Interpreter Instruction Scheduling

摘要

Whenever we extend the instruction set of an interpreter, we risk increased instruction cache miss penalties. We can alleviate this problem by selecting instructions from the instruction set and re-arranging them such that frequent instruction sequences are co-located in memory. We take these frequent instruction sequences from hot program traces of external programs and we report a maximum speedup by a factor of 1.142. Thus, interpreter instruction scheduling complements the improved efficiency of an extended instruction set by optimizing its instruction arrangement.