My CUDA kernel code is not working

Question

I try to make a small code to generate numbers and return the result in array but once I run this code it's not working, I have tried to use Nsight debugger to understand where is my problem but it freezes and closes immediately.

Could you help me please to understand where is the problem in this code?

__global__ void mykernel( int* PF_tmp, int* PL_tmp, int* QF_tmp, int* QL_tmp, 
                              int m[2], int p[5], int q[5], int i, int* n, 
                              int out[10][5], int N)
    {
        int id = blockDim.x * blockIdx.x + threadIdx.x;

        int idx = blockIdx.x;
        int idy = blockIdx.y;

        int w = idx/100;
        int x = idx%100;
        int y = idy;

        int z = threadIdx.x;

        int len = ((i * 2) + 5);


        // Fill PF_tmp & QF_tmp
        if( i > 0){
            for(int k = 0; k < (i * 2); k++)
            {
                p[k]   = PF_tmp[k];
                q[k]   = QF_tmp[k];
            }
        }

        // Fill X
        if( x > 10)
        {
            p[(i*2)] = (x - (x % 10)) / 10;
            p[(i*2)+1] = x % 10;
        }else{
            p[(i*2)] = 0;
            p[(i*2)+1] = x;
        }

        // Fill Y
        if( y > 10)
        {
            q[(i*2)] = (y - (y % 10)) / 10;
            q[(i*2)+1] = y % 10;
        }else{
            q[(i*2)] = 0;
            q[(i*2)+1] = y;
        }

        // Fill m
        p[(i * 2)+2] = m[0];
        q[(i * 2)+2] = m[1];

        // Fill W 
        if( w > 10)
        {
            p[(i*2)+3] = (w - (w % 10)) / 10;
            p[(i*2)+4] = w % 10;
        }else{
            p[(i*2)+3] = 0;
            p[(i*2)+4] = w;
        }

        // Fill Z 
        if( z > 10)
        {
            q[(i*2)+3] = (z - (z % 10)) / 10;
            q[(i*2)+4] = z % 10;
        }else{
            q[(i*2)+3] = 0;
            q[(i*2)+4] = z;
        }

        // Fill PL_tmp & QL_tmp
        if( i > 0)
        {
            for(int k = 0; k < (i * 2); k++)
            {
                p[(len-(i * 2))+k]   = PL_tmp[k];
                q[(len-(i * 2))+k]   = QL_tmp[k];
            }
        }

        if(id<10)
        {
            for(int k =0; k<5; k++)
                out[id][k] = p[k];
        }

    }



    int main()
    {
        cudaError err;
        dim3 blocks(10000, 100);
        dim3 threads(100); 

        int m[2] = {4,5};
        int hst_out[10][5];
        int p[5];
        int q[5];
        err = cudaMalloc((void **)&p, 5);
        err = cudaMalloc((void **)&q, 5);
        err = cudaMalloc((void **)&hst_out, 50);

        mykernel<<<blocks, threads>>>(NULL, NULL, NULL, NULL, m, p, q, 0, NULL, hst_out, 100000000);

        return 0;
    }

You should provide a complete code and use [proper cuda error checking](http://stackoverflow.com/questions/14038589/what-is-the-canonical-way-to-check-for-errors-using-the-cuda-runtime-api). — Robert Crovella, Apr 30 '15 at 03:27

Peter Teoh · Answer 1 · 2015-04-26T12:22:15.317

1

The error very obvious, it is all C programming.

when you declare

        int m[2] = {4,5};
        int hst_out[10][5];
        int p[5];
        int q[5];

now hst_out, p, q are not a pointer, but later it is used as a pointer:

        err = cudaMalloc((void **)&p, 5);
        err = cudaMalloc((void **)&q, 5);
        err = cudaMalloc((void **)&hst_out, 50);

so u should have declare it initially as a pointer instead, eg,

        int *p;

and used it as this way:

        err = cudaMalloc((void **)&p, 5*sizeof(int));

And notice too that the size you have declared is just 5 bytes....whereas I declared it as 5*sizeof(int).

For more example see:

http://cuda-programming.blogspot.sg/2013/03/how-to-avoid-uses-of-cudamalloc-in.html

edited Apr 26 '15 at 12:22

answered Apr 25 '15 at 07:29

Peter Teoh

6,337
4
42
58

I tried before to declare p, q, hst_out as pointers but the same result after run. I want to know why Nsight is not working with my code !! Thanks – al3ndaleeb Apr 25 '15 at 08:20
not sure how your new code looked like, perhaps u want to add as addendum? – Peter Teoh Apr 26 '15 at 12:20
I think I have reached to the problem. It's from my card Cuda 1.1 , It's very old and The cuda toolkit 6.5 is not compatible with this version of Cuda. I will try to upgrade the card. Thanks – al3ndaleeb Apr 26 '15 at 16:29
Actually, cuda toolkit 6.5 can be used with a cc1.1 device. – Robert Crovella Apr 30 '15 at 03:24

My CUDA kernel code is not working

1 Answers1