[petsc-dev] OpenMPI is a menace

[Bryce Lelbach]

On 2012.11.05 11.01, Christopher Bergstrom wrote:
> On 11/ 5/12 10:55 AM, Jed Brown wrote:
> >I just bumped the thread on this over on their mailing list.
> >
> >Educating users about LD_LIBRARY_PATH is a losing battle, in my opinion.
> (I wholly agree and I wish -rpath matched Solaris -R behavior.
> preload can be a slightly better alternative in some cases)  I'm
> still not getting exactly what's being done -
> 
> What  --prefix= is used to build openmpi?
> Where is it installed?
> Can the installed location be known at compile time?

So, I run into problems like this when building GCC on clusters that I do not
have root access to. GCC developers have told me that they won't add proper
rpath support because "LD_LIBRARY_PATH is the unix way".

I've found that simple bash scripts that add the proper flags/set LD_LIBRARY_PATH
can be effective, albeit hacky. For something like OpenMPI, this would probably
be a little trickier.

You could, hoever, write a simple compiler wrapper for the compiler that you use
to build openmpi. Something like:

	#! /bin/bash
	g++ -Wl,-rpath=/my/rpath $@

Then, just do CC=/path/to/wrapper when you build OpenMPI.

-- 
Bryce Adelstein-Lelbach aka wash
STE||AR Group, Center for Computation and Technology, LSU
--
860-808-7497 - Cell
225-578-6182 - Work (no voicemail)
--
stellar.cct.lsu.edu
boost-spirit.com
llvm.linuxfoundation.org
--
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20121104/9fb9df68/attachment.sig>