Re: 2nd beta preview looking good!

Date: Mon Aug 15 2005 - 18:51:44 PDT

  • Next message: Marc L. Smith: "Re: 2nd beta preview looking good!"
    On Mon, Aug 15, 2005 at 05:45:56PM -0400, Marc L. Smith wrote:
    > When I set SSH_SERVERS as you recommend, I get a runtime error:
    > ~~~
    > [email protected]:~/upc> upcc -T 8 helloworld2.upc
    > [email protected]:~/upc> upcrun helloworld2
    > Not enough machines in environment variable SSH_SERVERS to satisfy  
    > request for (8).
    > Only (4) machines available: n1 n2 n3 n4
    > *** AMUDP ERROR: Error spawning SPMD worker threads. Exiting...
    > [email protected]:~/upc>
    This is because you're not using pthreads, so an executable statically
    compiled for 8 UPC threads will use 8 processes, and thus needs 8 entries
    in your SSH_SERVERS/UPC_NODES.  If you add -pthreads, you should be OK:
         upcc -T 8 -pthreads helloworld2.upc
         upcrun helloworld2
    > default_cpus_per_node = 2
    > (currently this is blank)
    This ought to work in theory for you example above, but I've just
    discovered that it doesn't.  I've added a bug (#1213) to our bugzilla
    server about it.  It probably won't be fixed by the release, though.
    Generally, you'll want to use -pthreads to get the best performance on
    an SMP, and in that case this setting actually does work, so yes, you
    probably want to set it (assuming all your nodes are 2-way SMPs).
    > udp_spawn_nodes = n1 n2 n3 n4
    > (currently, this is blank)
    This is not actually for providing a list of node names--it's the
    command line that should be used when -nodes or -cpus-per-host is passed
    to upcrun.  I believe that this should currently be "%P %N %A", instead
    of blank, but the only effect of having it be blank (for programs
    compiled with -pthreads) is to cause some harmless warnings to be
    Thanks for rooting out these bugs!  
    Jason Duell             Future Technologies Group
    <jcduell_at_lbl_dot_gov>       Computational Research Division
    Tel: +1-510-495-2354    Lawrence Berkeley National Laboratory

  • Next message: Marc L. Smith: "Re: 2nd beta preview looking good!"