MIPT: Rapid exploration and evaluation for migrating sequential algorithms to multiprocessing systems with multi-port memories

Abstract

Research has shown that the memory load/store instructions consume an important part in execution time and energy consumption. Extracting available parallelism at different granularity has been an important approach for designing next generation highly parallel systems. In this work, we present MIPT, an architecture exploration framework that leverages instruction parallelism of memory and ALU operations from a sequential algorithm's execution trace. MIPT heuristics recommend memory port sizes and issue slot sizes for memory and ALU operations. Its custom simulator simulates and evaluates the recommended parallel version of the execution trace for measuring performance improvements versus dual port memory. MIPT's architecture exploration criteria is to improve performance by utilizing systems with multi-port memories and multi-issue ALUs. There exists design exploration tools such as Multi2Sim and Trimaran. These simulators offer customization of multi-port memory architectures but designers' initial starting points are usually unclear. Thus, MIPT can suggest initial starting point for customization in those design exploration systems. In addition, given same application with two different implementations, it is possible to compare their execution time by the MIPT simulator.