It's in the makefile in that repo:

    	g++ cpp.cpp -std=c++11 -Wall -O2 -march=native -o cpp

I was originally using -O3 but then someone found that -O2 is actually faster in this case.

When visited was changed from a vector<bool> to a bitset<T>, why is it not passed by reference anymore?

The most common call in the benchmark is to `getLongestPath<16>`. Passing 16 bits by value is cheaper and easier for the compiler to optimize (i.e., no aliasing) than passing by reference. I'd expect graphs with many nodes to behave differently.