Cactus supports CUDA programming in the option lists level via various CUDA_FOO variables but does not provide macros for __device__ and __host__ modifiers used by both CUDA and HIP accelerators.

It would be good if Cactus could provide those depending on which compiler is detected, Predefined defines to look for are (copying AMReX): CUDA_ARCH || HIP_DEVICE_COMPILE || SYCL_DEVICE_ONLY (in its include/AMReX_GpuQualifiers.H).

--
Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2737/define-cctk_device-and-cctk_host-in-cactus