Home

Yield Prolog Benchmarks

We benchmark Yield Prolog against other Prolog systems. When we optimize Yield Prolog in C# by allowing typed arguments, it can be faster than XSB, but not faster than Yap or Mercury.

All benchmarks use an Intel Core Solo U1500 1.33GHz, 1GB RAM under Windows Vista (Sony Vaio UX390N).

	Mercury csharp grade "optimized code"	Yap Prolog 6.2.0	Yield Prolog C# 2008 "optimized code"	Yield Prolog C# Mono 2.0 "optimized code"	XSB 3.1	SWI-Prolog 5.10.2	Yield Prolog Javascript Firefox 4 "optimized code"	JLog 1.3.4	Yield Prolog Python 2.7 "optimized code"	Yield Prolog IronPython 2.0RC2 "optimized code"
queens "optimized code" (seconds)	0.25	0.68	0.91	1.04	2.34	4.95	20.3	23.1	42.7	91.4
queens "naive code" (seconds)	0.27		7.06	8.93			194		297	1462

Queens

The "queens" benchmark runs the classic Prolog problem to find every solution of an N by N chess board where no queen can attack another. Here is the code for the Prolog benchmarks. (This comes from Glendon Holst's JLog example):

% Find all solutions of an 11 by 11 board.
% The \+ with the fail is a trick to make it find all solutions.
queensBenchmark :- time(\+ (queens(11, _Qs), fail)).
queens(N,Qs) :- rangeList(1,N,Ns), queens3(Ns,[],Qs).

queens3(UnplacedQs, SafeQs, Qs) :-
  selectq(Q, UnplacedQs, UnplacedQs1),
  \+ attack(Q,SafeQs),
  queens3(UnplacedQs1,[Q|SafeQs],Qs).
queens3([],Qs,Qs).

attack(X,Xs) :- attack3(X, 1, Xs).

attack3(X,N,[Y|_]) :- X =:= Y+N ; X =:= Y-N.
attack3(X,N,[_|Ys]) :- N1 is N+1, attack3(X,N1,Ys).

rangeList(M,N,[M]) :- M >= N, !.
rangeList(M,N,[M|Tail]) :- M1 is M+1, rangeList(M1,N,Tail).

selectq(X,[X|Xs],Xs).  
selectq(X,[Y|Ys],[Y|Zs]) :- selectq(X,Ys,Zs).

For the Yield Prolog benchmarks, the "naive code" literally translates the Prolog, and the function arguments are always assumed to be general. The "optimized code" allows the function arguments to be typed. For example, we require the first two arguments to rangeList to be integers so that we don't have to call YP.getValue(). Also, the attack function is only used by queens3 to check if there is an attack, and attack never binds any variables, so the "optimized code" converts it to a normal function hasAttack which returns a boolean.