303 組み合わせ最適化を用いたGPU-CPU協調コードチューニングツールgPotの提案
詳細
In developing software on GPU, it is important to find codes including parallelism, and execute these codes on GPU. We propose a system which accelerates general C-code used GPU partially by optimizing computing time circulating searching code regions which should be executed on GPU. We evaluated he system implemented this proposal method by applying a test program which has variant loop-structures. This paper discusses tuning technique for gPot based on quantitative performance derived.
