Matthew's Journal

One issue with converting from macros to functions is that these operations (essentially single bit testing and manipulation) are fairly trivial, such that the function call overhead can be many times higher than the cost of the operation itself. If you end up calling them a *lot* (like at the bottom of a nested loop) this overhead can become crippling to performance. Of course whether this is an issue for you or not depends on what exactly the client code is doing - ie you need to test it on some real world problems to find out.

Personally I'd go with the function name decoration with macros to choose a default set with undecorated versions of the names, but make the functions inlineable. If you put a key value in each matrix structure, higher level functions can choose the right low level implementation without needing to be duplicated for each one (or requiring an extra type parameter).