Damien L-G
Damien L-G
Ignoring HIP failures. I can't approve my own PR but I would be willing to merge as is.
> My issue with using ARBORX_ASSERT is that when a user compiles ArborX in release mode, this will never get noticed. As currently implemented you would always get the runtime...
Well looks like we still have the two branches so you need to target `develop`
> > `strip` rule is being re-run every time but I can live with it. > > Not sure that can be fixed. Every time it runs, it touches libperfetto-connector.so...
> just store the primitives and do the right thing without needing to provide another callback. What mechanism do you suggest? ADL?
> I guess your underlying question is how much slower is it to compile code with this extra template instantiations. My guess is not much, as those are standalone functions...
It influences the `find_package` cascade when we look for rocThrust. If absent, we end up building for all supported architectures. This is not applicable to CUDA. Not sure whether Kokkos...
> Do we always need this code or only when using rocthrust? No you are right, only when using rocThrust
I think for now we should stick with configuring with `-DGPU_TARGETS=...` CC @Rombur
> What about just using Kokkos::Sum instead of a scalar as the return value so that we are doing what was intended? Sure