Fig. 2

Sequre’s automatic compiler optimization workflow. Sequre transforms an analysis pipeline written in standard Python language into equivalent secure MPC programs through a set of compiler analysis and optimization modules. From the end-user’s perspective, the high-level syntax of Sequre is agnostic to MPC, because the individual operations are automatically replaced with the corresponding MPC routines by Sequre during compilation. The transformed code is then analyzed by the compiler in several passes to optimize the performance of the MPC pipeline. More specifically, Sequre’s optimizations include (1) replacing common code patterns with more efficient equivalents under the MPC setting, (2) restructuring arithmetic expressions to minimize redundant computation and the network overhead incurred by the Beaver partitioning operations, and (3) applying a set of higher-level performance optimizations, such as faster modulus and matrix operations. The resulting program is a high-performance executable ready to be executed by computing parties to carry out the desired analysis. The optimization techniques of Sequre are described in the “Methods” section and Additional file 1: Section 4