Refactoring and automated performance tuning of computational chemistry application codes

Computational chemistry codes such as GAMESS and MPQC have been under development for several years and are constantly evolving to include new science and adapt to new high performance computing (HPC) systems. Our work with these codes has given rise to two needs. One is to refactor the codes so that it is easier to optimize them. After profiling has identified performance critical regions, refactoring to outline those regions into separate routines facilitates performance tuning and porting to complex heterogeneous HPC architectures. The second need is for automated performance tuning. Because of the large number of both fine-grained and coarse-grained parameters for tuning performance on complex hierarchical and hybrid architectures, the search space for an optimal set of parameters becomes very large. This paper describes initial results on using refactoring tools to restructure MPQC and GAMESS and on using automated tools to tune performance on multicore and manycore architectures.