[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

gcc 3.0 slower for intel too



Guys,

On last report, gcc 3.0 was slowing down ATLAS on the athlon.  Well, it slows
down the Pentium III as well.  See:
   http://www.cs.utk.edu/~rwhaley/ATLAS/gcc30.html
for details.

If any of you know somebody who knows somebody, you may wanna talk with them.
Even better, if you have access to floating point intensive code of your own,
I urge you to see what gcc 3.0 does with it.  As long as the performance
problems are local to ATLAS, I am not confident gnu, in particular (ATLAS
seems to be at least somewhat on RedHat's radar, at least) will do anything
with this.  So, if you have optimized x87 code, give 3.0 a whirl, and if it
is slower than 2.9x, add to my bug report.  Getting this fixed is going to
beat having to tell users "get an old gcc or lose 10% of performance"; I've
had to do this on alphas, and it's not greeted warmly by the audience . . .

I guess the good news is that gcc 3.0 seems to be faster on the ev6, and
roughly the same speed on UltraSparcs . . .

Cheers,
Clint