[llvm] [AMDGPU] Elide bitcast fold i64 imm to build_vector (PR #154115)

[Matt Arsenault via llvm-commits]

================
@@ -5296,6 +5296,13 @@ SDValue AMDGPUTargetLowering::performRcpCombine(SDNode *N,
   return DCI.DAG.getConstantFP(One / Val, SDLoc(N), N->getValueType(0));
 }
 
+bool AMDGPUTargetLowering::canMov64bImm(uint64_t Val, SelectionDAG &DAG) const {
----------------
arsenm wrote:

Name could use work, and maybe this should work directly in terms of the original SDNode. Maybe isInt64ImmLegal? 

https://github.com/llvm/llvm-project/pull/154115