Thanks, Brian,
Reconfiguring LAM using "--without-memory-manager" fixed our problem.
We don't have Myrinet or InfiniBand. So no memory management is not a
concern.
Best regards,
Lily
On Mon, 2005-03-21 at 19:23, Brian Barrett wrote:
> On Mar 21, 2005, at 6:18 PM, Lily Li wrote:
>
> > We've encountered SIGSEGV in LAM's malloc() function problem with LAM
> > 7.1.1.
> > The same application works fine with LAM 7.0. I know our application
> > is not
> > very clean regarding using realloc() ( we may give a null pointer to
> > realloc()).
> > Is there a way that we don't use LAM's wrapper functions for the
> > malloc() etc. ?
> > e.g. how do we build our applications not to use LAM's malloc() ?
>
> That's kind of worrisome - usually ptmalloc shouldn't get in the way
> (and should handle realloc with a NULL first pointer just fine). To
> answer your question, you can give the option --without-memory-manager
> to configure to build without any internal memory management. Of
> course, you will not be able to use Myrinet or InfiniBand with LAM in
> that case.
>
> If you are using Linux, you might want to compile with --with-purify
> and use Valgrind or Purify to make sure you don't have some latent
> memory bug that can easily be cleaned up. The --with-purify option
> slows LAM down slightly by zeroing out all structures before use. LAM
> uses many structures with fields that are not always applicable (or may
> be filled in on the receiving side), so without the purify option, LAM
> will cause read from uninitialized warnings that aren't critical.
>
> Hope this helps,
>
> Brian
|
