Support reduce for bitwise ops

[Da1sypetals]

Is it planned to support reduction for bitwise ops for integer types? For example, reduce_and, reduce_xor, etc.

[LeiWang1999]

@Da1sypetals Thanks for the report! Currently no, but it’s a great first issue, and I think we can find somebody to give it a try.

[Da1sypetals]

Really looking forward to this feature! Our variant of sparse attention requires some of bitwise reduction operation, and we are currently investigating about possible solutions to implement our kernel, but we found triton to be quite slow in our use case. ❤️

[LeiWang1999]

cc @tzj-fxz if you have time?

[tzj-fxz]

OK, I'll have a try.

[LeiWang1999]

resolved, cc @Da1sypetals , close :)