Accelerating Cryptographic Applications Using Dynamically Reconfigurable Functional Units

Abstract

In this paper we propose and evaluate our platform to accelerate applications using custom instruction set extensions. We use a dynamically reconfigurable functional unit (DRFU) to execute the application specific custom instructions generated by our compiler framework. We explore two architectures with different computational granularities for the DRFU (look-up table and ALU based) and evaluate this framework using security and cryptographic applications as a case study. Our results indicate that the use of application specific instruction set extensions reduce code size by 10% and achieve a maximum speedup of 165% (41% on average).