[llvm] AMDGPU: Use pattern to select instruction for intrinsic llvm.fptrunc.round (PR #105761)

[Matt Arsenault via llvm-commits]

================
@@ -6957,8 +6957,7 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
     SDValue Result;
     Result = DAG.getNode(
         ISD::FPTRUNC_ROUND, sdl, VT, getValue(I.getArgOperand(0)),
-        DAG.getTargetConstant((int)*RoundMode, sdl,
-                              TLI.getPointerTy(DAG.getDataLayout())));
+        DAG.getTargetConstant((int)*RoundMode, sdl, MVT::i32));
----------------
arsenm wrote:

The type matters less for globalisel since the representation of immarg is different, but yes this should just consistently be i32. And this should still be pre-committed as a separate change 

https://github.com/llvm/llvm-project/pull/105761