Re: [PATCH 5/5] perf, x86: Add support for AMD family 15h core counters

[Robert Richter <robert.richter@amd.com>]

On 03.02.11 04:38:03, Peter Zijlstra wrote:
> On Thu, 2011-02-03 at 10:00 +0100, Robert Richter wrote:
> > 
> > 
> > ok, nb events may be implemented independent from core events in a
> > separate struct pmu.
> > 
> > I still would prefer a lookup table for counter addresses. Adding a
> > shift parameter to struct x86_pmu to do a
> > 
> >       addr = base + (index << shift)
> > 
> > seems to me a quite special solution that may not be reused in other
> > implementations
> 
> What other implementations? I hope people will not re-arrange the MSR
> layout on every new model, that'd be quite annoying.

I mean counters referred by index that cannot be derived from the base
address like fixed counters, BTS, IBS, P4... Often this is implemented
in if/else if paths.

> >  while a lookup table is more generic. I also don't
> > see a performance or memory impact there.
> 
> Well it is an extra pointer chase and data cache hit just to get
> something you can trivially compute.

Indeed, cache pollution is an argument.

> > Anyway, a shift parameter would work too. What do you think? 
> 
> I think the alternatives thing is probably nicest, except for having to
> write the bits in asm.

Will send an updated version.

-Robert

-- 
Advanced Micro Devices, Inc.
Operating System Research Center