[llvm] [X86] Lowering of load atomic vector of float via cast (PR #117189)

[Matt Arsenault via llvm-commits]

================
@@ -31158,6 +31158,14 @@ bool X86TargetLowering::needsCmpXchgNb(Type *MemType) const {
   return false;
 }
 
+TargetLoweringBase::AtomicExpansionKind
+X86TargetLowering::shouldCastAtomicLoadInIR(LoadInst *LI) const {
+  if (const auto VT = dyn_cast<VectorType>(LI->getType()))
+    if (VT->getElementType()->isFloatingPointTy())
----------------
arsenm wrote:

Should move this into the default implementation. Targets can opt out as the hook is removed 

https://github.com/llvm/llvm-project/pull/117189