Qus: BlockIdx and array index mapping

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I am having a problem of mapping gridIdx, blockIdx and array index. I have an array of size x which values will be filled in from after the kernel is run. When I define it in the host, I initialize each values in the array with 1. I print them out after copy the array back to the host, the values are still 1. I am thinking it may be the problem of mapping. Because I am not sure if I am doing right in trasfterring the blockIdx to array index. Could you please take a look at my code and let me know where the problem is? (I am using a 1D block and 1D thread.)

Host

   ...

 Â Â Â int x = 6;

 Â Â Â int* d = (int*)malloc(sizeof(int)*x);

 Â Â Â for(j=0; j<x; j++){

	d[j] = 1;

 Â Â Â } Â Â 

 Â Â Â int* dD = (int*)malloc(sizeof(int)*size);

 Â Â Â CUDA_SAFE_CALL(cudaMalloc((void**)&dD,size*sizeof(int)));

 Â Â Â CUDA_SAFE_CALL(cudaMemcpy(dD,d,size,cudaMemcpyHostToDevice));

 Â Â Â ...

 Â Â Â myKernel<<<dimGrid,dimBlock>>>(par1,par2,par3,...,d);

 Â Â Â 

 Â Â Â CUDA_SAFE_CALL(cudaMemcpy(d,dD,x,cudaMemcpyDeviceToHost));

 Â Â Â for(j=0;j<x;j++){

          printf("\n  d[%d]=%d",j,d[j]);

 Â Â Â }

 Â Â Â ...

Kernel…

   __global__ void myKernel(...,int* d){           

 Â Â Â Â Â Â Â int id = blockIdx.x*bz+threadIdx.x; Â Â Â 

 Â Â Â Â Â Â Â //It may have problems in the id. I would like to fill the values from d[0], d[1], ..... Since I have allocated memory for d in global memory, will the following be written to the global memory?

 Â Â Â Â Â Â Â d[id] = id;

 Â Â Â }

 Â Â Â

Appreciate!!

2

int* dD = (int*)malloc(sizeof(int)*size); should be int *dD;

int id = blockIdx.x*bz+threadIdx.x; should be
int id = threadIdx.x + __mul24(blockIdx.x, blockDim.x); (I guess you have bz = blockDim.x)

CUDA_SAFE_CALL(cudaMemcpy(d,dD,x,cudaMemcpyDeviceToHost)); needs to be
CUDA_SAFE_CALL(cudaMemcpy(d,dD,sizeof(int)*size, cudaMemcpyDeviceToHost));

You were only copying the first 6 bytes (one and a half int)

3

Thx for the reply. But it is still not working. It seems like the kerenl was not even executed. But there is no compile error.

4

did you compile in debug mode? then the CUDA_SAFE_CALL will tell you if the kernel exited with an error.