Is it possible to create a function that will work in parallel on blocks, where inside it is parallelized on threads?

Question

GPGPU

SM2 card version.
Question to people with experience, how can I solve this problem.
There is an image -> you need to divide it into grids -> into blocks into blocks -> into threads.
That is, you need to somehow run the __global__ or __host__ functions from under another function of the same kind.
Is there a solution to such a problem?
in compiler error

"calling a __global__ function from a __global__ function is only allowed on the compute_35 or above"

what to do on compute_20?

Answer 1

Buy a new card. The manual clearly states that recursion is not supported.