can I allocate CUDA memory in class attribute and then pass to a function on the cuda kernel?
Like this
MY class
class test{
public:
void *cudaData;
test:test(){
cudaMalloc( (void**)&cudaData, 10 * sizeof(double));
}
}
And after some time use this memory at kernel functions?
foo<<< 1, size >>>(my_array, m_cudaData);
its works?