Building

First clone HElib with git.

git clone [email protected]:homenc/HElib.git
cd HElib
git checkout [3e337a6]

Then, copy github_20241003.patch into HElib/ and apply the changes by

git apply github_20241003.patch

You need to replace the path path_of_fatboot.cpp in HElib/src/extractDigits.cpp by the directory path of fatboot.cpp. A directory called saved_ZZX needs to be created in this directory to store the computed digit extraction polynomials.

Then build and install HElib following the instructions in INSTALL.md, some dependencies like NTL and gmp may be needed. Remember to build in either Release or RelWithDebInfo mode.

You need to change the helib_DIR variable in CMakeLists.txt to the correct installation path of HElib.

Finally, switch to the directory of fatboot.cpp and build the test program by

cmake -S . -B build
cmake --build build

Running

If the building is successful, it will produce a binary file build/fatboot. Running ./build/fatboot -h will display the following help message

Usage: ./build/fatboot [i=<arg>] [h=<arg>] [t=<arg>] [newbts=<arg>] [newks=<arg>] [thick=<arg>] [repeat=<arg>] [rad2=<arg>] [baseline=<arg>] [cache=<arg>] [s2cFirst=<arg>]
  i        index of the chosen parameter set [ default=0 ]
  h        hwt of encapsulated key [ default=0 ]
  t        parameter t used in new bts [ default=0 ]
  newbts   if new bts is used [ default=0 ]
  newks    if new ks is used [ default=1 ]
  thick    if thick bts is used [ default=0 ]
  repeat   number of tests [ default=5 ]
  rad2     force to use radix-2 [ default=0 ]
  baseline to test baseline [ default=0 ]
  cache    use cache or not [ default=1 ]
  s2cFirst if SlotToCoeff is the first step in general bootstrapping [ default=0 ]

• Three parameter sets are available (given in Table 2 of the paper with ID I, II, III). By setting the argument i to an integer in 0 to 2, the corresponding parameter set is chosen.
• The Hamming weight of the encapsulated secret key needs to be assigned through the argument h. h is 26 for parameter sets I and II, and is 24 for parameter set II.
• Keep argument t as the default value 0.
• Always set newbts=1 so that the optimization of Ma et al. is used.
• Set thick=1 to test general bootstrapping. Set thick=0 to test thin bootstrapping.
• Argument repeat indicates the number of tests to run. The performance data will be averaged over these tests.
• Argument rad2 controls the decomposition style when $p\equiv 3\bmod 4$. rad2=0 means Bruun style while rad2=1 means Radix-2 style.
• Set baseline=1 to test the baselines for general or thin bootstrapping.
• Set cache=1 to precompute $\kappa(i)$ in linear transformations as double-CRTs. Refer to Section 5.2 for its discussion.
• Set s2cFirst=1 to evaluate SlotToCoeff as the first step in general bootstrapping. Refer to the comment below Table 5 for details.