Instruction Complexity in implicit-execution architectures: orthogonality, optimisation, and VLSI design

With the rapid emergence of Java as a machine independent programming environment, the stack-based processor model has entered a renaissance of interest after a long period of indifference from the general research community. The implicit execution mode employed in Java byte-code interpretation demands optimal execution of code on a stack-based virtual (or physical) machine model. One efficient solution is to utilise a stack-based microprocessor paradigm in which the associated native code mirrors the semantics of the object byte-code. This approach already has a past-history that spans the era of ALGOL through to present day FORTH-Engine implementations. However, the development of virtual and/or native semantic content of these environments has suffered a haphazard evolution, influenced more by convenience than by formalised concepts of stack-oriented execution. As a result, instruction-set schemes for stack based processing models have suffered from a lack of orthogonality that suits human programmers, but complicates automated code generation and optimisation. In this paper, we propose a classification scheme, a scaleable and symmetrical model for stack manipulation, which allows the degree of instruction complexity to be specified and the orthogonality of those functions to be visualised. Using a virtual machine simulator, VHDL models of CPU designs, and analysis of synthesised VHDL models, we show that the effects of varied degrees of instruction complexity have a clear and quantifiable impact on program efficiency, system performance, and VLSI logic characterisation.