All,
I've packaged lam and xmpi and use a single spec file to create the
relevant binary rpms. That said both lam and xmpi build perfectly when
using gcc. However, when I use the pgi or intel compilers the build of
lam works like a champ BUT the build of xmpi fails in different ways for
each compiler.
I've attached an xmpi_intel.out and xmpi_pgi.out that illustrates the
failures.
It feels like xmpi doesn't have the correct magic to build properly with
pgi (v4.0) or intel (v7.0) based lam mpiCC et al. (But I could be wrong =)
As my .out files suggest; succesfully building xmpi with intel compilers
feels much closer than pgi; as for what is broken I've yet to dig really
deep.
Intel:
At first glance it would appear the intel build just isn't using
-I/opt/intel/compiler70/ia32/include BUT.. if you add that the build of
xmpi_kiviat.cc fails much more horribly than not having access to sin and
cos protoypes from a <math.h>.
PGI:
I haven't _really_ fought with this as much as the intel... so as for why
the relevant files aren't being generated in order for the mv to succeed
I'm unsure. intel doesn't even attempt to mv files.. so I'm not sure what
games pgi is playing or causing. All the warnings with pgi do not make me
warm and fuzzy either.
Any help would be greatly appreciated.
--
Mike Snitzer msnitzer_at_[hidden]
Linux NetworX, Inc. http://www.lnxi.com
|
