Starting from a programmatic representation of a Markov decision process (MDP) in the PRISM syntax, we examine the task of synthesizing a policy in the form of a program for the MDP. The PRISM syntax allows us to specify MDPs concisely by partitioning the state space into regions with similar actions and transitions. In this work, we restrict ourselves to a subclass of two dimensional deterministic gridworlds partitioned into regions along linear predicates. We present an algorithm to synthesize programmatic policies to achieve reachability objectives which exploit the symmetries present in the specification of the MDP. Our programs use memory to track subgoals and navigate between the edges of regions to provide a concise representation of a policy. Our main result is a proof of an upper bound on the size of the synthesized programs. We also provide an implementation of our synthesis algorithm which is evaluated on randomly generated instances of gridworlds.