🚀 go-pugleaf

Date: Tue 3 Jul 84 19:09:25-PDT
From: Stuart Reges <uw-beaver!REGES@SU-SCORE.ARPA>
Subject: MacPascal Benchmark
To: info-mac@SU-SCORE.ARPA
Office: Margaret Jacks 210, 497-9798

A simple Pascal benchmark is given by the recursive fibonnaci calculation:

	program Fibonnaci (input, output);
	var number: integer;

	function fib (arg: integer): integer;
	begin
	    if arg <= 2 then fib := 1
	    else fib := fib (arg - 1) + fib (arg - 2);
	end;

	begin
	    repeat
		write ('number (0 to stop) ---> ');
		readln (number);
		if number <> 0 then writeln ('answer = ', fib (number));
	    until number = 0;
	end.

This benchmark involves addition and a lot of function calls.  The complexity
of this calculation is (phi)^n where phi = golden ratio = (SQRT(5) + 1)/2.  A
nice property of this benchmark is that every time you increase "n" by 5 you
increase the complexity by a factor of about 11.

Here are some figures I got from three different systems that will be used for
instruction at Stanford next Fall:

	"n"	Waterloo	MacPascal	MacPascal	DEC-20 Pascal
		Interpretter	(integer)	(longint)	Compiled
	-----------------------------------------------------------------------
	15	 50 seconds	 6.7 seconds	  5.9 seconds	
	20	540 seconds	70.5 seconds	 63.5 seconds	 0.2 seconds
	25					702.0 seconds	 1.7 seconds
	30							19.7 seconds

The Waterloo Pascal Interpretter runs on an IBM/PC.  The MacPascal version
using INTEGER could not calculate FIB (25) because it is greater then MAXINT.
I don't know why the LONGINT version should be faster.  That is
counter-intuitive for me.  The DEC-20 Pascal is Hedricks compiler from Rutgers.
The time reported is actual CPU time.  Other calculations were done with a
wrist-watch, but should be accurate to within a quarter second.

This is using the Beta version of MacPascal released last week.
-------