Remove Q type checks in Valhalla code

[hangshao0]

Currently many places in the VM rely on Q type check for the null-restricted value type. The Q type is going to be removed in the future. L type will be used for all value types, no matter it is null-restricted or not. For the fields, the Q type is going to be replaced with null-restricted attribute,

Also the keyword "primitive" is going to be removed.

The null restricted fields will be treated the same as Q type fields. For now, we need to add null-restricted attribute checks into the places where we are doing Q type field checks. After javac is updated, we can remove the Q type checks completely.

Some of our testing code is generating Q types and using "primitive", they need to be updated as well.

[hangshao0]

Also the keyword "primitive" is going to be removed.

When "primitive" is removed, J9ROMCLASS_IS_PRIMITIVE_VALUE_TYPE() will always be false. With the new javac, we can use J9ROMCLASS_IS_VALUE() instead.

Since only null-restricted field can be potentially flattened, we can add a check of null restricted field (J9FieldFlagIsNullRestricted) in J9_IS_FIELD_FLATTENED().

[dmitripivkine]

I am not sure right away, but I can check what is used by GC for recognition of VT "leaf object" (does not introduce new object references). We are going to need the way to make "leaf object" detection for VT objects in order to have GC optimization running.

[hangshao0]

Also the keyword "primitive" is going to be removed.

When "primitive" is removed, J9ROMCLASS_IS_PRIMITIVE_VALUE_TYPE() will always be false. With the new javac, we can use J9ROMCLASS_IS_VALUE() instead.

Not sure about the "we can use J9ROMCLASS_IS_VALUE() instead" part. There is a ACC_NON_ATOMIC flag introduced in the ImplicitCreation Attribute, do you know if that is introduced to the replace "primitive" keyword ? @TobiAjila

[theresa-m]

Before removing Q type checks I'm modifying existing Valhalla functional tests to be compatible as much as possible with the lw5 compiler.

edit: tests in src_lw5 compile with the latest version of lw5. Additional vm support is needed for them to pass.

No changes are needed for these files, they will remain in src:

• ValueTypeOptTests
• ValhallaAttributeTests
• ValhallaAttributeGenerator

These tests have been split into src_qtypes and src_lw5 and compile with lw5 commit dc715159feda32945d5c2351ede9c6be04a64d6f:

• NullRestrictedTypeOptTests
• ValueTypeSystemArraycopyTests
• ValueTypeTests part 1
• ValueTypeTests part 2
• ValueTypeTests part 3
• ValueTypeTests part 4 - includes ValueTypeGenerator split
• ValueTypeTests part 5
• ValueTypeArrayTests
• ValueTypeTestClasses
• DDRValueTypeTest
• DDRBackfillLayoutTest
• ValueTypeUnsafeTests
• aconst_init,withfield, removal

[theresa-m]

This is a new list of some of the vm tasks to be done once javac is updated for null restricted types. Aside from removing qtype and primitive references we will also need to:

• Enable null restricted (lw5) tests, remove qtype tests, and merge any vm changes that are not compatible in the current code base. These are being tracked here Valhalla: test updates for lw5 #18558

Edit: given that q is removed in the current javac and null restricted support is not yet there tests in src_qtypes have been updated and no longer have dependencies on q.

This pr can be closed once all references to q and primitive are removed.

[theresa-m]

#19799 will remove all remaining uses of 'q' and 'qtype' from OpenJ9. @a7ehuo is going to look at removing any q references from OMR.

The next step to close out this issue will be removing references to the old primitive flag for value types.

[a7ehuo]

@a7ehuo is going to look at removing any q references from OMR.

Opened eclipse/omr#7399

[theresa-m]

The following q and primitive reference are still outstanding before this issue can be closed:

ReferenceType naming

• Rename IS_REF_OR_VAL_SIGNATURE to IS_CLASS_SIGNATURE #20160
• Remove old ReferenceType error handling for value types #20162

J9_IS_J9CLASS_PRIMITIVE_VALUETYPE

• checkcast methods in cnathelp.cpp jitCheckCastForArrayStore should receive array class instead of base class #20184
• some uses to be updated in Unsafe api updates Update src_qtypes ValueTypeUnsafeTests and Unsafe library #19677
• use in J9::ClassEnv::isPrimitiveValueTypeClass

J9::ClassEnv::isPrimitiveValueTypeClass

• remaining uses are in compiler code

TR_J9VMBase::checkArrayCompClassPrimitiveValueType

• Remaining uses in compiler code

J9ClassIsPrimitiveValueType flag

• Remove DDR uses of J9ClassIsPrimitiveValueType #20182
• Remaining uses are either in comments or outdated methods and macros (J9_IS_J9CLASS_PRIMITIVE_VALUETYPE, J9::ClassEnv::isPrimitiveValueTypeClass, TR_J9VMBase::checkArrayCompClassPrimitiveValueType)

fyi @a7ehuo I'm sure you're already aware of these to be removed when it makes sense to do so, just wanted to summarize the remaining work that I'm aware of.