Simplify packing routines with macros and optimize packing by using index patterns

Implement PetscSFBcast with PetscSFBcastAndOp

Merge remote-tracking branch 'origin/barry/add-petscarrayxxx-ops' [PR #1771]

Add PetscArraycmp(), PetscArraycpy(), PetscArrayzero(), PetscArraymove(), update code to use new forms

Replace a couple multi-mallocs with a single malloc
Replace backwards loops in MatSetValues_*()

Add PetscArraycmp(), PetscArraycpy(), PetscArrayzero(), PetscArraymove(), update code to use new forms

Replace a couple multi-mallocs with a single malloc
Replace backwards loops in MatSetValues_*() with PetscArraymove()

Commit-type: style-fix, feature

show more ...

Merged in stefano_zampini/dmplex-speedup (pull request #1665)

Last speedups to the PLEX code with large number of MPI processes

Approved-by: Matthew Knepley <knepley@gmail.com>

PetscSFSetUp_Basic: fix coding standards

Merged in jed/feature-sf-char (pull request #1511)

Add signed char, unsigned char, and char support to SF

Approved-by: Jed Brown <jed@59a2.org>
Approved-by: BarryFSmith <bsmith@mcs.anl.gov>

Make SF eable to handle dumb types of any size

Add support of signed/unsigned char in SF

SF: add support for char types

Merged in dalcinl/plex-distribute (pull request #1454)

DMPLEX: Speedup DMPlexDistribute[Overlap]()

Approved-by: Matthew Knepley <knepley@gmail.com>

Merged in jczhang/vecscat-add-opts (pull request #1462)

Add two small optimizations in VecScatter and SF before release

Approved-by: BarryFSmith <bsmith@mcs.anl.gov>
Approved-by: Jed Brown <jed@59a

Merged in jczhang/vecscat-add-opts (pull request #1462)

Add two small optimizations in VecScatter and SF before release

Approved-by: BarryFSmith <bsmith@mcs.anl.gov>
Approved-by: Jed Brown <jed@59a2.org>

show more ...

Use an enum instead of a boolean to indicate operation in SF

Use persistent send/recv in SF

PETSCSFBASIC: Follow usual MPICH send/recv order in SetUp.

Follow MPICH pattern in the implementation of neighbor collectives

* Post all sends first
* Post all recvs second
* Handle messages to sel

PETSCSFBASIC: Follow usual MPICH send/recv order in SetUp.

Follow MPICH pattern in the implementation of neighbor collectives

* Post all sends first
* Post all recvs second
* Handle messages to self with memcpy
* Complete a single wait call

show more ...

Merge branch 'master' of bitbucket.org:petsc/petsc

Merge remote-tracking branch 'origin/jczhang/vecscat-in-sf' [PR #1396]

Remove {} around SETERRQ

Always use const data as key to look up links in SF

With this modification, we can have a VecScatter implementation that in VecScatterBegin(..,x,y,..),
one only accesses the read-only vector x and d

Always use const data as key to look up links in SF

With this modification, we can have a VecScatter implementation that in VecScatterBegin(..,x,y,..),
one only accesses the read-only vector x and does not access y - the vector to write. Access to y is
pushed to VecScatterEnd(..,x,y,..).

The benefit is that we can get rid of VecGet/RestoreArray(y,..) in VecScatterBegin().
For vectors that are on GPU or not petsc native (e.g., VecNest), getting/restoring arrays are quite expensive.

In other words, we do lazy data access -- we only access data when we need it. This discussion also applies to PetscSFXxxBegin/End.

show more ...

Handle MPI builtin types specially in SF to make the common case fast

We do not need complicated dup and compare logic for builtin types.
Note that VecScatter only has one type, MPIU_SCALAR.

Make incoming ranks in PetscSFSetUp_Basic() sorted

This is cheap and is needed by MatGetBrowsOfAoCols_MPIAIJ()

Add PetscSFGetLeafRanks()

Existing PetscSFGetRanks() only returns ranks this process will send to.
This new one gives ranks the process will receive from.

Add a basic implementation of PetscSFBcastAndOp

Merge branch 'master' into jpusztay/feature-swarm-symplectic-example

Merged petsc/petsc into master