I've tried on ORNL Wombat cluster which is aarch64 host with NVIDIA GPUs. Offloading works fine. All 19 tests in libomptarget passed.
Sounds good. Thanks for testing it.
Building the CUDA plugin on aarch64 hosts was enabled in the ykt branch ages ago, so the configuration has been tested extensively. LGTM.
I didn't have privilege to merge. Could someone do it for me?