https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126110
--- Comment #8 from Drea Pinski <pinskia at gcc dot gnu.org> --- (In reply to Drea Pinski from comment #7) > Created attachment 64945 [details] > Patch which kinda of works > > This patch works but it is very very hacky as it assumes query is always a > gimple_ranger (which it is currently). > > Basically what it does is resets the ranger via calling the deconstructor > and then calling an inplacement new. > And then updates the range for the groups. > > Basically the range for set_6 is not cached so VRP does the correct thing > and sets it to non-null as expected. I have not looked why VRP can't do what > DOM does when it comes to the cache though. Oh wait DOM does not use phi_analyzer, only VRP. So there is no cache at that point.
