Block Prefetching for Numerical Codes

Cache optimization is a crucial technique for most numerical code to exploit the performance of modern processors. It can be classified into improving access locality, and prefetching. Inherent algorithm constrains often limit the first approach which typically uses a blocking technique. While there exist automatic prefetching mechanism in hardware and/or compilers, they can not complement blocking with additional prefetching. We describe application controlled block prefetching, allowing to further improve on numerical code already optimized by blocking. It shows its benefits on both synthetic code and matrix multiplication by using the 2nd core of a dual-core processor as engine for executing block prefetching instructions.