On Fri, 18 Oct 2002, Stavros Macrakis wrote:
> Thanks for the explanation, which I received just after sending off my
> last mail!
>
> So extended precision does not implement a sticky bit? (Otherwise it
> would be storing correctly rounded results.) That's annoying.... Oh,
> well.
>
> GCL and Emacs Lisp must run with extended precision turned off. In a
> quick look, I couldn't find where they did this in the source; maybe
> they're just linked against a crt0 that takes care of that...? Or maybe
> the cygwin default setting is different from the Windows default
> setting?
>
> Thanks again,
>
> -s
>
It's possible to do this with inline code for each operation. The g77
that comes with gcc2.95 does that but it's slow. Later versions don't
and they're much faster.
In any event, it strikes me that an algorithm that depends on the right
precision is probably broken.
>
> PS About the volatile/optimization issues, I'm pretty familiar with
> them, having worked on a bunch of compilers over the years, which is why
> I used FMEM to force all intermediate results to memory.
>
>
Using volatile is more politically correct these days. Either way works.
10