My first program with CUDA need some help

[naveen290]

I have written a simple program which does nothing but takes in an array of 10 float’s returns an array of 10 float’s. In the kernel i am just trying to store some arbitrary value or just the thread index. I am still just trying to get a feel of CUDA. Unfortunately when i see the resultant array i see only two values and rest of it is garbage. I am attaching the code snippet, some help in debugging the issue will be of great help to me.

include <cuda_runtime.h>

include <cuda.h>

global void kernel(float *arrS)

{

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

arrS[idx] = idx;

}

extern “C” void launch(float *source)

{

// set the grid and block sizes

dim3 dimGrid(1); 

dim3 dimBlock(10); 

kernel<<<dimGrid,dimBlock>>>(source);

}

void main()

{

float arr[10] = {0,1,2,3,4,5,6,7,8,9};

float *arrS = NULL;

float dst[10];

cudaMalloc((void**)&arrS, sizeof(float) * 10);

cudaMemcpy((void*)arrS,(const void*)arr,10,cudaMemcpyHostToDevice);

launch(arrS);

cudaMemcpy((void*)dst,(const void*)arrS,10,cudaMemcpyDeviceToHost);

}

I am right now using CUDA 2.3 on 8600 GT on a windows machine.

[naveen290]

Can any body help me with this issue i am still trying to debug the problem. :(

[SPWorley]

Can any body help me with this issue i am still trying to debug the problem. :(

The transfer size is too small. The size is in bytes, not words. You did this properly for the malloc but not the transfer.

So change to

cudaMemcpy((void*)arrS,(const void*)arr,10*sizeof(float),cudaMemcpyHostToDevice);

launch(arrS);

cudaMemcpy((void*)dst,(const void*)arrS,10*sizeof(float),cudaMemcpyDeviceToHost);

[naveen290]

The transfer size is too small. The size is in bytes, not words. You did this properly for the malloc but not the transfer.

So change to

cudaMemcpy((void*)arrS,(const void*)arr,10*sizeof(float),cudaMemcpyHostToDevice);

launch(arrS);

cudaMemcpy((void*)dst,(const void*)arrS,10*sizeof(float),cudaMemcpyDeviceToHost);

Thank you SPWorley. That was really a silly mistake from me but was really struggling with this one. Other stuff was really running fine but not this