[patch] Fix SPEC slowdown from function specific branch

[Michael Meissner]

On Fri, Aug 08, 2008 at 06:28:23PM +0200, Michael Matz wrote:
> Hi,
> 
> On Fri, 8 Aug 2008, H.J. Lu wrote:
> 
> > > I sure hope not.  That flag can only be (re)set via tuning options, 
> > > i.e. using a different -mtune=xyz option.  I hope you aren't saying 
> > > that this is possible with function specific attributes because in 
> > > that case it
> > 
> > It was possible because we would call optimization_options for function 
> > specific attributes.
> 
> But that doesn't affect the current tune setting at all.  And only that 
> one can influence TARGET_SCHEDULE (the macro).  Perhaps you meant some 
> other option?
> 
> > > on the x86 backend does nearly nothing, while most things are decided 
> > > in override_options.  Still the former is used by the function 
> > > specific callbacks but the latter isn't.
> > >
> > > I wasn't able to really disentangle this, though, but did notice that 
> > > the patch I sent couldn't break anything which wouldn't be broken 
> > > right now anyway, so ...
> > >
> > 
> > It wasn't very pleasant. Please follow:
> > 
> > http://gcc.gnu.org/ml/gcc-patches/2008-07/msg01757.html
> 
> Yes, I've read over the thread.  But I don't see where the requirement for 
> moving the setting of both scheduling flags to optimization_options *on 
> x86* came from.  That thread dealt with ia64 breakages, introducing some 
> new target macro which then was replaced by this wrong move into 
> optimization_options.
> 
> If you could please provide an alternative patch that doesn't break 
> SPEC I'm happy to test it.

I would say short term install the patch.  This exposes an issue with function
specific options that I didn't think of (the target options reseting
optimization opitons, so I really need to combine saving/reseting both the
target options and optimization options at the same time).  I'm working on a
patch to fix the issue raised, but it will be a few days before I'm ready to
submit it, due to the fact I'm traveling.

-- 
Michael Meissner
email: gnu@the-meissners.org
http://www.the-meissners.org