Date: Tue, 06 Sep 1994 11:55:28 +1000
From: Bill Metzenthen <BILLM AT vaxc DOT cc DOT monash DOT edu DOT au>
Subject: Re: djgpp 1.12 math emulator
To: djgpp AT sun DOT soe DOT clarkson DOT edu


Pieter Kunst (kunst AT prl DOT philips DOT nl) wrote:

> Minxian Yang (s9001200 AT csdvax DOT csd DOT unsw DOT edu DOT au) writes:
> > I tried to port a Gamma function routine in Numerical Recipes Fortran to 
> > djgpp 1.11. The result was something like Gamma(7.0) = 712.906 (the true
> > value should be 720.000). My computer is an NEC 386D20. Though my problem
> > is not with djgpp's emulator, I do need some help on the precision of djgpp.
>
> Hello Minxian,
>
> I tried the sample program below (gamma.c) with DJGPP 1.12m1 to check the 
> gamma function given in Numerical Recipes. 
> Gamma(7) returns 719.999999868 on my PC, but I have a 486DX (which includes 
> a coprocessor). I'm sure some of the DJGPP users with a 386SX machine 
> (i.e. no coprocessor, just EMU387) can compile the program below to see what
> result is obtained with EMU387 (and post the result to you, as well as to the
> list).

If accuracy is of importance, I suggest that you try the alternative FPU
emulator which is in wmemu112.zip.  It was written for djgpp 1.11, but I
have no reports yet that it doesn't work with 1.12 (I use Linux these
days, so I don't keep up to date with the very latest djgpp, but I will
attempt to fix problems if people report them).

Under djgpp 1.11, I get the following result with Pieter's program:

gamma(7) = 719.999999868


--Bill