The software.

Linux and Unix programmers do not need a separate timer. They already have one in the shell. If luajit and sincosEuler9.lua are in the path, they can just type

time luajit sincosEuler9.lua

or the like at the prompt, and the time command will run the rest of the line and report back how much time it took. This will work wrong for us Windows programmers at the Dos box prompt, because our time command has a completely different meaning. I have therefore written a program in the Windows version of FreeBasic. The program is timer.bas and it looks like

dim a as single, b as single

a=timer
shell(command$)
b=timer

print (b-a);" seconds"

I have compiled it to an executable called timer.exe for our use. We can just type

timer luajit sincosEuler9.lua

to find out how long the sincosEuler9.lua program took. On my computer, using LuaJIT 2.0.0-beta1, and running three times, it took 5.402344 seconds, 5.078857 seconds, and 5.137695 seconds, or so says timer. That is much speedier than the previous LuaJIT, and Mr. Mike Pall deserves everybody’s thanks. The LuaJIT link is http://luajit.org/. My program is sincosEuler9.lua and it looks like

-- This program uses Euler's algorithm to solve the simultaneous
-- ordinary differential equations for the sine and cosine.  The
-- interval of integration is from zero to one.

local s=0 -- sine starts at zero
local c=1 -- cosine starts at one
local n=1e9 -- 1e9 steps
local h=1/n -- size of one step
local sNew -- a temporary location
local cNew -- another temporary location

	for j=1,n do
	sNew=s+c*h
	cNew=c-s*h
	s=sNew
	c=cNew
	end

print( s,c )

Back in 1958, when I first programmed on a computer, the great language was Fortran. I have written a program in it, sincosEuler9.f, to do the same thing that the Lua program does. It looks like

	double precision s,c,n,h,j,sNew,cNew

	n=1 000 000 000
	h=1.0d0/n
	s=0
	c=1

		do 17 j=1,n
		sNew=s+c*h
		cNew=c-s*h
		s=sNew
		c=cNew
17		continue

	print *,s,c
	end

Having used MinGW Fortran, that is g77, to compile the source, I ran the binary three times, and got 6.770508 seconds, 6.580322 seconds, and 6.564697 seconds. I think that I used the appropriate command line modifiers for speed, -O3 and -fomit-frame-pointer, when I compiled, but Fortran seems to be slower than LuaJIT for this algorithm.

Come to think of it, here is timer.zip containing all these files, including the index.htm file the reader is now reading. All the files in the zip are in the public domain. Of course, FreeBasic is not in the public domain. Its license is GNU General Public License (GPL), GNU Library or Lesser General Public License (LGPL). Mr. Mike Pall is the owner of LuaJIT. His copyright notice says Copyright (C) 2005-2009 Mike Pall.

The rant.

Before I wrote the timer.bas program, I believed everything I read on the Web about program times. I believed that Java was always slower than C, that interpreters were always slower than compiled code, that Lisp had been souped up to full hardware speed, that Lua was just a toy, that professional software was optimized better than amateur software, that you got what you paid for, that “extremely fast” meant nothing was faster, that compiled-ahead-of-time native code would always go faster than jit, and that code compiled by a later version of a compiler would be no slower than code compiled by an earlier version of the same compiler. Then I wrote the timer.bas program and compiled it and tried it out, and my youthful illusions evaporated.

Revision date and e-mail address.

This file is revised 2 March 2012. Comments both constructive and destructive come to me, Harold Kaplan,

       at     dot
smtw2gh  gmail   com

Harold Kaplan’s programming.htm