tdd11235813

Thanks a lot for the inspiring concept work. A more simple frontend would help not only the newcomers. It just makes coding more productive and readable. Code examples also would...

it seems, clang disables OpenMP, when clang(cuda) is involved. So no `_OPENMP` and pragma omp's are defined. When CUDA is disabled, then it works again. Tested with clang5+cuda8 and clang6+cuda9.0....

I will test with the new versions next week, when our Intel/Nvidia developer system is up. (On the cluster modules are not built properly or missing)

Using clang7 and CUDA10.0 I got a known issue: ``` clang-7: error: cannot find libdevice for sm_30. Provide path to different CUDA installation via --cuda-path, or pass -nocudalib to build...

12 hours ago they pushed the CUDA 10 commit :)

Sorry for not replying earlier. Just can report the last state. I got build errors when using clang-9 (patched-upstream) and cuda (tested 9.0.176, 9.2.88, 10.0.130). ### clang compilation ```bash spack...

thanks for the quick response. I will try the new version of clang. For the math the header `__clang_cuda_math_forward_declares.h` did not define `lgammaf`, but only `lgamma`. Same for `tgamma`. Looks...

ok, new version of alpaka and clang helped except for the math stuff, but adding missing functions to the aforementioned header works ... now I just get a glibc linker...

so you did not get the lgammaf error or did you fix the clang cuda header before as well?

I can't remember, and if upstream is working, then upstream is fine. I'll test the whole thing with upstream's version again.