script: rewrite the launch.py script in a more complete form : scripts/bench.py

.gitignore

@@ -32,3 +32,7 @@ machine_label.txt
 *.exe
 *.out
 *.app
+
+# generate default directories
+/runs
+/idefix

README.md

@@ -14,6 +14,83 @@ These tests were written for Idefix v1.1
 This directory contains scripts used to evaluate the single-process perfomance (frag_tests) as well as weak scaling tests for the lagrangian particles module.
 These tests were written for Idefix v2.0 with Clément Robert's particle module (unreleased at the time of writing).
 
+### scripts/bench.py
+
+A complete python script to :
+
+1. Build Idefix for the cluster (Sub-command **build**).
+2. Generate the job files for the given cluster (Sub-command **gen**).
+3. Launching the jobs (Sub-command **run**).
+4. Extracting the scalability results and plotting (Sub-command **extract**).
+
+In order to proceed, you need to first **download** and **compile** *Idefix*. You can either do it by hand or use the **build** sub-command of the script :
+
+    ./scripts/bench.py --cluster <CLUSTER> --commit <GIT-COMMIT> build
+
+Then you need to **generate** the job directory and files :
+
+    ./scripts/bench.py --cluster <CLUSTER> --commit <GIT-COMMIT> \
+        --account <ACCOUNT> --max-cores <MAX-CORES> --cores-per-node <CORES-PER-NODE>\
+        --problem-size <PROBLEM-SIZE> \
+        gen
+
+Note that when running on **GPUs**, the `core` options correspond to the number of GPUs and not cores.
+
+Check that the jobs are created the right way, then launch them by calling :
+
+    ./scripts/bench.py --cluster <CLUSTER> --commit <GIT-COMMIT> \
+        --account <ACCOUNT> --max-cores <MAX-CORES> --cores-per-node <CORES-PER-NODE> \
+        --problem-size <PROBLEM-SIZE> \
+        run
+
+You can then finally parse the logs to extract the perf and get some report files as output :
+
+    ./scripts/bench.py --cluster <CLUSTER> --commit <GIT-COMMIT> \
+        --account <ACCOUNT> --max-cores <MAX-CORES> --cores-per-node <CORES-PER-NODE> \
+        --problem-size <PROBLEM-SIZE> \
+        extract
+
+It will produce :
+
+ - `runs/<COMMIT>/<CLUSTER>/<DATE>/<PROBLEM-SIZE>/summary.json`
+ - `runs/<COMMIT>/<CLUSTER>/<DATE>/<PROBLEM-SIZE>/summary.dat`
+ - `runs/<COMMIT>/<CLUSTER>/<DATE>/<PROBLEM-SIZE>/summary.gnuplot`
+ - `runs/<COMMIT>/<CLUSTER>/<DATE>/<PROBLEM-SIZE>/summary.pdf`
+
+For a more concrete example, to use the **H100** of **Kraken**, you can use :
+
+    # compile
+    ./scripts/bench.py --cluster kraken-gpu/h100 --commit master build
+
+    # generate the jobs
+    ./scripts/bench.py --cluster kraken-gpu/h100 --commit master --account <ACCOUNT> \
+        --max-cores 8 --cores-per-node 2 --problem-size 256 \
+        gen
+
+    # submit the jobs to the cluster
+    ./scripts/bench.py --cluster kraken-gpu/h100 --commit master --account <ACCOUNT> \
+        --max-cores 8 --cores-per-node 2 --problem-size 256 \
+        run
+
+    # extract the performance from the logs
+    ./scripts/bench.py --cluster kraken-gpu/h100 --commit master --account <ACCOUNT> \
+        --max-cores 8 --cores-per-node 2 --problem-size 256 \
+        extract
+
+In case you want to build Idefix yourself, use the case in **./OrszagTang3D/setup/** to prepare your build dir and then use the
+`--build-directory` to configure in which directory your **idefix** binary and case files lands.
+
+    ./scripts/bench.py --build-directory ./my_idefix/build/ --cluster kraken-gpu/h100 \
+        --commit master --account <ACCOUNT> --max-cores 8 --cores-per-node 2 \
+        --problem-size 256 \
+        gen
+
+To extract the data from a past date, you can precise the date (following the format `year-month-day`) and force the plotting with gnuplot via `--plot`:
+
+    ./scripts/bench.py --cluster kraken-gpu/h100 --commit master --account <ACCOUNT> \
+        --max-cores 8 --cores-per-node 2 --problem-size 256 \
+        gen --date 2026-06-15 --plot
+
 ### scripts/run-bench
 
 A basic script to facilitate running the benchmark on a specific

clusters/kraken/cpu/mpi/env.sh

@@ -0,0 +1,5 @@
+module purge
+
+module load openmpi/5.0.8/gcc-14-2.0
+
+export IDEFIX_FLAGS="-DIdefix_MPI=ON"

clusters/kraken/cpu/mpi/script.oar

@@ -0,0 +1,18 @@
+#!/bin/bash
+#OAR -l /nodes=@nodes@/core=@core@,walltime=00:10:00
+#OAR --name @name@
+#OAR --project @account@
+#OAR --stdout idefix.out
+#OAR --stderr idefix.err
+
+# nettoyage des modules charges en interactif et herites par defaut
+module purge
+
+# chargement des modules
+module load openmpi/5.0.8/gcc-14-2.0*
+
+# echo des commandes lancees
+set -x
+
+# execution du code
+mpirun -np $((@nodes@ * @core@)) ./idefix

clusters/kraken/gpu/h100/env.sh

@@ -0,0 +1,7 @@
+module purge
+
+module load cuda/12.9.41.patched
+module load openmpi/5.0.8/cuda-12.9-gcc-14.2.0
+module load ucx/1.19.0-cuda
+
+export IDEFIX_FLAGS="-DKokkos_ENABLE_CUDA=ON -DKokkos_ARCH_HOPPER90=ON -DIdefix_MPI=ON"

clusters/kraken/gpu/h100/script.oar

@@ -0,0 +1,27 @@
+#!/bin/bash
+#OAR -l /nodes=@nodes@/gpu=@core@,walltime=00:10:00
+#OAR --name @name@
+#OAR --project @account@
+#OAR --stdout idefix.out
+#OAR --stderr idefix.err
+
+# nettoyage des modules charges en interactif et herites par defaut
+module purge
+
+# chargement des modules
+module load arch/h100
+module load cmake/3.30.1
+module load cuda/12.1.0
+module load openmpi/4.1.5-cuda
+
+# echo des commandes lancees
+set -x
+#export KOKKOS_PROFILE_LIBRARY=~/src/kokkos-tools/kp_nvprof_connector.so
+#which nsys
+#nsys -v
+# execution du code
+#export TMPDIR=$JOBSCRATCH
+#ln -s $JOBSCRATCH /tmp/nvidia
+
+# execution du code
+./idefix

clusters/localhost/cpu/mpi/env.sh

@@ -0,0 +1 @@
+export IDEFIX_FLAGS="-DIdefix_MPI=ON"

clusters/localhost/cpu/mpi/script.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+# echo des commandes lancees
+set -x
+#export KOKKOS_PROFILE_LIBRARY=~/src/kokkos-tools/kp_nvprof_connector.so
+#which nsys
+#nsys -v
+# execution du code
+#export TMPDIR=$JOBSCRATCH
+#ln -s $JOBSCRATCH /tmp/nvidia
+
+# execution du code
+mpirun -np @core@ ./idefix 2>&1 | tee idefix.out 

clusters/localhost/cpu/openmp/env.sh

@@ -0,0 +1 @@
+export IDEFIX_FLAGS="-DKokkos_ENABLE_OPENMP=ON"

clusters/localhost/cpu/openmp/script.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+# echo des commandes lancees
+set -x
+#export KOKKOS_PROFILE_LIBRARY=~/src/kokkos-tools/kp_nvprof_connector.so
+#which nsys
+#nsys -v
+# execution du code
+#export TMPDIR=$JOBSCRATCH
+#ln -s $JOBSCRATCH /tmp/nvidia
+
+# execution du code
+OMP_NUM_THREADS=@core@ OMP_PROC_BIND=spread ./idefix 2>&1 | tee idefix.out