Use flying edges for mesh generation #1741

perminder-17 · 2024-10-17T20:53:38Z

Resolves issue : #901
Developer Certificate of Origin
Version 1.1

Copyright (C) 2004, 2006 The Linux Foundation and its contributors.
1 Letterman Drive
Suite D4700
San Francisco, CA, 94129

Everyone is permitted to copy and distribute verbatim copies of this
license document, but changing it is not allowed.

Developer's Certificate of Origin 1.1

By making a contribution to this project, I certify that:

(a) The contribution was created in whole or in part by me and I
have the right to submit it under the open source license
indicated in the file; or

(b) The contribution is based upon previous work that, to the best
of my knowledge, is covered under an appropriate open source
license and I have the right under that license to submit that
work with modifications, whether created in whole or in part
by me, under the same open source license (unless I am
permitted to submit under a different license), as indicated
in the file; or

(c) The contribution was provided directly to me by some other
person who certified (a), (b) or (c) and I have not modified
it.

(d) I understand and agree that this project and the contribution
are public and that a record of the contribution (including all
personal information I submit with it, including my sign-off) is
maintained indefinitely and may be redistributed consistent with
this project or the open source license(s) involved.

Signed-off-by: Perminder <[email protected]>

avogadro/core/cube.cpp

avogadro/qtgui/meshgenerator.cpp

Signed-off-by: Perminder <[email protected]>

avogadro/core/cube.cpp

avogadro/qtgui/meshgenerator.cpp

github-actions · 2024-11-02T00:26:27Z

Here are the build results
macOS.dmg
Artifacts will only be retained for 90 days.

Signed-off-by: Perminder <[email protected]>

perminder-17 · 2024-11-03T02:03:51Z

avogadro/qtgui/meshgenerator.cpp

+            // Add triangles
+            const char* caseTri = caseTriangles[caseId]; // size 16
+            for(int idx = 0; caseTri[idx] != -1; idx += 3)
+            {
+
+
+                size_t globalIdx = globalIdxs[caseTri[idx]];
+                size_t globalIdx1 = globalIdxs[caseTri[idx + 1]];
+                size_t globalIdx2 = globalIdxs[caseTri[idx + 2]];
+
+
+
+                m_vertices.push_back(points[globalIdx]);
+                m_vertices.push_back(points[globalIdx1]);
+                m_vertices.push_back(points[globalIdx2]);
+
+                m_normals.push_back(normals[globalIdx]);
+                m_normals.push_back(normals[globalIdx1]);
+                m_normals.push_back(normals[globalIdx2]);    
+


Hi @cryos and @ghutchis , does this part looks okay to you guys. I think I have done all things right but my output comes out to be something like. Any suggestions? Do you think I am pushing for the vertices and normals in a right way?

perminder-17 · 2024-11-03T02:25:15Z

I can do cleanups in the code once I start getting the expected output.

Also tagging @matterhorn103 and @aerkiaga for their inputs.

Let me know if you guys have any feedbacks/suggestions?

avogadro/core/cube.cpp

avogadro/qtgui/meshgenerator.cpp

Signed-off-by: Perminder <[email protected]>

avogadro/core/cube.cpp

avogadro/core/mesh.h

avogadro/qtgui/meshgenerator.cpp

avogadro/qtplugins/meshes/meshes.cpp

matterhorn103 · 2024-11-06T18:24:43Z

@perminder-17 I'm afraid rendering is something I know absolutely nothing about

perminder-17 · 2024-11-06T18:26:11Z

@perminder-17 I'm afraid rendering is something I know absolutely nothing about

Aah... Thanks for your reply. No problem, I figured that out. It was like, we need to pass the triangles as well.

ghutchis · 2024-11-06T18:59:52Z

Great, I'll start a review tomorrow

Signed-off-by: Perminder <[email protected]>

avogadro/qtgui/meshgenerator.cpp

+    unsigned char edgeCaseZ = edgeCases[edgeCaseIdxZ];
+
+    // If the sum is odd, the edge along the z-axis is cut
+    if ((edgeCase + edgeCaseZ) % 2 == 1)


github-actions · 2024-11-07T00:27:00Z

Here are the build results
macOS.dmg
Artifacts will only be retained for 90 days.

avo-bot · 2024-11-08T02:20:43Z

This pull request has been mentioned on Avogadro Discussion. There might be relevant details there:

https://discuss.avogadro.cc/t/november-2024-live-updates/6446/1

Signed-off-by: Perminder <[email protected]>

avogadro/core/cube.cpp

… variables Signed-off-by: Perminder <[email protected]>

Signed-off-by: Perminder <[email protected]>

perminder-17 · 2024-11-10T01:59:24Z

Hi everyone, Here's the results after the implementation of flying-edges. requesting @ghutchis for the review, still needs some cleanups in the code, removal of unnecessary codeblocks and variables.

Marching cube takes 1.5 minutes to render the mesh

Flying edges takes only 3 seconds to render it. (could vary with different systems)

Also it produces a little smoother mesh (left->flying edges, right->marching cubes)

After Laplacian smoothing we have:

ghutchis · 2024-11-12T19:31:34Z

I'm happy to help do some performance tests - I'm not seeing the "Time taken" code at the moment in your pull request.

I have a range of molecule sizes and resolutions to give this a good benchmark.

Doing a rough stopwatch count, I'm seeing ~2-2.5x performance improvement, but that's including the whole thing, not just mesh generation.

I can say that with caffeine as an example, the flying edges version appears instantly on my MacBook even at "Very High" resolution. With a larger molecule, it's going from ~10s to display the VdW mesh to ~4.5s.

If you want to send me the patch you used for timing, I'd be happy to perform some real benchmarks tonight in release mode.

perminder-17 · 2024-11-12T21:12:59Z

All I see is debug mode in what you posted, if that is the case my point stands - benchmarks in debug mode are not very useful and we know the STL etc is very slow in debug. Comparing release numbers (which I also expect to be good) would be more informative.

Thanks @cryos for your insights. Really helps a lot and also thanks @ghutchis for the help in performance testing.

If I could say what code I used for checking the time, I used the chrono library https://en.cppreference.com/w/cpp/chrono/high_resolution_clock/now

If you include header #include<chrono> and replace your run() function with this, you can see the time taken.

#include <chrono> // at the top (header file)



// Rest code remains the same


// Replace the run function to test
 
void MeshGenerator::run() // you can directly copy the run() function and paste it with the original ones
{

  if (!m_cube || !m_mesh) {
    qDebug() << "No mesh or cube set - nothing to find isosurface of…";
    return;
  }
  m_mesh->setStable(false);
  m_mesh->clear();

  auto start = std::chrono::high_resolution_clock::now();
  FlyingEdgesAlgorithmPass1();
  FlyingEdgesAlgorithmPass2();
  FlyingEdgesAlgorithmPass3();
  FlyingEdgesAlgorithmPass4();

  auto end = std::chrono::high_resolution_clock::now();
  std::chrono::duration<double> elapsed = end - start;
  qDebug() << "Flying Edges Time: " << elapsed.count() << " s";    

  m_mesh->setVertices(points);
  m_mesh->setNormals(normals);
  m_mesh->setTriangles(tris);
  m_mesh->smooth(m_passes);
  m_mesh->setStable(true);

  points.resize(0);
  normals.resize(0);
  tris.resize(0);

}

also, if you want to test and compare the time with marching cubes.

here how you can check the time taken for marching cubes as well.

OLD CODE

#include <chrono> // similarly include this

// just a little changes in the run function 

void MeshGenerator::run()
{

// rest code remains the same.

 auto start = std::chrono::high_resolution_clock::now(); // add this

  for (int i = 0; i < m_dim.x() - 1; ++i) {
    for (int j = 0; j < m_dim.y() - 1; ++j) {
      for (int k = 0; k < m_dim.z() - 1; ++k) {
        marchingCube(Vector3i(i, j, k));
      }
    }
    if (m_vertices.capacity() < m_vertices.size() + m_dim.y() * m_dim.x() * 3) {
      m_vertices.reserve(m_vertices.capacity() * 2);
      m_normals.reserve(m_normals.capacity() * 2);
    }
    emit progressValueChanged(i);
  }

// add this
  auto end = std::chrono::high_resolution_clock::now();
  std::chrono::duration<double> elapsed = end - start;
  qDebug() << "Marching Cube Timing: " << elapsed.count() << " s";    

// rest code remains the same
}

perminder-17 · 2024-11-12T21:23:54Z

With a larger molecule, it's going from ~10s to display the VdW mesh to ~4.5s.

Also, after the merge of this, I can handle the merge conflicts (if any) in this PR #1692. If we add a progressive refinement, it won't take even 4.5 seconds to render and will generate the mesh in no time.

Do let me know if everything looks fine in the results, I could do many cleanups in the code (added too many debugging peices, unnecessary functions, variables haha). Still needs some formatting and cleanups.

Signed-off-by: Perminder <[email protected]>

perminder-17 · 2024-11-13T00:36:51Z

Btw @ghutchis I just noticed one bug (when you click on the calculate button twice for the same mesh generation it takes more time). It's because we were not clearing the memory. In the above commit I have fixed that as well. Sorry for the oversight.

ghutchis · 2024-11-13T01:52:48Z

I'm seeing ~2x - 2.2x speedup for mesh generation on a range of 7 molecules with 0.05Å resolution. It depends a lot on the size of the molecule and the shape (e.g., smaller molecules and more spherical / cubic see the greatest speedup).

I'll do a bit of profiling, but of course this is also the single-core implementation (vs. parallel)

perminder-17 · 2024-11-13T02:07:36Z

Correct! I looked at the results mentioned on the paper. They also have 2-2.2x speedups.

I'll do some cleanups in the code tomorrow.

ghutchis · 2024-11-13T20:15:37Z

I also checked the number of vertices for marching cubes vs. flying edges. It's getting exactly 6.0 times fewer vertices, so the mesh is much smaller at the same resolution (and thus faster to render, less memory, etc.)

Even on a really big DNA chunk, I'm seeing ~0.1s for mesh creation with the "automatic" resolution. (And yes, the progressive refinement could be a help here too.)

Signed-off-by: Perminder <[email protected]>

avogadro/core/cube.cpp

+
+float Cube::getData(int i, int j, int k) const
+{
+    int nx = m_points.x();


github-actions · 2024-11-14T14:00:20Z

Here are the build results
Avogadro2.AppImage
Ubuntu-Latest.tar.gz
macOS.dmg
Artifacts will only be retained for 90 days.

ghutchis

Mostly some small things with the comments and documentation consistency.

ghutchis · 2024-11-14T15:45:15Z

avogadro/core/cube.h

+  std::array<float, 3> computeGradient(int i, int j, int k) const;
+
+  /**
+   * Get the values of the eight corners of a cube cell.


I'm not quite sure I understand from the docs here. I've got some index (i, j, k) into the cube. What's the "cell" of the cube? Is it just going to get me the corners of the entire cube? If so, why do I need i, j, k parameters?

avogadro/core/cube.h

ghutchis · 2024-11-14T15:46:24Z

avogadro/core/cube.h

+  float getData(int i, int j, int k) const;
+
+  /**
+   * Get the positions of the eight corners of a cube cell.


Again, what's the "cell" here? Are these the min, max of the corners of the cube? If so, why do you need the i, j, k?

avogadro/core/mesh.h

ghutchis · 2024-11-14T15:47:37Z

avogadro/core/mesh.h

@@ -228,6 +232,7 @@ class AVOGADROCORE_EXPORT Mesh
  Core::Array<Vector3f> m_vertices;
  Core::Array<Vector3f> m_normals;
  Core::Array<Color3f> m_colors;
+  Core::Array<Vector3f> tris;


Please use m_triangles to be consistent with the other variables.

avogadro/qtgui/meshgenerator.h

ghutchis · 2024-11-14T15:55:43Z

avogadro/qtgui/meshgenerator.h

@@ -107,34 +173,44 @@ class AVOGADROQTGUI_EXPORT MeshGenerator : public QThread
  int progressMaximum() { return m_progmax; }

 signals:
+
  /**
   * The current value of the calculation's progress.
   */
  void progressValueChanged(int);

 protected:


Can we keep this? Sometimes it's helpful to get normals for arbitrary points.

ghutchis · 2024-11-14T15:55:53Z

avogadro/qtgui/meshgenerator.h

-   */
-  Vector3f normal(const Vector3f& pos);
+
+  unsigned long duplicate(const Vector3i& c, const Vector3f& pos);


Needs documentation.

ghutchis · 2024-11-14T15:57:22Z

avogadro/qtgui/meshgenerator.h

+  std::vector<unsigned char> cubeCases; //size ((m_dim.x() - 1) * (m_dim.y() - 1) * (m_dim.z() - 1))
+  std::vector<int> triCounter;  // size ((m_dim.y() - 1) * (m_dim.z() - 1))
+  std::vector<unsigned char> edgeCases; // size ((m_dim.x() - 1) * (m_dim.y()) * (m_dim.z()))
+  Core::Array<Vector3f> tris; // triangles of a mesh 


Please remove any un-used variables (from Marching Cubes)

Please also use m_triangles (or similar) to keep consistent with member variables through the rest of the code.

ghutchis · 2024-11-14T15:58:24Z

avogadro/qtgui/meshgenerator.h

+  static const unsigned char numTris[256]; 
+  static const bool isCut[256][12];  
+  static const char caseTriangles[256][16];
+  static const unsigned char edgeVertices[12][2];     


The static analysis says these class members are never used?
If they're used, please keep to m_edgeVertices etc. for consistency.

Signed-off-by: Perminder <[email protected]>

github-actions · 2024-11-15T08:27:13Z

Here are the build results
Avogadro2.AppImage
Ubuntu-Latest.tar.gz
macOS.dmg
Artifacts will only be retained for 90 days.

github-actions · 2024-11-15T14:28:43Z

Here are the build results
Avogadro2.AppImage
Ubuntu-Latest.tar.gz
macOS.dmg
Artifacts will only be retained for 90 days.

ghutchis · 2024-11-15T23:26:05Z

avogadro/core/cube.cpp

+
+  std::array<std::array<float, 3>, 8> pos;
+
+  float xpos = m_min.x() + (i * m_spacing.x());


I'm sorry - I still don't understand what this method is supposed to do.

xpos, ypos, zpos are the real-space versions of i, j, k indices
.. so the method gives me a cube that's one step in each of the x, y, z directions?

If so, shouldn't that be the doc string?

ghutchis · 2024-11-15T23:26:47Z

avogadro/core/cube.h

@@ -23,9 +24,7 @@ class Mutex;
 /**
 * @class Cube cube.h <avogadro/core/cube.h>
 * @brief Provide a data structure for regularly spaced 3D grids.
- * @author Marcus D. Hanwell


@author Marcus D. Hanwell @author Perminder Singh

ghutchis · 2024-11-16T00:18:09Z

avogadro/core/mesh.h

@@ -138,14 +142,6 @@ class AVOGADROCORE_EXPORT Mesh
   */
  const Core::Array<Vector3f>& normals() const;

-  /**


I still don't understand why you're removing numNormals() which seems like a useful method.

But i was unable to find where we were using numnormals() in our actual code?

We're not, currently. But since this is a core class, it's also API for future development.

ghutchis

Mostly resolved - please note the other comments, thanks.

Signed-off-by: Perminder <[email protected]>

github-actions · 2024-11-16T12:59:42Z

Here are the build results
Ubuntu-Latest.tar.gz
Avogadro2.AppImage
Artifacts will only be retained for 90 days.

github-actions · 2024-11-16T14:04:48Z

Here are the build results
Ubuntu-Latest.tar.gz
Avogadro2.AppImage
Artifacts will only be retained for 90 days.

Signed-off-by: Geoff Hutchison <[email protected]>

ghutchis

Sorry, I missed these last time. I think this fixes the compile on Windows.

ghutchis · 2024-11-16T17:16:26Z

avogadro/qtgui/meshgenerator.cpp

+          ge2.xstart += isCutCase[4];
+          ge2.ystart += isCutCase[7];
+        }
+        if(isXEnd and isYEnd)


&& not and - this is the build failure on Windows

ghutchis · 2024-11-16T17:16:45Z

avogadro/qtgui/meshgenerator.cpp

+        {
+          ge1.zstart += isCutCase[11];
+        }
+        if(isXEnd and isZEnd)


Same - use &&

ghutchis · 2024-11-16T17:17:00Z

avogadro/qtgui/meshgenerator.cpp

+        {
+          ge2.ystart += isCutCase[5];
+        }
+        if(isYEnd and isZEnd)


ghutchis · 2024-11-16T17:17:41Z

avogadro/qtgui/meshgenerator.cpp

+            if(isCutCase[11])
+            {
+                int idx = ge1.zstart + z1counter;
+                if(isXEnd and isYEnd)


&& not and

ghutchis · 2024-11-16T17:17:59Z

avogadro/qtgui/meshgenerator.cpp

+            if(isCutCase[5])
+            {
+                int idx = ge2.ystart + y2counter;
+                if(isXEnd and isZEnd)


&& not and

ghutchis · 2024-11-16T17:18:16Z

avogadro/qtgui/meshgenerator.cpp

+            if(isCutCase[6])
+            {
+                int idx = ge3.xstart + x3counter;
+                if(isYEnd and isZEnd)


&& not and

Signed-off-by: Perminder Singh <[email protected]>

perminder-17 · 2024-11-16T20:05:26Z

Done @ghutchis

github-actions · 2024-11-16T21:22:33Z

Here are the build results
Ubuntu-Latest.tar.gz
Avogadro2.AppImage
macOS.dmg
Win64.exe
Artifacts will only be retained for 90 days.

github-actions · 2024-11-16T23:39:15Z

Here are the build results
Avogadro2.AppImage
Ubuntu-Latest.tar.gz
Win64.exe
Artifacts will only be retained for 90 days.

adding-new-tables-for-flying-edges

69f9bee

Signed-off-by: Perminder <[email protected]>

ghutchis linked an issue Oct 17, 2024 that may be closed by this pull request

Consider switching to flying edges from parallel marching cubes #901

Closed

perminder-17 added 2 commits October 27, 2024 15:28

adding codes for flying-edges algo, still fixing left

4b9474b

Signed-off-by: Perminder <[email protected]>

some more fixes required

0e9ec3b

Signed-off-by: Perminder <[email protected]>

github-advanced-security bot found potential problems Nov 1, 2024

View reviewed changes

minor fixes

973fe51

Signed-off-by: Perminder <[email protected]>

github-advanced-security bot found potential problems Nov 1, 2024

View reviewed changes

some fixes

bcb02c4

Signed-off-by: Perminder <[email protected]>

perminder-17 commented Nov 3, 2024

View reviewed changes

github-advanced-security bot found potential problems Nov 3, 2024

View reviewed changes

avogadro/core/cube.cpp Fixed Show fixed Hide fixed

avogadro/qtgui/meshgenerator.cpp Fixed Show fixed Hide fixed

avogadro/qtgui/meshgenerator.cpp Fixed Show fixed Hide fixed

avogadro/qtgui/meshgenerator.cpp Fixed Show fixed Hide fixed

avogadro/qtgui/meshgenerator.cpp Fixed Show fixed Hide fixed

working

762e943

Signed-off-by: Perminder <[email protected]>

github-advanced-security bot found potential problems Nov 6, 2024

View reviewed changes

perminder-17 added 2 commits November 7, 2024 04:00

minor cleanusp and fixes

febd8e7

Signed-off-by: Perminder <[email protected]>

remove helper comments

46ca336

Signed-off-by: Perminder <[email protected]>

github-advanced-security bot found potential problems Nov 6, 2024

View reviewed changes

ghutchis changed the title ~~WIP : Consider switching to flying edges from parallel marching cubes~~ Use flying edges for mesh generation Nov 8, 2024

fixing laplace smoothing, only cleanups left

f37cd36

Signed-off-by: Perminder <[email protected]>

github-advanced-security bot found potential problems Nov 10, 2024

View reviewed changes

avogadro/core/cube.cpp Fixed Show fixed Hide fixed

avogadro/core/cube.cpp Fixed Show fixed Hide fixed

perminder-17 added 3 commits November 10, 2024 07:04

cleanups in the code done, still needs to remove unused functions and…

6507df7

… variables Signed-off-by: Perminder <[email protected]>

minor fixes

63324fe

Signed-off-by: Perminder <[email protected]>

minor cleaning

6afb548

Signed-off-by: Perminder <[email protected]>

perminder-17 marked this pull request as ready for review November 10, 2024 01:54

minor-fixes-on-calculating-the-mesh-again

8de667a

Signed-off-by: Perminder <[email protected]>

minor cleanups

f7bbfc8

Signed-off-by: Perminder <[email protected]>

github-advanced-security bot found potential problems Nov 14, 2024

View reviewed changes

avogadro/core/cube.cpp

float Cube::getData(int i, int j, int k) const

{

int nx = m_points.x();

Check notice

Code scanning / CodeQL

Unused local variable Note

Variable nx is not used.

ghutchis requested changes Nov 14, 2024

View reviewed changes

perminder-17 added 2 commits November 15, 2024 10:37

fixes on suggestions

0a60847

Signed-off-by: Perminder <[email protected]>

fixing

38dcb07

Signed-off-by: Perminder <[email protected]>

ghutchis reviewed Nov 15, 2024

View reviewed changes

ghutchis reviewed Nov 16, 2024

View reviewed changes

fixes

77d870e

Signed-off-by: Perminder <[email protected]>

Merge branch 'master' into flying-edges

e06bbc8

Signed-off-by: Geoff Hutchison <[email protected]>

ghutchis requested changes Nov 16, 2024

View reviewed changes

fixes and->&&

b33f001

Signed-off-by: Perminder Singh <[email protected]>

ghutchis merged commit 141745d into OpenChemistry:master Nov 16, 2024
22 of 23 checks passed


		std::array<std::array<float, 3>, 8> pos;

		float xpos = m_min.x() + (i * m_spacing.x());

Use flying edges for mesh generation #1741

Use flying edges for mesh generation #1741

Conversation

perminder-17 commented Oct 17, 2024

github-actions bot commented Nov 2, 2024

perminder-17 Nov 3, 2024 • edited Loading

Choose a reason for hiding this comment

perminder-17 commented Nov 3, 2024 • edited Loading

matterhorn103 commented Nov 6, 2024

perminder-17 commented Nov 6, 2024

ghutchis commented Nov 6, 2024

github-actions bot commented Nov 7, 2024

avo-bot commented Nov 8, 2024

perminder-17 commented Nov 10, 2024 • edited Loading

Marching cube takes 1.5 minutes to render the mesh

Flying edges takes only 3 seconds to render it. (could vary with different systems)

Also it produces a little smoother mesh (left->flying edges, right->marching cubes)

After Laplacian smoothing we have:

ghutchis commented Nov 12, 2024

perminder-17 commented Nov 12, 2024 • edited Loading

OLD CODE

perminder-17 commented Nov 12, 2024

perminder-17 commented Nov 13, 2024

ghutchis commented Nov 13, 2024

perminder-17 commented Nov 13, 2024 • edited Loading

ghutchis commented Nov 13, 2024

github-actions bot commented Nov 14, 2024

ghutchis left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

github-actions bot commented Nov 15, 2024

github-actions bot commented Nov 15, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ghutchis left a comment

Choose a reason for hiding this comment

github-actions bot commented Nov 16, 2024

github-actions bot commented Nov 16, 2024

ghutchis left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

perminder-17 commented Nov 16, 2024

github-actions bot commented Nov 16, 2024

github-actions bot commented Nov 16, 2024

perminder-17 Nov 3, 2024 •

edited

Loading

perminder-17 commented Nov 3, 2024 •

edited

Loading

perminder-17 commented Nov 10, 2024 •

edited

Loading

perminder-17 commented Nov 12, 2024 •

edited

Loading

perminder-17 commented Nov 13, 2024 •

edited

Loading