A High Performance Implementation of Prolog

We discuss an efficient implementation of the Warren Abstract Machine (WAM) [12] in detail. Special attention is given to data formats, memory layout, WAM optimizations and code generation techniques. A final section describes some hardware considerations for even higher performance execution. Currently the compiler produces code that runs at approximately 900,000 logical inferences per second (LIPS) on a single processor of an IBM 3090 using the naive reverse benchmark. Using several of the yet unimplemented optimizations, we expect this figure to top one million LIPS.