On Tue, 5 Aug 2025 06:32:23 GMT, Quan Anh Mai <qa...@openjdk.org> wrote:
>>> VectorNode::is_reinterpret_opcode returns true for Op_ReinterpretHF2S and
>>> Op_ReinterpretS2HF, which are very similar to the nodes in this PR, can you
>>> add these nodes to that method instead?
>>
>> You're suggesting to modify `is_reinterpret_opcode` to be like this, and
>> call that instead of `is_move_opcode`, right?
>>
>>
>> bool VectorNode::is_reinterpret_opcode(int opc) {
>> switch (opc) {
>> case Op_ReinterpretHF2S:
>> case Op_ReinterpretS2HF:
>> case Op_MoveF2I:
>> case Op_MoveD2L:
>> case Op_MoveL2D:
>> case Op_MoveI2F:
>> return true;
>> default:
>> return false;
>> }
>> }
>
>> You're suggesting to modify `is_reinterpret_opcode` to be like this, and
>> call that instead of `is_move_opcode`, right?
>
> Yes, that's right. I believe `VectorReinterpret` should be implemented for
> all pairs of vector species where both the input and output species are
> implemented. So, `VectorReinterpretNode::implemented` is unnecessary.
@merykitty thanks for the approval. I've run tier1-3 tests for 147633f and they
all passed, and the benchmark results are the same as in the description.
Thanks @chhagedorn for running the tests!
-------------
PR Comment: https://git.openjdk.org/jdk/pull/26457#issuecomment-3158785367
