[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Question

From: Dirk Laurie
Subject: Re: Question
Date: Mon, 8 Mar 1999 09:48:25 +0200 (SAT)

John W. Eaton wrote:
> On  5-Mar-1999, Emil Zagar <address@hidden> wrote:
> | I am using Octave and I have a question about function 'flops'.
> | 
> | Is it somehow possible to count floating point operations in Octave (like
> | in Matlab with 'flops' function)?
> Why do you want a flops count?
Researchers are often faced with comparing their own, super-duper,
best-ever, bleeding-edge algorithms to some outdated, mediocre,
inefficient and inaccurate technique that everybody has been using
up to then.  Claims that the new technique is faster are sometimes
substantiated by running one's own algorithm on one's own computer
and comparing it with the timings from the original authors' paper.

Octave users are naturally sophisticated enough not to be fooled
by the claim that Algorithm B, which takes 5 seconds on a 500MHz
Pentium III, is faster than algorithm A, which in 1968 needed 
5 minutes on an IBM 360 Model 40.  Other users might not be, and 
referees tend to reject numerical comparisons based on timing 
evidence.  Flops counts are better for this, but not the Matlab
kind that counts additions too.  


reply via email to

[Prev in Thread] Current Thread [Next in Thread]