[RFC/PATCH] Make powerpc64 use __thread for per-cpu variables

[Alan Modra]

On Thu, May 11, 2006 at 07:05:24AM +1000, Paul Mackerras wrote:
> No, Richard has a point, it's not the value that is the concern, it's
> the address, which gcc could assume is still valid after a barrier.
> Drat.

That may never happen, at least with a compiler that knows how to
optimise away the addi.  You're using -mtls-size=16 so all your accesses
should look like

	lwz rn,per_cpu_var at tprel(13)

gcc shouldn't think there is any reason to cache the address.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre