[1] | 1 | PROGRAM palm |
---|
| 2 | |
---|
[1036] | 3 | !--------------------------------------------------------------------------------! |
---|
| 4 | ! This file is part of PALM. |
---|
| 5 | ! |
---|
| 6 | ! PALM is free software: you can redistribute it and/or modify it under the terms |
---|
| 7 | ! of the GNU General Public License as published by the Free Software Foundation, |
---|
| 8 | ! either version 3 of the License, or (at your option) any later version. |
---|
| 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 | ! |
---|
| 17 | ! Copyright 1997-2012 Leibniz University Hannover |
---|
| 18 | !--------------------------------------------------------------------------------! |
---|
| 19 | ! |
---|
[484] | 20 | ! Current revisions: |
---|
[1] | 21 | ! ----------------- |
---|
[77] | 22 | ! |
---|
[1017] | 23 | ! |
---|
[77] | 24 | ! Former revisions: |
---|
| 25 | ! ----------------- |
---|
| 26 | ! $Id: palm.f90 1036 2012-10-22 13:43:42Z raasch $ |
---|
| 27 | ! |
---|
[1017] | 28 | ! 1015 2012-09-27 09:23:24Z raasch |
---|
[863] | 29 | ! Version number changed from 3.8 to 3.8a. |
---|
[1017] | 30 | ! OpenACC statements added + code changes required for GPU optimization |
---|
[863] | 31 | ! |
---|
[850] | 32 | ! 849 2012-03-15 10:35:09Z raasch |
---|
| 33 | ! write_particles renamed lpm_write_restart_file |
---|
| 34 | ! |
---|
[760] | 35 | ! 759 2011-09-15 13:58:31Z raasch |
---|
| 36 | ! Splitting of parallel I/O, cpu measurement for write_3d_binary and opening |
---|
| 37 | ! of unit 14 moved to here |
---|
| 38 | ! |
---|
[496] | 39 | ! 495 2010-03-02 00:40:15Z raasch |
---|
| 40 | ! Particle data for restart runs are only written if write_binary=.T.. |
---|
| 41 | ! |
---|
[226] | 42 | ! 215 2008-11-18 09:54:31Z raasch |
---|
| 43 | ! Initialization of coupled runs modified for MPI-1 and moved to external |
---|
| 44 | ! subroutine init_coupling |
---|
| 45 | ! |
---|
[198] | 46 | ! 197 2008-09-16 15:29:03Z raasch |
---|
| 47 | ! Workaround for getting information about the coupling mode |
---|
| 48 | ! |
---|
[110] | 49 | ! 108 2007-08-24 15:10:38Z letzel |
---|
| 50 | ! Get coupling mode from environment variable, change location of debug output |
---|
| 51 | ! |
---|
[77] | 52 | ! 75 2007-03-22 09:54:05Z raasch |
---|
[70] | 53 | ! __vtk directives removed, write_particles is called only in case of particle |
---|
[75] | 54 | ! advection switched on, open unit 9 for debug output, |
---|
| 55 | ! setting of palm version moved from modules to here |
---|
[1] | 56 | ! |
---|
[3] | 57 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
| 58 | ! |
---|
[1] | 59 | ! Revision 1.10 2006/08/04 14:53:12 raasch |
---|
| 60 | ! Distibution of run description header removed, call of header moved behind |
---|
| 61 | ! init_3d_model |
---|
| 62 | ! |
---|
| 63 | ! Revision 1.2 2001/01/25 07:15:06 raasch |
---|
| 64 | ! Program name changed to PALM, module test_variables removed. |
---|
| 65 | ! Initialization of dvrp logging as well as exit of dvrp moved to new |
---|
| 66 | ! subroutines init_dvrp_logging and close_dvrp (file init_dvrp.f90) |
---|
| 67 | ! |
---|
| 68 | ! Revision 1.1 1997/07/24 11:23:35 raasch |
---|
| 69 | ! Initial revision |
---|
| 70 | ! |
---|
| 71 | ! |
---|
| 72 | ! Description: |
---|
| 73 | ! ------------ |
---|
| 74 | ! Large-Eddy Simulation (LES) model for the convective boundary layer, |
---|
| 75 | ! optimized for use on parallel machines (implementation realized using the |
---|
| 76 | ! Message Passing Interface (MPI)). The model can also be run on vector machines |
---|
| 77 | ! (less well optimized) and workstations. Versions for the different types of |
---|
| 78 | ! machines are controlled via cpp-directives. |
---|
| 79 | ! Model runs are only feasible using the ksh-script mrun. |
---|
| 80 | !------------------------------------------------------------------------------! |
---|
| 81 | |
---|
| 82 | |
---|
| 83 | USE arrays_3d |
---|
| 84 | USE constants |
---|
[102] | 85 | USE control_parameters |
---|
[1] | 86 | USE cpulog |
---|
| 87 | USE dvrp_variables |
---|
| 88 | USE grid_variables |
---|
| 89 | USE indices |
---|
| 90 | USE interfaces |
---|
| 91 | USE model_1d |
---|
| 92 | USE particle_attributes |
---|
| 93 | USE pegrid |
---|
| 94 | USE spectrum |
---|
| 95 | USE statistics |
---|
| 96 | |
---|
[1015] | 97 | #if defined( __openacc ) |
---|
| 98 | USE OPENACC |
---|
| 99 | #endif |
---|
| 100 | |
---|
[1] | 101 | IMPLICIT NONE |
---|
| 102 | |
---|
| 103 | ! |
---|
| 104 | !-- Local variables |
---|
| 105 | CHARACTER (LEN=9) :: time_to_string |
---|
| 106 | CHARACTER (LEN=1) :: cdum |
---|
[559] | 107 | INTEGER :: i, run_description_header_i(80) |
---|
[1015] | 108 | #if defined( __openacc ) |
---|
| 109 | REAL, DIMENSION(100) :: acc_dum |
---|
| 110 | #endif |
---|
[1] | 111 | |
---|
[1036] | 112 | version = 'PALM 3.9' |
---|
[75] | 113 | |
---|
[1] | 114 | #if defined( __parallel ) |
---|
| 115 | ! |
---|
| 116 | !-- MPI initialisation. comm2d is preliminary set, because |
---|
| 117 | !-- it will be defined in init_pegrid but is used before in cpu_log. |
---|
| 118 | CALL MPI_INIT( ierr ) |
---|
| 119 | CALL MPI_COMM_SIZE( MPI_COMM_WORLD, numprocs, ierr ) |
---|
[206] | 120 | CALL MPI_COMM_RANK( MPI_COMM_WORLD, myid, ierr ) |
---|
[1] | 121 | comm_palm = MPI_COMM_WORLD |
---|
| 122 | comm2d = MPI_COMM_WORLD |
---|
| 123 | |
---|
| 124 | ! |
---|
[206] | 125 | !-- Initialize PE topology in case of coupled runs |
---|
| 126 | CALL init_coupling |
---|
[102] | 127 | #endif |
---|
| 128 | |
---|
[1015] | 129 | #if defined( __openacc ) |
---|
[102] | 130 | ! |
---|
[1015] | 131 | !-- Get the number of accelerator boards per node and assign the MPI processes |
---|
| 132 | !-- to these boards |
---|
| 133 | num_acc_per_node = ACC_GET_NUM_DEVICES( ACC_DEVICE_NVIDIA ) |
---|
| 134 | acc_rank = MOD( myid, num_acc_per_node ) |
---|
| 135 | CALL ACC_SET_DEVICE_NUM ( acc_rank, ACC_DEVICE_NVIDIA ) |
---|
| 136 | ! |
---|
| 137 | !-- Test output (to be removed later) |
---|
| 138 | WRITE (*,'(A,I4,A,I3,A,I3,A,I3)') '*** Connect MPI-Task ', myid,' to CPU ',& |
---|
| 139 | acc_rank, ' Devices: ', num_acc_per_node,& |
---|
| 140 | ' connected to:', & |
---|
| 141 | ACC_GET_DEVICE_NUM( ACC_DEVICE_NVIDIA ) |
---|
| 142 | #endif |
---|
| 143 | ! |
---|
| 144 | !-- Ensure that OpenACC first attaches the GPU devices by copying a dummy data |
---|
| 145 | !-- region |
---|
| 146 | !$acc data copyin( acc_dum ) |
---|
| 147 | |
---|
| 148 | ! |
---|
[1] | 149 | !-- Initialize measuring of the CPU-time remaining to the run |
---|
| 150 | CALL local_tremain_ini |
---|
| 151 | |
---|
| 152 | ! |
---|
| 153 | !-- Start of total CPU time measuring. |
---|
| 154 | CALL cpu_log( log_point(1), 'total', 'start' ) |
---|
| 155 | CALL cpu_log( log_point(2), 'initialisation', 'start' ) |
---|
| 156 | |
---|
| 157 | ! |
---|
[206] | 158 | !-- Open a file for debug output |
---|
| 159 | WRITE (myid_char,'(''_'',I4.4)') myid |
---|
| 160 | OPEN( 9, FILE='DEBUG'//TRIM( coupling_char )//myid_char, FORM='FORMATTED' ) |
---|
| 161 | |
---|
| 162 | ! |
---|
[1] | 163 | !-- Initialize dvrp logging. Also, one PE maybe split from the global |
---|
| 164 | !-- communicator for doing the dvrp output. In that case, the number of |
---|
| 165 | !-- PEs available for PALM is reduced by one and communicator comm_palm |
---|
| 166 | !-- is changed respectively. |
---|
| 167 | #if defined( __parallel ) |
---|
| 168 | CALL MPI_COMM_RANK( comm_palm, myid, ierr ) |
---|
| 169 | ! |
---|
[102] | 170 | !-- TEST OUTPUT (TO BE REMOVED) |
---|
| 171 | WRITE(9,*) '*** coupling_mode = "', TRIM( coupling_mode ), '"' |
---|
| 172 | CALL LOCAL_FLUSH( 9 ) |
---|
[215] | 173 | IF ( TRIM( coupling_mode ) /= 'uncoupled' ) THEN |
---|
| 174 | PRINT*, '*** PE', myid, ' Global target PE:', target_id, & |
---|
| 175 | TRIM( coupling_mode ) |
---|
| 176 | ENDIF |
---|
[102] | 177 | #endif |
---|
| 178 | |
---|
[108] | 179 | CALL init_dvrp_logging |
---|
| 180 | |
---|
[102] | 181 | ! |
---|
[108] | 182 | !-- Read control parameters from NAMELIST files and read environment-variables |
---|
| 183 | CALL parin |
---|
| 184 | |
---|
| 185 | ! |
---|
| 186 | !-- Determine processor topology and local array indices |
---|
| 187 | CALL init_pegrid |
---|
| 188 | |
---|
| 189 | ! |
---|
[1] | 190 | !-- Generate grid parameters |
---|
| 191 | CALL init_grid |
---|
| 192 | |
---|
| 193 | ! |
---|
| 194 | !-- Check control parameters and deduce further quantities |
---|
| 195 | CALL check_parameters |
---|
| 196 | |
---|
[667] | 197 | |
---|
[1] | 198 | ! |
---|
| 199 | !-- Initialize all necessary variables |
---|
| 200 | CALL init_3d_model |
---|
| 201 | |
---|
| 202 | ! |
---|
| 203 | !-- Output of program header |
---|
| 204 | IF ( myid == 0 ) CALL header |
---|
| 205 | |
---|
| 206 | CALL cpu_log( log_point(2), 'initialisation', 'stop' ) |
---|
| 207 | |
---|
| 208 | ! |
---|
| 209 | !-- Set start time in format hh:mm:ss |
---|
| 210 | simulated_time_chr = time_to_string( simulated_time ) |
---|
| 211 | |
---|
| 212 | ! |
---|
| 213 | !-- If required, output of initial arrays |
---|
| 214 | IF ( do2d_at_begin ) THEN |
---|
| 215 | CALL data_output_2d( 'xy', 0 ) |
---|
| 216 | CALL data_output_2d( 'xz', 0 ) |
---|
| 217 | CALL data_output_2d( 'yz', 0 ) |
---|
| 218 | ENDIF |
---|
| 219 | IF ( do3d_at_begin ) THEN |
---|
| 220 | CALL data_output_3d( 0 ) |
---|
| 221 | ENDIF |
---|
| 222 | |
---|
| 223 | ! |
---|
[1015] | 224 | !-- Declare and initialize variables in the accelerator memory with their |
---|
| 225 | !-- host values |
---|
| 226 | !$acc data copyin( diss, e, e_p, kh, km, pt, pt_p, q, ql, tend, te_m, tpt_m, tu_m, tv_m, tw_m, u, u_p, v, vpt, v_p, w, w_p ) & |
---|
| 227 | !$acc copyin( ddzu, ddzw, dd2zu, l_grid, l_wall, ptdf_x, ptdf_y, pt_init, rdf, rdf_sc, ug, vg, zu, zw ) & |
---|
| 228 | !$acc copyin( hom, qs, qsws, qswst, rif, rif_wall, shf, ts, tswst, us, usws, uswst, vsws, vswst, z0, z0h ) & |
---|
| 229 | !$acc copyin( fxm, fxp, fym, fyp, fwxm, fwxp, fwym, fwyp, nzb_diff_s_inner, nzb_diff_s_outer, nzb_diff_u ) & |
---|
| 230 | !$acc copyin( nzb_diff_v, nzb_s_inner, nzb_s_outer, nzb_u_inner ) & |
---|
| 231 | !$acc copyin( nzb_u_outer, nzb_v_inner, nzb_v_outer, nzb_w_inner ) & |
---|
| 232 | !$acc copyin( nzb_w_outer, wall_heatflux, wall_e_x, wall_e_y, wall_u, wall_v, wall_w_x, wall_w_y, wall_flags_0 ) |
---|
| 233 | ! |
---|
[495] | 234 | !-- Integration of the model equations using timestep-scheme |
---|
[1] | 235 | CALL time_integration |
---|
| 236 | |
---|
| 237 | ! |
---|
[495] | 238 | !-- If required, write binary data for restart runs |
---|
| 239 | IF ( write_binary(1:4) == 'true' ) THEN |
---|
[759] | 240 | |
---|
| 241 | CALL cpu_log( log_point(22), 'write_3d_binary', 'start' ) |
---|
| 242 | |
---|
| 243 | CALL check_open( 14 ) |
---|
| 244 | |
---|
| 245 | DO i = 0, io_blocks-1 |
---|
| 246 | IF ( i == io_group ) THEN |
---|
[1] | 247 | ! |
---|
[759] | 248 | !-- Write flow field data |
---|
| 249 | CALL write_3d_binary |
---|
| 250 | ENDIF |
---|
| 251 | #if defined( __parallel ) |
---|
| 252 | CALL MPI_BARRIER( comm2d, ierr ) |
---|
| 253 | #endif |
---|
| 254 | ENDDO |
---|
| 255 | |
---|
| 256 | CALL cpu_log( log_point(22), 'write_3d_binary', 'stop' ) |
---|
| 257 | |
---|
[495] | 258 | ! |
---|
| 259 | !-- If required, write particle data |
---|
[849] | 260 | IF ( particle_advection ) CALL lpm_write_restart_file |
---|
[495] | 261 | ENDIF |
---|
[1] | 262 | |
---|
| 263 | ! |
---|
| 264 | !-- If required, repeat output of header including the required CPU-time |
---|
| 265 | IF ( myid == 0 ) CALL header |
---|
| 266 | ! |
---|
| 267 | !-- If required, final user-defined actions, and |
---|
| 268 | !-- last actions on the open files and close files. Unit 14 was opened |
---|
| 269 | !-- in write_3d_binary but it is closed here, to allow writing on this |
---|
| 270 | !-- unit in routine user_last_actions. |
---|
| 271 | CALL cpu_log( log_point(4), 'last actions', 'start' ) |
---|
[759] | 272 | DO i = 0, io_blocks-1 |
---|
| 273 | IF ( i == io_group ) THEN |
---|
| 274 | CALL user_last_actions |
---|
| 275 | IF ( write_binary(1:4) == 'true' ) CALL close_file( 14 ) |
---|
| 276 | ENDIF |
---|
| 277 | #if defined( __parallel ) |
---|
| 278 | CALL MPI_BARRIER( comm2d, ierr ) |
---|
| 279 | #endif |
---|
| 280 | ENDDO |
---|
[1] | 281 | CALL close_file( 0 ) |
---|
| 282 | CALL close_dvrp |
---|
| 283 | CALL cpu_log( log_point(4), 'last actions', 'stop' ) |
---|
| 284 | |
---|
[102] | 285 | #if defined( __mpi2 ) |
---|
[1] | 286 | ! |
---|
[206] | 287 | !-- Test exchange via intercommunicator in case of a MPI-2 coupling |
---|
[102] | 288 | IF ( coupling_mode == 'atmosphere_to_ocean' ) THEN |
---|
| 289 | i = 12345 + myid |
---|
| 290 | CALL MPI_SEND( i, 1, MPI_INTEGER, myid, 11, comm_inter, ierr ) |
---|
| 291 | ELSEIF ( coupling_mode == 'ocean_to_atmosphere' ) THEN |
---|
| 292 | CALL MPI_RECV( i, 1, MPI_INTEGER, myid, 11, comm_inter, status, ierr ) |
---|
| 293 | PRINT*, '### myid: ', myid, ' received from atmosphere: i = ', i |
---|
| 294 | ENDIF |
---|
| 295 | #endif |
---|
| 296 | |
---|
| 297 | ! |
---|
[1015] | 298 | !-- Close the OpenACC dummy data region |
---|
| 299 | !$acc end data |
---|
| 300 | !$acc end data |
---|
| 301 | |
---|
| 302 | ! |
---|
[1] | 303 | !-- Take final CPU-time for CPU-time analysis |
---|
| 304 | CALL cpu_log( log_point(1), 'total', 'stop' ) |
---|
| 305 | CALL cpu_statistics |
---|
| 306 | |
---|
| 307 | #if defined( __parallel ) |
---|
| 308 | CALL MPI_FINALIZE( ierr ) |
---|
| 309 | #endif |
---|
| 310 | |
---|
| 311 | END PROGRAM palm |
---|