I would like to know if it is possible to load geometry from a VBO, for example, make a ITriangle_mesh from device data without any extra copy to the CPU?

To sum up my problem, I have input data on the host. It's send to the device, then an OpenCL process is launched. Finally I would like to raytrace the resulting geometry from the OpenCL process.

I try with Optix where the result is pretty good. I share data between OpenCL and Optix using VBO (the only way I found to share OpenCL and CUDA buffers).
How can I do the same with IRay ?