How to call cudaMalloc from a separate function? - cuda

Question

I'm learning cuda and try to write a function that allocate memory on the device in a similar way to that on the host. For example:
//host
float* allocate1D_float(int size)
{
float* array = (float*)malloc(size* sizeof(float));
if (array==NULL)
{
printf("\n Error allocating memory 1\n");
free(array);
exit(EXIT_FAILURE);
}
return array;
}
float *h_A = allocate1D_float(numElements);
//device
float* alloc_cuda1D_float(int numElements)
{
float *d_array = NULL;
size_t size = numElements * sizeof(float);
cudaError_t err = cudaSuccess;
err = cudaMalloc((void **)&d_array, size);
if (err != cudaSuccess)
{
fprintf(stderr, "Failed to allocate device vector (error code %s)!\n", cudaGetErrorString(err));
exit(EXIT_FAILURE);
}
return d_array;
}
float *d_A = alloc_cuda1D_float(int numElements);
However, nvcc keep saying
error: type name is not allowed
error: expected a ")"
for device function while the host function is ok. Hope you can help me to figure out the issue.
Thanks.

Answer 1

Regarding "type name is not allowed":
You did this correctly:
float *h_A = allocate1D_float(numElements);
But this is wrong:
float *d_A = alloc_cuda1D_float(int numElements);
^^^
This int shouldn't be here
So remove the int right in front of numElements
This of course has nothing to do with CUDA. Your host function call would have given a similar error if you attempted to put int where it doesn't belong in that call.