I have an app running so good in a machine with
Cuda compilation tools, release 4.0, V0.2.1221
and when I run this same app in a machine with
Cuda compilation tools, release 5.0, V0.2.1221
my app don’t work well. When I call the kernel and i want recover the results with
HANDLE_ERROR( cudaMemcpy (dev_failures, data->Failures, size * sizeof(unsigned int)*Nb_Errors_To_Simulate, cudaMemcpyHostToDevice) );
the function don’t copy nothing, what I get is only ‘garbage’, i.e, random numbers.
Can somebody tell me how I can solve this problem?
A last thing, when I compile my app in the version 5 of nvcc, im getting this warnings
warning: a host function(“Node::cuda_MyFunction”) redeclared with device, hence treated as a host device function
but in the code this function is declared as host device
Why I get this warning?