GL examples cause operating system call error

[farzeber]

My computer enviroment is no problem: win10 64bit + cuda10.0 + jcuda 10.0. Other samples are passed, but there are two bugs when I test all java source (directory: driver/gl.samples) related to opengl and JCudaRuntimeUnifiedMemory.java . The former reminds me with erro massage "CUDA_ERROR_OPERATING_SYSTEM on thread AWT-EventQueue-0", and after I debug it I only conclude that it comes from a wrong return of native methond "cuGraphicsGLRegisterBufferNative" ( it doesn't appear when I use visual studio 2012 to test the methond, I don't know why). The latter will fail with a not responding even a crash. Obviously, the system call isn't successful.

[jcuda]

First, it is not entirely clear which sample causes the error right now:

• The method cuGraphicsGLRegisterBuffer is only called in JCudaDriverSimpleJOGL, JCudaDriverSimpleLWJGL and JCudaDriverVolumeRendererJOGL, and not in the unified memory examples
• The error message/constant CUDA_ERROR_OPERATING_SYSTEM is actually a driver API error. For the runtime API, it would be cudaErrorOperatingSystem. So this could only be caused by the JCudaDriverUnifiedMemory sample (and not the runtime one)
• I just tried the GL samples as well as the unified memory examples, and could not immediately reproduce the error.

So in which example does which error appear exactly?

[farzeber]

Sorry, what I said isn't clear. There are two problems when I run the sample. One is that I will encounter error "CUDA_ERROR_OPERATING_SYSTEM" when I test JCudaDriverSimpleJOGL, JCudaDriverSimpleLWJGL and JCudaDriverVolumeRendererJOGL. The other is that my computer crashes when I run JCudaDriverUnifiedMemory.

[farzeber]

Maybe it isn't a common question, and I'm unlucy to encounter it. Good night! Don't worry it. I plan to choose pycuda for cuda accelerating

[jcuda]

Well, it's at least good to know that there might be an issue. The first guesses would be as usual ("Do you have the latest drivers installed" etc). But "CUDA_ERROR_OPERATING_SYSTEM" also raises the suspicion that it might be something specific for Windows 10. I could try to allocate some time to run the samples on a Win10 machine, but could not give a timeline here.

If you urgently need the specific features (GL buffer mapping and unified memory) on Windows 10, I could increase the priority here, but otherwise, I'd keep this issue mainly as a reminder, and see whether others report similar problems.