Home

Context Navigation

source: palm/trunk/SOURCE/init_pegrid.f90 @ 766

Last change on this file since 766 was 760, checked in by raasch, 13 years ago
last commit documented
Property svn:keywords set to `Id`
File size: 41.8 KB

Line
1	SUBROUTINE init_pegrid
2
3	!------------------------------------------------------------------------------!
4	! Current revisions:
5	! -----------------
6	!
7	!
8	! ATTENTION: nnz_x undefined problem still has to be solved!!!!!!!!
9	! TEST OUTPUT (TO BE REMOVED) logging mpi2 ierr values
10	!
11	! Former revisions:
12	! -----------------
13	! $Id: init_pegrid.f90 760 2011-09-15 14:37:54Z raasch $
14	!
15	! 759 2011-09-15 13:58:31Z raasch
16	!
17	! calculation of number of io_blocks and the io_group to which the respective
18	! PE belongs
19	!
20	! 755 2011-08-29 09:55:16Z witha
21	! 2d-decomposition is default for lcflow (ForWind cluster in Oldenburg)
22	!
23	! 722 2011-04-11 06:21:09Z raasch
24	! Bugfix: bc_lr/ns_cyc/dirrad/raddir replaced by bc_lr/ns, because variables
25	! are not yet set here; grid_level set to 0
26	!
27	! 709 2011-03-30 09:31:40Z raasch
28	! formatting adjustments
29	!
30	! 707 2011-03-29 11:39:40Z raasch
31	! bc_lr/ns replaced by bc_lr/ns_cyc/dirrad/raddir
32	!
33	! 667 2010-12-23 12:06:00Z suehring/gryschka
34	! Moved determination of target_id's from init_coupling
35	! Determination of parameters needed for coupling (coupling_topology, ngp_a,
36	! ngp_o) with different grid/processor-topology in ocean and atmosphere
37	! Adaption of ngp_xy, ngp_y to a dynamic number of ghost points.
38	! The maximum_grid_level changed from 1 to 0. 0 is the normal grid, 1 to
39	! maximum_grid_level the grids for multigrid, in which 0 and 1 are normal grids.
40	! This distinction is due to reasons of data exchange and performance for the
41	! normal grid and grids in poismg.
42	! The definition of MPI-Vectors adapted to a dynamic numer of ghost points.
43	! New MPI-Vectors for data exchange between left and right boundaries added.
44	! This is due to reasons of performance (10% faster).
45	!
46	! 646 2010-12-15 13:03:52Z raasch
47	! lctit is now using a 2d decomposition by default
48	!
49	! 622 2010-12-10 08:08:13Z raasch
50	! optional barriers included in order to speed up collective operations
51	!
52	! 438 2010-02-01 04:32:43Z raasch
53	! 2d-decomposition is default for Cray-XT machines
54	!
55	! 274 2009-03-26 15:11:21Z heinze
56	! Output of messages replaced by message handling routine.
57	!
58	! 206 2008-10-13 14:59:11Z raasch
59	! Implementation of a MPI-1 coupling: added __parallel within the __mpi2 part
60	! 2d-decomposition is default on SGI-ICE systems
61	!
62	! 197 2008-09-16 15:29:03Z raasch
63	! multigrid levels are limited by subdomains if mg_switch_to_pe0_level = -1,
64	! nz is used instead nnz for calculating mg-levels
65	! Collect on PE0 horizontal index bounds from all other PEs,
66	! broadcast the id of the inflow PE (using the respective communicator)
67	!
68	! 114 2007-10-10 00:03:15Z raasch
69	! Allocation of wall flag arrays for multigrid solver
70	!
71	! 108 2007-08-24 15:10:38Z letzel
72	! Intercommunicator (comm_inter) and derived data type (type_xy) for
73	! coupled model runs created, assign coupling_mode_remote,
74	! indices nxlu and nysv are calculated (needed for non-cyclic boundary
75	! conditions)
76	!
77	! 82 2007-04-16 15:40:52Z raasch
78	! Cpp-directive lcmuk changed to intel_openmp_bug, setting of host on lcmuk by
79	! cpp-directive removed
80	!
81	! 75 2007-03-22 09:54:05Z raasch
82	! uxrp, vynp eliminated,
83	! dirichlet/neumann changed to dirichlet/radiation, etc.,
84	! poisfft_init is only called if fft-solver is switched on
85	!
86	! RCS Log replace by Id keyword, revision history cleaned up
87	!
88	! Revision 1.28 2006/04/26 13:23:32 raasch
89	! lcmuk does not understand the !$ comment so a cpp-directive is required
90	!
91	! Revision 1.1 1997/07/24 11:15:09 raasch
92	! Initial revision
93	!
94	!
95	! Description:
96	! ------------
97	! Determination of the virtual processor topology (if not prescribed by the
98	! user)and computation of the grid point number and array bounds of the local
99	! domains.
100	!------------------------------------------------------------------------------!
101
102	USE control_parameters
103	USE fft_xy
104	USE grid_variables
105	USE indices
106	USE pegrid
107	USE poisfft_mod
108	USE poisfft_hybrid_mod
109	USE statistics
110	USE transpose_indices
111
112
113
114	IMPLICIT NONE
115
116	INTEGER :: gathered_size, i, id_inflow_l, id_recycling_l, ind(5), j, k, &
117	maximum_grid_level_l, mg_switch_to_pe0_level_l, mg_levels_x, &
118	mg_levels_y, mg_levels_z, nnx_y, nnx_z, nny_x, nny_z, nnz_x, &
119	nnz_y, numproc_sqr, nx_total, nxl_l, nxr_l, nyn_l, nys_l, &
120	nzb_l, nzt_l, omp_get_num_threads, subdomain_size
121
122	INTEGER, DIMENSION(:), ALLOCATABLE :: ind_all, nxlf, nxrf, nynf, nysf
123
124	INTEGER, DIMENSION(2) :: pdims_remote
125
126	LOGICAL :: found
127
128	!
129	!-- Get the number of OpenMP threads
130	!$OMP PARALLEL
131	#if defined( __intel_openmp_bug )
132	threads_per_task = omp_get_num_threads()
133	#else
134	!$ threads_per_task = omp_get_num_threads()
135	#endif
136	!$OMP END PARALLEL
137
138
139	#if defined( __parallel )
140
141	!
142	!-- Determine the processor topology or check it, if prescribed by the user
143	IF ( npex == -1 .AND. npey == -1 ) THEN
144
145	!
146	!-- Automatic determination of the topology
147	!-- The default on SMP- and cluster-hosts is a 1d-decomposition along x
148	IF ( host(1:3) == 'ibm' .OR. host(1:3) == 'nec' .OR. &
149	( host(1:2) == 'lc' .AND. host(3:5) /= 'sgi' .AND. &
150	host(3:4) /= 'xt' .AND. host(3:5) /= 'tit' .AND. &
151	host(3:6) /= 'flow' ) .OR. host(1:3) == 'dec' ) THEN
152
153	pdims(1) = numprocs
154	pdims(2) = 1
155
156	ELSE
157
158	numproc_sqr = SQRT( REAL( numprocs ) )
159	pdims(1) = MAX( numproc_sqr , 1 )
160	DO WHILE ( MOD( numprocs , pdims(1) ) /= 0 )
161	pdims(1) = pdims(1) - 1
162	ENDDO
163	pdims(2) = numprocs / pdims(1)
164
165	ENDIF
166
167	ELSEIF ( npex /= -1 .AND. npey /= -1 ) THEN
168
169	!
170	!-- Prescribed by user. Number of processors on the prescribed topology
171	!-- must be equal to the number of PEs available to the job
172	IF ( ( npex * npey ) /= numprocs ) THEN
173	WRITE( message_string, * ) 'number of PEs of the prescribed ', &
174	'topology (', npex*npey,') does not match & the number of ', &
175	'PEs available to the job (', numprocs, ')'
176	CALL message( 'init_pegrid', 'PA0221', 1, 2, 0, 6, 0 )
177	ENDIF
178	pdims(1) = npex
179	pdims(2) = npey
180
181	ELSE
182	!
183	!-- If the processor topology is prescribed by the user, the number of
184	!-- PEs must be given in both directions
185	message_string = 'if the processor topology is prescribed by the, ' // &
186	' user& both values of "npex" and "npey" must be given ' // &
187	'in the &NAMELIST-parameter file'
188	CALL message( 'init_pegrid', 'PA0222', 1, 2, 0, 6, 0 )
189
190	ENDIF
191
192	!
193	!-- The hybrid solver can only be used in case of a 1d-decomposition along x
194	IF ( pdims(2) /= 1 .AND. psolver == 'poisfft_hybrid' ) THEN
195	message_string = 'psolver = "poisfft_hybrid" can only be' // &
196	'& used in case of a 1d-decomposition along x'
197	CALL message( 'init_pegrid', 'PA0223', 1, 2, 0, 6, 0 )
198	ENDIF
199
200	!
201	!-- For communication speedup, set barriers in front of collective
202	!-- communications by default on SGI-type systems
203	IF ( host(3:5) == 'sgi' ) collective_wait = .TRUE.
204
205	!
206	!-- If necessary, set horizontal boundary conditions to non-cyclic
207	IF ( bc_lr /= 'cyclic' ) cyclic(1) = .FALSE.
208	IF ( bc_ns /= 'cyclic' ) cyclic(2) = .FALSE.
209
210	!
211	!-- Create the virtual processor grid
212	CALL MPI_CART_CREATE( comm_palm, ndim, pdims, cyclic, reorder, &
213	comm2d, ierr )
214	CALL MPI_COMM_RANK( comm2d, myid, ierr )
215	WRITE (myid_char,'(''_'',I4.4)') myid
216
217	CALL MPI_CART_COORDS( comm2d, myid, ndim, pcoord, ierr )
218	CALL MPI_CART_SHIFT( comm2d, 0, 1, pleft, pright, ierr )
219	CALL MPI_CART_SHIFT( comm2d, 1, 1, psouth, pnorth, ierr )
220
221	!
222	!-- Determine sub-topologies for transpositions
223	!-- Transposition from z to x:
224	remain_dims(1) = .TRUE.
225	remain_dims(2) = .FALSE.
226	CALL MPI_CART_SUB( comm2d, remain_dims, comm1dx, ierr )
227	CALL MPI_COMM_RANK( comm1dx, myidx, ierr )
228	!
229	!-- Transposition from x to y
230	remain_dims(1) = .FALSE.
231	remain_dims(2) = .TRUE.
232	CALL MPI_CART_SUB( comm2d, remain_dims, comm1dy, ierr )
233	CALL MPI_COMM_RANK( comm1dy, myidy, ierr )
234
235
236	!
237	!-- Find a grid (used for array d) which will match the transposition demands
238	IF ( grid_matching == 'strict' ) THEN
239
240	nxa = nx; nya = ny; nza = nz
241
242	ELSE
243
244	found = .FALSE.
245	xn: DO nxa = nx, 2*nx
246	!
247	!-- Meet conditions for nx
248	IF ( MOD( nxa+1, pdims(1) ) /= 0 .OR. &
249	MOD( nxa+1, pdims(2) ) /= 0 ) CYCLE xn
250
251	yn: DO nya = ny, 2*ny
252	!
253	!-- Meet conditions for ny
254	IF ( MOD( nya+1, pdims(2) ) /= 0 .OR. &
255	MOD( nya+1, pdims(1) ) /= 0 ) CYCLE yn
256
257
258	zn: DO nza = nz, 2*nz
259	!
260	!-- Meet conditions for nz
261	IF ( ( MOD( nza, pdims(1) ) /= 0 .AND. pdims(1) /= 1 .AND. &
262	pdims(2) /= 1 ) .OR. &
263	( MOD( nza, pdims(2) ) /= 0 .AND. dt_dosp /= 9999999.9 &
264	) ) THEN
265	CYCLE zn
266	ELSE
267	found = .TRUE.
268	EXIT xn
269	ENDIF
270
271	ENDDO zn
272
273	ENDDO yn
274
275	ENDDO xn
276
277	IF ( .NOT. found ) THEN
278	message_string = 'no matching grid for transpositions found'
279	CALL message( 'init_pegrid', 'PA0224', 1, 2, 0, 6, 0 )
280	ENDIF
281
282	ENDIF
283
284	!
285	!-- Calculate array bounds in x-direction for every PE.
286	!-- The last PE along x may get less grid points than the others
287	ALLOCATE( nxlf(0:pdims(1)-1), nxrf(0:pdims(1)-1), nynf(0:pdims(2)-1), &
288	nysf(0:pdims(2)-1), nnx_pe(0:pdims(1)-1), nny_pe(0:pdims(2)-1) )
289
290	IF ( MOD( nxa+1 , pdims(1) ) /= 0 ) THEN
291	WRITE( message_string, * ) 'x-direction: gridpoint number (',nx+1,') ',&
292	'is not an& integral divisor of the number ', &
293	'processors (', pdims(1),')'
294	CALL message( 'init_pegrid', 'PA0225', 1, 2, 0, 6, 0 )
295	ELSE
296	nnx = ( nxa + 1 ) / pdims(1)
297	IF ( nnx*pdims(1) - ( nx + 1) > nnx ) THEN
298	WRITE( message_string, * ) 'x-direction: nx does not match the', &
299	'requirements given by the number of PEs &used', &
300	'& please use nx = ', nx - ( pdims(1) - ( nnx*pdims(1) &
301	- ( nx + 1 ) ) ), ' instead of nx =', nx
302	CALL message( 'init_pegrid', 'PA0226', 1, 2, 0, 6, 0 )
303	ENDIF
304	ENDIF
305
306	!
307	!-- Left and right array bounds, number of gridpoints
308	DO i = 0, pdims(1)-1
309	nxlf(i) = i * nnx
310	nxrf(i) = ( i + 1 ) * nnx - 1
311	nnx_pe(i) = MIN( nx, nxrf(i) ) - nxlf(i) + 1
312	ENDDO
313
314	!
315	!-- Calculate array bounds in y-direction for every PE.
316	IF ( MOD( nya+1 , pdims(2) ) /= 0 ) THEN
317	WRITE( message_string, * ) 'y-direction: gridpoint number (',ny+1,') ', &
318	'is not an& integral divisor of the number of', &
319	'processors (', pdims(2),')'
320	CALL message( 'init_pegrid', 'PA0227', 1, 2, 0, 6, 0 )
321	ELSE
322	nny = ( nya + 1 ) / pdims(2)
323	IF ( nny*pdims(2) - ( ny + 1) > nny ) THEN
324	WRITE( message_string, * ) 'y-direction: ny does not match the', &
325	'requirements given by the number of PEs &used ', &
326	'& please use ny = ', ny - ( pdims(2) - ( nnx*pdims(2) &
327	- ( ny + 1 ) ) ), ' instead of ny =', ny
328	CALL message( 'init_pegrid', 'PA0228', 1, 2, 0, 6, 0 )
329	ENDIF
330	ENDIF
331
332	!
333	!-- South and north array bounds
334	DO j = 0, pdims(2)-1
335	nysf(j) = j * nny
336	nynf(j) = ( j + 1 ) * nny - 1
337	nny_pe(j) = MIN( ny, nynf(j) ) - nysf(j) + 1
338	ENDDO
339
340	!
341	!-- Local array bounds of the respective PEs
342	nxl = nxlf(pcoord(1))
343	nxra = nxrf(pcoord(1))
344	nxr = MIN( nx, nxra )
345	nys = nysf(pcoord(2))
346	nyna = nynf(pcoord(2))
347	nyn = MIN( ny, nyna )
348	nzb = 0
349	nzta = nza
350	nzt = MIN( nz, nzta )
351	nnz = nza
352
353	!
354	!-- Set switches to define if the PE is situated at the border of the virtual
355	!-- processor grid
356	IF ( nxl == 0 ) left_border_pe = .TRUE.
357	IF ( nxr == nx ) right_border_pe = .TRUE.
358	IF ( nys == 0 ) south_border_pe = .TRUE.
359	IF ( nyn == ny ) north_border_pe = .TRUE.
360
361	!
362	!-- Calculate array bounds and gridpoint numbers for the transposed arrays
363	!-- (needed in the pressure solver)
364	!-- For the transposed arrays, cyclic boundaries as well as top and bottom
365	!-- boundaries are omitted, because they are obstructive to the transposition
366
367	!
368	!-- 1. transposition z --> x
369	!-- This transposition is not neccessary in case of a 1d-decomposition along x,
370	!-- except that the uptream-spline method is switched on
371	IF ( pdims(2) /= 1 .OR. momentum_advec == 'ups-scheme' .OR. &
372	scalar_advec == 'ups-scheme' ) THEN
373
374	IF ( pdims(2) == 1 .AND. ( momentum_advec == 'ups-scheme' .OR. &
375	scalar_advec == 'ups-scheme' ) ) THEN
376	message_string = '1d-decomposition along x ' // &
377	'chosen but nz restrictions may occur' // &
378	'& since ups-scheme is activated'
379	CALL message( 'init_pegrid', 'PA0229', 0, 1, 0, 6, 0 )
380	ENDIF
381	nys_x = nys
382	nyn_xa = nyna
383	nyn_x = nyn
384	nny_x = nny
385	IF ( MOD( nza , pdims(1) ) /= 0 ) THEN
386	WRITE( message_string, * ) 'transposition z --> x:', &
387	'&nz=',nz,' is not an integral divisior of pdims(1)=', &
388	pdims(1)
389	CALL message( 'init_pegrid', 'PA0230', 1, 2, 0, 6, 0 )
390	ENDIF
391	nnz_x = nza / pdims(1)
392	nzb_x = 1 + myidx * nnz_x
393	nzt_xa = ( myidx + 1 ) * nnz_x
394	nzt_x = MIN( nzt, nzt_xa )
395
396	sendrecvcount_zx = nnx * nny * nnz_x
397
398	ELSE
399	!
400	!--- Setting of dummy values because otherwise variables are undefined in
401	!--- the next step x --> y
402	!--- WARNING: This case has still to be clarified!!!!!!!!!!!!
403	nnz_x = 1
404	nzb_x = 1
405	nzt_xa = 1
406	nzt_x = 1
407	nny_x = nny
408
409	ENDIF
410
411	!
412	!-- 2. transposition x --> y
413	nnz_y = nnz_x
414	nzb_y = nzb_x
415	nzt_ya = nzt_xa
416	nzt_y = nzt_x
417	IF ( MOD( nxa+1 , pdims(2) ) /= 0 ) THEN
418	WRITE( message_string, * ) 'transposition x --> y:', &
419	'&nx+1=',nx+1,' is not an integral divisor of ',&
420	'pdims(2)=',pdims(2)
421	CALL message( 'init_pegrid', 'PA0231', 1, 2, 0, 6, 0 )
422	ENDIF
423	nnx_y = (nxa+1) / pdims(2)
424	nxl_y = myidy * nnx_y
425	nxr_ya = ( myidy + 1 ) * nnx_y - 1
426	nxr_y = MIN( nx, nxr_ya )
427
428	sendrecvcount_xy = nnx_y * nny_x * nnz_y
429
430	!
431	!-- 3. transposition y --> z (ELSE: x --> y in case of 1D-decomposition
432	!-- along x)
433	IF ( pdims(2) /= 1 .OR. momentum_advec == 'ups-scheme' .OR. &
434	scalar_advec == 'ups-scheme' ) THEN
435	!
436	!-- y --> z
437	!-- This transposition is not neccessary in case of a 1d-decomposition
438	!-- along x, except that the uptream-spline method is switched on
439	nnx_z = nnx_y
440	nxl_z = nxl_y
441	nxr_za = nxr_ya
442	nxr_z = nxr_y
443	IF ( MOD( nya+1 , pdims(1) ) /= 0 ) THEN
444	WRITE( message_string, * ) 'transposition y --> z:', &
445	'& ny+1=',ny+1,' is not an integral divisor of ',&
446	'pdims(1)=',pdims(1)
447	CALL message( 'init_pegrid', 'PA0232', 1, 2, 0, 6, 0 )
448	ENDIF
449	nny_z = (nya+1) / pdims(1)
450	nys_z = myidx * nny_z
451	nyn_za = ( myidx + 1 ) * nny_z - 1
452	nyn_z = MIN( ny, nyn_za )
453
454	sendrecvcount_yz = nnx_y * nny_z * nnz_y
455
456	ELSE
457	!
458	!-- x --> y. This condition must be fulfilled for a 1D-decomposition along x
459	IF ( MOD( nya+1 , pdims(1) ) /= 0 ) THEN
460	WRITE( message_string, * ) 'transposition x --> y:', &
461	'& ny+1=',ny+1,' is not an integral divisor of ',&
462	'pdims(1)=',pdims(1)
463	CALL message( 'init_pegrid', 'PA0233', 1, 2, 0, 6, 0 )
464	ENDIF
465
466	ENDIF
467
468	!
469	!-- Indices for direct transpositions z --> y (used for calculating spectra)
470	IF ( dt_dosp /= 9999999.9 ) THEN
471	IF ( MOD( nza, pdims(2) ) /= 0 ) THEN
472	WRITE( message_string, * ) 'direct transposition z --> y (needed ', &
473	'for spectra):& nz=',nz,' is not an integral divisor of ',&
474	'pdims(2)=',pdims(2)
475	CALL message( 'init_pegrid', 'PA0234', 1, 2, 0, 6, 0 )
476	ELSE
477	nxl_yd = nxl
478	nxr_yda = nxra
479	nxr_yd = nxr
480	nzb_yd = 1 + myidy * ( nza / pdims(2) )
481	nzt_yda = ( myidy + 1 ) * ( nza / pdims(2) )
482	nzt_yd = MIN( nzt, nzt_yda )
483
484	sendrecvcount_zyd = nnx * nny * ( nza / pdims(2) )
485	ENDIF
486	ENDIF
487
488	!
489	!-- Indices for direct transpositions y --> x (they are only possible in case
490	!-- of a 1d-decomposition along x)
491	IF ( pdims(2) == 1 ) THEN
492	nny_x = nny / pdims(1)
493	nys_x = myid * nny_x
494	nyn_xa = ( myid + 1 ) * nny_x - 1
495	nyn_x = MIN( ny, nyn_xa )
496	nzb_x = 1
497	nzt_xa = nza
498	nzt_x = nz
499	sendrecvcount_xy = nnx * nny_x * nza
500	ENDIF
501
502	!
503	!-- Indices for direct transpositions x --> y (they are only possible in case
504	!-- of a 1d-decomposition along y)
505	IF ( pdims(1) == 1 ) THEN
506	nnx_y = nnx / pdims(2)
507	nxl_y = myid * nnx_y
508	nxr_ya = ( myid + 1 ) * nnx_y - 1
509	nxr_y = MIN( nx, nxr_ya )
510	nzb_y = 1
511	nzt_ya = nza
512	nzt_y = nz
513	sendrecvcount_xy = nnx_y * nny * nza
514	ENDIF
515
516	!
517	!-- Arrays for storing the array bounds are needed any more
518	DEALLOCATE( nxlf , nxrf , nynf , nysf )
519
520	!
521	!-- Collect index bounds from other PEs (to be written to restart file later)
522	ALLOCATE( hor_index_bounds(4,0:numprocs-1) )
523
524	IF ( myid == 0 ) THEN
525
526	hor_index_bounds(1,0) = nxl
527	hor_index_bounds(2,0) = nxr
528	hor_index_bounds(3,0) = nys
529	hor_index_bounds(4,0) = nyn
530
531	!
532	!-- Receive data from all other PEs
533	DO i = 1, numprocs-1
534	CALL MPI_RECV( ibuf, 4, MPI_INTEGER, i, MPI_ANY_TAG, comm2d, status, &
535	ierr )
536	hor_index_bounds(:,i) = ibuf(1:4)
537	ENDDO
538
539	ELSE
540	!
541	!-- Send index bounds to PE0
542	ibuf(1) = nxl
543	ibuf(2) = nxr
544	ibuf(3) = nys
545	ibuf(4) = nyn
546	CALL MPI_SEND( ibuf, 4, MPI_INTEGER, 0, myid, comm2d, ierr )
547
548	ENDIF
549
550	#if defined( __print )
551	!
552	!-- Control output
553	IF ( myid == 0 ) THEN
554	PRINT, ' processor topology *'
555	PRINT*, ' '
556	PRINT*, 'myid pcoord left right south north idx idy nxl: nxr',&
557	&' nys: nyn'
558	PRINT*, '------------------------------------------------------------',&
559	&'-----------'
560	WRITE (*,1000) 0, pcoord(1), pcoord(2), pleft, pright, psouth, pnorth, &
561	myidx, myidy, nxl, nxr, nys, nyn
562	1000 FORMAT (I4,2X,'(',I3,',',I3,')',3X,I4,2X,I4,3X,I4,2X,I4,2X,I3,1X,I3, &
563	2(2X,I4,':',I4))
564
565	!
566	!-- Receive data from the other PEs
567	DO i = 1,numprocs-1
568	CALL MPI_RECV( ibuf, 12, MPI_INTEGER, i, MPI_ANY_TAG, comm2d, status, &
569	ierr )
570	WRITE (*,1000) i, ( ibuf(j) , j = 1,12 )
571	ENDDO
572	ELSE
573
574	!
575	!-- Send data to PE0
576	ibuf(1) = pcoord(1); ibuf(2) = pcoord(2); ibuf(3) = pleft
577	ibuf(4) = pright; ibuf(5) = psouth; ibuf(6) = pnorth; ibuf(7) = myidx
578	ibuf(8) = myidy; ibuf(9) = nxl; ibuf(10) = nxr; ibuf(11) = nys
579	ibuf(12) = nyn
580	CALL MPI_SEND( ibuf, 12, MPI_INTEGER, 0, myid, comm2d, ierr )
581	ENDIF
582	#endif
583
584	#if defined( __parallel )
585	#if defined( __mpi2 )
586	!
587	!-- In case of coupled runs, get the port name on PE0 of the atmosphere model
588	!-- and pass it to PE0 of the ocean model
589	IF ( myid == 0 ) THEN
590
591	IF ( coupling_mode == 'atmosphere_to_ocean' ) THEN
592
593	CALL MPI_OPEN_PORT( MPI_INFO_NULL, port_name, ierr )
594
595	CALL MPI_PUBLISH_NAME( 'palm_coupler', MPI_INFO_NULL, port_name, &
596	ierr )
597
598	!
599	!-- Write a flag file for the ocean model and the other atmosphere
600	!-- processes.
601	!-- There seems to be a bug in MPICH2 which causes hanging processes
602	!-- in case that execution of LOOKUP_NAME is continued too early
603	!-- (i.e. before the port has been created)
604	OPEN( 90, FILE='COUPLING_PORT_OPENED', FORM='FORMATTED' )
605	WRITE ( 90, '(''TRUE'')' )
606	CLOSE ( 90 )
607
608	ELSEIF ( coupling_mode == 'ocean_to_atmosphere' ) THEN
609
610	!
611	!-- Continue only if the atmosphere model has created the port.
612	!-- There seems to be a bug in MPICH2 which causes hanging processes
613	!-- in case that execution of LOOKUP_NAME is continued too early
614	!-- (i.e. before the port has been created)
615	INQUIRE( FILE='COUPLING_PORT_OPENED', EXIST=found )
616	DO WHILE ( .NOT. found )
617	INQUIRE( FILE='COUPLING_PORT_OPENED', EXIST=found )
618	ENDDO
619
620	CALL MPI_LOOKUP_NAME( 'palm_coupler', MPI_INFO_NULL, port_name, ierr )
621
622	ENDIF
623
624	ENDIF
625
626	!
627	!-- In case of coupled runs, establish the connection between the atmosphere
628	!-- and the ocean model and define the intercommunicator (comm_inter)
629	CALL MPI_BARRIER( comm2d, ierr )
630	IF ( coupling_mode == 'atmosphere_to_ocean' ) THEN
631
632	CALL MPI_COMM_ACCEPT( port_name, MPI_INFO_NULL, 0, MPI_COMM_WORLD, &
633	comm_inter, ierr )
634	coupling_mode_remote = 'ocean_to_atmosphere'
635
636	ELSEIF ( coupling_mode == 'ocean_to_atmosphere' ) THEN
637
638	CALL MPI_COMM_CONNECT( port_name, MPI_INFO_NULL, 0, MPI_COMM_WORLD, &
639	comm_inter, ierr )
640	coupling_mode_remote = 'atmosphere_to_ocean'
641
642	ENDIF
643	#endif
644
645	!
646	!-- Determine the number of ghost point layers
647	IF ( scalar_advec == 'ws-scheme' .OR. momentum_advec == 'ws-scheme' ) THEN
648	nbgp = 3
649	ELSE
650	nbgp = 1
651	ENDIF
652
653	!
654	!-- Create a new MPI derived datatype for the exchange of surface (xy) data,
655	!-- which is needed for coupled atmosphere-ocean runs.
656	!-- First, calculate number of grid points of an xy-plane.
657	ngp_xy = ( nxr - nxl + 1 + 2 * nbgp ) * ( nyn - nys + 1 + 2 * nbgp )
658	CALL MPI_TYPE_VECTOR( ngp_xy, 1, nzt-nzb+2, MPI_REAL, type_xy, ierr )
659	CALL MPI_TYPE_COMMIT( type_xy, ierr )
660
661	IF ( TRIM( coupling_mode ) /= 'uncoupled' ) THEN
662
663	!
664	!-- Pass the number of grid points of the atmosphere model to
665	!-- the ocean model and vice versa
666	IF ( coupling_mode == 'atmosphere_to_ocean' ) THEN
667
668	nx_a = nx
669	ny_a = ny
670
671	IF ( myid == 0 ) THEN
672
673	CALL MPI_SEND( nx_a, 1, MPI_INTEGER, numprocs, 1, comm_inter, &
674	ierr )
675	CALL MPI_SEND( ny_a, 1, MPI_INTEGER, numprocs, 2, comm_inter, &
676	ierr )
677	CALL MPI_SEND( pdims, 2, MPI_INTEGER, numprocs, 3, comm_inter, &
678	ierr )
679	CALL MPI_RECV( nx_o, 1, MPI_INTEGER, numprocs, 4, comm_inter, &
680	status, ierr )
681	CALL MPI_RECV( ny_o, 1, MPI_INTEGER, numprocs, 5, comm_inter, &
682	status, ierr )
683	CALL MPI_RECV( pdims_remote, 2, MPI_INTEGER, numprocs, 6, &
684	comm_inter, status, ierr )
685	ENDIF
686
687	CALL MPI_BCAST( nx_o, 1, MPI_INTEGER, 0, comm2d, ierr )
688	CALL MPI_BCAST( ny_o, 1, MPI_INTEGER, 0, comm2d, ierr )
689	CALL MPI_BCAST( pdims_remote, 2, MPI_INTEGER, 0, comm2d, ierr )
690
691	ELSEIF ( coupling_mode == 'ocean_to_atmosphere' ) THEN
692
693	nx_o = nx
694	ny_o = ny
695
696	IF ( myid == 0 ) THEN
697
698	CALL MPI_RECV( nx_a, 1, MPI_INTEGER, 0, 1, comm_inter, status, &
699	ierr )
700	CALL MPI_RECV( ny_a, 1, MPI_INTEGER, 0, 2, comm_inter, status, &
701	ierr )
702	CALL MPI_RECV( pdims_remote, 2, MPI_INTEGER, 0, 3, comm_inter, &
703	status, ierr )
704	CALL MPI_SEND( nx_o, 1, MPI_INTEGER, 0, 4, comm_inter, ierr )
705	CALL MPI_SEND( ny_o, 1, MPI_INTEGER, 0, 5, comm_inter, ierr )
706	CALL MPI_SEND( pdims, 2, MPI_INTEGER, 0, 6, comm_inter, ierr )
707	ENDIF
708
709	CALL MPI_BCAST( nx_a, 1, MPI_INTEGER, 0, comm2d, ierr)
710	CALL MPI_BCAST( ny_a, 1, MPI_INTEGER, 0, comm2d, ierr)
711	CALL MPI_BCAST( pdims_remote, 2, MPI_INTEGER, 0, comm2d, ierr)
712
713	ENDIF
714
715	ngp_a = ( nx_a+1 + 2 * nbgp ) * ( ny_a+1 + 2 * nbgp )
716	ngp_o = ( nx_o+1 + 2 * nbgp ) * ( ny_o+1 + 2 * nbgp )
717
718	!
719	!-- Determine if the horizontal grid and the number of PEs in ocean and
720	!-- atmosphere is same or not
721	IF ( nx_o == nx_a .AND. ny_o == ny_a .AND. &
722	pdims(1) == pdims_remote(1) .AND. pdims(2) == pdims_remote(2) ) &
723	THEN
724	coupling_topology = 0
725	ELSE
726	coupling_topology = 1
727	ENDIF
728
729	!
730	!-- Determine the target PEs for the exchange between ocean and
731	!-- atmosphere (comm2d)
732	IF ( coupling_topology == 0 ) THEN
733	!
734	!-- In case of identical topologies, every atmosphere PE has exactly one
735	!-- ocean PE counterpart and vice versa
736	IF ( TRIM( coupling_mode ) == 'atmosphere_to_ocean' ) THEN
737	target_id = myid + numprocs
738	ELSE
739	target_id = myid
740	ENDIF
741
742	ELSE
743	!
744	!-- In case of nonequivalent topology in ocean and atmosphere only for
745	!-- PE0 in ocean and PE0 in atmosphere a target_id is needed, since
746	!-- data echxchange between ocean and atmosphere will be done only
747	!-- between these PEs.
748	IF ( myid == 0 ) THEN
749
750	IF ( TRIM( coupling_mode ) == 'atmosphere_to_ocean' ) THEN
751	target_id = numprocs
752	ELSE
753	target_id = 0
754	ENDIF
755
756	ENDIF
757
758	ENDIF
759
760	ENDIF
761
762
763	#endif
764
765	#else
766
767	!
768	!-- Array bounds when running on a single PE (respectively a non-parallel
769	!-- machine)
770	nxl = 0
771	nxr = nx
772	nxra = nx
773	nnx = nxr - nxl + 1
774	nys = 0
775	nyn = ny
776	nyna = ny
777	nny = nyn - nys + 1
778	nzb = 0
779	nzt = nz
780	nzta = nz
781	nnz = nz
782
783	ALLOCATE( hor_index_bounds(4,0:0) )
784	hor_index_bounds(1,0) = nxl
785	hor_index_bounds(2,0) = nxr
786	hor_index_bounds(3,0) = nys
787	hor_index_bounds(4,0) = nyn
788
789	!
790	!-- Array bounds for the pressure solver (in the parallel code, these bounds
791	!-- are the ones for the transposed arrays)
792	nys_x = nys
793	nyn_x = nyn
794	nyn_xa = nyn
795	nzb_x = nzb + 1
796	nzt_x = nzt
797	nzt_xa = nzt
798
799	nxl_y = nxl
800	nxr_y = nxr
801	nxr_ya = nxr
802	nzb_y = nzb + 1
803	nzt_y = nzt
804	nzt_ya = nzt
805
806	nxl_z = nxl
807	nxr_z = nxr
808	nxr_za = nxr
809	nys_z = nys
810	nyn_z = nyn
811	nyn_za = nyn
812
813	#endif
814
815	!
816	!-- Calculate number of grid levels necessary for the multigrid poisson solver
817	!-- as well as the gridpoint indices on each level
818	IF ( psolver == 'multigrid' ) THEN
819
820	!
821	!-- First calculate number of possible grid levels for the subdomains
822	mg_levels_x = 1
823	mg_levels_y = 1
824	mg_levels_z = 1
825
826	i = nnx
827	DO WHILE ( MOD( i, 2 ) == 0 .AND. i /= 2 )
828	i = i / 2
829	mg_levels_x = mg_levels_x + 1
830	ENDDO
831
832	j = nny
833	DO WHILE ( MOD( j, 2 ) == 0 .AND. j /= 2 )
834	j = j / 2
835	mg_levels_y = mg_levels_y + 1
836	ENDDO
837
838	k = nz ! do not use nnz because it might be > nz due to transposition
839	! requirements
840	DO WHILE ( MOD( k, 2 ) == 0 .AND. k /= 2 )
841	k = k / 2
842	mg_levels_z = mg_levels_z + 1
843	ENDDO
844
845	maximum_grid_level = MIN( mg_levels_x, mg_levels_y, mg_levels_z )
846
847	!
848	!-- Find out, if the total domain allows more levels. These additional
849	!-- levels are identically processed on all PEs.
850	IF ( numprocs > 1 .AND. mg_switch_to_pe0_level /= -1 ) THEN
851
852	IF ( mg_levels_z > MIN( mg_levels_x, mg_levels_y ) ) THEN
853
854	mg_switch_to_pe0_level_l = maximum_grid_level
855
856	mg_levels_x = 1
857	mg_levels_y = 1
858
859	i = nx+1
860	DO WHILE ( MOD( i, 2 ) == 0 .AND. i /= 2 )
861	i = i / 2
862	mg_levels_x = mg_levels_x + 1
863	ENDDO
864
865	j = ny+1
866	DO WHILE ( MOD( j, 2 ) == 0 .AND. j /= 2 )
867	j = j / 2
868	mg_levels_y = mg_levels_y + 1
869	ENDDO
870
871	maximum_grid_level_l = MIN( mg_levels_x, mg_levels_y, mg_levels_z )
872
873	IF ( maximum_grid_level_l > mg_switch_to_pe0_level_l ) THEN
874	mg_switch_to_pe0_level_l = maximum_grid_level_l - &
875	mg_switch_to_pe0_level_l + 1
876	ELSE
877	mg_switch_to_pe0_level_l = 0
878	ENDIF
879
880	ELSE
881
882	mg_switch_to_pe0_level_l = 0
883	maximum_grid_level_l = maximum_grid_level
884
885	ENDIF
886
887	!
888	!-- Use switch level calculated above only if it is not pre-defined
889	!-- by user
890	IF ( mg_switch_to_pe0_level == 0 ) THEN
891
892	IF ( mg_switch_to_pe0_level_l /= 0 ) THEN
893	mg_switch_to_pe0_level = mg_switch_to_pe0_level_l
894	maximum_grid_level = maximum_grid_level_l
895	ENDIF
896
897	ELSE
898	!
899	!-- Check pre-defined value and reset to default, if neccessary
900	IF ( mg_switch_to_pe0_level < mg_switch_to_pe0_level_l .OR. &
901	mg_switch_to_pe0_level >= maximum_grid_level_l ) THEN
902	message_string = 'mg_switch_to_pe0_level ' // &
903	'out of range and reset to default (=0)'
904	CALL message( 'init_pegrid', 'PA0235', 0, 1, 0, 6, 0 )
905	mg_switch_to_pe0_level = 0
906	ELSE
907	!
908	!-- Use the largest number of possible levels anyway and recalculate
909	!-- the switch level to this largest number of possible values
910	maximum_grid_level = maximum_grid_level_l
911
912	ENDIF
913
914	ENDIF
915
916	ENDIF
917
918	ALLOCATE( grid_level_count(maximum_grid_level), &
919	nxl_mg(maximum_grid_level), nxr_mg(maximum_grid_level), &
920	nyn_mg(maximum_grid_level), nys_mg(maximum_grid_level), &
921	nzt_mg(maximum_grid_level) )
922
923	grid_level_count = 0
924	nxl_l = nxl; nxr_l = nxr; nys_l = nys; nyn_l = nyn; nzt_l = nzt
925
926	DO i = maximum_grid_level, 1 , -1
927
928	IF ( i == mg_switch_to_pe0_level ) THEN
929	#if defined( __parallel )
930	!
931	!-- Save the grid size of the subdomain at the switch level, because
932	!-- it is needed in poismg.
933	ind(1) = nxl_l; ind(2) = nxr_l
934	ind(3) = nys_l; ind(4) = nyn_l
935	ind(5) = nzt_l
936	ALLOCATE( ind_all(5*numprocs), mg_loc_ind(5,0:numprocs-1) )
937	CALL MPI_ALLGATHER( ind, 5, MPI_INTEGER, ind_all, 5, &
938	MPI_INTEGER, comm2d, ierr )
939	DO j = 0, numprocs-1
940	DO k = 1, 5
941	mg_loc_ind(k,j) = ind_all(k+j*5)
942	ENDDO
943	ENDDO
944	DEALLOCATE( ind_all )
945	!
946	!-- Calculate the grid size of the total domain
947	nxr_l = ( nxr_l-nxl_l+1 ) * pdims(1) - 1
948	nxl_l = 0
949	nyn_l = ( nyn_l-nys_l+1 ) * pdims(2) - 1
950	nys_l = 0
951	!
952	!-- The size of this gathered array must not be larger than the
953	!-- array tend, which is used in the multigrid scheme as a temporary
954	!-- array
955	subdomain_size = ( nxr - nxl + 3 ) * ( nyn - nys + 3 ) * &
956	( nzt - nzb + 2 )
957	gathered_size = ( nxr_l - nxl_l + 3 ) * ( nyn_l - nys_l + 3 ) * &
958	( nzt_l - nzb + 2 )
959
960	IF ( gathered_size > subdomain_size ) THEN
961	message_string = 'not enough memory for storing ' // &
962	'gathered multigrid data on PE0'
963	CALL message( 'init_pegrid', 'PA0236', 1, 2, 0, 6, 0 )
964	ENDIF
965	#else
966	message_string = 'multigrid gather/scatter impossible ' // &
967	'in non parallel mode'
968	CALL message( 'init_pegrid', 'PA0237', 1, 2, 0, 6, 0 )
969	#endif
970	ENDIF
971
972	nxl_mg(i) = nxl_l
973	nxr_mg(i) = nxr_l
974	nys_mg(i) = nys_l
975	nyn_mg(i) = nyn_l
976	nzt_mg(i) = nzt_l
977
978	nxl_l = nxl_l / 2
979	nxr_l = nxr_l / 2
980	nys_l = nys_l / 2
981	nyn_l = nyn_l / 2
982	nzt_l = nzt_l / 2
983	ENDDO
984
985	ELSE
986
987	maximum_grid_level = 0
988
989	ENDIF
990
991	!
992	!-- Default level 0 tells exchange_horiz that all ghost planes have to be
993	!-- exchanged. grid_level is adjusted in poismg, where only one ghost plane
994	!-- is required.
995	grid_level = 0
996
997	#if defined( __parallel )
998	!
999	!-- Gridpoint number for the exchange of ghost points (y-line for 2D-arrays)
1000	ngp_y = nyn - nys + 1 + 2 * nbgp
1001
1002	!
1003	!-- Define new MPI derived datatypes for the exchange of ghost points in
1004	!-- x- and y-direction for 2D-arrays (line)
1005	CALL MPI_TYPE_VECTOR( nxr-nxl+1+2*nbgp, nbgp, ngp_y, MPI_REAL, type_x, &
1006	ierr )
1007	CALL MPI_TYPE_COMMIT( type_x, ierr )
1008	CALL MPI_TYPE_VECTOR( nxr-nxl+1+2*nbgp, nbgp, ngp_y, MPI_INTEGER, &
1009	type_x_int, ierr )
1010	CALL MPI_TYPE_COMMIT( type_x_int, ierr )
1011
1012	CALL MPI_TYPE_VECTOR( nbgp, ngp_y, ngp_y, MPI_REAL, type_y, ierr )
1013	CALL MPI_TYPE_COMMIT( type_y, ierr )
1014	CALL MPI_TYPE_VECTOR( nbgp, ngp_y, ngp_y, MPI_INTEGER, type_y_int, ierr )
1015	CALL MPI_TYPE_COMMIT( type_y_int, ierr )
1016
1017
1018	!
1019	!-- Calculate gridpoint numbers for the exchange of ghost points along x
1020	!-- (yz-plane for 3D-arrays) and define MPI derived data type(s) for the
1021	!-- exchange of ghost points in y-direction (xz-plane).
1022	!-- Do these calculations for the model grid and (if necessary) also
1023	!-- for the coarser grid levels used in the multigrid method
1024	ALLOCATE ( ngp_yz(0:maximum_grid_level), type_xz(0:maximum_grid_level),&
1025	type_yz(0:maximum_grid_level) )
1026
1027	nxl_l = nxl; nxr_l = nxr; nys_l = nys; nyn_l = nyn; nzb_l = nzb; nzt_l = nzt
1028
1029	!
1030	!-- Discern between the model grid, which needs nbgp ghost points and
1031	!-- grid levels for the multigrid scheme. In the latter case only one
1032	!-- ghost point is necessary.
1033	!-- First definition of MPI-datatypes for exchange of ghost layers on normal
1034	!-- grid. The following loop is needed for data exchange in poismg.f90.
1035	!
1036	!-- Determine number of grid points of yz-layer for exchange
1037	ngp_yz(0) = (nzt - nzb + 2) * (nyn - nys + 1 + 2 * nbgp)
1038
1039	!
1040	!-- Define an MPI-datatype for the exchange of left/right boundaries.
1041	!-- Although data are contiguous in physical memory (which does not
1042	!-- necessarily require an MPI-derived datatype), the data exchange between
1043	!-- left and right PE's using the MPI-derived type is 10% faster than without.
1044	CALL MPI_TYPE_VECTOR( nxr-nxl+1+2nbgp, nbgp(nzt-nzb+2), ngp_yz(0), &
1045	MPI_REAL, type_xz(0), ierr )
1046	CALL MPI_TYPE_COMMIT( type_xz(0), ierr )
1047
1048	CALL MPI_TYPE_VECTOR( nbgp, ngp_yz(0), ngp_yz(0), MPI_REAL, type_yz(0), &
1049	ierr )
1050	CALL MPI_TYPE_COMMIT( type_yz(0), ierr )
1051
1052	!
1053	!-- Definition of MPI-datatypes for multigrid method (coarser level grids)
1054	IF ( psolver == 'multigrid' ) THEN
1055	!
1056	!-- Definition of MPI-datatyoe as above, but only 1 ghost level is used
1057	DO i = maximum_grid_level, 1 , -1
1058
1059	ngp_yz(i) = (nzt_l - nzb_l + 2) * (nyn_l - nys_l + 3)
1060
1061	CALL MPI_TYPE_VECTOR( nxr_l-nxl_l+3, nzt_l-nzb_l+2, ngp_yz(i), &
1062	MPI_REAL, type_xz(i), ierr )
1063	CALL MPI_TYPE_COMMIT( type_xz(i), ierr )
1064
1065	CALL MPI_TYPE_VECTOR( 1, ngp_yz(i), ngp_yz(i), MPI_REAL, type_yz(i), &
1066	ierr )
1067	CALL MPI_TYPE_COMMIT( type_yz(i), ierr )
1068
1069	nxl_l = nxl_l / 2
1070	nxr_l = nxr_l / 2
1071	nys_l = nys_l / 2
1072	nyn_l = nyn_l / 2
1073	nzt_l = nzt_l / 2
1074
1075	ENDDO
1076
1077	ENDIF
1078	#endif
1079
1080	#if defined( __parallel )
1081	!
1082	!-- Setting of flags for inflow/outflow conditions in case of non-cyclic
1083	!-- horizontal boundary conditions.
1084	IF ( pleft == MPI_PROC_NULL ) THEN
1085	IF ( bc_lr == 'dirichlet/radiation' ) THEN
1086	inflow_l = .TRUE.
1087	ELSEIF ( bc_lr == 'radiation/dirichlet' ) THEN
1088	outflow_l = .TRUE.
1089	ENDIF
1090	ENDIF
1091
1092	IF ( pright == MPI_PROC_NULL ) THEN
1093	IF ( bc_lr == 'dirichlet/radiation' ) THEN
1094	outflow_r = .TRUE.
1095	ELSEIF ( bc_lr == 'radiation/dirichlet' ) THEN
1096	inflow_r = .TRUE.
1097	ENDIF
1098	ENDIF
1099
1100	IF ( psouth == MPI_PROC_NULL ) THEN
1101	IF ( bc_ns == 'dirichlet/radiation' ) THEN
1102	outflow_s = .TRUE.
1103	ELSEIF ( bc_ns == 'radiation/dirichlet' ) THEN
1104	inflow_s = .TRUE.
1105	ENDIF
1106	ENDIF
1107
1108	IF ( pnorth == MPI_PROC_NULL ) THEN
1109	IF ( bc_ns == 'dirichlet/radiation' ) THEN
1110	inflow_n = .TRUE.
1111	ELSEIF ( bc_ns == 'radiation/dirichlet' ) THEN
1112	outflow_n = .TRUE.
1113	ENDIF
1114	ENDIF
1115
1116	!
1117	!-- Broadcast the id of the inflow PE
1118	IF ( inflow_l ) THEN
1119	id_inflow_l = myidx
1120	ELSE
1121	id_inflow_l = 0
1122	ENDIF
1123	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
1124	CALL MPI_ALLREDUCE( id_inflow_l, id_inflow, 1, MPI_INTEGER, MPI_SUM, &
1125	comm1dx, ierr )
1126
1127	!
1128	!-- Broadcast the id of the recycling plane
1129	!-- WARNING: needs to be adjusted in case of inflows other than from left side!
1130	IF ( ( recycling_width / dx ) >= nxl .AND. &
1131	( recycling_width / dx ) <= nxr ) THEN
1132	id_recycling_l = myidx
1133	ELSE
1134	id_recycling_l = 0
1135	ENDIF
1136	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
1137	CALL MPI_ALLREDUCE( id_recycling_l, id_recycling, 1, MPI_INTEGER, MPI_SUM, &
1138	comm1dx, ierr )
1139
1140	#else
1141	IF ( bc_lr == 'dirichlet/radiation' ) THEN
1142	inflow_l = .TRUE.
1143	outflow_r = .TRUE.
1144	ELSEIF ( bc_lr == 'radiation/dirichlet' ) THEN
1145	outflow_l = .TRUE.
1146	inflow_r = .TRUE.
1147	ENDIF
1148
1149	IF ( bc_ns == 'dirichlet/radiation' ) THEN
1150	inflow_n = .TRUE.
1151	outflow_s = .TRUE.
1152	ELSEIF ( bc_ns == 'radiation/dirichlet' ) THEN
1153	outflow_n = .TRUE.
1154	inflow_s = .TRUE.
1155	ENDIF
1156	#endif
1157	!
1158	!-- At the outflow, u or v, respectively, have to be calculated for one more
1159	!-- grid point.
1160	IF ( outflow_l ) THEN
1161	nxlu = nxl + 1
1162	ELSE
1163	nxlu = nxl
1164	ENDIF
1165	IF ( outflow_s ) THEN
1166	nysv = nys + 1
1167	ELSE
1168	nysv = nys
1169	ENDIF
1170
1171	IF ( psolver == 'poisfft_hybrid' ) THEN
1172	CALL poisfft_hybrid_ini
1173	ELSEIF ( psolver == 'poisfft' ) THEN
1174	CALL poisfft_init
1175	ENDIF
1176
1177	!
1178	!-- Allocate wall flag arrays used in the multigrid solver
1179	IF ( psolver == 'multigrid' ) THEN
1180
1181	DO i = maximum_grid_level, 1, -1
1182
1183	SELECT CASE ( i )
1184
1185	CASE ( 1 )
1186	ALLOCATE( wall_flags_1(nzb:nzt_mg(i)+1, &
1187	nys_mg(i)-1:nyn_mg(i)+1, &
1188	nxl_mg(i)-1:nxr_mg(i)+1) )
1189
1190	CASE ( 2 )
1191	ALLOCATE( wall_flags_2(nzb:nzt_mg(i)+1, &
1192	nys_mg(i)-1:nyn_mg(i)+1, &
1193	nxl_mg(i)-1:nxr_mg(i)+1) )
1194
1195	CASE ( 3 )
1196	ALLOCATE( wall_flags_3(nzb:nzt_mg(i)+1, &
1197	nys_mg(i)-1:nyn_mg(i)+1, &
1198	nxl_mg(i)-1:nxr_mg(i)+1) )
1199
1200	CASE ( 4 )
1201	ALLOCATE( wall_flags_4(nzb:nzt_mg(i)+1, &
1202	nys_mg(i)-1:nyn_mg(i)+1, &
1203	nxl_mg(i)-1:nxr_mg(i)+1) )
1204
1205	CASE ( 5 )
1206	ALLOCATE( wall_flags_5(nzb:nzt_mg(i)+1, &
1207	nys_mg(i)-1:nyn_mg(i)+1, &
1208	nxl_mg(i)-1:nxr_mg(i)+1) )
1209
1210	CASE ( 6 )
1211	ALLOCATE( wall_flags_6(nzb:nzt_mg(i)+1, &
1212	nys_mg(i)-1:nyn_mg(i)+1, &
1213	nxl_mg(i)-1:nxr_mg(i)+1) )
1214
1215	CASE ( 7 )
1216	ALLOCATE( wall_flags_7(nzb:nzt_mg(i)+1, &
1217	nys_mg(i)-1:nyn_mg(i)+1, &
1218	nxl_mg(i)-1:nxr_mg(i)+1) )
1219
1220	CASE ( 8 )
1221	ALLOCATE( wall_flags_8(nzb:nzt_mg(i)+1, &
1222	nys_mg(i)-1:nyn_mg(i)+1, &
1223	nxl_mg(i)-1:nxr_mg(i)+1) )
1224
1225	CASE ( 9 )
1226	ALLOCATE( wall_flags_9(nzb:nzt_mg(i)+1, &
1227	nys_mg(i)-1:nyn_mg(i)+1, &
1228	nxl_mg(i)-1:nxr_mg(i)+1) )
1229
1230	CASE ( 10 )
1231	ALLOCATE( wall_flags_10(nzb:nzt_mg(i)+1, &
1232	nys_mg(i)-1:nyn_mg(i)+1, &
1233	nxl_mg(i)-1:nxr_mg(i)+1) )
1234
1235	CASE DEFAULT
1236	message_string = 'more than 10 multigrid levels'
1237	CALL message( 'init_pegrid', 'PA0238', 1, 2, 0, 6, 0 )
1238
1239	END SELECT
1240
1241	ENDDO
1242
1243	ENDIF
1244
1245	!
1246	!-- Calculate the number of groups into which parallel I/O is split.
1247	!-- The default for files which are opened by all PEs (or where each
1248	!-- PE opens his own independent file) is, that all PEs are doing input/output
1249	!-- in parallel at the same time. This might cause performance or even more
1250	!-- severe problems depending on the configuration of the underlying file
1251	!-- system.
1252	!-- First, set the default:
1253	IF ( maximum_parallel_io_streams == -1 .OR. &
1254	maximum_parallel_io_streams > numprocs ) THEN
1255	maximum_parallel_io_streams = numprocs
1256	ENDIF
1257
1258	!
1259	!-- Now calculate the number of io_blocks and the io_group to which the
1260	!-- respective PE belongs. I/O of the groups is done in serial, but in parallel
1261	!-- for all PEs belonging to the same group. A preliminary setting with myid
1262	!-- based on MPI_COMM_WORLD has been done in parin.
1263	io_blocks = numprocs / maximum_parallel_io_streams
1264	io_group = MOD( myid+1, io_blocks )
1265
1266
1267	END SUBROUTINE init_pegrid

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

Download in other formats:

| Impressum | ©Leibniz Universität Hannover |