On Mon, 11 Aug 2025 14:06:41 GMT, Galder Zamarreño <gal...@openjdk.org> wrote:
>> src/hotspot/share/opto/superword.cpp line 1635:
>>
>>> 1633: } else if (VectorNode::is_convert_opcode(opc)) {
>>> 1634: retValue = VectorCastNode::implemented(opc, size,
>>> velt_basic_type(p0->in(1)), velt_basic_type(p0));
>>> 1635: } else if (VectorNode::is_reinterpret_opcode(opc)) {
>>
>> How does this affect `Op_ReinterpretHF2S` that is also in
>> `VectorNode::is_reinterpret_opcode`?
>> I'm afraid that we need to test this with hardware or Intel's SDE, to make
>> sure we have it running on a VM that actually supports Float16. Otherwise
>> these instructions may not be used, and hence not tested right.
>>
>> @galderz Can you run the relevant tests?
>
> Would you run specific tiers in those platforms? Just hotspot compiler? Or
> individual tests such as `ConvF2HFIdealizationTests` and
> `TestFloat16ScalarOperations`?
Honestly, I don't know, I'd have to do the research myself. Probably focusing
on the Float16 tests would be good enough. No other test would really use
Float16, so running anything else would not be that useful probably.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/26457#discussion_r2266903230
