This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] External vectorization library (target) support

On Sun, 10 Dec 2006, Gabriel Dos Reis wrote:

> On Sun, 10 Dec 2006, Richard Guenther wrote:
> | Or we could do a techincal decision - but as there isn't even a
> | technical discussion about the matter I don't see things going
> | forward (neither from my peekings on the SC through my private
> | channels).
> I think we can proceed with the technical aspects. And leave the
> political aspect, if there really is one to others.
> How do we determine when we link libgc-math into programs?
> Could GCC assume that any standard C/C++ mathematical function is
> "defined" by GCc (e.g. libgcc-math) and therefore apply more knowledge
> in optimizations?

Yes, we could do so (basically get rid of TARGET_C99_FUNCTIONS
restrictions).  One way of doing it is with extensive configure
checking of the target library and providing only the missing bits,
another way is to use tricks with weak symbols so that libgcc-math
only "wins" in case the target support is not available.  I'm not
sure the latter will work on all targets, but at least for ELF
targets it should work.


Richard Guenther <>
Novell / SUSE Labs

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]