In low-level benchmarking, we sometimes resort to non-inlineable "sink" methods to escape dead-code elimination, like this:

    @Benchmark
    public void test() {
        doNothing(obj);
    }

    @CompilerControl(CompilerControl.Mode.DONT_INLINE)
    public void doNothing(Object obj) {
        // deliberately do nothing
    }

The performance of this method is very important, since we usually deal with nanosecond-scale benchmarks. Ideally, the generated code should contain a "ret" right away.
However, the generated code for doNothing contains prolog followed immediately with epilog:

                  [Verified Entry Point]
 10.93%    6.25%    0x00007f39f415fd80: mov    %eax,-0x14000(%rsp)
  3.76%    3.03%    0x00007f39f415fd87: push   %rbp
  1.92%    1.97%    0x00007f39f415fd88: sub    $0x30,%rsp
 10.42%   10.64%    0x00007f39f415fd8c: add    $0x30,%rsp
  2.88%    3.03%    0x00007f39f415fd90: pop    %rbp
 25.45%   31.68%    0x00007f39f415fd91: test   %eax,0x15df8369(%rip)        # 0x00007f3a09f58100
                                                                  ;   {poll_return}
  0.57%    0.47%    0x00007f39f415fd97: retq   

It seems that at least RSP operations are redundant, as well as saving/restoring RBP.
It would be interesting to see if we can remove these redundant ops, e.g.:

 *) Peephole MachPrologNode -> MachEpilogNode out completely;
 *) Macro-expand MachProlog/EpilogNode into the individual ops, and then peephole (sub $const, %reg) -> (add $const, %reg) and (push %reg) -> (pop %reg);
 *) Massage frame_size_in_bytes() so that it is zero for empty method;

Benchmark:
 http://cr.openjdk.java.net/~shade/8130398/EmptyMethod.java

Runnable JAR:
 http://cr.openjdk.java.net/~shade/8130398/benchmarks.jar

Output and disasssembly:
 http://cr.openjdk.java.net/~shade/8130398/perfasm.out