Daniel Arndt

Only `default_exec.overlap_mdrange_policy` is failing for two `Cuda` builds which looks unrelated/spurious.

Did you do a plain copy from `desul`?

My biggest concern is that we are nowadays mostly interested in performance on GPU backends and that the machines we are running CI on don't produce good/consistent enough results for...

Looks like some version of `std::unreachable` would also do the trick (at least for ICC), see https://godbolt.org/z/99T84x4cb.

Do you also see this problem with the `develop` or `release-candidate-3.7.00` branch?

I could reproduce and this is independent of `Kokkos`: ``` $ nvcc -arch=sm_70 -ccbin /usr/bin/g++-8 -Xcompiler -std=c++17 -x cu /home/6da/reproducer/main.cpp -c -o CMakeFiles/main.dir/main.cpp.o /usr/include/c++/8/bits/stl_pair.h(79): error: inline specifier allowed on function...

> thinkinghmmm, so this could be a bug from nvcc then? Somehow, we don't pass `-std=c++17` to `nvcc` directly and that's the problem, also see https://forums.developer.nvidia.com/t/ubuntu-20-04-gcc-9-3-cuda-toolkit-11-3-not-a-supported-combination/193486/10.

``` # Specify cmake version needed cmake_minimum_required(VERSION 3.20 FATAL_ERROR) # Set project name, language and c++ std project(test LANGUAGES CXX) #set(CMAKE_CXX_STANDARD 17) #set(CMAKE_CXX_STANDARD_REQUIRED ON) # Require OpenMP find_package(OpenMP REQUIRED) #...

For SYCL+CUDA, we currently get output such as ``` Devices: KOKKOS_ENABLE_SYCL: yes Runtime Configuration: macro KOKKOS_ENABLE_SYCL : defined Name: Tesla V100-PCIE-32GB Driver Version: CUDA 11.7 Is Host: 0 Is CPU:...

What is this used for?