On Wednesday 03 February 2010 10:24:59 pm Cunningham, David wrote:
> Hi, I had a quick look at the differences between KMeansCUDA's own
> DistributedRail and the system one, they are very similar except for some
> type system stuff which may or may not still be important.
>
> However the DistributedRail internally synchronises using clocks so you
> must be using clocks in your program. You should treat the collective
> operation like a next 'next' statement, it actually uses several next
> statements internally.
>
> Thanks
>
David,
thanks for your hint to use clocks. Apparently, I am still using the
DistributedRail class the wrong way. Following your hint to use the collective
operation like a "next" statement, I was calling it "collectively" from every
place. But this gives me an error message as printed below. Could you please
have a look ?
I actually stripped down the program to a small test example that I attached
to this mail. Running this program with C++ back end on two MPI tasks, I get
the following error message.
<map>
<host name="sirius" slots="1" max_slots="0">
<process rank="0"/>
<process rank="1"/>
</host>
</map>
<stdout rank="0">main: before ClockTest</stdout>
<stdout rank="0">Before next</stdout>
<stdout rank="0">After next</stdout>
<stdout rank="0">v_tmp(0)=0</stdout>
<stdout rank="1">Before next</stdout>
<stdout rank="1">After next</stdout>
<stdout rank="1">v_tmp(0)=1</stdout>
<stderr rank="0">0: xlupc transport: AMSend: handler returned NULL</stderr>
--------------------------------------------------------------------------
mpirun noticed that process rank 0 with PID 3572 on node sirius exited on
signal 6 (Aborted).
--------------------------------------------------------------------------
Does this ring a bell with you ?
This error message disappears, when the collective operation is commented out.
Also, I don't understand the error message. I am running on x86 (32 bit) with
g++ and OpenMPI. How comes that I get a message from "xlupc", which looks like
an xl-compiler (for UPC ?) ? I thought that xl-compilers only exist for POWER
CPUs.
--
Mit freundlichen Grüßen / Kind regards
Dr. Christoph Pospiech
High Performance & Parallel Computing
Phone: +49-351 86269826
Mobile: +49-171-765 5871
E-Mail: christoph.pospi...@de.ibm.com
-------------------------------------
IBM Deutschland GmbH
Vorsitzender des Aufsichtsrats: Erich Clementi
Geschäftsführung: Martin Jetter (Vorsitzender),
Reinhard Reschke, Christoph Grandpierre,
Klaus Lintelmann, Michael Diemer, Martina Koederitz
Sitz der Gesellschaft: Ehningen / Registergericht: Amtsgericht Stuttgart, HRB
14562 WEEE-Reg.-Nr. DE 99369940
import x10.util.DistributedRail;
public class Hello {
def ClockTest() {
val c = Clock.make();
val v_tmp:DistributedRail[Int]
= new DistributedRail[Int](4,
(i:Int) => {(0) as Int});
finish {
for (p in Place.places ) {
async (p) clocked (c) {
for (var k:Int=0; k<4; k++) {
v_tmp(k) = p.id();
}
Console.OUT.println("Before next");
next;
Console.OUT.println("After next");
Console.OUT.println("v_tmp(0)="+v_tmp(0));
v_tmp.collectiveReduce(Int.+);
}
}
c.drop();
}
Console.OUT.println("After finish");
Console.OUT.println("v_tmp(0)="+v_tmp(0));
}
public static def main(var args: Rail[String]!) {
Console.OUT.println("main: before ClockTest");
var h:Hello! = new Hello();
h.ClockTest();
Console.OUT.println("main: after ClockTest");
}
}------------------------------------------------------------------------------
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
http://p.sf.net/sfu/theplanet-com
_______________________________________________
X10-users mailing list
X10-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/x10-users
