This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Possible Enhancement - Forcing double prec. on Intel FPUs
- To: rth at cygnus dot com
- Subject: Re: Possible Enhancement - Forcing double prec. on Intel FPUs
- From: loesche at math dot tu-dresden dot de (Ralf Loesche)
- Date: Wed, 17 Jun 1998 10:36:22 +0200 (MET DST)
- Cc: egcs at cygnus dot com
>
> On Tue, Jun 16, 1998 at 06:30:24PM +0200, Ralf Loesche wrote:
> > This is exactly what I do from within Matlab.
> [...]
> > When you have some source code to compile that suffers from the
> > extended prec. it would be nice to tell the compiler to add the
> > appropiate call to __setfpucw() to the startup code.
> > Thereby removing the need to modify the sources.
>
> What you apparently missed is that my solution does get setfpucw
> called from the startup code, and so doesn't require modiying
> sources. You just add -lieee-dbl to the link line.
Yes, I missed that. Another question comes to my mind here:
Is there another elegant way (like the one you pointed out) to
compile just a few files in a huge project that they get executed
in double prec. while the remaining part still uses extended prec.?
To make it more clear: I use BLAS and LAPACK a lot. Some tests
(finding the largest element in a vector) fail unless they are
performed in double prec. or -ffloat-store is used. Thus I would like
to compile those libraries with different rounding modes for some
of the files. And, if possible, without changing the source code.
Your solution just fixes the problem for the whole executable.
Any hints?
Greetings, Ralf