Hardware Accelerated Bounding Volume Hierarchy

Is there a way to use the hardware accelerated 3d-tree (BVH) in a CUDA kernel for general computing?
In Vulkan, one can use ray query in a non-ray tracing pipeline (say in a compute shader) to use this facility.
It would be useful to have a more generalized way to use this fundamental data structure for general purpose computing.

It’s not currently exposed in CUDA.

As you have already indicated, it can be accessed (perhaps indirectly) using graphics ray tracing systems including Optix.