cuda 13.0 error, no suitable constructor exists

[mlohry]

On latest develop and cuda 13.0, I'm hitting this compiler error:

/home/mlohry/dev/build-cuda/external/raja/include/RAJA/policy/cuda/MemUtils_CUDA.hpp(147): error: no suitable constructor exists to convert from "int" to "cudaMemLocation"
      do { cudaError_t code_56792578 = cudaMemAdvise(ptr, nbytes, cudaMemAdviseSetPreferredLocation, device); if (code_56792578 != cudaSuccess && code_56792578 != cudaErrorNotReady) { static constexpr auto func_name_56792578 = "cudaMemAdvise"; static constexpr auto arg_names_56792578 = ::camp::experimental::cuda_get_api_arg_names(func_name_56792578); ::camp::reportError( "CUDA", cudaGetErrorString(code_56792578), func_name_56792578, arg_names_56792578, std::forward_as_tuple(ptr, nbytes, cudaMemAdviseSetPreferredLocation, device), "/home/mlohry/dev/build-cuda/external/raja/include/RAJA/policy/cuda/MemUtils_CUDA.hpp", 147); } } while (false)
                                                                                                     ^

/home/mlohry/dev/build-cuda/external/raja/include/RAJA/policy/cuda/MemUtils_CUDA.hpp(149): error: no suitable constructor exists to convert from "int" to "cudaMemLocation"
     ((int)-1)
     ^

Looks like a camp issue in CAMP_CUDA_API_INVOKE_AND_CHECK.

[MrBurmark]

I'm seeing that and an issue from the cmake as well.

CMake Error at tpl/RAJA/cmake/SetupPackages.cmake:40 (message):
  External CUB not found, CUB_DIR=.
Call Stack (most recent call first):
  tpl/RAJA/CMakeLists.txt:110 (include)

Cuda suggests we change our cmake, to link with CCCL::CCCL. https://nvidia.github.io/cccl/cccl/3.0_migration_guide.html
@white238 is this something that each project using cub, thrust, or anther component of cccl should fix or does it make sense to handle this in blt?

[white238]

I would need to sit down and verify it but it seems like all the components you mentioned are under the CUDAToolkit include and don't require a specific find for each, for example, find(CUB).

It is on my list to upgrade our, quite old at this point, CUDA support in BLT, but I was hesitant to do it before Sierra was retired and now I have just been overwhelmed.

[rhornung67]

@smithsg84 please verify that this is resolved and can be closed

[smithsg84]

@smithsg84 please verify that this is resolved and can be closed

The error compilation error is fixed by CAMP PR llnl/camp#196

CMake error is patched in #2011 in RAJA FindCUB.cmake but CCCL::CCCL linking would be cleaner and more modern CMake.

@rhornung67 I'd suggest making new issue to address CMake CCCL::CCCL changes and close this when the CAMP / RAJA PRs are merged.