Context migration in CUDA.NET

Posted by Vyacheslav on Stack Overflow See other posts from Stack Overflow or by Vyacheslav
Published on 2010-04-26T05:43:03Z Indexed on 2010/04/26 5:53 UTC
Read the original article Hit count: 827

Filed under:
|

I'm currently using CUDA.NET library by GASS. I need to initialize cuda arrays (actually cublas vectors, but it doesn't matters) in one CPU thread and use them in other CPU thread. But CUDA context which holding all initialized arrays and loaded functions, can be attached to only one CPU thread.

There is mechanism called context migration API to detach context from one thread and attach it to another. But i don't how to properly use it in CUDA.NET.

I tried something like this:

class Program
{
    private static float[] vector1, vector2;
    private static CUDA cuda;
    private static CUBLAS cublas;

    private static CUdeviceptr ptr;

    static void Main(string[] args)
    {
        cuda = new CUDA(false);
        cublas = new CUBLAS(cuda);
        cuda.Init();
        cuda.CreateContext(0);
        AllocateVectors();
        cuda.DetachContext();
        CUcontext context = cuda.PopCurrentContext();
        GetVectorFromDeviceAsync(context);
    }

    private static void AllocateVectors()
    {
        vector1 = new float[]{1f, 2f, 3f, 4f, 5f};
        ptr = cublas.Allocate(vector1.Length, sizeof (float));
        cublas.SetVector(vector1, ptr);

        vector2 = new float[5];
    }


    private static void GetVectorFromDevice(object objContext)
    {
        CUcontext localContext = (CUcontext) objContext;
        cuda.PushCurrentContext(localContext);
        cuda.AttachContext(localContext);

        //change vector somehow
        vector1[0] = -1;
        //copy changed vector to device
        cublas.SetVector(vector1, ptr);
        cublas.GetVector(ptr, vector2);
        CUDADriver.cuCtxPopCurrent(ref localContext);
    }

    private static void GetVectorFromDeviceAsync(CUcontext cUcontext)
    {
        Thread thread = new Thread(GetVectorFromDevice);
        thread.IsBackground = false;
        thread.Start(cUcontext);
    }
}

But execution fails on attempt to copy changed vector to device because context is not attached? Any ideas how i can get it work?

© Stack Overflow or respective owner

Related posts about c#

Related posts about cuda.net

  • Context migration in CUDA.NET

    as seen on Stack Overflow - Search for 'Stack Overflow'
    I'm currently using CUDA.NET library by GASS. I need to initialize cuda arrays (actually cublas vectors, but it doesn't matters) in one CPU thread and use them in other CPU thread. But CUDA context which holding all initialized arrays and loaded functions, can be attached to only one CPU thread. There… >>> More

  • CUDA Driver API vs. CUDA runtime

    as seen on Stack Overflow - Search for 'Stack Overflow'
    When writing CUDA applications, you can either work at the driver level or at the runtime level as illustrated on this image (The libraries are CUFFT and CUBLAS for advanced math): I assume the tradeoff between the two are increased performance for the low-evel API but at the cost of increased… >>> More

  • Allocate constant memory

    as seen on Stack Overflow - Search for 'Stack Overflow'
    I'm trying to set my simulation params in constant memory but without luck (CUDA.NET). cudaMemcpyToSymbol function returns cudaErrorInvalidSymbol. The first parameter in cudaMemcpyToSymbol is string... Is it symbol name? actualy I don't understand how it could be resolved. Any help appreciated. //init… >>> More

  • CUDA compare arrays

    as seen on Stack Overflow - Search for 'Stack Overflow'
    Hello. Trying to make an app that will compare 1-to-multiple bitmaps. there is one reference bitmap and multiple other bitmaps. Result from each compare should be new bitmap with diffs. Maybe comparing bitmaps rather as textures than arrays? My biggest problem is making kernel accept more than one… >>> More