Add support for experimental alternative of Dune::BCRSMatrix#7124
Add support for experimental alternative of Dune::BCRSMatrix#7124SoilRos wants to merge 1 commit into
Conversation
|
@akva2 I would like to eventually test the experimental mode in jenkins, but of course, ISTL would need to be pulled from the right branch. With markus, we thought that we could update the OPM branch from dune-istl ( |
|
jenkins build this please |
|
you find the opm ci docker container here, https://github.com/OPM/opm-utilities/tree/master/opm-ci-builder it currently pulls dune-istl from the ppa, we have to build it instead. we already build from source for the serial and debug-iter enabled configurations, so building from source for the "vanilla" is very doable. if it becomes a "hard" dependency i can build dune packages in the ppa, but i'd prefer to not do that. |
f5d49fc to
758e6f7
Compare
|
jenkins build this please |
758e6f7 to
c2502e5
Compare
|
jenkins build this please |
c2502e5 to
98d6b2f
Compare
|
jenkins build this please |
98d6b2f to
6e9a578
Compare
|
jenkins build this please |
1 similar comment
|
jenkins build this please |
3497317 to
1508f96
Compare
|
jenkins build this please |
1508f96 to
154a152
Compare
|
jenkins build this please |
154a152 to
7e97cb6
Compare
|
jenkins build this please |
7e97cb6 to
ccf9213
Compare
|
jenkins build this please |
|
(sorry for the noice with jenkins, I do not have a GPU locally to reproduce the erros) |
ccf9213 to
d2c4911
Compare
|
jenkins build this please |
Update matrix type definitions across the codebase to use Dune::IBCRSMatrix instead of the older Dune::BCRSMatrix when using experimental mode. This transition affects well models, thermal and tracer models, and the linear solver backends, including specialized support for external solvers like UMFPack and SuperLU.
d2c4911 to
28d613e
Compare
|
jenkins build this please |
Update matrix type definitions across the codebase to use
Dune::IBCRSMatrixinstead of the classicDune::BCRSMatrixwhen using experimental mode (see dune-istl !646). This transition affects well models, thermal and tracer models, and the linear solver backends, including specialized support for external solvers like UMFPack and SuperLU. The speedup from using the experimental mode is expected to be around 10% on distributed workloads (either MPI or multiple independent simulations working simultaneously).