since external packages use a variety of different types for 64 bit indices
(though all have the same bit values) casts are needed for some compilers to
prevent warning messages when using 64 bit ind

since external packages use a variety of different types for 64 bit indices
(though all have the same bit values) casts are needed for some compilers to
prevent warning messages when using 64 bit indices

show more ...

Merge branch 'knepley/example-ts-friction'

* knepley/example-ts-friction:
TS ex34: Now reproducing plots from Berkeley paper - Fixed coordinate DM screwups
DMDA: Fixed bug with 1D DMDAGetRay()

Merge branch 'knepley/example-ts-friction'

* knepley/example-ts-friction:
TS ex34: Now reproducing plots from Berkeley paper - Fixed coordinate DM screwups
DMDA: Fixed bug with 1D DMDAGetRay()
TS: Allow the DMDARay viewer to operator in LG mode - Added TSMonitorLGDMDARay() activated by -ts_monitor_lg_dmda_ray - Added TSMonitorLGCtx to TSMonitorDMDARayCtx
DMDA: Allow DMDAGetRay() to work in 1D
TS: Make TSMonitorLGCtx visible to other code
TS: Added an example of DR friction

show more ...

DMDA: Forgot to add new source file

DMDA: Add a matrix preallocation routine that works with PetscSection

DMDA: Add transitive closure
- Only working in 2D

Merge branch 'master' into knepley/feature-dmda-section

* master: (528 commits)
Doc: Fixed configure doc
Nightly summary: Added configure section.
TSSetErrorIfStepFails: fix __FUNCT__
TS: fi

Merge branch 'master' into knepley/feature-dmda-section

* master: (528 commits)
Doc: Fixed configure doc
Nightly summary: Added configure section.
TSSetErrorIfStepFails: fix __FUNCT__
TS: fix for TSGetRHSMats_Private() in the creation of the Arhs and Brhs matrices
protect logging of objects (which uses global variables) with PETSC_USE_LOG
error handler API had been previously changed but this example was not updated to new API.
DMPlex: Fixed bug in coordinate permutation
PetscSection: Doc fix for permutation
Bib: Added Top500 and fixed entry
fun3d: update PetscMallocValidate() usage
bugfix of MatCreateSeqAIJFromTriple() contributed by Nicolas Marsic <nicolas.marsic@montefiore.ulg.ac.be>
bib: rename ref with duplicate key
Webpage: Corrected '-dm_mat_type cusp' to 'dm_mat_type aijcusp'.
completed removing server code that was not removed in previous commit
removed PETSc webserver and configure package files that supported it
Allow calling MatGetBlockSize[s]() before matrix preallocation
SNESLINESEARCHBT: Set the norms when exiting early due to negligible step.
fix outdated PetscOptionsList
SNESQN: only monitor real part of dot product (fails with C++ complex)
Fix for multiple definitions in mat/color/interface/ custom fortran bindings
...

Conflicts:
config/builder.py
include/petscdmplex.h
src/dm/impls/da/dageometry.c
src/dm/impls/plex/examples/tests/ex4.c
src/dm/impls/plex/plex.c
src/dm/impls/plex/plexinterpolate.c

show more ...

Some minor fixes in comments, changed some float divisions to integer divisions and added error checking.

First attempt at sanitising chunking.

The rationale is to keep the old behaviour if possible, so we just add a if-
else after the chunk sizes are first calculated and change those sizes that
need to

First attempt at sanitising chunking.

The rationale is to keep the old behaviour if possible, so we just add a if-
else after the chunk sizes are first calculated and change those sizes that
need to be changed. [There is also a zeroing of the 6-element arrays in the
beginning of the function because we multiply them together to find the various
sizes and at least Cray seems to have random numbers in the stack when
entering the function, causing the sizes to be crazy.]

We first check if the average vector size exceeds 4 GiB and if it does, simply
use as few chunks as possible, splitting along slowest varying axis first and
if that is not enough, splitting the next axis, too etc. The patch *can* split
the fastest varying dimension, too, but that seems overkill: it would mean
that da_grid_x*dof > 4 GiB, which seems unlikely. It cannot split dof.

If the average Vec size is <= 4 GiB, but total Vec size is > 4 GiB, there is a
simple logic: first see if <local_z_size>*<global_y_size>*<global_x_size>*dof
is smaller than "target size". Target size comes from your formula earlier in
this thread. If that is so, use that, if not, then just use the local Vec size
as the chunk size. This could be improved by having an intermediate size
solution, too, but this solution gives *my* tests decent performance across
several global sizes, so it does not seem to be too bad. Now, the performance
we get is just ~3 GiB/s, using 36 OSTs on a lustre. Using one OST we get ~200
MiB/s, so the scaling is not that bad: we get half of 36*200 MiB/s, so we think
the performance is limited by the OSTs, not MPI-IO or HDF5 or chunking. Tests
with bigger systems would of course be needed to determine whether this
solution works for truly huge files, too: the biggest we tried are just over 50
GiB.

show more ...

Merge branch 'master' into barry/saws

Conflicts:
src/ksp/pc/impls/gamg/gamg.c
src/sys/classes/viewer/impls/ams/ams.c
src/sys/objects/pinit.c

DMDA: Fixed bug with 1D DMDAGetRay()

DMDA: Allow DMDAGetRay() to work in 1D

DMDA: Added DMDAGetCellPoint()

DMPlex: Added documentation

Merge branch 'barry/reduce-dmsetup-da-memoryusage' of bitbucket.org:petsc/petsc into barry/reduce-dmsetup-da-memoryusage

update hypre interface for const PetscInt ** argument to DMDAGetGlobalIndices()

Hypre: Fix cast for DMDAGetGlobalIndices()

removed use of DMDAGetGlobalIndices() for the DMDA computation of Jacobians
Replaced with call to DMGetLocalToGlobalMqppingBlock() and ISGetIndices() so these routines
no longer need the full set of

removed use of DMDAGetGlobalIndices() for the DMDA computation of Jacobians
Replaced with call to DMGetLocalToGlobalMqppingBlock() and ISGetIndices() so these routines
no longer need the full set of indices only the block version.

This in itself does not save memory but eventually will allow use to remove the storage of all the indices for dof > 1 with only the
storage of indices for the blocks

show more ...

removed idx and Nl from DM_DA object since the dm->ltogmap has the same information

this eliminates one memory usage of size dof * number of local grid points in DMSetUp_DA
Needed to add DMDARestore

removed idx and Nl from DM_DA object since the dm->ltogmap has the same information

this eliminates one memory usage of size dof * number of local grid points in DMSetUp_DA
Needed to add DMDARestoreGlobalIndices() to properly allow access to dm->logmap with DMDAGetGlobalIndices().
Though users could use DMGetGlobalToLocalMapping() followed by ISLocalGetGlobalMappingGetIndices() I have retained
DMDAGetGlobalIndices() to the code refactorization is less dramatic; the code could be changed later to eliminate
DMDAGetGlobalIndices()

show more ...

Merge branch 'jed/doc-restore-zeros-pointer'

* jed/doc-restore-zeros-pointer:
DMPlex: fix my mistake using NULL in place of (int) 0
doc: restore functions in PETSc zero their arguments

Conflict

Merge branch 'jed/doc-restore-zeros-pointer'

* jed/doc-restore-zeros-pointer:
DMPlex: fix my mistake using NULL in place of (int) 0
doc: restore functions in PETSc zero their arguments

Conflicts:
src/dm/impls/plex/plex.c

show more ...

DMDA: fix missing declarations and includes

DMDA: Added DMDAProjectFunction() and DMDAComputeL2Diff()

DMDA: Fixed 2D geometry
- Should reuse DMPlex routines

DMDA: Fix bug in 2D closure operation for cells

DMDA: Make closure operations also return the size

DMDA: Changed DMDACreateSection() to emulate DMPlexCreateSection()