Subject: ccm3.2 run-time error on RS6K and DEC
From: Erik Kluzek (erik@cgd.ucar.edu)
Date: Tue Mar 18 1997 - 17:08:20 MST
Erik:
I've encountered the same problem. here are some messages that address the problem.
-Steve Ghan
> The compilation line for running on an RS6K should read
>
> > FFLAGS = -c -g -qautodbl=dblpad -qflttrap=overflow:zero:invalid:enable -qnosave
>
> or
>
> > FFLAGS = -c -g -qrealsize=8 -qintsize=4
>
> rather than
>
> > FFLAGS = -c -g -qrealsize=8 -qintsize=4 -qflttrap=overflow:zero:invalid:enable -qnosave
>
> which immediately aborts with a segmentation error.
>
> Although one would normally run without floating point trapping (which
> is faster) so that option two could be used, serious errors cannot be
> trapped with option two. So the preferred method is option one, which
> can be exercised with or without floating point trapping.
My apologies for letting so much time lapse before responding to your email.
On our IBM machines, the -qautodbl=dblpad flag results in an immediate
segmentation fault as you describe above for option 3. The reason I believe
is that the integer data on the boundary datasets are written as 4-byte
entities, not 8-byte. Therefore the -qrealsize=8 -qintsize=4 flags are
mandatory.
Regards,
Jim Rosinski
CCM Core Group
This archive was generated by hypermail 2b27 : Thu Jun 01 2000 - 09:10:53 MDT