[llvm] Add LoopVectorizer support for `llvm.vector.partial.reduce.fadd` (PR #163975)

[Sander de Smalen via llvm-commits]

================
@@ -2827,7 +2827,9 @@ InstructionCost VPExpressionRecipe::computeCost(ElementCount VF,
                      TargetTransformInfo::getPartialReductionExtendKind(
                          ExtR->getOpcode()),
                      TargetTransformInfo::PR_None, std::nullopt, Ctx.CostKind,
-                     std::nullopt)
+                     RedTy->isFloatingPointTy()
+                         ? std::optional{ExtR->getFastMathFlags()}
+                         : std::nullopt)
----------------
sdesmalen-arm wrote:

My comment on passing fast-math flags here was actually related to my other comment on handling the simple `reduce.fadd(fpext(..))  `, so this change can be moved to the follow-up patch as well.

https://github.com/llvm/llvm-project/pull/163975