I am not sure static class members are allowed in device code. CUDA only allows a (generous, but not complete) subset of C++ features.
Appendix D of the toolkit 3.0 programming guide lists the supported C++ features, but does not mention static class data, so by implication, it’s unsupported.
Even if it was supported, you’d have an interesting question… is the static class data global across the device or local to the block?