1888/dense linop by MarcelKoch · Pull Request #2042 · ginkgo-project/ginkgo

MarcelKoch · 2026-07-02T16:11:03Z

This PR adds a Dense matrix format.

yhmtsai · 2026-08-18T11:17:11Z

-}
-
+                                        const LinOp* beta,
+                                        LinOp* x) const GKO_NOT_IMPLEMENTED;


because here is marked not implemented. the corresponding kernel and common hook can also be removed

The corresponding kernel is used in matrix::Dense.

yhmtsai

submit the partial review. Github seems to bring some changes in rename-dense here currently

yhmtsai · 2026-08-18T11:17:34Z

these two can also be removed

yhmtsai · 2026-08-18T11:18:04Z

can be removed

yhmtsai · 2026-08-18T11:23:10Z

-        if (dynamic_cast<const Diagonal<ValueType>*>(b)) {
-            exec->run(multivector::make_add_scaled_diag(
-                make_temporary_conversion<ValueType>(alpha)
-                    ->get_const_device_view(),
-                dynamic_cast<const Diagonal<ValueType>*>(b),
-                this->get_device_view()));
-        } else {


the diagonal part?

I'm not sure how much this is needed for the vectors. It is already available for dense so it can be done via vec->as_dense_view()->add_scaled(diag). But I can also add it here (or move it from dense to here).

yhmtsai · 2026-08-18T11:29:00Z

+ * @param exec  executor where the matrix should be allocated
+ * @param args  additional arguments for the matrix constructor
+ *
+ * @return the unique pointer of gko::matrix::MultiVector<ValueType>


Suggested change

* @return the unique pointer of gko::matrix::MultiVector<ValueType>

* @return the unique pointer of gko::matrix::Dense<ValueType>

yhmtsai · 2026-08-18T12:19:14Z

    if ((mode & permute_mode::rows) == permute_mode::rows) {
        // compute P * A
-        permutation_csr->apply(input, result);
+        result = permutation_csr->multiply(input);


check multiply interface

What are you asking here?

yhmtsai · 2026-08-18T12:20:40Z

    {
        csr1 = Csr::create(exec);
-        csr1->copy_from(dense1);
+        csr1->copy_from(dense1->as_const_dense_view().get());


should these be created as dense not multivector?

No, they are used as vectors everywhere but here.

yhmtsai · 2026-08-18T12:27:58Z

+    auto mtx = gko::share(MultiVector::create_with_config_of(
+        this->l_dense->as_multivector_view()));
+    this->l_dense->apply(this->u_dense->as_const_multivector_view(), mtx);
    auto ilu_factory =
        gko::preconditioner::Ilu<value_type, false, index_type>::build()
            .with_l_solver(LowerIsai::build())
            .with_u_solver(UpperIsai::build())
            .on(this->exec);
-    auto ilu = ilu_factory->generate(mtx);
+    auto ilu = ilu_factory->generate(mtx->as_const_dense_view());


this is how user will use dense/multivector.

I think here it could also be done with storing one multivector and one dense object, but I think I was lazy.

yhmtsai · 2026-08-18T12:41:18Z

-    auto permuted = gko::as<typename TestFixture::Mtx>(
-        this->star_mtx->permute(&perm_array));
+    auto permuted = this->star_mtx->as_multivector_view()
+                        ->permute(&perm_array)
+                        ->as_dense_view()
+                        ->clone();


interesting

I think the clone is necessary to keep the return value of permute alive.

yhmtsai · 2026-08-18T12:43:07Z

    using initial_guess_mode = gko::solver::initial_guess_mode;
-    auto ref_solver =
+    auto ref_factory =
        gko::solver::Ir<value_type>::build()
            .with_criteria(gko::stop::Iteration::build().with_max_iters(1u))
-            .on(this->exec)
-            ->generate(this->mtx);
+            .on(this->exec);


any reason not to generate ref_solver here?

don't remember. Seems to work without it, so I will revert it

This reverts commit b930916.

Signed-off-by: Marcel Koch <marcel.koch@kit.edu>

- diagonal extractable - scaled identity addable - add/sub scaled diagonal

MarcelKoch self-assigned this Jul 2, 2026

MarcelKoch mentioned this pull request Jul 3, 2026

Tracking Issue: MultiVector #1888

Open

9 tasks

MarcelKoch force-pushed the 1888/renaming branch from e0bbbd9 to 27a7815 Compare July 7, 2026 11:38

MarcelKoch force-pushed the 1888/dense-linop branch from 73996c7 to 1e611b2 Compare July 7, 2026 11:38

MarcelKoch force-pushed the 1888/dense-linop branch from fbd5d57 to 59a1fa6 Compare July 14, 2026 12:36

MarcelKoch force-pushed the 1888/renaming branch from 27a7815 to 65e737c Compare July 14, 2026 12:36

MarcelKoch force-pushed the 1888/dense-linop branch from 59a1fa6 to 7e96d52 Compare July 14, 2026 12:41

MarcelKoch force-pushed the 1888/renaming branch 2 times, most recently from 04359e0 to aa6583d Compare July 20, 2026 16:11

MarcelKoch force-pushed the 1888/dense-linop branch 2 times, most recently from 6c1727e to 8728fb4 Compare July 20, 2026 16:35

MarcelKoch force-pushed the 1888/renaming branch from aa6583d to 17b8605 Compare August 7, 2026 13:38

MarcelKoch force-pushed the 1888/dense-linop branch from bd7e3ef to 8a641ad Compare August 7, 2026 13:38

MarcelKoch changed the base branch from 1888/renaming to 1888/rename-dense August 7, 2026 13:38

MarcelKoch changed the base branch from 1888/rename-dense to 1888/renaming August 7, 2026 13:41

MarcelKoch force-pushed the 1888/renaming branch from 17b8605 to 230bb98 Compare August 7, 2026 13:45

MarcelKoch changed the base branch from 1888/renaming to 1888/rename-dense August 7, 2026 13:45

MarcelKoch force-pushed the 1888/dense-linop branch from 8a641ad to 8f00cd9 Compare August 17, 2026 16:00

MarcelKoch force-pushed the 1888/rename-dense branch from 17b8605 to da076d1 Compare August 17, 2026 16:17

MarcelKoch force-pushed the 1888/dense-linop branch from 8f00cd9 to e73b43c Compare August 17, 2026 16:24

MarcelKoch force-pushed the 1888/rename-dense branch from da076d1 to e5527df Compare August 17, 2026 16:24

MarcelKoch added this to the Ginkgo 2.0 milestone Aug 17, 2026

yhmtsai reviewed Aug 18, 2026

View reviewed changes

MarcelKoch force-pushed the 1888/rename-dense branch from e5527df to b930916 Compare August 18, 2026 12:39

yhmtsai reviewed Aug 18, 2026

View reviewed changes

MarcelKoch force-pushed the 1888/dense-linop branch 2 times, most recently from ea2b468 to 01e46b2 Compare August 18, 2026 14:57

MarcelKoch added 1:ST:ready-for-review This PR is ready for review 1:ST:skip-full-test labels Aug 18, 2026

MarcelKoch added 2 commits August 19, 2026 14:15

Revert "add Dense type alias"

ac00077

This reverts commit b930916.

implement dense matrix format

4c0cc61

MarcelKoch force-pushed the 1888/dense-linop branch from 01e46b2 to 9ef2419 Compare August 19, 2026 12:38

MarcelKoch force-pushed the 1888/rename-dense branch from b930916 to 17b823e Compare August 19, 2026 12:38

MarcelKoch added 7 commits August 19, 2026 14:55

add tests for dense

f4946be

Signed-off-by: Marcel Koch <marcel.koch@kit.edu>

remove multivector apply tests

8815817

move conversion functionality to dense

d5a7e90

move initialize functions to dense.hpp

2f0de80

fix conversion tests

5155a57

updates tests and examples

f5c2af4

Signed-off-by: Marcel Koch <marcel.koch@kit.edu>

move functionality to dense

00348a0

- diagonal extractable - scaled identity addable - add/sub scaled diagonal

MarcelKoch force-pushed the 1888/dense-linop branch from 9ef2419 to 00348a0 Compare August 19, 2026 12:56

	* @return the unique pointer of gko::matrix::MultiVector<ValueType>
	* @return the unique pointer of gko::matrix::Dense<ValueType>

Uh oh!

Conversation

MarcelKoch commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yhmtsai left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

MarcelKoch commented Jul 2, 2026 •

edited

Loading