Daniel Arndt

Guarantee parallel_scan to use two passes

7

Currently, we only guarantee that `parallel_scan` calls the functor with `is_final==true ` but not if it's called with `is_final==false`. Note that all backends apart from `Serial` use a two-pass implementation....

Creating an unmanaged View with Kokkos::Array fails

4

As reported on `slack` creating an unmanaged View with a `Kokkos::Array` extent fails, see https://godbolt.org/z/Yr38ax9YE. ```C++ #include #include int main() { using T = Kokkos::Array; std::vector w(10); Kokkos::View w_unmanaged(static_cast(nullptr), 1,...

Part of https://github.com/kokkos/kokkos/pull/6151 that only adds the tests from there. The tests check that kernels submitted to one execution space instance by multiple threads are not running concurrently.

Allow non-trivially-copyable comparators for SYCL

8

Related to https://github.com/kokkos/kokkos-core-wiki/pull/504 and https://github.com/kokkos/kokkos/pull/6801#issuecomment-1969888881. Even with changes like ```diff diff --git a/algorithms/src/sorting/impl/Kokkos_SortByKeyImpl.hpp b/algorithms/src/sorting/impl/Kokkos_SortByKeyImpl.hpp index 47c96a0a0..cd3548d4a 100644 --- a/algorithms/src/sorting/impl/Kokkos_SortByKeyImpl.hpp +++ b/algorithms/src/sorting/impl/Kokkos_SortByKeyImpl.hpp @@ -128,9 +128,13 @@ void sort_by_key_onedpl( auto policy =...

Remove MemorySpace::[de]allocate overloads without labels

6

I noticed in https://github.com/kokkos/kokkos/issues/6794#issuecomment-1957666389 that we call `MemorySpace::deallocate` without a label only from internal allocations where we can do better. Thus, this pull request suggests streamlining the interface by removing...

SYCL: Add support for Graphs

4

This pull request adds support for Graphs based on https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/experimental/sycl_ext_oneapi_graph.asciidoc. The current status is sufficient to pass all related unit tests. Note that: - immediate command lists are not supported...

Support completion query for execution spaces

35

Fixes #4209. This pull request adds the capability to query execution space instances for the status of the last event/kernel submitted to them. To this end, an enum class with...

Move `layout_iterate_type_selector` into `Impl` namespace

2

Follow up we should move the `layout_iterate_type_selector` into `Impl` namespace. There is no reason it should be public. Note this thing is used in stuff like ViewCopy to figure out...

Labels in fences create noticable overhead

11

@dalg24 I made a benchmark that looks like: ```c++ bool test_fence_with_kokkos(::benchmark::State&) { ... for( 100 times ) space.fence(); // will use the default message, and will use Kokkos (+ expect...

Cuda: Configuring fails with CMake 3.28.4 and later

1

With `CMake 3.28.4`, we are getting ```bash CMake Error at /app/kokkos-install/lib/cmake/Kokkos/KokkosConfig.cmake:41 (SET_TARGET_PROPERTIES): The link interface of target "CUDA::cudart" contains: CUDA::cudart_static_deps but the target was not found. Possible reasons include: *...