Static allocation problem on Mac

I’m trying to move some data into constant memory on a MacBook with a 650M card, and getting error 63, “OS call failed or not supported”. I have made a teensy little minimum program that demonstrates the problem:

#include <iostream> 

__device__ double testDouble; 
double hostDouble = 5.1;  

int main (int argc, char** argv) {
  cudaError_t err = cudaMemcpyToSymbol(testDouble, (void*) &hostDouble, sizeof(double));
  std::cout << cudaGetErrorString(err) << std::endl; 

  return 0;
}

This runs fine on my desktop (CUDA 4.2, C2050) but crashes on a MacBook (CUDA 5.0, 650M), with the error code 63 as noted. Does anyone know why this might be so, or what I can do to avoid it?

I just tried this out on my MBP Retina 15", and got “no error”. I’m not sure what the problem could be. Do other CUDA sample programs work on your Mac?