Ginkgo config solver #272
Ginkgo config solver #272
Conversation
Co-authored-by: Yu-Hsiang M. Tsai <yhmtsai@gmail.com>
|
Deployed test documentation to https://exasim-project.com/NeoFOAM/Build_PR_272 |
Co-authored-by: Yu-Hsiang M. Tsai <yhmtsai@gmail.com>
MarcelKoch
force-pushed
the
feat/ginkgo-config-solver
branch
from
398638c to
0c6cbe1
Compare
|
|
||
|
|
||
| NeoFOAM::la::ginkgo::BiCGStab<ValueType> solver(solution.exec(), fvSolution); | ||
| NeoFOAM::la::ginkgo::Solver<ValueType> solver(solution.exec(), fvSolution); |
There was a problem hiding this comment.
line 116-119 format seems to be incorrect
There was a problem hiding this comment.
clang format seems to be fine with it.
| if (auto node = parseAny(int {})) | ||
| { | ||
| return node; | ||
| } | ||
| if (auto node = parseAny((unsigned int) {})) | ||
| { | ||
| return node; | ||
| } |
There was a problem hiding this comment.
any other integer format from Dictionary?
There was a problem hiding this comment.
Possibly, but it doesn't make much sense right now to check for every possibility, since the dictionary is too unconstrained.
test/linearAlgebra/ginkgo.cpp
Outdated
|
|
||
| #if NF_WITH_GINKGO | ||
|
|
||
| bool operator==(const gko::config::pnode& a, const gko::config::pnode& b) |
There was a problem hiding this comment.
I create a pr in ginkgo-project/ginkgo#1801
|
|
||
| template<typename ExecSpace> | ||
| bool isNotKokkosThreads(ExecSpace ex) | ||
| bool isNotKokkosThreads([[maybe_unused]] ExecSpace ex) |
There was a problem hiding this comment.
| bool isNotKokkosThreads([[maybe_unused]] ExecSpace ex) | |
| bool isNotKokkosThreads(ExecSpace) |
it should also eliminate the warning by deleting the variable name.
but I am okay with the keywords
- fixes tests - fixes formatting Co-authored-by: Yu-Hsiang M. Tsai <yhmtsai@gmail.com>
This is necessary to prevent clashes with existing cmake targets from third-party libraries build through CPM.
| exec, gko::dim<2> {nrows, nrows}, valuesView, colIdxView, rowPtrView | ||
| return gko::share(gko::matrix::Csr<ValueType, IndexType>::create( | ||
| exec, | ||
| gko::dim<2> {nrows, nrows}, |
There was a problem hiding this comment.
nit: nrows from mtx.nRows() not from rhs?
There was a problem hiding this comment.
For now the matrices are assumed to be square. I think this will also stay like this for quite a while.
5 participants
Motivation
Replaces the Ginkgo solvers, by a single solver which is configured through a ginkgo property tree.
This allows to use any Ginkgo solver and preconditioner combination that is available.
TODO: check how the value type in the property tree should be handled.