This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Preliminary patch to fix some function specific option breakage


On Fri, Jul 25, 2008 at 06:04:10AM -0700, H.J. Lu wrote:
> On Fri, Jul 25, 2008 at 08:27:15AM -0400, Michael Meissner wrote:
> > I've been iterating with H. J. Lu about the breakage on the IA-64, which was
> > due to the cold attribute turning on options that that the backend had turned
> > off in OVERRIDE_OPTIONS.  This patch should fix the ia-64, and not enable
> > scheduling on the x86.
> > 
> > On the theory that there are other ports that have the same issue, this patch
> > also disables cold/hot attributes on ports that define OVERRIDE_OPTIONS but not
> > the new SECONDARY_OVERRIDE_OPTIONS until we can fix the ports.  At this point,
> > I just wrote the patch and did some preliminary checking on it, but I'm posting
> > it now to get feedback from the other port maintainers to see if it allows
> > their ports to bootstrap where previously it did not.
> 
> SECONDARY_OVERRIDE_OPTIONS is only used to modify optimization
> options. Can we use OVERRIDE_OPTIMIZATION_OPTIONS instead?

Or just fold the changes into OPTIMIZATION_OPTIONS and change the
documentation.  That probably is a better approach than creating yet another
target hook macro.

However, I can see that I will need to go over all of the ports with
OVERRIDE_OPTIONS and OPTIMIZATION_OPTIONS.  Sigh...

HJ, could you fire off a bootstrap build on the IA-64 with the patches (or some
varient) to see whether it fixes the current issues you have?

> > Index: gcc/config/i386/i386.h
> > ===================================================================
> > --- gcc/config/i386/i386.h	(revision 138117)
> > +++ gcc/config/i386/i386.h	(working copy)
> > @@ -476,6 +476,17 @@ enum calling_abi
> >  
> >  #define OVERRIDE_OPTIONS override_options (true)
> >  
> > +
> > +/* Like OVERRIDE_OPTIONS, but run after the optimize, hot, or cold attributes
> > +   are processed.  */
> > +
> > +#define SECONDARY_OVERRIDE_OPTIONS					\
> > +do {									\
> > +  if (!TARGET_SCHEDULE)							\
> > +    flag_schedule_insns_after_reload = flag_schedule_insns = 0;		\
> > +} while (0)
> > +
> > +
> 
> You can remove it from override_options in i386.c.

Yes.

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


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]