The instruction decode mechanism is a chain of "e--||(expression), e--|| ..." expressions. At the point where e-- is zero the expression will be evaluated.
The program makes heavy use of the equivalence between "index[pointer]" and "pointer[index]", presumably to save brackets.
switch(a) { case x: do1(); break; case y: do2(); break; default do3(); break; }
becomes
These structures can be seen in the minimized version (and a few other IOCCC winners :-).This code likely doesn't optimize quite as well. Several cute things you can do with switch statements cannot be translated so easily.