Compiling and running
Compilation
To compile STREAmS, you need at least a Fortran and a C compiler alongside a working installation of the Message Passing Interface (MPI) library.
In order to compile, first go to code
folder and search for a configuration file in the make-templates folder. The most basic CPU configuration
is makefile.inc.linux64_gnu
. Copy the template file to makefile.inc
.
cd code
cp make-templates/makefile.inc.linux64_gnu makefile.inc
A more complex configuration for NVIDIA GPUs on Marconi100 cluster at CINECA is makefile.inc.m100_cuda
.
In general, the configuration file contains 7 parameters:
EQUATION: solved equation: currently only singleideal (single-species Navier-Stokes) is supported
BACKEND: computing backend: gpu (CUDA Fortran for NVIDIA GPUs) or cpu (traditional CPU) are supported
COMPILE: compiler: nv-cuda (NVIDIA CUDA), nv (NVIDIA CPU), intel (Intel), gnu (GNU), ibmxl (IBM XL), cray-cuda (Cray CUDA)
MODE: compilation mode: opt (optimized, default) or debug (debug options enabled)
PREC: floating point precision: double (double precision, default), single (single precision)
CUDA_VER: CUDA version (meaningful if CUDA is enabled): e.g., 11.0
MPI_CMD: MPI wrapper command: this is an optional parameter to override the MPI wrapper command given by the chosen compiler
An example configuration file is available as makefile.inc.m100_cuda and allows you to easily compile the code on Marconi100 cluster at CINECA. Once you have a configuration file for your machine, you can create a link to that file to use it:
Once the configuration file is ready, it is possible to compile:
make
The executable streams_2.exe should be now in the code
folder.
Running
First prepare an input file for your equation. For singleideal equation, the input file is singleideal.ini. Some examples of input files are provided in the folder examples. For instance, copy the subsonic_channel case to your run folder.
mkdir RUN && cd RUN
cp examples/subsonic_channel/singleideal.ini .
To understand how to write an input file please go to the input section in the documentation.
To run the simulation you can use MPI launcher:
mpirun -np 4 ./streams_2.exe
Depending on the configuration of your machine, you may need to use a queue system to launch the simulation. For instance, on Marconi100 at CINECA, a SLURM submission script to run a simulation using GPUs may be:
#!/bin/bash
#SBATCH --partition=m100_usr_prod
#SBATCH --time=24:00:00
#SBATCH --cpus-per-task 32
#SBATCH --exclusive
#SBATCH --ntasks-per-node=4
#SBATCH --gpus-per-node=4
#SBATCH --ntasks-per-socket 2
#SBATCH --nodes 1
module purge
module load profile/global
module load gnu/8.4.0 hpc-sdk/2022--binary spectrum_mpi/10.4.0--binary
mpirun -n 4 -gpu ./streams_2.exe