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

Andrew Pinski pinskia at gmail.com
Thu Mar 22 08:16:00 EST 2007


On 3/21/07, Segher Boessenkool <segher at kernel.crashing.org> wrote:
> Are ucoded insns really _that_ bad on CBE?  You certainly
> should keep lha/lwa with -Os btw.

Yes as they cause the other hyperthread to stall while the waiting for
the microcoded instruction to finish and there is a 7cycle bubble for
the decoding stage.  So yes they are really that bad on the current
Cells.  I am trying to remember but I think the speed up is around
10-20%, maybe higher for multiple threaded programs.

-- Pinski



More information about the cbe-oss-dev mailing list