K312240@AEARN.BITNET (Klaus Kusche) (05/28/90)
Dear Mailing List: About Rick Stein's mail concerning parallelizing Maple: This is reasonable (Stephen Watt has parallelized Maple in 1986 on a network of VAXen as his PhD thesis), and we will most likely do that again in the near future, too. However, we will most likely not do it on transputers. The reason for that is not specific to Maple, but also impedes porting of many other interesting systems (Lisp's, Prolog's, expert systems, ...) which are written in C and in public domain for the academia: *** As long as there is no C compiler for transputers which compiles several ten thousand lines of standard Unix C code without modification and troubles, and generates reasonably efficient (and correct!) code, and comes with Unix-compatible libraries, such ports will not happen! They are not worth the trouble! *** A few examples: With Inmos parallel C, even XLISP (which is completely trivial code and compiles even on the most primitive toy C compilers) requires (for some mysterious reasons, the compiler just gives some misleading error messages for obviously correct and not at all tricky code parts) about a dozen source code changes. I was told that currently no C compiler for transputers exists which is able to compile the Unix distribution of TeX without too much hassle. ... If someone has experience in getting large Unix C applications to run on transputers (or failed to do so), please post! We are about to switch most of our work (except for the Strand-based projects) from transputers to some boxes which can be used and programmed, too. ************************************************************************ * Klaus Kusche * * Research Institute for Symbolic Computation * * Johannes Kepler University Tel: +43 7236 3231 67 * * A-4040 Linz Telex: (Austria) 22323 uni li a * * Austria (Europe) Fax: +43 7236 3338 30 * * * * Bitnet: K312240@AEARN * * Arpa/CS/Internet: K312240%AEARN.BITNET@CUNYVM.CUNY.EDU * * UUCP: .....!aearn.bitnet!K312240 * * Janet: k312240@earn.aearn or k312240%aearn@earn-relay * ************************************************************************
cleary@cpsc.ucalgary.ca (John Cleary) (05/29/90)
In article <9005280913.AA10653@tcgould.TN.CORNELL.EDU> K312240@AEARN.BITNET (Klaus Kusche) writes: > ....some bits deleted > >However, we will most likely not do it on transputers. The reason for >that is not specific to Maple, but also impedes porting of many >other interesting systems (Lisp's, Prolog's, expert systems, ...) >which are written in C and in public domain for the academia: > >*** As long as there is no C compiler for transputers which compiles >several ten thousand lines of standard Unix C code without modification >and troubles, and generates reasonably efficient (and correct!) code, >and comes with Unix-compatible libraries, such ports will not happen! >They are not worth the trouble! *** > We have been using a Meiko computing surface for about a year now for a large commercial system (A distributed simulation system using the TimeWarp algorithm). The code executed on the Transputer includes about 75,000 lines of C++ this is expanded to some larger number of lines of C. Of the C compilers we have used this has been the best, it generates good code, it is ANSI standard and has very few bugs. C++ generated code tends to stretch C compilers to the limit and this one stood up well. The main problem we have had with it is that the optimize option doesnt work. The rest of Meiko's software unfortunately isnt up to the same high standards. I dont recall the ancestry of this compiler nor do I know whether it can be used with other Transputer boxes. John G. Cleary Department of Computer Science University of Calgary 2500 University Drive N.W. Calgary Alberta T2N 1N4 Canada cleary@cpsc.UCalgary.ca Phone: (403)282-5711 or (403)220-6087
nick@lfcs.ed.ac.uk (Nick Rothwell) (05/29/90)
In article <9005280913.AA10653@tcgould.TN.CORNELL.EDU>, K312240@AEARN (Klaus Kusche) writes: >*** As long as there is no C compiler for transputers which compiles >several ten thousand lines of standard Unix C code without modification >and troubles, and generates reasonably efficient (and correct!) code, >and comes with Unix-compatible libraries, such ports will not happen! >They are not worth the trouble! *** Please define "standard Unix C code". I hope you aren't suggesting that code written in C for some unix will work on any other. If people write godawful non-portable code for unix, why should ports to the transputer be any easier? It's not the fault of the compilers. I wrote about 20000 lines of code in C under BSD unix. The port to the transputer was pretty immediate. This is because I didn't hack it. Nick. -- Nick Rothwell, Laboratory for Foundations of Computer Science, Edinburgh. nick@lfcs.ed.ac.uk <Atlantic Ocean>!mcsun!ukc!lfcs!nick ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ Ich weiss jetzt was kein Engel weiss