Daniel Arndt
The branch now uses the module for all unit tests in `core` and I have all tests passing for the `Serial` backend. Some notes: - The module file listing all...
> Cool that you got the tests to work -- you're far ahead of me (in parts because of our "unique" way of setting up tests). Can you show me...
> Ah, ok. So you're just referencing something you've just built in the same cmake call. I think that's about as far as I've gotten too, so far. I can...
> I see. I think I just hadn't gotten far enough yet to try that out. Is this last statement something that you do with an external project that imports...
Fixed in #8263.
https://godbolt.org/z/er5Ycb7M6 shows errors with thrust and -fno-exceptions for all CUDA versions and that is consistent with https://github.com/NVIDIA/thrust/blob/756c5afc0750f1413da05bd2b6505180e84c53d4/thrust/system/detail/system_error.inl#L85-L106. ``` # nvcc -x cu -std=c++17 -Xcompiler -fno-exceptions blabla.cpp && ./a.out /opt/nvidia/hpc_sdk/Linux_x86_64/23.7/cuda/12.2/include/thrust/system/detail/system_error.inl: In...
Currently, this pull request doesn't add flags for the `Makefile` build and that's the reason why CI passes. Adding `-Xcompiler -fno-exceptions` gives: ``` /app/kokkos_new/core/unit_test/TestViewAPI.hpp:1366:3: error: exception handling disabled, use '-fexceptions'...
`clang+Cuda` just seems to be fine: https://godbolt.org/z/eTcaT7PbG.
It seems we should just not test with -fno-exceptions with nvcc.
`SYCL` seems to be fine even on the testbeds. `-fno-exceptions` is visible in the compilation flags.