QUANTITATIVE ASSESSMENT OF MACHINE-STACK BEHAVIOUR FOR BETTER COMPUTER PERFORMANCE.

This paper presents an experiment to quantify stack behaviour during execution of a range of complementary programs. Through better understanding of stack behaviour, further optimisations can be made, not only improving stack machine efficiency, but perhaps influencing future designs both in RISC and CISC technologies. Mainstream technology has always been dominated by explicitly addressed register file architectures, with two clear philosophies predominant. The CISC school of thought demands complex instruction sets to reduce the semantic gap : 'More work for less code'. On the flip-side of the coin, RISC proponents believe simplicity and speed will succeed, even if more instructions are executed. The 'Stack machine' alternative, has been pushed to the back of the queue in terms of research and development. Stack machines abandon traditional register file concepts, and with them, the need for register addressing in program code. Instead, operands are, by default, found at the top of stack. The benefits of reducing instruction size and functional complexity offer potential for comparable performance to that of RISC and CISC architectures. Quantitative assessment of stack behaviour will clearly demonstrate statistical and probabilistic examples of stack actions, helping to guide future designs.