search

corelab-src / dacapo

Automatic Bootstrapping Management Compiler for FHE
MIT License
7 stars 0 forks source link

readme

hecate-compiler

Hecate (Homomorphic Encryption Compiler for Approximate TEnsor computation) is an optimizing compiler for the CKKS FHE scheme, built by Compiler Optimization Research Laboratory (Corelab) @ Yonsei University. Hecate is built on the top of Multi-Level Intermediate Representation (MLIR) compiler framework. We aim to support privacy-preserving machine learning and deep learning applications.

Installation

Requirements

Ninja   
git  
cmake >= 3.22.1  
python >= 3.10  
clang,clang++ >= 14.0.0

Install MLIR

git clone https://github.com/llvm/llvm-project.git
cd llvm-project
git checkout llvmorg-18.1.2
cmake -GNinja -Bbuild \
  -DCMAKE_C_COMPILER=clang \
  -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release \
  -DLLVM_ENABLE_PROJECTS=mlir -DLLVM_INSTALL_UTILS=ON \
  -DLLVM_TARGETS_TO_BUILD=host \
  llvm
cmake --build build
sudo cmake --install build
cd ..

Optional : Install Directory to maintain multiple versions or a debug build

git clone https://github.com/llvm/llvm-project.git
cd llvm-project
git checkout llvmorg-18.1.2
cmake -GNinja -Bbuild \ 
  -DCMAKE_C_COMPILER=clang \
  -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release\
  -DLLVM_ENABLE_PROJECTS=mlir -DLLVM_INSTALL_UTILS=ON \
  -DLLVM_TARGETS_TO_BUILD=host -DCMAKE_INSTALL_PREFIX=<MLIR_INSTALL>\
  llvm
cmake --build build
sudo cmake --install build
cd ..

Install SEAL

git clone https://github.com/microsoft/SEAL.git
cd SEAL
git checkout 4.0.0
cmake -S . -B build
cmake --build build
sudo cmake --install build
cd ..

Optional : Install Directory to maintain multiple versions or a debug build

git clone https://github.com/microsoft/SEAL.git
cd SEAL
git checkout 4.0.0
cmake -S . -B build -DCMAKE_INSTALL_PREFIX=<SEAL_INSTALL>
cmake --build build
sudo cmake --install build
cd ..

Build Hecate

git clone <this-repository>
cd <this-repository>
cmake -S . -B build 
cmake --build build

Optional : Install Directory to maintain multiple versions or a debug build

git clone <this-repository>
cd <this-repository>
cmake -S . -B build -DMLIR_ROOT=<MLIR_INSTALL> -DSEAL_ROOT=<SEAL_INSTALL>
cmake --build build

Configure Hecate

python3 -m venv .venv
source .venv/bin/activate
source config.sh

Install Hecate Python Binding

pip install -r requirements.txt
./install.sh

Tutorial

Trace the example python file to Encrypted ARiTHmetic IR

hc-trace <example-name>

e.g., 

hc-trace ResNet

Compile the traced Earth IR

hopts <pars|dacapo> <waterline:integer> <example-name> <library-name> <hardware>

e.g., 

hopts dacapo 40 ResNet HEAAN GPU

This command will print like this:

Estimated Latency: 13.980823 (sec)
Number of Bootstrapping: 19

Compile the traced Earth IR and Check the optimized code

hbt <pars|dacapo> <waterline:integer> <example-name> <library-name> <hardware>

e.g., 

hbt dacapo 40 ResNet HEAAN GPU

This command will print like this:

Estimated Latency: 13.980823 (sec)
Number of Bootstrapping: 19
===---------------------------====
  ... Execution Time Report ....

You can see the optimized code in "$hecate-compiler/examples/optimized/dacapo/ResNet.40.earth.mlir"

If you see an error message like "error: 'earth.bootstrap' op failed to infer returned types",\ just wait as it is in the normal compilation process.

Test code

We are currently conducting experiments using a modified version of the HEAAN library. However, due to licensing restrictions, we cannot offer interfaces for testing our compiled code directly with the HEAAN library. Therefore, we kindly ask you to attach a library that supports bootstrapping if you want to test compiled code.

In addition, we have included the SEAL library, which does not natively support bootstrapping. We've provided a wrapper (SEAL_HEVM.cpp) that modifies the bootstrapping algorithm into the decryption+encryption form. Please note that while this method allows you to test the results, it is not recommended from a privacy perspective. If you want to test with the SEAL library, refer to the example below:

First, need to generate trace code that matches SEAL parameters: \ In /examples/benchmarks/ResNet.py, "nt" : 2**16 ----> "nt" : 2**14

hc-trace ResNet
hbt dacapo 40 ResNet SEAL CPU
hc-test dacapo 40 ResNet SEAL CPU

This command will print like this:

======================================
---------------Option-----------------
compiler: dacapo
benchname: ResNet
waterline: 40
library: SEAL
device: CPU
---------------Result-----------------
latency: 53.7260615
rms: 0.00095152200605
======================================

Currently, this compiler only supports code generation and testing for the SEAL version only with ResNet. For other benchmarks, compilation with the HEAAN viersion is possible, but program testing is not due to library license. (Ongoing Research)

Papers

DaCapo: Automatic Bootstrapping Management for Efficient Fully Homomorphic Encryption\ Seonyoung Cheon, Yongwoo Lee, Ju Min Lee, Dongkwan Kim, Sunchul Jung, Taekyung Kim, Dongyoon Lee, and Hanjun Kim
33rd USENIX Security Symposium (USENIX Security), August 2024. [Prepublication]

ELASM: Error-Latency-Aware Scale Management for Fully Homomorphic Encryption [abstract]
Yongwoo Lee, Seonyoung Cheon, Dongkwan Kim, Dongyoon Lee, and Hanjun Kim
32nd USENIX Security Symposium (USENIX Security), August 2023. [Publication]

HECATE: Performance-Aware Scale Optimization for Homomorphic Encryption Compiler[IEEE Xplore]
Yongwoo Lee, Seonyeong Heo, Seonyoung Cheon, Shinnung Jeong, Changsu Kim, Eunkyung Kim, Dongyoon Lee, and Hanjun Kim
Proceedings of the 2022 International Symposium on Code Generation and Optimization (CGO), April 2022. [Publication]

Citations

@INPROCEEDINGS{lee:hecate:cgo,
  author={Lee, Yongwoo and Heo, Seonyeong and Cheon, Seonyoung and Jeong, Shinnung and Kim, Changsu and Kim, Eunkyung and Lee, Dongyoon and Kim, Hanjun},
  booktitle={2022 IEEE/ACM International Symposium on Code Generation and Optimization (CGO)}, 
  title={HECATE: Performance-Aware Scale Optimization for Homomorphic Encryption Compiler}, 
  year={2022},
  volume={},
  number={},
  pages={193-204},
  doi={10.1109/CGO53902.2022.9741265}}
@INPROCEEDINGS{lee:elasm:sec,
  title={{ELASM}: Error-Latency-Aware Scale Management for Fully Homomorphic Encryption},
  author={Lee, Yongwoo and Cheon, Seonyoung and Kim, Dongkwan and Lee, Dongyoon and Kim, Hanjun},
  booktitle={{32nd} USENIX Security Symposium (USENIX Security 23)},
 year={2023},
 address = {Anaheim, CA},
 publisher = {USENIX Association},
 month = aug
}
@INPROCEEDINGS{cheon:dacapo:sec,
  title={{DaCapo}: Automatic Bootstrapping Management for Efficient Fully Homomorphic Encryption},
  author={Cheon, Seonyoung and Lee, Yongwoo and Kim, Dongkwan and Lee, Ju Min and Jung, Sunchul and Kim, Taekyung and Lee, Dongyoon and Kim, Hanjun},
  booktitle={{33rd} USENIX Security Symposium (USENIX Security 24)},
 year={2024},
 address = {Philadelphia, CA},
 publisher = {USENIX Association},
 month = aug
}