How to use JCUDA to call CUBIN?

Question

I'm pretty new to trying to use JCUDA (http://www.jcuda.de) but need to work out how to call a CUBIN file from my Java code so I can use CUDA to perform calculations on large arrays of data in a Java program, as part of an assignment.

I've tried to follow the "JCudaDriverCubinSample" file but am really struggling to see how I can modify that example code to work for my own CUBIN file (note: the example does run properly however).

For example, I'm trying a simple kernel generated from the .CU file:

__global__ void multiply_array( int *a, int *b, int *c, int N ) {
    int tid = blockIdx.x * blockDim.x + threadIdx.x;    
    if (tid < N)
        c[tid] = a[tid] * b[tid];
}

This appears to build into a CUBIN file properly, but I can't seem to understand what changes I need to make to the example code to try and make this run from JCuda using the driver bindings.

Is anyone able to point me towards the solution, or towards material that would explain the code required a little more clearly? I'm finding the documentation on the JCuda website pretty sparse (but am really not adverse to reading, if there is a good resource to try and learn from).

Thanks!

Do you use JCudaDriverCubinSample.java? Did you replace the strings in `prepareCubinFile` and `cuModuleGetFunction` accordingly? Did you perpare the device memory as needed? Did you use the correct launch parameters? As JCuda is really just a binding of the original Cuda functions please refer to the official documentation found at http://developer.nvidia.com/object/cuda_3_2_downloads.html (programming guide, reference manual, examples, etc.). Oh and btw. could you describe in detail what errors you get? — Jonas Bötel, Mar 07 '11 at 18:50
I can alter the 'prepareCubinFile' and the 'cuModuleGetFunction' no problem, but I am struggling to understand how I am passing the parameters to the function, if that makes sense? The setting of certain sections of memory on device is never clearly explained, and I don't understand how I should alter this to pass two input arrays for example — obfuscation, Mar 07 '11 at 20:30

How to use JCUDA to call CUBIN?

0 Answers0