This is the mail archive of the 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: [PING] Loop-optimizer patches


> > > All these older patches are by Zdenek, but if approved I could take care
> > > of redoing bootstraps and regtests, and applying them:
> > >
> > >
> > > merging of equivalent invariants in rtl level invariant motion
> >
> > there is a minor bug in this version that I found during later tests,
> > I will try to prepare the fixed version soon.
> >
> > >
> > > reenables loop header predictor
> >
> > We have agreed with Honza that a better way is to move loop header
> > copying after the profiling pass, I will try to do this soon as well
> > (taking into account that I have about 5 patches I need to work
> > on in the queue now, "soon" may take some time :-( )  There already is some
> > profile updating code in tree-ssa-loop-ch, so it only needs to be
> > checked that it behaves sanely, and to run some benchmark to verify that
> > the pass reordering does not generate any unexpected regressions.
> Even with loop header copying after profiling, this patch to re-enable
> the predictor is still necessary.  Pat Haugen from IBM has an example
> where this predictor is necessary to get one of his bb-reorder fix to
> work.

could you send me the test?  This is impossible, therefore this is a bug
in profile updating code in loop header copying.


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