[cam-users] Anybody have CAM running on an Opteron?

[Jim Rosinski]

On Wed, 14 May 2003, Tim Carlson wrote:

> First of all I have less than an hour of experience with CAM ;)
> [...]
> So I get a segfault. If I run cam inside of gdb and do a backtrace, I get
> this information
> 
> Program received signal SIGSEGV, Segmentation fault.
> 0x00000000401867ad in swapn8b ()
> (gdb) bt
> #0  0x00000000401867ad in swapn8b ()
> #1  0x000000004018925d in ncx_getn_double_double ()
> #2  0x000000004018dbbb in getNCvx_double_double ()
> #3  0x000000004018ddec in getNCv_double ()
> 
> If I redo the configuration and enable debugging
> 
> Enable compiler debugging options? [n]: y
> Enabling compiler debugging options.
> 
> the code fails to compile and I get this error
> [...]
> PGF90-S-0000-Internal compiler error. getsname: bad stype for       1 (/home/tim/cam1/models/lnd/clm2/src/biogeophys/BalanceCheckMod.F90: 116)
> PGF90-S-0000-Internal compiler error. getsname: bad stype for       1 (/home/tim/cam1/models/lnd/clm2/src/biogeophys/BalanceCheckMod.F90: 213)
>   0 inform,   0 warnings,   2 severes, 0 fatal for balancecheck

A couple of suggestions: First try running at a very low resolution
(e.g. T5).  If this works it is an indication that the seg fault is the
result of exceeding some system memory limit.  If T5 datasets are available
with the model code this should be easy to try.

If memory appears to be a problem I'd increase it with things like "limit
stacksize unlimited", "limit memoryuse unlimited", etc. (in csh).  This only
affects the soft limits.  If the limits are still too small there are
boot-time options to increase the hard limits.

Your backtrace doesn't show what CAM routine was calling the netcdf library
routine that seg faulted.  As long as it isn't one of the BalanceCheck
routines, I would just compile those 2 routines without the debug option,
relink and procede from there.  Note that pg products provide a window-based
debugger, called pgdbg, that you can use.  Those "internal compiler error"
messages from pgf90 are really annoying, but since the compiler doesn't
complain about the routines when debug is false I wouldn't worry about them.

Hope at least some of this helps.

Regards,

Jim Rosinski