[llvm] [LoopUnroll] Introduce parallel accumulators when unrolling FP reductions. (PR #166630)

[Florian Hahn via llvm-commits]

================
@@ -1256,14 +1257,19 @@ llvm::canParallelizeReductionWhenUnrolling(PHINode &Phi, Loop *L,
     return std::nullopt;
   RecurKind RK = RdxDesc.getRecurrenceKind();
   // Skip unsupported reductions.
-  // TODO: Handle additional reductions, including FP and min-max
-  // reductions.
-  if (!RecurrenceDescriptor::isIntegerRecurrenceKind(RK) ||
+  // TODO: Handle additional reductions, including min-max reductions.
+  if (!(RecurrenceDescriptor::isIntegerRecurrenceKind(RK) ||
+        RecurrenceDescriptor::isFloatingPointRecurrenceKind(RK)) ||
       RecurrenceDescriptor::isAnyOfRecurrenceKind(RK) ||
       RecurrenceDescriptor::isFindIVRecurrenceKind(RK) ||
       RecurrenceDescriptor::isMinMaxRecurrenceKind(RK))
     return std::nullopt;
 
+  if (RecurrenceDescriptor::isFloatingPointRecurrenceKind(RK)) {
+    if (!RdxDesc.getFastMathFlags().allowReassoc())
+      return std::nullopt;
+  }
----------------
fhahn wrote:

Do we need the check here? I *think* the recurrence descriptor analysis should take care of checking the required fast-math flags (in `RecurrenceDescriptor::isRecurrenceInstr`), and return RecurKind::None, if the required flags are missing.

https://github.com/llvm/llvm-project/pull/166630