[PATCH] D45336: Apply accumulator to fadd/fmul experimental vector reductions (PR36734)

[Gonzalo BG via Phabricator via llvm-commits]

gnzlbg added inline comments.


================
Comment at: lib/Transforms/Utils/LoopUtils.cpp:1545
+    if (Op != Instruction::ICmp && Op != Instruction::FCmp) {
+      // Floating point operations had to be 'fast' to enable the reduction.
+      Result = addFastMathFlag(
----------------
Could you add a note explaining why this is the case?

If this isn't the case adding fast math flags is going to have a lot of unexpected side-effects for the users. Is there a way to verify this or assert this?


Repository:
  rL LLVM

https://reviews.llvm.org/D45336