RFR: 8363989: AArch64: Add missing backend support of VectorAPI expand operation [v2]

[erifan]

On Thu, 4 Sep 2025 08:00:14 GMT, erifan <duke at openjdk.org> wrote:

>> test/hotspot/jtreg/compiler/vectorapi/VectorExpandTest.java line 48:
>> 
>>> 46:     static final VectorSpecies<Float> F_SPECIES = FloatVector.SPECIES_MAX;
>>> 47:     static final VectorSpecies<Long> L_SPECIES = LongVector.SPECIES_MAX;
>>> 48:     static final VectorSpecies<Double> D_SPECIES = DoubleVector.SPECIES_MAX;
>> 
>> Would it make sense to run these tests with various vector sizes?
>> Because it seems your algorithm depends on `vector_length_in_bytes` in the prefix sum algo.
>
> Since we already have correctness tests for `expand` on **all vector types** under `test/jdk/jdk/incubator/vector/`, such as https://github.com/openjdk/jdk/blob/986ecff5f9b16f1b41ff15ad94774d65f3a4631d/test/jdk/jdk/incubator/vector/Byte128VectorTests.java#L5375, this test primarily verifies that the expected IR is generated. So, I think this is sufficient?
> 
> I've tested this PR locally on a 128-bit SVE2 machine, a 256-bit SVE machine, and a 512-bit QEMU environment, and all tests passed.

By the way, `vector_length_in_bytes` doesn't affect the IR generation.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/26740#discussion_r2324223475