From CPU to GPU in 80 days: Work Package 2

Dear Community

With Work Package 2, things get interesting: we will launch an attempt to implement the AcceleratedLattice, a version of the MultiBlockLattice which is GPU ready, with a structure-of-array and AA-pattern implementation. On the side of community involvement, we encourage you

Finally, everything related to this project is summarized on the project page.


1 Like

Work Package 2 is finally complete, after a two-week delay. Different issues had to be overcome, but there we are: the AcceleratedLattice is implemented. It replaces the MultiBlockLattice and can be accelerated either through OpenMP or through parallel algorithms. For the first time in the history of Palabos, it is possible to run hybrid MPI / multi-threaded simulations.

We abandoned for now the plan to implement the AcceleratedLattice with help of the AA-pattern, which became a bit technical in combination with Palabos’ MPI communication approach. For now, the AcceleratedLattice will be implemented through a double-population implementation.

To learn more about the AcceleratedLattice, its principles, how to use it, and the performance measurements, please read the corresponding readme file.