[ale] Trapping a fpe?

[Justin Russell]

FYI,

Here is some info I found on trapping a floating-point exception under
Linux on Intel (works differently on Alpha).  I tried to use signal to
trap SIGFPE, but for some reason (I'm still not clear on this), SIGFPE
is not produced when, for example, a divide-by-zero is encountered.  It
just produces an "Inf" and keeps on going (the old nan nan nan
syndrome). I did find this reference, however, and the following works
great:



Floating point trapping

Build with gcc -c trapfpe.c and link with trapfpe.o. Ultimately, the
code could be made somewhat more portable by using the C9X facilities
(fesetenv() declared in fenv.h if special
masks are defined).
(Adapted from: info g77 'Trouble' 'Missing Features' 'Floating-point
Exception Handling')
                           trapfpe.c
 #include <fpu_control.h>
 static void __attribute__ ((constructor)) trapfpe (void)
 {
 fpu_control_t cw =
 _FPU_DEFAULT & ~(_FPU_MASK_IM | _FPU_MASK_ZM | _FPU_MASK_OM);
 _FPU_SETCW(cw);
 }



A core file is dumped and the code dies (with a signal 8) when a fp
exception is encountered making it MUCH easier to debug with gdb.  The
web page I found this on also has a lot of other great info on various
*NIX compilers.  Here is the link:

http://www.nag.co.uk/projects/Julius/Useful/CompilerTricks.html

I hope this will help others.

--
-----------------------------------------------
 Justin W. Russell
 Engineer Senior, Aerodynamics/CFD
 Lockheed Martin Aeronautics Company
 86 South Cobb Drive
 Marietta, GA  30063-0685

 Tel:  (770) 494-6303
 Fax:  (770) 494-3055
-----------------------------------------------


--
To unsubscribe: mail majordomo at ale.org with "unsubscribe ale" in message body.