Is tic/toc really reliable?

Hello,

I am comparing 3 ways of calculating a simple stuff (cutting b in half until b<a), and tic/toc gives me VERY improbable results.

Here is my entire code (R2014b):

_____________________________________________

function [W,L,LC]=while_vs_log(a,b)
% test which one of a log2 and a while loop is the fastest to reduce b until it unders a.
if (b<a)  fprintf('b must be >= a\n'); return; end
if (a<=0) fprintf('a must be >0\n');   return; end

%%log compact
ticLC = tic;
LC    = b/2^ceil(log2(b/a));
tocLC = toc(ticLC);

%%log
ticL = tic;
n    = ceil(log2(b/a));
L    = b/2^n;
tocL = toc(ticL);

%%while
ticW = tic;
while (b>a) b=b/2; end
W=b;
tocW = toc(ticW);

%%compare
fprintf('while loop  : %e µs\n',  1e6*tocW);
fprintf('log2        : %e µs\n',  1e6*tocL);
fprintf('log2 compact: %e µs\n\n',1e6*tocLC);

end

_____________________________________________

I called the function several times, changing absolutely nothing between each call.

Have a look at the output: sometimes, the while loop and the log2 computation appear to take the EXACT same time, which is absolutely unlikely.

My question is: how can it be? And how can I trust tic/toc if it sometimes fail? Or do I use it the wrong way?

[William Frane]

When using tic/toc to time relatively simple code, it's typical to run the code N times (for large values of N) between tic and toc and then calculate the average time. See the section "Time Smaller Sections of Code" on this page.

[C.J. Harris]

Also depends on your operating system. Unless you are using a real-time operating system you are going to get jitter which is much larger than the microsecond time-scales you are trying to measure.

[B. C.]

@William : Yes, I will average several run times later, this was just a quick start.

@C.J. : Ok, I had no idea about jitter before.

Thanks to both of you.

[Andrew Newell]

See also Timing code in Matlab.