1 | SUBROUTINE pres |
---|
2 | |
---|
3 | !------------------------------------------------------------------------------! |
---|
4 | ! Current revisions: |
---|
5 | ! ----------------- |
---|
6 | ! |
---|
7 | ! |
---|
8 | ! Former revisions: |
---|
9 | ! ----------------- |
---|
10 | ! $Id: pres.f90 779 2011-11-07 14:30:01Z maronga $ |
---|
11 | ! |
---|
12 | ! 778 2011-11-07 14:18:25Z fricke |
---|
13 | ! New allocation of tend when multigrid is used and the collected field on PE0 |
---|
14 | ! has more grid points than the subdomain of an PE. |
---|
15 | ! |
---|
16 | ! 719 2011-04-06 13:05:23Z gryschka |
---|
17 | ! Bugfix in volume flow control for double cyclic boundary conditions |
---|
18 | ! |
---|
19 | ! 709 2011-03-30 09:31:40Z raasch |
---|
20 | ! formatting adjustments |
---|
21 | ! |
---|
22 | ! 707 2011-03-29 11:39:40Z raasch |
---|
23 | ! Calculation of weighted average of p is now handled in the same way |
---|
24 | ! regardless of the number of ghost layers (advection scheme), |
---|
25 | ! multigrid and sor method are using p_loc for iterative advancements of |
---|
26 | ! pressure, |
---|
27 | ! localsum calculation modified for proper OpenMP reduction, |
---|
28 | ! bc_lr/ns replaced by bc_lr/ns_cyc |
---|
29 | ! |
---|
30 | ! 693 2011-03-08 09:..:..Z raasch |
---|
31 | ! bugfix: weighting coefficient added to ibm branch |
---|
32 | ! |
---|
33 | ! 680 2011-02-04 23:16:06Z gryschka |
---|
34 | ! bugfix: collective_wait |
---|
35 | ! |
---|
36 | ! 675 2011-01-19 10:56:55Z suehring |
---|
37 | ! Removed bugfix while copying tend. |
---|
38 | ! |
---|
39 | ! 673 2011-01-18 16:19:48Z suehring |
---|
40 | ! Weighting coefficients added for right computation of the pressure during |
---|
41 | ! Runge-Kutta substeps. |
---|
42 | ! |
---|
43 | ! 667 2010-12-23 12:06:00Z suehring/gryschka |
---|
44 | ! New allocation of tend when ws-scheme and multigrid is used. This is due to |
---|
45 | ! reasons of perforance of the data_exchange. The same is done with p after |
---|
46 | ! poismg is called. |
---|
47 | ! nxl-1, nxr+1, nys-1, nyn+1 replaced by nxlg, nxrg, nysg, nyng when no |
---|
48 | ! multigrid is used. Calls of exchange_horiz are modified. |
---|
49 | ! bugfix: After pressure correction no volume flow correction in case of |
---|
50 | ! non-cyclic boundary conditions |
---|
51 | ! (has to be done only before pressure correction) |
---|
52 | ! Call of SOR routine is referenced with ddzu_pres. |
---|
53 | ! |
---|
54 | ! 622 2010-12-10 08:08:13Z raasch |
---|
55 | ! optional barriers included in order to speed up collective operations |
---|
56 | ! |
---|
57 | ! 151 2008-03-07 13:42:18Z raasch |
---|
58 | ! Bugfix in volume flow control for non-cyclic boundary conditions |
---|
59 | ! |
---|
60 | ! 106 2007-08-16 14:30:26Z raasch |
---|
61 | ! Volume flow conservation added for the remaining three outflow boundaries |
---|
62 | ! |
---|
63 | ! 85 2007-05-11 09:35:14Z raasch |
---|
64 | ! Division through dt_3d replaced by multiplication of the inverse. |
---|
65 | ! For performance optimisation, this is done in the loop calculating the |
---|
66 | ! divergence instead of using a seperate loop. |
---|
67 | ! |
---|
68 | ! 75 2007-03-22 09:54:05Z raasch |
---|
69 | ! Volume flow control for non-cyclic boundary conditions added (currently only |
---|
70 | ! for the north boundary!!), 2nd+3rd argument removed from exchange horiz, |
---|
71 | ! mean vertical velocity is removed in case of Neumann boundary conditions |
---|
72 | ! both at the bottom and the top |
---|
73 | ! |
---|
74 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
75 | ! |
---|
76 | ! Revision 1.25 2006/04/26 13:26:12 raasch |
---|
77 | ! OpenMP optimization (+localsum, threadsum) |
---|
78 | ! |
---|
79 | ! Revision 1.1 1997/07/24 11:24:44 raasch |
---|
80 | ! Initial revision |
---|
81 | ! |
---|
82 | ! |
---|
83 | ! Description: |
---|
84 | ! ------------ |
---|
85 | ! Compute the divergence of the provisional velocity field. Solve the Poisson |
---|
86 | ! equation for the perturbation pressure. Compute the final velocities using |
---|
87 | ! this perturbation pressure. Compute the remaining divergence. |
---|
88 | !------------------------------------------------------------------------------! |
---|
89 | |
---|
90 | USE arrays_3d |
---|
91 | USE constants |
---|
92 | USE control_parameters |
---|
93 | USE cpulog |
---|
94 | USE grid_variables |
---|
95 | USE indices |
---|
96 | USE interfaces |
---|
97 | USE pegrid |
---|
98 | USE poisfft_mod |
---|
99 | USE poisfft_hybrid_mod |
---|
100 | USE statistics |
---|
101 | |
---|
102 | IMPLICIT NONE |
---|
103 | |
---|
104 | INTEGER :: i, j, k, sr |
---|
105 | |
---|
106 | REAL :: ddt_3d, localsum, threadsum, d_weight_pres |
---|
107 | |
---|
108 | REAL, DIMENSION(1:2) :: volume_flow_l, volume_flow_offset |
---|
109 | REAL, DIMENSION(1:nzt) :: w_l, w_l_l |
---|
110 | |
---|
111 | |
---|
112 | CALL cpu_log( log_point(8), 'pres', 'start' ) |
---|
113 | |
---|
114 | |
---|
115 | ddt_3d = 1.0 / dt_3d |
---|
116 | d_weight_pres = 1.0 / weight_pres(intermediate_timestep_count) |
---|
117 | |
---|
118 | ! |
---|
119 | !-- Multigrid method expects array d to have one ghost layer. |
---|
120 | !-- |
---|
121 | IF ( psolver == 'multigrid' ) THEN |
---|
122 | |
---|
123 | DEALLOCATE( d ) |
---|
124 | ALLOCATE( d(nzb:nzt+1,nys-1:nyn+1,nxl-1:nxr+1) ) |
---|
125 | |
---|
126 | ! |
---|
127 | !-- Since p is later used to hold the weighted average of the substeps, it |
---|
128 | !-- cannot be used in the iterative solver. Therefore, its initial value is |
---|
129 | !-- stored on p_loc, which is then iteratively advanced in every substep. |
---|
130 | IF ( intermediate_timestep_count == 1 ) THEN |
---|
131 | DO i = nxl-1, nxr+1 |
---|
132 | DO j = nys-1, nyn+1 |
---|
133 | DO k = nzb, nzt+1 |
---|
134 | p_loc(k,j,i) = p(k,j,i) |
---|
135 | ENDDO |
---|
136 | ENDDO |
---|
137 | ENDDO |
---|
138 | ENDIF |
---|
139 | |
---|
140 | ELSEIF ( psolver == 'sor' .AND. intermediate_timestep_count == 1 ) THEN |
---|
141 | |
---|
142 | ! |
---|
143 | !-- Since p is later used to hold the weighted average of the substeps, it |
---|
144 | !-- cannot be used in the iterative solver. Therefore, its initial value is |
---|
145 | !-- stored on p_loc, which is then iteratively advanced in every substep. |
---|
146 | p_loc = p |
---|
147 | |
---|
148 | ENDIF |
---|
149 | |
---|
150 | ! |
---|
151 | !-- Conserve the volume flow at the outflow in case of non-cyclic lateral |
---|
152 | !-- boundary conditions |
---|
153 | !-- WARNING: so far, this conservation does not work at the left/south |
---|
154 | !-- boundary if the topography at the inflow differs from that at the |
---|
155 | !-- outflow! For this case, volume_flow_area needs adjustment! |
---|
156 | ! |
---|
157 | !-- Left/right |
---|
158 | IF ( conserve_volume_flow .AND. ( outflow_l .OR. outflow_r ) ) THEN |
---|
159 | |
---|
160 | volume_flow(1) = 0.0 |
---|
161 | volume_flow_l(1) = 0.0 |
---|
162 | |
---|
163 | IF ( outflow_l ) THEN |
---|
164 | i = 0 |
---|
165 | ELSEIF ( outflow_r ) THEN |
---|
166 | i = nx+1 |
---|
167 | ENDIF |
---|
168 | |
---|
169 | DO j = nys, nyn |
---|
170 | ! |
---|
171 | !-- Sum up the volume flow through the south/north boundary |
---|
172 | DO k = nzb_2d(j,i)+1, nzt |
---|
173 | volume_flow_l(1) = volume_flow_l(1) + u(k,j,i) * dzw(k) |
---|
174 | ENDDO |
---|
175 | ENDDO |
---|
176 | |
---|
177 | #if defined( __parallel ) |
---|
178 | IF ( collective_wait ) CALL MPI_BARRIER( comm1dy, ierr ) |
---|
179 | CALL MPI_ALLREDUCE( volume_flow_l(1), volume_flow(1), 1, MPI_REAL, & |
---|
180 | MPI_SUM, comm1dy, ierr ) |
---|
181 | #else |
---|
182 | volume_flow = volume_flow_l |
---|
183 | #endif |
---|
184 | volume_flow_offset(1) = ( volume_flow_initial(1) - volume_flow(1) ) & |
---|
185 | / volume_flow_area(1) |
---|
186 | |
---|
187 | DO j = nysg, nyng |
---|
188 | DO k = nzb_2d(j,i)+1, nzt |
---|
189 | u(k,j,i) = u(k,j,i) + volume_flow_offset(1) |
---|
190 | ENDDO |
---|
191 | ENDDO |
---|
192 | |
---|
193 | ENDIF |
---|
194 | |
---|
195 | ! |
---|
196 | !-- South/north |
---|
197 | IF ( conserve_volume_flow .AND. ( outflow_n .OR. outflow_s ) ) THEN |
---|
198 | |
---|
199 | volume_flow(2) = 0.0 |
---|
200 | volume_flow_l(2) = 0.0 |
---|
201 | |
---|
202 | IF ( outflow_s ) THEN |
---|
203 | j = 0 |
---|
204 | ELSEIF ( outflow_n ) THEN |
---|
205 | j = ny+1 |
---|
206 | ENDIF |
---|
207 | |
---|
208 | DO i = nxl, nxr |
---|
209 | ! |
---|
210 | !-- Sum up the volume flow through the south/north boundary |
---|
211 | DO k = nzb_2d(j,i)+1, nzt |
---|
212 | volume_flow_l(2) = volume_flow_l(2) + v(k,j,i) * dzw(k) |
---|
213 | ENDDO |
---|
214 | ENDDO |
---|
215 | |
---|
216 | #if defined( __parallel ) |
---|
217 | IF ( collective_wait ) CALL MPI_BARRIER( comm1dx, ierr ) |
---|
218 | CALL MPI_ALLREDUCE( volume_flow_l(2), volume_flow(2), 1, MPI_REAL, & |
---|
219 | MPI_SUM, comm1dx, ierr ) |
---|
220 | #else |
---|
221 | volume_flow = volume_flow_l |
---|
222 | #endif |
---|
223 | volume_flow_offset(2) = ( volume_flow_initial(2) - volume_flow(2) ) & |
---|
224 | / volume_flow_area(2) |
---|
225 | |
---|
226 | DO i = nxlg, nxrg |
---|
227 | DO k = nzb_v_inner(j,i)+1, nzt |
---|
228 | v(k,j,i) = v(k,j,i) + volume_flow_offset(2) |
---|
229 | ENDDO |
---|
230 | ENDDO |
---|
231 | |
---|
232 | ENDIF |
---|
233 | |
---|
234 | ! |
---|
235 | !-- Remove mean vertical velocity |
---|
236 | IF ( ibc_p_b == 1 .AND. ibc_p_t == 1 ) THEN |
---|
237 | IF ( simulated_time > 0.0 ) THEN ! otherwise nzb_w_inner not yet known |
---|
238 | w_l = 0.0; w_l_l = 0.0 |
---|
239 | DO i = nxl, nxr |
---|
240 | DO j = nys, nyn |
---|
241 | DO k = nzb_w_inner(j,i)+1, nzt |
---|
242 | w_l_l(k) = w_l_l(k) + w(k,j,i) |
---|
243 | ENDDO |
---|
244 | ENDDO |
---|
245 | ENDDO |
---|
246 | #if defined( __parallel ) |
---|
247 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
248 | CALL MPI_ALLREDUCE( w_l_l(1), w_l(1), nzt, MPI_REAL, MPI_SUM, & |
---|
249 | comm2d, ierr ) |
---|
250 | #else |
---|
251 | w_l = w_l_l |
---|
252 | #endif |
---|
253 | DO k = 1, nzt |
---|
254 | w_l(k) = w_l(k) / ngp_2dh_outer(k,0) |
---|
255 | ENDDO |
---|
256 | DO i = nxlg, nxrg |
---|
257 | DO j = nysg, nyng |
---|
258 | DO k = nzb_w_inner(j,i)+1, nzt |
---|
259 | w(k,j,i) = w(k,j,i) - w_l(k) |
---|
260 | ENDDO |
---|
261 | ENDDO |
---|
262 | ENDDO |
---|
263 | ENDIF |
---|
264 | ENDIF |
---|
265 | |
---|
266 | ! |
---|
267 | !-- Compute the divergence of the provisional velocity field. |
---|
268 | CALL cpu_log( log_point_s(1), 'divergence', 'start' ) |
---|
269 | |
---|
270 | IF ( psolver == 'multigrid' ) THEN |
---|
271 | !$OMP PARALLEL DO SCHEDULE( STATIC ) |
---|
272 | DO i = nxl-1, nxr+1 |
---|
273 | DO j = nys-1, nyn+1 |
---|
274 | DO k = nzb, nzt+1 |
---|
275 | d(k,j,i) = 0.0 |
---|
276 | ENDDO |
---|
277 | ENDDO |
---|
278 | ENDDO |
---|
279 | ELSE |
---|
280 | !$OMP PARALLEL DO SCHEDULE( STATIC ) |
---|
281 | DO i = nxl, nxra |
---|
282 | DO j = nys, nyna |
---|
283 | DO k = nzb+1, nzta |
---|
284 | d(k,j,i) = 0.0 |
---|
285 | ENDDO |
---|
286 | ENDDO |
---|
287 | ENDDO |
---|
288 | ENDIF |
---|
289 | |
---|
290 | localsum = 0.0 |
---|
291 | threadsum = 0.0 |
---|
292 | |
---|
293 | #if defined( __ibm ) |
---|
294 | !$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum) |
---|
295 | !$OMP DO SCHEDULE( STATIC ) |
---|
296 | DO i = nxl, nxr |
---|
297 | DO j = nys, nyn |
---|
298 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
299 | d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * ddx + & |
---|
300 | ( v(k,j+1,i) - v(k,j,i) ) * ddy + & |
---|
301 | ( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) ) * ddt_3d & |
---|
302 | * d_weight_pres |
---|
303 | ENDDO |
---|
304 | ! |
---|
305 | !-- Additional pressure boundary condition at the bottom boundary for |
---|
306 | !-- inhomogeneous Prandtl layer heat fluxes and temperatures, respectively |
---|
307 | !-- dp/dz = -(dtau13/dx + dtau23/dy) + g*pt'/pt0. |
---|
308 | !-- This condition must not be applied at the start of a run, because then |
---|
309 | !-- flow_statistics has not yet been called and thus sums = 0. |
---|
310 | IF ( ibc_p_b == 2 .AND. sums(nzb+1,4) /= 0.0 ) THEN |
---|
311 | k = nzb_s_inner(j,i) |
---|
312 | d(k+1,j,i) = d(k+1,j,i) + ( & |
---|
313 | ( usws(j,i+1) - usws(j,i) ) * ddx & |
---|
314 | + ( vsws(j+1,i) - vsws(j,i) ) * ddy & |
---|
315 | - g * ( pt(k+1,j,i) - sums(k+1,4) ) / & |
---|
316 | sums(k+1,4) & |
---|
317 | ) * ddzw(k+1) * ddt_3d * d_weight_pres |
---|
318 | ENDIF |
---|
319 | |
---|
320 | ! |
---|
321 | !-- Compute possible PE-sum of divergences for flow_statistics |
---|
322 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
323 | threadsum = threadsum + ABS( d(k,j,i) ) |
---|
324 | ENDDO |
---|
325 | |
---|
326 | ENDDO |
---|
327 | ENDDO |
---|
328 | |
---|
329 | localsum = localsum + threadsum * dt_3d * & |
---|
330 | weight_pres(intermediate_timestep_count) |
---|
331 | |
---|
332 | !$OMP END PARALLEL |
---|
333 | #else |
---|
334 | IF ( ibc_p_b == 2 .AND. sums(nzb+1,4) /= 0.0 ) THEN |
---|
335 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
336 | !$OMP DO SCHEDULE( STATIC ) |
---|
337 | DO i = nxl, nxr |
---|
338 | DO j = nys, nyn |
---|
339 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
340 | d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * ddx + & |
---|
341 | ( v(k,j+1,i) - v(k,j,i) ) * ddy + & |
---|
342 | ( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) ) * ddt_3d & |
---|
343 | * d_weight_pres |
---|
344 | ENDDO |
---|
345 | ENDDO |
---|
346 | ! |
---|
347 | !-- Additional pressure boundary condition at the bottom boundary for |
---|
348 | !-- inhomogeneous Prandtl layer heat fluxes and temperatures, respectively |
---|
349 | !-- dp/dz = -(dtau13/dx + dtau23/dy) + g*pt'/pt0. |
---|
350 | !-- This condition must not be applied at the start of a run, because then |
---|
351 | !-- flow_statistics has not yet been called and thus sums = 0. |
---|
352 | DO j = nys, nyn |
---|
353 | k = nzb_s_inner(j,i) |
---|
354 | d(k+1,j,i) = d(k+1,j,i) + ( & |
---|
355 | ( usws(j,i+1) - usws(j,i) ) * ddx & |
---|
356 | + ( vsws(j+1,i) - vsws(j,i) ) * ddy & |
---|
357 | - g * ( pt(k+1,j,i) - sums(k+1,4) ) / & |
---|
358 | sums(k+1,4) & |
---|
359 | ) * ddzw(k+1) * ddt_3d & |
---|
360 | * d_weight_pres |
---|
361 | ENDDO |
---|
362 | ENDDO |
---|
363 | !$OMP END PARALLEL |
---|
364 | |
---|
365 | ELSE |
---|
366 | |
---|
367 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
368 | !$OMP DO SCHEDULE( STATIC ) |
---|
369 | DO i = nxl, nxr |
---|
370 | DO j = nys, nyn |
---|
371 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
372 | d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * ddx + & |
---|
373 | ( v(k,j+1,i) - v(k,j,i) ) * ddy + & |
---|
374 | ( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) ) * ddt_3d & |
---|
375 | * d_weight_pres |
---|
376 | ENDDO |
---|
377 | ENDDO |
---|
378 | ENDDO |
---|
379 | !$OMP END PARALLEL |
---|
380 | |
---|
381 | ENDIF |
---|
382 | |
---|
383 | ! |
---|
384 | !-- Compute possible PE-sum of divergences for flow_statistics |
---|
385 | !$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum) |
---|
386 | !$OMP DO SCHEDULE( STATIC ) |
---|
387 | DO i = nxl, nxr |
---|
388 | DO j = nys, nyn |
---|
389 | DO k = nzb+1, nzt |
---|
390 | threadsum = threadsum + ABS( d(k,j,i) ) |
---|
391 | ENDDO |
---|
392 | ENDDO |
---|
393 | ENDDO |
---|
394 | localsum = localsum + threadsum * dt_3d * & |
---|
395 | weight_pres(intermediate_timestep_count) |
---|
396 | !$OMP END PARALLEL |
---|
397 | #endif |
---|
398 | |
---|
399 | ! |
---|
400 | !-- For completeness, set the divergence sum of all statistic regions to those |
---|
401 | !-- of the total domain |
---|
402 | sums_divold_l(0:statistic_regions) = localsum |
---|
403 | |
---|
404 | CALL cpu_log( log_point_s(1), 'divergence', 'stop' ) |
---|
405 | |
---|
406 | ! |
---|
407 | !-- Compute the pressure perturbation solving the Poisson equation |
---|
408 | IF ( psolver(1:7) == 'poisfft' ) THEN |
---|
409 | |
---|
410 | ! |
---|
411 | !-- Enlarge the size of tend, used as a working array for the transpositions |
---|
412 | IF ( nxra > nxr .OR. nyna > nyn .OR. nza > nz ) THEN |
---|
413 | DEALLOCATE( tend ) |
---|
414 | ALLOCATE( tend(1:nza,nys:nyna,nxl:nxra) ) |
---|
415 | ENDIF |
---|
416 | |
---|
417 | ! |
---|
418 | !-- Solve Poisson equation via FFT and solution of tridiagonal matrices |
---|
419 | IF ( psolver == 'poisfft' ) THEN |
---|
420 | ! |
---|
421 | !-- Solver for 2d-decomposition |
---|
422 | CALL poisfft( d, tend ) |
---|
423 | ELSEIF ( psolver == 'poisfft_hybrid' ) THEN |
---|
424 | ! |
---|
425 | !-- Solver for 1d-decomposition (using MPI and OpenMP). |
---|
426 | !-- The old hybrid-solver is still included here, as long as there |
---|
427 | !-- are some optimization problems in poisfft |
---|
428 | CALL poisfft_hybrid( d ) |
---|
429 | ENDIF |
---|
430 | |
---|
431 | ! |
---|
432 | !-- Resize tend to its normal size |
---|
433 | IF ( nxra > nxr .OR. nyna > nyn .OR. nza > nz ) THEN |
---|
434 | DEALLOCATE( tend ) |
---|
435 | ALLOCATE( tend(nzb:nzt+1,nysg:nyng,nxlg:nxrg) ) |
---|
436 | ENDIF |
---|
437 | |
---|
438 | ! |
---|
439 | !-- Store computed perturbation pressure and set boundary condition in |
---|
440 | !-- z-direction |
---|
441 | !$OMP PARALLEL DO |
---|
442 | DO i = nxl, nxr |
---|
443 | DO j = nys, nyn |
---|
444 | DO k = nzb+1, nzt |
---|
445 | tend(k,j,i) = d(k,j,i) |
---|
446 | ENDDO |
---|
447 | ENDDO |
---|
448 | ENDDO |
---|
449 | |
---|
450 | ! |
---|
451 | !-- Bottom boundary: |
---|
452 | !-- This condition is only required for internal output. The pressure |
---|
453 | !-- gradient (dp(nzb+1)-dp(nzb))/dz is not used anywhere else. |
---|
454 | IF ( ibc_p_b == 1 ) THEN |
---|
455 | ! |
---|
456 | !-- Neumann (dp/dz = 0) |
---|
457 | !$OMP PARALLEL DO |
---|
458 | DO i = nxlg, nxrg |
---|
459 | DO j = nysg, nyng |
---|
460 | tend(nzb_s_inner(j,i),j,i) = tend(nzb_s_inner(j,i)+1,j,i) |
---|
461 | ENDDO |
---|
462 | ENDDO |
---|
463 | |
---|
464 | ELSEIF ( ibc_p_b == 2 ) THEN |
---|
465 | ! |
---|
466 | !-- Neumann condition for inhomogeneous surfaces, |
---|
467 | !-- here currently still in the form of a zero gradient. Actually |
---|
468 | !-- dp/dz = -(dtau13/dx + dtau23/dy) + g*pt'/pt0 would have to be used for |
---|
469 | !-- the computation (cf. above: computation of divergences). |
---|
470 | !$OMP PARALLEL DO |
---|
471 | DO i = nxlg, nxrg |
---|
472 | DO j = nysg, nyng |
---|
473 | tend(nzb_s_inner(j,i),j,i) = tend(nzb_s_inner(j,i)+1,j,i) |
---|
474 | ENDDO |
---|
475 | ENDDO |
---|
476 | |
---|
477 | ELSE |
---|
478 | ! |
---|
479 | !-- Dirichlet |
---|
480 | !$OMP PARALLEL DO |
---|
481 | DO i = nxlg, nxrg |
---|
482 | DO j = nysg, nyng |
---|
483 | tend(nzb_s_inner(j,i),j,i) = 0.0 |
---|
484 | ENDDO |
---|
485 | ENDDO |
---|
486 | |
---|
487 | ENDIF |
---|
488 | |
---|
489 | ! |
---|
490 | !-- Top boundary |
---|
491 | IF ( ibc_p_t == 1 ) THEN |
---|
492 | ! |
---|
493 | !-- Neumann |
---|
494 | !$OMP PARALLEL DO |
---|
495 | DO i = nxlg, nxrg |
---|
496 | DO j = nysg, nyng |
---|
497 | tend(nzt+1,j,i) = tend(nzt,j,i) |
---|
498 | ENDDO |
---|
499 | ENDDO |
---|
500 | |
---|
501 | ELSE |
---|
502 | ! |
---|
503 | !-- Dirichlet |
---|
504 | !$OMP PARALLEL DO |
---|
505 | DO i = nxlg, nxrg |
---|
506 | DO j = nysg, nyng |
---|
507 | tend(nzt+1,j,i) = 0.0 |
---|
508 | ENDDO |
---|
509 | ENDDO |
---|
510 | |
---|
511 | ENDIF |
---|
512 | |
---|
513 | ! |
---|
514 | !-- Exchange boundaries for p |
---|
515 | CALL exchange_horiz( tend, nbgp ) |
---|
516 | |
---|
517 | ELSEIF ( psolver == 'sor' ) THEN |
---|
518 | |
---|
519 | ! |
---|
520 | !-- Solve Poisson equation for perturbation pressure using SOR-Red/Black |
---|
521 | !-- scheme |
---|
522 | CALL sor( d, ddzu_pres, ddzw, p_loc ) |
---|
523 | tend = p_loc |
---|
524 | |
---|
525 | ELSEIF ( psolver == 'multigrid' ) THEN |
---|
526 | |
---|
527 | ! |
---|
528 | !-- Solve Poisson equation for perturbation pressure using Multigrid scheme, |
---|
529 | !-- array tend is used to store the residuals, logical exchange_mg is used |
---|
530 | !-- to discern data exchange in multigrid ( 1 ghostpoint ) and normal grid |
---|
531 | !-- ( nbgp ghost points ). |
---|
532 | |
---|
533 | !-- If the number of grid points of the gathered grid, which is collected |
---|
534 | !-- on PE0, is larger than the number of grid points of an PE, than array |
---|
535 | !-- tend will be enlarged. |
---|
536 | IF ( gathered_size > subdomain_size ) THEN |
---|
537 | DEALLOCATE( tend ) |
---|
538 | ALLOCATE( tend(nzb:nzt_mg(mg_switch_to_pe0_level)+1,nys_mg( & |
---|
539 | mg_switch_to_pe0_level)-1:nyn_mg(mg_switch_to_pe0_level)+1,& |
---|
540 | nxl_mg(mg_switch_to_pe0_level)-1:nxr_mg( & |
---|
541 | mg_switch_to_pe0_level)+1) ) |
---|
542 | ENDIF |
---|
543 | |
---|
544 | CALL poismg( tend ) |
---|
545 | |
---|
546 | IF ( gathered_size > subdomain_size ) THEN |
---|
547 | DEALLOCATE( tend ) |
---|
548 | ALLOCATE( tend(nzb:nzt+1,nysg:nyng,nxlg:nxrg) ) |
---|
549 | ENDIF |
---|
550 | |
---|
551 | ! |
---|
552 | !-- Restore perturbation pressure on tend because this array is used |
---|
553 | !-- further below to correct the velocity fields |
---|
554 | DO i = nxl-1, nxr+1 |
---|
555 | DO j = nys-1, nyn+1 |
---|
556 | DO k = nzb, nzt+1 |
---|
557 | tend(k,j,i) = p_loc(k,j,i) |
---|
558 | ENDDO |
---|
559 | ENDDO |
---|
560 | ENDDO |
---|
561 | |
---|
562 | ENDIF |
---|
563 | |
---|
564 | ! |
---|
565 | !-- Store perturbation pressure on array p, used for pressure data output. |
---|
566 | !-- Ghost layers are added in the output routines (except sor-method: see below) |
---|
567 | IF ( intermediate_timestep_count == 1 ) THEN |
---|
568 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
569 | !$OMP DO |
---|
570 | DO i = nxl-1, nxr+1 |
---|
571 | DO j = nys-1, nyn+1 |
---|
572 | DO k = nzb, nzt+1 |
---|
573 | p(k,j,i) = tend(k,j,i) * & |
---|
574 | weight_substep(intermediate_timestep_count) |
---|
575 | ENDDO |
---|
576 | ENDDO |
---|
577 | ENDDO |
---|
578 | !$OMP END PARALLEL |
---|
579 | |
---|
580 | ELSE |
---|
581 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
582 | !$OMP DO |
---|
583 | DO i = nxl-1, nxr+1 |
---|
584 | DO j = nys-1, nyn+1 |
---|
585 | DO k = nzb, nzt+1 |
---|
586 | p(k,j,i) = p(k,j,i) + tend(k,j,i) * & |
---|
587 | weight_substep(intermediate_timestep_count) |
---|
588 | ENDDO |
---|
589 | ENDDO |
---|
590 | ENDDO |
---|
591 | !$OMP END PARALLEL |
---|
592 | |
---|
593 | ENDIF |
---|
594 | |
---|
595 | ! |
---|
596 | !-- SOR-method needs ghost layers for the next timestep |
---|
597 | IF ( psolver == 'sor' ) CALL exchange_horiz( p, nbgp ) |
---|
598 | |
---|
599 | ! |
---|
600 | !-- Correction of the provisional velocities with the current perturbation |
---|
601 | !-- pressure just computed |
---|
602 | IF ( conserve_volume_flow .AND. ( bc_lr_cyc .OR. bc_ns_cyc ) ) THEN |
---|
603 | volume_flow_l(1) = 0.0 |
---|
604 | volume_flow_l(2) = 0.0 |
---|
605 | ENDIF |
---|
606 | |
---|
607 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
608 | !$OMP DO |
---|
609 | DO i = nxl, nxr |
---|
610 | DO j = nys, nyn |
---|
611 | DO k = nzb_w_inner(j,i)+1, nzt |
---|
612 | w(k,j,i) = w(k,j,i) - dt_3d * & |
---|
613 | ( tend(k+1,j,i) - tend(k,j,i) ) * ddzu(k+1) * & |
---|
614 | weight_pres(intermediate_timestep_count) |
---|
615 | ENDDO |
---|
616 | DO k = nzb_u_inner(j,i)+1, nzt |
---|
617 | u(k,j,i) = u(k,j,i) - dt_3d * & |
---|
618 | ( tend(k,j,i) - tend(k,j,i-1) ) * ddx * & |
---|
619 | weight_pres(intermediate_timestep_count) |
---|
620 | ENDDO |
---|
621 | DO k = nzb_v_inner(j,i)+1, nzt |
---|
622 | v(k,j,i) = v(k,j,i) - dt_3d * & |
---|
623 | ( tend(k,j,i) - tend(k,j-1,i) ) * ddy * & |
---|
624 | weight_pres(intermediate_timestep_count) |
---|
625 | ENDDO |
---|
626 | ! |
---|
627 | !-- Sum up the volume flow through the right and north boundary |
---|
628 | IF ( conserve_volume_flow .AND. bc_lr_cyc .AND. bc_ns_cyc .AND. & |
---|
629 | i == nx ) THEN |
---|
630 | !$OMP CRITICAL |
---|
631 | DO k = nzb_2d(j,i) + 1, nzt |
---|
632 | volume_flow_l(1) = volume_flow_l(1) + u(k,j,i) * dzw(k) |
---|
633 | ENDDO |
---|
634 | !$OMP END CRITICAL |
---|
635 | ENDIF |
---|
636 | IF ( conserve_volume_flow .AND. bc_ns_cyc .AND. bc_lr_cyc .AND. & |
---|
637 | j == ny ) THEN |
---|
638 | !$OMP CRITICAL |
---|
639 | DO k = nzb_2d(j,i) + 1, nzt |
---|
640 | volume_flow_l(2) = volume_flow_l(2) + v(k,j,i) * dzw(k) |
---|
641 | ENDDO |
---|
642 | !$OMP END CRITICAL |
---|
643 | ENDIF |
---|
644 | |
---|
645 | ENDDO |
---|
646 | ENDDO |
---|
647 | !$OMP END PARALLEL |
---|
648 | |
---|
649 | ! |
---|
650 | !-- Conserve the volume flow |
---|
651 | IF ( conserve_volume_flow .AND. ( bc_lr_cyc .AND. bc_ns_cyc ) ) THEN |
---|
652 | |
---|
653 | #if defined( __parallel ) |
---|
654 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
655 | CALL MPI_ALLREDUCE( volume_flow_l(1), volume_flow(1), 2, MPI_REAL, & |
---|
656 | MPI_SUM, comm2d, ierr ) |
---|
657 | #else |
---|
658 | volume_flow = volume_flow_l |
---|
659 | #endif |
---|
660 | |
---|
661 | volume_flow_offset = ( volume_flow_initial - volume_flow ) / & |
---|
662 | volume_flow_area |
---|
663 | |
---|
664 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
665 | !$OMP DO |
---|
666 | DO i = nxl, nxr |
---|
667 | DO j = nys, nyn |
---|
668 | DO k = nzb_u_inner(j,i) + 1, nzt |
---|
669 | u(k,j,i) = u(k,j,i) + volume_flow_offset(1) |
---|
670 | ENDDO |
---|
671 | DO k = nzb_v_inner(j,i) + 1, nzt |
---|
672 | v(k,j,i) = v(k,j,i) + volume_flow_offset(2) |
---|
673 | ENDDO |
---|
674 | ENDDO |
---|
675 | ENDDO |
---|
676 | |
---|
677 | !$OMP END PARALLEL |
---|
678 | |
---|
679 | ENDIF |
---|
680 | |
---|
681 | ! |
---|
682 | !-- Exchange of boundaries for the velocities |
---|
683 | CALL exchange_horiz( u, nbgp ) |
---|
684 | CALL exchange_horiz( v, nbgp ) |
---|
685 | CALL exchange_horiz( w, nbgp ) |
---|
686 | |
---|
687 | ! |
---|
688 | !-- Compute the divergence of the corrected velocity field, |
---|
689 | !-- a possible PE-sum is computed in flow_statistics |
---|
690 | CALL cpu_log( log_point_s(1), 'divergence', 'start' ) |
---|
691 | sums_divnew_l = 0.0 |
---|
692 | |
---|
693 | ! |
---|
694 | !-- d must be reset to zero because it can contain nonzero values below the |
---|
695 | !-- topography |
---|
696 | IF ( topography /= 'flat' ) d = 0.0 |
---|
697 | |
---|
698 | localsum = 0.0 |
---|
699 | threadsum = 0.0 |
---|
700 | |
---|
701 | !$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum) |
---|
702 | !$OMP DO SCHEDULE( STATIC ) |
---|
703 | #if defined( __ibm ) |
---|
704 | DO i = nxl, nxr |
---|
705 | DO j = nys, nyn |
---|
706 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
707 | d(k,j,i) = ( u(k,j,i+1) - u(k,j,i) ) * ddx + & |
---|
708 | ( v(k,j+1,i) - v(k,j,i) ) * ddy + & |
---|
709 | ( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) |
---|
710 | ENDDO |
---|
711 | DO k = nzb+1, nzt |
---|
712 | threadsum = threadsum + ABS( d(k,j,i) ) |
---|
713 | ENDDO |
---|
714 | ENDDO |
---|
715 | ENDDO |
---|
716 | #else |
---|
717 | DO i = nxl, nxr |
---|
718 | DO j = nys, nyn |
---|
719 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
720 | d(k,j,i) = ( u(k,j,i+1) - u(k,j,i) ) * ddx + & |
---|
721 | ( v(k,j+1,i) - v(k,j,i) ) * ddy + & |
---|
722 | ( w(k,j,i) - w(k-1,j,i) ) * ddzw(k) |
---|
723 | threadsum = threadsum + ABS( d(k,j,i) ) |
---|
724 | ENDDO |
---|
725 | ENDDO |
---|
726 | ENDDO |
---|
727 | #endif |
---|
728 | |
---|
729 | localsum = localsum + threadsum |
---|
730 | !$OMP END PARALLEL |
---|
731 | |
---|
732 | ! |
---|
733 | !-- For completeness, set the divergence sum of all statistic regions to those |
---|
734 | !-- of the total domain |
---|
735 | sums_divnew_l(0:statistic_regions) = localsum |
---|
736 | |
---|
737 | CALL cpu_log( log_point_s(1), 'divergence', 'stop' ) |
---|
738 | |
---|
739 | CALL cpu_log( log_point(8), 'pres', 'stop' ) |
---|
740 | |
---|
741 | |
---|
742 | |
---|
743 | END SUBROUTINE pres |
---|