TFHE-rs v1.4: GPU Performance Breakthrough and More

TFHE-rs v1.4 significantly enhances the performance and usability of Zama's open-source fully homomorphic encryption (FHE) library across CPU, GPU, and HPU backends. It introduces a range of improvements, including a twofold boost in GPU operation speed and a 45% reduction in HPU integer operation latency, alongside introducing friendlier parameter APIs through MetaParameters structures. The update also optimizes keyswitching operations using 32-bit integers, offers a new ReRand feature for improved security, and enhances features such as key-value store and multi-bit noise squashing. Furthermore, the GPU backend now benefits from a reworked integer logarithm and encrypted random generation, along with a new multi-GPU algorithm that quadruples the speed of 64-bit integer division. On the HPU side, clock frequencies have been increased, and new SIMD operations have been introduced, boosting the performance of tasks like ERC20 transfers. These advancements make TFHE-rs more efficient and accessible, facilitating the application of privacy-enhancing technologies in real-world scenarios.