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>
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
makethis builds the executeables for 3 versions:
- 2047m1l226
- 2047m4l205
- 2047m6l194
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-summaryBy 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
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 arithmeticEach 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.outIf DENABLE_CT_TESTING=ON, checkct versions of the executable are created
for all versions, which can be validated with valgrind.
e.G.:
valgrind ./main/checkct-2047m6l194.mainWe 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
./greedywombats.py