Enhancing Image Processing Performance for PCID in a Heterogeneous Network of Multi-core Processors

摘要

The Phy sically-Constrained Iterative Deconvolution (PCID) image deblurring code i s being ported to heterogeneous networks of multi-core systems, including Intel Xeons and IBM Cell Broadband Engines. This paper reports results from experiments using the JAWS supercomputer at MHPCC (60 TFLOPS of dual -dual Xeon nodes linked with Infiniband) and the Cell Cluster at AFRL in Rome, NY. The Cell Cluster has 52 TFLOPS of Playstation 3 (PS3) nodes with IBM Cell Broadband Engine m ulti-cores and 15 dual-quad Xeon head nodes. The interconnect fabric includes Infiniband, 10 Gigabit Ethernet and 1 Gigabit Ethernet to each of the 336 PS3s. The results compare approaches to parallelizing FFT executions across the Xeons and the Cell's Synergistic Processing Elements (SPEs) for fram e-level im age p rocessing. The experiments included In tel's Perfo rmance Prim itives an d Math Kern el Library, and FFTW3.2.