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: [tree-ssa] PHI_ARG_CAPACITY != block degree


Hello,

> > *************** rewrite_block (basic_block bb)
> > *** 783,789 ****
> >   	  tree currdef;
> >   
> >   	  /* Ignore PHI nodes that have already been renamed.  */
> > ! 	  if (PHI_NUM_ARGS (phi) == PHI_ARG_CAPACITY (phi))
> >   	    continue;
> >   
> Oh, bugger.  I inadvertently left this behind.  This is wrong now that
> we have a proper way of detecting whether a variable needs to be
> renamed.  In here we should have a test similar to that done in
> tree-ssa.c:prepare_operand_for_rename.  We should be testing whether
> PHI_RESULT (phi) is present in vars_to_rename.
> 
> In any case, could you add a test case where you found this to be a
> problem?  I'm curious about what kinds of failures does this generate.

it only happened to me on tree-ssa-cfg-branch after merge from tree-ssa.
It probably cannot be reproduced on tree-ssa, since the only pass that
uses add_phi_arg is ssa_pre and rewrite_into_ssa is not called after it.

Zdenek


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