bimbart@kulcs.uucp (Bart Demoen) (04/10/89)
In response to <10073@megaron.arizona.edu> debray@arizona.edu (Saumya K. Debray) : > commercial Prolog vendor[*] managed to compile append/3 into > sufficiently few machine instructions that code for the whole > procedure fit into the instruction cache of the 68020, giving huge > LIPS numbers. A competing vendor[*] apparently felt compelled to > ... with in-line native code there is not really much choice: either the code runs in the cache, or not; with threaded code, one can carefully tune the system - by placing code of specific intermediate instructions at the rigth place or, as was done by some Prolog implementor(s?), invent benchmark-special-purpose intermediate instructions (like for instance taking together the WRITE mode of GET_LIST A3 + UNIFY_XVAL A4 + UNIFY_XVAR A3) - and then certain benchmarks will run completely in the cache > I have friends at both places so have I, and one of my friends over there said enthousiastically 2 years ago in Melbourne (ICLP4) at the stand where their Prolog was demonstrated: Wow, naive reverse could run completely in the cache of the SUN3, I wonder how fast it will be ! > ... and also because aforementioned friends are quite likely to referee my > papers in the future ... even your enemies have to recognise that your papers are good, but I doubt that you have any enemies bimbart@kulcs