Home

Context Navigation

source: palm/trunk/SOURCE/init_pegrid.f90 @ 254

Last change on this file since 254 was 254, checked in by heinze, 16 years ago
Output of messages replaced by message handling routine.
Property svn:keywords set to `Id`
File size: 34.0 KB

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

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

Download in other formats:

| Impressum | ©Leibniz Universität Hannover |