Is there any way to write Linux kernel module and using CUDA features?
I want to make my Linux kernel as fast as possible.
Great idea! I wonder why nobody thought of it before! I would have thought all that linear algebra and computational geometry in the Linux kernel would be a prime candidate to offload to a heterogeneous accelerator.
In fact, why stop there? Why not port the whole kernel to the GPU. Now that would be something…
The sarcasm from the previous commenter aside, there is no supported way to call CUDA from inside the kernel. People have asked about this in the past, in particular regarding encryption and other computational things that go on in kernel space.