Changeset 2644 for trunk/LMDZ.MARS

Timestamp:

Mar 21, 2022, 5:00:11 PM (3 years ago)

Author:

romain.vande

Message:

LMDZ_MARS:
OMP : update the batch example for Occigen and the README associated
RV

Location:

trunk/LMDZ.MARS

Files:

• README (modified) (4 diffs)
• deftank/occigen/README (modified) (1 diff)
• deftank/occigen/run_month1 (modified) (2 diffs)

trunk/LMDZ.MARS/README

@@ -2466 +2466 @@
 microphysics (just like water clouds). The amplitude is spantCO2, also read from callphys.def
 - The previous logical microphysco2 has been removed.
-- Cloud opacity at 1µm is now computed in the co2cloud.F routine
+- Cloud opacity at 1µm is now computed in the co2cloud.F routine
 - Most of the co2 ice clouds scheme writediagfi are now in co2clouds.F
 - Some cleaning of the CO2 ice clouds routine has been done. Not perfect yet!
@@ -2482 +2482 @@
   if a gravity wave propagates to the mesosphere or not (wheter the wave saturates on its way or not). 
   The sub-grid temperature distribution is cancelled if the saturation index value of the column (between 12 and 80 km) is > 0.1, and is applied (+/- 3K) otherwise. if the keyword satindexco2=.true. in callphys.def (only applies if CLFvaryingCO2=.true. anyway). If set to .false., deactivate this filter and the sub grid T distribution applies everywhere and everytime. See comments in co2clouds.F
-- All that you need to launch a GCM run with co2 clouds is described in co2clouds.F comments. Ehouarn Millour and Anni Määttänen have the files. 
+- All that you need to launch a GCM run with co2 clouds is described in co2clouds.F comments. Ehouarn Millour and Anni Määttänen have the files. 
  
 == 15/11/2017 == EM
@@ -3661 +3661 @@
 Implementation of the IRabs-to-VISext dust scenario conversion, depending on the GCM effective radius :
 - new flag 'reff_driven_IRtoVIS_scenario', false by default, must be set to true to use this dynamic conversion
-  (otherwise, the coefficient takes the constant value of 2.6, eqv to reff=1.5µm, as before)
+  (otherwise, the coefficient takes the constant value of 2.6, eqv to reff=1.5\B5m, as before)
   A specific line is added in deftank/callphys.def.GCM6
 - this flag requires the 'dustiropacity' to be set to 'tes' to match the IR scenario's wavelength.
@@ -3668 +3668 @@
   (only at the first call to callradite during each physical timestep)
 - change read_dust_scenario into module read_dust_scenario_mod
+
+== 21/03/2022 == RV
+OMP : update the batch example for Occigen and the README associated

trunk/LMDZ.MARS/deftank/occigen/README

@@ -3 +3 @@
 a job "run_month2" which runs a simulation for month #2 which then possibly
 launches a job "run_month3" etc.
+
+OMP : in "run_month1" the number denoted A, B and C should be choose according
+to these rules : 
+_ AxB = number of MPI threads should be .LE. to #lat/2
+_ C = number of OMP tasks should be .LE. to #lev/10
+_ BxC compatible to the node specification (ex; here BDW28 so .LT. 28 )
+_ Be carefull to change the line "export OMP_NUM_THREADS=" to the correct number C
+
 The starting and ending months of the simulation can be set in "run_month1"
 One should adapt the script (and underlying "run0" and "run.def.ref") to

trunk/LMDZ.MARS/deftank/occigen/run_month1

@@ -1 +1 @@
 #!/bin/bash
-#SBATCH --nodes=1
-#SBATCH --ntasks-per-node=24
+#SBATCH --nodes=6                          #A
+#SBATCH --ntasks-per-node=4                #B
+#SBATCH --cpus-per-task=5                  #C
 #SBATCH -J clim
 #SBATCH --time=13:00:00
 #SBATCH --output run_month1.%j.out
 ##SBATCH --mem=8GB
-#SBATCH --constraint=[HSW24|BDW28]
+#SBATCH --constraint=[BDW28]
 #SBATCH --exclusive
 
@@ -32 +33 @@
 
 \rm -f  error; touch error
+
+# environment setup for OpenMP:
+export I_MPI_DOMAIN=auto
+export I_MPI_PIN_RESPECT_CPUSET=0
+#Make sure that OMP_NUM_THREADS = cpus-per-task * KMP_HW_SUBSET
+export KMP_HW_SUBSET=1T
+export OMP_NUM_THREADS=5                    #C
+export OMP_STACKSIZE=400M
+export KMP_AFFINITY=granularity=fine,compact,1,0,verbose
 
 case $true_num in