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

[Benjamin Maxwell via llvm-commits]

================
@@ -5798,17 +5798,25 @@ InstructionCost AArch64TTIImpl::getPartialReductionCost(
       (!ST->isNeonAvailable() || !ST->hasDotProd()))
     return Invalid;
 
-  if ((Opcode != Instruction::Add && Opcode != Instruction::Sub) ||
+  if ((Opcode != Instruction::Add && Opcode != Instruction::Sub &&
+       Opcode != Instruction::FAdd) ||
       OpAExtend == TTI::PR_None)
     return Invalid;
 
+  // Floating-point partial reductions are invalid if `reassoc` and `contract`
+  // are not allowed.
+  if (Opcode == Instruction::FAdd &&
+      (!FMF->allowReassoc() || !FMF->allowContract()))
----------------
MacDue wrote:

Assumptions should be checked via assertions (that state the assumption) rather than deliberate errors (a `std::optional` error is not very helpful, and it's unclear if it was intentional or a bug). 

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