[llvm] [IA][RISCV] Add support for vp.load/vp.store with shufflevector (PR #135445)

[Min-Yih Hsu via llvm-commits]

================
@@ -10585,7 +10610,19 @@ SDValue RISCVTargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
                   ContainerVT.getScalarSizeInBits();
     EVT VecTupTy = MVT::getRISCVVectorTupleVT(Sz, NF);
 
-    SDValue VL = DAG.getConstant(VT.getVectorNumElements(), DL, XLenVT);
+    // Masked: (pointer, mask, vl)
+    // Non-masked: (pointer, vl)
+    bool IsMasked = Op.getNumOperands() > 4;
+    SDValue VL = Op.getOperand(Op.getNumOperands() - 1);
+    SDValue Mask =
+        IsMasked ? Op.getOperand(3) : getAllOnesMask(ContainerVT, VL, DL, DAG);
+    MVT MaskVT = Mask.getSimpleValueType();
+    if (MaskVT.isFixedLengthVector()) {
----------------
mshockwave wrote:

oh that's a good point -- we always need to convert the mask to scalable vector (after we remove unmasked intrinsics). I just found that I didn't not remove the check in the follow-up PR, I'll do that shortly.

https://github.com/llvm/llvm-project/pull/135445