	CALL MPI_ALLREDUCE( w_l_l(1), w_l(1), nzt, MPI_REAL, MPI_SUM, &
278	comm2d, ierr )
279	#else
280	w_l = w_l_l
281	#endif
282	DO k = 1, nzt
283	w_l(k) = w_l(k) / ngp_2dh_outer(k,0)
284	ENDDO
285	DO i = nxlg, nxrg
286	DO j = nysg, nyng
287	DO k = nzb_w_inner(j,i)+1, nzt
288	w(k,j,i) = w(k,j,i) - w_l(k)
289	ENDDO
290	ENDDO
291	ENDDO
292	ENDIF
293	ENDIF
294
295	!
296	!-- Compute the divergence of the provisional velocity field.
297	CALL cpu_log( log_point_s(1), 'divergence', 'start' )
298
299	IF ( psolver == 'multigrid' ) THEN
300	!$OMP PARALLEL DO SCHEDULE( STATIC )
301	DO i = nxl-1, nxr+1
302	DO j = nys-1, nyn+1
303	DO k = nzb, nzt+1
304	d(k,j,i) = 0.0
305	ENDDO
306	ENDDO
307	ENDDO
308	ELSE
309	!$OMP PARALLEL DO SCHEDULE( STATIC )
310	!$acc kernels present( d )
311	!$acc loop
312	DO i = nxl, nxr
313	DO j = nys, nyn
314	!$acc loop vector(32)
315	DO k = nzb+1, nzt
316	d(k,j,i) = 0.0
317	ENDDO
318	ENDDO
319	ENDDO
320	!$acc end kernels
321	ENDIF
322
323	localsum = 0.0
324	threadsum = 0.0
325
326	#if defined( __ibm )
327	!$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum)
328	!$OMP DO SCHEDULE( STATIC )
329	DO i = nxl, nxr
330	DO j = nys, nyn
331	DO k = nzb_s_inner(j,i)+1, nzt
332	d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * ddx + &
333	( v(k,j+1,i) - v(k,j,i) ) * ddy + &
334	( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) ) * ddt_3d &
335	* d_weight_pres
336	ENDDO
337	!
338	!-- Compute possible PE-sum of divergences for flow_statistics
339	DO k = nzb_s_inner(j,i)+1, nzt
340	threadsum = threadsum + ABS( d(k,j,i) )
341	ENDDO
342
343	ENDDO
344	ENDDO
345
346	localsum = localsum + threadsum * dt_3d * &
347	weight_pres(intermediate_timestep_count)
348
349	!$OMP END PARALLEL
350	#else
351
352	!$OMP PARALLEL PRIVATE (i,j,k)
353	!$OMP DO SCHEDULE( STATIC )
354	!$acc kernels present( d, ddzw, nzb_s_inner, u, v, w )
355	!$acc loop
356	DO i = nxl, nxr
357	DO j = nys, nyn
358	!$acc loop vector(32)
359	DO k = 1, nzt
360	IF ( k > nzb_s_inner(j,i) ) THEN
361	d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * ddx + &
362	( v(k,j+1,i) - v(k,j,i) ) * ddy + &
363	( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) ) * ddt_3d &
364	* d_weight_pres
365	ENDIF
366	ENDDO
367	ENDDO
368	ENDDO
369	!$acc end kernels
370	!$OMP END PARALLEL
371
372	!
373	!-- Compute possible PE-sum of divergences for flow_statistics
374	!$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum)
375	!$OMP DO SCHEDULE( STATIC )
376	DO i = nxl, nxr
377	DO j = nys, nyn
378	DO k = nzb+1, nzt
379	threadsum = threadsum + ABS( d(k,j,i) )
380	ENDDO
381	ENDDO
382	ENDDO
383	localsum = localsum + threadsum * dt_3d * &
384	weight_pres(intermediate_timestep_count)
385	!$OMP END PARALLEL
386	#endif
387
388	!
389	!-- For completeness, set the divergence sum of all statistic regions to those
390	!-- of the total domain
391	sums_divold_l(0:statistic_regions) = localsum
392
393	CALL cpu_log( log_point_s(1), 'divergence', 'stop' )
394
395	!
396	!-- Compute the pressure perturbation solving the Poisson equation
397	IF ( psolver(1:7) == 'poisfft' ) THEN
398
399	!
400	!-- Solve Poisson equation via FFT and solution of tridiagonal matrices
401	IF ( psolver == 'poisfft' ) THEN
402	!
403	!-- Solver for 2d-decomposition
404	CALL poisfft( d, tend )
405
406	ELSEIF ( psolver == 'poisfft_hybrid' ) THEN
407	!
408	!-- Solver for 1d-decomposition (using MPI and OpenMP).
409	!-- The old hybrid-solver is still included here, as long as there
410	!-- are some optimization problems in poisfft
411	CALL poisfft_hybrid( d )
412
413	ENDIF
414
415	!
416	!-- Store computed perturbation pressure and set boundary condition in
417	!-- z-direction
418	!$OMP PARALLEL DO
419	!$acc kernels present( d, tend )
420	!$acc loop
421	DO i = nxl, nxr
422	DO j = nys, nyn
423	!$acc loop vector( 32 )
424	DO k = nzb+1, nzt
425	tend(k,j,i) = d(k,j,i)
426	ENDDO
427	ENDDO
428	ENDDO
429	!$acc end kernels
430
431	!
432	!-- Bottom boundary:
433	!-- This condition is only required for internal output. The pressure
434	!-- gradient (dp(nzb+1)-dp(nzb))/dz is not used anywhere else.
435	IF ( ibc_p_b == 1 ) THEN
436	!
437	!-- Neumann (dp/dz = 0)
438	!$OMP PARALLEL DO
439	!$acc kernels present( nzb_s_inner, tend )
440	DO i = nxlg, nxrg
441	DO j = nysg, nyng
442	tend(nzb_s_inner(j,i),j,i) = tend(nzb_s_inner(j,i)+1,j,i)
443	ENDDO
444	ENDDO
445	!$acc end kernels
446
447	ELSE
448	!
449	!-- Dirichlet
450	!$OMP PARALLEL DO
451	!$acc kernels present( tend )
452	DO i = nxlg, nxrg
453	DO j = nysg, nyng
454	tend(nzb_s_inner(j,i),j,i) = 0.0
455	ENDDO
456	ENDDO
457	!$acc end kernels
458
459	ENDIF
460
461	!
462	!-- Top boundary
463	IF ( ibc_p_t == 1 ) THEN
464	!
465	!-- Neumann
466	!$OMP PARALLEL DO
467	!$acc kernels present( tend )
468	DO i = nxlg, nxrg
469	DO j = nysg, nyng
470	tend(nzt+1,j,i) = tend(nzt,j,i)
471	ENDDO
472	ENDDO
473	!$acc end kernels
474
475	ELSE
476	!
477	!-- Dirichlet
478	!$OMP PARALLEL DO
479	!$acc kernels present( tend )
480	DO i = nxlg, nxrg
481	DO j = nysg, nyng
482	tend(nzt+1,j,i) = 0.0
483	ENDDO
484	ENDDO
485	!$acc end kernels
486
487	ENDIF
488
489	!
490	!-- Exchange boundaries for p
491	IF ( numprocs == 1 ) THEN ! workaround for single-core GPU runs
492	on_device = .TRUE. ! to be removed after complete porting
493	ELSE ! of ghost point exchange
494	!$acc update host( tend )
495	ENDIF
496	CALL exchange_horiz( tend, nbgp )
497	IF ( numprocs == 1 ) THEN ! workaround for single-core GPU runs
498	on_device = .FALSE. ! to be removed after complete porting
499	ELSE ! of ghost point exchange
500	!$acc update device( tend )
501	ENDIF
502
503	ELSEIF ( psolver == 'sor' ) THEN
504
505	!
506	!-- Solve Poisson equation for perturbation pressure using SOR-Red/Black
507	!-- scheme
508	CALL sor( d, ddzu_pres, ddzw, p_loc )
509	tend = p_loc
510
511	ELSEIF ( psolver == 'multigrid' ) THEN
512
513	!
514	!-- Solve Poisson equation for perturbation pressure using Multigrid scheme,
515	!-- array tend is used to store the residuals, logical exchange_mg is used
516	!-- to discern data exchange in multigrid ( 1 ghostpoint ) and normal grid
517	!-- ( nbgp ghost points ).
518
519	!-- If the number of grid points of the gathered grid, which is collected
520	!-- on PE0, is larger than the number of grid points of an PE, than array
521	!-- tend will be enlarged.
522	IF ( gathered_size > subdomain_size ) THEN
523	DEALLOCATE( tend )
524	ALLOCATE( tend(nzb:nzt_mg(mg_switch_to_pe0_level)+1,nys_mg( &
525	mg_switch_to_pe0_level)-1:nyn_mg(mg_switch_to_pe0_level)+1,&
526	nxl_mg(mg_switch_to_pe0_level)-1:nxr_mg( &
527	mg_switch_to_pe0_level)+1) )
528	ENDIF
529
530	CALL poismg( tend )
531
532	IF ( gathered_size > subdomain_size ) THEN
533	DEALLOCATE( tend )
534	ALLOCATE( tend(nzb:nzt+1,nysg:nyng,nxlg:nxrg) )
535	ENDIF
536
537	!
538	!-- Restore perturbation pressure on tend because this array is used
539	!-- further below to correct the velocity fields
540	DO i = nxl-1, nxr+1
541	DO j = nys-1, nyn+1
542	DO k = nzb, nzt+1
543	tend(k,j,i) = p_loc(k,j,i)
544	ENDDO
545	ENDDO
546	ENDDO
547
548	ENDIF
549
550	!
551	!-- Store perturbation pressure on array p, used for pressure data output.
552	!-- Ghost layers are added in the output routines (except sor-method: see below)
553	IF ( intermediate_timestep_count == 1 ) THEN
554	!$OMP PARALLEL PRIVATE (i,j,k)
555	!$OMP DO
556	!$acc kernels present( p, tend, weight_substep )
557	!$acc loop
558	DO i = nxl-1, nxr+1
559	DO j = nys-1, nyn+1
560	!$acc loop vector( 32 )
561	DO k = nzb, nzt+1
562	p(k,j,i) = tend(k,j,i) * &
563	weight_substep(intermediate_timestep_count)
564	ENDDO
565	ENDDO
566	ENDDO
567	!$acc end kernels
568	!$OMP END PARALLEL
569
570	ELSE
571	!$OMP PARALLEL PRIVATE (i,j,k)
572	!$OMP DO
573	!$acc kernels present( p, tend, weight_substep )
574	!$acc loop
575	DO i = nxl-1, nxr+1
576	DO j = nys-1, nyn+1
577	!$acc loop vector( 32 )
578	DO k = nzb, nzt+1
579	p(k,j,i) = p(k,j,i) + tend(k,j,i) * &
580	weight_substep(intermediate_timestep_count)
581	ENDDO
582	ENDDO
583	ENDDO
584	!$acc end kernels
585	!$OMP END PARALLEL
586
587	ENDIF
588
589	!
590	!-- SOR-method needs ghost layers for the next timestep
591	IF ( psolver == 'sor' ) CALL exchange_horiz( p, nbgp )
592
593	!
594	!-- Correction of the provisional velocities with the current perturbation
595	!-- pressure just computed
596	!$acc update host( u, v, w )
597	IF ( conserve_volume_flow .AND. ( bc_lr_cyc .OR. bc_ns_cyc ) ) THEN
598	volume_flow_l(1) = 0.0
599	volume_flow_l(2) = 0.0
600	ENDIF
601
602	!$OMP PARALLEL PRIVATE (i,j,k)
603	!$OMP DO
604	!$acc kernels present( ddzu, nzb_u_inner, nzb_v_inner, nzb_w_inner, tend, u, v, w, weight_pres )
605	!$acc loop
606	DO i = nxl, nxr
607	DO j = nys, nyn
608	!$acc loop vector( 32 )
609	DO k = 1, nzt
610	IF ( k > nzb_w_inner(j,i) ) THEN
611	w(k,j,i) = w(k,j,i) - dt_3d * &
612	( tend(k+1,j,i) - tend(k,j,i) ) * ddzu(k+1) * &
613	weight_pres(intermediate_timestep_count)
614	ENDIF
615	ENDDO
616	!$acc loop vector( 32 )
617	DO k = 1, nzt
618	IF ( k > nzb_u_inner(j,i) ) THEN
619	u(k,j,i) = u(k,j,i) - dt_3d * &
620	( tend(k,j,i) - tend(k,j,i-1) ) * ddx * &
621	weight_pres(intermediate_timestep_count)
622	ENDIF
623	ENDDO
624	!$acc loop vector( 32 )
625	DO k = 1, nzt
626	IF ( k > nzb_v_inner(j,i) ) THEN
627	v(k,j,i) = v(k,j,i) - dt_3d * &
628	( tend(k,j,i) - tend(k,j-1,i) ) * ddy * &
629	weight_pres(intermediate_timestep_count)
630	ENDIF
631	ENDDO
632
633	ENDDO
634	ENDDO
635	!$acc end kernels
636	!$OMP END PARALLEL
637
638	!
639	!-- Sum up the volume flow through the right and north boundary
640	IF ( conserve_volume_flow .AND. bc_lr_cyc .AND. bc_ns_cyc .AND. &
641	nxr == nx ) THEN
642
643	!$OMP PARALLEL PRIVATE (j,k)
644	!$OMP DO
645	DO j = nys, nyn
646	!$OMP CRITICAL
647	DO k = nzb_2d(j,nx) + 1, nzt
648	volume_flow_l(1) = volume_flow_l(1) + u(k,j,nx) * dzw(k)
649	ENDDO
650	!$OMP END CRITICAL
651	ENDDO
652	!$OMP END PARALLEL
653
654	ENDIF
655
656	IF ( conserve_volume_flow .AND. bc_ns_cyc .AND. bc_lr_cyc .AND. &
657	nyn == ny ) THEN
658
659	!$OMP PARALLEL PRIVATE (i,k)
660	!$OMP DO
661	DO i = nxl, nxr
662	!$OMP CRITICAL
663	DO k = nzb_2d(ny,i) + 1, nzt
664	volume_flow_l(2) = volume_flow_l(2) + v(k,ny,i) * dzw(k)
665	ENDDO
666	!$OMP END CRITICAL
667	ENDDO
668	!$OMP END PARALLEL
669
670	ENDIF
671
672	!
673	!-- Conserve the volume flow
674	IF ( conserve_volume_flow .AND. ( bc_lr_cyc .AND. bc_ns_cyc ) ) THEN
675
676	#if defined( __parallel )
677	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
678	CALL MPI_ALLREDUCE( volume_flow_l(1), volume_flow(1), 2, MPI_REAL, &
679	MPI_SUM, comm2d, ierr )
680	#else
681	volume_flow = volume_flow_l
682	#endif
683
684	volume_flow_offset = ( volume_flow_initial - volume_flow ) / &
685	volume_flow_area
686
687	!$OMP PARALLEL PRIVATE (i,j,k)
688	!$OMP DO
689	DO i = nxl, nxr
690	DO j = nys, nyn
691	DO k = nzb_u_inner(j,i) + 1, nzt
692	u(k,j,i) = u(k,j,i) + volume_flow_offset(1)
693	ENDDO
694	DO k = nzb_v_inner(j,i) + 1, nzt
695	v(k,j,i) = v(k,j,i) + volume_flow_offset(2)
696	ENDDO
697	ENDDO
698	ENDDO
699
700	!$OMP END PARALLEL
701
702	ENDIF
703
704	!
705	!-- Exchange of boundaries for the velocities
706	IF ( numprocs == 1 ) THEN ! workaround for single-core GPU runs
707	on_device = .TRUE. ! to be removed after complete porting
708	ELSE ! of ghost point exchange
709	!$acc update host( u, v, w )
710	ENDIF
711	CALL exchange_horiz( u, nbgp )
712	CALL exchange_horiz( v, nbgp )
713	CALL exchange_horiz( w, nbgp )
714	IF ( numprocs == 1 ) THEN ! workaround for single-core GPU runs
715	on_device = .FALSE. ! to be removed after complete porting
716	ELSE ! of ghost point exchange
717	!$acc update device( u, v, w )
718	ENDIF
719
720	!
721	!-- Compute the divergence of the corrected velocity field,
722	!-- a possible PE-sum is computed in flow_statistics
723	CALL cpu_log( log_point_s(1), 'divergence', 'start' )
724	sums_divnew_l = 0.0
725
726	!
727	!-- d must be reset to zero because it can contain nonzero values below the
728	!-- topography
729	IF ( topography /= 'flat' ) d = 0.0
730
731	localsum = 0.0
732	threadsum = 0.0
733
734	!$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum)
735	!$OMP DO SCHEDULE( STATIC )
736	#if defined( __ibm )
737	DO i = nxl, nxr
738	DO j = nys, nyn
739	DO k = nzb_s_inner(j,i)+1, nzt
740	d(k,j,i) = ( u(k,j,i+1) - u(k,j,i) ) * ddx + &
741	( v(k,j+1,i) - v(k,j,i) ) * ddy + &
742	( w(k,j,i) - w(k-1,j,i) ) * ddzw(k)
743	ENDDO
744	DO k = nzb+1, nzt
745	threadsum = threadsum + ABS( d(k,j,i) )
746	ENDDO
747	ENDDO
748	ENDDO
749	#else
750	!$acc kernels present( d, ddzw, nzb_s_inner, u, v, w )
751	!$acc loop
752	DO i = nxl, nxr
753	DO j = nys, nyn
754	!$acc loop vector( 32 )
755	DO k = 1, nzt
756	IF ( k > nzb_s_inner(j,i) ) THEN
757	d(k,j,i) = ( u(k,j,i+1) - u(k,j,i) ) * ddx + &
758	( v(k,j+1,i) - v(k,j,i) ) * ddy + &
759	( w(k,j,i) - w(k-1,j,i) ) * ddzw(k)
760	ENDIF
761	ENDDO
762	ENDDO
763	ENDDO
764	!$acc end kernels
765	!
766	!-- Compute possible PE-sum of divergences for flow_statistics
767	!$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum)
768	!$OMP DO SCHEDULE( STATIC )
769	DO i = nxl, nxr
770	DO j = nys, nyn
771	DO k = nzb_s_inner(j,i)+1, nzt
772	threadsum = threadsum + ABS( d(k,j,i) )
773	ENDDO
774	ENDDO
775	ENDDO
776	#endif
777
778	localsum = localsum + threadsum
779	!$OMP END PARALLEL
780
781	!
782	!-- For completeness, set the divergence sum of all statistic regions to those
783	!-- of the total domain
784	sums_divnew_l(0:statistic_regions) = localsum
785
786	CALL cpu_log( log_point_s(1), 'divergence', 'stop' )
787
788	CALL cpu_log( log_point(8), 'pres', 'stop' )
789
790
791	END SUBROUTINE pres

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

Download in other formats:

| Impressum | ©Leibniz Universität Hannover |