To catch issues like JDK-8301491 and JDK-8302976 (and many others in the past) where a bug in an intrinsic leads to different results between interpreted and compiled code, we should improve our test coverage of intrinsified methods.

Ideally, we have a test that invokes each intrinsified method with the entire range of possible input values and compares the result between the compiled and interpreted versions. Given that this is not feasible for all intrinsics, we should at least add a test that generates random input and boundary values like max/min values, NaN etc.

The problem is that intrinsified methods often omit certain checks that are present in the Java version, so we need to make sure to either special case or call them with a wrapper method.