[PATCH] D58251: Extend "idempotent" atomicrmw optimizations to floating point

[Philip Reames via Phabricator via llvm-commits]

reames marked 2 inline comments as done.
reames added inline comments.


================
Comment at: lib/Transforms/InstCombine/InstCombineAtomicRMW.cpp:33
+      return false;
+    };
+
----------------
jfb wrote:
> @scanon for comment.
@scanon ping on the FP question only


================
Comment at: lib/Transforms/InstCombine/InstCombineAtomicRMW.cpp:90
+    RMWI.setOperation(AtomicRMWInst::FAdd);
+    RMWI.setOperand(1, ConstantFP::getNegativeZero(RMWI.getType()));
+    return &RMWI;
----------------
jfb wrote:
> I think you want to canonicalize to integer or with zero a well, and have casts. It's still just a load at the end of the day.
Can do so, but do you mind if I do that in a separate change just for risk reduction?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D58251/new/

https://reviews.llvm.org/D58251