Abstract

Execution profiling is a common aid for improving software performance. Measurements based on instruction and memory access traces show that Prolog programs can spend significant time unifying and backtracking. Trace generation is time consuming and expensive while conventional timing techniques aren't well suited to Prolog. This paper presents an efficient profiling technique for Prolog programs that isolates the time spent in unification and backtracking. The technique employs a unique combination of static and dynamic analysis. Run-time overhead is restricted to updating counters. Execution timings are computed following program execution by combining the flow information provided by the counters with a static timing analysis of the program. The method is suitable for modeling the gross timing behavior of a program and resolves time expenditures to the clausal level.

Metadata

publication

1987 Symposium on Logic Programming: Proceedings, August 31-September 4 …, 1987

year

1987

publication date

1987

authors

Michael M Gorlick, Carl F Kesselman

link

https://scholar.google.com/scholar?cluster=13254096971976318954&hl=en&oi=scholarr

journal

1987 Symposium on Logic Programming: Proceedings, August 31-September 4, 1987, Hyatt on Union Square, San Francisco, California

pages

426

publisher

IEEE Computer Society Press