PlanarTriangulation: reusable ISweepLineCache for the sweep-line triangulation - #6542
Draft
Grantim wants to merge 6 commits into
Draft
PlanarTriangulation: reusable ISweepLineCache for the sweep-line triangulation#6542Grantim wants to merge 6 commits into
Grantim wants to merge 6 commits into
Conversation
…ngulation SweepLineQueue becomes reusable behind a new ISweepLineCache interface with a makeSweepLineCache() factory: init() resets a run while every internal buffer, including the topology (via new MeshTopology::clear()) and the projected points, keeps its capacity. triangulateDisjointContours and fillContours2DPlan accept an optional cache, and new triangulateDisjointContoursTopology builds the patch connectivity for a HoleFillPlan inside the cache without materializing a Mesh at all. HoleFillPlanner passes an optional cache through, and getPlanarHoleFillPlans injects per-worker caches persisting across calls. Plans stay bit-identical; plan-batch microbenchmarks improve by 10-30%, and MeshBoolean with cMinSweptHoleSize lowered to 4 (routing every hole through the swept-line path) improves by about 5%. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Replace the function-static per-worker caches in getPlanarHoleFillPlans with a cache member created on the first swept run, like optimalStepsCache_: ownership is local to the planner, which is already thread-confined, instead of thread-keyed static state. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… being the cache All reusable buffers move into the nested SweepLineQueue::Cache (the only ISweepLineCache implementation), and the queue references one given to its constructor: the queue is the algorithm, the cache is the storage. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Per-call ETS planners reuse a member cache only 1-3 times while destroying every per-thread cache serially at the join, which measured slower than master's per-hole temporaries (MeshBoolean with cMinSweptHoleSize=3: 41 vs 36 ms median; without the member: parity). The cache remains an opt-in fillContours2DPlan parameter for callers that keep one across many calls. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This reverts commit 84e30ef: the removal was measured on cuts with ~30 holes, where per-call cache provisioning had nothing to amortize against. On cuts that flood holes (MeshInspector's Boolean Benchmark spheres produce ~9.7k holes per getPlanarHoleFillPlans call, ~500 per worker) the per-worker reuse is deep and the cache wins 7-8% of the whole boolean with cMinSweptHoleSize lowered to 3, while staying neutral at the shipping threshold. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The ETS destructor frees every worker's planner (grown sweep-line cache and triangulation maps) serially on the calling thread, which dominates small batches: MRTest MeshBoolean with cMinSweptHoleSize=3 improves 41 -> 38 ms (master 36), while deep-batch cuts keep their win (Boolean Benchmark spheres: 17.8 vs master 19.5 ms median). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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.
SweepLineQueuebecomes reusable behind a newISweepLineCacheinterface with amakeSweepLineCache()factory:init()resets a run while every internal buffer, including the topology (via newMeshTopology::clear()) and the projected points, keeps its capacity. The queue itself is now a cheap per-run object working on aSweepLineQueue::Cache(the only implementation of the interface) that holds all the reusable buffers.triangulateDisjointContoursandfillContours2DPlanaccept an optional cache as a defaulted last parameter (ISweepLineCache* cache = nullptr), so existing callers and bound signatures gain the parameter without new overloads.ISweepLineCacheis abstract (pure virtual dtor): instances come only from the factory.triangulateDisjointContoursTopologybuilds the patch connectivity for aHoleFillPlaninside the cache without materializing aMeshat all.HoleFillPlannerowns a cache member created on the first swept run (like itsoptimalStepsCache_), so each worker ofgetPlanarHoleFillPlansreuses the buffers across all the holes it plans within a call; the per-worker planners are then released in a parallel pass, since freeing them serially in the ETS destructor dominates small batches.Plans stay bit-identical (FNV hash over plan items, verified against master). Measured with
cMinSweptHoleSizelowered to 3 so that every hole goes through the swept path: on MeshInspector's Boolean Benchmark workload (3366-vert spheres, DifferenceAB, ~9.7k per-face holes per cut, ~500 per worker) the whole boolean improves by 7-8% (median 19.5 -> 17.8 ms, interleaved rounds); on MRTest MeshBoolean (~30 holes per cut, 1-3 per worker, so nothing to amortize the cache against) it costs about 5% (38 vs 36 ms). At the shipping threshold both are at parity, and a serial 30-hole plan batch with a persistent cache goes 0.229 -> 0.184 ms.🤖 Generated with Claude Code