[llvm] [DAG] Combine `store + vselect` to `masked_store` (PR #145176)

[Abhishek Kaushik via llvm-commits]

================
@@ -22534,6 +22534,55 @@ SDValue DAGCombiner::visitATOMIC_STORE(SDNode *N) {
   return SDValue();
 }
 
+static SDValue foldToMaskedStore(StoreSDNode *Store, SelectionDAG &DAG,
+                                 const SDLoc &Dl) {
+  if (!Store->isSimple() || !ISD::isNormalStore(Store))
+    return SDValue();
+
+  SDValue StoredVal = Store->getValue();
+  SDValue StorePtr = Store->getBasePtr();
+  SDValue StoreOffset = Store->getOffset();
+  EVT VT = Store->getMemoryVT();
+  unsigned AddrSpace = Store->getAddressSpace();
----------------
abhishek-kaushik22 wrote:

I can add that test, but what targets have `ISD::MSTORE` legal and use different address spaces? (I tried AMDGPU and NVPTX but the `llvm.masked.store` intrinsic is converted to a bunch of jumps)

https://github.com/llvm/llvm-project/pull/145176