Dynamic autotuning of adaptive fast multipole methods on hybrid multicore CPU & GPU systems

摘要

We discuss an implementation of adaptive fast multipole methods targetinghybrid multicore CPU- and GPU-systems. From previous experiences with thecomputational profile of our version of the fast multipole algorithm, suitableparts are off-loaded to the GPU, while the remaining parts are threaded andexecuted concurrently by the CPU. The parameters defining the algorithm affectsthe performance and by measuring this effect we are able to dynamically balancethe algorithm towards optimal performance. Our setup uses the dynamic nature ofthe computations and is therefore of general character.