An environment for evaluating architectures for spatially mapped computation: System architecture and initial results

An environment which addresses several problems in evaluating massively parallel array architectures is described. A realistic workload including a series of applications currently being used as building blocks in vision research has been constructed. Both flexibility in architectural parameter selection and simulation efficiency are maintained by combining virtual machine emulation with trace driven simulation. The tradeoff between fairness to diverse target architectures and programmability of the test programs is addressed through the use of operator and application libraries. Initial results are presented indicating the appropriate balance between register file and cache to optimize performance under varying levels of processor element virtualization