A Video Compression Case Study on a Reconfigurable VLIW Architecture

In this paper, we investigate the benefits of a flexible,application-specific instruction set by adding a run-timeReconfigurable Functional Unit (RFU) to a VLIWprocessor. Preliminary results on the motion estimationstage in an MPEG4 video encoder are presented. Withthe RFU modeled at functional level and under realisticassumptions on execution latency, technology scaling andreconfiguration penalty, we explore different RFUinstructions at fine-grain (instruction-level) and coarse-grain(loop-level) granularity to speedup the applicationexecution. The memory bandwidth bottleneck, typical forstreaming applications, is alleviated through thecombined adoption of custom prefetch patterninstructions and an extent of local memory. Performanceevaluations indicate up to 8x improvement, with loop-leveloptimizations is achieved under variousarchitectural assumptions.