WAM-CL, 3x slower SBCL, 17x faster than CLISP, potentually

/*

NOTES:

* WAM-CL currently produces code 6 times slower than the handwritten code

* Handwritten Prolog is 2-3 slower than SBCL

* If WAM-CL becomes fast as handwritten code,
** it will be 17 times faster than CLISP
** it will be 6 times faster than ECL




(defun fib (n)
  (if (> n 1)
    (+ (fib (- n 1))
       (fib (- n 2)))
    1))

*/

% WAM-CL
fibc(A, K) :- !,
        B=[[bv(n, [A|_])]],
        sym_arg_val_envc(n, A, C, B),
        >(C, 1, D),
        (   D\=[]
        ->  sym_arg_val_envc(n, A, E, B),
            -(E, 1, F),
            fibc(F, I),
            sym_arg_val_envc(n, A, G, B),
            -(G, 2, H),
            fibc(H, J),
            +(I, J, L),
            K=L
        ;   K=1
        ).
fibc(_, _) :- '<<=='(fibc(n),if(n>1, fibc(n-1)+fibc(n-2), 1)).


% HANDWRITTEN

fibp2(N, F) :-
        N =< 1
        -> F = 1
        ;
        N1 is N-1,
        N2 is N-2,
        fibp2(N1, F1),
        fibp2(N2, F2),
        F is F1+F2.





% SBCL 1.3.1
% * (time (fib 38))
% 1.264000 seconds of total run time (1.264000 user, 0.000000 system)

% YAP-Prolog (Hand written)
% ?- time(fibp2(38,O)).
% 3.124 CPU in 3.148 seconds ( 99% CPU)

% YAP-Lisp (WAM-CL)
% ?- time(fibc(38,O)).
% 20.184 CPU in 20.340 seconds ( 99% CPU)

% SWI-Prolog (Hand written)
% ?- timel(fibp2(38,O)).
% 24.558 CPU in 24.826 seconds (99% CPU, 18027611 Lips)

% ECL 15.3.7
% > (time (fib 38))
% run time  : 25.516 secs (real time : 26.290 secs)

% CLISP 2.49
% (time (fib 38))
% 53.0 sec.

% SWI-Lisp (WAM-CL)
% ?- time(fibc(38,O)).
% 113.043 CPU in 114.324 seconds (99% CPU, 15665558 Lips)