Rotom is a tensor compiler that autovectorizes high-level programs into optimized homomorphic encryption circuits—searching layouts, applying rolls and rewrites, and cutting rotation cost.
Built for automatic layout assignment
Turn tensor programs into HE-friendly packed vectors automatically instead of hand-packing every operand.
Explore assignments and lightweight layout transforms—including roll-style operators—to unlock better protocols.
Optimize toward practical circuits: evaluations show large wins vs prior autovectorization and hand-tuned baselines.
Peer-reviewed research
Edward Chen, Fraser Brown, Wenting Zheng.
Bridging Usability and Performance: A Tensor Compiler for Autovectorizing Homomorphic
Encryption.
IACR ePrint 2025/1319 · Published elsewhere: Usenix Security 2026.
@misc{cryptoeprint:2025/1319,
author = {Edward Chen and Fraser Brown and Wenting Zheng},
title = {Bridging Usability and Performance: A Tensor Compiler for Autovectorizing Homomorphic Encryption},
howpublished = {Cryptology {ePrint} Archive, Paper 2025/1319},
year = {2025},
url = {https://eprint.iacr.org/2025/1319}
}
Open source — contributions welcome
Rotom lives on GitHub under a BSD-3-Clause license. We encourage the community to experiment, file issues, share benchmarks, and send pull requests—whether you’re extending frontends or backends, tightening layout assignment, improving docs, or polishing the web visualizer.