Advent of Code 2017 optimized C solutions. The AoC creator issued a challenge to solve each of the puzzles within 250 ms. I decided to take that a step further and do the whole month within that limit, and managed to get it down to 195 ms.
Here are the timings from an example run on a 4.20 GHz i7-7700K CPU. The total is greater than the sum of the individual days because of rounding.
Day 1 28 μs
Day 2 15 μs
Day 3 1 μs
Day 4 63 μs
Day 5 36,131 μs
Day 6 82 μs
Day 7 117 μs
Day 8 60 μs
Day 9 66 μs
Day 10 87 μs
Day 11 55 μs
Day 12 100 μs
Day 13 3 μs
Day 14 4,574 μs
Day 15 119,350 μs
Day 16 127 μs
Day 17 122 μs
Day 18 206 μs
Day 19 21 μs
Day 20 798 μs
Day 21 17 μs
Day 22 28,369 μs
Day 23 231 μs
Day 24 3,241 μs
Day 25 1,270 μs
------------------
Total 195,145 μs
The input parsing is ugly not because of optimization, but rather as a result of continuing my self-imposed challenge during December to avoid using any Perl modules (or in the case of my C implementation, library functions.) I later did away with that restriction because it was obscuring the intent of my code too much, but I never did bother going back to rewrite the input parsing.
In some of the solutions (especially where the naive solution is already blindingly fast) I decide to get a little silly. So when you see that SIMD bitonic sort network in Day 4, I did it because I thought it was interesting, not because it was particularly fast.