Daniel Arndt
> On 2nd thought I am tempted to just go ahead and add support for EXTRA_ARGS (I'd keep ENVIRONMENT) Sure, added in the last commit.
@ndellingwood I fixed the order so that the options are applied past test generation. The syntax should be ``` cmake -D Kokkos_UnitTest_Random_ENVIRONMENT="GTEST_FILTER=-serial.Random_XorShift1024_0" .. ```
> @masterleinad should I also test with these changed pulled into #6164 to confirm that the options are recognized after the TriBITS removal? I expect that you would run into...
All the changes here are in the branch for Kokkos-standalone so it shouldn't actually have any impact for a Trilinos build as of now. 🤦
> So close this and merge into the overhaul PR? Yeah I checked that the ENVIRONMENT option works as expected for `Kokkos` and `Trilinos`.
```C++ Kokkos::View device_view("device_view", 10); Kokkos::View host_view = Kokkos::create_mirror_view_and_copy(Kokkos::HostSpace{}, device_view); Kokkos::View host_view1 = Kokkos::create_mirror_view(Kokkos::HostSpace{}, device_view); //Kokkos::View host_view2 = Kokkos::create_mirror_view(device_view); ``` fails only in the last lines (since that returns a View...
```diff diff --git a/core/src/HIP/Kokkos_HIP_Space.hpp b/core/src/HIP/Kokkos_HIP_Space.hpp index 63590f278..7dff79f00 100644 --- a/core/src/HIP/Kokkos_HIP_Space.hpp +++ b/core/src/HIP/Kokkos_HIP_Space.hpp @@ -305,10 +305,11 @@ static_assert(Kokkos::Impl::MemorySpaceAccess::assignable); template struct MemorySpaceAccess { - enum : bool { assignable = false };...
PETSc tries to wrap a host pointer in a DualView which is pretty hard on MI300A. Normally, ``` View v_h(host_ptr, host_size); auto v_d = Kokkos::create_mirror_view_and_copy(MemorySpace{}, v_h); Kokkos::DualView dual_view(v_d, v_h); ```...
`Kokkos` also has a (portable) SIMD implementation mirroring the std::simd interface.
#17874 includes all the pull requests suggested to cherry-pick above.