RFR: 8308248: Revisit alignment of layout constants on 32-bit platforms [v3]

[Maurizio Cimadamore]

On Thu, 18 May 2023 09:28:14 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

>> The FFM API exposes layout constants for Java primitives. Among those there are constants for `JAVA_LONG` and `JAVA_DOUBLE`. Currently, the alignment of these layouts is set the same as their size (e.g. 8 bytes).
>> 
>> This is obviously correct on 64-bit platforms, but on 32-bit platform it is not, as such platforms cannot guarantee that doubles and longs will be always 64-bit aligned. This will also result in problems when trying to use e.g. `JAVA_DOUBLE` to model a C double for the linker API on 32-bit platforms.
>> 
>> For these reasons, it would be preferable to define the alignment of `JAVA_LONG` and `JAVA_DOUBLE` constants as `ADDRESS.byteSize()`. 
>> 
>> This patch rectifies alignment of those layout constants to reflect platform-dependent constraints. It also fixes the maximum alignment constraint supported by heap segments, so that it is 4 for long[] and double[] on 32-bit platforms.
>
> Maurizio Cimadamore has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Address review comments
>   Actually tweak alignment of ValueLayout.OfLong/ValueLayout.OfDouble

src/java.base/share/classes/jdk/internal/foreign/layout/ValueLayouts.java line 298:

> 296: 
> 297:         public static OfLong of(ByteOrder order) {
> 298:             return new OfLongImpl(order, ADDRESS_SIZE_BITS, Optional.empty());

Whoops - I forgot these all-important changes :-)

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

PR Review Comment: https://git.openjdk.org/jdk/pull/14007#discussion_r1197604492