This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: bug in dominance algorithm
- To: Jan Hubicka <jh at suse dot cz>
- Subject: Re: bug in dominance algorithm
- From: Michael Matz <matzmich at cs dot tu-berlin dot de>
- Date: Fri, 15 Jun 2001 07:59:10 +0200 (MET DST)
- cc: <gcc at gcc dot gnu dot org>
On Fri, 15 Jun 2001, I wrote:
> Hi,
>
[... and more stuff ...]
Ugh. I was already answering the opening post, but only then got aware,
that there also was a thread following ;-)
So you already noticed, that the current behaviour is correct, but could
become better, if the simulation of the old behaviour (with the bitvector
approach) would be relaxed. I.e. the conclusion from "A postdoms B" to "A
is reachable from B" already is true internally, but this is
sometimes screwed by that simulation of old behaviour.
Nevertheless note, that we _never_ get this any better for blocks without
_any_ successors. For those postdoms really have no sane meaning (for
blocks without path to EXIT, but with successors there are definitions
possible which seem sane, e.g. the postdom of such node is the postdom of
that node interpreted in that part-tree of the forest containing this
node, and this is, what is currently done, but nodes without successors
are the roots of there trees, and tree-roots have no dominator).
Ciao,
Michael.