Gr/wedge improve #180
Merged
Gr/wedge improve #180
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Added Wedge between primal 1s. Also optimized computations for wedge products using for loops.
Added specialized 0-1 primal wedge and optimized preallocations for 1-1 primal wedge.
It should now work with the new Lie Derivative and Interior Product.
Instead of computing a large matrix inverse, inv_hodge_star for dual1 is now computed by backslash operator.
Wedge now returns Vector and not Offset array, which impacted matrix multiplication performance.
These substitutions should ultimately remove the need for dedicated functions to compute those operators. Also added experimental functions to have decapodes generate d0 and d1.
Also added SparseArrays as dependency
Got rid of the generate function equivalents.
New implementation uses the decapodes version of the exterior derivative.
Also added a hodge_diag implementation.
Woe to anyone who debugs this.
Even though this seems to be slower, we can much more easily tell which vertices are edge centers.
These new type annotation fix the return of ANY type before, and the new. Blessed be those who debug this.
We are still missing the Geometric Hodge 1 for 2D and the Diagonal Hodge 0 for 2D could use some more work.
The boundary implemented here is just the adjoint of the exterior derivative.
This could be further extended to work with other in-place binary operations.
Using views seems to improve performance in most cases, but not quite all. For example, views on the points/dual_points in Geo Hdg 1. Also got rid of older prototypes of hodge_diag 1.
Sped up wedge02 and wedge11 preallocations and calculations greatly. Made some modifications to derivbound to remove the recompute check.
commit c204376 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Sun Oct 8 13:10:13 2023 -0400 Fix diff_adv link and some poiseuille params (#150) * Fix diff_adv link and some poiseuille params commit 53b23a5 Merge: abb94f8 fc8c8f8 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Sat Oct 7 21:33:11 2023 -0400 Merge pull request #142 from AlgebraicJulia/llm/ascii commit fc8c8f8 Author: Luke Morris <lukelukemorrismorris@gmail.com> Date: Tue Sep 5 16:19:01 2023 -0400 Add advection Vec equivalent commit abb94f8 Merge: 670a74e cbb91dc Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Sep 1 19:43:07 2023 -0400 Merge pull request #145 from AlgebraicJulia/llm/burger Add Burger's Equation Decapode commit cbb91dc Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Sep 1 17:03:06 2023 -0400 Add Burger's Equation Decapode commit c235acb Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Sep 1 10:10:54 2023 -0400 Fix comment in vec_to_dec commit c972e5a Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Aug 31 09:11:54 2023 -0400 Escape backslash in ascii doc commit be6a614 Author: Luke Morris <lukelukemorrismorris@gmail.com> Date: Mon Aug 28 15:49:43 2023 -0400 Replace vector calculus operators with DEC equivalents commit b437458 Author: Luke Morris <lukelukemorrismorris@gmail.com> Date: Mon Aug 28 12:55:38 2023 -0400 Loop through incident in unicode\! commit 7dadf60 Author: Luke Morris <lukelukemorrismorris@gmail.com> Date: Fri Aug 25 16:59:51 2023 -0400 Add ASCII support for Unicode ops Add doc for ASCII equivalents
Also added ability to set DataType of DEC operators and added special support for inverse hodge 1
The addition to contract operators to only limit contractions to a set of operators, namely DEC matrices, was modified here to fix issues.
Also changed operator indices to use Int32s instead of Int64. This should be changed later to allow user to set their types, with Int32 being default.
Also tuned Geometric Hodge 1 since I haven't found a good way to set a type for the points arrays. Setting the type to a Vector of Point3 of Float64 brings down generation time to third of the version before this.
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## main #180 +/- ##
==========================================
- Coverage 88.85% 81.50% -7.36%
==========================================
Files 4 12 +8
Lines 359 1611 +1252
==========================================
+ Hits 319 1313 +994
- Misses 40 298 +258 ☔ View full report in Codecov by Sentry. |
Also added tests for more operators and fixed inverse Geometric Hodge to be negated. Commented out wedge functions that were never being called.
These tests cover that the operators used by Gensim are all working properly and are well implemented. This also fixed an oversight were Geometric Inverse Hodge 1 was missing a -1.
lukem12345
reviewed
Dec 5, 2023
I suspect I've set them to be too strict so some variances in how the floating point calculations are done may be enough to tip it to fail.
Fixed an error occuring in the docs where the Poiseuille model was not resolving the Laplace de Rham operator due to it not being supported in the 1D case. This worked on main however due to the 2D resolving rules being the default and also covering this case. Also added some changes to the Geo Hodge 1.
Also changed Benchmark to use Ico5
|
Resolves issue #93 by using contract_operators. |
Documentation seemed to make correctly at this version. They should however be updated to use the latest version.
commit 83efc30 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 23:31:09 2024 -0500 Update grigoriev.md commit 53aa53d Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 22:45:29 2024 -0500 Update grigoriev.md commit ef030ff Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 16:06:55 2024 -0500 Clean readme commit cfe25fb Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 14:59:05 2024 -0500 Delete examples/poiseuille directory commit 62806ad Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 14:56:46 2024 -0500 The Klausmeier Model (#152) * Initial Klausmeier example * Remove "" from markdown links * Add newlines between reference bullets * Use raw html for figure * Call hide directive on Decapode table * Use 600 simulation time * using ComponentArrays with DiagEq * Use property accessor commit 8e9a050 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 14:11:24 2024 -0500 Remove "stable" docs temporarily commit 4cafa46 Merge: 3395d01 f83b374 Author: quffaro <cuffaro.matt@gmail.com> Date: Fri Jan 12 12:47:25 2024 -0500 Merge pull request #183 from AlgebraicJulia/migrate-diagrammatic-equations Migrate diagrammatic equations commit f83b374 Author: Matt <cuffaro.matt@gmail.com> Date: Fri Jan 12 11:02:07 2024 -0500 migrated test/composition.jl to DE#main commit 7e4552e Author: Matt Cuffaro <cuffaro.matt@gmail.com> Date: Thu Jan 11 14:28:37 2024 -0500 docs build v2 commit 7163881 Author: Matt Cuffaro <cuffaro.matt@gmail.com> Date: Thu Jan 11 13:50:01 2024 -0500 docs build commit 6dda9b1 Author: Matt <cuffaro.matt@gmail.com> Date: Tue Jan 9 13:56:38 2024 -0500 addressed comments commit 22de7db Author: Matt <cuffaro.matt@gmail.com> Date: Tue Jan 9 08:19:41 2024 -0500 now reliant on DiagrammaticEquations@v0.1.0 commit 4db11fa Author: Matt <cuffaro.matt@gmail.com> Date: Fri Jan 5 15:15:50 2024 -0500 tests passing on dev copy of DiaEq commit 41d8b9b Merge: 57cf75e 3395d01 Author: Matt <cuffaro.matt@gmail.com> Date: Fri Jan 5 14:41:28 2024 -0500 resolved merge conflicts commit 57cf75e Author: Matt <cuffaro.matt@gmail.com> Date: Tue Dec 19 16:41:24 2023 -0500 tests pass with dev'd de.jl commit c9f3a2c Author: Matt Cuffaro <cuffaro.matt@gmail.com> Date: Tue Dec 19 13:56:51 2023 -0500 working through tests commit 46c615a Author: Matt <cuffaro.matt@gmail.com> Date: Tue Dec 19 11:13:41 2023 -0500 deleted files commit fc35a87 Author: Matt Cuffaro <cuffaro.matt@gmail.com> Date: Mon Nov 20 14:42:01 2023 -0500 first pass at decapodes dependencing on DEq commit a25f9c6 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Oct 20 10:26:18 2023 -0400 Internalize collate and compile Collage commit fa8130d Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Oct 12 15:30:53 2023 -0400 Test naming convention commit 3fe8ca4 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Oct 12 15:24:52 2023 -0400 Update tangent pointers in collage commit 7ef1c77 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Oct 12 14:54:34 2023 -0400 Demonstrate collage boundaries in a simulation commit decf186 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Oct 12 13:33:00 2023 -0400 Use Op2 to encode boundaries commit afe414d Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Mon Oct 9 12:10:20 2023 -0400 Add restriction morphisms as Op1s
lukem12345
approved these changes
Feb 2, 2024
lukem12345
added a commit
that referenced
this pull request
Feb 3, 2024
Some old commit messages: * started new branch for resource sharing * continuing work on this feature * working on resource loading code * adding buildkite to decapodes * removed file not part of this feature * testing buildkite with echo hello * fixed jobscript.sh * changed jobscript for multithreading * changed jobscript for multithreading, 64 cores * 32 cores instead * pmapping files in docs/make.jl * 32 cores * Initial upload of CISM notebook. Need to convert to .md format and verify results consistent with cise.ufl.edu/~luke.morris/cism.html * Gr/wedge improve (#180) * General wedge improvements Added Wedge between primal 1s. Also optimized computations for wedge products using for loops. * Further improvements to wedge Added specialized 0-1 primal wedge and optimized preallocations for 1-1 primal wedge. * Moved DEC into operators.jl * Added support for Lie Derivative and Inner Product * Fixed type_inference It should now work with the new Lie Derivative and Interior Product. * Changed inv_hodge Instead of computing a large matrix inverse, inv_hodge_star for dual1 is now computed by backslash operator. * Fixed wedge product Wedge now returns Vector and not Offset array, which impacted matrix multiplication performance. * Added substitutions for 2D Lie and InterProd These substitutions should ultimately remove the need for dedicated functions to compute those operators. Also added experimental functions to have decapodes generate d0 and d1. * Improved d2 operator Also added SparseArrays as dependency * Added 1D Lie and Interior Sub Rules Got rid of the generate function equivalents. * Added improved lapace deRham New implementation uses the decapodes version of the exterior derivative. * Further improved wedge Also added a hodge_diag implementation. * Greatly improved the performance of hodge_diag1 Woe to anyone who debugs this. * Attempt to use indexing trick Even though this seems to be slower, we can much more easily tell which vertices are edge centers. * Fixed typing of hodge_diag 1 These new type annotation fix the return of ANY type before, and the new. Blessed be those who debug this. * Added support for new hodges We are still missing the Geometric Hodge 1 for 2D and the Diagonal Hodge 0 for 2D could use some more work. * Added boundary and some GeoInvHdg1 The boundary implemented here is just the adjoint of the exterior derivative. * Added support for in-place wedge This could be further extended to work with other in-place binary operations. * Improved Hodge_Diag 0 * Added views and more typing Using views seems to improve performance in most cases, but not quite all. For example, views on the points/dual_points in Geo Hdg 1. Also got rid of older prototypes of hodge_diag 1. * Further improvements to wedge Sped up wedge02 and wedge11 preallocations and calculations greatly. Made some modifications to derivbound to remove the recompute check. * Squashed commit of the following: commit c204376 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Sun Oct 8 13:10:13 2023 -0400 Fix diff_adv link and some poiseuille params (#150) * Fix diff_adv link and some poiseuille params commit 53b23a5 Merge: abb94f8 fc8c8f8 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Sat Oct 7 21:33:11 2023 -0400 Merge pull request #142 from AlgebraicJulia/llm/ascii commit fc8c8f8 Author: Luke Morris <lukelukemorrismorris@gmail.com> Date: Tue Sep 5 16:19:01 2023 -0400 Add advection Vec equivalent commit abb94f8 Merge: 670a74e cbb91dc Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Sep 1 19:43:07 2023 -0400 Merge pull request #145 from AlgebraicJulia/llm/burger Add Burger's Equation Decapode commit cbb91dc Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Sep 1 17:03:06 2023 -0400 Add Burger's Equation Decapode commit c235acb Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Sep 1 10:10:54 2023 -0400 Fix comment in vec_to_dec commit c972e5a Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Aug 31 09:11:54 2023 -0400 Escape backslash in ascii doc commit be6a614 Author: Luke Morris <lukelukemorrismorris@gmail.com> Date: Mon Aug 28 15:49:43 2023 -0400 Replace vector calculus operators with DEC equivalents commit b437458 Author: Luke Morris <lukelukemorrismorris@gmail.com> Date: Mon Aug 28 12:55:38 2023 -0400 Loop through incident in unicode\! commit 7dadf60 Author: Luke Morris <lukelukemorrismorris@gmail.com> Date: Fri Aug 25 16:59:51 2023 -0400 Add ASCII support for Unicode ops Add doc for ASCII equivalents * Added contract operators support * Merge branch 'main' into gr/wedge_improve * Fixed errors caused by rebasing * Got rid of 1D/2D dec_generate * Added contract operator support Also added ability to set DataType of DEC operators and added special support for inverse hodge 1 * Fixed issues with Contract Operators The addition to contract operators to only limit contractions to a set of operators, namely DEC matrices, was modified here to fix issues. * Added float types for Wedge Also changed operator indices to use Int32s instead of Int64. This should be changed later to allow user to set their types, with Int32 being default. * Added tests for 1D ops and more open_operators Also tuned Geometric Hodge 1 since I haven't found a good way to set a type for the points arrays. Setting the type to a Vector of Point3 of Float64 brings down generation time to third of the version before this. * Added some tests for gensim * Added tests for operator contraction in Gensim Also added tests for more operators and fixed inverse Geometric Hodge to be negated. Commented out wedge functions that were never being called. * Fixed Geometric Hodge 1 tests * Changed gitignore to match main * Added more tests These tests cover that the operators used by Gensim are all working properly and are well implemented. This also fixed an oversight were Geometric Inverse Hodge 1 was missing a -1. * Tests covering some more functionality * Loosened tolerances on operator tests I suspect I've set them to be too strict so some variances in how the floating point calculations are done may be enough to tip it to fail. * Removed show in operator tests * Restored V_1 in operator tests * Added the Benchmark file Fixed an error occuring in the docs where the Poiseuille model was not resolving the Laplace de Rham operator due to it not being supported in the 1D case. This worked on main however due to the 2D resolving rules being the default and also covering this case. Also added some changes to the Geo Hodge 1. * Added Sparse/Static Arrays to compat Also changed Benchmark to use Ico5 * Fixed CairoMake to 0.10 for Docs Documentation seemed to make correctly at this version. They should however be updated to use the latest version. * Squashed commit of the following: commit 83efc30 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 23:31:09 2024 -0500 Update grigoriev.md commit 53aa53d Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 22:45:29 2024 -0500 Update grigoriev.md commit ef030ff Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 16:06:55 2024 -0500 Clean readme commit cfe25fb Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 14:59:05 2024 -0500 Delete examples/poiseuille directory commit 62806ad Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 14:56:46 2024 -0500 The Klausmeier Model (#152) * Initial Klausmeier example * Remove "" from markdown links * Add newlines between reference bullets * Use raw html for figure * Call hide directive on Decapode table * Use 600 simulation time * using ComponentArrays with DiagEq * Use property accessor commit 8e9a050 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Jan 19 14:11:24 2024 -0500 Remove "stable" docs temporarily commit 4cafa46 Merge: 3395d01 f83b374 Author: quffaro <cuffaro.matt@gmail.com> Date: Fri Jan 12 12:47:25 2024 -0500 Merge pull request #183 from AlgebraicJulia/migrate-diagrammatic-equations Migrate diagrammatic equations commit f83b374 Author: Matt <cuffaro.matt@gmail.com> Date: Fri Jan 12 11:02:07 2024 -0500 migrated test/composition.jl to DE#main commit 7e4552e Author: Matt Cuffaro <cuffaro.matt@gmail.com> Date: Thu Jan 11 14:28:37 2024 -0500 docs build v2 commit 7163881 Author: Matt Cuffaro <cuffaro.matt@gmail.com> Date: Thu Jan 11 13:50:01 2024 -0500 docs build commit 6dda9b1 Author: Matt <cuffaro.matt@gmail.com> Date: Tue Jan 9 13:56:38 2024 -0500 addressed comments commit 22de7db Author: Matt <cuffaro.matt@gmail.com> Date: Tue Jan 9 08:19:41 2024 -0500 now reliant on DiagrammaticEquations@v0.1.0 commit 4db11fa Author: Matt <cuffaro.matt@gmail.com> Date: Fri Jan 5 15:15:50 2024 -0500 tests passing on dev copy of DiaEq commit 41d8b9b Merge: 57cf75e 3395d01 Author: Matt <cuffaro.matt@gmail.com> Date: Fri Jan 5 14:41:28 2024 -0500 resolved merge conflicts commit 57cf75e Author: Matt <cuffaro.matt@gmail.com> Date: Tue Dec 19 16:41:24 2023 -0500 tests pass with dev'd de.jl commit c9f3a2c Author: Matt Cuffaro <cuffaro.matt@gmail.com> Date: Tue Dec 19 13:56:51 2023 -0500 working through tests commit 46c615a Author: Matt <cuffaro.matt@gmail.com> Date: Tue Dec 19 11:13:41 2023 -0500 deleted files commit fc35a87 Author: Matt Cuffaro <cuffaro.matt@gmail.com> Date: Mon Nov 20 14:42:01 2023 -0500 first pass at decapodes dependencing on DEq commit a25f9c6 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Fri Oct 20 10:26:18 2023 -0400 Internalize collate and compile Collage commit fa8130d Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Oct 12 15:30:53 2023 -0400 Test naming convention commit 3fe8ca4 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Oct 12 15:24:52 2023 -0400 Update tangent pointers in collage commit 7ef1c77 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Oct 12 14:54:34 2023 -0400 Demonstrate collage boundaries in a simulation commit decf186 Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Thu Oct 12 13:33:00 2023 -0400 Use Op2 to encode boundaries commit afe414d Author: Luke Morris <70283489+lukem12345@users.noreply.github.com> Date: Mon Oct 9 12:10:20 2023 -0400 Add restriction morphisms as Op1s * Removed code shifted to other packages * Delete meshes.jl include, add operators include * Fix compat between Decapodes and CombiSpaces with @GeorgeR227 * Undo stub code deletion --------- Co-authored-by: Luke Morris <lukelukemorrismorris@gmail.com> * Set version to 0.5.1 * Some fixes for Decapodes Docs (#196) * Some fixes for Decapodes Docs Was still us SphericalMesh/Gridmeshes and packages were outdated. Add compats for various packages. * Moved Point3D to shallow_ice * finishing this feature * Rename ipynb to qmd * Use quarto convert, tweak links, and editorialize * Rename qmd md, add to make * Fix imports, define ^ * Remove docs/Manifest.toml from repo 7163881 * Initial upload of CISM notebook. Need to convert to .md format and verify results consistent with cise.ufl.edu/~luke.morris/cism.html * Rename ipynb to qmd * Use quarto convert, tweak links, and editorialize * Rename qmd md, add to make * Fix imports, define ^ * Up comp at on DiagEqs --------- Co-authored-by: AlgebraicJulia Bot <129184742+algebraicjuliabot@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This pull request will added improved versions of common operators in the DEC. This includes operators such the Exterior Derivative, Boundary Operator, Dual Derivative, both Diagonal/Geometric Hodges and their inverses and Wedge Products. Any more complicated operators, such as the Laplace de Rham, Co-Differential, Interior Product and Lie Derivative are dealt with by exposing their interiors to the compiler. This is done in open_operators.
Due to the slow downs caused by the open_operators, since it blows single operators apart into multiple operators that all have to be computed with, this pull request also implements operator contraction support, the opposite of the expand_operators. The contract operators here is fairly simple in that it can only contract operators in a straight line in the flow graph but this can be replaced later with a more robust contraction system.
Various other optimizations to the compiler are added such as switching the negation function calls to multiplication by -1, which saves allocations, as well as the ability to convert operators to use user specified float types, normally Float64 or Float32. The operators have also been changed to use Int32 indices instead of Int64 as in the mesh ACSet.