Valhalla implementation required for latest OpenJDK/Valhalla updates

[JasonFengJ9]

With latest OpenJDK/Valhalla updates, following OpenJ9 Valhalla implementations are required:

• jdk.internal.misc.Unsafe

  • public final Object getAndSetReference(Object obj, long offset, Class<?> valueType, Object value)
  • public final Object getAndSetReferenceAcquire(Object obj, long offset, Class<?> valueType, Object value)
  • public final Object getAndSetReferenceRelease(Object obj, long offset, Class<?> valueType, Object value)
  • public void notifyStrictStaticAccess(Class<?> clz, long staticFieldOffset, boolean writing)

• JVM_CopyOfSpecialArray

• JVM_IsAtomicArray

• JVM_NewNullRestrictedNonAtomicArray

  • the initialValue which can't be null

• JVM_NewNullRestrictedAtomicArray

  • support an additional parameter initialValue which can't be null

• Resolve test failures

  • expecting implicitly constructible classes as inputs

    java.lang.IllegalArgumentException
           at java.base/jdk.internal.value.ValueClass.newNullRestrictedAtomicArray(Native Method)
  

  • gc test failures due to lack of support for newNullRestrictedAtomicArray. Current workaround is to keep the manual initialization

     java.lang.AssertionError: expected [-1114130] but found [0]
           at org.testng.Assert.fail(Assert.java:96)
    

Related to

• merging latest OpenJDK contents - https://github.com/JasonFengJ9/openj9-openjdk-jdk.valuetypes/tree/mergetmp, (Valhalla) Merge master HEAD into openj9-staging ibmruntimes/openj9-openjdk-jdk.valuetypes#20
• Remove Valhalla workaround for OpenJCEPlus JDK24 references
• Valhalla support to merge latest OpenJDK contents #22641

FYI @hangshao0 @theresa-m @tajila

[theresa-m]

Some notes on test updates #22641 (review)

newNullRestrictedAtomicArray must support an extra field initVal.

Also both newNullRestrictedAtomicArray and newNullRestrictedNonAtomicArray should not allow the initVal to be null.

[tajila]

Can we add stubs for these @JasonFengJ9 or do we need a full implementation?

[theresa-m]

A full implementation for JVM_NewNullRestrictedAtomicArray implementation is at least needed and waiting on strict field support to be able to replace implicit creation. I've also seen the Unsafe methods causing some OpenJDK tests to fail.