October 2, 2025
Fully Homomorphic Encryption (FHE) enables evaluating computations over ciphertexts without revealing the encrypted data. A key challenge when compiling for FHE is dealing with control flow: the usual semantics require the party evaluating the program to learn something about each branch, which violates the basic premise of secure computations. Standard “multiplexer”-based compilers solve this problem by generating programs that require the evaluator to execute every possible path, obviating the need to know which path is correct. Thus, they provide oblivious semantics at the cost of producing unwieldy circuits that are difficult to effectively optimize. We present COIL, an FHE compiler that addresses the control flow challenge by restructuring how control flow in circuits is interpreted, replacing the use of multiplexers with path forests, and enabling a series of path-dependent optimizations that result in more efficient realizations of complex branching kernels. We demonstrate on a variety of benchmarks that COIL outperforms other state-of-the-art FHE compilation techniques, often by more than an order of magnitude.
About Raghav Malik
Raghav Malik is a (recently-defended) sixth-year PhD candidate working with Milind Kulkarni at Purdue University. His research focuses on building better compilers and programming languages for privacy-preserving computation.