Extensible microprocessor without interlocked pipeline stages (emips), the reconfigurable microprocessor

Abstract

In this thesis we propose to realize the performance benefits of applicationspecific hardware optimizations in a general-purpose, multi-user system environment using a dynamically extensible microprocessor architecture. We have called our dynamically extensible microprocessor design the Extensible Microprocessor without Interlocked Pipeline Stages, or eMIPS. The eMIPS architecture uses the interaction of fixed and configurable logic available in modern Field Programmable Gate Array (FPGA). This interaction is used to address the limitations of current microprocessor architectures based solely on Application Specific Integrated Circuits (ASIC). These limitations include inflexibility, size, and application specific performance optimization. The eMIPS system allows multiple secure extensions to load dynamically and to plug into the stages of a pipelined central processing unit (CPU) data path, thereby extending the core instruction set of the microprocessor. Extensions can also be used to realize on-chip peripherals, and if area permits, even multiple cores. Extension instructions reduce dramatically the execution time of frequently executed instruction patterns. These new functionalities we have developed can be exploited by patching the binaries of existing applications, without any changes to the compilers. A FPGA based workstation prototype and a flexible simulation system implementating this design demonstrates speedups of 2x-3x on a set of applications that include video games, real-time programs and the SPEC2000 integer benchmarks. eMIPS is the first realized workstation based entirely on a dynamically extensible microprocessor that is safe for general purpose, multi-user applications. By exposing the individual stages of the data path, eMIPS allows optimizations not previously possible. This includes permitting safe and coherent accesses to memory from within an extension, optimizing multi-branched blocks, and throwing precise and restart able exceptions from within an extension. This work describes a simplified implementation of an extensible microprocessor architecture based on the Microprocessor without Interlocked Pipeline Stages (MIPS) Reduced Instruction Set Computer (RISC) architecture. The concepts and methods contained within this thesis may be applied to other similar architectures. Given this simplified prototype we look forward to propose how this architecture will be expanded as it matures.