kokkos-kernels
kokkos-kernels copied to clipboard
kokkos
Kokkos C++ Performance Portability Programming Ecosystem: Math Kernels - Provides BLAS, Sparse BLAS and Graph Kernels
Building tests in Kokkos Kernels exhibits low parallelism when the CUDA and OpenMP spaces are enabled. Consider two different ETI configurations: "small" ``` float, double layoutLeft, layoutRight offset size_t ordinal...
Mainly provided so that Luc can test this out on a HIP platform
With kokkos develop commit: ``` commit 506088b55d4196b31dd81287240feff7051b6e07 Merge: d418fa5 4d06211 Author: Damien L-G Date: Tue Feb 8 10:59:33 2022 -0500 Merge pull request #4762 from crtrott/rmeove-gcc-toolchain-from-makefile-kokkos Remove gcc-toolchain auto add...
SpTRSV (sparse tri solve) should have a serial algorithm that does no setup, and solves using forward substitution. KokkosSparse_trsv.hpp already contains an implementation like this, but it should be callable...
In the constructor, https://github.com/kokkos/kokkos-kernels/blob/e4529f966d2bad5c4cdaec6911bef44a8013d5fb/src/sparse/KokkosSparse_CrsMatrix.hpp#L618 the pointer for `rowmap` may need to be `size_type*` @lucbv
A merge to kokkos-kernels on 1/19/2022 triggered a -Werror breaking the clang/7+cuda/9.2 and clang/8+cuda/10 nightly builds, it looks like changes to related `Kokkos_InnerProductSpaceTraits.hpp` possibly from #1266 **Sample output**: ``` In...
here is a starting point reference for the algorithms we would like to implement: https://faculty.cc.gatech.edu/~echow/pubs/parilut.pdf There is an implementation of the above algorithm in [Ginkgo](https://ginkgo-project.github.io/). Eventually the algorithm might be...
@benc303 reports failures when running Trilinos tests of Kokkos Kernels on A100. I am not sure where we can go to investigate that issue but I will look into that.
Kokkos::Impl functions should be avoided in Kokkos Kernels if possible since they are susceptible to change without warning. Further more in the particular instance of `Kokkos::Impl::throw_runtime_exception` there is no reason...
Nightly test failures with Serial backend of various batched gemm tests with ibm/xl/16.1.1 on Power9
Various batched gemm tests have been failing in Serial build nightly tests on Power9 with xl/16.1.1: **Failing tests**: ``` : [ FAILED ] 4 tests, listed below: 2: [ FAILED...