[PATCH] D16263: [CodeGenPrepare] Also consider metadata uses

[Hal Finkel via llvm-commits]

hfinkel added a comment.

In http://reviews.llvm.org/D16263#328811, @loladiro wrote:

> As far as I can tell, DWARF values are implicitly truncated. The problem is not the final DWARF representation but our IR representation. Passes like SROA rely on the sizing information of IR and debug information to match


Why does it make this assumption? Do other passes that widen variables (e.g. indvarsimplify) also trigger this problem?

> , which is what the upcoming Verifier stuff is supposed to ensure. Of course, SROA will never run after CodeGenPrepare, so it's not an actual problem, but it would be very annoying to not be able to have the Verifier check this invariant, because of this one thing. I do think it might be reasonable to have an (`OP_trunc size`) for the DIExpression. It could get ignored by the backend, but would allow us to do proper verification.





Repository:
  rL LLVM

http://reviews.llvm.org/D16263