[mpich-discuss] For beginning MPI users...

[kiss attila]

Hi

  I am looking for a debugging/profiling tool for Mpi, but this needs
"Microsoft Compute Cluster Pack SDK". It doesn't even want to install
itself with mpich2. :( Anyway the screenshots look nice. By the way
what other debug/profile tools are available with mpich2 under
windows? Thanks.

regards
K.A. Albert



2009/2/27 Ganesh <ganesh at cs.utah.edu>:
> Hi,
>
> I am new to this group, and notice that many folks are learning how to
> program in MPI. I have a small tip for them.
>
> Once they get their code to compile (*), they may wish to try using ISP
> which was announced in this list just recently. ISP is extremely easy to use
> (its user manual shows the 2-3 commands you need - i.e. you do ispcc instead
> of mpicc , and ispcc calls mpicc from within -- etc.)
>
> Once you run even your simplest MPI programs using ISP, you can watch all
> the possible send/receive matches etc. graphically. In fact, much more than
> a debugger, it shows all possible interleavings, and for each interleaving
> (execution possibility) it shows what send / receive matches can occur.
> ISP's Java GUI has tool tips and you can also watch the source lines using
> the source window. The source window has  tabbed and split views, and you
> can further see the communication matches through color codes.
>
> You may also place C "assert" statements within your code, and ISP will
> check those assertions for all possible executions of your code. Finally you
> may catch deadlocks and MPI object leaks. All this for doing "ispcc" instead
> of "mpicc"
>
> (see user manual at
> http://www.cs.utah.edu/formal_verification/ISP-release )
>
> If you follow the instructions in our user manual and run into problems, we
> will be most happy to help you.
>
> Ganesh Gopalakrishnan
>
> (*) well, you can even provide a "raw" file to ISP . ISP calls MPICH from
> within, and arranges for a lot of the environment settings and flags. So you
> can even see all the "file not found" messages and correct them from within
> ISP.  Once your code runs OK in ISP, then you can shift the code to a larger
> cluster and do your performance tuning experiments. That is, it may be
> easier to get rid of "file not found"s as well as deadlocks from within ISP.
>
> p.s. For those doing MPI code development for their research / classes, we
> are very eager to work with you. We want to see how ISP fares, whether it
> does not work as advertised, or if it surprised you by finding a bug. Please
> drop us a note and we will work with you!
>
> Our email is isp-dev at cs.utah.edu
>
> --
>
>
>
>
>
>
> --
>
>
>
>