| commit | e52b850ba2095181e43b9e4e9236d4d5794a64d6 | |
| author | Brett Simmers <bsimmers@fb.com> | |
| Sun, 18 Oct 2015 22:29:40 +0000 (18 15:29 -0700) | ||
| committer | hhvm-bot <hhvm-bot@fb.com> | |
| Sun, 18 Oct 2015 23:00:28 +0000 (18 16:00 -0700) | ||
| tree | 5c1c97f00c3d239d6dd0956fed60bc0b006e9ee4 | treesnapshot (tar.gz zip) |
| parent | 1c9f7452a2cc22af723066a261622b044221643b | commitdiff |
Fix bug in ptr_subtype()
Summary: It used to think Unk was a subtype of Ref, because remove_ref(Ref) ==
Unk. Special case Unk before we get to the remove_ref() call, like other users
of remove_ref() do.
This was causing us to think that PtrToGen - PtrToRefGen was Bottom, messing up
alias analysis and causing load/store-elim to misbehave.
Reviewed By: alexmalyshev
Differential Revision: D2554135
fb-gh-sync-id: 0637ff0ae17002dbac260bd67b5718f586098d61
Summary: It used to think Unk was a subtype of Ref, because remove_ref(Ref) ==
Unk. Special case Unk before we get to the remove_ref() call, like other users
of remove_ref() do.
This was causing us to think that PtrToGen - PtrToRefGen was Bottom, messing up
alias analysis and causing load/store-elim to misbehave.
Reviewed By: alexmalyshev
Differential Revision: D2554135
fb-gh-sync-id: 0637ff0ae17002dbac260bd67b5718f586098d61