- The project is encrypted computing, where the encrpyted data is computed
- We provide
- python client code to do encoding, encryption, decryption, decoding and
- server host code and kernel code for receiving data from client and sending it into kernel accelerator to do homomorphic computation
- The implementation is referenced from BFV-RNS
- One should read it before getting into this project
- We optimize individual kernel code used in
src/kernelto achieve nearly the same performance as this paper did using HLS- Lots of code restructure, array partition, unrolling, refer to the
docs/doc.txtfor further details
- Lots of code restructure, array partition, unrolling, refer to the
- We parametrized the code so that it can be easily tested under different settings
- We did not perform system level optimization and intend to do it as future work
├── LICENSE
├── README.md
├── build
│ └── README.md
├── docs
│ └── doc.txt
├── impl_result
│ ├── K_ntt_csynth.rpt
│ ├── NTT_csynth.rpt
│ ├── README.md
│ ├── add_sub_csynth.rpt
│ ├── butterfly_csynth.rpt
│ └── mul_mod_csynth.rpt
├── src
│ ├── client
│ │ ├── README.md // see the README.md to set up
│ │ ├── afterDense100_8192.txt
│ │ ├── afterSquare_8192.txt
│ │ ├── bfv_rns.py // BFV_RNS scheme python code
│ │ ├── bias100_8192.txt
│ │ ├── client.py // client code using BFV-RNS
│ │ ├── cnn8192.txt.zip
│ │ ├── ct.txt
│ │ ├── data.h
│ │ ├── data.py
│ │ ├── defs.py
│ │ ├── dense100_8192.txt
│ │ ├── dense10_8192.txt
│ │ ├── encrypt8192.txt
│ │ ├── gk.txt.partaa
│ │ ├── gk.txt.partab
│ │ ├── gk.txt.partac
│ │ ├── gk.txt.partad
│ │ ├── gk.txt.partae
│ │ ├── inference.py // reference code of how to write encrypted inference
│ │ ├── inference_multiprocess.py // reference code of how to write encrypted inference
│ │ ├── mask8192.txt
│ │ ├── model8192.txt.zip
│ │ ├── rlk.txt
│ │ ├── test.txt.partaa
│ │ ├── test.txt.partab
│ │ ├── test.txt.partac
│ │ ├── test.txt.partad
│ │ ├── test.txt.partae
│ │ └── util.py
│ ├── host
│ │ ├── global.h
│ │ ├── help_functions.cpp
│ │ ├── help_functions.h
│ │ ├── host.cpp // Host code for dealing with networking and kernel interaction
│ │ ├── inference.cpp // HE primitives or even higher level code examples
│ │ ├── inference.h
│ │ └── model_weight.h
│ └── kernel // Kernel functions
│ ├── K_add.cpp
│ ├── K_apply_galois.cpp
│ ├── K_mov.cpp
│ ├── K_mulConst.cpp
│ ├── K_mulWise.cpp
│ ├── K_ntt.cpp
│ ├── lift.cpp
│ ├── mod.cpp
│ ├── mod.h
│ ├── rns.h
│ └── scale.cpp
└── tests
└── README.md
- Follow
README.mdinsrc/client/to recover all files needed- Check out
client.cppto see how to send and receive text.txt,rlk.txt,gk.txt, andct.xtare pickled file, wheretext.txt: class including encoder and encryptorrlk.txt: relinearization keygk.txt: galois keyct.txt: ciphertext
- Check out
- Vitis: include
host.cppand all codes underkernel/and build kernel codes (Emulation-SW, EMulation-HW and Hardware)
- Server run the built code
- run
client.pyundersrc/clientand wait for the result - Change
host/inference.cppto do different tests
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent