ThreadVectorRange with a range

[kyungjoo-kim]

I found for TeamThreadRange(member, ibeg, iend) but I could not see ThreadVectorRange(member, ibeg, iend). Do we have a specifc reason not to have that ? If everything is okay, would it be possible to add a constructor with a range arguments (member, begin, end) ?

[crtrott]

OK OK we will do it 👎 ...

We didn't have that initially because it is better for the compiler and alignment etc. to have vectorized loops which go from 0 to N. But

• compilers are getting better
• too many people ask for this

So we might as well do it.

[kyungjoo-kim]

Oh.... I see why you prefer not to have it. Thanks.

I meant to use the ThreadVectorRange on cuda, which can remove unnecessary offset calculation.

Thank you again.

[crtrott]

Kyungjoo I totally get it, and I think we should just do it, but document that if possible use the compile time zero offset variant.

[bartgol]

I just thought I'd need this feature too, and I stumbled on this issue. Any rough idea if/when this could end up in master (or at least develop)? To be clear, I'm not trying to put pressure, I'm just trying to understand if I can wait for this, or if I should find a workaround.

Thanks!