Automatic Array Alignment in Parallel Matlab Scripts

We present the ParAL system which compiles Matlab scripts into C programs with calls to a parallel run-time library. The novel feature of the compiler is the optimisation of array alignment which reduces or eliminates unnecessary communication overheads. We have evaluated this technique on several Matlab codes. For comparison, the same applications were hand-coded using the PBLAS library. The aligned codes were on average 43% faster then the misaligned codes, with the speedup factor of almost 4 achieved in some cases. This optimisation technique enabled ordinary Matlab scripts to run at a similar speed as manually optimised PBLAS codes. 1. Introduction Many practical applications can be implemented using matrix operations as building blocks. The need for software infrastructure which supports development of such applications has resulted in many array programming tools and environments. Libraries of subroutines such as LINPACK were among the first such tools, and the Matlab language w..