ia64: Don't use static branch prediction
Bernd Schmidt
bernds@redhat.co.uk
Sat Oct 28 09:57:00 GMT 2000
On 28 Oct 2000, Chris G. Demetriou wrote:
> bernds@redhat.co.uk (Bernd Schmidt) writes:
> > /* Guess top and bottom 10% statically predicted. */
>
> does this comment deserve updating? (I'd guess so.)
Yes. Darn - this always happens to me. I really ought to start reading
comments.
> > - else if (pred_val < REG_BR_PROB_BASE * 9 / 10)
> > + else if (pred_val < REG_BR_PROB_BASE / 100 * 98)
>
> in practice, the ordering here is OK because of the (large-ish) value
> of REG_BR_PROB_BASE... but why switch it around?
Paranoia about overflow :)
Bernd
More information about the Gcc-patches
mailing list