Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Build alpaka device code for the CUDA and ROCm back-ends in "GPU only" mode. In this mode, functions marked as ALPAKA_FN_ACC are marked as __device__ functions, and are compiled only for the corresponding GPU device back-ends. Currently, functions marked as ALPAKA_FN_ACC are marked as __host__ __device__ functions, and may be compiled for both device and host back-ends. The latter leads to linker errors in kernels that use device symbols like threadIdx, blockIdx, etc. that are compiled for the ROCm back-end.
- Loading branch information