[llvm] [AMDGPU] Convert 64-bit sra to 32-bit if shift amt >= 32 (PR #144421)

[Shilei Tian via llvm-commits]

================
@@ -4151,32 +4151,97 @@ SDValue AMDGPUTargetLowering::performShlCombine(SDNode *N,
 
 SDValue AMDGPUTargetLowering::performSraCombine(SDNode *N,
                                                 DAGCombinerInfo &DCI) const {
-  if (N->getValueType(0) != MVT::i64)
+  SDValue RHS = N->getOperand(1);
+  ConstantSDNode *CRHS = dyn_cast<ConstantSDNode>(RHS);
+  EVT VT = N->getValueType(0);
+  SDValue LHS = N->getOperand(0);
+  SelectionDAG &DAG = DCI.DAG;
+  SDLoc SL(N);
+
+  if (VT.getScalarType() != MVT::i64)
     return SDValue();
 
-  const ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N->getOperand(1));
-  if (!RHS)
+  // for C >= 32
----------------
shiltian wrote:

```suggestion
  // For C >= 32
```

https://github.com/llvm/llvm-project/pull/144421