[jdk18] RFR: 8278267: ARM32: several vector test failures for ASHR [v3]

[Hao Sun]

On Tue, 4 Jan 2022 04:27:08 GMT, Dean Long <dlong at openjdk.org> wrote:

>> I think it's safe to add them to avoid mis-optimizations at IR level.
>> 
>> Per my understanding, two ShiftV nodes with the same vector elements but with different `_is_var_shift` field values, are two different nodes, and they cannot be merged/treated as the same node.
>
> That makes sense.  Is it possible to write a test for that?

Generally speaking `hash()` and `cmp()` need be overrided if there are new field members. E.g., LoadLNode, VectorMaskCmpNode, VectorTestNode, etc. 

It could happen theoretically. But I'm afraid it's not easy to design a test case as it might involve some internal optimizations in a crafty way to construct such two kinds of **ShiftV** nodes.

-------------

PR: https://git.openjdk.java.net/jdk18/pull/41