Experimental Physics and
Industrial Control System

[email protected] (Jeff Hill) · Fri, 12 Mar 1999 17:56:08 -0700

All,

This message is in reference to "devVXStats" which is not a part of
EPICS base but may nevertheless be in use at a number of EPICS
sites. This code is EPICS device support which is used to monitor an 
IOC's state of health.

Symptom:
Mysterious floating point value corruption in an analog input record
which initially defies any rational explanation.

Cause:
After some considerable effort I have discovered that a floating point
value corruption occurring in our Intel-based IOCs was caused by a 
missing function prototype for the function "sqrt()" in devVXStats.c.
It appears the Intel floating point unit became corrupted in a task
spawned off by devVXStats.c for the purpose of calculating the IOC's
CPU consumption. The "sqrt()" call was used by devVXStats.c to 
consume CPU and thereby estimate CPU consumption in a low priority 
task, and so the code was not interested in the result from "sqrt()". 
The compiler assumed that the function returned type "int" when in fact 
a double precision floating point number was returned, and this 
appears to have corrupted the floating point unit. The result was that
another task in the system produced a not-a-number IEEE floating point
value in the ESLO field of an analog input record during EPICS
initialization. When the record processed the NaN value was propagated
to the VAL field of the ai record.

Resolution:
Here is the patch we applied which eliminated the problem:

D:\hill\leda\share\rp\src>cvs diff -r 1.2 -r 1.1 devVXStats.c
Index: devVXStats.c
===================================================================
RCS file: /usr/local/cvsroot/share/rp/src/devVXStats.c,v
retrieving revision 1.2
retrieving revision 1.1
diff -r1.2 -r1.1
24d23
< #include <math.h>

Jeff
______________________________________________________
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