Adds demo using DMPlex #7

bishtgautam · 2022-11-08T18:24:03Z

Adds a demo that uses TS and DMPlex
Only works serially
The 2D partial dam break problem is implemented
The code can be run via

./ex2 -dm_view  -hu 10.0 -hd 5.0 -save -dt 0.01 \
-ts_max_time 0.01 -ts_monitor -Nx 200 -Ny 200 -save -b -ts_max_time 0.1

and the output will be nearly identical with O(1-e-14) differences between ex1.c that is run via

./ex1 -dt 0.01 -ts_max_time 0.1 -dm_view -save -Lx 200 -Ly 200 -save 1 -ts_monitor -b

This heavily borrows data structure from the TDycore project.

Presently, a DM (of type DMPlex) is created and then information from it is extracted in saved in RDy<*> mesh data structures.

swe_roe/ex2.c

jeff-cohere · 2022-11-09T23:29:34Z

swe_roe/ex2.c

+  ierr = PetscOptionsBegin(user->comm, NULL, "2D Mesh Options", "");
+  PetscCall(ierr);
+  {
+    PetscCall(PetscOptionsInt("-Nx", "Number of cells in X", "", user->Nx, &user->Nx, NULL));


Are we really supposed to call PetscCall for every call to a PETSc function? I wonder if we can configure PETSc to fail on unsuccessful operations instead of having to check ierr everywhere. This isn't an issue with this PR, just something I've been curious about lately. I can look more into this.

I think we might be able to use this error handler to abort when an error is encountered. That would preclude the need for all the error flag checking. Just an idea.

You can use the abort handler, but then you don't get a reliable stack trace. Some day we'll use Rust for systems programming and then it's the ? operator. It's true that stack traces can be obtained using libunwind, but PETSc hasn't gone all-in on that route (it adds complexity to the build and may not work everywhere).

jeff-cohere · 2022-11-09T23:33:38Z

Looks pretty good so far (and I've verified for myself that it builds and runs). I will take a closer look at this tomorrow early in the day if that works for you.

jeff-cohere

Looks pretty good to me. I've left some comments for your consideration, but I don't see anything that should prevent this from going in.

jeff-cohere · 2022-11-10T15:15:23Z

swe_roe/ex2.c

+
+typedef struct {
+  PetscReal X[3];
+} RDyCoordinate;


What about RDyPoint?

Or maybe RDyNode? We should also store other attributes in the Node or Point type, such as boundary code to define boundary and outlet.

I like the idea of storing only the coordinates in a "point" type. We could then have a "node" type that associates other information with a point, embedding the point in the node like this:

typedef struct { RDyPoint point; // (other stuff) } RDyNode;

Then functions that only care about the coordinates could focus on RDyPoint, and anything that needs the additional data could use RDyNode instead.

jeff-cohere · 2022-11-10T15:17:39Z

swe_roe/ex2.c

+  CELL_QUAD_TYPE     /* hexahedron cell for a 3D cell */
+} RDyCellType;
+
+typedef struct {


Several of the comments for fields in this struct are copy-pasted and should probably be made more specific.

jeff-cohere · 2022-11-10T15:22:34Z

swe_roe/ex2.c

+
+  PetscReal *area; /* area of the cell                */
+
+} RDyCell;


What about RDyCells, RDyVertices, and RDyEdges for these three structs, instead of the singular forms? That makes clear that they are arrays of data for several items instead of one. (Not a big deal, but naming is hard, so I think it deserves some thought.)

jeff-cohere · 2022-11-10T15:24:00Z

swe_roe/ex2.c

+  RDyMesh *mesh;
+};
+
+PetscErrorCode RDyInitialize_IntegerArray_1D(PetscInt *array_1D, PetscInt ndim_1, PetscInt init_value) {


I have some thoughts on arrays that we could discuss when we talk about the library itself.

donghuix

It looks great to me. I just have some minor comments.

swe_roe/ex2.c

donghuix · 2022-11-10T03:27:25Z

swe_roe/ex2.c

+    PetscReal xc = cells->centroid[icell].X[1];
+    PetscReal yc = cells->centroid[icell].X[0];
+    if (yc < bu && xc >= bl && xc < br) {
+      b_ptr[icell * 3 + 0] = 1.0;


Since we set the dof=3 to account for 3 unknowns in the equations, the global vector size is number_of_cells * dof. But the boundary code (b_ptr) suppose to have dof=1. It is OK for the demo code, but it will be better in the future if we can allocate b_ptr with the size = number_of_cells. What do you think?

I agree we need a better way to have no-flow boundaries embedded in the domain.

Can we make them as physical curves (Face Sets) in the mesh?

Face set was exactly what I was thinking of using. We had a similar discussion on TDycores-Project/TDycore#225.

In the RDycore lib repo (which we haven't created yet), we would go with TDycores-Project/TDycore#225 (comment) and TDycores-Project/TDycore#225 (comment).

donghuix · 2022-11-10T03:41:13Z

swe_roe/show_results.m

+close all
+
+if ~exist('PetscBinaryRead')
+    error(['Please add PETSc MATLAB files before calling this script via: ' char(10) ...


I am now getting the PETSC_DIR though [~,PETSC_DIR] = system('echo $PETSC_DIR');
Then we can add path to PetscBinaryRead: addpath([PETSC_DIR(1:end-1) '/share/petsc/matlab/']);

@donghuix What do you think about a6887fa#diff-d78a0fbb7b2633122c4842162cbc315ff849e79a59ee8c9d8931aa1ecac04cd3R11-R19?

It looks great to me. Thanks.

donghuix · 2022-11-10T18:35:50Z

swe_roe/ex2.c

+
+typedef struct {
+  PetscReal X[3];
+} RDyCoordinate;


Or maybe RDyNode? We should also store other attributes in the Node or Point type, such as boundary code to define boundary and outlet.

…roblems into bishtgautam/dmplex * 'bishtgautam/dmplex' of https://github.com/RDycore/toy-problems: Minor fixes for PETScv3.18.1

bishtgautam added 7 commits November 8, 2022 11:54

Initial check-in of demo with DMPlex

534d595

Presently, a DM (of type DMPlex) is created and then information from it is extracted in saved in RDy<*> mesh data structures.

Adds code to set IC

bd79386

Corrects the indices of edges and vertices

98a634a

Adds TS solver

746574d

Adds comments

30c3168

Adds code to save cell IDs

9bec22c

Fixes formatting

9d9b794

bishtgautam force-pushed the bishtgautam/dmplex branch from 3215f55 to 9d9b794 Compare November 8, 2022 19:55

bishtgautam added 8 commits November 9, 2022 14:03

Adds the reflecting wall

429a41f

Accounts for edge length and cell area

999b203

Fix formatting

9dd9f78

Changes the output filename

ea8d1f8

Updates the MATLAB script to visualize the output

dfb325f

Generalizes MATLAB plotting script

295ec12

Minor modification to output from ex2

a967f90

Fixes formatting

0d49164

bishtgautam changed the title ~~[WIP] Adds demo using DMPlex~~ Adds demo using DMPlex Nov 9, 2022

bishtgautam requested review from donghuix and jeff-cohere November 9, 2022 22:59

jeff-cohere reviewed Nov 9, 2022

View reviewed changes

swe_roe/ex2.c Outdated Show resolved Hide resolved

jeff-cohere reviewed Nov 9, 2022

View reviewed changes

I think the natural order works now

bcd18eb

jeff-cohere approved these changes Nov 10, 2022

View reviewed changes

Minor fixes for PETScv3.18.1

17d0734

donghuix reviewed Nov 10, 2022

View reviewed changes

knepley and others added 4 commits November 11, 2022 08:56

Merge branch 'bishtgautam/dmplex' of https://github.com/RDycore/toy-p…

806f83a

…roblems into bishtgautam/dmplex * 'bishtgautam/dmplex' of https://github.com/RDycore/toy-problems: Minor fixes for PETScv3.18.1

Do not distribute by default

926b8c0

PetscSF code block

9d2c088

Delete VecViews

107eeee

bishtgautam added 2 commits November 15, 2022 08:45

Address comments during review

a6887fa

Fixes formatting

aa040fb

bishtgautam requested a review from donghuix November 15, 2022 16:55

bishtgautam merged commit 2805d44 into main Nov 15, 2022

bishtgautam deleted the bishtgautam/dmplex branch November 15, 2022 17:12

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Adds demo using DMPlex #7

Adds demo using DMPlex #7

bishtgautam commented Nov 8, 2022 •

edited

Loading

jeff-cohere Nov 9, 2022

jeff-cohere Nov 9, 2022

jedbrown Nov 10, 2022

jeff-cohere commented Nov 9, 2022 •

edited

Loading

jeff-cohere left a comment

jeff-cohere Nov 10, 2022

donghuix Nov 10, 2022

jeff-cohere Nov 14, 2022

jeff-cohere Nov 10, 2022

jeff-cohere Nov 10, 2022 •

edited

Loading

jeff-cohere Nov 10, 2022 •

edited

Loading

donghuix left a comment

donghuix Nov 10, 2022

bishtgautam Nov 15, 2022

jedbrown Nov 15, 2022

bishtgautam Nov 15, 2022

donghuix Nov 10, 2022

bishtgautam Nov 15, 2022

donghuix Nov 15, 2022

donghuix Nov 10, 2022

Adds demo using DMPlex #7

Adds demo using DMPlex #7

Conversation

bishtgautam commented Nov 8, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jeff-cohere commented Nov 9, 2022 • edited Loading

jeff-cohere 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

jeff-cohere Nov 10, 2022 • edited Loading

Choose a reason for hiding this comment

jeff-cohere Nov 10, 2022 • edited Loading

Choose a reason for hiding this comment

donghuix 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

Choose a reason for hiding this comment

bishtgautam commented Nov 8, 2022 •

edited

Loading

jeff-cohere commented Nov 9, 2022 •

edited

Loading

jeff-cohere Nov 10, 2022 •

edited

Loading

jeff-cohere Nov 10, 2022 •

edited

Loading