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: [tree-ssa] RFC: Making control flow more explicit


Hello,

> > I am not really sure what is the exact meaning of the parent.  Can't you
> > run into problems in cases like
> > 
> > if (c1)
> >   goto bla;
> > 
> > if (c2)
> >   {
> >     something;
> >     bla:
> >     something_else;
> >   }
> > ?
> > 
> You shouldn't.  What would be the dead statement here?

sorry, I did not really think about what I write into the example.
What I wanted to point out is that with presence of gotos "parent"
loses sense, so it has to be handled somehow anyway (or if it is not,
something is wrong).

> > Anyway, losing concept of parent relationship seems to be relatively
> > small cost in comparison to other gains.
> > 
> The only reason why we use the parent relation is to avoid spending time
> building the control dependence information.  That can be costly.  IIRC,
> Jeff Law had done some experiments with that, but they don't seem to be
> in CVS.

It should not be; it is just some playing with dominators and
post-dominators, isn't it?  Note that the problem might be just that
our current implementation of dominator tree is not very efficient
(although using et-trees seemed like a cool idea to me originally :-(

Zdenek


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