On Sat, Mar 19, 2022 at 8:27 PM Jeff Law <jeffreya...@gmail.com> wrote: > > > > On 2/2/2022 2:27 AM, Richard Biener wrote: > > On Tue, Feb 1, 2022 at 7:41 PM Aldy Hernandez <al...@redhat.com> wrote: > >> Ping > > I didn't quite get Jeffs comment, so I waited (sorry). I've meanwhile added > Sorry. IIRC the concern was whether or not we need to do something > special for irreducible regions. In that case which edge gets marked as > the backedge depends on graph traversal order. My suggestion was to use > the dominance relationship instead of edge flags.
Ah. Yes - it depends on what situation we are trying to detect here. If we are just trying to identify PHI arguments from "backedges" using dominance info is good. If we are trying to detect backedges from a CFG walk then we have to use a backedge DFS that matches our CFG walk, otherwise the backedges might not match ours for irreducible regions. Richard. > Jeff >
