You mention multiple failures here - but include log for only one case. Ideally lanugage compilers (c,c++,fortran) should interoperate with each other - but without it - one has to know the 'compatibility libraries' to use.
-lstdc++ is a gnu/c++ library. For PGI - the library names could be diffferent. --with-cxxlib-autodetect=1 - which is configure default - tries to determine this list - but with cray modules - this causes multiple listing of all libraries that are setup via modules - sometimes causing grief - hence you are using --with-cxxlib-autodetect=0 on cray builds. Can you send me the following info? cd src/benchmarks cc -v sizeof.c CC -v sizeof.c Also you can try --with-cxxlib-autodetect=1 - and see if it works. Satish On Thu, 14 May 2015, Mark Adams wrote: > We have Titan working but we have a problem that we need lstdc++ to get > PETSc to configure. Otherwise it fails to build fortran. This seems to > only be needed for GNU. This is for PGI. If we remove this C++ works, > otherwise there is a missing external. I guess I can just deploy two > version, on for fortran and one for C++ (I have a C++ and a fortran code > that I support). > > > On Tue, May 12, 2015 at 9:36 AM, Barry Smith <[email protected]> wrote: > > > > > Remind me again how much the US tax payers spend on this machine each > > year. > > > > > > Begin forwarded message: > > > > *From: *Mark Adams <[email protected]> > > *Subject: **Re: [petsc-users] configure error on Titan with Intel* > > *Date: *May 12, 2015 at 11:12:55 AM CDT > > *To: *Barry Smith <[email protected]> > > *Cc: *Jed Brown <[email protected]>, petsc-users <[email protected]>, > > "David Trebotich" <[email protected]>, "D'Azevedo, Ed F." <[email protected]> > > > > > > > >> Notes: ./configure ran fine and detected -lhwloc in some standard > >> system install location under normal circumstances it couldn't just > >> disappear for a different example. > >> > >> > > I configured in an interactive shell, so on a compute node. I tried to > > 'make ex56' on a login node, as usual. So I am guessing that it would have > > worked if I had made it a compute node. They have an inconstancy, I'm > > guessing. I can try it all on a compute node .... > > > > Mark > > > > > >> > >> > / > >> > > >> > On Fri, May 8, 2015 at 5:51 PM, Jed Brown <[email protected]> wrote: > >> > Satish Balay <[email protected]> writes: > >> > > >> > > Also - Perhaps with intel compilers - the recommended blas is > >> > > something other than ACML? [like MKL?]. Something to check.. > >> > > >> > MKL (and anything built using the Intel compiler) has the "run slower on > >> > AMD" feature. For any given operation, ACML won't necessarily be > >> > faster, but at least it's not intentionally crippled. For most of what > >> > PETSc does, any such MKL/ACML difference is irrelevant. > >> > > >> > <configure.log><make.log> > >> > >> > > > > >
