CVC Seminar Wed Aug 31 1:15-2:30pm ACES 4.304

Deukhyun Cha

Parallelizing molecular energy and force calculations using cilk++ and CUDA

Parallelization is an obvious tool for accelerating serial algorithms, but effective results often require a further treatment of the algorithm. Many commodity PCs provide high performance multi-threading environment including multi-core CPUs and programmable GPUs. In this talk, I will explain an idea to parallelize our octree-based approximation method for molecular energetics and force calculations using cilk++ and CUDA with efficient multi-thread programming. I will begin with a brief review of the cilk++ and CUDA programming model. After explaining the octree-based approximation algorithm, I will talk about how to parallelize it and some optimization issues. We will conclude with some experimental results for selected molecular energetics and force computations.