https://hgpu.org/?p=25214
Study and evaluation of improved automatic GPU offloading method