~jon-hill/fluidity/diagnostic

~jon-hill/fluidity/diagnostic_check » Changes from revision 4307

From Revision 4285 to 4266

Rev	Summary	Authors	Date
4285	BUG-FIX: unsafe storage of halos on petsc_csr_matrix. Only a... BUG-FIX: unsafe storage of halos on petsc_csr_matrix. Only a pointer to the passed in halo was stored in the allocate routines, which means that the halo would be invalidated if the halo is deallocated before the matrix.	Stephan Kramer	10 years ago
4284	BUG FIX: making sure that if val isn't found in the Python f... BUG FIX: making sure that if val isn't found in the Python funciton, the user gets an error, not a seg fault	Jon Hill	10 years ago
4283	BUG FIX: Minor change to paths in the distclean target of th... BUG FIX: Minor change to paths in the distclean target of the top level Makefile to deal with a change in the spatialindex codebase where the 'regressiontests' directory has been renamed to 'tests'. This meant that some .deps files which are not removed by spatialindex' own distclean were not being cleaned up by Fluidity's distclean.	Tim Greaves	10 years ago
4282	Updating Makefile dependencies which were missed from the Zo... Updating Makefile dependencies which were missed from the Zoltan tweaks merge.	Tim Greaves	10 years ago
4281	BUG-FIX: when setting a vector field from python, check for ... BUG-FIX: when setting a vector field from python, check for PyErr_Occurred() directly after the python is evaluated. Previously it tested the shape of the returned value first, so that the error message from python would get lost.	Stephan Kramer	10 years ago
4280	This branch fixes a Zoltan bug that caused major issues with... This branch fixes a Zoltan bug that caused major issues with load balancing and node locking, particularly in three-dimensional simulations. The cause of this problem was that two levels of halo existed during the adapt process. It is important that only one halo exists during the adapt process to allow for adapting to occur as close to the partition boundaries as possible. in Assemble/Adapt_State.F90 we now strip the level 2 halo prior to calling the adapt_mesh routines. Following the adapt the 2nd halo is rebuilt. Two new functions were created to do this. These are strip_l2_halo and create_l2_halo. These functions are very similar to strip_level_2_halo and sam_drive within Sam_Integration.F90. strip_l2_halo is the same as create_l2_halo except that it works only on a coordinate field and metric rather than on an entire state and metric. create_l2_halo is a stripped down version of sam_drive. A new function sam_migrate_halo2() is added to libadaptivity/load_balance/src/fluidity_sam.cpp which does the same as sam_migrate() but skips the decomp step. this is called instead of sam_migrate within create_l2_halo. Additionally, some changes were required in the interpolation routines to avoid post-adapt interpolation within the level 2 halo. Halo 2 elements are allowed to adapt. Because of this there is no guarantee that target mesh elements in halo 2 will be a subset of donor elements. We cannot therefore pass information across from the donor to target mesh within this region. Instead, we only update the level 1 halo, and then do a halo update to fix the halo 2 information. A number of other changes were made to how zoltan works: - global_min_quality is now always passed into zoltan_drive whereas before it was only passed in if zoltan_additional_adapt_iterations was enabled. This means that the global_min_quality is always calculated and printed to the fluidity log. This is really useful output. - a function has been added that gradually reduces the required element quality tolerance if nodes are locked for several iterations (potentially could take this out now that it is all working much better) - The method of searching for poor-quality elements and determining edge weights has been changed. This allows for variable searching depths. - Exposed a few more zoltan api options in the schema. Can now choose whether to REPARTITION/PARTITION for each adapt stage. - Algorithm for finding maximum load balance tolerance that will not produce empty partitions is much improved in speed and performance (again we may be able to remove this completely now the halo issue is fixed)	Sam Parkinson	10 years ago
4279	Minor adjustment to manual only Minor adjustment to manual only	Jon Hill	10 years ago
4278	One word documentation change. One word documentation change.	Cian Wilson	10 years ago
4277	Merge in removal of the use_compressible_projection_method o... Merge in removal of the use_compressible_projection_method option which was partially deprecated and unnecessary.	Cian Wilson	10 years ago
4276	Modify this test so that is passes with binary gmsh. Thanks ... Modify this test so that is passes with binary gmsh. Thanks to Tim for pointing out the issue.	Rhodri Davies	10 years ago
4275	Fixes for openmp build. Fixes for openmp build.	Stephan Kramer	10 years ago
4274	Bug-fix: fix viscous free surface in periodic domains. Bug-fix: fix viscous free surface in periodic domains.	Stephan Kramer	10 years ago
4273	Merge branch into trunk, thus fixing spurious warning bug. Merge branch into trunk, thus fixing spurious warning bug.	Rhodri Davies	10 years ago
4272	Updating nprocs in meshconv_test to correspond to the resour... Updating nprocs in meshconv_test to correspond to the resources actually required by the running test.	Tim Greaves	10 years ago
4271	Updating femtools/Makefiles.dependencies following r.4270. Updating femtools/Makefiles.dependencies following r.4270.	Tim Greaves	10 years ago
4270	1) Removed ALGENCAN 1) Removed ALGENCAN 2) testharness now figures out where itself lives (and therefore what the fluidity python path is) in the event that import fluidity.regressiontest fails. This means you do not have to edit PYTHONPATH. 3) In the event PETSC_DIR is not set, autoconf now seeks a sensible choice on the host system. Tested on Ubuntu 13.04. 4) Use tcmalloc when available and debugging is not enabled.	Tim Greaves	10 years ago
4269	Make fluidity build with petsc 3.4 Make fluidity build with petsc 3.4 One, infrequently diagnostics solver option (diagnostics/monitors/preconditioned_residual_graph) is not yet supported with 3.4 Need some more time to figure out how to do this with KSPMONITORLGCreate(). Not really important for next release. Passes at least "make test", i.e. short tests.	Stephan Kramer	10 years ago
4268	Improved support for internal boundaries, in particular 2d a... Improved support for internal boundaries, in particular 2d adaptivity. After this merge, you can: * read in a gmsh or triangle mesh with an internal boundary directly. * use 2d adaptivity with an internal boundary This functionality is tested in a new test, internal_boundary_adaptivity. NOTE that the actual applying of internal boundaries to the equations is not very well tested. This depends very much on which discretisation is used and which type of bcs. It should therefore be carefully evaluated for the case you want to apply it to. You will probably have most luck with strong bcs, with weak bcs your mileage may vary.	Stephan Kramer	10 years ago
4267	Updates within the testing infrastructure, primarily to appl... Updates within the testing infrastructure, primarily to apply lessons learned when attempting to run these tests in a cluster environment, but also removing some no-longer-needed tests. This also adapts testharness to allow explicitly selecting PBS-run tests.	Tim Greaves	10 years ago
4266	Update Makefile.dependencies. Update Makefile.dependencies.	Stephan Kramer	10 years ago