Tools, demo, and test vectors for SMAC.

Compile

Make sure that you have CMake, clang++, and git available in your environment. Maybe a new version of g++ also works, but I didn't use it. Remember to read the Issues when you start to compile.

git clone https://github.com/jarro2783/cxxopts.git ./src/3rd/cxxopts

git clone https://github.com/google/or-tools.git ./src/3rd/or-tools

mkdir build
cd build
CC=clang CXX=clang++ cmake ..
make mac_universal
make smac_demo

To compile the code for performance, run

cd src
clang++ smac.cpp -o smac_perf -march=native -O3
./smac_perf

In lines 39-41 of smac.cpp, you can uncomment other versions and recompile to test SMAC1, SMAC1x4, and SMAC1x8. g++ should also work. Also, on my computer, g++ produces a faster executable than clang++.

Run

The help manual of ./build/bin/mac_universal shows everything you need (./build/bin/mac_universal -h). Examples are as follows. Please remember that these examples are just examples! The runtime depends on the hardware and we did not use only 16 threads. A stronger sigma usually means a longer runtime. For some cases, we even used 48 threads (and about 48G RAM) on AMD Zen to run for a few days. However, sometimes we also found that 12 threads was better than 32 threads. It could be because of the strategy used by OR-Tools. So, play with your hardware and try more different parameters~

To test t+3 for all sigmas in a file called dataset.txt, run the following command that uses 16 threads, outputs a detailed log, and stops after 57600 seconds or when a trail with 19 active sboxes is found.

mac_universal --file ./dataset.txt --round 3 --thread 16 --threshold 19 -t 57600 -l 1

If you don't specify the threshold and timeout, it will run until the optimal trail is confirmed.

mac_universal --file ./dataset.txt --round 3 --thread 16 -l 1

Note that a special format of the file is required! Otherwise you will get nothing or random stuff. The content of the file should look like the following one where each sigma starts with 'sg={'. This is due to some legacy format at the beginning of our research.

sg={0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15}
sg={0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15}

To test t+3 for only one sigma, simply replace the --file option with -s like

mac_universal -s 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 --round 3 --thread 16 --threshold 19 -t 57600 -l 1

Dummy clocks are given by the option --dpi=. The value to this option is a list of indices where the constant msg is compressed. For example, to test SMAC-1/2 and t+5, run

mac_universal -s 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 --round 5 --thread 16 --threshold 19 -t 57600 -l 1 --dpi=1,3

which means the clocks 1 and 3 are dummy rounds taking the constant msg.

To generate patterns of sigmas that have the same differential trails, use -l 2.

To run the demo of SMAC, use ./build/bin/smac_demo.

Test vectors

The whole list of test vectors is in test-vectors.txt.

Issues

• It's normal that some errors appear when you clone OR-Tools. No worries, just try more times.
• It's normal that you spend a little bit long time on compiling the program at the first time. Take a coffee and have a break.
• It's random if your complier complains about abseil. I don't know why and how to solve. I only had this once and a re-compilation solved. However, some people couldn't solve it for a long time.
• I use C++20, so please use a modern C++ compiler. Otherwise, unknown errors could raise. FYI, mine is clang++-15. Systems are Ubuntu-22.04 and Ubuntu-18.04 (WSL version, but should be fine for real Ubuntu).
• As you can see in the CMakeLists.txt, CMake version older than 3.27 is not supported. This is mostly due to OR-Tools. Also, I use new version of CMake. To make sure that everything works correctly, please follow this requirement and update your CMake via Kitware's repo if you're using Ubuntu.
• I only tested mac_universal on my Intel i7-11xxx and the AMD Zen in the data center. Not sure if it works on other platform. Feel free to report any problem not listed above.