hyesungji / ckks-gpu-core Goto Github PK

This repository includes GPU implementations of the core functions for CKKS (e.g., base conversions, NTT, key-switching, etc.) to reproduce the results in the paper Over 100x Faster Bootstrapping in Fully Homomorphic Encryption through Memory-centric Optimization with GPUs submitted to TCHES 2021.

The implementations included largely contribute to CRYPTOLAB INC. and HEAAN.

• NVIDIA GPU (compute capability over 7.0)
• cmake v3.18 or over.

Tested under ubuntu 18.04 with g++ 9.3.0 and NVIDIA V100.

Check out this notebook.

Alternatively, use the included dockerfile (make sure NVIDIA Container Toolkit is installed).

docker build -t ckks-gpu-core .
docker run --gpus all --rm ckks-gpu-core build/benchmark/Benchmark

Wonkyung Jung [email protected]