David -

No problem.  Generally, you can get away with that trick whenever you're
compiling shared libraries and using the system compiler.  The biggest
advantage is that the chance of the code being mis-compiled goes down
significantly when using the system compiler.  The kernel is a good stress
test that all the atomics got implemented properly :).

Brian

On 3/20/12 1:24 PM, "Gunter, David O" <d...@lanl.gov> wrote:

>You are right, Brian. I failed to build from a fresh untar, so I had some
>leftover intel-compiled bits.
>
>Your suggestion seems to work so I'll pass it along to the user to try
>out.
>
>Thanks!
>
>-david
>--
>David Gunter
>HPC-3: Infrastructure Team
>Los Alamos National Laboratory
>
>
>
>
>On Mar 20, 2012, at 9:52 AM, Barrett, Brian W wrote:
>
>> That doesn't make a whole lot of sense; what compile / link line is
>> resulting in that error message?  The error message is saying that
>> libmpi.so depends on an Intel built-in function, but since you built
>> libmpi.so with gcc, that shouldn't happen.  Are you sure that libmpi.so
>> wasn't build against the Intel compilers?
>> 
>> Brian
>> 
>> On 3/20/12 11:35 AM, "Gunter, David O" <d...@lanl.gov> wrote:
>> 
>>> I wish it were that easy.  When I go that route, I get error messages
>>> like the following when trying to compile the parallel code with Intel:
>>> 
>>> libmpi.so:  undefined reference to `__intel_sse2_strcpy'
>>> 
>>> and other messages for every single Intel-implemented standard
>>>C-function.
>>> 
>>> -david
>>> --
>>> David Gunter
>>> HPC-3: Infrastructure Team
>>> Los Alamos National Laboratory
>>> 
>>> 
>>> 
>>> 
>>> On Mar 20, 2012, at 8:18 AM, Barrett, Brian W wrote:
>>> 
>>>> On 3/20/12 10:06 AM, "Gunter, David O" <d...@lanl.gov> wrote:
>>>> 
>>>>> I need to build ompi-1.4.3 (or the newer 1.4.5) with an older Intel
>>>>> 10.0
>>>>> compiler, but on a newer system in which the default g++ headers are
>>>>> incompatible with Intel. Thus the C and Fortran compilers function
>>>>> normally but the Intel C++ compiler fails to build even a simple
>>>>>"hello
>>>>> world" code.
>>>>> 
>>>>> Is there a way to build OMPI without a C++ compiler?  I tried using
>>>>>the
>>>>> --disable-mpi-cxx and --disable-mpi-cxx-seek flags but these are just
>>>>> for
>>>>> the resulting bindings. The configure step still continues to search
>>>>> for
>>>>> a working C++ compiler.
>>>>> 
>>>>> Yes, I know I can upgrade the Intel compiler but we don't have that
>>>>>as
>>>>> an
>>>>> option in this case.
>>>> 
>>>> Unfortunately, you're a bit out of luck.  Open MPI 1.4.x requires C++
>>>> even
>>>> if you're not building the C++ bindings.  This is not true of 1.5.x
>>>>and
>>>> later.
>>>> 
>>>> If you don't need the C++ bindings, I would build Open MPI with GNU C
>>>> and
>>>> C++ and Intel Fortran.  After building, edit
>>>> <PREFIX>/share/openmpi/mpicc-wrapper-data.txt to change the
>>>>compiler=gcc
>>>> line to compiler=<intel C compiler>.  There's not going to be much
>>>> performance difference between the two compilers for Open MPI itself.
>>>> GNU
>>>> C and Intel C are link compatible, so that should work out for you and
>>>> the
>>>> users will never be the wiser.
>>>> 
>>>> Brian
>>>> 
>>>> -- 
>>>> Brian W. Barrett
>>>> Dept. 1423: Scalable System Software
>>>> Sandia National Laboratories
>>>> 
>>>> 
>>>> 
>>>> 
>>>> 
>>>> 
>>>> _______________________________________________
>>>> users mailing list
>>>> us...@open-mpi.org
>>>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>>> 
>>> 
>>> _______________________________________________
>>> users mailing list
>>> us...@open-mpi.org
>>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>>> 
>>> 
>> 
>> 
>> -- 
>>  Brian W. Barrett
>>  Dept. 1423: Scalable System Software
>>  Sandia National Laboratories
>> 
>> 
>> 
>> 
>> 
>> 
>> _______________________________________________
>> users mailing list
>> us...@open-mpi.org
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>
>
>_______________________________________________
>users mailing list
>us...@open-mpi.org
>http://www.open-mpi.org/mailman/listinfo.cgi/users
>
>


-- 
  Brian W. Barrett
  Dept. 1423: Scalable System Software
  Sandia National Laboratories






Reply via email to