For efficiency, Prolog code is typically compiled to abstract machine code before it is run.

Many different abstract machine architectures and variants have been proposed for efficient execution of Prolog programs. These include:

• WAM, the Warren Abstract Machine
• TOAM, an abstract machine used in B-Prolog.
• ZIP, used for example as the basis for the VM of SWI-Prolog
• VAM, a research architecture developped in Vienna.

All widely used Prolog interpreters use argument indexing to efficiently select suitable clauses.

Users can typically rely on at least first argument indexing, meaning that clauses can be efficiently told apart by the functor and arity of the outermost term of the first argument. In calls where that argument is sufficiently instantiated, matching clauses can essentially be selected in constant time via hashing on that argument.

More recently, JIT indexing has been implemented in more systems, enabling dynamic indexing on any argument that is sufficiently instantiated when the predicate is called.

Virtually all Prolog systems implement tail call optimization (TCO). This means that predicate calls that are in a tail position can be executed in constant stack space if the predicate is deterministic.

Tail recursion optimization (TRO) is a special case of tail call optimization.