Improve performance - #51
Merged
Merged
Conversation
Cache dense log magnitudes and sweep the grid for heuristic covers. Preserve traversal order while storing boost entries contiguously and deriving balance counts from the grid. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Build Woodbury grids directly from `A` and construct edge-list supports only for the other solver paths. Count support first to select the solver. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Assemble the sparse Woodbury correction in reusable CSC storage and multiply its symmetric upper triangle directly. Separate Boolean pattern scans from floating-point loops. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Use `Int32` labels for dense boost entries when dimensions fit, and form already-bounded bucket indices without clamping. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Use grid support counts to avoid a second traversal when support is full; retain component traversal otherwise. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Deficits passed to `bucketof` are already in range, so compute the index without clamping. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Certify returned hard covers in `iscover` arithmetic by uniformly inflating factors for the worst rounding shortfall. Leave initializer starts unchanged. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Record each continuation stage's exit reason and final decrease; warn when it reaches `maxiter` while still descending. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Build working-precision schedules ending at `1e8`: eight stages for exact solvers and four for `:lsqr`. An explicit `κs` overrides the default. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Start LSQR continuation from the heuristic cover; exact solvers retain the unweighted start. An empty schedule still returns the unweighted fit. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Cache one log magnitude per support entry across heuristic sweeps while preserving traversal order and results. The dense-grid path is unchanged. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Match `SparseMatrixCSC` symmetry partners with per-column cursors in O(nnz), preserving generic-check behavior for zeros and roundoff. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Right-precondition `Float64` LSQR with a reusable sparse Cholesky factor of the weighted normal matrix. A ridge handles singular support graphs. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Flatten support into compact row, column, and log-magnitude arrays, using a SIMD logarithm approximation. Preserve traversal order and certify returned covers in linear arithmetic. Assisted-by: Claude Fable 5 (claude-fable-5) <noreply@anthropic.com>
Use CHOLMOD's symbolic fill estimate to cap Cholesky preconditioner storage at `fillbudget`. Fall back to diagonal preconditioning and report the selected regime in the statistics. Assisted-by: Claude Opus (claude-opus-5) <noreply@anthropic.com>
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #51 +/- ##
==========================================
+ Coverage 98.39% 98.92% +0.52%
==========================================
Files 13 15 +2
Lines 2937 3643 +706
==========================================
+ Hits 2890 3604 +714
+ Misses 47 39 -8 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
These 15 commits improve runtime performance over all major code paths. In some cases they adjust the cap on iterations upward, but that was found to be justified by the goal of having a consistent tolerance. Preconditioning improvements often compensate by reducing the number of iterations to below what they were before this PR series began.