Somewhat Efficient Computations with CKKS - or secCKKS in short - is a fork of the OpenFHE library that implements functional bootstrapping for CKKS mostly following the blueprint given in Alexandru et al., paper titled General Functional Bootstrapping using CKKS. The original OpenFHE library provides FHE capabilities for BFV/BGV, CKKS and TFHE scheme, and I extended it by adding:

• Complex number support for CKKS

• Functional bootstrapping for CKKS for cleartexts up to 8 bits

The main benefit of functional bootstrapping is that it allows to evaluate arbitrary Look-Up Tables (LUTs) while bootstrapping and resetting the noises in ciphertexts to a nominal level. As such, it allows to perform a wider range of computations, including evaluating nice functions such as activation functions for neural networks.

This project is still in active development.

OpenFHE should be compiled with 64 bits integer, as the 128 bits version is still under development.

You can refer to OpenFHE's guide on installation for your specific operating system:

• Linux

• MacOS

• Windows

Some unit test will fail with the current version of secCKKS (serialization). The project was tested on Ubuntu 24.04.1 LTS and compiled with clang++.

I highly recommend that you start by looking at the following example : CKKS functional bootstrapping

The example will be built automatically (if not set BUILD_EXAMPLES to ON with set(BUILD_EXAMPLES ON) in the main CMakeLists.txt) and the executable will be located in build/bin/examples/pke/, so you can run it from the project folder with:

./build/bin/examples/pke/ckks-functional-bootstrapping

The basic syntax of CKKS functional bootstrapping is the following: to evaluate an 8 bits LUT of the identity function with first order Hermite interpolation:

int p = pow(2, 4);
auto f = [](double x) -> double { return x; };
int hermite_order = 1;
auto ctxtResult = cc->EvalFuncBootstrap(ctxt, f, p, hermite_order);

Please pay attention to the fact that EvalFuncBootstrapSetup should be called instead of EvalBootstrapSetup to use functional bootstrapping.

For performances (especially multi-value bootstrapping), you should use the option:

export OMP_MAX_ACTIVE_LEVELS=4

• Fixed complex support for encoding

• Fixed complex constant/ciphertext addition/multiplication

• Added complex support for EvalPoly (linear)

• Added support for real->complex function interpolation via EvalChebyshev (linear and PS)

• Added Hermite coefficients calculation and EvalHermiteFunction for Hermite interpolation for up to order 3

• Added StC-first Bootstrapping (experimental)

• Added Functional Bootstrapping with use of both real and imaginary components

• Added Multi-Value Functional bootstrapping

• Added Tree base method for Functional Bootstrapping (WIP)

• Removed noise estimation and noise estimation unit tests

• Added the FBTS option to round at decryption

• Parallelization of (MV) Functional Bootstrapping with OpenMP

This project is licensed under the BSD-2 License - see the LICENSE file for details.

This project was developed at CEA-List.