Feature Request: HERMES

[ludns]

Implementing HERMES (fast ring packing) on GPU would enable LWE ciphertexts to be packed into a CKKS or BGV/BFV ciphertext to benefit from SIMD acceleration. It's currently quite slow on CPU (~10s), but would most probably benefit a lot from GPU acceleration.

Given the really competitive CKKS bit & gate bootstrapping results from this library, LWE->RLWE ring packing would enable batch bootstrapping of AP/GINX style ciphertexts.

[Alisah-Ozcan]

Good idea! But before implementing this, I need to understand how it works mathematically. If there's any reference code or implementation available, feel free to share it. I might consider implementing it and integrating it into the library in the future.