Re: [PATCH 2/2] tcg: use QTree instead of GTree

[Emilio Cota <cota@braap.org>]

On Wed, Jan 11, 2023 at 12:10:53 +0000, Daniel P. Berrangé wrote:
> On Tue, Jan 10, 2023 at 10:55:36PM -0500, Emilio Cota wrote:
> > Performance impact on linux-user:
> > - ~2% slowdown in spec06
> > - 1.05% slowdown in Nbench-int
> > - 4.51% slowdown in Nbench-fp
> 
> What do you get *before* applying this patch, if you just run
> linux-user with G_SLICE=always-malloc set ?
> 
> Also what libc impl were you testing with ? glibc or musl or something
> else ?

I've now done a few runs of nbench and the measurements I'm getting
are within noise. So I'd say no perf difference.
For SPEC I don't have time right now although I could do more runs
if you think they're needed.

I'm tempted for v2 to just remove these macrobenchmark numbers, since
with perf we can see that little time is spent on gtree anyway,
and at the moment I don't have time to do proper benchmarking.

Thanks,
		E.