My solutions to the Advent of Code of 2023, 2024 and 2025 in C++20.
Some of the days have only the solutions of part 2, as I wrote it over the part 1 solutions.
When a day has two solutions is because the part 1 and part 2 solutions are very different and I have not written the part 2 over the part 1 solution.
I haven't uploaded the inputs as mentioned in the AoC web
I've tried to make them run as fast as possible, so there is the script called perf.py that compiles everything and runs it.
If in every folder of every problem there is an input file with the name of the program, for example 1.in for day 1 of any year, in linux you could run the script given that you have g++ with C++20 and Python.
Every program is compiled with -O2 and the time measured is execution time, not launch program time, with the AOC.h utility that I made
There is also a new option in the perf.py script: --profiled, that makes use of the -fprofile-generate and -fprofile-use options of the g++ compiler, but since the profiling is made with the real problem input, I don't know if that could be considered cheating in some sort of way. Anyways I've created that option since it reduces the time a lot on some branch-heavy, low-memory days, like day 10 2025
With my Ubuntu 24 WSL2 32gb ram 13th gen i7 machine the sum of the execution times of every program is consistently under 1.2 seconds, getting under a second with profiling enabled
I've also run the script in an old laptop with Ubuntu 22 WSL2 installed and this are the results:
With --profiled (<2.5 seconds) |
Without profiling (<3 seconds) |
|---|---|
 |
 |
