RFR: 8358521: Optimize vector operations by reassociating broadcasted inputs [v2]

[Jatin Bhateja]

On Tue, 10 Feb 2026 02:29:11 GMT, Xiaohong Gong <xgong at openjdk.org> wrote:

>> Jatin Bhateja has updated the pull request incrementally with one additional commit since the last revision:
>> 
>>   Review comments resolutions
>
> src/hotspot/share/opto/vectornode.cpp line 1317:
> 
>> 1315:   }
>> 1316:   return nullptr;
>> 1317: }
> 
> Reassociation can be a separate optimization for vector nodes to me. Do you think it's better to split it as a separate change following the broadcast optimization? We can add specific tests for it.

Reassociation are only applicable to integral vector operation, for vector API we only use fp-fast model semantics for reductions of floating point vector, I have now added tests to cover all possible transformations so that we can include both associated transforms in same patch.

> test/hotspot/jtreg/compiler/vectorapi/TestVectorBroadcastReassociations.java line 136:
> 
>> 134:      * ======================= */
>> 135: 
>> 136:     static final VectorSpecies<Long> LSP = LongVector.SPECIES_256;
> 
> Why not using the `SPECIES_PREFERRED` instead like the int species? The max vector size for AArch64 NEON (asimd) is 128-bit. We have to add another condition `MaxVectorSize >= 32` for following IR tests if using `SPECIES_256`.

Done

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

PR Review Comment: https://git.openjdk.org/jdk/pull/25617#discussion_r2785891612
PR Review Comment: https://git.openjdk.org/jdk/pull/25617#discussion_r2785892467