[MPICH2 Req #2556] Re: [MPICH] Parallel I/O problems on 64-bit machine (pleasehelp:-( )
Peter Diamessis
pjd38 at cornell.edu
Fri Jun 2 14:15:43 CDT 2006
Hi Rajeev,
Thanks so much for the pointer. I feel quite embarrassed that I didn't
notice it
and I apologize if I took too much of your time. I had foolishly patterned
the
code after the example you have in the MPICH-2 book (figure 3.33). I guess
this code worked deceptively well on a 32-bit machine to never draw my
attention to my mistake.
So now the code does work perfectly when compiled with either MPICH1
(v1.2.6)
and MPICH2 (v2.1.0.3) but still has problems with MPICH1 -v1.2.7p1.
I don't know if this is related to Michael's comment.
I'm also finding that a separate postprocessing code that I have works fine
with MPICH1 but not MPICH2, uggghhh :-) I'll look into this myself however,
before I bother you again.
Again, I really appreciate the help and I hope I wasn't that much of a
hassle.
Cheers,
Peter
----- Original Message -----
From: "Michael Kluskens" <mkluskens at ieee.org>
To: <mpich-discuss at mcs.anl.gov>
Sent: Friday, June 02, 2006 9:15 AM
Subject: Re: [MPICH2 Req #2556] Re: [MPICH] Parallel I/O problems on 64-bit
machine (pleasehelp:-( )
>
> On May 31, 2006, at 8:40 PM, Rajeev Thakur wrote:
>
>> Peter,
>> The problem is a simple one, but it took me a very long time to
>> figure
>> out. You forgot to add the "ierr" parameter to MPI_File_write_all :-).
>> It's
>> a common mistake people make in Fortran programs, and I should have been
>> more vigilant, but I did all sorts of debugging and stripped your code
>> down
>> to until there was nothing left in it before I noticed the missing
>> parameter
>> :-).
>>
>> Rajeev
>
>
> Shouldn't this be spotted automatically by the compiler or is the
> interface file for MPICH2 incomplete?
>
> I'm testing 3 different MPI libraries and I see a wide range of
> differences between them in this regard.
>
> Michael
>
More information about the mpich-discuss
mailing list