Implementation of Recurrence Calculation on a Heterogeneous Multicore Processor

Heterogeneous multi core architectures are considered as a prominent next generation processor. Cell B.E. (Broadband Engine) was originally designed for a processor suitable for streaming applications such as media processing, but it also can be applied to other high performance computing applications. In order to confirm the effectiveness of the Cell B.E. for other HPC applications, the parallel version of recurrence equation solver and SAXPY routine are implemented and evaluated on the system using the processor. By using the overlapping technique of the DMA processing and the computation, the performance of over 1400 MFLOPS can be achieved with the array size of 107 for the SAXPY routine. On the other hand, a recurrence equation is one of algorithms which cannot be parallelized straightforwardly. We implement it on the Cell B.E. by using the P-scheme algorithm and achieves the speedup of about 4 with 6 SPEs compared with the normal solver with using 1 PPE.