[MPICH] MPICH2+MPE on windows

David Ashton ashton at mcs.anl.gov
Tue Jul 12 13:09:51 CDT 2005


James,

MPI_Pcontrol is not correctly implemented in the wrappers for Windows yet.
If you want to use it you will have to link with mpich2mpe.lib, mpe.lib and
the channel library of your choice like mpich2.lib.

We'll fix MPI_Pcontrol in mpi.lib for the next release.

-David Ashton

-----Original Message-----
From: owner-mpich-discuss at mcs.anl.gov
[mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of James S Perrin
Sent: Tuesday, July 12, 2005 8:07 AM
To: mpich-discuss at mcs.anl.gov
Subject: Re: [MPICH] MPICH2+MPE on windows

Hi,
	One other thing I forgot to mention. Using this method to add
logging (which I 
like a lot, thanks!) is that MPI_Pcontrol() seems to be rendered
ineffective. I 
call MPI_Pcontrol(0) immediately after MPI_Init() but still get a full log.

James

James S Perrin wrote:

> Hi David,
> 
>     I found that "-env MPI_WRAP_DLL_NAME mpich2mpe" or rather better 
> "-genv MPI_WRAP_DLL_NAME mpich2mpe" worked for me. Maybe windows also 
> looks fo "%s.dll"
> 
>     I think you've got a little mixed up as MPICH2_CHANNEL seem to be 
> for setting things like "ib" (inifiband) :) MPI_WRAP_DLL_NAME can be set 
> "mpe" can be set to as the code also looks for dll's of the pattern 
> "mpich2%s.dll"
> 
> Regards
> James
> 
> David Ashton wrote:
> 
>> James,
>>
>> Thanks for pointing out the bug that -log is not global.  If you specify
>> then environment variable, MPI_WRAP_DLL_NAME, then you must specify the
>> entire name - mpich2mpe.dll.  If you use the MPICH2_CHANNEL environment
>> variable then you can use the short name, mpe.
>>
>> -David Ashton
>>
>> -----Original Message-----
>> From: owner-mpich-discuss at mcs.anl.gov
>> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of James S Perrin
>> Sent: Monday, July 11, 2005 8:46 AM
>> To: mpich-discuss at mcs.anl.gov
>> Subject: [MPICH] MPICH2+MPE on windows
>>
>> Hi,
>>     This is a follow up to an earlier thread (Status of MPE2 on Windows)
>> where it was stated that to run an mpi job with profiling you can use the
>> (undocumented) -log  to mpiexec (smpd only). This didn't work for my 
>> master/slave job until
>> I realised -log isn't a global option and has to be specified for each 
>> group
>> of processes.
>>
>>     However I noticed while digging around in the source code -log sets
>> the wrapper dll to "mpe" whereas the suggested default is "mpich2mpe" 
>> (src/util/multichannel/mpi.c). I was wondering is there any difference, I
>> didn't notice any when passing "-env MPI_WRAP_DLL_NAME mpich2mped" to 
>> mpiexec?
>>
>> Regards
>> James
> 
> 
> 


-- 
----------------------------------------------------------------------------
-
James S. Perrin,                   | email: james.perrin at manchester.ac.uk
Manchester Visualization Centre,   | www.sve.man.ac.uk/General/Staff/perrin
Kilburn Building, The University,  | tel: +44 161 275 6945
Manchester, England. M13 9PL.      | fax: +44 161 275 6800/6040
----------------------------------------------------------------------------
-
"The test of intellect is the refusal to belabour the obvious" -Alfred
Bester
----------------------------------------------------------------------------
-





More information about the mpich-discuss mailing list