Changeset 1523 for trunk

Timestamp:

03/26/20 15:24:04 (4 years ago)

Author:

aclsce

Message:

• JeanZay? : Modified method to launch MPMD code to avoid too big run_file configuration file (SLURM has a number of characters limitation).

File:

• trunk/libIGCM/libIGCM_sys/libIGCM_sys_jeanzay.ksh (modified) (4 diffs)

trunk/libIGCM/libIGCM_sys/libIGCM_sys_jeanzay.ksh

@@ -1288 +1288 @@
     # ...
     # example of run_file :
-    # 0 taskset -c 0-4 ./script_lmdz.x.ksh
-    # 1 taskset -c 5-9 ./script_lmdz.x.ksh
-    #...
-    # 32 taskset -c 0 ./script_opa.xx.ksh
-    # 33 taskset -c 1 ./script_opa.xx.ksh
-    #...
-    # 62 taskset -c 30 ./script_xios.x.ksh
+    # 0-70 ./prog_lmdz.x.sh %o %t
+    # 71-430 ./prog_opa.xx.sh %o %t
+    # 431-431 ./prog_xios.x.sh %o %t
+    # examples of prog_file :
+    # prog_lmdz.x.sh :
+    # (( init = 0 + $1 ))
+    # (( index = init * 10 ))
+    # (( slot = index % 40 ))
+    # taskset -c $slot-$((slot + 10 - 1)) ./script_lmdz.x.ksh
+    # that will become
+    # taskset -c 0-9 ./script_lmdz.x.ksh
+    # ...
+    # with script_lmdz.x.ksh
+    # export OMP_STACKSIZE=3g
+    # export OMP_PLACES=cores
+    # export OMP_NUM_THREADS=10
+    # ./lmdz.x > out_lmdz.x.out.${SLURM_PROCID} 2>out_lmdz.x.err.${SLURM_PROCID}
 
     #  Hosts treatment
@@ -1305 +1315 @@
         rank=0
         current_core=0
+        current_core_mpi=0
 
       for comp in ${config_ListOfComponents[*]} ; do
@@ -1354 +1365 @@
           IGCM_sys_Chmod u+x script_${ExeNameOut}.ksh
 
-          # Complete run_file
+        # Build run_file
+        # Only if the component has an executable
+        if ( [ "X${ExeNameOut}" != X\"\" ] ) ; then
+
+          eval comp_proc_mpi_loc=\${${comp}_PROC_MPI}
+          (( end_core = ${current_core_mpi} + ${comp_proc_mpi_loc} - 1 ))
+          echo "${current_core_mpi}-${end_core} ./prog_${ExeNameOut}.sh %o %t" >> run_file
+          (( current_core_mpi = ${end_core} + 1 ))
+        fi
+
           if [ ${comp_proc_nod_loc} -gt 1 ] ; then
               (( offset_comp_proc_loc =  NB_CORE_PER_NODE / (comp_proc_mpi_loc / comp_proc_nod_loc) ))
@@ -1361 +1381 @@
           fi
 
+          # Build configuration file
+ 
+          echo "#!/bin/sh" > prog_${ExeNameOut}.sh
+          echo "(( init = $current_core + \$1 ))" >> prog_${ExeNameOut}.sh
+          echo "(( index = init * $comp_proc_omp_loc ))" >> prog_${ExeNameOut}.sh
+          echo "(( slot = index % 40 ))" >> prog_${ExeNameOut}.sh
+          echo "echo ${ExeNameOut} taskset -c \$slot"-"\$((slot + $comp_proc_omp_loc - 1))" >> prog_${ExeNameOut}.sh
+          echo "taskset -c \$slot"-"\$((slot + $comp_proc_omp_loc - 1)) ./script_${ExeNameOut}.ksh" >> prog_${ExeNameOut}.sh
+
+          IGCM_sys_Chmod u+x prog_${ExeNameOut}.sh
+
+        # Build hostlist file
+
           for nb_proc_mpi in `seq 0 $(($comp_proc_mpi_loc-1))`; do
              (( index_host = current_core / NB_CORE_PER_NODE ))
              host_value=${listnodes[${index_host}]}
-             (( slot =  current_core % NB_CORE_PER_NODE ))
                echo "$host_value" >> hostlist
-               echo "$rank taskset -c $slot-$((slot+comp_proc_omp_loc-1)) ./script_${ExeNameOut}.ksh" >> run_file
-             (( rank = rank + 1 ))
              (( current_core = current_core + offset_comp_proc_loc ))
           done