[PATCH] D28875: AArch64LoadStoreOptimizer: Update kill flags when merging stores

[Chad Rosier via Phabricator via llvm-commits]

mcrosier added a comment.

Given we want to deprecate the use of kill flags and AFAIK there are no other passes that use these flags after the load/store optimizer, I'm okay with the lossiness of this solution.



================
Comment at: lib/Target/AArch64/AArch64LoadStoreOptimizer.cpp:707
+      unsigned Reg = getLdStRegOp(*I).getReg();
+      for (MachineInstr &MI : make_range(NextI, Paired))
+        MI.clearRegisterKills(Reg, TRI);
----------------
If the two instructions being paired happen to be adjacent to one another, NextI will point to the instruction after 'Paired'.  Is this going to behave as expected with make_range()?


Repository:
  rL LLVM

https://reviews.llvm.org/D28875