[llvm] [DAG][X86] Avoid slow masked-store folds on Zen (PR #214187)

[Simon Pilgrim via llvm-commits]

================
@@ -24402,6 +24323,7 @@ static SDValue foldToMaskedStore(StoreSDNode *Store, SelectionDAG &DAG,
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
 
   if (!TLI.isOperationLegalOrCustom(ISD::MSTORE, VT) ||
+      !TLI.shouldFoldStoreToMaskedStore(VT) ||
       !TLI.allowsMisalignedMemoryAccesses(VT, AddrSpace, Alignment))
----------------
RKSimon wrote:

An alternative would be for the allowsMisalignedMemoryAccesses callback to be tweaked to be aware of the memory opcode - either way I'm surprised we don't check the IsFast result from allowsMisalignedMemoryAccesses here.

https://github.com/llvm/llvm-project/pull/214187