kokkos-kernels
kokkos-kernels copied to clipboard
kokkos
Kokkos C++ Performance Portability Programming Ecosystem: Math Kernels - Provides BLAS, Sparse BLAS and Graph Kernels
Use only one fragment per team when the matrix block size is small enough.
A revision of #1330. Using CUDA backend as an example (this is applied to all backends) Break the monolithic `unit_test/cuda/Test_Cuda_Sparse.cpp` into multiple translation units, in this case * `unit_test/cuda/Test_Cuda_Sparse.cpp` includes...
@lucbv @fnrizzi @kliegeois Similar to #1433, this PR moves `TeamGemv` and `TeamVectorGemv` from `KokkosBatched` to `KokkosBlas` - except from batch processing (rank-3 `A` matrix) variants. ## Status * Implementation: -...
@e10harvey @brian-kelley Unless someone explicitly requests to keep a one of these branches open, I will go ahead and delete them. Active: - [fix_docs](https://github.com/kokkos/kokkos-kernels/tree/fix_docs) [open] - [issue1344](https://github.com/kokkos/kokkos-kernels/tree/issue1344) [merged] - [half-precision](https://github.com/kokkos/kokkos-kernels/tree/half-precision)...
c++11 deprecated implicit copy-assign operators when a copy-constructor is defined. array_sum_reduce is actually okay with implicit versions both copy-ctor and copy-assign, so don't define either one ourselves
Now that we require c++17 support, use [[fallthrough]] to suppress `-Wimplicit-fallthrough` and friends
Hello I have compiled the 'kokkos' project for AMD ZEN2 architecture (as reported [here](https://en.wikichip.org/wiki/amd/epyc/7402)) (-DKokkos_ARCH_ZEN2=On) and with OpenMP enabled (-DKokkos_ENABLE_OPENMP=On) After building successfully, I compiled the 'kokkos-kernels' project too, with...
The destination of this deep copy is the raw argument pointer (an input argument), when instead it should probably be the `values` member of the BlockCrsMatrix class: https://github.com/kokkos/kokkos-kernels/blob/564dccb339d8d1528c2bb948abdac0c6e48e09d5/src/sparse/KokkosSparse_BlockCrsMatrix.hpp#L929 Furthermore, the...
This PR removes `Experimental::BlockCrsMatrix` and everything that references it. The replacement is intended to be `Experimental::BsrMatrix`, which matches its interface and has a data layout consistent with third-party library (TPL)...
- The path to the kokkos-kernels source directory was missing - Doxygen was not being run
