Experimental Physics and
Industrial Control System

Jeff Hill <[email protected]> · Thu, 30 May 1996 09:08:01 -0600

the vxWorks Users Group Exploder wrote:
> 
> 
> > We are using an MVE177 with 68060 CPU and seeing greatly degraded FP
> > performance when dealing with numbers whose magnitude are less than
> > FLT_MIN.  This can cause a fabs() call to go from 1 microsec to 53
> > microsec!  Similar degradation occurs for all FP Math operations.
> >
> > Has anyone else seen this and can suggest a workaround.  This is of
> > great concern to us in a high speed (1KHz) application which is math
> > intensive.
> 
> Yes, you are getting underflow exception errors.  You'll have to figure
> out how to turn that off.  This is a problem on 167 boards as well.
> 
> We had first noticed this when comparing performance of applications
> running single-precision vs. double-precision calculations.  We were
> surprised to see that the single-precision version ran slower.
> 
> The reason was that it (a simulated controls system) was doing such a
> good job that the errors in the system were below FLT_MIN, thereby
> generating underflow exceptions.  It took the double-precision
> application much longer to hit its underflow limit, so it ran faster
> before that point.
> 
> We "fixed" the problem by injection noise on the order of FLT_MIN into
> the system, since we couldn't disable exception-handling for
> underflow.
> 

-- 

______________________________________________________________________
Jeffrey O. Hill			Internet	[email protected]
LANL MS H820			Voice		505 665 1831
Los Alamos, NM 87545 USA 	FAX		505 665 5107

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System