kokkos

kokkos copied to clipboard

OpenACC `parallel_for` and `parallel_reduce`

1

This PR advances the integration of the OpenACC backend into kokkos. It introduces `parallel_for` and `parallel_reduce` for `RangePolicy` and bumps the incremental tests to level 5. Pedro did the implementation...

JBludau

team impl of std algorithms

4

# Contains team-level implementations for: ## NonModifying sequence algorithms - `count_if` ## Modifying sequence algorithms (complete) - `fill`, `fill_n`, `replace`, `replace_if`, `replace_copy`, `replace_copy_if`, `copy`, `copy_n`, `copy_if`, `copy_backward`, `transform`, `generate`, `generate_n`,...

fnrizzi

add cmake flag to enable mdspan and include mdspan as a tpl

9

This PR adds `KOKKOS_ENABLE_IMPL_MDSPAN` and `KOKKOS_ENABLE_EXTERNAL_MDSPAN` flags. Additionally, mdspan sources are bundled as a tpl. This is ongoing work for #4421 Note: the license on the mdspan sources caused a...

nmm0

Trilinos compile question CUDA + MPI

9

Greetings, I was trying to compile Trilinos with CUDA and OpenMPI using the following: export OMPI_CXX=${trilinosdir}/packages/kokkos/bin/nvcc_wrapper cmake -DCMAKE_BUILD_TYPE=RELEASE -DTrilinos_MUST_FIND_ALL_TPL_LIBS=TRUE -DTPL_ENABLE_MPI=ON -DCMAKE_CXX_STANDARD:STRING=17 -DTrilinos_ENABLE_Kokkos=ON -DTPL_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_UVM=OFF -DTrilinos_ENABLE_OpenMP=OFF -DTrilinos_ENABLE_Amesos2=ON -DTrilinos_ENABLE_Belos=ON -DTrilinos_ENABLE_MueLu=ON -DTrilinos_ENABLE_ROL=ON -DTrilinos_ENABLE_Ifpack2=ON...

Adrian-Diaz

Seg. fault on ppc64le with clang 15

6

This occurs with both version 3.6.01 and well as the master branch: ``` cmake .. -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_INSTALL_PREFIX=`pwd`/../kokkos_install -DKokkos_ENABLE_OPENMP=On -DKokkos_ARCH_HSW=On -DKokkos_ENABLE_TESTS=On make -j install OMP_PROC_BIND=TRUE OMP_NUM_THREADS=16 ./core/unit_test/KokkosCore_UnitTest_OpenMP --gtest_color=no --gtest_filter=openmp.task_fib_single ... Note:...

scottbiersdorff

- Keeps `partiton_master` deprecated. - Introduces `Kokkos_ENABLE_DISABLED_CODE_4`.

masterleinad

Avoid instantiating default constructor of value type when WithoutInitializing is given

2

Trying #2024 on top of #5277. Fixes #1873.

masterleinad

Team- and thread-level sort, sort_by_key

2

(Addresses #645) Add sort functions that can be called from device, and exploit team and thread level parallelism. The new functions use bitonic sort, which is good for this because...

brian-kelley

Fixup CUDA arch auto detection in Trilinos

1

Tentative fix for https://github.com/kokkos/kokkos/issues/5320#issuecomment-1209623403

dalg24

Remove dummy arguments for ViewCtorProp

2

This pull request removes support for the dummy `std::integral_constant` `ViewCtor` properties and replaces the mechanism to add properties (that weren't existent already) with a separate function. This avoids writing much...

masterleinad