I’ve a question regarding DirectX interop with cuda. I implemented a grafic output using OpenGL and DirectX ontop of a simulation that is using CUDA (toolkit 3.2) to compute data. The Implementation is quite similar to the SimpleGL and SimpleD3D9 samples.
My question is now: Is it posible to reregister a d3d vertex buffer after a device lost event without calling cudaThreadExit? Without cudaThreadExit cudaGraphicsD3D9RegisterResource returns unknown error. Since the simulation has huge data stored in GPU memory it is not practicable to memcpy the whole data and states to the host everytime the d3d device is lost. On the OpenGL side the probleme does not exist. Why?
Thanks for any help.