| 19180038 | 11-Jul-2011 |
Matthew G Knepley <knepley@gmail.com> |
Added MPI GPU assembly (almost working)
- Using KSP ex4 to test
Hg-commit: 47b63e64abf7eb11ce9a217ac5edc2375647124e |
| a7f19b73 | 27-Jun-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
commit after merge
Hg-commit: d900e2595c9c592b0b11ea851fe12c0b94974881 |
| f15663dc | 27-Jun-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
cleaned up MatMultAdd_SeqAIj() to match MatMult_SeqAIJ()
changed KSPSolve_LSQR() to only compute matrix norm requested (which is currently never) rather than for each solve
Hg-commit: b43bb77e3c20c7
cleaned up MatMultAdd_SeqAIj() to match MatMult_SeqAIJ()
changed KSPSolve_LSQR() to only compute matrix norm requested (which is currently never) rather than for each solve
Hg-commit: b43bb77e3c20c759ac5e2fafd53ef058199516fe
show more ...
|
| f4b4226b | 27-Jun-2011 |
Matthew G Knepley <knepley@gmail.com> |
More comments in ex4, and added aijAssemble.cu to makefile
Hg-commit: 69b7e629032d1875121dfa56d83fdd4fb52bccf0 |
| 47adae6f | 26-Jun-2011 |
Matthew G Knepley <knepley@gmail.com> |
Solve system in ex4
Hg-commit: f358294c14916e2c9a5ddef4f4d3f4e66d4392c3 |
| d751767b | 26-Jun-2011 |
Matthew G Knepley <knepley@gmail.com> |
Now assembling 150K rows/1M unknowns in < 0.33s on crappy laptop
- Need a comparison to naive assembly
Hg-commit: deca59027988654e50cb48e632042b59c5cdd1bc |
| 80d50684 | 26-Jun-2011 |
Matthew G Knepley <knepley@gmail.com> |
More progress on KSP ex4
- Can now use different sized DAs
Hg-commit: a0bb4fc13a0d856b97424f41bc9511199ed6dab5 |
| 6acf6a18 | 25-Jun-2011 |
Matthew G Knepley <knepley@gmail.com> |
Added a small example of GPU assembly
- Will extend soon for Lanzhou conference
Hg-commit: e9bc1707d933002a8530dd703a2e05731dc53f9d |
| 0ca81413 | 23-Jun-2011 |
Kerry Stevens <kstevens@spetjhiong.com> |
added more pthread vector functions
added a pthread matrix function (not yet operational)
Hg-commit: 402caddad6a4ab372111fbc5baac892290efd412 |
| 588e25ff | 13-Jun-2011 |
Matthew G Knepley <knepley@gmail.com> |
Replaced my code that some idiot removed in changeset b056329a6646
Hg-commit: b3087da7604d964098b43a3e023ea99c87110fbb |
| 71cd77b2 | 13-Jun-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
added support for DM handlingly finite difference computation of Jacobian if no Jacobian is provided
Hg-commit: b9e17f19f814636f3cab90745af0ff3b00101a76 |
| 1d019be2 | 08-Jun-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
never use variable flops to count flops
Hg-commit: 67e37857a3320060a2831e3ee3c9d7ee523d1407 |
| 7dca3203 | 08-Jun-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
undid my bad changes to MatPtAPNumeric_SeqAIJ_SeqAIJ()
Hg-commit: a0171b9668502a53972cf65a0ff8de5bc1cf89ef |
| 342cc144 | 08-Jun-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
another very minor optimization to MatPtAPNumeric_SeqAIJ_SeqAIJ()
Hg-commit: 59f6c353c7c0fc2874e878019be7d24bde979c71 |
| 52ebfde1 | 08-Jun-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
minor optimizations to MatPtAPNumeric_SeqAIJ_SeqAIJ
Hg-commit: 66e57c801e984ad49db4733e22aefde520ec02a3 |
| da65836d | 08-Jun-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
forgot
Hg-commit: 25130f99e0f2fe091c27a52ed2306548b4504951 |
| 44bb0d42 | 02-Jun-2011 |
Hong Zhang <hzhang@mcs.anl.gov> |
a bug in superlu with options.RowPerm=LargeDiag causing src/ksp/ksp/examples/tutorials/runex52_superlu crashes.
Set default to options.RowPerm = NOROWPERM until the bug is fixed in superlu.
Hg-commi
a bug in superlu with options.RowPerm=LargeDiag causing src/ksp/ksp/examples/tutorials/runex52_superlu crashes.
Set default to options.RowPerm = NOROWPERM until the bug is fixed in superlu.
Hg-commit: 98757329eb0a29d387ad3318c783b3c357f5e944
show more ...
|
| d0660788 | 31-May-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
added matrix object to MatCheckPivot routines to allow displaying more information on failure
PCEXOTIC turns off DM object in PCMG so that PCSetUp_MG() doesn't use the DM object to generate incorrect
added matrix object to MatCheckPivot routines to allow displaying more information on failure
PCEXOTIC turns off DM object in PCMG so that PCSetUp_MG() doesn't use the DM object to generate incorrect work vectors
SNESSetDM,KSPSetDM and PCSetDM allow using 0 dm to clear current DM.
Hg-commit: d3ce2e5ec6f3ec4502b397e3155085dddbdcd62e
show more ...
|
| 4dcab191 | 27-May-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
added DMClearGlobalVectors() that destroys all the stashed global vectors
added DM_SNESVI basics for supporting the use of DM with the VI solvers basic version with no multigrid works
Hg-commit: 3ba
added DMClearGlobalVectors() that destroys all the stashed global vectors
added DM_SNESVI basics for supporting the use of DM with the VI solvers basic version with no multigrid works
Hg-commit: 3baea364d1194fdae279c6bd14ddd04795eaaea2
show more ...
|
| 992144d0 | 15-May-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
moved localtoglobal mapping bmapping of matrices and vectors into the PetscLayout object
Hg-commit: 9101686a27a1dc39bda221d0212c5e4e16e2b091 |
| aa5ea44d | 15-May-2011 |
Barry Smith <bsmith@mcs.anl.gov> |
MatDuplicate_XXX() and MatGetVecs_XXX() now reference the previous objects PetscLayout instead of making a new one for the created objects
Hg-commit: 09acbe7f87158c1125044b0743c911c532c88a17 |
| 23508aee | 13-May-2011 |
Paul Mullowney <paulm@txcorp.com> |
Fixing error codes. Tests pass.
Hg-commit: bdf819c75c20406e5a6865f364261735fcc4ff47 |
| 2576faa2 | 21-Jan-2012 |
Jed Brown <jed@59A2.org> |
Fixes for insertion error options.
Option should be set if "real" preallocation information was
provided. If PETSC_DECIDE was used, then do not error. The option is
inherited through a DisAssemble_*
Fixes for insertion error options.
Option should be set if "real" preallocation information was
provided. If PETSC_DECIDE was used, then do not error. The option is
inherited through a DisAssemble_* operation.
Hg-commit: e675ef955f926f7569d867832ae3d74f566041a1
show more ...
|
| 237e4908 | 21-Jan-2012 |
Paul Mullowney <paulm@txcorp.com> |
Slight fix
Hg-commit: dc8047fc3dc01bc1a31029d61360b3e26d997f63 |
| 29f304a0 | 21-Jan-2012 |
Paul Mullowney <paulm@txcorp.com> |
MatCUSPARSEAnalysisAndCopyToGPU is broken into 3 functions now. This should be more manageable.
Hg-commit: 2742d5d07e9625c6a7b0c4673c8740e267596436 |