[llvm] [AMDGPU] Do not treat bitcast across FP types as canonicality-preserving (PR #203560)

[Arseniy Obolenskiy via llvm-commits]

================
@@ -15643,8 +15643,17 @@ SDValue SITargetLowering::performRcpCombine(SDNode *N,
 }
 
 bool SITargetLowering::isCanonicalized(SelectionDAG &DAG, SDValue Op,
-                                       SDNodeFlags UserFlags,
-                                       unsigned MaxDepth) const {
+                                       SDNodeFlags UserFlags, unsigned MaxDepth,
+                                       EVT QueryVT) const {
+  EVT VT = Op.getValueType();
+  if (VT.isFloatingPoint()) {
----------------
aobolensk wrote:

Added assert for the case when there is no `QueryVT` (original overload). As for the new one, it is not always true and it is a subject of checking

https://github.com/llvm/llvm-project/pull/203560