[FLASH-USERS] Flash 4 issues with mpich

[Klaus Weide]

On Sat, 2 Aug 2014, Pierre Gourdain wrote:

> Hi,
> 
> I am trying to run examples from the Flash4 code and all the ones I tried so
> far (Sedov,Noh,CurrentSheet) do not work with MPI when more than two processes
> are involved. The common errors are zero density at some points in the grid or
> an illegal free memory. Less than or equal to two work fine (including mesh
> refinement) and I can read the outputs in Visit.
> 
> All libraries have been compiled using the newest  intel fortran and c++
> compilers, including mpich. I have many other codes which run with mpich using
> this compiler. Does any one have some experience in resolving this issue?

Pierre,

What you describe (i.e., FLASH working correctly with numprocs = 1 and 2, 
but failing with numprocs > 2 ) seems be a new problem that we have not 
encountered before.  There is no place in the FLASH code that I'm aware of 
where the code path depends excplictly on whether numprocs > 2.

We do test the FLASH code regularly with Intel compilers, but not the 
latest versiosn - we are mostly using 12.1.3 (as well as 12.1.0 and 10.1
in some older configurations).

Maybe your problem depends on the compiler version?
Does the problem persist when you simplify your complier flags - in 
particular, by lowering the optimization level (and maybe removing some 
advanced optimization options) in Makefile.h ?

It would be useful if you could send a log file, together with 
at least Makefile.h and Flash.h, for a simple one of your problems 
that fails, together with error messages generated.

Klaus