Mini-Aevol : A mini-application based on the Aevol simulator

A reduced version (from a model and implementation point of view) of Aevol.

DO NOT USE IT TO SIMULATE BIOLOGICAL RESULTS ! See http://www.aevol.fr for that !

It must be used only to test HPC optimization of the code (parallel, vector, porting to new architecture...).

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Prerequisites

You will need a unix base system. Sorry for Windows users, maybe try with Cygwin

You will also need to install zlib (and its headers):

• On Debian, Ubuntu, Mint, ... (apt): zlib1g-dev
• On Arch (pacman): zlib
• On Fedora (dnf, rpm): zlib-devel

To use cuda implementation, you will need to install the cuda-toolkit:

• On all distributions: cuda

Compilation

The compilation is straightforward

mkdir build
cd build
cmake ..
make

It will produced the executable micro_aevol_cpu.

If CUDA toolkit is available on your system, you can build the software with GPU support

cmake .. -DUSE_CUDA=on
make

It will produced the executable micro_aevol_gpu.

Running a simulation

A help is given to explain the different parameters when using option -H or --help.

Basically, you must create a directory to store the simulation files (backup/checkpointing and stats files) and then run the simulation

mkdir simulation_example_1
cd simulation_example_1
PATH/TO/micro_aevol_cpu

You can also resume a simulation from a backup/checkpointing file (for example, resuming from the generation 1000):

cd simulation_example_1
PATH/TO/micro_aevol_cpu -r 1000

Model and Implementation

These slides give a short presentation of the model and the purpose of this project can

Authors

• Jonathan Rouzaud-Cornabas - Initial work
• Laurent Turpin

For the authors of Aevol software, see http://www.aevol.fr

License

This project is licensed under the GPLv2 License

Profilers:

In order to profile, we need to use debug mode.

compiler options: cmake PATH -DCMAKE_BUILD_TYPE=Debug (for profiling) cmake PATH -DCMAKE_BUILD_TYPE=RelWithDebug (for profiling and with some optimization) cmake PATH -DCMAKE_BUILD_TYPE=Release (with better perf on CPU, not optimizinig the perf on CPU)

profiling tools to install: CPU: gporf intel vtune

GPU: nsight systems (nvprof)
     nsight compute (needs root)

To find the indexed parts to improve, search for "//! HPC i" where i is the index PS: for each step, please check the magnitude of the loop times