benchmarking confusion

[Stephen Weeks]

> How big is the vector?

10,000 ints, i.e. 40,000 bytes.  I have appended the code below.

Here are the times for another machine as well.

MhZ		400	400	733	
		eponym	nobill	starlinux
local-flatten 	time	time	time
true		7.3	7.1	3.7
false		7.8	7.4	3.5

> If it is large enough, then you can get pimped by
> caching effects.
...
> Could this explain things?

Except I don't understand why I would consistently see one executable do better
than the other on one machine, given that the assembly code (for the hot loop)
is the same.


structure Main =
   struct
      open Vector
	 
      fun rev v =
	 let
	    val n = length v
	 in
	    tabulate (n, fn i => sub (v, n - 1 - i))
	 end

      fun doit () =
	 let
	    val v = tabulate (10000, fn i => i)
	    fun loop n =
	       if n < 0
		  then ()
	       else
		  if 0 = sub (rev (rev v), 0)
		     then loop (n - 1)
		  else raise Fail "bug"
	 in loop 10000
	 end
   end

val _ = Main.doit ()