Per-step allocation and inner-loop cleanups in the step pipeline - #989
Closed
marknefedov wants to merge 1 commit into
Closed
Per-step allocation and inner-loop cleanups in the step pipeline#989marknefedov wants to merge 1 commit into
marknefedov wants to merge 1 commit into
Conversation
- carry_warmstart_data: hoist the manifold-level subshape_pos1 lookup out of the per-point matching loop (it is invariant there, and always None for cluster manifolds). - Narrow phase: keep the serial pair-transition buffer as a persistent NarrowPhase scratch field (mem::take + restore) so its capacity is reused across steps instead of reallocating each step. - Substep: replace the joint island event drain().collect() with the take + restore idiom already used for multibody chain events, removing a per-substep Vec allocation. - Staged island solver: resize the generic multibody solver velocity vectors in place and zero them instead of allocating two fresh DVectors every step. rapier3d test suite passes.
Author
|
Closing this one: the measured effect on the benchmark scenes was within noise, and I'd rather not ask for review time on changes I can't demonstrate. The benchmarks in #988 stand on their own; if a scene shows these allocations mattering I'll reopen with numbers. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Four changes removing per-step allocator traffic and one loop-invariant lookup:
carry_warmstart_data: the manifold-levelsubshape_pos1()lookup moved out of the per-point matching loop. It is invariant there and alwaysNonefor cluster manifolds, but was re-checked for every candidate point of an O(prev_points x target_points) loop.NarrowPhasescratch field (take and restore), so its capacity survives across steps. Same idiom the function already uses forupdate_candidatesandsolver_graph_dirty.drain().collect()replaced with the take-and-restore idiom used three lines below for multibody chain events. Removes a per-substep Vec.DVector::zerosevery step. Reallocation only happens when the total DoF count grows.On the benchmark scenes from #988 the effect was within noise (-1.6% on the pyramid at p = 0.05, wider on the terrain scene). Withdrawn for lack of demonstrable numbers.