Coarsen and Snapping #514

Angelyr · 2025-08-06T22:06:20Z

Changes to coarsen

improvement to independent set
fixed bug that made it impossible to collapse edge in both directions
fixed bug that made it slower to collapse in one direction in parallel
added multi-step coarsen

Changes to snapping

fixed bug with snapping coarsening the mesh too mush
fixed bug where not all regions were included in first problem plane
should snap vertices in most cases
added quality settings for split collapse

More testing

added test for common errors in coarsen, refinement and snapping

Most of the these changes to snapping and coarsen came from Li's thesis with some minor exceptions:

In the coarsen procedure in Li's thesis it describes getting the shortest edge next to a vertex and collapsing it. We found we had better results if we try collapsing the shortest edge and then grab the next shortest until one succeeds.
Another difference is that Li's thesis mentions collapses, but it has very little detail in how those collapses are selected, so our logic for collapses is taken from our old mesh adapt library.
Li's thesis also describes using FaceSwap in snapping and vertex repositioning in snapping and coarsen, but we haven't created either of those operators yet.

cwsmith · 2025-08-09T01:13:48Z

/runtests

cwsmith · 2025-08-09T01:15:52Z

@Angelyr Thank you. I pushed the sim version change to develop and once it appears in master tomorrow morning runtests will use it.

cwsmith · 2025-08-11T17:07:27Z

/runtests

cwsmith · 2025-08-11T17:09:20Z

@Angelyr Would you please take a look at the failing test:

 8/95 Test  #8: bezierRefine .....................***Failed    0.23 sec

in the (gcc-10, g++-10, Debug, OFF, 20, ON) config?

Angelyr · 2025-08-11T17:16:12Z

@cwsmith Yes I will take a look.

github-actions · 2025-08-11T17:17:26Z

Build Log
Simmetrix Test Result: success
Simmetrix + CGNS Test Result: success

Angelyr · 2025-08-13T18:49:57Z

@cwsmith I was struggling to recreate the issue locally, but I re-ran the test and it passed

cwsmith · 2025-08-13T18:59:48Z

OK. Thanks for the heads up. A couple things we could try come to mind:

run the test a bunch of times with ctest --repeat-until-fail 100
check for memory errors in the test with valgrind

Angelyr · 2025-08-13T20:03:42Z

@cwsmith I fixed an uninitialized value in 2d cases. That might have been the issue

cwsmith · 2025-08-13T20:18:14Z

@Angelyr Looks good. All the github hosted ci tests are passing. Thank you.

(Note: I still need to read through the changes)

joshia5 · 2025-08-14T15:48:26Z

ma/ma.cc

As the changes fall into following categories (or more) so add comments like this throughout the code that's changed:

key logic changes to adapt as per Li's thesis

key logic changes to adapt as discussed in meeting - a sentence that summarizes discussion

bug fixes in operators

memory issues or low level software fixes
and so on..

Angelyr

I added comments to aid with reviewing the PR

ma/ma.cc

Angelyr · 2025-08-15T20:21:42Z

ma/maMatch.cc

+/* we are starting to support a few operations on matched
+meshes, sincluding snapping+UR. this should prevent snapping
+from modifying any matched entities */


Moved comment from snapping to relevant function

Angelyr · 2025-08-15T20:24:47Z

ma/maMatchedSnapper.cc

-MatchedSnapper::MatchedSnapper(Adapt* a, Tag* st, bool is)
+MatchedSnapper::MatchedSnapper(Adapt* a, Tag* st)


The option for simple snapping has been removed. This option would stop the snapping procedure from collapsing resulting in vertices that didn't snap.

Angelyr · 2025-08-15T20:30:06Z

ma/maSnap.cc

+static bool isCapstoneMesh(apf::Mesh* m)
+{
+  #ifdef HAVE_CAPSTONE
+    if (dynamic_cast<apf::MeshCAP*>(m)) return true;
+    else return false;
+  #else
+    (void)m;
+    return false;
+  #endif
+}
+


Most the changes in this file involve the replacement of a compile time capstone flag with a runtime capstone flag to ensure that these changes are being used on non-capstone meshes

Angelyr · 2025-08-15T20:31:17Z

ma/maSnap.cc

+  if (period < 0.5) //partial range can't be periodic
+    return (1-t)*a + t*b;


Solved bug caused by meshes in capstone can have partial ranges that aren't periodic

Angelyr · 2025-08-15T20:35:35Z

ma/maSnap.cc

+  bool snapped = true;
+  int prevTagged = 0;
+  while (snapped) {
+    int tagged = tagVertsToSnap(a, tag);
+    if (tagged == prevTagged) break;
+    prevTagged = tagged;
+    if (a->mesh->hasMatching())
+      snapped = snapMatchedVerts(a, tag, successCount);
+    else
+      snapped = snapAllVerts(a, tag, snapper);
+  }
+
+  a->input->shouldForceAdaptation = shouldForce;
  return successCount;
 }

+int collect(Adapt* a, int val) {
+  return a->mesh->getPCU()->Add<long>(val);
+}
+
 void snap(Adapt* a)
 {
-  if ( ! a->input->shouldSnap)
+  if (!a->input->shouldSnap)
    return;
  double t0 = pcu::Time();
-  Tag* tag;
-  /* we are starting to support a few operations on matched
-     meshes, including snapping+UR. this should prevent snapping
-     from modifying any matched entities */
+  Tag* snapTag = a->mesh->createDoubleTag("ma_snap", 3);
  preventMatchedCavityMods(a);
-  long targets = tagVertsToSnap(a, tag);
-  long success = snapTaggedVerts(a, tag);
-  snapLayer(a, tag);
-  apf::removeTagFromDimension(a->mesh, tag, 0);
-  a->mesh->destroyTag(tag);
-  double t1 = pcu::Time();
-  print(a->mesh->getPCU(), "snapped in %f seconds: %ld targets, %ld non-layer snaps",
-    t1 - t0, targets, success);
-  if (a->hasLayer)
-    checkLayerShape(a->mesh, "after snapping");
-}
-
-void visualizeGeometricInfo(Mesh* m, const char* name)
-{
-  Tag* dimensionTag = m->createIntTag("ma_geom_dim",1);
-  Tag* idTag = m->createIntTag("ma_geom_id",1);
-  apf::Field* field = apf::createLagrangeField(m,"ma_param",apf::VECTOR,1);
-  apf::Field* targetField = apf::createLagrangeField(m,"ma_target",apf::VECTOR,1);
-  Iterator* it = m->begin(0);
-  Entity* v;
-  while ((v = m->iterate(it)))
-  {
-    Model* c = m->toModel(v);
-    int dimension = m->getModelType(c);
-    m->setIntTag(v,dimensionTag,&dimension);
-    int id = m->getModelTag(c);
-    m->setIntTag(v,idTag,&id);
-    Vector p;
-    Vector xp = getPosition(m, v);
-    m->getParam(v,p);
-    if (dimension == 2 || dimension == 1) {
-      Vector x;
-      m->isParamPointInsideModel(c, &p[0], x);
-      apf::setVector(targetField, v, 0, x - xp);
-    }
-    else {
-      Vector x(0., 0., 0.);
-      apf::setVector(targetField, v, 0, x);
-    }
-    apf::setVector(field,v,0,p);
-  }
-  m->end(it);
-  apf::writeVtkFiles(name,m);
-  it = m->begin(0);
-  while ((v = m->iterate(it)))
+  int toSnap = tagVertsToSnap(a, snapTag);
+  if (toSnap)
  {
-    m->removeTag(v,dimensionTag);
-    m->removeTag(v,idTag);
+    Snapper snapper(a, snapTag);
+    snapTaggedVerts(a, snapTag, snapper);
+    snapLayer(a, snapTag);
+
+    double t1 = pcu::Time();
+    print(a->mesh->getPCU(), "ToSnap %d - Moved %d - Failed %d - CollapseToVtx %d - Collapse %d - Swap %d - SplitCollapse %d - completed in %f seconds",
+              toSnap, collect(a,snapper.numSnapped), collect(a,snapper.numFailed), collect(a,snapper.numCollapseToVtx), collect(a,snapper.numCollapse), collect(a,snapper.numSwap), collect(a,snapper.numSplitCollapse), t1 - t0);
+    if (a->hasLayer)
+      checkLayerShape(a->mesh, "after snapping");
  }
-  m->end(it);
-  m->destroyTag(dimensionTag);
-  m->destroyTag(idTag);
-  apf::destroyField(field);
-  apf::destroyField(targetField);
+  a->mesh->destroyTag(snapTag);
 }
-


Simplified old snapping logic since it was overly complicated

Angelyr · 2025-08-15T20:41:08Z

ma/maSnapper.cc

-      	dists.push_back(minDist + 1.0 + tol);
+      	dists.push_back(newDirection.getLength());


fixed bug that prevented some coplanar regions from being included in first problem plane

Angelyr · 2025-08-15T20:44:29Z

test/capVol.cc

-      adaptMesh, apf::GRAPH, apf::PARTITION
+      mesh, apf::GRAPH, apf::PARTITION


Angelyr · 2025-08-15T20:49:03Z

ma/maCollapse.cc

-  if (tryThisDirection(qualityToBeat))
+  if (tryThisDirectionNoCancel(qualityToBeat))
    return true;
-  if ( ! getFlag(adapt,vertToKeep,COLLAPSE))
-    return false;
-  std::swap(vertToKeep,vertToCollapse);
-  computeElementSets();
-  if (!adapt->input->shouldForceAdaptation)
-    qualityToBeat = std::min(adapt->input->goodQuality,
-        std::max(getOldQuality(),adapt->input->validQuality));
+  else destroyNewElements();
+
+  if (getFlag(adapt,vertToKeep,COLLAPSE)) {
+    std::swap(vertToKeep,vertToCollapse);
+    computeElementSets();
+    if (tryThisDirectionNoCancel(qualityToBeat))
+      return true;
+    else destroyNewElements();
+  }

-  return tryThisDirection(qualityToBeat);
+  unmark();
+  return false;


There was a bug in this function that prevented collapsing an edge in either direction. These changes result in both vertices being considered for collapse.

Angelyr · 2025-08-15T20:49:26Z

ma/maCollapse.cc

-  PCU_ALWAYS_ASSERT( ! m->isShared(v[0]));
-  PCU_ALWAYS_ASSERT( ! m->isShared(v[1]));
+  if (!getFlag(a, v[0], DONT_COLLAPSE)) //Allow collapse in one direction
+    PCU_ALWAYS_ASSERT( ! m->isShared(v[0]));
+  if (!getFlag(a, v[1], DONT_COLLAPSE)) //Allow collapse in one direction
+    PCU_ALWAYS_ASSERT( ! m->isShared(v[1]));


Bug that prevented collapsing and edge in one direction in parallel when only the collapsing vertex was owned by the current process

cwsmith · 2025-08-26T14:16:41Z

@Angelyr Would you please move your comments above into the code if they are not already there?

@joshia5 Are you done with your review?

joshia5 · 2025-08-26T15:22:14Z

@cwsmith I'll take a look

joshia5 · 2025-08-28T14:59:18Z

ma/maSize.h

L22,23 since we defined these as input parameters, I'm inclined to have them in maInput unless I can think a good reason why not, tbd

Mark mentioned that we should not be modifying this value at this point, I added a comment about it.

joshia5 · 2025-08-28T15:07:56Z

ma/maCoarsen.cc

@@ -253,4 +257,211 @@ bool coarsen(Adapt* a)
  return true;
 }

+//Measure and edge lenght and stores the result so it doesn't have to be calculated again


we should have been computing edge lengths downstream in markEdgesToCollapse/Split , I was suggesting to add/store the tag where that is done. or are those markedges modified and now calling this new getLength?

The confusion is that there are two separate implementations of coarsen in this file and only one of them uses this getLength function. It would not be useful in the old implementation of coarsen since that implementation only calculates the length once anyway.

ma/maDBG.h

ma/maDBG.cc

Angelyr · 2025-08-30T19:24:53Z

@bobpaw Feel free to take a look at the code now

Nightly Bot and others added 30 commits March 10, 2023 07:16

Merging develop into master

8436909

Merging develop into master

97523f8

Merging develop into master

b2ec4a5

Merging develop into master

b9af579

Merging develop into master

909a234

Merging develop into master

8e7ea44

Merging develop into master

51541e8

Merging develop into master

73e848d

Merging develop into master

4cc2da9

Merging develop into master

d7cc6c8

Merging develop into master

c6fb1d0

Merging develop into master

ea85d83

Merging develop into master

c0c30f4

Merging develop into master

6769092

Merging develop into master

49f143f

Merging develop into master

68861da

Merge branch 'develop'

73d55d1

Merging develop into master

88892c5

Merging develop into master

1ab42dd

Merging develop into master

3d621c8

Merging develop into master

df37b8d

Merging develop into master

cf0a06f

Merging develop into master

274835f

Merging develop into master

8959c59

Merging develop into master

8449430

Merging develop into master

324e9e7

Merge branch 'develop'

61ffc9b

Merge branch 'develop'

c861f9a

Merge branch 'develop'

d79c52c

Merge branch 'develop'

c6254bf

2d get invalid

4163701

joshia5 reviewed Aug 14, 2025

View reviewed changes

Angelyr commented Aug 15, 2025

View reviewed changes

cwsmith added the v4.1.0 changes included in the 4.1.0 release label Aug 26, 2025

cwsmith mentioned this pull request Aug 26, 2025

MeshAdapt: add mesh modification operator counts #481

Open

3 tasks

cwsmith mentioned this pull request Aug 26, 2025

PUMI 4.1.0 Release #515

Open

Angelyr linked an issue Aug 27, 2025 that may be closed by this pull request

ac/adapt-changes dpw6 #513

Open

apply coarsening first, revert changes

41e827e

joshia5 reviewed Aug 28, 2025

View reviewed changes

ma/maDBG.h Show resolved Hide resolved

joshia5 reviewed Aug 28, 2025

View reviewed changes

ma/maDBG.cc Show resolved Hide resolved

moved comments from github

d465479

Angelyr added 2 commits September 3, 2025 14:17

fix tolerance bug

38dcf6e

remove unused

7988f5f

		MatchedSnapper::MatchedSnapper(Adapt* a, Tag* st, bool is)
		MatchedSnapper::MatchedSnapper(Adapt* a, Tag* st)

		if (period < 0.5) //partial range can't be periodic
		return (1-t)a + tb;

		dists.push_back(minDist + 1.0 + tol);
		dists.push_back(newDirection.getLength());

		adaptMesh, apf::GRAPH, apf::PARTITION
		mesh, apf::GRAPH, apf::PARTITION

Coarsen and Snapping #514

Are you sure you want to change the base?

Coarsen and Snapping #514

Uh oh!

Conversation

Angelyr commented Aug 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cwsmith commented Aug 9, 2025

Uh oh!

cwsmith commented Aug 9, 2025

Uh oh!

cwsmith commented Aug 11, 2025

Uh oh!

cwsmith commented Aug 11, 2025

Uh oh!

Angelyr commented Aug 11, 2025

Uh oh!

github-actions bot commented Aug 11, 2025

Uh oh!

Angelyr commented Aug 13, 2025

Uh oh!

cwsmith commented Aug 13, 2025

Uh oh!

Angelyr commented Aug 13, 2025

Uh oh!

cwsmith commented Aug 13, 2025 • 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!

Angelyr left a comment

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!

cwsmith commented Aug 26, 2025

Uh oh!

joshia5 commented Aug 26, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

joshia5 Aug 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Angelyr commented Aug 30, 2025

Uh oh!

Uh oh!

Angelyr commented Aug 6, 2025 •

edited

Loading

cwsmith commented Aug 13, 2025 •

edited

Loading

joshia5 Aug 28, 2025 •

edited

Loading