The code needs OCCA to run (https://github.com/libocca/occa )
Install stable version of OCCA first.
git clone https://github.com/libocca/occa.git -b 0.2
Enter occa directory.
Type make.
Set environmental variables
export LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:${PWD}/lib"
PATH+=":${PWD}/bin"
export OCCA_DIR=/path/to/folder/occa/
You can also add the commands to your .bashrc
export OCCA_DIR=/path/to/folder/occa/
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$OCCA_DIR/lib
To run the code in the repo:
cd BK10
make
./BK10 512 10
[to simulate mass-matrix-vector multiplication on a mesh with 512 elements and polynomial degree 10 using 12^3 quadrature nodes]
cd BK30
make
./BPK0 512 10
[to simulate stiffness-matrix-vector multiplication on a mesh with 512 elements and polynomial degree 10 using 12^3 quadrature nodes]
cd BK50
make
./BK50 512 10
[to simulate stiffness-matrix-vector multiplication on a mesh with 512 elements and polynomial degree 10 using 11^3 quadrature nodes]
The results vary depending on the hardware. For example, on an NVIDIA Tesla P100 and V100 we expect the results to be close to the roofline based on device-to-device data copy but on Fermi and Kepler class cards the performance is far from the peak. Examples: Results on NVIDIA K40c and Results on NVIDIA V100
Email [email protected]