Modern compilers use complex optimizations. It is often a problem for programmers to understand how source code should be written to enable optimizations. Interactive tools which guide programmers to higher performance are very important. We have developed such a tool that helps programmers modify their code to allow for aggressive optimization. In this paper, we extend it to support high level memory optimizations such as matrix reorganization. We evaluate the tool using two benchmarks and four dierent compilers. We show that it can guide the programmer to 22.9% higher performance.
Proceedings of 5th Workshop on Programmability Issues for Heterogeneous Multicores (multiprog-12), 2012
Main Research Area:
5th Workshop on Programmability Issues for Heterogeneous Multicores (MULTIPROG 2012)