edwin@maui.cs.ucla.edu (08/31/88)
I need a FORTRAN compiler or preprocessor that can implement non-standard arithmetic (i.e. INTERVAL arithmetic) for error analysis. Can anyone tell me how to obtain FORTRAN-SC, FORTRAN 8x, AUGMENT I & II or any other compiler or pre-compiler that will do the job? Bob Tisdale (213) 825-6010.
williamo@hpcupt1.HP.COM (William O'Saughnessy) (08/31/88)
The method that I have used to implement Interval arithmetic is to declare the interval variables as complex and then provide my own library at link time that replaces the compiler complex arithmetic routines with interval arithmetic routines. This trick will work almost any compiler that does not generate inline code for complex arithmetic. Good Luck. Bill O'Shaughnessy
bruces@marque.mu.edu (Bruce Stephens) (09/02/88)
I love the idea of using COMPLEX to do interval arithmetic. I suspect it would involve a fair amount of fiddling though. I'm involved in a project on differentiation arithmetic, which involves storing not just two numbers (the ends of an interval), but any number (corresponding to partial derivatives of various orders). Hence COMPLEX numbers are not sufficiently big - and I need to do genuine operator overloading. FORTRAN-SC is an IBM product - the Fortran equivalent of PASCAL-SC. It allows data structures, operator overloading, dynamic memory allocation, and all that sort of stuff. Unfortunately it only works on IBM machines and if you're interested get in contact with them. (They also have ACRITH, which does interval computations very accurately.) AUGMENT seems to be largely dead. There were a lot of papers published on it, but virtually all came from the place where it was made (the MRC). On top of that, it was written some time before Fortran-77. Porting it has caused several people a great deal of wasted effort - it seems to be very difficult (in spite of the descriptions). I don't know about AUGMENT II, maybe this is an updated version. If you don't mind changing languages, then use one that allows operator overloading, (ADA, C++, Fortran-SC, ALGOL-68, ...). If you have to use Fortran, then maybe you ought to wait until the preprocessor that four of us designed over August is written. It was designed to satisfy the needs of people wanting to do interval arithmetic (and interval Taylor series arithmetic) as well as my application of partial derivatives. The design specifications will be published as a University of Bristol mathematics departement internal report, I'm not sure exactly when. I can be reached until 8 Sept here at Marquette, and subsequently at Bristol University, email address below. Bruce. ---------------------------------------------------------------- Bruce Stephens. (University of Bristol) UUCP: ...!marque.mu.edu!bruces JANET: StephensBru@uk.ac.bristol.msa (Please use UKACRL from the US, UKC costs money!)