high performance computing on graphics processing units: hgpu.org

When implementing a function mapping on the contemporary GPU, several contradictory performance factors affecting distribution of computation into GPU kernels have to be balanced. A decomposition-fusion scheme suggest to decompose computational problem to be solved by several simple functions implemented as standalone kernels and some of these functions later fuse into more complex kernels to improve memory locality. In this paper, a prototype of source-to-source compiler automating the fusion phase is presented and the impact of fusions generated by the compiler as well as compiler efficiency is experimentally evaluated.