[llvm] [AMDGPU] Implement llvm.lrint intrinsic lowering (PR #98931)

[Matt Arsenault via llvm-commits]

================
@@ -4326,6 +4326,16 @@ bool SelectionDAGLegalize::ExpandNode(SDNode *Node) {
     // targets where it is not needed.
     Results.push_back(Node->getOperand(0));
     break;
+  case ISD::LRINT:
+  case ISD::LLRINT: {
+    SDValue Arg = Node->getOperand(0);
+    EVT ArgVT = Arg.getValueType();
+    EVT ResVT = Node->getValueType(0);
+    SDLoc dl(Node);
+    SDValue RoundNode = DAG.getNode(ISD::FROUNDEVEN, dl, ArgVT, Arg);
----------------
arsenm wrote:

Ditto, just use ISD::FRINT until we consolidate these 

https://github.com/llvm/llvm-project/pull/98931