Efficiency of local fns and closures

["David A. Moon" <Moon@STONY-BROOK.SCRC.SYMBOLICS.COM>]

    Date: Sun, 22 Oct 89 20:10:17 EDT
    From: barmar@Think.COM

    I was doing loops of 100,000 and 1,000,000 repetitions, ran them a couple
    of times and saw consistent results.  My functions were pretty simple, but
    it's possible that I wasn't factoring everything out.  This is similar to
    what I used to determine speed of accessing various types of variable in a
    local function:

    (defun test (x n)
      (flet ((test1 (x) (declare (ignore x))) ; access no variable
	     (test2 (x) x) ; access local variable
	     (test3 () x)) ; access lexical variable
	(without-interrupts (time (loop repeat n do (test1 1))))
	(without-interrupts (time (loop repeat n do (test2 1))))
	(without-interrupts (time (loop repeat n do (test3))))))

    I guess comparing TEST1 to either of the others includes the difference
    between RETURN-NIL and RETURN-STACK.  But I think it should give an
    accurate measurement when comparing TEST2 and TEST3.

More importantly, you are comparing the speed of calling a closure with
one argument and accessing a parent lexical variable, to the speed of
calling a non-closure function with no arguments and accessing a local
variable, and then calling that the difference between accessing a
parent lexical variable and accessing a local variable.  The difference
you are reporting is the sum of three differences: two types of called
function, two numbers of arguments, and two types of variable access.