Throw IdentityException if operand of monent is a value type object

[hzongaro]

Previously, specifying an instance of a value type class on a monitorenter bytecode would result in an IllegalMonitorStateException; a java.lang.IdentityException must now be thrown instead.

This commit introduces a non-helper, <isIdentityObject>, that checks that the operand of a TR::monent instruction is an instance of an identity class rather than a value type class. If it is a value type class, it throws an IdentityException. The actual test is expanded to test the flags field of the object's class during the Tree Lowering optimization pass.

The result of <isIdentityObject> is tested on a ZEROCHK operation that is generated before the monent during IL generation.

This change includes only required functional changes that allow JIT-compiled code to detect whether it should throw an IdentityException. A planned improvement to Value Propagation will fold calls to <isIdentityObject> if the argument is recognized to be an instance of a value type class or recognized to be an instance of an identity type class at compile-time.

This change depends on eclipse-omr/omr#7674

Fixes #20984

[hzongaro]

I have marked this pull request as draft, as it depends on eclipse-omr/omr#7674. However, I believe it is complete and ready for review.

@a7ehuo, @0xdaryl, may I ask you to review this change?
@theresa-m, @hangshao0, may I ask you to review the VM-related portions of this change?

[a7ehuo]

An extra space in necessary. The

[hzongaro]

That was intentional. I always use two spaces after the period at the end of sentence, or after a colon. It's a style that was common using typewriters that printed using monospace fonts like Courier, and as we usually look at our code in Courier font, I prefer to continue to follow that convention.

[a7ehuo]

Thank you very much for the fix! Just some minor comments.

[a7ehuo]

iconst node doesn't seem indented correctly to align with the iloadi node. I also wonder if putting the comment at the iand node could make it clearer.

 * @verbatim
   n88n  PassThrough
   n99n    iand                     // Test whether class is an identity class
   n98n      iloadi  <isClassFlags>
   n97n        aloadi  <vft-symbol>
   n77n          aload x
   n96n      iconst 0x80000
 * @endverbatim

[a7ehuo]

I'm not sure if TR_identityException should be guarded with J9VM_OPT_VALHALLA_VALUE_TYPES in this file. First, the JIT code generally doesn't guard Valhalla specific change with build flags such as J9VM_OPT_VALHALLA_VALUE_TYPES. Secondly, anywhere else that references TR_identityException is not guarded with this flag either.

[hzongaro]

I think at one point, I had the entire definition of old_slow_jitThrowIdentityException enclosed in an

#if defined(J9VM_OPT_VALHALLA_VALUE_TYPES)

and so I conditionally compiled these uses of jitThrowIdentityException. But then I ended up with link errors on a couple of platforms when building without valhalla enabled, so I adjusted the definition of old_slow_jitThrowIdentityException to allow it to compile even without value types enabled. But I forgot to remove these ifdefs.

Yes, I think they can safely be removed. Thanks for pointing that out!

[a7ehuo]

Same comment as the above regarding to J9VM_OPT_VALHALLA_VALUE_TYPES.

[hzongaro]

Same as the previous comment - I think I had left this ifdef in because of the way I defined old_slow_jitThrowIdentityException in an earlier version of the code. I will remove it.