[PATCH] D158273: [NewGVN] Don't ignore poison in cyclic PHIs

[Nuno Lopes via Phabricator via llvm-commits]

nlopes added inline comments.


================
Comment at: llvm/test/Transforms/NewGVN/cycle_poison.ll:8
+; CHECK:       1:
+; CHECK-NEXT:    [[TMP2:%.*]] = phi i32 [ poison, [[TMP1]] ], [ undef, [[TMP0:%.*]] ]
+; CHECK-NEXT:    br label [[TMP1]]
----------------
this could be folded to just undef.


================
Comment at: llvm/test/Transforms/NewGVN/cycle_poison.ll:16
+  %3 = lshr i32 %2, 973496514
+  br label %1
+}
----------------
can the test case have a terminating loop pls?
non-terminating loops are evil for verification and semantics-wise.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D158273/new/

https://reviews.llvm.org/D158273