[Beowulf] ifort and gfortran

[Nifty Tom Mitchell]

On Mon, Feb 23, 2009 at 12:39:33PM -0500, Mark Hahn wrote:
>> When gfortran is used, then we get
>> "segmentation fault" in the middle/during the
>> parameter setting stage of the computation.
>> I would be grateful if you
>> have any idea on how to treat this problem.
>
> compile with gfortran and add -g.  then run inside gdb.
> when compiled, are there any warning messages?

And.

Make sure that the libs you are linking to are compiled with the
same compiler (think ABI compatible, not just API).   Subtle things
like Boolean tests have different underlying representations for True
and False.   Strings, args(), memcpy and other bits... differ between FORTRAN
compilers.  I saw some binary compat code bits get checked into Open
MPI so search there for this topic and build a check list to simplify
debugging with gdb.

The logical tests against True/ False can be interesting to debug because
compilers may reduce some logical expression to an equivalent test depending
on the optimization level.

Some source code bugs surface only at different optimization levels so walk through 
the various levels of optimization.  -O0, -O1, ... etc.

Even "house libs" can be involved, so look there for mismatches too
(GotoBLAS has a lot of permutations for a long list of reasons, some 
of which are ABI).


-- 
	T o m  M i t c h e l l 
	Found me a new hat, now what?