GCC 2.95 and floating point precision

[Toon Moene]

Stephen L Moshier wrote:
> 
> GCC-2.95 is more sensitive than any previous GCC version has been
> to the floating point pointer aliasing feature-bug of ANSI C.

:-)

> The program you referenced contains code that is not valid ANSI C,
> such as the following:
> 
>   long double ChopVar;
>   *Lo = *((unsigned long int *) &ChopVar);
>   *Hi = (*(((unsigned long int *) &ChopVar) + 1)) & ~2147483648U;
> 
> You may not do this any more.  It is verboten.  You are supposed to use
> a union containing the float and the int together.

Phew - it's questionable whether I would ever have found that.  Thanks,
Stephen ...

Using either -O1 or -O2 -fno-strict-aliasing delivers the same result. 
Far be it from me to proclaim that it's the _correct_ result - I only
know the 20 leading digits of pi (and that's including the `3').

HTH,

-- 
Toon Moene (toon@moene.indiv.nluug.nl)
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Phone: +31 346 214290; Fax: +31 346 214286
GNU Fortran: http://world.std.com/~burley/g77.html