Daniel Arndt

Results 161 issues of Daniel Arndt

Related to #301. This pull request introduces a class that handles data transfer in CellDataTransfer when cells are coarsened or refined by assigning the value at the closest quadrature point.

Part of #8117 to get started. Out of all the Kokkos functionalities, `StdAlgorithms` seems to be easiest to provide a module for. This pull request adds a build with clang-19...

C++20

Related to #8117 (g++ in particular). This pull request avoids more TU-local symbols in headers that are problematic when creating C++20 modules. In particular, - avoid unnamed namespaces in headers....

Fixes https://github.com/kokkos/kokkos/issues/8081. [kokkoscore.ccm](https://github.com/kokkos/kokkos/pull/8117/files#diff-eb9e4e20ea79de568a260e5ec06a35460fd690765ff40093729f4b06e02cc886) needs some structuring but otherwise this should be good enough for a first stab at using modules that works with `clang++` to gather some feedback for the...

C++20

We noticed in #7890 that we don't actually test with `Kokkos_ENABLE_LARGE_MEM_TESTS=ON` and had the respective tests failing. This pull request proposes to enable these tests for for some of the...

oneAPI 2025.1.0 added preliminary support for device virtual functions in `SYCL`, at least, with JIT compilation, see https://github.com/intel/llvm/blob/02ba869938b7c77eb7a6a88b9bbbbcc866057084/sycl/doc/extensions/proposed/sycl_ext_oneapi_virtual_functions.asciidoc. This pull request adds support for that feature in `Kokkos` via a...

Addresses #7412. This pull requests distinguishes between - `HostSpace` - `Default::ExecutionSpace::memory_space` - `SharedSpace` - `SharedHostPinnedSpace` ~~and aborts if we pass in a pointer from a different memory space for the...

#7582 will require `CMake` 3.21. and we our current guards are ``` $ git grep -n "CMAKE_VERSION" cmake cmake/KokkosConfigCommon.cmake.in:15: if(CMAKE_VERSION VERSION_LESS "3.18") cmake/KokkosConfigCommon.cmake.in:229: if(CMAKE_VERSION VERSION_GREATER_EQUAL 3.17) cmake/Modules/FindTPLCUDA.cmake:10:if(KOKKOS_CXX_HOST_COMPILER_ID STREQUAL NVHPC AND...

Our CI shows that `team_scratch_1_queues` is very expensive ``` 7: [ OK ] cuda.team_scratch_1_streams (24035 ms) 5: [ OK ] hip.team_scratch_1_streams (43609 ms) 11: [ OK ] sycl.team_scratch_1_queues (148337 ms)...

Required for #7961. The current approach for `DualViews`'s `modified_flags` increments the host or device value to one higher than the other flags value in the respective `modify` call and it's...