[lworld] RFR: 8349162: [lworld] VM Flags controlling flattening need an update

[Vladimir Kozelkov]

On Fri, 31 Jan 2025 20:17:48 GMT, Frederic Parain <fparain at openjdk.org> wrote:

> This patch is mostly about VM flags update and internal renaming.
> 
> Renaming a few symbols to make them consistent with previous code cleanup:
>   - data_for_oop() -> payload_address()
>   - first_field_offset() -> payload_offset()
>   - LayoutKind::PAYLOAD -> LayoutKind::BUFFERED (it is only used to describe layout in buffered values, and it makes it an adjective like the other LayoutKind values)
>   
> VM flags controlling flattening:
>   - size based VM flags are removed:  FlatArrayElementMaxSize, InlineFieldMaxFlatSize. For now, FlatArrayElementMaxOops is still present because some compiler tests rely on it, but it will be removed eventually
>   - InlineArrayAtomicAccess has been removed, use AlwaysAtomicAccesses instead
>   - new VM flags to control the generation of the new layouts: NonAtomicValueFlattening, AtomicValueFlattening, NullableValueFlattening. Note that that those flags control the generation of the layouts, not the way they are used (see flags below)
>   - new VM flags to control use of flat layouts: UseFlatField, UseFlatArray
>   
> Example:
>   To enable flattening of nullable fields, the flag combination is -XX:+NullableValueFlattening -XX:+UseFlatField
> 
> The new VM flags controlling flattening are flags for testing or troubleshooting (they should probably be turned into DIAGNOSTIC flags before preview). They are not designed for performance tuning.
> 
> Feel free to propose new names for those flags if you think they are not explicit enough.
> 
> Thank you,
> 
> Fred

src/hotspot/share/oops/layoutKind.hpp line 1:

> 1: /*

LayoudKind - typo

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

PR Review Comment: https://git.openjdk.org/valhalla/pull/1345#discussion_r1938233148