cv::Matx in cuda kernel

Passing the matrix is not the problem here, the problem is automatic variables (so just normal variables) of cv::Matx declared inside a kernel doesn’t compile.