On Sun, 24 Feb 2002, Dave Seff wrote:

>  I took gcc 2.95.3, 2.96-80 (current cooker version)

It is not.

> It it obvious that gcc 2.95.3 creates code that is twice as fast as the
>  latter versions. I ran this test after reading this article:
> http://www.cs.utk.edu/~rwhaley/ATLAS/gcc30.html

That "atlas problem" should be fixed in 3.0.4-2mdk. Jan Hubicka, who made
the patch, had to say on the gcc mailing-list:

"There was nothing wrong with the scheduler and the analysis on page are
somewhat missleading. Real problem was that gcc "forgotten" about
posibility of using memory operand in certain cases of commutative i387 fp
instructions requiring one additional register. (this happent as result of
two independent major change sin the compiler) This register is not
available in the loop curefully written for 8 registers and causes the
performance drop."

BTW, the patch also helps the P4 FPU, for that particular case...

Bye,
Gwenole.


Reply via email to