I'm starting, need some help

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, I’m new to CUDA programming, and I’d like to get some help to make a program that would take an array, modify it and return it back to host.

public static CUDA cuda;

public static CUfunction myFunction;

int[] myArray = new int[4];

myArray[0] = 10;

myArray[1] = 20;

myArray[2] = 30;

myArray[3] = 40;

cuda = new CUDA(0, true);

cuda.LoadModule("test.cubin");

myFunction = cuda.GetModuleFunction("myFunction");

CUdeviceptr dpt = cuda.CopyHostToDevice<int>(myArray);

cuda.SetParameter(myFunction, 0, myArray);

cuda.Launch(myFunction);

cuda.CopyDeviceToHost<int>(dpt, myArray);

/* I'd like the CUDA program to add 10 to every item

in the array, so i would get myArray[0] = 20, myArray[1] = 30, etc

*/

Wouldn’t the C# code be something like this?

Thanks in advance! :">

2

Im not too familiar with the .net extension for cuda so i will leave that for someone else

The cuda program will look like:

__global__ void vecAdd(float* in, float*out)

{

idx = blockDim.x*blockIdx.x+threadIdx.x

if(idx>=NUM_ITEMS)

return;

out[idx]=in[idx]+10.0f;

}

Obviously this does not do the task “in place” (one buffer in, one buffer out) but you can modify it easily to do it in place.

Hope this helps.

3

I’ve been trying this code, but I get a cudaUnknownError in my code at this line.

cuda.CopyDeviceToHost(dpt, myArray);

Is there something I’m doing wrong in my code (well, it’s obvious there is something External Media ). I’d like to get some help, I can’t pass pointers on C#, so that’s why I try to pass the whole array as a parameter. Is that wrong?

Thanks again :rolleyes:

EDIT

Even if I pass the pointer instead of the array it still won’t work.

fixed (int* iAr = inArray)

				{

					cuda.SetParameter(vecAdd, 0, (uint)iAr);

					cuda.SetParameterSize(vecAdd, (uint)inArray.Length);

				}
4 
...

int[] myArray = new int[4];

...

CUdeviceptr dpt = cuda.CopyHostToDevice<int>(myArray);

cuda.SetParameter(myFunction, 0, myArray);

cuda.Launch(myFunction);

Hi,

I have never used CUDA.net, but I know of both, used independently… I guess the problem is that in cuda.SetParameter you are passing “myArray”, that is a C# managed array. I guess you need to pass instead the device pointer there - dpt.

Easy to try - let us know whether I’ve got it correctly.