Re: floating point problem

[dewar@merv.cs.nyu.edu (Robert Dewar)]

Joel said

<<Ouch.  Actually all intel 486DX chips have a math co-processor however
386DX chips and 486SX chips do not.  I wouldn't call computers a few
years old ancient just yet. A 286 you might be justified calling ancient
but even that seems to be stretching it. Also, corret me if I am wrong
but isn't floating point support provided by software emulation on such
machines?  It may be that your problem isn't.  I'd go find a machine that
doesn't have floating point to try your program on just to be sure that
there is a problem at all.>>

  As I noted in my previous post, GNAT will work on any x86 which has a
  floating-point processor, or with any operating system environment that
  transparently emulates the floating-point.
  
  GNAT will definitely NOT work in some environments. In particular, the
  DOS version of GNAT on DJGPP does not work on such machines, so if you
  try the DOS version on an ancient machine without a coprocessor it
  definitely will NOT work.
  
  (as for whether a 386 is ancient at this stage, that's a matter of
  terminology, you can barely give away a 386 based machine at this
  stage!)
  
  Given that one can buy a brand new machine that can run GNAT just fine
  for a few hundred dollars these days, this has never been a priority
  (getting GNAT to work on non-coprocessor machines under DOS), and of
  course as time goes on, the utility of such an effort diminishes.

<<it maybe that your problem isn't>>

   We don't have a problem. As far as we are concerned, GNAT simply does
   not support x86 machines with no floating-point emulation, and we do
   not intend to do anything about it!
   
   If some volunteer wants to work on this problem, that would be useful to
   some people, but better work fast, or even Joel will be willing to agree
   about ancient :-)
   
   Going back to cheap PC's, if you want to run GNAT, you really need a minimum
   of a 486 with 16 megs of memory. Yes, I know you *can* run on a 386, and
   I know you *can* run in 8 megs, but these days when 16 megs of memory costs
   $79, you have to value your time very cheaply for this to be worthwhile.
   
   In general memory can make a HUGE difference in compilation speeds with
   GNAT. GNAT is NOT written to compile fast on small machines! How much
   memory you need depends on the environment. For running simple programs,
   16 meg on a PC, or 32 megs on a Unix work station is adequate, but for
   serious large programs, I would recommend double this. We recently had
   a customer complain that a compilation took 4 hours. We recommended
   adding memory (to go from 32megs to 64megs on a SPARC, costing less
   than $300 I think), and the compilation time fell to 5 minutes.

Robert Dewar
Ada Core Technologies