I have upgraded my application to use CUDA for updating VBOs. Spatial coherency of the terrain enables reusing data already stored in the neighboring blocks. But to shorten the story, I have a very serious problem accessing resident buffers. External Image
For a (Bindless Graphics) resident buffer, function cudaGraphicsGLRegisterBuffer() returns cudaErrorMemoryAllocation!
It works for normal VBOs. Currently I’m using 259.09 drivers.
Another problem was returning cudaErrorPriorLaunchFailure by cudaGraphicsMapResources() call, for a normal VBO. While I was experimenting with small demo application with two VBOs everything worked perfectly, but when I tried to implement it in my application with thousands of VBOs with complex organization, the problems started to arise. External Image
Please, answer me is it possible to use CUDA along with Bindless Graphics anyway?
If you haven’t done so already, you should be applying for a registered developer account. Then you can report such mission-critical things through their nVidia’s bug tracking interface. I’ve usually had quick and professional turnaround on the two issues I’ve raised so far.
I think the GPU Computing Developer program is the right one, your problem seems to be an interoperability problem with CUDA and not really an OpenGL issue.