Re: [mpich-discuss] beginner for remote mem access (sufeng)

5 Jul 2013

      Hi, Jeff

Thanks a lot for your reply. Sorry about subject issue.

1.
MPI_Win_fence is not similar to MPI_Barrier in some cases.  The
MPICH implementation can turn some calls into no-ops and others into
MPI_Reduce_scatter.
I am not quite sure about "turn some calls into no-ops and others into
MPI_Reduce_scatter" Could you please give an example if possible? Another
thing is if a process create a window for other process, but the data
access is available after some operations. Should I use MPI_Win_fence or
MPI_Barrier to sync? or other methods?

2.
Regarding "use MPI window as thread level", I really don't
understand your thinking at all.  MPI RMA is not shared memory nor is
MPI a threading model.
Sorry for uncleared statements, right now I would like to do multithreads
and MPI hybrid programming. let me give an example: I have 3 processes,
each one has 8 threads. thread 0 in process 0 creates a RMA window. if I
would like all other threads to access it, should I use thread 0 in process
1, thread 0 in process 2 to MPI_Get the data from window, then use shared
memory for internal threads to load the data? i am not sure what is the
proper way for RMA in hybrid model.

Thank you so much!

Sufeng

On Fri, Jul 5, 2013 at 6:15 AM, <[email protected]> wrote:
...
Send discuss mailing list submissions to
        [email protected]
To subscribe or unsubscribe via the World Wide Web, visit
        https://lists.mpich.org/mailman/listinfo/discuss
or, via email, send a message with subject or body 'help' to
        [email protected]
You can reach the person managing the list at
        [email protected]
When replying, please edit your Subject line so it is more specific
than "Re: Contents of discuss digest..."
Today's Topics:
1. Re:  beginner for remote mem access (Jeff Hammond)
   2.  Running configure on Suse Enterprise 10
      ([email protected])
----------------------------------------------------------------------
Message: 1
Date: Wed, 3 Jul 2013 14:50:44 -0500
From: Jeff Hammond <[email protected]>
To: [email protected]
Subject: Re: [mpich-discuss] beginner for remote mem access
Message-ID:
        <CAGKz=uJN2iH3Pxjxyq=YzrP=
[email protected]>
Content-Type: text/plain; charset=ISO-8859-1
1) When you reply to the digest emails, change the subject to that of
the thread you care to respond to.  That will make it much easier for
others to track the thread.
2) MPI_Win_fence is not similar to MPI_Barrier in some cases.  The
MPICH implementation can turn some calls into no-ops and others into
MPI_Reduce_scatter.
3) Regarding "use MPI window as thread level", I really don't
understand your thinking at all.  MPI RMA is not shared memory nor is
MPI a threading model.
Jeff
On Wed, Jul 3, 2013 at 2:46 PM, Sufeng Niu <[email protected]> wrote:
...
Thanks a lot for your guys reply. I should initial b array, I figure out
the
...
MPI_Win_fence should be called by all processes, which will solve the
assertion failed issue.
it seems that MPI_Win_fence is very similar to MPI_Barrier, Can I say
MPI_Win_fence is the RMA version of MPI_Barrier?
sorry about so many questions. Can I use MPI window as thread level?
let's
say I have 4 processes, each process has 8 threads. if only one thread is
used to create mem access window, other process copy it and internal 8
threads use shared memory to perform some operations. Is that ok?
Thanks a lot!
On Wed, Jul 3, 2013 at 1:28 PM, <[email protected]> wrote:
...
Send discuss mailing list submissions to
        [email protected]
To subscribe or unsubscribe via the World Wide Web, visit
        https://lists.mpich.org/mailman/listinfo/discuss
or, via email, send a message with subject or body 'help' to
        [email protected]
You can reach the person managing the list at
        [email protected]
When replying, please edit your Subject line so it is more specific
than "Re: Contents of discuss digest..."
Today's Topics:
1.  beginner for remote mem access (Sufeng Niu)
   2. Re:  beginner for remote mem access (Yi Gu)
   3. Re:  beginner for remote mem access (Yi Gu)
   4. Re:  beginner for remote mem access (Jeff Hammond)
----------------------------------------------------------------------
Message: 1
Date: Wed, 3 Jul 2013 13:12:46 -0500
From: Sufeng Niu <[email protected]>
To: [email protected]
Subject: [mpich-discuss] beginner for remote mem access
Message-ID:
<CAFNNHkwxqXUB_+b8_tZ=L5K7VcDVhsz4ZMDDMGF55v-FAhgeqg@mail.gmail.com>
Content-Type: text/plain; charset="iso-8859-1"
Hi,
I am a beginner and just try to use remote memory access, and I wrote a
simple program to test it:
#include "mpi.h"
#include <stdio.h>
#define SIZE 8
int main(int argc, char *argv[])
{
        int numtasks, rank, source=0, dest, tag=1, i;
        float a[64] =
        {
                1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0,
12.0, 13.0, 14.0, 15.0, 16.0,
                1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0,
12.0, 13.0, 14.0, 15.0, 16.0,
                1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0,
12.0, 13.0, 14.0, 15.0, 16.0,
                1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0,
12.0, 13.0, 14.0, 15.0, 16.0,
        };
        float b[SIZE];
MPI_Status stat;
MPI_Init(&argc,&argv);
        MPI_Comm_rank(MPI_COMM_WORLD, &rank);
        MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
MPI_Win win;
// check processor rank
        char processor_name[MPI_MAX_PROCESSOR_NAME];
        int name_len;
        MPI_Get_processor_name(processor_name, &name_len);
        printf("-- processor %s, rank %d out of %d processors\n",
processor_name, rank, numtasks);
MPI_Barrier(MPI_COMM_WORLD);
if (numtasks == 4) {
                if (rank == 0) {
                        printf("create window \n");
                        MPI_Win_create(a, 8*sizeof(float),
sizeof(float),
...
MPI_INFO_NULL, MPI_COMM_WORLD, &win);
}
                else {
                        MPI_Win_create(MPI_BOTTOM, 0, sizeof(float),
MPI_INFO_NULL, MPI_COMM_WORLD, &win);
                }
MPI_Win_fence(0, win);
if (rank == 1){
                        MPI_Get(b, SIZE, MPI_FLOAT, 0, 8, SIZE,
MPI_FLOAT,
win);
MPI_Win_fence(0, win);
                }
printf("rank= %d  b= %3.1f %3.1f %3.1f %3.1f %3.1f %3.1f
%3.1f %3.1f\n", rank,b[0],b[1],b[2],b[3],b[4],b[5],b[6],b[7]);
        }
        else
                printf("Must specify %d processors.
Terminating.\n",SIZE);
MPI_Win_free(&win);
        MPI_Finalize();
}
However the terminal gives some odd results:
rank= 0  b= 0.0 0.0 0.0 0.0 0.0 0.0 -71847793475452928.0 0.0
rank= 2  b= 0.0 0.0 0.0 0.0 0.0 0.0 222086852849451401216.0 0.0
rank= 3  b= 0.0 0.0 0.0 0.0 0.0 0.0 -74882.4 0.0
rank= 1  b= 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0
the rank 1 is correct results, but others should be all zero.
terminal also give some comments: "Assertion failed in file
src/mpid/ch3/src/ch3u_rma_sync.c at line 5061: win_ptr->my_counter >= 0
internal ABORT - process 0"
another question is if I use remote memory access. all process which
does
not create window for share must add additional line:
MPI_Win_create(MPI_BOTTOM, 0, data_type, MPI_INFO_NULL, MPI_COMM_WORLD,
&win); correct?
Thanks a lot!
--
Best Regards,
Sufeng Niu
ECASP lab, ECE department, Illinois Institute of Technology
Tel: 312-731-7219

Sufeng Niu

tags

participants (1)