[llvm] [AMDGPU] Fix crash on strict fptrunc to bf16 (PR #208037)

[Arseniy Obolenskiy via llvm-commits]

================
@@ -8694,6 +8696,10 @@ SDValue SITargetLowering::lowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const {
   // hardware f32 -> bf16 instruction.
   EVT F32VT = SrcVT.changeElementType(*DAG.getContext(), MVT::f32);
   SDValue Rod = expandRoundInexactToOdd(F32VT, Src, DL, DAG);
+  if (IsStrict)
----------------
aobolensk wrote:

Added

https://github.com/llvm/llvm-project/pull/208037