On Tue, Dec 7, 2021 at 3:16 PM Andrew MacLeod <amacl...@redhat.com> wrote: > > On 12/7/21 02:12, Richard Biener wrote: > > On Mon, Dec 6, 2021 at 7:39 PM Andrew MacLeod <amacl...@redhat.com> wrote: > >> On > >> Well, its only does the fill now when there is range info located on an > >> outgoing edge of the dominator. Its still used, just on a much smaller > >> portion of the graph. > >> > >> We could do even better if we knew whether one edge was involved. ie > >> > >> bb3: > >> a = foo() > >> if (a > 4) > >> blah1; bb4 > >> else > >> blah2; bb5 > >> bb6: > >> = a; > >> > >> The use of a in bb6 will look to bb3 as the dominator, but if it knew > >> that both edges are used in that dominance relation (ie, neither > >> outgoing edge dominates bb6), it wouldn't have to calculate a range for > >> 'a'. > >> > >> But as it stands, it cant really tell the above situation from: > >> > >> bb3: > >> a = foo() > >> if (a > 4) > >> = a bb4 > >> > >> In this case, only the one edge is used, and we DO need to calculate a > >> range for a. The edge from 3->4 does dominate bb4 > >> > >> In both cases, bb3 is the dominator, and bb3 can generate an outgoing > >> range, but only in one case do we need to calculate a range. > >> > >> What would be really useful would be to be able to tell if an edge > >> dominates a block :-) > >> > >> I have thoughts on this for next release, and may overhaul the cache... > >> but I don't think there is any such facility in the dominator subsystem? > > Well, dominance of an edge is dominance of the edge destination if the > > destination has only a single non-backedge. If the destination has more > > than one non-backedge then the edge does not dominate anything. So > > to generally answer dominance queries for edges you have to have > > backedges marked. > > > Are back edges not always marked if dominance information is available?
no > OR does one need to do something to ensure they are up to date. mark_dfs_back_edges () > And if an edge is marked as DFS_BACK, is that guaranteed to be true > always, regardless of dominance info, assuming no CFG manipulation is > going on? well, it's guaranteed after the above call. CFG manipulations do not keep the flag up-to-date. Richard. > > Thanks > > Andrew >
