cudaErrorInvalidValue while copying from host to Device

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

I’m quite new to CUDA, though I have good background in programming.

I’m working on a project where I use matrices of increasingly smaller size. That’s how it looks like:

// host

float *u[MAX_GRID_LEVELS];

float *v[MAX_GRID_LEVELS];

// device

float *uCUDA[MAX_GRID_LEVELS];

float *vCUDA[MAX_GRID_LEVELS];

int nx[MAX_GRID_LEVELS];

int ny[MAX_GRID_LEVELS];

int levels = 0;

// ... initialization stuff ...

nx[levels] = 65;

nx[levels] = 65;

// allocate memory on CPU for each grid

// and initialize to 0

while(( nx[levels] > 2 ) && (ny[levels] > 2) && (levels < MAX_GRID_LEVELS))

{

      size_t sizeCPU = nx[levels] * ny[levels] * sizeof(float);

u[levels] = new float[ sizeCPU ];

      v[levels] = new float[ sizeCPU ];

memset( u[levels], 0, sizeCPU );

      memset( v[levels], 0, sizeCPU );

nx[levels+1] = (nx[levels] + 1)/2;

      ny[levels+1] = (ny[levels] + 1)/2;

levels++;

}

// Allocate memory on GPU for each grid

// and initialize grids with 0.

for(int k = 0; k <= levels; k++) {

      size_t sizeGPU = nx[k] * ny[k] * sizeof(float);

cudasafe( cudaMalloc( (void**) &uCUDA[k], sizeGPU ), "cudaMallocU" );

      cudasafe( cudaMalloc( (void**) &vCUDA[k], sizeGPU ), "cudaMallocV" );

cudasafe( cudaMemcpy( uCUDA[k], u[k], sizeGPU, cudaMemcpyHostToDevice ), "cudaMemcpyU" );

      cudasafe( cudaMemcpy( vCUDA[k], v[k], sizeGPU, cudaMemcpyHostToDevice ), "cudaMemcpyV" );

cudasafe( cudaMemset( *uCUDA, 0, sizeGPU ), "cudaMemsetU" );

      cudasafe( cudaMemset( *vCUDA, 0, sizeGPU ), "cudaMemsetV" );

}

cudasafe is a small routine used to catch and print any cudaError generated

void cudasafe( cudaError_t error, char *message)

Well, the cudaMalloc doesn’t report any error, while I get a CUDA error 11 (cudaErrorInvalidValue) on the first cudaMemcpy.

I can’t figure out why. is the way I allocate memory for each grid correct? could that be the cause of the error?

Please, any help would be appreciated