1 | !> @file shared_memory_io_mod.f90 |
---|
2 | !--------------------------------------------------------------------------------------------------! |
---|
3 | ! This file is part of the PALM model system. |
---|
4 | ! |
---|
5 | ! PALM is free software: you can redistribute it and/or modify it under the terms of the GNU General |
---|
6 | ! Public License as published by the Free Software Foundation, either version 3 of the License, or |
---|
7 | ! (at your option) any later version. |
---|
8 | ! |
---|
9 | ! PALM is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the |
---|
10 | ! implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General |
---|
11 | ! Public License for more details. |
---|
12 | ! |
---|
13 | ! You should have received a copy of the GNU General Public License along with PALM. If not, see |
---|
14 | ! <http://www.gnu.org/licenses/>. |
---|
15 | ! |
---|
16 | ! Copyright 1997-2020 Leibniz Universitaet Hannover |
---|
17 | !--------------------------------------------------------------------------------------------------! |
---|
18 | ! |
---|
19 | ! |
---|
20 | ! Current revisions: |
---|
21 | ! ----------------- |
---|
22 | ! |
---|
23 | ! |
---|
24 | ! Former revisions: |
---|
25 | ! ----------------- |
---|
26 | ! $Id: shared_memory_io_mod.f90 4628 2020-07-29 07:23:03Z raasch $ |
---|
27 | ! extensions required for MPI-I/O of particle data to restart files |
---|
28 | ! |
---|
29 | ! 4620 2020-07-22 14:11:16Z raasch |
---|
30 | ! bugfix: variable definition changed |
---|
31 | ! |
---|
32 | ! 4618 2020-07-22 11:21:08Z raasch |
---|
33 | ! unused variable removed |
---|
34 | ! |
---|
35 | ! |
---|
36 | ! Additions for cyclic fill mode |
---|
37 | ! |
---|
38 | ! |
---|
39 | ! File re-formatted to follow the PALM coding standard |
---|
40 | ! |
---|
41 | ! |
---|
42 | ! Initial version (Klaus Ketelsen) |
---|
43 | ! |
---|
44 | ! |
---|
45 | ! |
---|
46 | ! Description: |
---|
47 | ! ------------ |
---|
48 | !> Handle MPI-IO or NetCDF-IO shared memory arrays. |
---|
49 | !> This module performs the organization of new communicators, adapted PE-grids and allocation of |
---|
50 | !> shared memory arrays. The IO itself is not done here. |
---|
51 | !--------------------------------------------------------------------------------------------------! |
---|
52 | MODULE shared_memory_io_mod |
---|
53 | |
---|
54 | #if defined( __parallel ) |
---|
55 | #if defined( __mpifh ) |
---|
56 | INCLUDE "mpif.h" |
---|
57 | #else |
---|
58 | USE MPI |
---|
59 | #endif |
---|
60 | #endif |
---|
61 | |
---|
62 | USE, INTRINSIC :: ISO_C_BINDING |
---|
63 | |
---|
64 | USE control_parameters, & |
---|
65 | ONLY: maximum_grid_level, & |
---|
66 | message_string, & |
---|
67 | mg_switch_to_pe0_level |
---|
68 | |
---|
69 | |
---|
70 | USE indices, & |
---|
71 | ONLY: nbgp, & |
---|
72 | nnx, & |
---|
73 | nny, & |
---|
74 | nnz, & |
---|
75 | nx, & |
---|
76 | nxl, & |
---|
77 | nxlg, & |
---|
78 | nxr, & |
---|
79 | nxrg, & |
---|
80 | ny, & |
---|
81 | nyn, & |
---|
82 | nyng, & |
---|
83 | nys, & |
---|
84 | nysg, & |
---|
85 | nzb, & |
---|
86 | nzt |
---|
87 | |
---|
88 | USE kinds, & |
---|
89 | ONLY: dp, & |
---|
90 | idp, & |
---|
91 | isp, & |
---|
92 | iwp, & |
---|
93 | sp, & |
---|
94 | wp |
---|
95 | |
---|
96 | USE pegrid, & |
---|
97 | ONLY: comm1dx, & |
---|
98 | comm1dy, & |
---|
99 | comm2d, & |
---|
100 | ierr, & |
---|
101 | myid, & |
---|
102 | myidx, & |
---|
103 | myidy, & |
---|
104 | npex, & |
---|
105 | npey, & |
---|
106 | numprocs, & |
---|
107 | pdims, & |
---|
108 | pleft, & |
---|
109 | pnorth, & |
---|
110 | pright, & |
---|
111 | psouth, & |
---|
112 | sendrecvcount_xy |
---|
113 | |
---|
114 | #if defined( __parallel ) |
---|
115 | USE pegrid, & |
---|
116 | ONLY: pcoord, & |
---|
117 | reorder |
---|
118 | #endif |
---|
119 | |
---|
120 | IMPLICIT NONE |
---|
121 | |
---|
122 | PRIVATE |
---|
123 | |
---|
124 | SAVE |
---|
125 | |
---|
126 | ! |
---|
127 | !-- Type to store information about the domain decomposition grid |
---|
128 | TYPE, PUBLIC :: domain_decomposition_grid_features !< |
---|
129 | |
---|
130 | INTEGER(iwp) :: comm2d !< |
---|
131 | INTEGER(iwp) :: myid !< |
---|
132 | INTEGER(iwp) :: nnx !< |
---|
133 | INTEGER(iwp) :: nny !< |
---|
134 | INTEGER(iwp) :: nx !< |
---|
135 | INTEGER(iwp) :: nxl !< |
---|
136 | INTEGER(iwp) :: nxr !< |
---|
137 | INTEGER(iwp) :: ny !< |
---|
138 | INTEGER(iwp) :: nyn !< |
---|
139 | INTEGER(iwp) :: nys !< |
---|
140 | INTEGER(iwp) :: numprocs !< |
---|
141 | |
---|
142 | CONTAINS |
---|
143 | |
---|
144 | PROCEDURE, PASS(this), PUBLIC :: activate_grid_from_this_class |
---|
145 | PROCEDURE, PASS(this), PUBLIC :: save_grid_into_this_class |
---|
146 | |
---|
147 | END TYPE domain_decomposition_grid_features |
---|
148 | |
---|
149 | ! |
---|
150 | !-- Class definition for shared memory instances. |
---|
151 | !-- For every use of shared memory IO, one instance of this class is created. |
---|
152 | TYPE, PUBLIC :: sm_class !< |
---|
153 | |
---|
154 | INTEGER(iwp) :: nr_io_pe_per_node = 2 !< typical configuration, 2 sockets per node |
---|
155 | LOGICAL :: no_shared_Memory_in_this_run !< |
---|
156 | |
---|
157 | INTEGER(iwp) :: comm_model !< communicator of this model run |
---|
158 | ! |
---|
159 | !-- Variables for the shared memory communicator |
---|
160 | INTEGER(iwp), PUBLIC :: comm_shared !< communicator for processes with shared array |
---|
161 | INTEGER(iwp), PUBLIC :: sh_npes !< |
---|
162 | INTEGER(iwp), PUBLIC :: sh_rank !< |
---|
163 | |
---|
164 | LOGICAL, PUBLIC :: iam_io_pe = .TRUE. !< This PE is an IO-PE |
---|
165 | ! |
---|
166 | !-- Variables for the I/O virtual grid |
---|
167 | INTEGER(iwp), PUBLIC :: comm_io !< Communicator for all IO processes |
---|
168 | INTEGER(iwp), PUBLIC :: io_npes !< |
---|
169 | INTEGER(iwp), PUBLIC :: io_rank !< |
---|
170 | ! |
---|
171 | !-- Variables for the node local communicator |
---|
172 | INTEGER(iwp) :: comm_node !< Communicator for all processes of current node |
---|
173 | INTEGER(iwp) :: io_pe_global_rank !< |
---|
174 | INTEGER(iwp) :: n_npes !< |
---|
175 | INTEGER(iwp) :: n_rank !< |
---|
176 | |
---|
177 | TYPE(domain_decomposition_grid_features), PUBLIC :: io_grid !< io grid features, depending on reading from prerun or restart run |
---|
178 | |
---|
179 | |
---|
180 | CONTAINS |
---|
181 | |
---|
182 | PRIVATE |
---|
183 | |
---|
184 | PROCEDURE, PASS(this), PUBLIC :: is_sm_active |
---|
185 | PROCEDURE, PASS(this), PUBLIC :: sm_adjust_outer_boundary |
---|
186 | PROCEDURE, PASS(this), PUBLIC :: sm_free_shared |
---|
187 | PROCEDURE, PASS(this), PUBLIC :: sm_init_comm |
---|
188 | PROCEDURE, PASS(this), PUBLIC :: sm_init_part |
---|
189 | PROCEDURE, PASS(this), PUBLIC :: sm_node_barrier |
---|
190 | #if defined( __parallel ) |
---|
191 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_1d_64 |
---|
192 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_1d_32 |
---|
193 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_1di |
---|
194 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_2d_64 |
---|
195 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_2d_32 |
---|
196 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_2di |
---|
197 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_3d_64 |
---|
198 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_3d_32 |
---|
199 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_3di_32 |
---|
200 | PROCEDURE, PASS(this), PUBLIC :: sm_allocate_shared_3di_64 |
---|
201 | |
---|
202 | GENERIC, PUBLIC :: sm_allocate_shared => & |
---|
203 | sm_allocate_shared_1d_64, sm_allocate_shared_1d_32, & |
---|
204 | sm_allocate_shared_2d_64, sm_allocate_shared_2d_32, & |
---|
205 | sm_allocate_shared_2di, sm_allocate_shared_3d_64, & |
---|
206 | sm_allocate_shared_3d_32, sm_allocate_shared_1di, & |
---|
207 | sm_allocate_shared_3di_32, sm_allocate_shared_3di_64 |
---|
208 | #endif |
---|
209 | END TYPE sm_class |
---|
210 | |
---|
211 | |
---|
212 | CONTAINS |
---|
213 | |
---|
214 | |
---|
215 | !--------------------------------------------------------------------------------------------------! |
---|
216 | ! Description: |
---|
217 | ! ------------ |
---|
218 | !> Create the communicator for shared memory groups and IO-PEs. |
---|
219 | !> Setup the grid for shared memory IO. |
---|
220 | !--------------------------------------------------------------------------------------------------! |
---|
221 | SUBROUTINE sm_init_comm( this, sm_active, comm_input ) |
---|
222 | |
---|
223 | IMPLICIT NONE |
---|
224 | |
---|
225 | CLASS(sm_class), INTENT(INOUT) :: this !< pointer to access internal variables of this call |
---|
226 | INTEGER, INTENT(IN), OPTIONAL :: comm_input !< main model communicator (comm2d) can optional be set |
---|
227 | |
---|
228 | #if defined( __parallel ) |
---|
229 | INTEGER :: color |
---|
230 | INTEGER :: max_n_npes !< maximum number of PEs/node |
---|
231 | #endif |
---|
232 | |
---|
233 | LOGICAL, INTENT(IN) :: sm_active !< flag to activate shared-memory IO |
---|
234 | |
---|
235 | IF ( PRESENT( comm_input ) ) THEN |
---|
236 | this%comm_model = comm_input |
---|
237 | ELSE |
---|
238 | this%comm_model = comm2d |
---|
239 | ENDIF |
---|
240 | |
---|
241 | this%no_shared_memory_in_this_run = .NOT. sm_active |
---|
242 | this%comm_io = this%comm_model ! preset in case of non shared-memory-IO |
---|
243 | |
---|
244 | IF ( this%no_shared_memory_in_this_run ) THEN |
---|
245 | this%iam_io_pe = .TRUE. |
---|
246 | RETURN |
---|
247 | ENDIF |
---|
248 | |
---|
249 | #if defined( __parallel ) |
---|
250 | ! |
---|
251 | !-- Determine, how many MPI threads are running on a node |
---|
252 | this%iam_io_pe = .FALSE. |
---|
253 | CALL MPI_COMM_SPLIT_TYPE( this%comm_model, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL, & |
---|
254 | this%comm_node, ierr ) |
---|
255 | CALL MPI_COMM_SIZE( this%comm_node, this%n_npes, ierr ) |
---|
256 | CALL MPI_COMM_RANK( this%comm_node, this%n_rank, ierr ) |
---|
257 | |
---|
258 | CALL MPI_ALLREDUCE( this%n_npes, max_n_npes, 1, MPI_INTEGER, MPI_MAX, this%comm_model, ierr ) |
---|
259 | ! |
---|
260 | !-- Decide, if the configuration can run with shared-memory IO |
---|
261 | IF ( max_n_npes > 64 ) THEN |
---|
262 | ! |
---|
263 | !-- Special configuration on the HLRN-IV system with 4 shared memory blocks/node |
---|
264 | this%nr_io_pe_per_node = 4 |
---|
265 | |
---|
266 | ELSEIF ( max_n_npes <= 32 ) THEN |
---|
267 | ! |
---|
268 | !-- No shared memory IO with less than 32 threads/node |
---|
269 | this%no_shared_memory_in_this_run = .TRUE. |
---|
270 | this%iam_io_pe = .TRUE. |
---|
271 | RETURN |
---|
272 | ENDIF |
---|
273 | |
---|
274 | ! |
---|
275 | !-- No shared memory IO with small setups |
---|
276 | IF ( nx < 24 .OR. ny < 24 ) THEN |
---|
277 | this%no_shared_memory_in_this_run = .TRUE. |
---|
278 | this%iam_io_pe = .TRUE. |
---|
279 | RETURN |
---|
280 | ENDIF |
---|
281 | |
---|
282 | ! |
---|
283 | !-- Divide a node into shared memory groups, depending on the virtual x-y grid |
---|
284 | CALL compute_color( color ) |
---|
285 | ! |
---|
286 | !-- If no shared memory IO possible, nothing is left to be done here. |
---|
287 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
288 | |
---|
289 | ! |
---|
290 | !-- Setup the shared memory area |
---|
291 | CALL MPI_COMM_SPLIT( this%comm_node, color, 0, this%comm_shared, ierr ) |
---|
292 | CALL MPI_COMM_SIZE( this%comm_shared, this%sh_npes, ierr ) |
---|
293 | CALL MPI_COMM_RANK( this%comm_shared, this%sh_rank, ierr ) |
---|
294 | |
---|
295 | ! |
---|
296 | !-- Setup the communicator across the nodes depending on the shared memory rank. |
---|
297 | !-- All threads with shared memory rank 0 will be I/O threads. |
---|
298 | color = this%sh_rank |
---|
299 | CALL MPI_COMM_SPLIT( this%comm_model, color, 0, this%comm_io, ierr ) |
---|
300 | |
---|
301 | IF ( this%comm_io /= MPI_COMM_NULL ) THEN |
---|
302 | CALL MPI_COMM_SIZE( this%comm_io, this%io_npes, ierr ) |
---|
303 | CALL MPI_COMM_RANK( this%comm_io, this%io_rank, ierr ) |
---|
304 | ELSE |
---|
305 | this%io_npes = -1 |
---|
306 | this%io_rank = -1 |
---|
307 | ENDIF |
---|
308 | |
---|
309 | IF ( this%sh_rank == 0 ) THEN |
---|
310 | this%iam_io_pe = .TRUE. |
---|
311 | this%io_pe_global_rank = myid |
---|
312 | ENDIF |
---|
313 | CALL MPI_BCAST( this%io_pe_global_rank, 1, MPI_INTEGER, 0, this%comm_shared, ierr ) |
---|
314 | |
---|
315 | #else |
---|
316 | this%iam_io_pe = .TRUE. |
---|
317 | #endif |
---|
318 | |
---|
319 | ! write(9,'(a,8i7)') ' end of sm_init_comm ',this%sh_rank,this%sh_npes,this%io_rank,this%io_npes,this%io_pe_global_rank |
---|
320 | ! write(9,*) 'This process is IO Process ',this%iam_io_pe |
---|
321 | |
---|
322 | #if defined( __parallel ) |
---|
323 | CONTAINS |
---|
324 | |
---|
325 | SUBROUTINE compute_color( color ) |
---|
326 | |
---|
327 | IMPLICIT NONE |
---|
328 | |
---|
329 | INTEGER(iwp), INTENT(OUT) :: color !< |
---|
330 | |
---|
331 | INTEGER(iwp) :: group_start !< |
---|
332 | INTEGER(iwp) :: my_color !< |
---|
333 | INTEGER(iwp) :: n !< |
---|
334 | INTEGER(iwp) :: pe !< |
---|
335 | INTEGER(iwp) :: sh_group_size !< |
---|
336 | |
---|
337 | INTEGER(iwp), DIMENSION(4,0:this%n_npes-1) :: local_dim_s !< |
---|
338 | INTEGER(iwp), DIMENSION(4,0:this%n_npes-1) :: local_dim_r !< |
---|
339 | |
---|
340 | TYPE(domain_decomposition_grid_features), DIMENSION(32) :: node_grid !< |
---|
341 | |
---|
342 | ! |
---|
343 | !-- No shared memory I/O on one node jobs |
---|
344 | IF ( numprocs < this%n_npes ) THEN |
---|
345 | this%no_shared_memory_in_this_run = .TRUE. |
---|
346 | RETURN |
---|
347 | ENDIF |
---|
348 | |
---|
349 | local_dim_s = 0 |
---|
350 | local_dim_s(1,this%n_rank) = nxl |
---|
351 | local_dim_s(2,this%n_rank) = nxr |
---|
352 | local_dim_s(3,this%n_rank) = nys |
---|
353 | local_dim_s(4,this%n_rank) = nyn |
---|
354 | |
---|
355 | node_grid%nyn = -1 |
---|
356 | ! |
---|
357 | !-- Distribute the x-y layout of all cores of a node to all node processes |
---|
358 | CALL MPI_ALLREDUCE( local_dim_s, local_dim_r, SIZE( local_dim_s ), MPI_INTEGER, MPI_SUM, & |
---|
359 | this%comm_node, ierr ) |
---|
360 | sh_group_size = ( max_n_npes + this%nr_io_pe_per_node - 1 ) / this%nr_io_pe_per_node |
---|
361 | |
---|
362 | pe = 0 |
---|
363 | my_color = 1 ! color is used to split the shared memory communicator into a communicator for |
---|
364 | ! io groups |
---|
365 | group_start = pe |
---|
366 | node_grid(my_color)%nxl = local_dim_r(1,group_start) |
---|
367 | node_grid(my_color)%nxr = local_dim_r(2,group_start) |
---|
368 | node_grid(my_color)%nys = local_dim_r(3,group_start) |
---|
369 | |
---|
370 | DO n = 1, this%n_npes-1 |
---|
371 | |
---|
372 | pe = n |
---|
373 | IF ( n > 0 .AND. MOD( n,sh_group_size ) == 0 ) THEN |
---|
374 | ! |
---|
375 | !-- If group boundary, start new IO group |
---|
376 | node_grid(my_color)%nyn = local_dim_r(4,pe-1) |
---|
377 | my_color = my_color + 1 |
---|
378 | group_start = pe |
---|
379 | node_grid(my_color)%nxl = local_dim_r(1,group_start) |
---|
380 | node_grid(my_color)%nxr = local_dim_r(2,group_start) |
---|
381 | node_grid(my_color)%nys = local_dim_r(3,group_start) |
---|
382 | |
---|
383 | ELSEIF ( local_dim_r(1,pe) /= node_grid(my_color)%nxl ) THEN |
---|
384 | ! |
---|
385 | !-- If nxl changes, start new IO group |
---|
386 | node_grid(my_color)%nyn = local_dim_r(4,pe-1) |
---|
387 | my_color = my_color+1 |
---|
388 | group_start = pe |
---|
389 | node_grid(my_color)%nxl = local_dim_r(1,group_start) |
---|
390 | node_grid(my_color)%nxr = local_dim_r(2,group_start) |
---|
391 | node_grid(my_color)%nys = local_dim_r(3,group_start) |
---|
392 | ENDIF |
---|
393 | ! |
---|
394 | !-- Save values for local PE |
---|
395 | IF ( this%n_rank == pe ) THEN ! |
---|
396 | color = my_color |
---|
397 | ENDIF |
---|
398 | IF ( n == this%n_npes-1 ) node_grid(my_color)%nyn = local_dim_r(4,pe) |
---|
399 | |
---|
400 | ENDDO |
---|
401 | |
---|
402 | IF ( this%n_rank == 0 ) THEN |
---|
403 | color = 1 |
---|
404 | ENDIF |
---|
405 | |
---|
406 | this%io_grid = node_grid(color) |
---|
407 | this%io_grid%nnx = this%io_grid%nxr - this%io_grid%nxl + 1 |
---|
408 | this%io_grid%nny = this%io_grid%nyn - this%io_grid%nys + 1 |
---|
409 | |
---|
410 | END SUBROUTINE compute_color |
---|
411 | #endif |
---|
412 | |
---|
413 | END SUBROUTINE sm_init_comm |
---|
414 | |
---|
415 | |
---|
416 | ! |
---|
417 | !-- TODO: short description required, about the meaning of the following routine |
---|
418 | !-- part must be renamed particles! |
---|
419 | SUBROUTINE sm_init_part( this ) |
---|
420 | |
---|
421 | IMPLICIT NONE |
---|
422 | |
---|
423 | CLASS(sm_class), INTENT(INOUT) :: this !< pointer to access internal variables of this call |
---|
424 | |
---|
425 | #if defined( __parallel ) |
---|
426 | INTEGER(iwp) :: color !< |
---|
427 | INTEGER(iwp) :: ierr !< |
---|
428 | INTEGER(iwp) :: max_n_npes !< maximum number of PEs/node |
---|
429 | #endif |
---|
430 | |
---|
431 | LOGICAL :: sm_active !< |
---|
432 | |
---|
433 | |
---|
434 | sm_active = .TRUE. ! particle IO always uses shared memory |
---|
435 | this%comm_model = comm2d |
---|
436 | |
---|
437 | this%no_shared_memory_in_this_run = .NOT. sm_active |
---|
438 | this%comm_io = this%comm_model ! preset in case of non shared-memory-IO |
---|
439 | |
---|
440 | IF ( this%no_shared_memory_in_this_run ) THEN |
---|
441 | this%iam_io_pe = .TRUE. |
---|
442 | RETURN |
---|
443 | ENDIF |
---|
444 | |
---|
445 | #if defined( __parallel ) |
---|
446 | ! |
---|
447 | !-- Determine, how many MPI threads are running on a node |
---|
448 | this%iam_io_pe = .FALSE. |
---|
449 | CALL MPI_COMM_SPLIT_TYPE( this%comm_model, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL, & |
---|
450 | this%comm_node, ierr ) |
---|
451 | CALL MPI_COMM_SIZE( this%comm_node, this%n_npes, ierr ) |
---|
452 | CALL MPI_COMM_RANK( this%comm_node, this%n_rank, ierr ) |
---|
453 | |
---|
454 | CALL MPI_ALLREDUCE( this%n_npes, max_n_npes, 1, MPI_INTEGER, MPI_MAX, this%comm_model, ierr ) |
---|
455 | |
---|
456 | ! |
---|
457 | !-- TODO: better explanation |
---|
458 | !-- It has to be testet, if using memory blocks for an IO process (MPI shared Memory), or if it is |
---|
459 | !-- even better to use the complete node for MPI shared memory (this%nr_io_pe_per_node = 1). |
---|
460 | !- In the latter case, the access to the MPI shared memory buffer is slower, the number of |
---|
461 | !-- particles to move between threads will be much smaller. |
---|
462 | IF ( max_n_npes > 64 ) THEN |
---|
463 | ! |
---|
464 | !-- Special configuration on the HLRN-IV system with 4 shared memory blocks/node |
---|
465 | this%nr_io_pe_per_node = 4 |
---|
466 | ENDIF |
---|
467 | |
---|
468 | IF ( this%nr_io_pe_per_node == 1 ) THEN |
---|
469 | ! |
---|
470 | !-- This branch is not realized so far |
---|
471 | this%iam_io_pe = ( this%n_rank == 0 ) |
---|
472 | this%comm_shared = this%comm_node |
---|
473 | CALL MPI_COMM_SIZE( this%comm_shared, this%sh_npes, ierr ) |
---|
474 | CALL MPI_COMM_RANK( this%comm_shared, this%sh_rank, ierr ) |
---|
475 | |
---|
476 | ELSEIF( this%nr_io_pe_per_node == 2 ) THEN |
---|
477 | |
---|
478 | this%iam_io_pe = ( this%n_rank == 0 .OR. this%n_rank == this%n_npes/2 ) |
---|
479 | IF ( this%n_rank < this%n_npes/2 ) THEN |
---|
480 | color = 1 |
---|
481 | ELSE |
---|
482 | color = 2 |
---|
483 | ENDIF |
---|
484 | CALL MPI_COMM_SPLIT( this%comm_node, color, 0, this%comm_shared, ierr ) |
---|
485 | CALL MPI_COMM_SIZE( this%comm_shared, this%sh_npes, ierr ) |
---|
486 | CALL MPI_COMM_RANK( this%comm_shared, this%sh_rank, ierr ) |
---|
487 | |
---|
488 | ELSEIF( this%nr_io_pe_per_node == 4 ) THEN |
---|
489 | |
---|
490 | this%iam_io_pe = ( this%n_rank == 0 .OR. this%n_rank == this%n_npes/4 .OR. & |
---|
491 | this%n_rank == this%n_npes/2 .OR. this%n_rank == (3*this%n_npes)/4 ) |
---|
492 | IF ( this%n_rank < this%n_npes/4 ) THEN |
---|
493 | color = 1 |
---|
494 | ELSEIF( this%n_rank < this%n_npes/2 ) THEN |
---|
495 | color = 2 |
---|
496 | ELSEIF( this%n_rank < (3*this%n_npes)/4 ) THEN |
---|
497 | color = 3 |
---|
498 | ELSE |
---|
499 | color = 4 |
---|
500 | ENDIF |
---|
501 | CALL MPI_COMM_SPLIT( this%comm_node, color, 0, this%comm_shared, ierr ) |
---|
502 | CALL MPI_COMM_SIZE( this%comm_shared, this%sh_npes, ierr ) |
---|
503 | CALL MPI_COMM_RANK( this%comm_shared, this%sh_rank, ierr ) |
---|
504 | |
---|
505 | ELSE |
---|
506 | |
---|
507 | WRITE( *, * ) 'shared_memory_io_mod: internal error' |
---|
508 | WRITE( *, * ) 'only 2 or 4 shared memory groups per node are allowed' |
---|
509 | STOP |
---|
510 | |
---|
511 | ENDIF |
---|
512 | |
---|
513 | ! |
---|
514 | !-- Setup the shared memory area |
---|
515 | CALL MPI_COMM_SPLIT( this%comm_node, color, 0, this%comm_shared, ierr ) |
---|
516 | CALL MPI_COMM_SIZE( this%comm_shared, this%sh_npes, ierr ) |
---|
517 | CALL MPI_COMM_RANK( this%comm_shared, this%sh_rank, ierr ) |
---|
518 | |
---|
519 | ! |
---|
520 | !-- Setup the communicator across the nodes depending on the shared memory rank. |
---|
521 | !-- All threads with shared memory rank 0 will be I/O threads. |
---|
522 | color = this%sh_rank |
---|
523 | CALL MPI_COMM_SPLIT( this%comm_model, color, 0, this%comm_io, ierr ) |
---|
524 | |
---|
525 | IF ( this%comm_io /= MPI_COMM_NULL ) THEN |
---|
526 | CALL MPI_COMM_SIZE( this%comm_io, this%io_npes, ierr ) |
---|
527 | CALL MPI_COMM_RANK( this%comm_io, this%io_rank, ierr ) |
---|
528 | ELSE |
---|
529 | this%io_npes = -1 |
---|
530 | this%io_rank = -1 |
---|
531 | ENDIF |
---|
532 | |
---|
533 | IF ( this%sh_rank == 0 ) THEN |
---|
534 | this%iam_io_pe = .TRUE. |
---|
535 | this%io_pe_global_rank = myid |
---|
536 | ENDIF |
---|
537 | CALL MPI_BCAST( this%io_pe_global_rank, 1, MPI_INTEGER, 0, this%comm_shared, ierr ) |
---|
538 | |
---|
539 | #else |
---|
540 | this%iam_io_pe = .FALSE. |
---|
541 | #endif |
---|
542 | |
---|
543 | ! write(9,'(a,8i7)') 'sm_init_comm_part ',this%sh_rank,this%sh_npes,this%io_rank,this%io_npes |
---|
544 | |
---|
545 | END SUBROUTINE sm_init_part |
---|
546 | |
---|
547 | !--------------------------------------------------------------------------------------------------! |
---|
548 | ! Description: |
---|
549 | ! ------------ |
---|
550 | !> Function to return if shared Memory IO is active. |
---|
551 | !--------------------------------------------------------------------------------------------------! |
---|
552 | FUNCTION is_sm_active( this ) RESULT( ac ) |
---|
553 | |
---|
554 | IMPLICIT NONE |
---|
555 | |
---|
556 | CLASS(sm_class), INTENT(inout) :: this !< |
---|
557 | |
---|
558 | LOGICAL :: ac !< |
---|
559 | |
---|
560 | ac = .NOT. this%no_shared_memory_in_this_run |
---|
561 | |
---|
562 | END FUNCTION is_sm_active |
---|
563 | |
---|
564 | |
---|
565 | #if defined( __parallel ) |
---|
566 | |
---|
567 | !--------------------------------------------------------------------------------------------------! |
---|
568 | ! Description: |
---|
569 | ! ------------ |
---|
570 | !> Allocate shared 1d-REAL (64 Bit) array on ALL threads |
---|
571 | !--------------------------------------------------------------------------------------------------! |
---|
572 | SUBROUTINE sm_allocate_shared_1d_64( this, p1, d1, d2, win ) |
---|
573 | |
---|
574 | IMPLICIT NONE |
---|
575 | |
---|
576 | CLASS(sm_class), INTENT(inout) :: this |
---|
577 | |
---|
578 | INTEGER(iwp) :: disp_unit |
---|
579 | INTEGER(iwp), INTENT(IN) :: d1 |
---|
580 | INTEGER(iwp), INTENT(IN) :: d2 |
---|
581 | INTEGER(iwp), SAVE :: pe_from = 0 |
---|
582 | INTEGER(iwp), INTENT(OUT) :: win |
---|
583 | |
---|
584 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size |
---|
585 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize |
---|
586 | |
---|
587 | INTEGER, DIMENSION(1) :: buf_shape |
---|
588 | |
---|
589 | REAL(dp), DIMENSION(:), POINTER :: buf |
---|
590 | REAL(dp), DIMENSION(:), POINTER :: p1 |
---|
591 | |
---|
592 | TYPE(C_PTR), SAVE :: base_ptr |
---|
593 | TYPE(C_PTR), SAVE :: rem_ptr |
---|
594 | |
---|
595 | |
---|
596 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
597 | ! |
---|
598 | !-- Allocate shared memory on node rank 0 threads. |
---|
599 | IF ( this%sh_rank == pe_from ) THEN |
---|
600 | wsize = d2 - d1 + 1 |
---|
601 | ELSE |
---|
602 | wsize = 1 |
---|
603 | ENDIF |
---|
604 | wsize = wsize * dp ! please note, size is always in bytes, independently of the displacement |
---|
605 | ! unit |
---|
606 | |
---|
607 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, dp, MPI_INFO_NULL, this%comm_shared,base_ptr, win, ierr ) |
---|
608 | ! |
---|
609 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
610 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
611 | ! |
---|
612 | !-- Convert C- to Fortran-pointer |
---|
613 | buf_shape(1) = d2 - d1 + 1 |
---|
614 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
615 | p1(d1:) => buf |
---|
616 | ! |
---|
617 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
618 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
619 | |
---|
620 | END SUBROUTINE sm_allocate_shared_1d_64 |
---|
621 | |
---|
622 | |
---|
623 | !--------------------------------------------------------------------------------------------------! |
---|
624 | ! Description: |
---|
625 | ! ------------ |
---|
626 | !> Allocate shared 1d-REAL (32 Bit) array on ALL threads |
---|
627 | !--------------------------------------------------------------------------------------------------! |
---|
628 | SUBROUTINE sm_allocate_shared_1d_32( this, p1, d1, d2, win ) |
---|
629 | |
---|
630 | IMPLICIT NONE |
---|
631 | |
---|
632 | CLASS(sm_class), INTENT(inout) :: this |
---|
633 | |
---|
634 | INTEGER(iwp) :: disp_unit |
---|
635 | INTEGER(iwp), INTENT(IN) :: d1 |
---|
636 | INTEGER(iwp), INTENT(IN) :: d2 |
---|
637 | INTEGER(iwp), SAVE :: pe_from = 0 |
---|
638 | INTEGER(iwp), INTENT(OUT) :: win |
---|
639 | |
---|
640 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size |
---|
641 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize |
---|
642 | |
---|
643 | INTEGER, DIMENSION(1) :: buf_shape |
---|
644 | |
---|
645 | REAL(sp), DIMENSION(:), POINTER :: buf |
---|
646 | REAL(sp), DIMENSION(:), POINTER :: p1 |
---|
647 | |
---|
648 | TYPE(C_PTR), SAVE :: base_ptr |
---|
649 | TYPE(C_PTR), SAVE :: rem_ptr |
---|
650 | |
---|
651 | |
---|
652 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
653 | ! |
---|
654 | !-- Allocate shared memory on node rank 0 threads. |
---|
655 | IF ( this%sh_rank == pe_from ) THEN |
---|
656 | wsize = d2 - d1 + 1 |
---|
657 | ELSE |
---|
658 | wsize = 1 |
---|
659 | ENDIF |
---|
660 | wsize = wsize * sp ! Please note, size is always in bytes, independently of the displacement |
---|
661 | ! unit |
---|
662 | |
---|
663 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, sp, MPI_INFO_NULL, this%comm_shared,base_ptr, win, ierr ) |
---|
664 | ! |
---|
665 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
666 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
667 | ! |
---|
668 | !-- Convert C- to Fortran-pointer |
---|
669 | buf_shape(1) = d2 - d1 + 1 |
---|
670 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
671 | p1(d1:) => buf |
---|
672 | ! |
---|
673 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
674 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
675 | |
---|
676 | END SUBROUTINE sm_allocate_shared_1d_32 |
---|
677 | |
---|
678 | |
---|
679 | !--------------------------------------------------------------------------------------------------! |
---|
680 | ! Description: |
---|
681 | ! ------------ |
---|
682 | !> Allocate shared 1d-INTEGER array on ALL threads |
---|
683 | !--------------------------------------------------------------------------------------------------! |
---|
684 | SUBROUTINE sm_allocate_shared_1di( this, p1, d1, d2, win ) |
---|
685 | |
---|
686 | IMPLICIT NONE |
---|
687 | |
---|
688 | CLASS(sm_class), INTENT(inout) :: this |
---|
689 | |
---|
690 | INTEGER(iwp) :: disp_unit |
---|
691 | INTEGER(iwp), INTENT(IN) :: d1 |
---|
692 | INTEGER(iwp), INTENT(IN) :: d2 |
---|
693 | INTEGER(iwp), SAVE :: pe_from = 0 |
---|
694 | INTEGER(iwp), INTENT(OUT) :: win |
---|
695 | |
---|
696 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size |
---|
697 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize |
---|
698 | |
---|
699 | INTEGER, DIMENSION(1) :: buf_shape |
---|
700 | |
---|
701 | INTEGER(iwp), DIMENSION(:), POINTER :: buf |
---|
702 | INTEGER(iwp), DIMENSION(:), POINTER :: p1 |
---|
703 | |
---|
704 | TYPE(C_PTR), SAVE :: base_ptr |
---|
705 | TYPE(C_PTR), SAVE :: rem_ptr |
---|
706 | |
---|
707 | |
---|
708 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
709 | ! |
---|
710 | !-- Allocate shared memory on node rank 0 threads. |
---|
711 | IF ( this%sh_rank == pe_from ) THEN |
---|
712 | wsize = d2 - d1 + 1 |
---|
713 | ELSE |
---|
714 | wsize = 1 |
---|
715 | ENDIF |
---|
716 | wsize = wsize * iwp ! Please note, size is always in bytes, independently of the displacement |
---|
717 | ! unit |
---|
718 | |
---|
719 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, iwp, MPI_INFO_NULL, this%comm_shared,base_ptr, win, ierr ) |
---|
720 | ! |
---|
721 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
722 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
723 | ! |
---|
724 | !-- Convert C- to Fortran-pointer |
---|
725 | buf_shape(1) = d2 - d1 + 1 |
---|
726 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
727 | p1(d1:) => buf |
---|
728 | ! |
---|
729 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
730 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
731 | |
---|
732 | END SUBROUTINE sm_allocate_shared_1di |
---|
733 | |
---|
734 | |
---|
735 | !--------------------------------------------------------------------------------------------------! |
---|
736 | ! Description: |
---|
737 | ! ------------ |
---|
738 | !> Allocate shared 2d-REAL array on ALL threads (64 Bit) |
---|
739 | !--------------------------------------------------------------------------------------------------! |
---|
740 | SUBROUTINE sm_allocate_shared_2d_64( this, p2, n_nxlg, n_nxrg, n_nysg, n_nyng, win ) |
---|
741 | |
---|
742 | IMPLICIT NONE |
---|
743 | |
---|
744 | CLASS(sm_class), INTENT(INOUT) :: this |
---|
745 | |
---|
746 | INTEGER(iwp) :: disp_unit |
---|
747 | INTEGER(iwp), INTENT(IN) :: n_nxlg |
---|
748 | INTEGER(iwp), INTENT(IN) :: n_nxrg |
---|
749 | INTEGER(iwp), INTENT(IN) :: n_nyng |
---|
750 | INTEGER(iwp), INTENT(IN) :: n_nysg |
---|
751 | INTEGER(iwp), SAVE :: pe_from = 0 |
---|
752 | INTEGER(iwp), INTENT(OUT) :: win |
---|
753 | |
---|
754 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size |
---|
755 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize |
---|
756 | |
---|
757 | INTEGER(iwp), DIMENSION(2) :: buf_shape |
---|
758 | |
---|
759 | REAL(dp), DIMENSION(:,:), POINTER :: buf |
---|
760 | REAL(dp), DIMENSION(:,:), POINTER :: p2 |
---|
761 | |
---|
762 | TYPE(C_PTR), SAVE :: base_ptr |
---|
763 | TYPE(C_PTR), SAVE :: rem_ptr |
---|
764 | |
---|
765 | |
---|
766 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
767 | ! |
---|
768 | !-- Allocate shared memory on node rank 0 threads. |
---|
769 | IF ( this%sh_rank == pe_from ) THEN |
---|
770 | wsize = ( n_nyng - n_nysg + 1 ) * ( n_nxrg - n_nxlg + 1 ) |
---|
771 | ELSE |
---|
772 | wsize = 1 |
---|
773 | ENDIF |
---|
774 | |
---|
775 | wsize = wsize * dp ! Please note, size is always in bytes, independently of the displacement |
---|
776 | ! unit |
---|
777 | |
---|
778 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, 8, MPI_INFO_NULL, this%comm_shared, base_ptr, win, ierr ) |
---|
779 | ! |
---|
780 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
781 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
782 | ! |
---|
783 | !-- Convert C- to Fortran-pointer |
---|
784 | buf_shape(2) = n_nyng - n_nysg + 1 |
---|
785 | buf_shape(1) = n_nxrg - n_nxlg + 1 |
---|
786 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
787 | p2(n_nxlg:, n_nysg:) => buf |
---|
788 | ! |
---|
789 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
790 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
791 | |
---|
792 | END SUBROUTINE sm_allocate_shared_2d_64 |
---|
793 | |
---|
794 | |
---|
795 | !--------------------------------------------------------------------------------------------------! |
---|
796 | ! Description: |
---|
797 | ! ------------ |
---|
798 | !> Allocate shared 2d-REAL (32 Bit) array on ALL threads |
---|
799 | !--------------------------------------------------------------------------------------------------! |
---|
800 | SUBROUTINE sm_allocate_shared_2d_32( this, p2, n_nxlg, n_nxrg, n_nysg, n_nyng, win ) |
---|
801 | |
---|
802 | IMPLICIT NONE |
---|
803 | |
---|
804 | CLASS(sm_class), INTENT(INOUT) :: this |
---|
805 | |
---|
806 | INTEGER(iwp) :: disp_unit |
---|
807 | INTEGER(iwp), INTENT(IN) :: n_nxlg |
---|
808 | INTEGER(iwp), INTENT(IN) :: n_nxrg |
---|
809 | INTEGER(iwp), INTENT(IN) :: n_nyng |
---|
810 | INTEGER(iwp), INTENT(IN) :: n_nysg |
---|
811 | INTEGER(iwp), SAVE :: pe_from = 0 |
---|
812 | INTEGER(iwp), INTENT(OUT) :: win |
---|
813 | |
---|
814 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size |
---|
815 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize |
---|
816 | |
---|
817 | INTEGER(iwp), DIMENSION(2) :: buf_shape |
---|
818 | |
---|
819 | REAL(sp), DIMENSION(:,:), POINTER :: buf |
---|
820 | REAL(sp), DIMENSION(:,:), POINTER :: p2 |
---|
821 | |
---|
822 | TYPE(C_PTR), SAVE :: base_ptr |
---|
823 | TYPE(C_PTR), SAVE :: rem_ptr |
---|
824 | |
---|
825 | |
---|
826 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
827 | ! |
---|
828 | !-- Allocate shared memory on node rank 0 threads. |
---|
829 | IF ( this%sh_rank == pe_from ) THEN |
---|
830 | wsize = ( n_nyng - n_nysg + 1 ) * ( n_nxrg - n_nxlg + 1 ) |
---|
831 | ELSE |
---|
832 | wsize = 1 |
---|
833 | ENDIF |
---|
834 | |
---|
835 | wsize = wsize * sp ! Please note, size is always in bytes, independently of the displacement |
---|
836 | ! unit |
---|
837 | |
---|
838 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, dp, MPI_INFO_NULL, this%comm_shared, base_ptr, win, ierr ) |
---|
839 | ! |
---|
840 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
841 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
842 | ! |
---|
843 | !-- Convert C- to Fortran-pointer |
---|
844 | buf_shape(2) = n_nyng - n_nysg + 1 |
---|
845 | buf_shape(1) = n_nxrg - n_nxlg + 1 |
---|
846 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
847 | p2(n_nxlg:, n_nysg:) => buf |
---|
848 | ! |
---|
849 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
850 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
851 | |
---|
852 | END SUBROUTINE sm_allocate_shared_2d_32 |
---|
853 | |
---|
854 | |
---|
855 | !--------------------------------------------------------------------------------------------------! |
---|
856 | ! Description: |
---|
857 | ! ------------ |
---|
858 | !> Allocate shared 2d-INTEGER array on ALL threads |
---|
859 | !--------------------------------------------------------------------------------------------------! |
---|
860 | SUBROUTINE sm_allocate_shared_2di( this, p2i, n_nxlg, n_nxrg, n_nysg, n_nyng, win ) |
---|
861 | |
---|
862 | IMPLICIT NONE |
---|
863 | |
---|
864 | CLASS(sm_class), INTENT(inout) :: this !< |
---|
865 | |
---|
866 | INTEGER(iwp) :: disp_unit !< |
---|
867 | INTEGER(iwp), INTENT(IN) :: n_nxlg !< |
---|
868 | INTEGER(iwp), INTENT(IN) :: n_nxrg !< |
---|
869 | INTEGER(iwp), INTENT(IN) :: n_nyng !< |
---|
870 | INTEGER(iwp), INTENT(IN) :: n_nysg !< |
---|
871 | INTEGER(iwp), SAVE :: pe_from = 0 !< |
---|
872 | INTEGER(iwp), INTENT(OUT) :: win !< |
---|
873 | |
---|
874 | INTEGER(kind=MPI_ADDRESS_KIND) :: rem_size !< |
---|
875 | INTEGER(kind=MPI_ADDRESS_KIND) :: wsize !< |
---|
876 | |
---|
877 | INTEGER(iwp), DIMENSION(2) :: buf_shape !< |
---|
878 | |
---|
879 | INTEGER(iwp), DIMENSION(:,:), POINTER :: buf !< |
---|
880 | INTEGER(iwp), DIMENSION(:,:), POINTER :: p2i !< |
---|
881 | |
---|
882 | TYPE(C_PTR), SAVE :: base_ptr !< |
---|
883 | TYPE(C_PTR), SAVE :: rem_ptr !< |
---|
884 | |
---|
885 | |
---|
886 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
887 | ! |
---|
888 | !-- Allocate shared memory on node rank 0 threads. |
---|
889 | IF ( this%sh_rank == pe_from ) THEN |
---|
890 | wsize = ( n_nyng - n_nysg + 1 ) * ( n_nxrg - n_nxlg + 1 ) |
---|
891 | ELSE |
---|
892 | wsize = 1 |
---|
893 | ENDIF |
---|
894 | |
---|
895 | wsize = wsize * 4 ! Please note, size is always in bytes, independently of the displacement |
---|
896 | ! unit |
---|
897 | |
---|
898 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, 4, MPI_INFO_NULL, this%comm_shared, base_ptr, win, ierr ) |
---|
899 | ! |
---|
900 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
901 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
902 | ! |
---|
903 | !-- Convert C- to Fortran-pointer |
---|
904 | buf_shape(2) = n_nyng - n_nysg + 1 |
---|
905 | buf_shape(1) = n_nxrg - n_nxlg + 1 |
---|
906 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
907 | p2i(n_nxlg:, n_nysg:) => buf |
---|
908 | ! |
---|
909 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
910 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
911 | |
---|
912 | END SUBROUTINE sm_allocate_shared_2di |
---|
913 | |
---|
914 | |
---|
915 | !--------------------------------------------------------------------------------------------------! |
---|
916 | ! Description: |
---|
917 | ! ------------ |
---|
918 | !> Allocate shared 3d-REAL (64 Bit) array on ALL threads |
---|
919 | !--------------------------------------------------------------------------------------------------! |
---|
920 | SUBROUTINE sm_allocate_shared_3d_64( this, p3, d1s, d1e, d2s, d2e, d3s, d3e, win ) |
---|
921 | |
---|
922 | IMPLICIT NONE |
---|
923 | |
---|
924 | CLASS(sm_class), INTENT(inout) :: this !< |
---|
925 | |
---|
926 | INTEGER :: disp_unit !< |
---|
927 | INTEGER, INTENT(IN) :: d1e !< |
---|
928 | INTEGER, INTENT(IN) :: d1s !< |
---|
929 | INTEGER, INTENT(IN) :: d2e !< |
---|
930 | INTEGER, INTENT(IN) :: d2s !< |
---|
931 | INTEGER, INTENT(IN) :: d3e !< |
---|
932 | INTEGER, INTENT(IN) :: d3s !< |
---|
933 | INTEGER, SAVE :: pe_from = 0 !< |
---|
934 | INTEGER, INTENT(OUT) :: win !< |
---|
935 | |
---|
936 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size !< |
---|
937 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize !< |
---|
938 | |
---|
939 | INTEGER, DIMENSION(3) :: buf_shape !< |
---|
940 | |
---|
941 | REAL(dp), DIMENSION(:,:,:), POINTER :: buf !< |
---|
942 | REAL(dp), DIMENSION(:,:,:), POINTER :: p3 !< |
---|
943 | |
---|
944 | TYPE(C_PTR), SAVE :: base_ptr !< |
---|
945 | TYPE(C_PTR), SAVE :: rem_ptr !< |
---|
946 | |
---|
947 | |
---|
948 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
949 | ! |
---|
950 | !-- Allocate shared memory on node rank 0 threads. |
---|
951 | IF ( this%sh_rank == pe_from ) THEN |
---|
952 | wsize = ( d3e - d3s + 1 ) * ( d2e - d2s + 1 ) * ( d1e - d1s + 1 ) |
---|
953 | ELSE |
---|
954 | wsize = 1 |
---|
955 | ENDIF |
---|
956 | |
---|
957 | wsize = wsize * dp ! Please note, size is always in bytes, independently of the displacement |
---|
958 | ! unit |
---|
959 | |
---|
960 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, dp, MPI_INFO_NULL, this%comm_shared, base_ptr, win, ierr ) |
---|
961 | ! |
---|
962 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
963 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
964 | ! |
---|
965 | !-- Convert C- to Fortran-pointer |
---|
966 | buf_shape(3) = d3e - d3s + 1 |
---|
967 | buf_shape(2) = d2e - d2s + 1 |
---|
968 | buf_shape(1) = d1e - d1s + 1 |
---|
969 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
970 | p3(d1s:,d2s:,d3s:) => buf |
---|
971 | ! |
---|
972 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
973 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
974 | |
---|
975 | END SUBROUTINE sm_allocate_shared_3d_64 |
---|
976 | |
---|
977 | |
---|
978 | !--------------------------------------------------------------------------------------------------! |
---|
979 | ! Description: |
---|
980 | ! ------------ |
---|
981 | !> Allocate shared 3d-REAL (32 Bit) array on ALL threads |
---|
982 | !--------------------------------------------------------------------------------------------------! |
---|
983 | SUBROUTINE sm_allocate_shared_3d_32( this, p3, d1s, d1e, d2s, d2e, d3s, d3e, win ) |
---|
984 | |
---|
985 | IMPLICIT NONE |
---|
986 | |
---|
987 | CLASS(sm_class), INTENT(inout) :: this |
---|
988 | |
---|
989 | INTEGER :: disp_unit |
---|
990 | INTEGER, INTENT(IN) :: d1e |
---|
991 | INTEGER, INTENT(IN) :: d1s |
---|
992 | INTEGER, INTENT(IN) :: d2e |
---|
993 | INTEGER, INTENT(IN) :: d2s |
---|
994 | INTEGER, INTENT(IN) :: d3e |
---|
995 | INTEGER, INTENT(IN) :: d3s |
---|
996 | INTEGER, SAVE :: pe_from = 0 |
---|
997 | INTEGER, INTENT(OUT) :: win |
---|
998 | |
---|
999 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size |
---|
1000 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize |
---|
1001 | |
---|
1002 | INTEGER, DIMENSION(3) :: buf_shape |
---|
1003 | |
---|
1004 | REAL(sp), DIMENSION(:,:,:), POINTER :: buf |
---|
1005 | REAL(sp), DIMENSION(:,:,:), POINTER :: p3 |
---|
1006 | |
---|
1007 | TYPE(C_PTR), SAVE :: base_ptr |
---|
1008 | TYPE(C_PTR), SAVE :: rem_ptr |
---|
1009 | |
---|
1010 | |
---|
1011 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
1012 | ! |
---|
1013 | !-- Allocate shared memory on node rank 0 threads. |
---|
1014 | IF ( this%sh_rank == pe_from ) THEN |
---|
1015 | wsize = ( d3e - d3s + 1 ) * ( d2e - d2s + 1 ) * ( d1e - d1s + 1 ) |
---|
1016 | ELSE |
---|
1017 | wsize = 1 |
---|
1018 | ENDIF |
---|
1019 | |
---|
1020 | wsize = wsize * sp ! Please note, size is always in bytes, independently of the displacement |
---|
1021 | ! unit |
---|
1022 | |
---|
1023 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, sp, MPI_INFO_NULL, this%comm_shared, base_ptr, win, ierr ) |
---|
1024 | ! |
---|
1025 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
1026 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
1027 | ! |
---|
1028 | !-- Convert C- to Fortran-pointer |
---|
1029 | buf_shape(3) = d3e - d3s + 1 |
---|
1030 | buf_shape(2) = d2e - d2s + 1 |
---|
1031 | buf_shape(1) = d1e - d1s + 1 |
---|
1032 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
1033 | p3(d1s:,d2s:,d3s:) => buf |
---|
1034 | ! |
---|
1035 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
1036 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
1037 | |
---|
1038 | END SUBROUTINE sm_allocate_shared_3d_32 |
---|
1039 | |
---|
1040 | |
---|
1041 | !--------------------------------------------------------------------------------------------------! |
---|
1042 | ! Description: |
---|
1043 | ! ------------ |
---|
1044 | !> Allocate shared 3d-REAL (32 bit) array on ALL threads |
---|
1045 | !--------------------------------------------------------------------------------------------------! |
---|
1046 | SUBROUTINE sm_allocate_shared_3di_32( this, p3, d1s, d1e, d2s, d2e, d3s, d3e, win ) |
---|
1047 | |
---|
1048 | IMPLICIT NONE |
---|
1049 | |
---|
1050 | CLASS(sm_class), INTENT(inout) :: this |
---|
1051 | |
---|
1052 | INTEGER :: disp_unit |
---|
1053 | INTEGER, INTENT(IN) :: d1e |
---|
1054 | INTEGER, INTENT(IN) :: d1s |
---|
1055 | INTEGER, INTENT(IN) :: d2e |
---|
1056 | INTEGER, INTENT(IN) :: d2s |
---|
1057 | INTEGER, INTENT(IN) :: d3e |
---|
1058 | INTEGER, INTENT(IN) :: d3s |
---|
1059 | INTEGER, SAVE :: pe_from = 0 |
---|
1060 | INTEGER, INTENT(OUT) :: win |
---|
1061 | |
---|
1062 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size |
---|
1063 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize |
---|
1064 | |
---|
1065 | INTEGER, DIMENSION(3) :: buf_shape |
---|
1066 | |
---|
1067 | INTEGER(isp), DIMENSION(:,:,:), POINTER :: buf |
---|
1068 | INTEGER(isp), DIMENSION(:,:,:), POINTER :: p3 |
---|
1069 | |
---|
1070 | TYPE(C_PTR), SAVE :: base_ptr |
---|
1071 | TYPE(C_PTR), SAVE :: rem_ptr |
---|
1072 | |
---|
1073 | |
---|
1074 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
1075 | ! |
---|
1076 | !-- Allocate shared memory on node rank 0 threads. |
---|
1077 | IF ( this%sh_rank == pe_from ) THEN |
---|
1078 | wsize = ( d3e - d3s + 1 ) * ( d2e - d2s + 1 ) * ( d1e - d1s + 1 ) |
---|
1079 | ELSE |
---|
1080 | wsize = 1 |
---|
1081 | ENDIF |
---|
1082 | |
---|
1083 | wsize = wsize * isp ! Please note, size is always in bytes, independently of the displacement |
---|
1084 | ! unit |
---|
1085 | |
---|
1086 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, isp, MPI_INFO_NULL, this%comm_shared, base_ptr, win, ierr ) |
---|
1087 | ! |
---|
1088 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
1089 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
1090 | ! |
---|
1091 | !-- Convert C- to Fortran-pointer |
---|
1092 | buf_shape(3) = d3e - d3s + 1 |
---|
1093 | buf_shape(2) = d2e - d2s + 1 |
---|
1094 | buf_shape(1) = d1e - d1s + 1 |
---|
1095 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
1096 | p3(d1s:,d2s:,d3s:) => buf |
---|
1097 | ! |
---|
1098 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
1099 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
1100 | |
---|
1101 | END SUBROUTINE sm_allocate_shared_3di_32 |
---|
1102 | |
---|
1103 | |
---|
1104 | !--------------------------------------------------------------------------------------------------! |
---|
1105 | ! Description: |
---|
1106 | ! ------------ |
---|
1107 | !> Allocate shared 3d-REAL (64 bit) array on ALL threads |
---|
1108 | !--------------------------------------------------------------------------------------------------! |
---|
1109 | SUBROUTINE sm_allocate_shared_3di_64( this, p3, d1s, d1e, d2s, d2e, d3s, d3e, win ) |
---|
1110 | |
---|
1111 | IMPLICIT NONE |
---|
1112 | |
---|
1113 | CLASS(sm_class), INTENT(inout) :: this !< |
---|
1114 | |
---|
1115 | INTEGER :: disp_unit !< |
---|
1116 | INTEGER, INTENT(IN) :: d1e !< |
---|
1117 | INTEGER, INTENT(IN) :: d1s !< |
---|
1118 | INTEGER, INTENT(IN) :: d2e !< |
---|
1119 | INTEGER, INTENT(IN) :: d2s !< |
---|
1120 | INTEGER, INTENT(IN) :: d3e !< |
---|
1121 | INTEGER, INTENT(IN) :: d3s !< |
---|
1122 | INTEGER, SAVE :: pe_from = 0 !< |
---|
1123 | INTEGER, INTENT(OUT) :: win !< |
---|
1124 | |
---|
1125 | INTEGER(KIND=MPI_ADDRESS_KIND) :: rem_size !< |
---|
1126 | INTEGER(KIND=MPI_ADDRESS_KIND) :: wsize !< |
---|
1127 | |
---|
1128 | INTEGER, DIMENSION(3) :: buf_shape !< |
---|
1129 | |
---|
1130 | INTEGER(idp), DIMENSION(:,:,:), POINTER :: buf !< |
---|
1131 | INTEGER(idp), DIMENSION(:,:,:), POINTER :: p3 !< |
---|
1132 | |
---|
1133 | TYPE(C_PTR), SAVE :: base_ptr !< |
---|
1134 | TYPE(C_PTR), SAVE :: rem_ptr !< |
---|
1135 | |
---|
1136 | |
---|
1137 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
1138 | ! |
---|
1139 | !-- Allocate shared memory on node rank 0 threads. |
---|
1140 | IF ( this%sh_rank == pe_from ) THEN |
---|
1141 | wsize = ( d3e - d3s + 1 ) * ( d2e - d2s + 1 ) * ( d1e - d1s + 1 ) |
---|
1142 | ELSE |
---|
1143 | wsize = 1 |
---|
1144 | ENDIF |
---|
1145 | |
---|
1146 | wsize = wsize * idp ! Please note, size is always in bytes, independently of the displacement |
---|
1147 | ! unit |
---|
1148 | |
---|
1149 | CALL MPI_WIN_ALLOCATE_SHARED( wsize, idp, MPI_INFO_NULL, this%comm_shared, base_ptr, win, ierr ) |
---|
1150 | ! |
---|
1151 | !-- Get C-pointer of the memory located on node-rank pe_from (sh_rank == pe_from) |
---|
1152 | CALL MPI_WIN_SHARED_QUERY( win, pe_from, rem_size, disp_unit, rem_ptr, ierr ) |
---|
1153 | ! |
---|
1154 | !-- Convert C- to Fortran-pointer |
---|
1155 | buf_shape(3) = d3e - d3s + 1 |
---|
1156 | buf_shape(2) = d2e - d2s + 1 |
---|
1157 | buf_shape(1) = d1e - d1s + 1 |
---|
1158 | CALL C_F_POINTER( rem_ptr, buf, buf_shape ) |
---|
1159 | p3(d1s:,d2s:,d3s:) => buf |
---|
1160 | ! |
---|
1161 | !-- Allocate shared memory in round robin on all PEs of a node. |
---|
1162 | pe_from = MOD( pe_from, this%sh_npes ) |
---|
1163 | |
---|
1164 | END SUBROUTINE sm_allocate_shared_3di_64 |
---|
1165 | |
---|
1166 | #endif |
---|
1167 | |
---|
1168 | |
---|
1169 | !--------------------------------------------------------------------------------------------------! |
---|
1170 | ! Description: |
---|
1171 | ! ------------ |
---|
1172 | !> ??? |
---|
1173 | !--------------------------------------------------------------------------------------------------! |
---|
1174 | SUBROUTINE sm_adjust_outer_boundary( this ) |
---|
1175 | |
---|
1176 | IMPLICIT NONE |
---|
1177 | |
---|
1178 | CLASS(sm_class), INTENT(inout) :: this !< |
---|
1179 | |
---|
1180 | |
---|
1181 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
1182 | |
---|
1183 | IF ( this%io_grid%nxl == 0 ) THEN |
---|
1184 | this%io_grid%nxl = this%io_grid%nxl - nbgp |
---|
1185 | this%io_grid%nnx = this%io_grid%nnx + nbgp |
---|
1186 | ENDIF |
---|
1187 | |
---|
1188 | IF ( this%io_grid%nxr == nx .OR. npex == -1 ) THEN ! npex == -1 if -D__parallel not set |
---|
1189 | this%io_grid%nxr = this%io_grid%nxr + nbgp |
---|
1190 | this%io_grid%nnx = this%io_grid%nnx + nbgp |
---|
1191 | ENDIF |
---|
1192 | |
---|
1193 | IF ( this%io_grid%nys == 0 ) THEN |
---|
1194 | this%io_grid%nys = this%io_grid%nys - nbgp |
---|
1195 | this%io_grid%nny = this%io_grid%nny + nbgp |
---|
1196 | ENDIF |
---|
1197 | |
---|
1198 | IF ( this%io_grid%nyn == ny .OR. npey == -1 ) THEN ! npey == -1 if -D__parallel not set |
---|
1199 | this%io_grid%nyn = this%io_grid%nyn + nbgp |
---|
1200 | this%io_grid%nny = this%io_grid%nny + nbgp |
---|
1201 | ENDIF |
---|
1202 | |
---|
1203 | this%io_grid%nxl = this%io_grid%nxl + nbgp |
---|
1204 | this%io_grid%nxr = this%io_grid%nxr + nbgp |
---|
1205 | this%io_grid%nys = this%io_grid%nys + nbgp |
---|
1206 | this%io_grid%nyn = this%io_grid%nyn + nbgp |
---|
1207 | this%io_grid%nnx = this%io_grid%nnx |
---|
1208 | this%io_grid%nny = this%io_grid%nny |
---|
1209 | |
---|
1210 | END SUBROUTINE sm_adjust_outer_boundary |
---|
1211 | |
---|
1212 | |
---|
1213 | !--------------------------------------------------------------------------------------------------! |
---|
1214 | ! Description: |
---|
1215 | ! ------------ |
---|
1216 | !> Deallocate shared aray and free related window. |
---|
1217 | !--------------------------------------------------------------------------------------------------! |
---|
1218 | SUBROUTINE sm_free_shared( this, win ) |
---|
1219 | |
---|
1220 | IMPLICIT NONE |
---|
1221 | |
---|
1222 | CLASS(sm_class), INTENT(inout) :: this !< |
---|
1223 | |
---|
1224 | INTEGER(iwp), INTENT(INOUT) :: win !< |
---|
1225 | |
---|
1226 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
1227 | #if defined( __parallel ) |
---|
1228 | CALL MPI_WIN_FREE( win, ierr ) |
---|
1229 | #endif |
---|
1230 | win = -1 |
---|
1231 | |
---|
1232 | END SUBROUTINE sm_free_shared |
---|
1233 | |
---|
1234 | |
---|
1235 | !--------------------------------------------------------------------------------------------------! |
---|
1236 | ! Description: |
---|
1237 | ! ------------ |
---|
1238 | !> ... |
---|
1239 | !--------------------------------------------------------------------------------------------------! |
---|
1240 | SUBROUTINE sm_node_barrier( this ) |
---|
1241 | |
---|
1242 | IMPLICIT NONE |
---|
1243 | |
---|
1244 | CLASS(sm_class), INTENT(inout) :: this !< |
---|
1245 | |
---|
1246 | |
---|
1247 | IF ( this%no_shared_memory_in_this_run ) RETURN |
---|
1248 | |
---|
1249 | #if defined( __parallel ) |
---|
1250 | CALL MPI_BARRIER( this%comm_shared, ierr ) |
---|
1251 | #endif |
---|
1252 | |
---|
1253 | END SUBROUTINE sm_node_barrier |
---|
1254 | |
---|
1255 | |
---|
1256 | SUBROUTINE save_grid_into_this_class( this ) |
---|
1257 | |
---|
1258 | IMPLICIT NONE |
---|
1259 | |
---|
1260 | CLASS(domain_decomposition_grid_features), INTENT(inout) :: this !< |
---|
1261 | |
---|
1262 | this%myid = myid !< |
---|
1263 | this%nnx = nnx !< |
---|
1264 | this%nny = nny !< |
---|
1265 | this%nx = nx !< |
---|
1266 | this%nxl = nxl !< |
---|
1267 | this%nxr = nxr !< |
---|
1268 | this%ny = ny !< |
---|
1269 | this%nyn = nyn !< |
---|
1270 | this%nys = nys !< |
---|
1271 | this%numprocs = numprocs !< |
---|
1272 | this%comm2d = comm2d !< |
---|
1273 | |
---|
1274 | END SUBROUTINE save_grid_into_this_class |
---|
1275 | |
---|
1276 | |
---|
1277 | SUBROUTINE activate_grid_from_this_class( this ) |
---|
1278 | |
---|
1279 | IMPLICIT NONE |
---|
1280 | |
---|
1281 | CLASS(domain_decomposition_grid_features), INTENT(inout) :: this !< |
---|
1282 | |
---|
1283 | myid = this%myid !< |
---|
1284 | nnx = this%nnx !< |
---|
1285 | nny = this%nny !< |
---|
1286 | nx = this%nx !< |
---|
1287 | nxl = this%nxl !< |
---|
1288 | nxr = this%nxr !< |
---|
1289 | ny = this%ny !< |
---|
1290 | nyn = this%nyn !< |
---|
1291 | nys = this%nys !< |
---|
1292 | numprocs = this%numprocs !< |
---|
1293 | comm2d = this%comm2d !< |
---|
1294 | |
---|
1295 | END SUBROUTINE activate_grid_from_this_class |
---|
1296 | |
---|
1297 | END MODULE shared_memory_io_mod |
---|