[RegCNET] Floating point exception / Arithmetic exception error

[Alexander Bryan]

Dear RegCNET,

I am getting a bizarre error when running the "icbc" preprocessing script
-- here is a snippet of the terminal output:

...
 Read in fields at Date: 1979-06-15 00:00:00 UTC
 Read in fields at Date: 1979-06-15 06:00:00 UTC
 Read in fields at Date: 1979-06-15 12:00:00 UTC
/var/spool/torque/mom_priv/jobs/17552728.nyx.arc-ts.umich.edu.SC: line 16:
84122 Floating point exception./icbc US25km_hourly.in

This error suggests I'm dividing by zero somewhere, or something else
illegal. I ran with the debugger (gdb), which pointed me to the "qs"
calculation in the humid1fv subroutine in PreProc/ICBC/mod_humid.f90 --
line 122 in RegCM-4.3.4.

....
 Read in fields at Date: 1979-06-15 00:00:00 UTC
 Read in fields at Date: 1979-06-15 06:00:00 UTC
 Read in fields at Date: 1979-06-15 12:00:00 UTC

Program received signal SIGFPE, Arithmetic exception.
0x0000000000487818 in mod_humid::humid1fv (t=..., q=..., p3d=..., ni=192,
nj=145, nk=38) at ./mod_humid.f90:123
122               qs = sep2*satvp/(p3d(i,j,k)-satvp)   ! SAT. MIXING RATIO
Missing separate debuginfos, use: debuginfo-install
glibc-2.12-1.149.el6_6.9.x86_64 zlib-1.2.3-29.el6.x86_64

I put a print statement in before the qs calculation to print out i, j, k,
p3d(i,j,k), and satvp to figure out which (i, j, k) value it crashes and
which quantity is zero.

This is what's bizarre... When I format the print statement output (e.g., print
'(3I5, 2F10.3)', i, j, ...) versus leave it unformatted (e.g., print *, i,
j, ...), the icbc script seems to stop at different values of i, j, and k:

formatted: 80, 64, 37
unformatted: 80, 75, 26

Why would simply changing print statement formatting cause the script to
seemingly crash in two different places?

Any insights on why else I might get a floating point (arithmetic)
exception error?

Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ictp.it/pipermail/regcnet/attachments/20151110/826dec4b/attachment.html>