Version of Mpi and fblaslapack

[Satish Balay]

petsc-2.2.1 was released in Aug2004 - hence tested with compilers
available at that time - whereas petsc-2.3.3 is tested with compilers
available to us as of May2007 [some old compilers, some new compilers]

So it is possible that there are some issues with certain
combinations.  We don't have a document that lists these
compatibilities or incompatibilities.

If you encounter problems - let us know at petsc-maint at mcs.anl.gov -
we can help you resolve them.

We generally hope that users migrate to newer versions of PETSc, so
that we don't have to deal with such issues that come up with aging
old versions of software..

Satish


On Fri, 12 Oct 2007, Trivedi, Jagruti CIV 470000D, 474300D wrote:

>  
> 
> Yes, I am sorry. I did misunderstood the situation.
> 
> Best Regards,
> 
> ...Jagruti
> -----Original Message-----
> From: owner-petsc-users at mcs.anl.gov
> [mailto:owner-petsc-users at mcs.anl.gov] On Behalf Of Barry Smith
> Sent: Friday, October 12, 2007 10:42
> To: petsc-users at mcs.anl.gov
> Subject: RE: Version of Mpi and fblaslapack
> 
> 
>   You are totally misunderstanding the situation! PETSc should work with
> virtually ANY release of gcc or other C or C++ compiler and ANY mpicc
> and mpif77 unless they have bugs or are not installed properly.
> 
>   The point is you can not mix incompatible compilers or MPIs when you
> install two or more different packages to work together. This has
> nothing to do with PETSc. If you build Diffpack and Superlu_dist with
> different MPI's it would not work.
> 
>    Barry
> 
> 
> On Fri, 12 Oct 2007, Trivedi, Jagruti CIV 470000D, 474300D wrote:
> 
> >  
> > Satish,
> > 
> > That is not the point. The question is how to find out which version 
> > of compilers such as gcc(or c ) , C++ mpicc , mpif77 is needed for 
> > petsc 2.2.1, 2.3.0, 2.3.2 etc?
> > 
> > I do not want to find out after the fact that oh I needed so and so 
> > compiler version.
> > 
> > If possible, can this information be listed as part of Release Notes? 
> > If this information is already available, can you point me to the 
> > right place.
> > 
> > Thanks,
> > 
> > ...Jagruti
> > -----Original Message-----
> > From: owner-petsc-users at mcs.anl.gov
> > [mailto:owner-petsc-users at mcs.anl.gov] On Behalf Of Satish Balay
> > Sent: Friday, October 12, 2007 7:23
> > To: petsc-users at mcs.anl.gov
> > Subject: Re: Version of Mpi and fblaslapack
> > 
> > On Fri, 12 Oct 2007, Mads Hoel wrote:
> > > 
> > > I found out that linking PETSc with another library (Diffpack) 
> > > compiled with different versions of MPI will crash when calling 
> > > InitializePetsc(...). My solution was to drop running 
> > > config/configure
> > 
> > > with --download-mpi option, but have MPI point to the correct 
> > > installation of MPI using the --with-mpi-dir=/path/to/mpi option.
> > 
> > Yes - you cannot mix 2 different versions of MPI [or any other 
> > package] to create a single binary. This is equvalent to using include
> 
> > files from
> > package-version1 - and linking with package-version2. [which is 
> > obviously wrong.]
> > 
> > In the same way, its best to install ALL packages that go into a 
> > binary with the same compiler set [c,c++,fortran].
> > 
> > Usually compilers & MPI are the basic blocks that many external 
> > packages use, so in this case one can first install MPI [with the 
> > correct desired compilers]. And then install all other packages [used 
> > by the application] with the MPI wrappers.
> > 
> > And all external packages built by PETSc will use the compilers/MPI 
> > specified to PETSc.
> > 
> > ./config/configure.py --with-cc=mpicc --with-fc=mpif90 or the shortcut
> 
> > equivalent of the above:
> > ./config/configure.py --with-mpi-dir=
> > 
> > Satish
> > 
> > 
> 
>