Interpreter vs. Compiler Performance at Run-Time

Abstract

Both ``interpreter'' and ``compiler'' cover a wide range of implementation techniques, and in some cases one can argue where to draw the line. Consequently, both approaches also have a wide performance spread, leading to comparable performance in some cases. The present work looks closely at the SwiftForth JIT compiler and the Gforth interpreter. Even with dynamic superinstructions and IP-update optimizations, Gforth suffers from interpretive overhead, but that is compensated by Gforth's additional sophistication in caching stack items in registers, an optimization that SwiftForth does not employ.