PulpFHE

Complex Instruction Set Extensions for FHE Processors

What is PulpFHE

PulpFHE is an extension unit with optimized non-linear functions for FHE processors. This project is a deployment of the new instructions set on its sister project, Juliet.

This project introduces several optimizations and extensions to the existing Juliet architecutre.

Prerequisites

1. Install TFHE
2. Update library path if necessary:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<TFHE_INSTALL_DIR>/lib
export LIBRARY_PATH=$LIBRARY_PATH:<CUFHE_DIR>/cufhe/bin
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:<CUFHE_DIR>/cufhe/include

How to Run

1. Navigate to the client directory, open the ppscript.sh file, and edit the wordsize variable to specify the bit size of the data.
   • PulpFHE supports 8, 16, and 32-bit data.
2. Edit the preAux.txt file to input the plaintext data to be encrypted and processed.
3. From the PulpFHE directory, launch the server by running bash runme.sh.
4. From another terminal, navigate to the cloud_enc directory and run the processor by running the interperter using the following command: python3 pulpfhe_interpreter.py <program-name> [wordsize] [number-of-registers].
   • The <program-name> is the name of the assembly file in the Benchmarks directory.
   • The [wordsize] is the size of the plaintext data in bits. The default is 8.
   • The [number-of-registers] is the number of registers that processor will use. The defualt is 128.
   • Please note that when passing <program-name>, just pass the filename and the interperter will automatically fetch it from the Benchmarks directory.
5. To print the output, go to the client directory, and run ./decrypt ../cloud_enc/output.data <wordsize>.

The companion upgraded HE-Java compiler can be found at HE-Java-Compiler

Trustworthy Computing Group