- 30 Jul, 2020 4 commits
-
-
Dominik Thoennes authored
-
Dominik Thoennes authored
-
Daniel Drzisga authored
-
Dominik Thoennes authored
fix docu image See merge request !345
-
- 29 Jul, 2020 7 commits
-
-
Dominik Thoennes authored
elementwise vs constant op See merge request !344
-
Daniel Drzisga authored
-
Dominik Thoennes authored
-
Nils Kohl authored
-
Dominik Thoennes authored
-
Dominik Thoennes authored
-
Marcus Mohr authored
Mixed forms for P2-P1 Stokes with blending Closes #126 See merge request !343
-
- 27 Jul, 2020 4 commits
-
-
Marcus Mohr authored
-
Marcus Mohr authored
Comparison of local element matrix between HyTeG and FEniCS with "challenging" affine blending looks good.
-
Marcus Mohr authored
This is another step towards the implementation including blending. We test that the results we get from our HyTeG form match those from the FEniCS form. Next step will be to include the blending.
-
Marcus Mohr authored
-
- 26 Jul, 2020 2 commits
-
-
Dominik Thoennes authored
fix coverage reports See merge request !340
-
Dominik Thoennes authored
-
- 22 Jul, 2020 1 commit
-
- 13 Jul, 2020 1 commit
-
- 08 Jul, 2020 1 commit
-
-
Dominik Thoennes authored
improve coverage; enable clang 10; change walberla module to relative path See merge request !339
-
- 07 Jul, 2020 3 commits
-
-
Dominik Thoennes authored
-
Dominik Thoennes authored
-
Dominik Thoennes authored
-
- 06 Jul, 2020 7 commits
-
-
Dominik Thoennes authored
-
Dominik Thoennes authored
-
Nils Kohl authored
-
Nils Kohl authored
-
Dominik Thoennes authored
-
Dominik Thoennes authored
-
Dominik Thoennes authored
-
- 03 Jul, 2020 6 commits
-
-
Nils Kohl authored
-
Nils Kohl authored
-
Nils Kohl authored
-
Nils Kohl authored
-
Nils Kohl authored
Objective was the reduction of global communication to a minimum. The migration did not scale well above 10k processes in some experiments. Major changes - not only the primitive -> rank map, but also the number of of expected primitives after migration is required for migration (wrapped by new class MigrationInfo) - added helper function that gathers the number of expected primitives if it cannot be easily calculated by the load balancing strategy (this method is expensive / scales badly, though) - replaced global reduction of migration info by neighborhood communication, gathering the future ranks of neighbor primitives before the actual migration - replaces global p2p message passing by using new PackageBufferSystem (no need to send empty messages to all processes anymore) - renamed copyDistribution() to reverseDistribution() because it fits better, also massively refactored the code - some minor test fixes
-
Nils Kohl authored
This buffer system is intended to allow communication if the sender ranks are not knowns on the receiver side, but the number of packages that are expected, is. Such functionality is not available in walberla. Thus, this might be ported to the walberla core eventually.
-
- 23 Jun, 2020 3 commits
-
-
Dominik Thoennes authored
use gcc:9 image for coverage job See merge request !337
-
Dominik Thoennes authored
- 22 Jun, 2020 1 commit
-
-
Nils Kohl authored
-