Daniel Arndt
Daniel Arndt
> This work could be useful for multi-core CPUs also. I'd be curious how you would distribute the execution range for host parallel backends with this implementation.
This needs to be rebased and probably needs a changelog entry.
Looks like this changes the output of step-40 in some way.
If it passes the CI now, I don't have any problems with enabling the feature for NVIDIA GPUs as well.
I'm not working on it.
We briefly discussed strategies in https://github.com/kokkos/kokkos/pull/8301#issuecomment-3180321593.
> Once we are happy with the direction and the PR #16745 is merged, I will also write a changelog for this PR. Do you want to write a changelog...
``` /spack-installs/binutils/2.41/gcc/8.5.0/base/xt4vsa7/bin/ld: CMakeFiles/Kokkos_CoreUnitTest_Cuda_ViewSupport.dir/cuda/TestCuda_View_ViewEqualityOperator.cpp.o: in function `Kokkos::Impl::BV::BasicView::BasicView()': /home/runner/_work/kokkos/kokkos/kokkos/core/src/View/Kokkos_BasicView.hpp:237:(.text+0xe04): undefined reference to `Kokkos::Impl::ReferenceCountedDataHandle::ReferenceCountedDataHandle()' /spack-installs/binutils/2.41/gcc/8.5.0/base/xt4vsa7/bin/ld: /home/runner/_work/kokkos/kokkos/kokkos/core/src/View/Kokkos_BasicView.hpp:237:(.text+0xe0c): undefined reference to `Kokkos::Experimental::layout_right_padded::mapping::mapping()' ```
It turns out that virtual destructors are not yet supported.
> I am not sure about how it is done for some parts, however. I find the modules name `kokkos` hard to read, I propose `kokkos_`, or `kokkos.` instead. `kokkos:`...