VUDA is a header-only library that allows developers to write GPU-accelerated applications using the Vulkan API. It provides a CUDA Runtime API interface, closely following the specifications of the CUDA runtime. To use VUDA, developers can include the vuda.hpp header file and utilize the vuda:: namespace. Alternatively, they can use the vuda_runtime.hpp file which wraps and redirects all CUDA functionality. The code provided showcases an example of using VUDA, including allocating memory on the device, copying data to the device, running a kernel, and copying the result back to the host.
https://github.com/jgbit/vuda