Re: spin_unlock optimization

Gabor Lenart (lgb@oxygene.terra.vein.hu)
Thu, 30 Dec 1999 00:27:38 +0100

On Wed, Dec 29, 1999 at 07:44:02PM +0100, Oliver Henning wrote:
> Hi,
>
> as was mentioned on Kernel Traffic, the spin_unlock optimization (movb $0,%0
> instead of lock; btrl $0,%0) is not used at last. But could there be a
> detection routine (I know this is not a completely new and innovative idea)
> for those older PPro CPUs that require the slow version so that in the sane
> cases the optimization would be used. At least, IMHO it should be a
> compile-time option (defaulting to the slow version) to activate the fast
> version.

Yes. Runtime option is not acceptable imho. movb $0,%0 can be fast but if
you need a conditional instruction to test if the fast version can be used
uses more and more cycles.

-- 
 ---[ LGB/DC ]------------[ University Of Veszprém ]------[ Lénárt Gábor ]---
 "The truth is out there"  "We're l{ea,i}ving together" "The future is dark."
 ---[ 30/2270823 ]--------[ http://lgb.hal.vein.hu ]---------[ 87/477074 ]---
 finger lgb@hal2000.hal.vein.hu for more  !LINUX!  SMS : lgblgb@westel900.net


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/