Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] MPI_Allreduce for INTEGER8?
From: Harald Anlauf (anlauf_at_[hidden])
Date: 2011-01-26 15:23:51


On 01/24/2011 11:28 PM, Harald Anlauf wrote:
> Hi,
>
> MPI_Allreduce works for me with MPI_INTEGER8 for all OpenMPI
> versions up to 1.4.3. However, with OpenMPI 1.5.1 I get a
> failure at runtime:
>
> [proton:23642] *** An error occurred in MPI_Allreduce: the reduction operation MPI_SUM is not defined on the MPI_INTEGER8 datatype
> [proton:23642] *** on communicator MPI_COMM_WORLD
> [proton:23642] *** MPI_ERR_OP: invalid reduce operation
> [proton:23642] *** MPI_ERRORS_ARE_FATAL (your MPI job will now abort)

Since I got no reply yet, I have attached an enhanced test case.

With openmpi-1.5.1 and np=1, also tested with gfortran:

 Real kind, digits: 8 53
 Integer kind, bits: 8 64
 Default Integer : 4 32
 Sum[real(8)]: 1.0000000000000000 2.0000000000000000
3.0000000000000000
 Sum[integer(4)]: 1 2 3
[proton:16920] *** An error occurred in MPI_Allreduce: the reduction
operation MPI_SUM is not defined on the MPI_INTEGER8 datatype
[proton:16920] *** on communicator MPI_COMM_WORLD
[proton:16920] *** MPI_ERR_OP: invalid reduce operation
[proton:16920] *** MPI_ERRORS_ARE_FATAL (your MPI job will now abort)

With openmpi-1.4.3:

 Real kind, digits: 8 53
 Integer kind, bits: 8 64
 Default Integer : 4 32
 Sum[real(8)]: 1.0000000000000000 2.0000000000000000
3.0000000000000000
 Sum[integer(4)]: 1 2 3
 Sum[integer(8)]: 1 2
       3

That's clearly a regression.

Harald