Efficient bytecode optimizations for a multicore Java co-processor system

As the bytecode produced by the Java compiler is unoptimized, the bytecode generated from certain types of idiomatic Java code is inefficient for execution in an interpreter. This effect is amplified in a co-processor system, in which a single processor must process heap accesses and virtual method calls from multiple threads. Two types of optimizations are presented which can be performed directly on bytecode during class loading and which do not require a large amount of processing time. These optimizations are shown to improve performance greatly, up to 29 % in the Embedded Caffeinemark Benchmark suite. Even higher improvements were measured in multithreaded programs.