Introduce mlir::hasEffect and refactor existing usage to use utility.
NFC.
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
I think this change alters the semantics, and won't work if the op has deallocate + some other effects on val. Those type of ops still qualify as dealloc ops.
Comment Actions
Yes, it does tighten the semantics. An op that has, say, a read effect and a dealloc effect on the same value will no longer be treated as a dealloc by this utility. We probably need to introduce hasEffect<....>(user, value) and use that.
Comment Actions
We probably need to introduce hasEffect<....>(user, value) and use that.
This is basically the code implemented in-line right now, right?