Currently, the IR-framework has the limitation that we can only specify the count of IR nodes as an absolute number or relative to an absolute number. For many tests this is limiting, since we only want to express that from one phase to the next the number of some IR node has decreased by one since the optimization we are testing succeeded. Often there will be unrelated nodes not touched by the optimization that we currently have to include in the count to make the test accurate. However, this sometimes leads to the test failing on different platforms or with different flags that will generate more or less nodes of interest than specified in the test. One example of this is JDK-8361040.

Thus, @IR annotations should allow specifying a difference to other IR annotations on a test. Either by relative position of the annotations or by some kind of explicit id.