TFHE-rs v0.5: Detecting Overflows, Running on GPU and More
Blog post from Zama
The latest update of TFHE-rs introduces significant advancements aimed at enhancing performance and reliability in homomorphic encryption. Key features include GPU acceleration, which leverages CUDA to boost cryptographic operations for unsigned integers, and the implementation of overflow detection in homomorphic operations that captures overflow scenarios for improved reliability, albeit with a slight performance trade-off compared to standard operations. Additionally, the update marks a commitment to backward compatibility, facilitating smoother transitions for developers working with different versions of TFHE-rs. The update also includes enhancements like optimized keyswitch operations and accelerated addition for vectors of ciphertexts, achieving substantial speed improvements. As part of the improvements, TFHE-rs now supports easier manipulation of large integers through the Residue Number System (RNS) and introduces a homomorphic circuits simulator for faster and more efficient debugging.