Is correct to declare __shared__ a variable (not pointer/array) in CUDA (C) kernel?

Question

In my kernel function I defined:

__shared__ float shared_var;

which is not a pointer/array (as it is shown in examples NVIDIA docs). Then I initia