On 8/11/07, Jonathan Bober <[EMAIL PROTECTED]> wrote:
> (3a)What I think might be the best idea, at least on Linux, is to change
> the compilation settings for quad double so that the fpu fix is not
> needed. There are two ways to do this: If a processor supports sse2,
> then passing gcc -march=whatever -msse2 -mfpmath=sse (maybe the -march
> isn't needed) will cause gcc to use sse registers and instructions for
> doubles, and these have the proper precision. In fact, gcc already does
> this by default for x86-64 cpus, so the quad double package doesn't even
> need the fpu fix on those architectures. Also, this has the added
> benefit of being faster.
>
> (3b)For processors that don't support sse2, gcc can be passed the
> -ffloat-store option, which fixes the problem by storing doubles in
> memory after every operation, ensuring that they are always correctly
> rounded to double precision. This slows things down a little bit, but
> would probably be much simpler than option (1).
>
> Personally, I think that I like options 3a and 3b. These would probably
> require rewriting the configure script for quad double. I don't know how
> to do that, but it probably isn't that hard.

>From what you say, it sounds like we should *definitely* do 3 if it works,
and as soon as possible no less.   That said, I am not personally going to do
this because I don't understand the issues well enough, or how to test
that I haven't messed anything up.  I hope somebody will volunteer
to create a new quaddouble package that implements your idea
*and* tests that it really works.   Moreover, whoever does that should
definitely write to the quaddouble authors and ask them to add
instructions to their README -- currently the quaddouble README
strongly suggests that the only option is using their f_fpu_fix_start
functions.

 -- William

--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://sage.scipy.org/sage/ and http://modular.math.washington.edu/sage/
-~----------~----~----~----~------~----~------~--~---

Reply via email to