This is the mail archive of the gcc@gcc.gnu.org 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: some remarks about gcc4.0


On Thu, 06 Jan 2005 08:20:21 -0800, Mark Mitchell <mark@codesourcery.com> wrote:
> IA32 is an important platform for GCC.  Performance is important for
> GCC. IA32 performance is important for GCC.  However, IA32 SSE
> performance is just one kind of performance.
I understand that IA32 is just one of the many supported targets.
On the other hand the trend for current x86 cpu is to deprecate the
use of the fpu entirely for SSE only operations, so it's bound to
become more critical; it's already the case of k8 & p4.

> So, we'll make an effort to fix problems
> with IA32 SSE code generation that were not present in previous releases
> -- but we'd probably not worry about shipping GCC 4.0 just because there
> were missed optimizations that had also been missed in the past.
That's what i got and the reason why i worried.
You have to admit that pre-4.0 gcc versions were not up to the job
when dealing with SSE: lots of gotcha, poor performance etc (ie
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14776 etc...)
So i really doubt they make a valid standard to measure against.
Add on top of that there's little automatic testing done for SSE atm
(again as far as i know from my own experience ie
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18944,
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18073).

Anyway, i can compile/patch a gcc snapshot and deal with asm... just
don't expect the bulk of your IA32 users to.


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