dCTIDH: Fast & Deterministic CTIDH

This repository contains auxiliary material for the paper "dCTIDH: Fast & Deterministic CTIDH".

Authors:

• Fabio Campos <campos@sopmac.de>
• Andreas Hellenbrand <andreas.hellenbrand@hs-rm.de>
• Michael Meyer <michael@random-oracles.org>
• Krijn Reijnders <krijn@q1q1.nl>

Overview

This implementation is based on the code from secsidh.

Building

We tested our code with GCC 12 on Debian 12. Furthermore, the implementation makes use of the ADX (ADOX and ADCX) instructions, so you need an Intel Broadwell/AMD ZEN CPU or newer.

# Only necessary first time (generally)
mkdir build && cd build
cmake ..

# If you want with instrumentation for constant-time behavior testing, 
#the default value is OFF. Valgrind development files are used for this build option.
cmake -DENABLE_CT_TESTING=ON ..

# Building
make

This builds the executeables for 3 versions:

• 2047m1l226
• 2047m4l205
• 2047m6l194

benchmarking

Automated Benchmarking

The project includes automated benchmark targets that make it easy to run and analyze benchmarks for all enabled parameter sets:

# Run benchmarks for a specific parameter set
make benchmark-ctidh-2047m1l226

# Run all benchmarks and display a summary
make benchmark

# Show just the summary of previously run benchmarks 
make benchmark-summary

By default, benchmarks run with 100 iterations, which will take several hours. You can change this by setting the SECSIDH_BENCHMARK_RUNS option:

# Configure with 5 benchmark runs
cmake -DSECSIDH_BENCHMARK_RUNS=5 ..

The benchmark results are saved to files in the build directory:

• Raw logs: benchmark-ctidh-<param_set>.log
• Analysis results: benchmark-ctidh-<param_set>-analysis.log

Manual Benchmarking

You can also run benchmarks manually using the executable options: when in build:

usage: 	
    ./main/ctidh-2047m1l226.main                            # for a quick test
	./main//ctidh-2047m1l226.main -bact [number of runs]    # run benchmark for the action
	./main//ctidh-2047m1l226.main -bfp [number of runs]     # run benchmark for fp arithmetic

Each version contains benchmarking tools for the action, as well as the finite-field arithmetic, which can be used with -bact, resp. -bfp.

The action benchmarks can be analyzed using the analyze_bench.py script:

./main/ctidh-2047m1l226.main -bact 100 > bench_action.out
python3 ../analyze_bench.py < bench_action.out 

The analyze_bench.py script supports different output formats:

# Default grid format for terminal viewing
python3 ../analyze_bench.py < bench_action.out

# CSV format for importing into spreadsheets
python3 ../analyze_bench.py --format=csv < bench_action.out

# LaTeX format
python3 ../analyze_bench.py --format=latex < bench_action.out

constant-time check

If DENABLE_CT_TESTING=ON, checkct versions of the executable are created for all versions, which can be validated with valgrind.

when in build:

cmake -DENABLE_CT_TESTING=ON ..

make  # creates all versions

make checkct-2047m1l226.main  # for single version
make checkct-2047m4l205.main
make checkct-2047m6l194.main

# run valgrind test
valgrind ./main/checkct-2047m1l226.main
valgrind ./main/checkct-2047m4l205.main
valgrind ./main/checkct-2047m6l194.main

Remark: There seems to be a Valgrind issue with some combinations of GCC versions and modern CPUs due to missing AVX instructions. See the details here. If you encounter issues, we recommend trying again with GCC 12, as used in our setup.

parameter search and new primes

We use greedy to find optimal configurations. The script explors the keyspace for primes with 151 to 226 ell_i and 1 to 18 batches. We recomend to split up the search, as this will take a while (up to a month using 4 jobs with 48 threads each).

cd scripts/greedy/
./greedywombats.py

To add the files needed for a new prime as part of a new parameter set, use the scripts in scripts/new_prime

Licenses

Code in this repository that does not indicate otherwise is placed in the public domain. The code in this repository is based on secsidh, which uses the same license as this work: secsidh License (CC0)