Re: mtune=nocona

[Vladimir Makarov <vmakarov redhat com>]

Tomas Mraz wrote:

> On Thu, 2005-06-09 at 10:15 -0400, Dan Williams wrote:
>  
>
> > The question, distilled:
> >
> > Would you like: (a) Small speed hit for Opteron users, OR (b) much
> > larger speed hit for EM64T users?
> >    
>
> The question is how small is the small hit and how large is the large
> one. Percentage numbers anyone?
>
>  
I have an old data for gcc3.4.  Data should be practically the same for 
gcc4 because tuninng affect back back end optimizations which have not 
been changed since gcc3.4.

In brief, when we use tuning to AMD64, SPECFP2000 (floating point 
programs) is 28% worse and SPECINT2000 
(integer benchmarks) is 0.6% worse on Intel nocona processor.

When we use tuning to Intel Nocona, SPECFP2000 is 2.7% and SPECINT2000 
is 1.6% worse on Opteron.

Code tuned for Intel Nocona parctically always has smaller size.

So I think tunning to nocona by default is a right decision.