Re: two-face collective scenario

Thomas Boenisch (Thomas.Boenisch@RUS.Uni-Stuttgart.DE)
Fri, 31 Jan 1997 17:29:44 +0100

Date: Fri, 31 Jan 1997 10:07:02 -0600 (CST)
From: Tony Skjellum <tony@aurora.cs.msstate.edu>
Cc: mpi-external@mcs.anl.gov, mpi-coll@mcs.anl.gov,
Thomas Boenisch <Thomas.Boenisch@RUS.Uni-Stuttgart.DE>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII

Tony Skjellum wrote:
> What is the motivation of using "info" instead of a request. Elsewhere in
> MPI, request serves the purpose of an abstract base class. Why invent a
> new object here, which behaves just like a request? Just because its only
> valid use is with the other "book end", it is still of the same flavor as
> a request.
> -Tony

info is wrong. This should be ierror, because it's a fortran code.

the correct example is:

call MPI_REDUCE_START(teilresl2, resl2, 1, REALx, MPI_SUM, 0,
. MPI_COMM_WORLD, ierror)
call MPI_ALLREDUCE_START(teilresl1, resl1, 1, REALx, MPI_SUM,
. mpi_comm_world_a, ierror)
call MPI_ALLREDUCE_START(teilresl0, resl0, 1, REALx, MPI_MAX,
. mpi_comm_world_b, ierror)

c...solv the equationsystem
call parallelsolver( rm000, rmf00, rmb00, rm0f0, rm0b0, rm00f,
. rm00b, dq, rhs, nsubit, nhofhausit )

c...computation of the CFL-Number with resl1
call MPI_ALLREDUCE_END(mpi_comm_world_a, resl1, ierror)
...

c...output of all residuals
call MPI_REDUCE_END(MPI_COMM_WORLD, resl2,ierror)
call MPI_ALLREDUCE_END(mpi_comm_world_b, resl0, ierror)

Thomas

Computing Center University of Stuttgart Phone: ++49(0)711/6852509
HLRS Fax : ++49(0)711/6788363
Parallel Computing E-Mail: thomas.boenisch@rus.uni-stuttgart.de