Design Space Exploration for Java Processors with Cross-Profiling

Most processors are used in embedded systems, where the processor architectures are diverse due to optimizations for different application domains. The main challenge for embedded system processors is the right balance between performance and chip size, which directly relates to cost. An early estimation of the performance for a new design is of paramount importance. In this paper we propose cross-profiling for that performance estimation, which can be accomplished very early in the design phase. We evaluate our approach in the context of a Java processor for embedded systems using cross-profiling on a standard desktop Java virtual machine. We explore the performance impact of various processor design choices and optimizations, such as different caches strategies or pipeline organizations, and come up with an improved processor design that yields speedups of up to 40% on standard Java benchmarks. Comparing the generated cross-profiles with the execution of benchmarks in real hardware confirms that our approach is sound.