Messages in this thread | | | From | David Mosberger <> | Date | Wed, 11 Aug 2004 01:47:33 -0700 | Subject | Re: [PATCH] [LSM] Rework LSM hooks |
| |
>>>>> On Tue, 10 Aug 2004 16:31:12 -0400 (EDT), James Morris <jmorris@redhat.com> said:
James> On Tue, 10 Aug 2004, Chris Wright wrote: >> Thanks, James. Since these are the only concrete numbers and >> they are in the noise, I see no compelling reason to change to >> unlikely().
James> There may be some way to make it ia64 specific. Is it a cpu James> issue, or compiler?
I'm pretty sure the "unlikely()" part could be dropped with little/no downside. The part that's relatively expensive (10 cycles when mispredicted) is the indirect call. GCC doesn't handle this well on ia64 and as a result, most indirect calls are mispredicted.
An alternative solution might be to have a call_likely() macro, where you could predict the most likely target of an indirect call. Perhaps that could help other platforms as well.
--david - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |