[mpich-discuss] Running MPI jobs without mpiexec

[Amar K. Patel]

Hi Jayesh,

Thanks for suggesting the two options earlier. I have been exploring them 
since your recommendations and I would appreciate if you could answer 
these follow-up questions. 

1. I have been looking at MPI_Comm_spawn_multiple and MPI_Comm_spawn under 
MPI dynamic processes and I have concluded that for either of these 
options to work, the callee binary should be an exe. Am I correct to 
conclude this? 

Meaning that if I have the following line in my main code (call it 
myCode2.exe), 

MPI_Comm_spawn_multiple(NHOST, progs, MPI_ARGVS_NULL, nproc, infos, 0, 
MPI_COMM_WORLD, &intercomm, MPI_ERRCODES_IGNORE);

the binary specified under "Progs" argument should be an exe and not a dll 
or a lib file. Is this correct ? 

Here is a little background on what I am trying to do here.
I am trying to call a shared library (dll) that needs to be run on 
multiple processors but I do not want the main calling program to be run 
on multiple PCs. So basically, I need to run the main program 
(myCode2.exe) as a singleton process which at run time needs to call a dll 
(myCode1.dll) and that dll needs to be run on multiple processors. 
I have tried putting in MPI_Comm_spawn_multiple in the main program and 
tried to call the dll through MPI_Comm_spawn_multiple but I get a runtime 
error indicating that the program being called by MPI_Comm_spawn_multiple 
is not an executable. 

2. Your other suggestion was launching mpiexec from the main program 
(myCode2.exe). Are you suggesting using the system() function in 
myCode2.exe to launch mpiexec which then launches myCode1?  Wont this 
require myCode1 to be an exe file to be launched with mpiexec command? In 
my case myCode1 used to be compiled as a standalone exe file before, but 
we would now wish to compile it as a dll that is loaded as a shared 
library and called by another larger executable at run time.

Thanks for your help.

Amar 





Jayesh Krishna <jayesh at mcs.anl.gov> 
Sent by: mpich-discuss-bounces at mcs.anl.gov
08/04/2011 03:28 PM
Please respond to
mpich-discuss at mcs.anl.gov


To
mpich-discuss at mcs.anl.gov
cc
PATEL_AMAR_K at cat.com
Subject
Re: [mpich-discuss] Running MPI jobs without mpiexec




Caterpillar: Confidential Green Retain Until: 09/03/2011 



Hi,
 Two options come to my mind,

# Take a look at MPI dynamic processes (
http://www.mpi-forum.org/docs/mpi21-report/node197.htm#Node197).
# Launch mpiexec from myCode2.

Regards,
Jayesh
----- Original Message -----
From: "Amar K. Patel" <PATEL_AMAR_K at cat.com>
To: mpich-discuss at mcs.anl.gov
Sent: Thursday, August 4, 2011 11:35:58 AM
Subject: [mpich-discuss] Running MPI jobs without mpiexec



Hello, 

I have MPI program that runs fine when it started with the command mpiexec 
-n # myCode1.exe. I now need to call myCode1 from another code (say 
myCode2) in such a way that myCode1 is a dll file which is loaded and 
called when myCode2 is being executed. myCode2 is started without using 
mpiexec commad. 

With this new setup I would like to call myCode1.dll and be able to use 
the MPI features that have been embedded in it such that myCode1.dll can 
be run on more than one CPU. Is such a thing possible ? If so, then I 
would appreciate if someone could point me to a documentation or an 
example that illustrates how to implement the changes needed in myCode1 
and/or myCode2. 

Here is an crude version of myCode1 and myCode2 that I am using. 

void myCode1() 
{ 
int rank; 
MPI_Init(x, x); 
MPI_Comm_rank(MPI_COMM_WORLD, &rank); 
...... 
...... 
...... 
...... 
...... 
...... 
...... 
...... 
MPI_Finalize(); 
return; 
} 


int myCode2(int argc, char **argv) 
{ 
myCode1(); 
return 0; 
} 
_______________________________________________
mpich-discuss mailing list
mpich-discuss at mcs.anl.gov
https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
_______________________________________________
mpich-discuss mailing list
mpich-discuss at mcs.anl.gov
https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20110815/48d713ab/attachment.htm>