Compiling computations to constraints for verified computation

We present a compiler that automates the task of converting high-level code to constraint sets of the form accepted by the Ginger and Zaatar protocols for verified computation. Performing the conversion from high-level code to constraints by hand is prone to human error and therefore not practical for large computations. This paper evaluates the performance of the compiler and the effectiveness of its optimizations on reducing the size of the constraint set. We show that the compiler can produce constraint sets for a number of interesting computations, including DNA sequence alignment of 200 nucleotide sequences and partitionabout-medoids clustering of 100-dimensional data into two clusters.