search

mupq / pqm4

Post-quantum crypto library for the ARM Cortex-M4
296 stars 75 forks source link

Add context info to benchmark.md #359

Open franslundberg opened 1 month ago

franslundberg commented 1 month ago

Great to see the benchmark results in benchmark.md. Would be great to have some more info on how the tests where performed. On what platform and with what compiler settings. For example, for flash size, was the -Os flag used? Not super-easy to find out by looking at the code that generated it. At least not easy for me.

rpls commented 1 month ago

Good idea actually. The info is somewhat scattered throughout the README.md, and not up-to-date. I'll think about a way on how to automatically include this with the scripts.

FYI: benchmarks are created with the benchmark.py script without any additional optimization flags to the script. For pqm4, this results in compilation with -O3. By default, the compilation is done in one translation unit, so similar to, but not quite, LTO. For cycle count-based benchmarks, the target boards are clocked low-enough, that the flash doesn't need any wait-states. Default board right now is the Nucleo-L4R5ZI. The used compiler changes but we aim to use the "official" GNU GCC Toolchain provided by ARM in the most up-to-date version.

franslundberg commented 1 month ago

Thanks for the info. Much appreciated.