RFR: 8371187: [BigEndian Platforms] Vector lane reversal error [v3]

[Martin Doerr]

On Thu, 22 Jan 2026 06:29:26 GMT, Varada M <varadam at openjdk.org> wrote:

>> src/jdk.incubator.vector/share/classes/jdk/incubator/vector/AbstractVector.java line 185:
>> 
>>> 183:         // NOTE:  This assumes that convert0('X')
>>> 184:         // respects REGISTER_ENDIAN order.
>>> 185:         return convert0('X', vspecies().withLanes(laneType)).maybeSwapOnConverted(java.nio.ByteOrder.nativeOrder(), vspecies());
>> 
>> Would `swapIfNeeded` be a better name?
>> I think we could determine the ByteOrder where it is used. Then, we don't have to pass it any more.
>
> I agree, looks more clean. I have renamed  and made the suggested changes

Thanks! My idea was to move `java.nio.ByteOrder.nativeOrder()` to `subLanesToSwap` where the only usage is AFAICS.

>> src/jdk.incubator.vector/share/classes/jdk/incubator/vector/AbstractVector.java line 217:
>> 
>>> 215:         int sBytes = srcSpecies.elementSize();
>>> 216:         int tBytes = vspecies().elementSize();
>>> 217:         if (sBytes == tBytes || (sBytes % tBytes) != 0) {
>> 
>> What do we do if it's not divisible? Should we better throw an exception?
>
> I have tried adding the exception but many tests failed!
> what I understood is asVectorRawTemplate is called in many cases which eventually calls swapIfNeeded and exception is thrown for the widening cases. When we reinterpret smaller elements into larger ones we don't need to swap elements so we could opt out.

Ok, a comment would be helpful.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/28425#discussion_r2716289852
PR Review Comment: https://git.openjdk.org/jdk/pull/28425#discussion_r2716283133