[PATCH] D140906: [DebugInfo] Replace UndefValue with PoisonValue in AssignmentTrackingAnalysis

[Scott Linder via Phabricator via llvm-commits]

scott.linder added a comment.

Code LGTM, thank you! One small request is to move some comments around and update `llvm/docs/AssignmentTracking.md` to reflect the change.

Something like:

  The first three parameters look and behave like an llvm.dbg.value. ID is a reference to a store (see next section). Address is the destination address of the store and it is modified by AddressExpression. An empty/undef/poison Address represents <...>.



================
Comment at: llvm/include/llvm/IR/IntrinsicInst.h:451
   void setValue(Value *V);
+  bool isAddressKillLocation();
   /// \name Casting methods
----------------
I vote for a short doc comment mentioning that this is the only correct way to ask a `DbgAssignIntrinsic` whether it "kills" the address-based location.


================
Comment at: llvm/lib/CodeGen/AssignmentTrackingAnalysis.cpp:1478
+    if (DAI.isAddressKillLocation()) {
       // Address may be undef to indicate that although the store does take
       // place, this part of the original store has been elided.
----------------
Somewhat stale reference to `undef` directly here. Could the comment just be folded into a general one on `isAddressKillLocation` itself?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D140906/new/

https://reviews.llvm.org/D140906