1 | !> @file timestep.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: timestep.f90 4717 2020-09-30 22:27:40Z Giersch $ |
---|
27 | ! Fixes and optimizations of OpenMP parallelization, formatting of OpenMP |
---|
28 | ! directives (J. Resler) |
---|
29 | ! |
---|
30 | ! 4564 2020-06-12 14:03:36Z raasch |
---|
31 | ! Vertical nesting method of Huq et al. (2019) removed |
---|
32 | ! |
---|
33 | ! 4540 2020-05-18 15:23:29Z raasch |
---|
34 | ! File re-formatted to follow the PALM coding standard |
---|
35 | ! |
---|
36 | ! 4444 2020-03-05 15:59:50Z raasch |
---|
37 | ! Bugfix: cpp-directives for serial mode added |
---|
38 | ! |
---|
39 | ! 4360 2020-01-07 11:25:50Z suehring |
---|
40 | ! Added missing OpenMP directives |
---|
41 | ! |
---|
42 | ! 4233 2019-09-20 09:55:54Z knoop |
---|
43 | ! OpenACC data update host removed |
---|
44 | ! |
---|
45 | ! 4182 2019-08-22 15:20:23Z scharf |
---|
46 | ! Corrected "Former revisions" section |
---|
47 | ! |
---|
48 | ! 4101 2019-07-17 15:14:26Z gronemeier |
---|
49 | ! - Consider 2*Km within diffusion criterion as Km is considered twice within the diffusion of e, |
---|
50 | ! - in RANS mode, instead of considering each wind component individually use the wind speed of 3d |
---|
51 | ! wind vector in CFL criterion |
---|
52 | ! - Do not limit the increase of dt based on its previous value in RANS mode |
---|
53 | ! |
---|
54 | ! 3658 2019-01-07 20:28:54Z knoop |
---|
55 | ! OpenACC port for SPEC |
---|
56 | ! |
---|
57 | ! Revision 1.1 1997/08/11 06:26:19 raasch |
---|
58 | ! Initial revision |
---|
59 | ! |
---|
60 | ! |
---|
61 | ! Description: |
---|
62 | ! ------------ |
---|
63 | !> Compute the time step under consideration of the FCL and diffusion criterion. |
---|
64 | !--------------------------------------------------------------------------------------------------! |
---|
65 | SUBROUTINE timestep |
---|
66 | |
---|
67 | |
---|
68 | USE arrays_3d, & |
---|
69 | ONLY: dzu, & |
---|
70 | dzw, & |
---|
71 | kh, & |
---|
72 | km, & |
---|
73 | u, & |
---|
74 | u_stokes_zu, & |
---|
75 | v, & |
---|
76 | v_stokes_zu, & |
---|
77 | w |
---|
78 | |
---|
79 | USE control_parameters, & |
---|
80 | ONLY: cfl_factor, & |
---|
81 | dt_3d, & |
---|
82 | dt_fixed, & |
---|
83 | dt_max, & |
---|
84 | galilei_transformation, & |
---|
85 | message_string, & |
---|
86 | rans_mode, & |
---|
87 | stop_dt, & |
---|
88 | timestep_reason, & |
---|
89 | u_gtrans, & |
---|
90 | use_ug_for_galilei_tr, & |
---|
91 | v_gtrans |
---|
92 | |
---|
93 | #if defined( __parallel ) |
---|
94 | USE control_parameters, & |
---|
95 | ONLY: coupling_mode, & |
---|
96 | terminate_coupled, & |
---|
97 | terminate_coupled_remote |
---|
98 | #endif |
---|
99 | |
---|
100 | USE cpulog, & |
---|
101 | ONLY: cpu_log, & |
---|
102 | log_point |
---|
103 | |
---|
104 | USE grid_variables, & |
---|
105 | ONLY: dx, & |
---|
106 | dx2, & |
---|
107 | dy, & |
---|
108 | dy2 |
---|
109 | |
---|
110 | USE indices, & |
---|
111 | ONLY: nxl, & |
---|
112 | nxlg, & |
---|
113 | nxr, & |
---|
114 | nxrg, & |
---|
115 | nyn, & |
---|
116 | nyng, & |
---|
117 | nys, & |
---|
118 | nysg, & |
---|
119 | nzb, & |
---|
120 | nzt |
---|
121 | |
---|
122 | USE interfaces |
---|
123 | |
---|
124 | USE kinds |
---|
125 | |
---|
126 | USE bulk_cloud_model_mod, & |
---|
127 | ONLY: dt_precipitation |
---|
128 | |
---|
129 | USE pegrid |
---|
130 | |
---|
131 | USE pmc_interface, & |
---|
132 | ONLY: nested_run |
---|
133 | |
---|
134 | USE statistics, & |
---|
135 | ONLY: flow_statistics_called, & |
---|
136 | hom, & |
---|
137 | u_max, & |
---|
138 | u_max_ijk, & |
---|
139 | v_max, & |
---|
140 | v_max_ijk, & |
---|
141 | w_max, & |
---|
142 | w_max_ijk |
---|
143 | |
---|
144 | IMPLICIT NONE |
---|
145 | |
---|
146 | INTEGER(iwp) :: i !< |
---|
147 | INTEGER(iwp) :: j !< |
---|
148 | INTEGER(iwp) :: k !< |
---|
149 | INTEGER(iwp) :: km_max_ijk(3) = -1 !< index values (i,j,k) of location where km_max occurs |
---|
150 | INTEGER(iwp) :: kh_max_ijk(3) = -1 !< index values (i,j,k) of location where kh_max occurs |
---|
151 | |
---|
152 | LOGICAL :: stop_dt_local !< local switch for controlling the time stepping |
---|
153 | |
---|
154 | REAL(wp) :: div !< |
---|
155 | REAL(wp) :: dt_diff !< |
---|
156 | REAL(wp) :: dt_diff_l !< |
---|
157 | REAL(wp) :: dt_u !< |
---|
158 | REAL(wp) :: dt_u_l !< |
---|
159 | REAL(wp) :: dt_v !< |
---|
160 | REAL(wp) :: dt_v_l !< |
---|
161 | REAL(wp) :: dt_w !< |
---|
162 | REAL(wp) :: dt_w_l !< |
---|
163 | REAL(wp) :: km_max !< maximum of Km in entire domain |
---|
164 | REAL(wp) :: kh_max !< maximum of Kh in entire domain |
---|
165 | REAL(wp) :: u_gtrans_l !< |
---|
166 | REAL(wp) :: v_gtrans_l !< |
---|
167 | |
---|
168 | REAL(wp), DIMENSION(2) :: uv_gtrans_l !< |
---|
169 | #if defined( __parallel ) |
---|
170 | REAL(wp), DIMENSION(2) :: uv_gtrans !< |
---|
171 | REAL(wp), DIMENSION(3) :: reduce !< |
---|
172 | REAL(wp), DIMENSION(3) :: reduce_l !< |
---|
173 | #endif |
---|
174 | REAL(wp), DIMENSION(nzb+1:nzt) :: dxyz2_min !< |
---|
175 | !$ACC DECLARE CREATE(dxyz2_min) |
---|
176 | |
---|
177 | |
---|
178 | CALL cpu_log( log_point(12), 'calculate_timestep', 'start' ) |
---|
179 | |
---|
180 | ! |
---|
181 | !-- In case of Galilei-transform not using the geostrophic wind as translation velocity, compute the |
---|
182 | !-- volume-averaged horizontal velocity components, which will then be subtracted from the |
---|
183 | !-- horizontal wind for the time step and horizontal advection routines. |
---|
184 | IF ( galilei_transformation .AND. .NOT. use_ug_for_galilei_tr ) THEN |
---|
185 | IF ( flow_statistics_called ) THEN |
---|
186 | ! |
---|
187 | !-- Horizontal averages already existent, just need to average them vertically. |
---|
188 | u_gtrans = 0.0_wp |
---|
189 | v_gtrans = 0.0_wp |
---|
190 | DO k = nzb+1, nzt |
---|
191 | u_gtrans = u_gtrans + hom(k,1,1,0) |
---|
192 | v_gtrans = v_gtrans + hom(k,1,2,0) |
---|
193 | ENDDO |
---|
194 | u_gtrans = u_gtrans / REAL( nzt - nzb, KIND = wp ) |
---|
195 | v_gtrans = v_gtrans / REAL( nzt - nzb, KIND = wp ) |
---|
196 | ELSE |
---|
197 | ! |
---|
198 | !-- Averaging over the entire model domain. |
---|
199 | u_gtrans_l = 0.0_wp |
---|
200 | v_gtrans_l = 0.0_wp |
---|
201 | DO i = nxl, nxr |
---|
202 | DO j = nys, nyn |
---|
203 | DO k = nzb+1, nzt |
---|
204 | u_gtrans_l = u_gtrans_l + u(k,j,i) |
---|
205 | v_gtrans_l = v_gtrans_l + v(k,j,i) |
---|
206 | ENDDO |
---|
207 | ENDDO |
---|
208 | ENDDO |
---|
209 | uv_gtrans_l(1) = u_gtrans_l / REAL( (nxr-nxl+1) * (nyn-nys+1) * (nzt-nzb), KIND = wp ) |
---|
210 | uv_gtrans_l(2) = v_gtrans_l / REAL( (nxr-nxl+1) * (nyn-nys+1) * (nzt-nzb), KIND = wp ) |
---|
211 | #if defined( __parallel ) |
---|
212 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
213 | CALL MPI_ALLREDUCE( uv_gtrans_l, uv_gtrans, 2, MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
214 | u_gtrans = uv_gtrans(1) / REAL( numprocs, KIND = wp ) |
---|
215 | v_gtrans = uv_gtrans(2) / REAL( numprocs, KIND = wp ) |
---|
216 | #else |
---|
217 | u_gtrans = uv_gtrans_l(1) |
---|
218 | v_gtrans = uv_gtrans_l(2) |
---|
219 | #endif |
---|
220 | ENDIF |
---|
221 | ENDIF |
---|
222 | |
---|
223 | ! |
---|
224 | !-- Determine the maxima of the velocity components, including their grid index positions. |
---|
225 | CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, u, 'abs', 0.0_wp, u_max, u_max_ijk ) |
---|
226 | CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, v, 'abs', 0.0_wp, v_max, v_max_ijk ) |
---|
227 | CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, w, 'abs', 0.0_wp, w_max, w_max_ijk ) |
---|
228 | |
---|
229 | IF ( .NOT. dt_fixed ) THEN |
---|
230 | ! |
---|
231 | !-- Variable time step: |
---|
232 | !-- Calculate the maximum time step according to the CFL-criterion |
---|
233 | dt_u_l = 999999.9_wp |
---|
234 | dt_v_l = 999999.9_wp |
---|
235 | dt_w_l = 999999.9_wp |
---|
236 | |
---|
237 | IF ( .NOT. rans_mode ) THEN |
---|
238 | ! |
---|
239 | !-- Consider each velocity component individually |
---|
240 | |
---|
241 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
242 | !$ACC COPY(dt_u_l, dt_v_l, dt_w_l, u_stokes_zu, v_stokes_zu) & |
---|
243 | !$ACC REDUCTION(MIN: dt_u_l, dt_v_l, dt_w_l) & |
---|
244 | !$ACC PRESENT(u, v, w, dzu) |
---|
245 | !$OMP PARALLEL DO PRIVATE(i,j,k) & |
---|
246 | !$OMP REDUCTION(MIN: dt_u_l, dt_v_l, dt_w_l) |
---|
247 | DO i = nxl, nxr |
---|
248 | DO j = nys, nyn |
---|
249 | DO k = nzb+1, nzt |
---|
250 | dt_u_l = MIN( dt_u_l, ( dx / ( ABS( u(k,j,i) - u_gtrans + u_stokes_zu(k) ) & |
---|
251 | + 1.0E-10_wp ) ) ) |
---|
252 | dt_v_l = MIN( dt_v_l, ( dy / ( ABS( v(k,j,i) - v_gtrans + v_stokes_zu(k) ) & |
---|
253 | + 1.0E-10_wp ) ) ) |
---|
254 | dt_w_l = MIN( dt_w_l, ( dzu(k) / ( ABS( w(k,j,i) ) + 1.0E-10_wp ) ) ) |
---|
255 | ENDDO |
---|
256 | ENDDO |
---|
257 | ENDDO |
---|
258 | |
---|
259 | ELSE |
---|
260 | ! |
---|
261 | !-- Consider the wind speed at the scalar-grid point |
---|
262 | !-- !> @note Considering the wind speed instead of each individual wind component is only a |
---|
263 | !-- !> workaround so far. This has to be changed in the future. |
---|
264 | |
---|
265 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
266 | !$ACC COPY(dt_u_l, u_stokes_zu, v_stokes_zu) & |
---|
267 | !$ACC REDUCTION(MIN: dt_u_l) & |
---|
268 | !$ACC PRESENT(u, v, w, dzu) |
---|
269 | !$OMP PARALLEL DO PRIVATE(i,j,k) & |
---|
270 | !$OMP REDUCTION(MIN: dt_u_l) |
---|
271 | DO i = nxl, nxr |
---|
272 | DO j = nys, nyn |
---|
273 | DO k = nzb+1, nzt |
---|
274 | dt_u_l = MIN( dt_u_l, ( MIN( dx, dy, dzu(k) ) / ( SQRT( & |
---|
275 | ( 0.5 * ( u(k,j,i) + u(k,j,i+1) ) - u_gtrans + u_stokes_zu(k) )**2 & |
---|
276 | + ( 0.5 * ( v(k,j,i) + v(k,j+1,i) ) - v_gtrans + v_stokes_zu(k) )**2 & |
---|
277 | + ( 0.5 * ( w(k,j,i) + w(k-1,j,i) ) )**2 ) + 1.0E-10_wp ) ) ) |
---|
278 | ENDDO |
---|
279 | ENDDO |
---|
280 | ENDDO |
---|
281 | |
---|
282 | dt_v_l = dt_u_l |
---|
283 | dt_w_l = dt_u_l |
---|
284 | |
---|
285 | ENDIF |
---|
286 | |
---|
287 | #if defined( __parallel ) |
---|
288 | reduce_l(1) = dt_u_l |
---|
289 | reduce_l(2) = dt_v_l |
---|
290 | reduce_l(3) = dt_w_l |
---|
291 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
292 | CALL MPI_ALLREDUCE( reduce_l, reduce, 3, MPI_REAL, MPI_MIN, comm2d, ierr ) |
---|
293 | dt_u = reduce(1) |
---|
294 | dt_v = reduce(2) |
---|
295 | dt_w = reduce(3) |
---|
296 | #else |
---|
297 | dt_u = dt_u_l |
---|
298 | dt_v = dt_v_l |
---|
299 | dt_w = dt_w_l |
---|
300 | #endif |
---|
301 | |
---|
302 | ! |
---|
303 | !-- Compute time step according to the diffusion criterion. |
---|
304 | !-- First calculate minimum grid spacing which only depends on index k. When using the dynamic |
---|
305 | !-- subgrid model, negative km are possible. |
---|
306 | dt_diff_l = 999999.0_wp |
---|
307 | |
---|
308 | !$ACC PARALLEL LOOP PRESENT(dxyz2_min, dzw) |
---|
309 | DO k = nzb+1, nzt |
---|
310 | dxyz2_min(k) = MIN( dx2, dy2, dzw(k) * dzw(k) ) * 0.125_wp |
---|
311 | ENDDO |
---|
312 | |
---|
313 | !$OMP PARALLEL DO private(i,j,k) reduction(MIN: dt_diff_l) |
---|
314 | !$ACC PARALLEL LOOP COLLAPSE(3) PRIVATE(i,j,k) & |
---|
315 | !$ACC COPY(dt_diff_l) REDUCTION(MIN: dt_diff_l) & |
---|
316 | !$ACC PRESENT(dxyz2_min, kh, km) |
---|
317 | DO i = nxl, nxr |
---|
318 | DO j = nys, nyn |
---|
319 | DO k = nzb+1, nzt |
---|
320 | dt_diff_l = MIN( dt_diff_l, dxyz2_min(k) / ( MAX( kh(k,j,i), 2.0_wp * & |
---|
321 | ABS( km(k,j,i) ) ) + 1E-20_wp ) ) |
---|
322 | ENDDO |
---|
323 | ENDDO |
---|
324 | ENDDO |
---|
325 | #if defined( __parallel ) |
---|
326 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
327 | CALL MPI_ALLREDUCE( dt_diff_l, dt_diff, 1, MPI_REAL, MPI_MIN, comm2d, ierr ) |
---|
328 | #else |
---|
329 | dt_diff = dt_diff_l |
---|
330 | #endif |
---|
331 | |
---|
332 | ! |
---|
333 | !-- The time step is the minimum of the 3-4 components and the diffusion time step minus a |
---|
334 | !-- reduction (cfl_factor) to be on the safe side. |
---|
335 | !-- The time step must not exceed the maximum allowed value. |
---|
336 | dt_3d = cfl_factor * MIN( dt_diff, dt_u, dt_v, dt_w, dt_precipitation ) |
---|
337 | dt_3d = MIN( dt_3d, dt_max ) |
---|
338 | |
---|
339 | ! |
---|
340 | !-- Remember the restricting time step criterion for later output. |
---|
341 | IF ( MIN( dt_u, dt_v, dt_w ) < dt_diff ) THEN |
---|
342 | timestep_reason = 'A' |
---|
343 | ELSE |
---|
344 | timestep_reason = 'D' |
---|
345 | ENDIF |
---|
346 | |
---|
347 | ! |
---|
348 | !-- Set flag if the time step becomes too small. |
---|
349 | IF ( dt_3d < ( 0.00001_wp * dt_max ) ) THEN |
---|
350 | stop_dt = .TRUE. |
---|
351 | |
---|
352 | ! |
---|
353 | !-- Determine the maxima of the diffusion coefficients, including their grid index positions. |
---|
354 | CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, km, 'abs', 0.0_wp, km_max, & |
---|
355 | km_max_ijk ) |
---|
356 | CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, kh, 'abs', 0.0_wp, kh_max, & |
---|
357 | kh_max_ijk ) |
---|
358 | |
---|
359 | WRITE( message_string, * ) 'Time step has reached minimum limit.', & |
---|
360 | '&dt = ', dt_3d, ' s Simulation is terminated.', & |
---|
361 | '&dt_u = ', dt_u, ' s', & |
---|
362 | '&dt_v = ', dt_v, ' s', & |
---|
363 | '&dt_w = ', dt_w, ' s', & |
---|
364 | '&dt_diff = ', dt_diff, ' s', & |
---|
365 | '&u_max = ', u_max, ' m/s k=', u_max_ijk(1), & |
---|
366 | ' j=', u_max_ijk(2), ' i=', u_max_ijk(3), & |
---|
367 | '&v_max = ', v_max, ' m/s k=', v_max_ijk(1), & |
---|
368 | ' j=', v_max_ijk(2), ' i=', v_max_ijk(3), & |
---|
369 | '&w_max = ', w_max, ' m/s k=', w_max_ijk(1), & |
---|
370 | ' j=', w_max_ijk(2), ' i=', w_max_ijk(3), & |
---|
371 | '&km_max = ', km_max, ' m2/s2 k=', km_max_ijk(1), & |
---|
372 | ' j=', km_max_ijk(2), ' i=', km_max_ijk(3), & |
---|
373 | '&kh_max = ', kh_max, ' m2/s2 k=', kh_max_ijk(1), & |
---|
374 | ' j=', kh_max_ijk(2), ' i=', kh_max_ijk(3) |
---|
375 | CALL message( 'timestep', 'PA0312', 0, 1, 0, 6, 0 ) |
---|
376 | ! |
---|
377 | !-- In case of coupled runs inform the remote model of the termination and its reason, |
---|
378 | !-- provided the remote model has not already been informed of another termination reason |
---|
379 | !-- (terminate_coupled > 0). |
---|
380 | #if defined( __parallel ) |
---|
381 | IF ( coupling_mode /= 'uncoupled' .AND. terminate_coupled == 0 ) THEN |
---|
382 | terminate_coupled = 2 |
---|
383 | IF ( myid == 0 ) THEN |
---|
384 | CALL MPI_SENDRECV( terminate_coupled, 1, MPI_INTEGER, target_id, 0, & |
---|
385 | terminate_coupled_remote, 1, MPI_INTEGER, target_id, 0, & |
---|
386 | comm_inter, status, ierr ) |
---|
387 | ENDIF |
---|
388 | CALL MPI_BCAST( terminate_coupled_remote, 1, MPI_INTEGER, 0, comm2d, ierr) |
---|
389 | ENDIF |
---|
390 | #endif |
---|
391 | ENDIF |
---|
392 | |
---|
393 | ! |
---|
394 | !-- In case of nested runs all parent/child processes have to terminate if one process has set |
---|
395 | !-- the stop flag, i.e. they need to set the stop flag too. |
---|
396 | IF ( nested_run ) THEN |
---|
397 | stop_dt_local = stop_dt |
---|
398 | #if defined( __parallel ) |
---|
399 | CALL MPI_ALLREDUCE( stop_dt_local, stop_dt, 1, MPI_LOGICAL, MPI_LOR, MPI_COMM_WORLD, ierr ) |
---|
400 | #endif |
---|
401 | ENDIF |
---|
402 | |
---|
403 | ! |
---|
404 | !-- Ensure a smooth value (two significant digits) of the timestep. |
---|
405 | div = 1000.0_wp |
---|
406 | DO WHILE ( dt_3d < div ) |
---|
407 | div = div / 10.0_wp |
---|
408 | ENDDO |
---|
409 | dt_3d = NINT( dt_3d * 100.0_wp / div ) * div / 100.0_wp |
---|
410 | |
---|
411 | ENDIF |
---|
412 | |
---|
413 | CALL cpu_log( log_point(12), 'calculate_timestep', 'stop' ) |
---|
414 | |
---|
415 | END SUBROUTINE timestep |
---|