[Darshan-users] Darshan and AIX?
Phil Carns
carns at mcs.anl.gov
Thu Feb 10 16:13:44 CST 2011
Hi Julian,
We haven't made a new release yet, but I wanted to let you know that
these changes are in place in the trunk version of Darshan now:
- (experimental) LD_PRELOAD support, see libdarshan.so
- removed sizeof test in autoconf, no need for cross compile
- replace MPIO_Request with MPI_Request
- removed c++ comments
Those things still don't help any if you can't get either the -wrap
linker argument or ld preloading to work on AIX, but maybe it will make
Darshan a little easier to work with if you experiment with it again in
the future :)
-Phil
On 12/22/2010 09:17 AM, Phil Carns wrote:
> Thanks for the bug reports. Tat stuff all sounds fixable.
>
> On AIX, does mpicc do static linking or dynamic linking by default?
>
> If it is the latter, then you might be able to use LD_PRELOAD for your
> wrappers. (google tells me maybe this LDR_PRELOAD on AIX? I haven't
> a clue). This isn't ready yet, but we are working on an LD_PRELOAD
> compatible library in svn trunk right now. It should be available at
> least in experimental form in the next release sometime after the
> holidays.
>
> -Phil
>
> On 12/22/2010 04:20 AM, Julian Kunkel wrote:
>> Thanks guys for your help,
>> I modified configure by setting cross_compiling=yes
>> host and build did not work (maybe I have not experience with it, or
>> it detects that it should compile for the same host and switches to
>> no). now I was able to run configure properly.
>>
>> To get around a problem with mpicc (under IBM PE it is called mpcc) I
>> created a wrapper bash script which just runs mpcc "$@"
>>
>> On IBMs PE it seems that MPIO_Request should be MPI_Request, otherwise
>> the compiler argues.
>> int MPI_File_iread_at(MPI_File fh, MPI_Offset offset, void *buf,
>> int count, MPI_Datatype datatype, MPI_Request *request);
>>
>> I wonder if this issue will appear with other MPI libraries as well,
>> as it seems the MPI Standard uses here MPI_Request as Datatype.
>>
>> There are two C++ style (//) comments in lib/darshan-mpi-io.c
>>
>> At this point I have issues on compiling the parser with getopt, I
>> might continue to investigate this issues next year.
>> As we get a small Power7 Testsystem which runs under RedHat it might
>> be a better choice to test it there ;-)
>>
>> Regards,
>> Julian
>>
>> P.S. Have a merry chrismas.
>>
>> 2010/12/22 Rob Latham<robl at mcs.anl.gov>:
>>> On Tue, Dec 21, 2010 at 11:51:45AM +0100, Julian Kunkel wrote:
>>>> Hi,
>>>> has anyone suceeded or experience to compile Darshan on an AIX (6.1)
>>>> system so far?
>>>>
>>>> One problem seems to be that mpicc is used to compile the
>>>> particular test for:
>>>> checking size of long int... configure: error: cannot compute
>>>> sizeof (long int)
>>>> However, running an MPI executable is not possible (easily) on the
>>>> login node.
>>> one way around that is to put configure into cross compile mode with
>>> the --build and --host flags. kind of icky but will at least use
>>> compile-time only tests. we might have other configure tests that
>>> execute code, and cross compile mode will find those as well.
>>>
>>>> I'm not so sure about the wrap (linker) flag used to intercept calls,
>>>> probably another method is required in AIX, so I would be glad about
>>>> experience from other AIX users.
>>> I think you're it, Julian.
>>>
>>> ==rob
>>>
>>> --
>>> Rob Latham
>>> Mathematics and Computer Science Division
>>> Argonne National Lab, IL USA
>>>
>> _______________________________________________
>> Darshan-users mailing list
>> Darshan-users at lists.mcs.anl.gov
>> https://lists.mcs.anl.gov/mailman/listinfo/darshan-users
>
> _______________________________________________
> Darshan-users mailing list
> Darshan-users at lists.mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/darshan-users
More information about the Darshan-users
mailing list