Stack operations folding in Java processors

Abstract

Traditionally, the performance of a stack machine has been limited by the true data dependency. A performance enhancement mechanism, stack operations folding, was used in Sun Microelectronics' picoJava-I design, and it can fold up to 60% of all stack operations. The authors use the Java bytecode language as the target machine language, and study Java instruction folding on a proposed folding model, the POC model, which is used to illustrate the theoretical folding operations. Various practical folding strategies based on the POC model are introduced and evaluated. Statistical data show that the 4-foldable strategy eliminates 84% of all stack operations, and the 2-, 3-, and 4-foldable strategies result in overall program speedups of 1.22, 1.32 and 1.34, respectively, as compared to a stack machine without folding. Furthermore, the 4-foldable strategy is the most practical and cost effective of a Java stack machine design with a decoder width of 8 bytes. Circuit simulation results show that a 100 MHz 4-foldable folding mechanism can be realized with 0.6 µm CMOS standard cells, or 240 MHz with 0.25 µm CMOS technology.