Scalar shifts and the like are compiled in line by your compiler. A library function call would take longer than any reasonable compiler generated code. The main thing to watch is that you specify the architecture you want to optimize; for example, if you specify 'gcc -march=pentium4', the compiler will know about the P4 preference for a sequence of adds over a left shift or multiply.
Another thing to watch is that you don't force unnecessary conversions between short int and int data types, and don't over-use short int.
Only a vector library offers enough advantage for special instructions, to overcome the overhead of function calls.
