[lworld] RFR: 8376135: [lworld] Add JIT support for NULLABLE_NON_ATOMIC_FLAT layout [v3]

[Quan Anh Mai]

On Tue, 27 Jan 2026 09:25:22 GMT, Quan Anh Mai <qamai at openjdk.org> wrote:

>> Hi,
>> 
>> This PR implements the support of `NULLABLE_NON_ATOMIC_FLAT` layout in the JITs. There is nothing to do in C2. In C1, I need to implement loading/storing nullable value of a non-atomic field.
>> 
>> The test `TestValueClasses` is failing with `-XX:-TieredCompilation`, which I suspect is due to the substitutability test handling padding bytes incorrectly.
>> 
>> Please kindly review what there are for now, thanks a lot.
>
> Quan Anh Mai has updated the pull request incrementally with two additional commits since the last revision:
> 
>  - no_null_marker
>  - rename null_free in InlineTypeNode::initialize_fields to has_null_marker

I have pushed a commit that should have fixed the remaining issues:

- The code to unpack value objects must zero all the field values if the object is `null` because the runtime requires a canonical representation of `null`.
- Remove `ciInlineKlass::must_be_atomic` in favor of a more consolidated `ciField::is_atomic`.
- The `null_free` value in `InlineTypeNode::initialize_fields` merely means that there is no projection for the null marker. This is the reason why we just pass `true` to the recursive call for subfields. I have renamed the parameter. There is also a variable hiding there. The expectation is that if the field is null free, and the holder object is null free, than the field value is not null. This is incorrect, given:

    ```
    class A {
        B v;
    }

    value class B {
        null-free C v;
    }
    ```

Then, `A.v.v` is nullable because `A.v` is nullable. To fix this, I remove the cast that informs the compiler that the oop is null-free.

- `Unsafe_newSpecialArray` should throw exception for `NULLABLE_NON_ATOMIC`. Currently, it sneaks past the check and causes a crash
- Fix some tests that did not expect the existence of `NULLABLE_NON_ATOMIC`.
- Problemlist `serviceability/jvmti/valhalla/HeapDump/HeapDump.java` : [JDK-8376461](https://bugs.openjdk.org/browse/JDK-8376461)
- Clean up as suggested by @TobiHartmann

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

PR Comment: https://git.openjdk.org/valhalla/pull/1970#issuecomment-3804108652