Daniel Arndt
The pull request is best understood looking at the changes to `Kokkos_Core_fwd.hpp` and then `Kokkos_Parallel_Reduce.hpp`: - The new signature for the internal classes is ``` template ::execution_space> class ParallelReduce; ```...
> t certainly does not support CUDA 11 > I did some archeology and found #3106 > There is no more reason to have it What about clang++ using CUDA...
> Does that actually work? I don't know about `macOS` support but we test `clang-11` with `CUDA-10.1` in our CI.
Also, see https://llvm.org/docs/CompileCudaWithLLVM.html#id3: > CUDA is supported since llvm 3.9. Clang currently supports CUDA 7.0 through 11.5. If clang detects a newer CUDA version, it will issue a warning and...
> Lets drop the CUDA/APPLE combo. Don't think that matters right now, and might not in the future either I removed the related macro guards.
`HIP` failing with ``` [ RUN ] hip.reduce_device_view_range_policy 4: /var/jenkins/workspace/Kokkos/core/unit_test/TestReductions_DeviceView.hpp:89: Failure 4: Expected: (time1) < (time_fence1), actual: 0.000702964 vs 0.000493095 ``` is clearly unrelated (and spurious).
> 2) our commits in the previous PR did not conform to the author naming convention used in deal.ii, which seems to be a requirement in order to run the...
/rebuild
Looks like this needs https://github.com/dealii/dealii/pull/14339 and thus needs to be rebased.
Looks like we need `cuda-11` for `gcc 9` instead.