A performance comparison of several superscalar processor models with a VLIW processor

Superscalar and VLIW processors can both execute multiple instructions each cycle. Each employs a different instruction scheduling method to achieve multiple instruction execution. Superscalar processors schedule instructions dynamically, and VLIW processors execute statically scheduled instructions. This paper quantitatively compares various superscalar processor architectures with a very long instruction word (VLIW) architecture developed at the University of California, Irvine. An architectural overview and performance analysis of the superscalar processor models and VIPER, a VLIW processor designed to take advantage of the parallelizing capabilities of percolation scheduling, are presented. The motivation for this comparison is to study the capability of a dynamically scheduled processor to obtain the same performance achieved by a statically scheduled processor, and examine the hardware resources required by each.