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: Performance with Intel Foster chip


On Friday 22 February 2002 04:33, Jan Hubicka wrote:
> > Paolo Carlini wrote:
> > >Thomas Hiller wrote:
> > >>Hi,
> > >>
> > >>will gcc 3.1 support the architecture of the Intel Foster chip ?
> > >
> > >What's "Foster"? The code name of a new member of the P4 family?
> > >
> > >P.
> >
> > That's the name of the upcoming server chip (to be announced at CeBit).
> > Let's say it's a P4 for servers.
> > What's special about this chip is that it has two execution units in the
> > core. And I'm not sure, whether the current gcc can make use of these
> > two units.
The 2 units are not completely separate; they share execution units and 
cache, and split the Store Allocation Buffers between them (3 each).
>
> I believe these units are visiable via hyperthreading functionallity, so
> any code that is multithreaded can take advantage of that (with some kernel
> side tunning).
linux kernel 2.4.17 is supposed to support this, if the BIOS is correct.  
Evidently, some of the BIOS have not been checked out with linux.  Nor has 
the question been resolved whether Windows XP Pro will support dual 
Hyper-Threaded CPU's, nor has any interest been showed in changing stock 
Windows binutils to support stack alignments.
>
> GCC 3.1 has basic support for tunning for P4 architecture. Not perfect, but
> it is at the place.  I plan to run comparison agains P3 optimized code soon
> to see what it does in practice.
>
> Concerning floating point operations, the P4 and -mfpmath=sse is huge win.
>
> Honza
>
> > - Thomas


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