[MPICH] MPICH2 Questions: Read input files from current directory and output control

[Yusong Wang]

I recompiled my program with the mpicc under the MPICH2 directory (The
previous version was compiled with mpicc of MPICH1) and
all the problems were gone. The lesson I learned is that the mpicc and
mpiexec have to be from same version.  
 
For simple program, the program could run, but may have a problem by
combining the mpicc of MPICH1 and mpiexec of MPICH2, even though they
are built on the same architecture and same gcc compilers.

Thanks for your help!

Yusong  


On Thu, 2006-04-20 at 15:32 -0500, Ralph Butler wrote:
> I have not been able to produce a problem like yours.  Let me explain  
> what I have tried and why I think
> it mimics what you are trying to do.  You can tell me where I am  
> going wrong.  I ran this:
>      mpiexec -n 2 /bin/cat tempin
> where tempin is a tiny test file in the current dir.  I also tested  
> like this:
>      mpiexec -n 2 -wdir `pwd` /bin/cat `pwd`/tempin
> In this latter example, the `pwd` stuff should not be necessary since  
> the default wdir is current dir.
> It seems to me that both do the same thing as your program.  Your  
> "cat" program (Plegant) actually reads
> a separate file which tells it to read the equivalent of tempin.   
> But, then I would assume that it just opens
> the second file and reads it like cat does.  There is one obvious  
> possible problem, i.e. the file being
> read (FFTB.lte) is actually not on each box doing the reading.
> 
> It might be interesting to see if the examples that I tried work for  
> you.  If so, then there may be some
> unique manner in which your application is opening the file for reading.
> 
> --ralph
> 
> On Apr 20, 2006, at 1:34 PM, Yusong Wang wrote:
> 
> > Ralph,
> >
> > Thanks for your response. I hope I can clear the confusions:
> >
> > On Thu, 2006-04-20 at 13:09 -0500, Ralph Butler wrote:
> >> Hi Yusong:
> >>
> >> I cannot figure out what you are attempting to do because I am not
> >> sure what all the parts actually are.
> >> For example, Pelegant seems to be an executable (pgm or script) that
> >> you are attempting to run giving
> >> it run_p.ele as an argument.
> > Yes, you are right. Pelegant is the executable file.
> >>   But, I am not sure of that.  If I am
> >> right, then it is not clear what the role of
> >> run_p.ele is.
> > run_p.ele is a text input file (used for parsing several commands for
> > Pelegant), which specifies which lattice file (used for describe the
> > layout of an accelerator) will be used.
> >>  You say that FFTB.lte "is given" in run_p.ele.  But, I
> >> don't know if you mean that it is a pre-existing
> >> file that is merely mentioned by name there, or if it is written by
> >> the executing program, or created via a
> >> here-document in a script, etc.
> > Both run_p.ele and FFTB.lte are pre-existed in the current directory.
> >>   Can you explain how to reproduce
> >> the problem using simple cmds such as
> >> cat or printenv??  If not, please explain what the individual files
> >> are and how they interact.
> > Pelegant parses run_p.ele, which tells Pelegant to use FFTB.lte as the
> > lattice file.
> >>
> >> Thanks.
> >> --ralph
> >>
> >> On Apr 20, 2006, at 11:23 AM, Yusong Wang wrote:
> >>
> >>> Hi,
> >>>
> >>> I am trying to run my program with MPICH2 under MPD environment.
> >>> The cpi
> >>> example went well. I have two questions for my own examples:
> >>>
> >>> 1) There is another file (FFTB.lte, in the current directory),
> >>> which is
> >>> given in run_p.ele. How can mpiexec find it without giving the path
> >>> explicitly in the run_p.ele? (I am dealing with more than a  
> >>> hundred of
> >>> these kind of examples, it will be convenient if mpiexec can find  
> >>> the
> >>> file itself.) It seems to me that the -wdir can only find the binary
> >>> file instead of other text input files.
> >>>
> >>> [ywang25 at weed23 FFTB1]$ mpiexec -machinefile ./myMachines -np 4 - 
> >>> wdir
> >>> `pwd` $ebin/Pelegant `pwd`/run_p.ele
> >>>
> >>> Unable to find file FFTB.lte
> >>>
> >>> 2) The output from all the slave processes are printed out,  
> >>> although I
> >>> have redirected them to "/dev/null".
> >>>
> >>> I don't have any problem with the mpirun of MPICH1. Those are
> >>> something
> >>> I need to solve before I can run my program with MPICH2.
> >>>
> >>> Thanks for your help.
> >>>
> >>> Yusong
> >>>
> >>>
> >>>
> >>
> >
> >
>