Re: [FLASH-USERS] FW: 3D RM Problem

Shuang -

On Tue, Apr 29, 2003 at 04:05:05PM -0400, Shuang Zhang wrote:
> Thanks Tomek!
>
> Unfortunately, we still haven't got Prism
> working for MPI debugging on our system.

You would need to the code under debugger anyway. To help in the
curent case, can you restart the code from some recent checkpoint file
and demonstrate that the code crashes in the same way? If so, you
should obtain a checkpoint file just prior to the crash - this will
shorten runtime necessary to reproduce the problem. The final step is
to run the code under control of dbx and get the line
number/instruction where the code actually crashes. For this you would
need to run unoptimized version with some set of debug flags (to
include symbolic information required by debugger).

> Here is what "dbx rm_3d core" gives:
>
> ************************************************
> detected a multithreaded program
> t@1 (l@1) terminated by signal FPE (invalid floating point operation)
> 0x00059dd4: hydro_1d+0x1e1c: ld [%sp + 0x26c], %o4
> (/opt2/SUNWspro/bin/../WS6U2/bin/sparcv9/dbx) where
> current thread: t@1
> =>[1] hydro_1d(0x151aeba0, 0x151aec28, 0x151af420, 0x151aed38, 0x151aedc0,
> 0x151aee48), at 0x59dd4
> [2] hydro_sweep(0x2a0588, 0x3198b68, 0x3199dd8, 0x10, 0x50, 0x6), at
> 0x63994
> [3] hydro_3d(0x54114, 0xffbef2f0, 0x65, 0x235536, 0x54, 0x23328a), at
> 0x55ae0
> [4] evolve_(0x23381e, 0x233232, 0x54, 0x23323b, 0x0, 0x54), at 0x541bc
> [5] MAIN_(0xaae61, 0xffbef464, 0xffbef470, 0x0, 0x0, 0x0), at 0x48b4c
> [6] main(0x2, 0xffbef464, 0xffbef470, 0x29ec00, 0x0, 0x0), at 0x489f4
> **************************************************
>
> Dose this look like memory overflow?

No, it does not. But "invalid" usually indicates sqrt(negative
argument). My guess is that for some reason either density or
pressure gets negative and the code crashes in computation of sound
speed. Well, I speculate here, obviously.

I would recommend you to wait for another 10 days and try your problem
again but with FLASH2.3 - it is just around the corner - and it has
some options that may help to run your simulation smoothly.

Tomek

-- 
Tue, 15:19 CDT (20:19 GMT), Apr-29-2003
_______________________________________________________________________________
   Tomasz Plewa                                      www:   flash.uchicago.edu
   Computational Physics and Validation Group        email: tomek@uchicago.edu
   The ASCI Flash Center, The University of Chicago  phone: 773.834.3227
   5640 South Ellis, RI 475, Chicago, IL 60637       fax:   773.834.3230
_______________________________________________________________________________