> 2011/4/29 Sharad Singhai (????????? ???????????????) <sing...@google.com>: > > On Thu, Apr 28, 2011 at 4:38 PM, Richard Guenther > > <richard.guent...@gmail.com> wrote: > >> > >> On Fri, Apr 29, 2011 at 1:34 AM, Xinliang David Li <davi...@google.com> > >> wrote: > >> > Sharad can provide some some performance data -- we have seen up to 2% > >> > degradation to with tracer turned on for one of google's most > >> > important program. Perhaps Sharad can collect some SPEC numbers. > >> > > >> > I agree a better approach should be to fix the problem in tracer > >> > instead of turning it off in trunk. > >> > >> Esp. not turning it off for profile-use only where it should have the most > >> precise input. > >> > > > > Yes. As I explained, I am not proposing to turn -ftracer off in the trunk. > > Ah, sorry for the confusion on my side. If you can back up the change > with SPEC performance numbers it's ok for trunk as well. But indeed > coverage of -ftracer will be zero then, and I wonder what it is useful > for if it doesn't even do good with FDO ...
Tracer definitely did measurable SPEC speedups at a time it was implemented. So we are most likely running into some bug, like misupdated profile or tracer confusing something more important. If you provide me some testcase, I can try to debug it. Honza
