Apurdue.138 net.applic utzoo!decvax!pur-ee!purdue!dar Sat Dec 19 18:12:05 1981 FFP program analysis As Scott Baden (ucbvax!baden) alluded earlier, there is an FFP interpreter up and running at Purdue. The intent is to gather some statistics about the relative frequencies of various language operators as a prelude to the design and analysis of some parallel hardware to support applicative languages. I would be interested in any ideas about what suitable machine primitives should be. The absence of side-effects in functional programs has caused a gleam in the eye of more than one hardware designer. As a practical matter, the transition from a sequential phase of processing to a parallel phase of processing requires either program/data copying and distribution to multiple processors or sharing with its consequent memory contention. It is actually possible for the startup overhead to negate any possible time reduction one might hope to obtain from parallel evaluation. It seems that the ratio processing time/communication overhead is critical. By taking a small number of FFP programs and assigning execution times to the language primitives, I've obtained a few plots of theoretical parallelism vs time. The curves tend to have many large "spikes." That is, parallelism vs time changes both dramatically and frequently. This may be a function of the programs I have or the assignment of execution times. Any conflicting evidence/ideas/comments out there? Also, does anyone have any FP/FFP programs they would be willing to let me have for analysis? Dan Reed (purdue!dar) ----------------------------------------------------------------- gopher://quux.org/ conversion by John Goerzen of http://communication.ucsd.edu/A-News/ This Usenet Oldnews Archive article may be copied and distributed freely, provided: 1. There is no money collected for the text(s) of the articles. 2. The following notice remains appended to each copy: The Usenet Oldnews Archive: Compilation Copyright (C) 1981, 1996 Bruce Jones, Henry Spencer, David Wiseman.