This is the mail archive of the gcc@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: Jump threading and loops


> Hello,
> 
> > Jump threading is one of optimizations that has potential of turning
> > loop into irreducible region.  THis appears to happen in several cases
> > in the gcc bootstrap and seems to be reason for warning on uninitialized
> > variable i in reload1.c for instance (I am going to look into this
> > deeper).  Would be possible for you to consider some strategy of not
> > doing the forwarding when it breaks loops?  It seems to me that loop
> > optimization is more important than threading in this particular case..
> 
> I also think it is a good idea to avoid this, and it it trivial to do --
> just mark back edges, and forbid threading over basic blocks that are
> entered by one.
We also may consider to do busy work and thread when all the entry edges
actually thread same way, but for that we need loop structure, I guess.

On my favorite topic of week, perhaps we can add an sanity check verifying
that functions with no irreducible region at input produce no
irreducible region at output?

Honza
> 
> Zdenek


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