[1682] | 1 | !> @file pres.f90 |
---|
[2000] | 2 | !------------------------------------------------------------------------------! |
---|
[2696] | 3 | ! This file is part of the PALM model system. |
---|
[1036] | 4 | ! |
---|
[2000] | 5 | ! PALM is free software: you can redistribute it and/or modify it under the |
---|
| 6 | ! terms of the GNU General Public License as published by the Free Software |
---|
| 7 | ! Foundation, either version 3 of the License, or (at your option) any later |
---|
| 8 | ! version. |
---|
[1036] | 9 | ! |
---|
| 10 | ! PALM is distributed in the hope that it will be useful, but WITHOUT ANY |
---|
| 11 | ! WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR |
---|
| 12 | ! A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
---|
| 13 | ! |
---|
| 14 | ! You should have received a copy of the GNU General Public License along with |
---|
| 15 | ! PALM. If not, see <http://www.gnu.org/licenses/>. |
---|
| 16 | ! |
---|
[2718] | 17 | ! Copyright 1997-2018 Leibniz Universitaet Hannover |
---|
[2000] | 18 | !------------------------------------------------------------------------------! |
---|
[1036] | 19 | ! |
---|
[484] | 20 | ! Current revisions: |
---|
[1212] | 21 | ! ------------------ |
---|
[1932] | 22 | ! |
---|
[3013] | 23 | ! |
---|
[1919] | 24 | ! Former revisions: |
---|
| 25 | ! ----------------- |
---|
[3016] | 26 | ! $Id: pres.f90 3016 2018-05-09 10:53:37Z Giersch $ |
---|
| 27 | ! Dollar sign added before Id |
---|
| 28 | ! |
---|
| 29 | ! 2696 2017-12-14 17:12:51Z kanani |
---|
[3013] | 30 | ! To avoid jumps while plotting w-profiles w level nzt+1 is set to w level nzt |
---|
| 31 | ! after velocity modifications through the pressure solver were carried out |
---|
| 32 | ! |
---|
| 33 | ! 2696 2017-12-14 17:12:51Z kanani |
---|
[2716] | 34 | ! Corrected "Former revisions" section |
---|
[2696] | 35 | ! |
---|
[2716] | 36 | ! 2696 2017-12-14 17:12:51Z kanani |
---|
| 37 | ! Change in file header (GPL part) |
---|
| 38 | ! poismg_noopt modularized (MS) |
---|
| 39 | ! |
---|
[2696] | 40 | ! 2298 2017-06-29 09:28:18Z raasch |
---|
[2298] | 41 | ! comment changed + some formatting |
---|
| 42 | ! |
---|
| 43 | ! 2233 2017-05-30 18:08:54Z suehring |
---|
[1919] | 44 | ! |
---|
[2233] | 45 | ! 2232 2017-05-30 17:47:52Z suehring |
---|
| 46 | ! Adjustments to new topography and surface concept |
---|
| 47 | ! |
---|
[2119] | 48 | ! 2118 2017-01-17 16:38:49Z raasch |
---|
| 49 | ! OpenACC directives and related code removed |
---|
| 50 | ! |
---|
[2074] | 51 | ! 2073 2016-11-30 14:34:05Z raasch |
---|
| 52 | ! openmp bugfix for calculation of new divergence |
---|
| 53 | ! |
---|
[2038] | 54 | ! 2037 2016-10-26 11:15:40Z knoop |
---|
| 55 | ! Anelastic approximation implemented |
---|
| 56 | ! |
---|
[2001] | 57 | ! 2000 2016-08-20 18:09:15Z knoop |
---|
| 58 | ! Forced header and separation lines into 80 columns |
---|
| 59 | ! |
---|
[1933] | 60 | ! 1932 2016-06-10 12:09:21Z suehring |
---|
| 61 | ! Initial version of purely vertical nesting introduced. |
---|
| 62 | ! |
---|
[1932] | 63 | ! 1931 2016-06-10 12:06:59Z suehring |
---|
| 64 | ! Rename multigrid into multigrid_noopt and multigrid_fast into multigrid |
---|
| 65 | ! |
---|
[1930] | 66 | ! 1929 2016-06-09 16:25:25Z suehring |
---|
| 67 | ! Bugfix: weight_substep for initial call, replace by local variable |
---|
| 68 | ! |
---|
[1919] | 69 | ! 1918 2016-05-27 14:35:57Z raasch |
---|
[1918] | 70 | ! sum of divergence is also calculated when pres is called before the initial |
---|
| 71 | ! first time step, |
---|
| 72 | ! stearing is modified by using intermediate_timestep_count = 0 in order to |
---|
| 73 | ! determine if pres is called before the first initial timestep, |
---|
| 74 | ! bugfix: in case of Neumann conditions for pressure both at bottom and top, |
---|
| 75 | ! mean vertical velocity is also removed for the first time step |
---|
| 76 | ! bugfix for calculating divergences |
---|
[1321] | 77 | ! |
---|
[1909] | 78 | ! 1908 2016-05-25 17:22:32Z suehring |
---|
| 79 | ! New divergence for output into RUN_CONTROL file is calculated only at last |
---|
| 80 | ! Runge-Kutta step |
---|
| 81 | ! |
---|
[1846] | 82 | ! 1845 2016-04-08 08:29:13Z raasch |
---|
| 83 | ! nzb_2d replace by nzb_u|v_inner |
---|
| 84 | ! |
---|
[1800] | 85 | ! 1799 2016-04-05 08:35:55Z gronemeier |
---|
| 86 | ! Bugfix: excluded third dimension from horizontal volume flow calculation |
---|
| 87 | ! |
---|
[1763] | 88 | ! 1762 2016-02-25 12:31:13Z hellstea |
---|
| 89 | ! Introduction of nested domain feature |
---|
| 90 | ! |
---|
[1683] | 91 | ! 1682 2015-10-07 23:56:08Z knoop |
---|
| 92 | ! Code annotations made doxygen readable |
---|
| 93 | ! |
---|
[1576] | 94 | ! 1575 2015-03-27 09:56:27Z raasch |
---|
| 95 | ! poismg_fast + respective module added, adjustments for psolver-queries |
---|
| 96 | ! |
---|
[1343] | 97 | ! 1342 2014-03-26 17:04:47Z kanani |
---|
| 98 | ! REAL constants defined as wp-kind |
---|
| 99 | ! |
---|
[1321] | 100 | ! 1320 2014-03-20 08:40:49Z raasch |
---|
[1320] | 101 | ! ONLY-attribute added to USE-statements, |
---|
| 102 | ! kind-parameters added to all INTEGER and REAL declaration statements, |
---|
| 103 | ! kinds are defined in new module kinds, |
---|
| 104 | ! old module precision_kind is removed, |
---|
| 105 | ! revision history before 2012 removed, |
---|
| 106 | ! comment fields (!:) to be used for variable explanations added to |
---|
| 107 | ! all variable declaration statements |
---|
[708] | 108 | ! |
---|
[1319] | 109 | ! 1318 2014-03-17 13:35:16Z raasch |
---|
| 110 | ! module interfaces removed |
---|
| 111 | ! |
---|
[1307] | 112 | ! 1306 2014-03-13 14:30:59Z raasch |
---|
| 113 | ! second argument removed from routine poisfft |
---|
| 114 | ! |
---|
[1258] | 115 | ! 1257 2013-11-08 15:18:40Z raasch |
---|
| 116 | ! openacc loop and loop vector clauses removed, independent clauses added, |
---|
| 117 | ! end parallel replaced by end parallel loop |
---|
| 118 | ! |
---|
[1222] | 119 | ! 1221 2013-09-10 08:59:13Z raasch |
---|
| 120 | ! openACC porting of reduction operations, loops for calculating d are |
---|
| 121 | ! using the rflags_s_inner multiply flag instead of the nzb_s_inner loop index |
---|
| 122 | ! |
---|
[1213] | 123 | ! 1212 2013-08-15 08:46:27Z raasch |
---|
| 124 | ! call of poisfft_hybrid removed |
---|
| 125 | ! |
---|
[1118] | 126 | ! 1117 2013-03-27 11:15:36Z suehring |
---|
| 127 | ! Bugfix in OpenMP parallelization. |
---|
| 128 | ! |
---|
[1114] | 129 | ! 1113 2013-03-10 02:48:14Z raasch |
---|
| 130 | ! GPU-porting of several loops, some loops rearranged |
---|
| 131 | ! |
---|
[1112] | 132 | ! 1111 2013-03-08 23:54:10Z |
---|
| 133 | ! openACC statements added, |
---|
| 134 | ! ibc_p_b = 2 removed |
---|
| 135 | ! |
---|
[1093] | 136 | ! 1092 2013-02-02 11:24:22Z raasch |
---|
| 137 | ! unused variables removed |
---|
| 138 | ! |
---|
[1037] | 139 | ! 1036 2012-10-22 13:43:42Z raasch |
---|
| 140 | ! code put under GPL (PALM 3.9) |
---|
| 141 | ! |
---|
[1004] | 142 | ! 1003 2012-09-14 14:35:53Z raasch |
---|
| 143 | ! adjustment of array tend for cases with unequal subdomain sizes removed |
---|
| 144 | ! |
---|
[1] | 145 | ! Revision 1.1 1997/07/24 11:24:44 raasch |
---|
| 146 | ! Initial revision |
---|
| 147 | ! |
---|
| 148 | ! |
---|
| 149 | ! Description: |
---|
| 150 | ! ------------ |
---|
[1682] | 151 | !> Compute the divergence of the provisional velocity field. Solve the Poisson |
---|
| 152 | !> equation for the perturbation pressure. Compute the final velocities using |
---|
| 153 | !> this perturbation pressure. Compute the remaining divergence. |
---|
[1] | 154 | !------------------------------------------------------------------------------! |
---|
[1682] | 155 | SUBROUTINE pres |
---|
| 156 | |
---|
[1] | 157 | |
---|
[1320] | 158 | USE arrays_3d, & |
---|
[2037] | 159 | ONLY: d, ddzu, ddzu_pres, ddzw, dzw, p, p_loc, rho_air, rho_air_zw, & |
---|
| 160 | tend, u, v, w |
---|
[1320] | 161 | |
---|
| 162 | USE control_parameters, & |
---|
[3012] | 163 | ONLY: bc_lr_cyc, bc_ns_cyc, conserve_volume_flow, coupling_mode, & |
---|
| 164 | dt_3d, gathered_size, ibc_p_b, ibc_p_t, & |
---|
| 165 | intermediate_timestep_count, intermediate_timestep_count_max, & |
---|
| 166 | mg_switch_to_pe0_level, nest_domain, outflow_l, outflow_n, & |
---|
| 167 | outflow_r, outflow_s, psolver, subdomain_size, topography, & |
---|
| 168 | volume_flow, volume_flow_area, volume_flow_initial |
---|
[1320] | 169 | |
---|
| 170 | USE cpulog, & |
---|
| 171 | ONLY: cpu_log, log_point, log_point_s |
---|
| 172 | |
---|
| 173 | USE grid_variables, & |
---|
| 174 | ONLY: ddx, ddy |
---|
| 175 | |
---|
| 176 | USE indices, & |
---|
| 177 | ONLY: nbgp, ngp_2dh_outer, nx, nxl, nxlg, nxl_mg, nxr, nxrg, nxr_mg, & |
---|
[2232] | 178 | ny, nys, nysg, nys_mg, nyn, nyng, nyn_mg, nzb, nzt, nzt_mg, & |
---|
| 179 | wall_flags_0 |
---|
[1320] | 180 | |
---|
| 181 | USE kinds |
---|
| 182 | |
---|
[1] | 183 | USE pegrid |
---|
[1933] | 184 | |
---|
| 185 | USE pmc_interface, & |
---|
| 186 | ONLY: nesting_mode |
---|
[1] | 187 | |
---|
[1320] | 188 | USE poisfft_mod, & |
---|
| 189 | ONLY: poisfft |
---|
| 190 | |
---|
[1575] | 191 | USE poismg_mod |
---|
| 192 | |
---|
[2696] | 193 | USE poismg_noopt_mod |
---|
| 194 | |
---|
[1320] | 195 | USE statistics, & |
---|
| 196 | ONLY: statistic_regions, sums_divnew_l, sums_divold_l, weight_pres, & |
---|
| 197 | weight_substep |
---|
| 198 | |
---|
[2232] | 199 | USE surface_mod, & |
---|
| 200 | ONLY : bc_h |
---|
| 201 | |
---|
[1] | 202 | IMPLICIT NONE |
---|
| 203 | |
---|
[1682] | 204 | INTEGER(iwp) :: i !< |
---|
| 205 | INTEGER(iwp) :: j !< |
---|
| 206 | INTEGER(iwp) :: k !< |
---|
[2232] | 207 | INTEGER(iwp) :: m !< |
---|
[1] | 208 | |
---|
[1682] | 209 | REAL(wp) :: ddt_3d !< |
---|
[1918] | 210 | REAL(wp) :: d_weight_pres !< |
---|
[1682] | 211 | REAL(wp) :: localsum !< |
---|
| 212 | REAL(wp) :: threadsum !< |
---|
[1918] | 213 | REAL(wp) :: weight_pres_l !< |
---|
[1929] | 214 | REAL(wp) :: weight_substep_l !< |
---|
[1] | 215 | |
---|
[1762] | 216 | REAL(wp), DIMENSION(1:3) :: volume_flow_l !< |
---|
| 217 | REAL(wp), DIMENSION(1:3) :: volume_flow_offset !< |
---|
[1682] | 218 | REAL(wp), DIMENSION(1:nzt) :: w_l !< |
---|
| 219 | REAL(wp), DIMENSION(1:nzt) :: w_l_l !< |
---|
[1] | 220 | |
---|
[1933] | 221 | LOGICAL :: nest_domain_nvn !< |
---|
[1] | 222 | |
---|
[1933] | 223 | |
---|
[1] | 224 | CALL cpu_log( log_point(8), 'pres', 'start' ) |
---|
| 225 | |
---|
[85] | 226 | |
---|
[1918] | 227 | ! |
---|
| 228 | !-- Calculate quantities to be used locally |
---|
[1342] | 229 | ddt_3d = 1.0_wp / dt_3d |
---|
[1918] | 230 | IF ( intermediate_timestep_count == 0 ) THEN |
---|
| 231 | ! |
---|
| 232 | !-- If pres is called before initial time step |
---|
[1929] | 233 | weight_pres_l = 1.0_wp |
---|
| 234 | d_weight_pres = 1.0_wp |
---|
| 235 | weight_substep_l = 1.0_wp |
---|
[1918] | 236 | ELSE |
---|
[1929] | 237 | weight_pres_l = weight_pres(intermediate_timestep_count) |
---|
| 238 | d_weight_pres = 1.0_wp / weight_pres(intermediate_timestep_count) |
---|
| 239 | weight_substep_l = weight_substep(intermediate_timestep_count) |
---|
[1918] | 240 | ENDIF |
---|
[85] | 241 | |
---|
[1] | 242 | ! |
---|
[707] | 243 | !-- Multigrid method expects array d to have one ghost layer. |
---|
| 244 | !-- |
---|
[1575] | 245 | IF ( psolver(1:9) == 'multigrid' ) THEN |
---|
[667] | 246 | |
---|
[1] | 247 | DEALLOCATE( d ) |
---|
[667] | 248 | ALLOCATE( d(nzb:nzt+1,nys-1:nyn+1,nxl-1:nxr+1) ) |
---|
[707] | 249 | |
---|
| 250 | ! |
---|
| 251 | !-- Since p is later used to hold the weighted average of the substeps, it |
---|
| 252 | !-- cannot be used in the iterative solver. Therefore, its initial value is |
---|
| 253 | !-- stored on p_loc, which is then iteratively advanced in every substep. |
---|
[1762] | 254 | IF ( intermediate_timestep_count <= 1 ) THEN |
---|
[707] | 255 | DO i = nxl-1, nxr+1 |
---|
| 256 | DO j = nys-1, nyn+1 |
---|
| 257 | DO k = nzb, nzt+1 |
---|
| 258 | p_loc(k,j,i) = p(k,j,i) |
---|
| 259 | ENDDO |
---|
| 260 | ENDDO |
---|
| 261 | ENDDO |
---|
[667] | 262 | ENDIF |
---|
| 263 | |
---|
[1918] | 264 | ELSEIF ( psolver == 'sor' .AND. intermediate_timestep_count <= 1 ) THEN |
---|
[707] | 265 | |
---|
| 266 | ! |
---|
| 267 | !-- Since p is later used to hold the weighted average of the substeps, it |
---|
| 268 | !-- cannot be used in the iterative solver. Therefore, its initial value is |
---|
| 269 | !-- stored on p_loc, which is then iteratively advanced in every substep. |
---|
| 270 | p_loc = p |
---|
| 271 | |
---|
[1] | 272 | ENDIF |
---|
| 273 | |
---|
| 274 | ! |
---|
[75] | 275 | !-- Conserve the volume flow at the outflow in case of non-cyclic lateral |
---|
| 276 | !-- boundary conditions |
---|
[106] | 277 | !-- WARNING: so far, this conservation does not work at the left/south |
---|
| 278 | !-- boundary if the topography at the inflow differs from that at the |
---|
| 279 | !-- outflow! For this case, volume_flow_area needs adjustment! |
---|
| 280 | ! |
---|
| 281 | !-- Left/right |
---|
[709] | 282 | IF ( conserve_volume_flow .AND. ( outflow_l .OR. outflow_r ) ) THEN |
---|
[680] | 283 | |
---|
[1342] | 284 | volume_flow(1) = 0.0_wp |
---|
| 285 | volume_flow_l(1) = 0.0_wp |
---|
[106] | 286 | |
---|
| 287 | IF ( outflow_l ) THEN |
---|
| 288 | i = 0 |
---|
| 289 | ELSEIF ( outflow_r ) THEN |
---|
| 290 | i = nx+1 |
---|
| 291 | ENDIF |
---|
| 292 | |
---|
| 293 | DO j = nys, nyn |
---|
| 294 | ! |
---|
| 295 | !-- Sum up the volume flow through the south/north boundary |
---|
[2232] | 296 | DO k = nzb+1, nzt |
---|
| 297 | volume_flow_l(1) = volume_flow_l(1) + u(k,j,i) * dzw(k) & |
---|
| 298 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 299 | BTEST( wall_flags_0(k,j,i), 1 ) & |
---|
| 300 | ) |
---|
[106] | 301 | ENDDO |
---|
| 302 | ENDDO |
---|
| 303 | |
---|
| 304 | #if defined( __parallel ) |
---|
[680] | 305 | IF ( collective_wait ) CALL MPI_BARRIER( comm1dy, ierr ) |
---|
[106] | 306 | CALL MPI_ALLREDUCE( volume_flow_l(1), volume_flow(1), 1, MPI_REAL, & |
---|
| 307 | MPI_SUM, comm1dy, ierr ) |
---|
| 308 | #else |
---|
| 309 | volume_flow = volume_flow_l |
---|
| 310 | #endif |
---|
[709] | 311 | volume_flow_offset(1) = ( volume_flow_initial(1) - volume_flow(1) ) & |
---|
[106] | 312 | / volume_flow_area(1) |
---|
| 313 | |
---|
[667] | 314 | DO j = nysg, nyng |
---|
[2232] | 315 | DO k = nzb+1, nzt |
---|
| 316 | u(k,j,i) = u(k,j,i) + volume_flow_offset(1) & |
---|
| 317 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 318 | BTEST( wall_flags_0(k,j,i), 1 ) & |
---|
| 319 | ) |
---|
[106] | 320 | ENDDO |
---|
| 321 | ENDDO |
---|
| 322 | |
---|
| 323 | ENDIF |
---|
| 324 | |
---|
| 325 | ! |
---|
| 326 | !-- South/north |
---|
[709] | 327 | IF ( conserve_volume_flow .AND. ( outflow_n .OR. outflow_s ) ) THEN |
---|
[106] | 328 | |
---|
[1342] | 329 | volume_flow(2) = 0.0_wp |
---|
| 330 | volume_flow_l(2) = 0.0_wp |
---|
[75] | 331 | |
---|
[106] | 332 | IF ( outflow_s ) THEN |
---|
| 333 | j = 0 |
---|
| 334 | ELSEIF ( outflow_n ) THEN |
---|
[75] | 335 | j = ny+1 |
---|
[106] | 336 | ENDIF |
---|
| 337 | |
---|
| 338 | DO i = nxl, nxr |
---|
[75] | 339 | ! |
---|
[106] | 340 | !-- Sum up the volume flow through the south/north boundary |
---|
[2232] | 341 | DO k = nzb+1, nzt |
---|
| 342 | volume_flow_l(2) = volume_flow_l(2) + v(k,j,i) * dzw(k) & |
---|
| 343 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 344 | BTEST( wall_flags_0(k,j,i), 2 ) & |
---|
| 345 | ) |
---|
[75] | 346 | ENDDO |
---|
[106] | 347 | ENDDO |
---|
| 348 | |
---|
[75] | 349 | #if defined( __parallel ) |
---|
[680] | 350 | IF ( collective_wait ) CALL MPI_BARRIER( comm1dx, ierr ) |
---|
[75] | 351 | CALL MPI_ALLREDUCE( volume_flow_l(2), volume_flow(2), 1, MPI_REAL, & |
---|
| 352 | MPI_SUM, comm1dx, ierr ) |
---|
| 353 | #else |
---|
| 354 | volume_flow = volume_flow_l |
---|
| 355 | #endif |
---|
| 356 | volume_flow_offset(2) = ( volume_flow_initial(2) - volume_flow(2) ) & |
---|
[106] | 357 | / volume_flow_area(2) |
---|
[75] | 358 | |
---|
[667] | 359 | DO i = nxlg, nxrg |
---|
[2232] | 360 | DO k = nzb+1, nzt |
---|
| 361 | v(k,j,i) = v(k,j,i) + volume_flow_offset(2) & |
---|
| 362 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 363 | BTEST( wall_flags_0(k,j,i), 2 ) & |
---|
| 364 | ) |
---|
[75] | 365 | ENDDO |
---|
[106] | 366 | ENDDO |
---|
[75] | 367 | |
---|
| 368 | ENDIF |
---|
| 369 | |
---|
[76] | 370 | ! |
---|
[1762] | 371 | !-- Remove mean vertical velocity in case that Neumann conditions are |
---|
[1933] | 372 | !-- used both at bottom and top boundary, and if not a nested domain in a |
---|
| 373 | !-- normal nesting run. In case of vertical nesting, this must be done. |
---|
| 374 | !-- Therefore an auxiliary logical variable nest_domain_nvn is used here, and |
---|
| 375 | !-- nvn stands for non-vertical nesting. |
---|
[1918] | 376 | !-- This cannot be done before the first initial time step because ngp_2dh_outer |
---|
| 377 | !-- is not yet known then. |
---|
[1933] | 378 | nest_domain_nvn = nest_domain |
---|
| 379 | IF ( nest_domain .AND. nesting_mode == 'vertical' ) THEN |
---|
| 380 | nest_domain_nvn = .FALSE. |
---|
| 381 | ENDIF |
---|
| 382 | |
---|
| 383 | IF ( ibc_p_b == 1 .AND. ibc_p_t == 1 .AND. & |
---|
| 384 | .NOT. nest_domain_nvn .AND. intermediate_timestep_count /= 0 ) & |
---|
[1918] | 385 | THEN |
---|
| 386 | w_l = 0.0_wp; w_l_l = 0.0_wp |
---|
| 387 | DO i = nxl, nxr |
---|
| 388 | DO j = nys, nyn |
---|
[2232] | 389 | DO k = nzb+1, nzt |
---|
| 390 | w_l_l(k) = w_l_l(k) + w(k,j,i) & |
---|
| 391 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 392 | BTEST( wall_flags_0(k,j,i), 3 ) & |
---|
| 393 | ) |
---|
[76] | 394 | ENDDO |
---|
| 395 | ENDDO |
---|
[1918] | 396 | ENDDO |
---|
[76] | 397 | #if defined( __parallel ) |
---|
[1918] | 398 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
| 399 | CALL MPI_ALLREDUCE( w_l_l(1), w_l(1), nzt, MPI_REAL, MPI_SUM, & |
---|
| 400 | comm2d, ierr ) |
---|
[76] | 401 | #else |
---|
[1918] | 402 | w_l = w_l_l |
---|
[76] | 403 | #endif |
---|
[1918] | 404 | DO k = 1, nzt |
---|
| 405 | w_l(k) = w_l(k) / ngp_2dh_outer(k,0) |
---|
| 406 | ENDDO |
---|
| 407 | DO i = nxlg, nxrg |
---|
| 408 | DO j = nysg, nyng |
---|
[2232] | 409 | DO k = nzb+1, nzt |
---|
| 410 | w(k,j,i) = w(k,j,i) - w_l(k) & |
---|
| 411 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 412 | BTEST( wall_flags_0(k,j,i), 3 ) & |
---|
| 413 | ) |
---|
[76] | 414 | ENDDO |
---|
| 415 | ENDDO |
---|
[1918] | 416 | ENDDO |
---|
[76] | 417 | ENDIF |
---|
[75] | 418 | |
---|
| 419 | ! |
---|
[1] | 420 | !-- Compute the divergence of the provisional velocity field. |
---|
| 421 | CALL cpu_log( log_point_s(1), 'divergence', 'start' ) |
---|
| 422 | |
---|
[1575] | 423 | IF ( psolver(1:9) == 'multigrid' ) THEN |
---|
[2232] | 424 | !$OMP PARALLEL DO SCHEDULE( STATIC ) PRIVATE (i,j,k) |
---|
[1] | 425 | DO i = nxl-1, nxr+1 |
---|
| 426 | DO j = nys-1, nyn+1 |
---|
| 427 | DO k = nzb, nzt+1 |
---|
[1342] | 428 | d(k,j,i) = 0.0_wp |
---|
[1] | 429 | ENDDO |
---|
| 430 | ENDDO |
---|
| 431 | ENDDO |
---|
| 432 | ELSE |
---|
[2232] | 433 | !$OMP PARALLEL DO SCHEDULE( STATIC ) PRIVATE (i,j,k) |
---|
[1003] | 434 | DO i = nxl, nxr |
---|
| 435 | DO j = nys, nyn |
---|
| 436 | DO k = nzb+1, nzt |
---|
[1342] | 437 | d(k,j,i) = 0.0_wp |
---|
[1] | 438 | ENDDO |
---|
| 439 | ENDDO |
---|
| 440 | ENDDO |
---|
| 441 | ENDIF |
---|
| 442 | |
---|
[1342] | 443 | localsum = 0.0_wp |
---|
| 444 | threadsum = 0.0_wp |
---|
[1] | 445 | |
---|
| 446 | #if defined( __ibm ) |
---|
| 447 | !$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum) |
---|
| 448 | !$OMP DO SCHEDULE( STATIC ) |
---|
| 449 | DO i = nxl, nxr |
---|
| 450 | DO j = nys, nyn |
---|
[2232] | 451 | DO k = nzb+1, nzt |
---|
[2037] | 452 | d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * rho_air(k) * ddx + & |
---|
| 453 | ( v(k,j+1,i) - v(k,j,i) ) * rho_air(k) * ddy + & |
---|
| 454 | ( w(k,j,i) * rho_air_zw(k) - & |
---|
| 455 | w(k-1,j,i) * rho_air_zw(k-1) ) * ddzw(k) & |
---|
[2232] | 456 | ) * ddt_3d * d_weight_pres & |
---|
| 457 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 458 | BTEST( wall_flags_0(k,j,i), 0 ) & |
---|
| 459 | ) |
---|
[1] | 460 | ENDDO |
---|
| 461 | ! |
---|
| 462 | !-- Compute possible PE-sum of divergences for flow_statistics |
---|
[2232] | 463 | DO k = nzb+1, nzt |
---|
| 464 | threadsum = threadsum + ABS( d(k,j,i) ) & |
---|
| 465 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 466 | BTEST( wall_flags_0(k,j,i), 0 ) & |
---|
| 467 | ) |
---|
[1] | 468 | ENDDO |
---|
| 469 | |
---|
| 470 | ENDDO |
---|
| 471 | ENDDO |
---|
| 472 | |
---|
[1918] | 473 | IF ( intermediate_timestep_count == intermediate_timestep_count_max .OR. & |
---|
| 474 | intermediate_timestep_count == 0 ) THEN |
---|
| 475 | localsum = localsum + threadsum * dt_3d * weight_pres_l |
---|
[1908] | 476 | ENDIF |
---|
[1] | 477 | !$OMP END PARALLEL |
---|
| 478 | #else |
---|
| 479 | |
---|
[1111] | 480 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
| 481 | !$OMP DO SCHEDULE( STATIC ) |
---|
| 482 | DO i = nxl, nxr |
---|
| 483 | DO j = nys, nyn |
---|
| 484 | DO k = 1, nzt |
---|
[2037] | 485 | d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * rho_air(k) * ddx + & |
---|
| 486 | ( v(k,j+1,i) - v(k,j,i) ) * rho_air(k) * ddy + & |
---|
| 487 | ( w(k,j,i) * rho_air_zw(k) - & |
---|
| 488 | w(k-1,j,i) * rho_air_zw(k-1) ) * ddzw(k) & |
---|
[2232] | 489 | ) * ddt_3d * d_weight_pres & |
---|
| 490 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 491 | BTEST( wall_flags_0(k,j,i), 0 ) & |
---|
| 492 | ) |
---|
[1] | 493 | ENDDO |
---|
| 494 | ENDDO |
---|
[1111] | 495 | ENDDO |
---|
| 496 | !$OMP END PARALLEL |
---|
[1] | 497 | |
---|
| 498 | ! |
---|
[1908] | 499 | !-- Compute possible PE-sum of divergences for flow_statistics. Carry out |
---|
| 500 | !-- computation only at last Runge-Kutta substep. |
---|
[1918] | 501 | IF ( intermediate_timestep_count == intermediate_timestep_count_max .OR. & |
---|
| 502 | intermediate_timestep_count == 0 ) THEN |
---|
[1908] | 503 | !$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum) |
---|
| 504 | !$OMP DO SCHEDULE( STATIC ) |
---|
| 505 | DO i = nxl, nxr |
---|
| 506 | DO j = nys, nyn |
---|
| 507 | DO k = nzb+1, nzt |
---|
| 508 | threadsum = threadsum + ABS( d(k,j,i) ) |
---|
| 509 | ENDDO |
---|
[1] | 510 | ENDDO |
---|
| 511 | ENDDO |
---|
[1918] | 512 | localsum = localsum + threadsum * dt_3d * weight_pres_l |
---|
[1908] | 513 | !$OMP END PARALLEL |
---|
| 514 | ENDIF |
---|
[1] | 515 | #endif |
---|
| 516 | |
---|
| 517 | ! |
---|
| 518 | !-- For completeness, set the divergence sum of all statistic regions to those |
---|
| 519 | !-- of the total domain |
---|
[1918] | 520 | IF ( intermediate_timestep_count == intermediate_timestep_count_max .OR. & |
---|
| 521 | intermediate_timestep_count == 0 ) THEN |
---|
[1908] | 522 | sums_divold_l(0:statistic_regions) = localsum |
---|
[1918] | 523 | ENDIF |
---|
[1] | 524 | |
---|
| 525 | CALL cpu_log( log_point_s(1), 'divergence', 'stop' ) |
---|
| 526 | |
---|
| 527 | ! |
---|
| 528 | !-- Compute the pressure perturbation solving the Poisson equation |
---|
[1575] | 529 | IF ( psolver == 'poisfft' ) THEN |
---|
[1] | 530 | |
---|
| 531 | ! |
---|
| 532 | !-- Solve Poisson equation via FFT and solution of tridiagonal matrices |
---|
[1575] | 533 | CALL poisfft( d ) |
---|
[1212] | 534 | |
---|
[1] | 535 | ! |
---|
| 536 | !-- Store computed perturbation pressure and set boundary condition in |
---|
| 537 | !-- z-direction |
---|
[2232] | 538 | !$OMP PARALLEL DO PRIVATE (i,j,k) |
---|
[1] | 539 | DO i = nxl, nxr |
---|
| 540 | DO j = nys, nyn |
---|
| 541 | DO k = nzb+1, nzt |
---|
| 542 | tend(k,j,i) = d(k,j,i) |
---|
| 543 | ENDDO |
---|
| 544 | ENDDO |
---|
| 545 | ENDDO |
---|
| 546 | |
---|
| 547 | ! |
---|
| 548 | !-- Bottom boundary: |
---|
| 549 | !-- This condition is only required for internal output. The pressure |
---|
| 550 | !-- gradient (dp(nzb+1)-dp(nzb))/dz is not used anywhere else. |
---|
| 551 | IF ( ibc_p_b == 1 ) THEN |
---|
| 552 | ! |
---|
[2232] | 553 | !-- Neumann (dp/dz = 0). Using surfae data type, first for non-natural |
---|
| 554 | !-- surfaces, then for natural and urban surfaces |
---|
| 555 | !-- Upward facing |
---|
| 556 | !$OMP PARALLEL DO PRIVATE( i, j, k ) |
---|
| 557 | DO m = 1, bc_h(0)%ns |
---|
[2298] | 558 | i = bc_h(0)%i(m) |
---|
| 559 | j = bc_h(0)%j(m) |
---|
| 560 | k = bc_h(0)%k(m) |
---|
[2232] | 561 | tend(k-1,j,i) = tend(k,j,i) |
---|
[1] | 562 | ENDDO |
---|
[2232] | 563 | ! |
---|
| 564 | !-- Downward facing |
---|
| 565 | !$OMP PARALLEL DO PRIVATE( i, j, k ) |
---|
| 566 | DO m = 1, bc_h(1)%ns |
---|
[2298] | 567 | i = bc_h(1)%i(m) |
---|
| 568 | j = bc_h(1)%j(m) |
---|
| 569 | k = bc_h(1)%k(m) |
---|
[2232] | 570 | tend(k+1,j,i) = tend(k,j,i) |
---|
| 571 | ENDDO |
---|
[1] | 572 | |
---|
| 573 | ELSE |
---|
| 574 | ! |
---|
[2232] | 575 | !-- Dirichlet. Using surface data type, first for non-natural |
---|
| 576 | !-- surfaces, then for natural and urban surfaces |
---|
| 577 | !-- Upward facing |
---|
| 578 | !$OMP PARALLEL DO PRIVATE( i, j, k ) |
---|
| 579 | DO m = 1, bc_h(0)%ns |
---|
[2298] | 580 | i = bc_h(0)%i(m) |
---|
| 581 | j = bc_h(0)%j(m) |
---|
| 582 | k = bc_h(0)%k(m) |
---|
[2232] | 583 | tend(k-1,j,i) = 0.0_wp |
---|
[1] | 584 | ENDDO |
---|
[2232] | 585 | ! |
---|
| 586 | !-- Downward facing |
---|
| 587 | !$OMP PARALLEL DO PRIVATE( i, j, k ) |
---|
| 588 | DO m = 1, bc_h(1)%ns |
---|
[2298] | 589 | i = bc_h(1)%i(m) |
---|
| 590 | j = bc_h(1)%j(m) |
---|
| 591 | k = bc_h(1)%k(m) |
---|
[2232] | 592 | tend(k+1,j,i) = 0.0_wp |
---|
| 593 | ENDDO |
---|
[1] | 594 | |
---|
| 595 | ENDIF |
---|
| 596 | |
---|
| 597 | ! |
---|
| 598 | !-- Top boundary |
---|
| 599 | IF ( ibc_p_t == 1 ) THEN |
---|
| 600 | ! |
---|
| 601 | !-- Neumann |
---|
[2232] | 602 | !$OMP PARALLEL DO PRIVATE (i,j,k) |
---|
[667] | 603 | DO i = nxlg, nxrg |
---|
| 604 | DO j = nysg, nyng |
---|
[1] | 605 | tend(nzt+1,j,i) = tend(nzt,j,i) |
---|
| 606 | ENDDO |
---|
| 607 | ENDDO |
---|
| 608 | |
---|
| 609 | ELSE |
---|
| 610 | ! |
---|
| 611 | !-- Dirichlet |
---|
[2232] | 612 | !$OMP PARALLEL DO PRIVATE (i,j,k) |
---|
[667] | 613 | DO i = nxlg, nxrg |
---|
| 614 | DO j = nysg, nyng |
---|
[1342] | 615 | tend(nzt+1,j,i) = 0.0_wp |
---|
[1] | 616 | ENDDO |
---|
| 617 | ENDDO |
---|
| 618 | |
---|
| 619 | ENDIF |
---|
| 620 | |
---|
| 621 | ! |
---|
| 622 | !-- Exchange boundaries for p |
---|
[667] | 623 | CALL exchange_horiz( tend, nbgp ) |
---|
[1] | 624 | |
---|
| 625 | ELSEIF ( psolver == 'sor' ) THEN |
---|
| 626 | |
---|
| 627 | ! |
---|
| 628 | !-- Solve Poisson equation for perturbation pressure using SOR-Red/Black |
---|
| 629 | !-- scheme |
---|
[707] | 630 | CALL sor( d, ddzu_pres, ddzw, p_loc ) |
---|
| 631 | tend = p_loc |
---|
[1] | 632 | |
---|
[1575] | 633 | ELSEIF ( psolver(1:9) == 'multigrid' ) THEN |
---|
[1] | 634 | |
---|
| 635 | ! |
---|
| 636 | !-- Solve Poisson equation for perturbation pressure using Multigrid scheme, |
---|
[2298] | 637 | !-- array tend is used to store the residuals. |
---|
[778] | 638 | |
---|
| 639 | !-- If the number of grid points of the gathered grid, which is collected |
---|
| 640 | !-- on PE0, is larger than the number of grid points of an PE, than array |
---|
| 641 | !-- tend will be enlarged. |
---|
| 642 | IF ( gathered_size > subdomain_size ) THEN |
---|
| 643 | DEALLOCATE( tend ) |
---|
| 644 | ALLOCATE( tend(nzb:nzt_mg(mg_switch_to_pe0_level)+1,nys_mg( & |
---|
| 645 | mg_switch_to_pe0_level)-1:nyn_mg(mg_switch_to_pe0_level)+1,& |
---|
| 646 | nxl_mg(mg_switch_to_pe0_level)-1:nxr_mg( & |
---|
| 647 | mg_switch_to_pe0_level)+1) ) |
---|
| 648 | ENDIF |
---|
| 649 | |
---|
[1575] | 650 | IF ( psolver == 'multigrid' ) THEN |
---|
| 651 | CALL poismg( tend ) |
---|
| 652 | ELSE |
---|
[1931] | 653 | CALL poismg_noopt( tend ) |
---|
[1575] | 654 | ENDIF |
---|
[707] | 655 | |
---|
[778] | 656 | IF ( gathered_size > subdomain_size ) THEN |
---|
| 657 | DEALLOCATE( tend ) |
---|
| 658 | ALLOCATE( tend(nzb:nzt+1,nysg:nyng,nxlg:nxrg) ) |
---|
| 659 | ENDIF |
---|
| 660 | |
---|
[1] | 661 | ! |
---|
| 662 | !-- Restore perturbation pressure on tend because this array is used |
---|
| 663 | !-- further below to correct the velocity fields |
---|
[707] | 664 | DO i = nxl-1, nxr+1 |
---|
| 665 | DO j = nys-1, nyn+1 |
---|
| 666 | DO k = nzb, nzt+1 |
---|
| 667 | tend(k,j,i) = p_loc(k,j,i) |
---|
| 668 | ENDDO |
---|
| 669 | ENDDO |
---|
| 670 | ENDDO |
---|
[667] | 671 | |
---|
[1] | 672 | ENDIF |
---|
| 673 | |
---|
| 674 | ! |
---|
[707] | 675 | !-- Store perturbation pressure on array p, used for pressure data output. |
---|
| 676 | !-- Ghost layers are added in the output routines (except sor-method: see below) |
---|
[1918] | 677 | IF ( intermediate_timestep_count <= 1 ) THEN |
---|
[707] | 678 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
| 679 | !$OMP DO |
---|
| 680 | DO i = nxl-1, nxr+1 |
---|
| 681 | DO j = nys-1, nyn+1 |
---|
| 682 | DO k = nzb, nzt+1 |
---|
| 683 | p(k,j,i) = tend(k,j,i) * & |
---|
[1929] | 684 | weight_substep_l |
---|
[673] | 685 | ENDDO |
---|
[1] | 686 | ENDDO |
---|
[707] | 687 | ENDDO |
---|
| 688 | !$OMP END PARALLEL |
---|
| 689 | |
---|
[1762] | 690 | ELSEIF ( intermediate_timestep_count > 1 ) THEN |
---|
[707] | 691 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
| 692 | !$OMP DO |
---|
| 693 | DO i = nxl-1, nxr+1 |
---|
| 694 | DO j = nys-1, nyn+1 |
---|
| 695 | DO k = nzb, nzt+1 |
---|
| 696 | p(k,j,i) = p(k,j,i) + tend(k,j,i) * & |
---|
[1929] | 697 | weight_substep_l |
---|
[673] | 698 | ENDDO |
---|
| 699 | ENDDO |
---|
[707] | 700 | ENDDO |
---|
| 701 | !$OMP END PARALLEL |
---|
| 702 | |
---|
| 703 | ENDIF |
---|
[673] | 704 | |
---|
[707] | 705 | ! |
---|
| 706 | !-- SOR-method needs ghost layers for the next timestep |
---|
| 707 | IF ( psolver == 'sor' ) CALL exchange_horiz( p, nbgp ) |
---|
[682] | 708 | |
---|
[1] | 709 | ! |
---|
| 710 | !-- Correction of the provisional velocities with the current perturbation |
---|
| 711 | !-- pressure just computed |
---|
[709] | 712 | IF ( conserve_volume_flow .AND. ( bc_lr_cyc .OR. bc_ns_cyc ) ) THEN |
---|
[1342] | 713 | volume_flow_l(1) = 0.0_wp |
---|
| 714 | volume_flow_l(2) = 0.0_wp |
---|
[1] | 715 | ENDIF |
---|
[707] | 716 | |
---|
[1] | 717 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
| 718 | !$OMP DO |
---|
[673] | 719 | DO i = nxl, nxr |
---|
[1] | 720 | DO j = nys, nyn |
---|
[2118] | 721 | |
---|
[2232] | 722 | DO k = nzb+1, nzt |
---|
| 723 | w(k,j,i) = w(k,j,i) - dt_3d * & |
---|
| 724 | ( tend(k+1,j,i) - tend(k,j,i) ) * ddzu(k+1) & |
---|
| 725 | * weight_pres_l & |
---|
| 726 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 727 | BTEST( wall_flags_0(k,j,i), 3 ) & |
---|
| 728 | ) |
---|
[1] | 729 | ENDDO |
---|
[2118] | 730 | |
---|
[2232] | 731 | DO k = nzb+1, nzt |
---|
| 732 | u(k,j,i) = u(k,j,i) - dt_3d * & |
---|
| 733 | ( tend(k,j,i) - tend(k,j,i-1) ) * ddx & |
---|
| 734 | * weight_pres_l & |
---|
| 735 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 736 | BTEST( wall_flags_0(k,j,i), 1 ) & |
---|
| 737 | ) |
---|
[1] | 738 | ENDDO |
---|
[2118] | 739 | |
---|
[2232] | 740 | DO k = nzb+1, nzt |
---|
| 741 | v(k,j,i) = v(k,j,i) - dt_3d * & |
---|
| 742 | ( tend(k,j,i) - tend(k,j-1,i) ) * ddy & |
---|
| 743 | * weight_pres_l & |
---|
| 744 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 745 | BTEST( wall_flags_0(k,j,i), 2 ) & |
---|
| 746 | ) |
---|
[673] | 747 | ENDDO |
---|
[1] | 748 | |
---|
| 749 | ENDDO |
---|
| 750 | ENDDO |
---|
| 751 | !$OMP END PARALLEL |
---|
[1113] | 752 | |
---|
| 753 | ! |
---|
[3012] | 754 | !-- The vertical velocity is not set to zero at nzt + 1 for nested domains |
---|
| 755 | !-- Instead it is set to the values of nzt (see routine vnest_boundary_conds |
---|
| 756 | !-- or pmci_interp_tril_t) BEFORE calling the pressure solver. To avoid jumps |
---|
| 757 | !-- while plotting profiles w at the top has to be set to the values in the |
---|
| 758 | !-- height nzt after above modifications. Hint: w level nzt+1 does not impact |
---|
| 759 | !-- results. |
---|
| 760 | IF (nest_domain .OR. coupling_mode == 'vnested_fine') THEN |
---|
| 761 | w(nzt+1,:,:) = w(nzt,:,:) |
---|
| 762 | ENDIF |
---|
| 763 | |
---|
| 764 | ! |
---|
[1113] | 765 | !-- Sum up the volume flow through the right and north boundary |
---|
[2232] | 766 | IF ( conserve_volume_flow .AND. bc_lr_cyc .AND. bc_ns_cyc .AND. & |
---|
[1113] | 767 | nxr == nx ) THEN |
---|
| 768 | |
---|
| 769 | !$OMP PARALLEL PRIVATE (j,k) |
---|
| 770 | !$OMP DO |
---|
| 771 | DO j = nys, nyn |
---|
| 772 | !$OMP CRITICAL |
---|
[2232] | 773 | DO k = nzb+1, nzt |
---|
| 774 | volume_flow_l(1) = volume_flow_l(1) + u(k,j,nxr) * dzw(k) & |
---|
| 775 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 776 | BTEST( wall_flags_0(k,j,nxr), 1 )& |
---|
| 777 | ) |
---|
[1113] | 778 | ENDDO |
---|
| 779 | !$OMP END CRITICAL |
---|
| 780 | ENDDO |
---|
| 781 | !$OMP END PARALLEL |
---|
| 782 | |
---|
| 783 | ENDIF |
---|
| 784 | |
---|
[2232] | 785 | IF ( conserve_volume_flow .AND. bc_ns_cyc .AND. bc_lr_cyc .AND. & |
---|
[1113] | 786 | nyn == ny ) THEN |
---|
| 787 | |
---|
| 788 | !$OMP PARALLEL PRIVATE (i,k) |
---|
| 789 | !$OMP DO |
---|
| 790 | DO i = nxl, nxr |
---|
| 791 | !$OMP CRITICAL |
---|
[2232] | 792 | DO k = nzb+1, nzt |
---|
| 793 | volume_flow_l(2) = volume_flow_l(2) + v(k,nyn,i) * dzw(k) & |
---|
| 794 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 795 | BTEST( wall_flags_0(k,nyn,i), 2 )& |
---|
| 796 | ) |
---|
[1113] | 797 | ENDDO |
---|
| 798 | !$OMP END CRITICAL |
---|
| 799 | ENDDO |
---|
| 800 | !$OMP END PARALLEL |
---|
| 801 | |
---|
| 802 | ENDIF |
---|
[673] | 803 | |
---|
[1] | 804 | ! |
---|
| 805 | !-- Conserve the volume flow |
---|
[707] | 806 | IF ( conserve_volume_flow .AND. ( bc_lr_cyc .AND. bc_ns_cyc ) ) THEN |
---|
[1] | 807 | |
---|
| 808 | #if defined( __parallel ) |
---|
[622] | 809 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
[1] | 810 | CALL MPI_ALLREDUCE( volume_flow_l(1), volume_flow(1), 2, MPI_REAL, & |
---|
| 811 | MPI_SUM, comm2d, ierr ) |
---|
| 812 | #else |
---|
| 813 | volume_flow = volume_flow_l |
---|
| 814 | #endif |
---|
| 815 | |
---|
[1799] | 816 | volume_flow_offset(1:2) = ( volume_flow_initial(1:2) - volume_flow(1:2) ) / & |
---|
| 817 | volume_flow_area(1:2) |
---|
[1] | 818 | |
---|
| 819 | !$OMP PARALLEL PRIVATE (i,j,k) |
---|
| 820 | !$OMP DO |
---|
| 821 | DO i = nxl, nxr |
---|
| 822 | DO j = nys, nyn |
---|
[2232] | 823 | DO k = nzb+1, nzt |
---|
| 824 | u(k,j,i) = u(k,j,i) + volume_flow_offset(1) & |
---|
| 825 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 826 | BTEST( wall_flags_0(k,j,i), 1 ) & |
---|
| 827 | ) |
---|
[719] | 828 | ENDDO |
---|
[2232] | 829 | DO k = nzb+1, nzt |
---|
| 830 | v(k,j,i) = v(k,j,i) + volume_flow_offset(2) & |
---|
| 831 | * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 832 | BTEST( wall_flags_0(k,j,i), 2 ) & |
---|
| 833 | ) |
---|
[667] | 834 | ENDDO |
---|
[1] | 835 | ENDDO |
---|
| 836 | ENDDO |
---|
[667] | 837 | |
---|
[1] | 838 | !$OMP END PARALLEL |
---|
| 839 | |
---|
| 840 | ENDIF |
---|
| 841 | |
---|
| 842 | ! |
---|
| 843 | !-- Exchange of boundaries for the velocities |
---|
[667] | 844 | CALL exchange_horiz( u, nbgp ) |
---|
| 845 | CALL exchange_horiz( v, nbgp ) |
---|
| 846 | CALL exchange_horiz( w, nbgp ) |
---|
[1] | 847 | |
---|
| 848 | ! |
---|
| 849 | !-- Compute the divergence of the corrected velocity field, |
---|
[1908] | 850 | !-- A possible PE-sum is computed in flow_statistics. Carry out computation |
---|
| 851 | !-- only at last Runge-Kutta step. |
---|
[1918] | 852 | IF ( intermediate_timestep_count == intermediate_timestep_count_max .OR. & |
---|
| 853 | intermediate_timestep_count == 0 ) THEN |
---|
[1908] | 854 | CALL cpu_log( log_point_s(1), 'divergence', 'start' ) |
---|
| 855 | sums_divnew_l = 0.0_wp |
---|
[1] | 856 | |
---|
| 857 | ! |
---|
[1908] | 858 | !-- d must be reset to zero because it can contain nonzero values below the |
---|
| 859 | !-- topography |
---|
| 860 | IF ( topography /= 'flat' ) d = 0.0_wp |
---|
[1] | 861 | |
---|
[1908] | 862 | localsum = 0.0_wp |
---|
| 863 | threadsum = 0.0_wp |
---|
[1] | 864 | |
---|
[1908] | 865 | !$OMP PARALLEL PRIVATE (i,j,k) FIRSTPRIVATE(threadsum) REDUCTION(+:localsum) |
---|
[2073] | 866 | #if defined( __ibm ) |
---|
[1908] | 867 | !$OMP DO SCHEDULE( STATIC ) |
---|
| 868 | DO i = nxl, nxr |
---|
| 869 | DO j = nys, nyn |
---|
[2232] | 870 | DO k = nzb+1, nzt |
---|
| 871 | d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * rho_air(k) * ddx + & |
---|
| 872 | ( v(k,j+1,i) - v(k,j,i) ) * rho_air(k) * ddy + & |
---|
| 873 | ( w(k,j,i) * rho_air_zw(k) - & |
---|
| 874 | w(k-1,j,i) * rho_air_zw(k-1) ) * ddzw(k) & |
---|
| 875 | ) * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 876 | BTEST( wall_flags_0(k,j,i), 0 ) & |
---|
| 877 | ) |
---|
[1908] | 878 | ENDDO |
---|
| 879 | DO k = nzb+1, nzt |
---|
| 880 | threadsum = threadsum + ABS( d(k,j,i) ) |
---|
| 881 | ENDDO |
---|
[1] | 882 | ENDDO |
---|
| 883 | ENDDO |
---|
| 884 | #else |
---|
[2073] | 885 | !$OMP DO SCHEDULE( STATIC ) |
---|
[1908] | 886 | DO i = nxl, nxr |
---|
| 887 | DO j = nys, nyn |
---|
[2232] | 888 | DO k = nzb+1, nzt |
---|
[2037] | 889 | d(k,j,i) = ( ( u(k,j,i+1) - u(k,j,i) ) * rho_air(k) * ddx + & |
---|
| 890 | ( v(k,j+1,i) - v(k,j,i) ) * rho_air(k) * ddy + & |
---|
| 891 | ( w(k,j,i) * rho_air_zw(k) - & |
---|
| 892 | w(k-1,j,i) * rho_air_zw(k-1) ) * ddzw(k) & |
---|
[2232] | 893 | ) * MERGE( 1.0_wp, 0.0_wp, & |
---|
| 894 | BTEST( wall_flags_0(k,j,i), 0 ) & |
---|
| 895 | ) |
---|
[1908] | 896 | ENDDO |
---|
[1113] | 897 | ENDDO |
---|
| 898 | ENDDO |
---|
| 899 | ! |
---|
[1908] | 900 | !-- Compute possible PE-sum of divergences for flow_statistics |
---|
[2073] | 901 | !$OMP DO SCHEDULE( STATIC ) |
---|
[1908] | 902 | DO i = nxl, nxr |
---|
| 903 | DO j = nys, nyn |
---|
| 904 | DO k = nzb+1, nzt |
---|
| 905 | threadsum = threadsum + ABS( d(k,j,i) ) |
---|
| 906 | ENDDO |
---|
[1] | 907 | ENDDO |
---|
| 908 | ENDDO |
---|
| 909 | #endif |
---|
[667] | 910 | |
---|
[1908] | 911 | localsum = localsum + threadsum |
---|
| 912 | !$OMP END PARALLEL |
---|
[1] | 913 | |
---|
| 914 | ! |
---|
[1908] | 915 | !-- For completeness, set the divergence sum of all statistic regions to those |
---|
| 916 | !-- of the total domain |
---|
| 917 | sums_divnew_l(0:statistic_regions) = localsum |
---|
[1] | 918 | |
---|
[1908] | 919 | CALL cpu_log( log_point_s(1), 'divergence', 'stop' ) |
---|
[1] | 920 | |
---|
[1908] | 921 | ENDIF |
---|
| 922 | |
---|
[1] | 923 | CALL cpu_log( log_point(8), 'pres', 'stop' ) |
---|
| 924 | |
---|
| 925 | |
---|
| 926 | END SUBROUTINE pres |
---|