| 1 | !> @file check_for_restart.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 | ! Current revisions: |
|---|
| 20 | ! ----------------- |
|---|
| 21 | ! |
|---|
| 22 | ! |
|---|
| 23 | ! Former revisions: |
|---|
| 24 | ! ----------------- |
|---|
| 25 | ! $Id: check_for_restart.f90 4542 2020-05-19 15:45:12Z moh.hefny $ |
|---|
| 26 | ! file re-formatted to follow the PALM coding standard |
|---|
| 27 | ! |
|---|
| 28 | ! 4360 2020-01-07 11:25:50Z suehring |
|---|
| 29 | ! Corrected "Former revisions" section |
|---|
| 30 | ! |
|---|
| 31 | ! 3655 2019-01-07 16:51:22Z knoop |
|---|
| 32 | ! Error messages revised |
|---|
| 33 | ! |
|---|
| 34 | ! Revision 1.1 1998/03/18 20:06:51 raasch |
|---|
| 35 | ! Initial revision |
|---|
| 36 | ! |
|---|
| 37 | ! |
|---|
| 38 | ! Description: |
|---|
| 39 | ! ------------ |
|---|
| 40 | !> Set stop flag, if restart is neccessary because of expiring cpu-time or if it is forced by user. |
|---|
| 41 | !--------------------------------------------------------------------------------------------------! |
|---|
| 42 | SUBROUTINE check_for_restart |
|---|
| 43 | |
|---|
| 44 | |
|---|
| 45 | USE control_parameters, & |
|---|
| 46 | ONLY: coupling_mode, dt_restart, end_time, message_string, run_description_header, & |
|---|
| 47 | simulated_time, terminate_coupled, terminate_coupled_remote, terminate_run, & |
|---|
| 48 | termination_time_needed, time_restart, time_since_reference_point, write_binary |
|---|
| 49 | |
|---|
| 50 | USE kinds |
|---|
| 51 | |
|---|
| 52 | USE pegrid |
|---|
| 53 | |
|---|
| 54 | USE pmc_interface, & |
|---|
| 55 | ONLY: comm_world_nesting, cpl_id, nested_run |
|---|
| 56 | |
|---|
| 57 | IMPLICIT NONE |
|---|
| 58 | |
|---|
| 59 | INTEGER :: global_communicator !< global communicator to be used here |
|---|
| 60 | |
|---|
| 61 | LOGICAL :: terminate_run_l !< |
|---|
| 62 | LOGICAL :: do_stop_now = .FALSE. !< |
|---|
| 63 | LOGICAL :: do_restart_now = .FALSE. !< |
|---|
| 64 | |
|---|
| 65 | REAL(wp) :: remaining_time !< |
|---|
| 66 | |
|---|
| 67 | |
|---|
| 68 | ! |
|---|
| 69 | !-- Check remaining CPU-time |
|---|
| 70 | CALL local_tremain( remaining_time ) |
|---|
| 71 | |
|---|
| 72 | ! |
|---|
| 73 | !-- If necessary set a flag to stop the model run |
|---|
| 74 | terminate_run_l = .FALSE. |
|---|
| 75 | IF ( remaining_time <= termination_time_needed .AND. write_binary ) THEN |
|---|
| 76 | |
|---|
| 77 | terminate_run_l = .TRUE. |
|---|
| 78 | ENDIF |
|---|
| 79 | |
|---|
| 80 | ! |
|---|
| 81 | !-- Set the global communicator to be used (depends on the mode in which PALM is running) |
|---|
| 82 | IF ( nested_run ) THEN |
|---|
| 83 | global_communicator = comm_world_nesting |
|---|
| 84 | ELSE |
|---|
| 85 | global_communicator = comm2d |
|---|
| 86 | ENDIF |
|---|
| 87 | |
|---|
| 88 | #if defined( __parallel ) |
|---|
| 89 | ! |
|---|
| 90 | !-- Make a logical OR for all processes. Stop the model run if at least one process has reached the |
|---|
| 91 | !-- time limit. |
|---|
| 92 | IF ( collective_wait ) CALL MPI_BARRIER( global_communicator, ierr ) |
|---|
| 93 | CALL MPI_ALLREDUCE( terminate_run_l, terminate_run, 1, MPI_LOGICAL, MPI_LOR, & |
|---|
| 94 | global_communicator, ierr ) |
|---|
| 95 | #else |
|---|
| 96 | terminate_run = terminate_run_l |
|---|
| 97 | #endif |
|---|
| 98 | |
|---|
| 99 | ! |
|---|
| 100 | !-- Output that job will be terminated |
|---|
| 101 | IF ( terminate_run .AND. myid == 0 ) THEN |
|---|
| 102 | WRITE( message_string, * ) 'run will be terminated because it is ', & |
|---|
| 103 | 'running out of job cpu limit & ', & |
|---|
| 104 | 'remaining time: ', remaining_time, ' s &', & |
|---|
| 105 | 'termination time needed:', termination_time_needed, ' s' |
|---|
| 106 | CALL message( 'check_for_restart', 'PA0163', 0, 1, 0, 6, 0 ) |
|---|
| 107 | ENDIF |
|---|
| 108 | |
|---|
| 109 | ! |
|---|
| 110 | !-- In case of coupled runs inform the remote model of the termination and its reason, provided the |
|---|
| 111 | !-- remote model has not already been informed of another termination reason (terminate_coupled > 0) |
|---|
| 112 | !-- before, or vice versa (terminate_coupled_remote > 0). |
|---|
| 113 | IF ( terminate_run .AND. TRIM( coupling_mode ) /= 'uncoupled' .AND. & |
|---|
| 114 | terminate_coupled == 0 .AND. terminate_coupled_remote == 0 ) THEN |
|---|
| 115 | |
|---|
| 116 | terminate_coupled = 3 |
|---|
| 117 | |
|---|
| 118 | #if defined( __parallel ) |
|---|
| 119 | IF ( myid == 0 ) THEN |
|---|
| 120 | CALL MPI_SENDRECV( terminate_coupled, 1, MPI_INTEGER, & |
|---|
| 121 | target_id, 0, & |
|---|
| 122 | terminate_coupled_remote, 1, MPI_INTEGER, & |
|---|
| 123 | target_id, 0, & |
|---|
| 124 | comm_inter, status, ierr ) |
|---|
| 125 | ENDIF |
|---|
| 126 | CALL MPI_BCAST( terminate_coupled_remote, 1, MPI_INTEGER, 0, comm2d, ierr ) |
|---|
| 127 | #endif |
|---|
| 128 | ENDIF |
|---|
| 129 | |
|---|
| 130 | |
|---|
| 131 | ! |
|---|
| 132 | !-- Check if a flag file exists that forces a termination of the model. |
|---|
| 133 | IF ( myid == 0 ) THEN |
|---|
| 134 | INQUIRE(FILE="DO_STOP_NOW", EXIST=do_stop_now) |
|---|
| 135 | INQUIRE(FILE="DO_RESTART_NOW", EXIST=do_restart_now) |
|---|
| 136 | |
|---|
| 137 | IF ( do_stop_now .OR. do_restart_now ) THEN |
|---|
| 138 | |
|---|
| 139 | terminate_run_l = .TRUE. |
|---|
| 140 | |
|---|
| 141 | WRITE( message_string, * ) 'run will be terminated because user ', & |
|---|
| 142 | 'forced a job finalization using a flag', & |
|---|
| 143 | 'file:', & |
|---|
| 144 | '&DO_STOP_NOW: ', do_stop_now, & |
|---|
| 145 | '&DO_RESTART_NOW: ', do_restart_now |
|---|
| 146 | CALL message( 'check_for_restart', 'PA0398', 0, 0, 0, 6, 0 ) |
|---|
| 147 | |
|---|
| 148 | ENDIF |
|---|
| 149 | ENDIF |
|---|
| 150 | |
|---|
| 151 | |
|---|
| 152 | #if defined( __parallel ) |
|---|
| 153 | ! |
|---|
| 154 | !-- Make a logical OR for all processes. Stop the model run if a flag file has been detected above. |
|---|
| 155 | IF ( collective_wait ) CALL MPI_BARRIER( global_communicator, ierr ) |
|---|
| 156 | CALL MPI_ALLREDUCE( terminate_run_l, terminate_run, 1, MPI_LOGICAL, MPI_LOR, & |
|---|
| 157 | global_communicator, ierr ) |
|---|
| 158 | #else |
|---|
| 159 | terminate_run = terminate_run_l |
|---|
| 160 | #endif |
|---|
| 161 | |
|---|
| 162 | ! |
|---|
| 163 | !-- In case of coupled runs inform the remote model of the termination and its reason, provided the |
|---|
| 164 | !-- remote model has not already been informed of another termination reason (terminate_coupled > 0) |
|---|
| 165 | !-- before, or vice versa (terminate_coupled_remote > 0). |
|---|
| 166 | IF ( terminate_run .AND. coupling_mode /= 'uncoupled' .AND. & |
|---|
| 167 | terminate_coupled == 0 .AND. terminate_coupled_remote == 0 ) THEN |
|---|
| 168 | |
|---|
| 169 | terminate_coupled = 6 |
|---|
| 170 | |
|---|
| 171 | #if defined( __parallel ) |
|---|
| 172 | IF ( myid == 0 ) THEN |
|---|
| 173 | CALL MPI_SENDRECV( terminate_coupled, 1, MPI_INTEGER, & |
|---|
| 174 | target_id, 0, & |
|---|
| 175 | terminate_coupled_remote, 1, MPI_INTEGER, & |
|---|
| 176 | target_id, 0, & |
|---|
| 177 | comm_inter, status, ierr ) |
|---|
| 178 | ENDIF |
|---|
| 179 | CALL MPI_BCAST( terminate_coupled_remote, 1, MPI_INTEGER, 0, comm2d, ierr ) |
|---|
| 180 | #endif |
|---|
| 181 | |
|---|
| 182 | ENDIF |
|---|
| 183 | |
|---|
| 184 | ! |
|---|
| 185 | !-- Set the stop flag also, if restart is forced by user settings |
|---|
| 186 | IF ( time_restart /= 9999999.9_wp .AND. time_restart < time_since_reference_point ) THEN |
|---|
| 187 | |
|---|
| 188 | ! |
|---|
| 189 | !-- Restart is not neccessary, if the end time of the run (given by the user) has been reached. |
|---|
| 190 | IF ( simulated_time < end_time ) THEN |
|---|
| 191 | terminate_run = .TRUE. |
|---|
| 192 | ! |
|---|
| 193 | !-- Increment restart time, if forced by user, otherwise set restart time to default (no user |
|---|
| 194 | !-- restart). |
|---|
| 195 | IF ( dt_restart /= 9999999.9_wp ) THEN |
|---|
| 196 | time_restart = time_restart + dt_restart |
|---|
| 197 | ELSE |
|---|
| 198 | time_restart = 9999999.9_wp |
|---|
| 199 | ENDIF |
|---|
| 200 | |
|---|
| 201 | WRITE( message_string, * ) 'run will be terminated due to user ', & |
|---|
| 202 | 'settings of ', & |
|---|
| 203 | 'restart_time / dt_restart, ', & |
|---|
| 204 | 'new restart time is: ', time_restart, ' s' |
|---|
| 205 | CALL message( 'check_for_restart', 'PA0164', 0, 0, 0, 6, 0 ) |
|---|
| 206 | |
|---|
| 207 | ! |
|---|
| 208 | !-- In case of coupled runs inform the remote model of the termination and its reason, |
|---|
| 209 | !-- provided the remote model has not already been informed of another termination reason |
|---|
| 210 | !-- (terminate_coupled > 0) before, or vice versa (terminate_coupled_remote > 0). |
|---|
| 211 | IF ( coupling_mode /= 'uncoupled' .AND. terminate_coupled == 0 & |
|---|
| 212 | .AND. terminate_coupled_remote == 0 ) THEN |
|---|
| 213 | |
|---|
| 214 | IF ( dt_restart /= 9999999.9_wp ) THEN |
|---|
| 215 | terminate_coupled = 4 |
|---|
| 216 | ELSE |
|---|
| 217 | terminate_coupled = 5 |
|---|
| 218 | ENDIF |
|---|
| 219 | #if defined( __parallel ) |
|---|
| 220 | IF ( myid == 0 ) THEN |
|---|
| 221 | CALL MPI_SENDRECV( terminate_coupled, 1, MPI_INTEGER, & |
|---|
| 222 | target_id, 0, & |
|---|
| 223 | terminate_coupled_remote, 1, MPI_INTEGER, & |
|---|
| 224 | target_id, 0, & |
|---|
| 225 | comm_inter, status, ierr ) |
|---|
| 226 | ENDIF |
|---|
| 227 | CALL MPI_BCAST( terminate_coupled_remote, 1, MPI_INTEGER, 0, comm2d, ierr ) |
|---|
| 228 | #endif |
|---|
| 229 | ENDIF |
|---|
| 230 | ELSE |
|---|
| 231 | time_restart = 9999999.9_wp |
|---|
| 232 | ENDIF |
|---|
| 233 | ENDIF |
|---|
| 234 | |
|---|
| 235 | ! |
|---|
| 236 | !-- If the run is stopped, set a flag file which is necessary to initiate the start of a |
|---|
| 237 | !-- continuation run, except if the user forced to stop the run without restart. |
|---|
| 238 | IF ( terminate_run .AND. myid == 0 .AND. cpl_id == 1 .AND. .NOT. do_stop_now) THEN |
|---|
| 239 | |
|---|
| 240 | OPEN ( 90, FILE='CONTINUE_RUN', FORM='FORMATTED' ) |
|---|
| 241 | WRITE ( 90, '(A)' ) TRIM( run_description_header ) |
|---|
| 242 | CLOSE ( 90 ) |
|---|
| 243 | |
|---|
| 244 | ENDIF |
|---|
| 245 | |
|---|
| 246 | |
|---|
| 247 | END SUBROUTINE check_for_restart |
|---|
