[FLASH-BUGS] Running flash on a SUNblade.

Richard Bower r.g.bower at durham.ac.uk
Fri Jan 24 09:15:19 CST 2003


Hi,
	we are really keen to get FLASH running on our 24 processor 
SUNfire and 128 processor SUNblade cluster.

It comiles happily, and we almost have it running successfully.
Would you like our Makefile?   We are interested in doing some bench 
marking as well.

The compilation has thrown up a few problems which I'll describe below. 
Have these been seen elsewhere?


1)  line 197 of init_global_parms.F90  contains the following line:

   call get_parm_from_context(global_parm_context, "eos_mode", eos_mode)

however, eos_mode is declared as a parameter. At run time, the programme 
segmenation faults:

Process 0: signal SIGSEGV (Segmentation Fault) in procedure 
"get_int_parm_from_context" at "runtime_parameters.F90":562 

We have solved this by commenting out line 197.



2)  The appears to be a problem with the code using variables which have 
not been explicitly initialised.

In hydro_1d.F90 we had to explicitly add the lines at the begin of the
subroutine (insert at line 142)

    ul = 0.0 ; ur = 0.0
    pl = 0.0 ; pr = 0.0
    vl = 0.0 ; vr = 0.0
    utl = 0.0 ; utr = 0.0
    uttl = 0.0 ; uttr = 0.0
    rhol = 0.0 ; rhor = 0.0
    gamcl = 0.0 ; gamcr = 0.0
    gamel = 0.0 ; gamer = 0.0
    xnl = 0.0 ; xnr = 0.0

this appears to fix the problem, but its not clear whether we also need 
to modify other routines.

Presumably this has not shown up on other compilers because the variable 
is set to 0 when it is declared.  This does not happen on the SUN.

Our worry is that there maybe other variables which are implicitly 
assumed to be zero, but for which the side-effects are less obvious.



Thanks for your help!  Richard



-------------------------------------------------------------------------
Dr Richard Bower,                                      Tel. 0191 374 2146
Dept. of Physics, University of Durham,                Fax. 0191 374 7465
South Road, Durham, DH1 3LE, UK			   R.G.Bower at durham.ac.uk
-------------------------------------------------------------------------
 




More information about the flash-bugs mailing list