Various statistics presented in the thesis are based on analysis of benchmark programs undertaken by the author or reported by others. The purpose of this section is to summarise the nature of these benchmarks and to justify their relavence to this work.
The sources of data are sumarised in table A.1.
Reference Processor Applications [ENDE93] SPARC ccreg (C compiler) (Appendix C) (Load/Store architecture with register windows) compress (data compression) dhry10 (synthetic benchmark) espresso (logic minimisation) fgrep (text search) gcc (C compiler) (same as [HENN90]) rleflip (image manipulation) [HENN90] DLX gcc (C compiler) (figure 2.17) (MIPS-like Load/Store architecture) Tex (text formatter) Spice (analogue circuit simulator) [JOHN91] MIPS R2000 5diff (text file comparison) (table 3-2) (Load/Store architecture) awk (pattern scanning and processing) ccom (optimizing C compiler) compress (file compressing using Lempel-Ziv encoding) doduc (Monte-Carlo simulation, double precision floating point) espresso (logic minimization) gnuchess (computer chess program) grep (reports occurences of a string in one or more text files) irsim (delay simulator for VLIS layouts) latex (document preparation system) linpack (linear equation solver, double precision floating point) nroff (text formatter for a typewritter-like device) simple (hyrodynamics code) spice2g6 (circuit simulator) troff (text formatter for typesetting device) wolf (standard-cell placement using simulated annealing) whetstone (standard floating point benchmark, double precision floating point) yacc (compiles a context-free grammar into LR(1) parser tables)In all cases the programs have been compiled with optimising compilers.
The applications used are intended to be typical of those commonly used in workstation environments. Some applications of low power processors have workstation-like characteristics (e.g. portable computers) and in these cases these benchmarks are relevant. In other applications with a different type of workload (e.g. signal processing) these results are less relevant and should be used with more caution.