rename '-vec_mpi1' -> 'vecscatter_mpi1'

rm mpi1 from _p_Mat

add runtime option '-vec_mpi1' to activate mpi1 Vecscatter

bugfix for VecScatterEnd(): add MPI_Barrier()

use mpi1 for DMCreateMatrix_Network: isNest; see src/ksp/ksp/examples/tutorials/network/runex1_nest_2

bugfix: MPI_Barrier(PETSC_COMM_WORLD) -> MPI_Barrier(PetscObjectComm((PetscObject)ctx))

add Mvctx_mpi1 for mat-mat ops and MatCreateSubMatrix

rm debugging VecView()

additional rename

add vscat_mpi1.c

bugfix for fortran build

update makefile for newly added files

add vpscat_mpi1.c vpscat_mpi1.c for implementing mpi1 -- copied from master and renamed all functions with _MPI1

fix memeory leak

Need to free local shared win object
Copied value from incorrect location

Funded-by: P-ECP
Time: 0 hours

Updates for VecScatterCopy() with local shared memory copies

Funded-by: P-ECP
Time: 2 hours

Since sharedoffsetwin is only needed during the VecScatterCreate() stage not during the VecScatter phases
remove it from the to/from data structures and only create it locally

Funded-by: P-ECP
Time:

Since sharedoffsetwin is only needed during the VecScatterCreate() stage not during the VecScatter phases
remove it from the to/from data structures and only create it locally

Funded-by: P-ECP
Time: .2 hours

show more ...

additions for VecScatter in the reverse direction to work with shared memory

Funded-by: P-ECP
Time: 2 hours

Improve DMDA test example to increase flexibility of scatter types
Allocated enough space in VecScatter for shared memory buffer

Funded-by: P-ECP
Time: 2 hour

Insure all VecScatter datastructures are freed when using MPI windows for local communication

Funded-by: P-ECP
Time: 1 hour

Yet another memory misallocation size
Clean set of new performance numbers for MPICH and Intel

Funded-by: P-ECP
Time: .5 hours

Debugging, to->sharedspacestarts was allocated before its size (msize) was determined
causing occasional crashes

Funded-by: P-ECP
Time: 1 hour

Debugging and bug-fix. Accessing into the shared memory buffer has to be multipled by bs

Funded-by: P-ECP
Time: .5 hours

Removed generation of send and receive messages for processes that are on the same shared memory
Hence the code now seperates the needed VecScatter communication for that within common shared memory

Removed generation of send and receive messages for processes that are on the same shared memory
Hence the code now seperates the needed VecScatter communication for that within common shared memory
and that which requires the usual MPI message passing.

Funded-by: P-ECP
Time: 1 hour

show more ...

Debugging incorrect use of PetscMalloc()
USE_DYNAMIC not HAVE_DYNAMIC

Funded-by: P-ECP
Time: 1 hour