| bb86044d | 10-May-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
SF: do not use MPI_Type_free when MPI_Type_dup() was not available
Type handling is fragile with MPICH1, but this should make normal usage
work. |
| f94c336b | 09-May-2013 |
Matthew G. Knepley <knepley@gmail.com> |
Vec: Fixes for test output, now Builder is clean on regression in src/vec |
| 146574ab | 09-May-2013 |
Barry Smith <bsmith@mcs.anl.gov> |
fixed bug where XXXViewFromOptions() used prefix of object rather than appropriate prefix of KSP or SNES |
| 74fa8af5 | 08-May-2013 |
Satish Balay <balay@mcs.anl.gov> |
Merge commit '4a73821034819dcad1293eaa084b5d44928428de' |
| 4a738210 | 08-May-2013 |
Satish Balay <balay@mcs.anl.gov> |
cusp: fix VecGetArray() as its switching codepath to cusp [when it shouldn't]
This fixes memory leak in src/ksp/ksp/examples/tutorials, runex43_3 etc.
[with cusp/txpetscgpu enabled] |
| 11513a83 | 07-May-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
Merge branch 'jed/stack-call-blas'
Fix performance bug that calling BLAS takes an amount of time dependent
on the number of memory allocations that the program has made.
* jed/stack-call-blas:
Pe
Merge branch 'jed/stack-call-blas'
Fix performance bug that calling BLAS takes an amount of time dependent
on the number of memory allocations that the program has made.
* jed/stack-call-blas:
PetscStackCallBLAS: new macro instead of PetscStackCall (skip CHKMEMQ)
Stack: consolidate handling of PetscStack
show more ...
|
| 16554520 | 07-May-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
Merge branch 'knepley/plex'
Improved support for discretizing PDEs. Cell geometry and FVM. Little to
no topology.
* knepley/plex:
SNES ex12: Fix leak
TS ex11: error if dim != DIM because memory
Merge branch 'knepley/plex'
Improved support for discretizing PDEs. Cell geometry and FVM. Little to
no topology.
* knepley/plex:
SNES ex12: Fix leak
TS ex11: error if dim != DIM because memory corruption occurs otherwise
DMPlex: All quadrilateral geometry tests working
DMPlex: remove unused variable
DMPlex: Hexahedron geometry tests now pass
DMPlex: More geometry work
DMPlex: use PETSC_HAVE_TGAMMA to get the build working on windows
configure: add test for tgamma() [this is missing on windows]
DMPlex: Added random transform tests for tets - Fixed memory leak
DMPlex: Fixed 3D geometry
DMPlex: FVM geometry tests now work for triangles
DMPlex: one more win32/complex fix
DMPlex: Enhanced geometry tests - Added random transforms of the reference triangle in 2D
DMPlex: Fixed complex problems in geometry calculation
DMPlex: Fixed Fortran test output
DMDT: M_PI --> PETSC_PI
DMPlex: Reorganized geometry testing
TS ex11: Put in 3D inflow and outflow ids from the command line
DMPlex: Added DMPlexRestoreCone/Support() to F90 interface definition header - Also added to test ex1f90
TS ex11: Need to make these ids input from the command line
DMPlex: Add Fortran bindings
DMPlex: Reorder tet output since I order tets differently
Builder: Link Fortran examples with Fortran linker
DMPlex: Fixed leak in example
DMPlex: Tutorial now shows how to output VTK
PetscViewer: Added VTK type for Fortran
DMPlex: Remove parameter names from prototype
DMPlex: Geometry fixes for C++ complex
DMAKKT: Remove from build
DMPlex: Now DMPlexCreateSection() only has an F90 interface
SNES: add Fortran bindings to DMSNESSet{Function,Jacobian}
SNES: remove superfluous casting in DM local Fortran bindings
SNES: compile zdmlocalsnesf.c
DMDT: Fix Fortran wrapper
DMPlex: Added Fortran interface for DMPlexMatSetClosure()
PetscSection: Added docs, enabled Fortran wrappers
DM: Added Fortran bindings for DMSNESSetFunction/JacobianLocal()
SNES ex12: Looks like 3D Neumann conditions are working
DMPlex: Fix volume determination for 2D in 3D
DMPlex: Better error reporting
SNES ex12: Boundary integration seems to work in 2D
DMPlex+FEM: Fix for boundary integration
TS ex11: Better error reporting
DMPlex ex8: Added FVM geometry tests
DMPlex: Fixed FVM geometry for 2D in 3D
SNES ex62: Now using PetscDT quadrature
DMDT: Translated FIAT's Gauss-Jacobi quadrature
DMPlex: Fix geometry tests - Fix tests for 2D projection
DMPlex: Geometry fixes
DMPlex: Small fixes for geometry
DMPlex: Fixed damn prototype
PetscSection: Added VecSetValuesSectionF90() - Fixed bug in Vec Fortran header
DMPlex: Added Fortran defines for additional insert modes
DMPlex: Silence warning in DMPlexInvertCells_Internal()
DMPlex: Functions passed to evaluation routines now return void and pass results in arguments
DMPlex: Now we give an explicit embedding dimension to the FVM geometry methods
DMPlex: Fixed complication with complex - Started to add 3D geometry stuff
TS ex11: Replace edge geometry with call to DMPlexComputeCellGeometryFVM()
DMPlex: Added normal argument to DMPlexComputeCellGeometryFVM() - This is only calculated for faces - Added calculation for faces in 2D
DMPlex: Damn makefiles
DMPlex: Added declaration for DMPlexCreateCGNS()
TS ex11: Reorganized computation of geometry in preparation for 3D
Builder: Now individual tests can have requirements - The key is 'requires', and it takes a list of package names - Also fixed up showSingleRun()
TS ex11: Use new DMPlexComputeCellGeometryFVM()
DMPlex: Added DMPlexComputeCellGeometryFVM() - Added internal volume methods
DMPlex: fix const and int vs. PetscInt
DMPlex: Better error reporting
DMPlex+ExodusII: Fixed reading of quads - Was broken by 2e1b13c25062c3c40593ce7412c5cd227259ade7
DMPlex: Hex cell geometry was broken - I don't see how this test passed before
DMPlex: Regression cleanup
DMPlex: Regression cleanup
SNES ex12: Fixed sign for Neumann BC
DMPlex+FEM: The dimension does not change for boundary elements
DMPlex: Geometry now works for boundary elements
DMPlex+FEM: Fixed quadrature coordinate handling for boundary integrals
SNES ex12: Reorganized tests, still working on Neumann conditions - Fixed boundary face label
DMPlex: Turn off CGNS by default
DMPlex: Added 1D cell geometry
DMPlex: DMPlexCreateCGNS()
DMPlex: Updated test ex1 output
DMPlex: Fixed compiler problems with complex
SNES ex12: Now does boundary integration - Still not verifying exact solution for Neumann conditions
DMPlex: Fix DMPlexComputeCellGeometry() to handle lower dimensional cells
DMPlex: Fixed error in projection from 3D to 2D - Need to check for case where normal is already z
DMPlex: Adding boundary integration
DMPlex: Fixed cell inversion for TetGen to commute with mesh interpolation
SNES ex12: Cleaning up ex12 testing
DMPlex: Added integration over boundary to DMPlexComputeResidualFEM() - Added quadBd, f0/1BdFuncs to PetscFEM struct - Needs to be tested
FEM Generation: We now allow a *_bd.h header that holds boundary discreization
Builder: Fixed regression requirements
Builder: Fixed specification of multiple tests to run
DMPlex ex7: Orientations now working properly
DMPlex: Changed orientation convention and fixed many bugs
SNEX ex12: Ignore generated headers
SNEX ex12 and ex52: Remove CTetGen warnings from test output - Fixed this in CTetGen repository
DMPlex: Fix default |J| for DMPlexComputeCellGeometry()
DMPlex: Fix for cell geometry (unbelievable screwup)
DMPlex ex7: Check that |J| is nonzero for each cell in interpolated mesh
DMPlex: Allow DMPlexComputeCellGeometry() to work with interpolated meshes
DMPlex: Added prototype for DMPlexCopyCoordinates(), and docs for that and DMPlexInterpolate()
DMPlex: Added F90 interface for DMPlexComputeCellGeometry()
Conflicts:
include/petscdmplex.h
show more ...
|
| 8b83055f | 05-May-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
PetscStackCallBLAS: new macro instead of PetscStackCall (skip CHKMEMQ)
When CHKMEMQ occurs on every call to BLAS or Lapack, we sometimes see
unacceptable performance degradation (orders of magnitude
PetscStackCallBLAS: new macro instead of PetscStackCall (skip CHKMEMQ)
When CHKMEMQ occurs on every call to BLAS or Lapack, we sometimes see
unacceptable performance degradation (orders of magnitude).
PetscStackCallBLAS is currently identical to PetscStackCall except for
skipping CHKMEMQ, but such checking could be made optional in the
future.
show more ...
|
| be366a87 | 03-May-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
Merge branch 'jed/fix-vecload-hdf5-bs'
* jed/fix-vecload-hdf5-bs:
VecLoad_HDF5: uninitialized block size is indicated by bs==-1, not bs==1 |
| 8b3dd749 | 03-May-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
SF: use #requiresdefine to guard window (needed by builder.py)
Fix for 6f2c30af2f7b15a20f2fd3f121bdd916d76bf921, which was inadequate
because MPI_Win_create is tested by MPI.py rather than through t
SF: use #requiresdefine to guard window (needed by builder.py)
Fix for 6f2c30af2f7b15a20f2fd3f121bdd916d76bf921, which was inadequate
because MPI_Win_create is tested by MPI.py rather than through the
generic "functions" interface, thus BuildSystem does not know that it is
a function.
show more ...
|
| 6f2c30af | 01-May-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
SF: use #requiresfunction to guard 'window' impl |
| 37ef07e6 | 28-Apr-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
VecLoad_HDF5: uninitialized block size is indicated by bs==-1, not bs==1 |
| 7d71e715 | 26-Apr-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
Merge branch 'jed/sf-mpich1'
The last release of MPICH1 had some MPI-2 functions without essential
parts. They strictly adhere to MPIX_* naming now so that this problem
does not occur.
* jed/sf-mp
Merge branch 'jed/sf-mpich1'
The last release of MPICH1 had some MPI-2 functions without essential
parts. They strictly adhere to MPIX_* naming now so that this problem
does not occur.
* jed/sf-mpich1:
SF: define MPIU_REPLACE to work around missing MPI_REPLACE
SF: define and work around missing MPI_COMBINER_DUP [reconfigure-needed]
show more ...
|
| 1368a9f5 | 26-Apr-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
Merge branch 'jed/fix-vecload-hdf5'
* jed/fix-vecload-hdf5:
VecLoad: ugly hack to re-enable VecLoad_HDF5 |
| e7fb2b78 | 25-Apr-2013 |
Jed Brown <jedbrown@mcs.anl.gov> |
VecLoad: ugly hack to re-enable VecLoad_HDF5
It was disabled in parent commit. |
| 8bfbc91c | 19-Apr-2013 |
Jed Brown <jed@59A2.org> |
SF: define MPIU_REPLACE to work around missing MPI_REPLACE |
| b2aac118 | 19-Apr-2013 |
Jed Brown <jed@59A2.org> |
SF: define and work around missing MPI_COMBINER_DUP [reconfigure-needed]
A standards-compliant MPI cannot provide MPI_Type_get_envelope without
also defining MPI_COMBINER_DUP, because the latter is
SF: define and work around missing MPI_COMBINER_DUP [reconfigure-needed]
A standards-compliant MPI cannot provide MPI_Type_get_envelope without
also defining MPI_COMBINER_DUP, because the latter is necessary to
interpret the results of the former. However, a version of MPICH1 does
exactly this so we need a configure test and a workaround in the code.
show more ...
|
| 74d4b734 | 16-Apr-2013 |
Matthew G. Knepley <knepley@gmail.com> |
PetscSection: Added docs, enabled Fortran wrappers |
| 3260fda8 | 15-Apr-2013 |
Jed Brown <jed@59A2.org> |
Merge branch 'maint'
* maint:
AO: fix PETSC_NULL_OBJECT in aocreatebasicis_ and aocreatememoryscalableis_ |
| 30403741 | 14-Apr-2013 |
Karl Rupp <rupp@iue.tuwien.ac.at> |
VecEqual: Updated documentation to make clear that no round-off errors are considered. |
| 4957363a | 14-Apr-2013 |
Karl Rupp <rupp@iue.tuwien.ac.at> |
Merge branch 'karlrupp/fix-cusp-mdot' |
| 62a20339 | 13-Apr-2013 |
Jed Brown <jed@59A2.org> |
Merge branch 'jed/aijcusparse-icc'
* jed/aijcusparse-icc:
MatSeqAIJCUSPARSE: make private functions static
txpetscgpu: upgrade to version 0.1.0
aijcusparse : fixed MatGetFactor and other small
Merge branch 'jed/aijcusparse-icc'
* jed/aijcusparse-icc:
MatSeqAIJCUSPARSE: make private functions static
txpetscgpu: upgrade to version 0.1.0
aijcusparse : fixed MatGetFactor and other small issues in this class
ksp tests: new tests for ksp algorithms using aijcusparse methods
vscatcusp : fixed compiler warnings
MatSeqAIJCUSPARSE: white space, style issues, static and extern functions
aijcusparse: ICC/Cholesky preconditioners
Conflicts:
src/mat/impls/aij/seq/seqcusparse/aijcusparse.cu (PetscObjectComposeFunction usage)
show more ...
|
| 710b1ebf | 12-Apr-2013 |
Matthew G. Knepley <knepley@gmail.com> |
PetscSection: Added VecSetValuesSectionF90()
- Fixed bug in Vec Fortran header |
| d52f9e1a | 12-Apr-2013 |
Karl Rupp <rupp@iue.tuwien.ac.at> |
CUSP: Improved mdot() performance by another 10 percent. |
| 20cbb662 | 11-Apr-2013 |
Jed Brown <jed@59A2.org> |
Merge branch 'jed/fix-locdir'
* jed/fix-locdir:
LOCDIR: fix all outdated paths and normalize trailing slash |