[Cbe-oss-dev] GCC inline memcpy() from local store causes bad page fault

[Gwenole Beauchesne]

Hi,

> I am trying to remember but I think the speed up is around 10-20%, maybe
> higher for multiple threaded programs.

On former benchmarks I could make, the performance improvement with
-mno-gen-microcode (the default for -mtune=cell) was +10% on average with
a peak at 34%.

At a larger scale (distribution wise), it was notable enough that KDE
loads faster than the stock FC5 GNOME, even if it was not built with
--hash-style=gnu. Of course, this was my impression. ;-)

Regards,
Gwenole.