Abstract
An effective method to implement particle filter algorithm on Cell
Broadband Engine in parallel way has been proposed. First problem for the implementaion is resampling procedure of the algorithm, which is
difficult to parallelize, and second one is local memory of
computational core being limited to 256KB. Further subject is effective use of Single Instruction Multiple Data (SIMD) operations. In this paper, we propose following ideas for the implementaion. DMA (Direct Memory Access) list transfer is effectively used to circumvent the limited capacity of local memory in computational core. In resampling, estimation accuracy is maintained and improvement in high performance is achieved by residual resampling for each computational core individualy. SIMD operations are effectively used with Structure of Arrays form. Through performance evaluation experiments, much higher performance of Cell Broadband Engine compared with conventional CPUs has been shown.
