MultifrontalSolver #2327
Merged
MultifrontalSolver #2327
Conversation
This file contains hidden or 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
feat(linear): Finalize MultifrontalSolver implementation and clean up feat(linear): Finalize MultifrontalSolver with benchmarks (Phase 5) feat(linear): Implement in-place back-substitution (Phase 4) feat(linear): Implement imperative multifrontal elimination (Phase 3) feat(linear): Implement iterative value loading (Phase 2) test(linear): Add balanced_smoother test case to MultifrontalSolver tests feat(linear): Implement MultifrontalSolver structure and initialization (Phase 1)
Tighten Clique API Clique compilation unit calculateSeparatorKeys Lazy separatorKeys print methods Make Clique more classy Add streaming
Contributor
There was a problem hiding this comment.
Pull request overview
This PR introduces a new imperative-style multifrontal solver (MultifrontalSolver) that optimizes Gaussian factor graph elimination by pre-allocating memory and reusing the structure across iterations. The implementation achieves approximately 4x speedup over the standard GTSAM elimination approach for iterative optimization scenarios.
Key changes:
- New
MultifrontalSolverclass that pre-builds the symbolic junction tree and pre-allocates all necessary matrices - New
MultifrontalCliqueclass for imperative clique operations (eliminate, solve) - Extended
VectorValueswith a newinsertmethod for concatenated vectors with explicit key ordering - Added
blockViewmethod toSymmetricBlockMatrixfor non-copying block access
Reviewed changes
Copilot reviewed 9 out of 9 changed files in this pull request and generated 23 comments.
Show a summary per file
| File | Description |
|---|---|
gtsam/linear/MultifrontalSolver.h |
Header defining the main MultifrontalSolver class with methods for construction, loading, elimination, and solving |
gtsam/linear/MultifrontalSolver.cpp |
Implementation of MultifrontalSolver including symbolic tree construction and memory pre-allocation |
gtsam/linear/MultifrontalClique.h |
Header for the imperative clique structure with methods for matrix operations and back-substitution |
gtsam/linear/MultifrontalClique.cpp |
Implementation of clique operations including fillAb, eliminateClique, and solveClique |
gtsam/linear/VectorValues.h |
Added new insert method signature for concatenated vectors |
gtsam/linear/VectorValues.cpp |
Implementation of the new insert method that segments a vector by keys and dimensions |
gtsam/base/SymmetricBlockMatrix.h |
Added blockView method for efficient non-copying block access |
gtsam/linear/tests/testMultifrontalSolver.cpp |
Unit tests covering constructor, load, eliminate, and solve operations |
timing/timeMultifrontalSolver.cpp |
Benchmark comparing MultifrontalSolver performance against standard GTSAM elimination |
Member
Author
|
@copilot open a new pull request to apply changes based on the comments in this thread |
Contributor
Co-authored-by: dellaert <10515273+dellaert@users.noreply.github.com>
- Revert rightHandSideScratch_ back to rhsScratch_ - Change separatorSolutionScratch_ to separatorScratch_ - Remove currentClique reference, use clique-> directly Co-authored-by: dellaert <10515273+dellaert@users.noreply.github.com>
Address code review comments: documentation and naming conventions
ProfFan
approved these changes
Dec 30, 2025
| } | ||
| } | ||
|
|
||
| void MultifrontalClique::eliminate() { |
Member
Author
|
I’ll implement both TODOs in follow-up PR |
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.
In order to speed up optimization, I created a fully imperative multifrontal solver that nevertheless is in the style of GTSAM. However, it creates the structure and allocates memory only once, and then can be re-used in subsequent iterations. For a simple chain with 500 nodes, with metis ordering (tree of$\log_2(500)$ deep) it is between 3.5 and 4 times as fast:
Update
After storing pointers into the vectorValues data structure, we are now 5-6 times as fast (on Mac):
Also, I realized GTSAM is using TBB, whereas new solver is single-threaded for now.