TGeoParallelWorld improvements #23
Merged
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
This should allow to speed up the safety calculation in the parallel world (by pruning away uncessary safety calls which are anyway beyond the limit).
Optimization of check if 2 geometry paths are the same, through
(a) simple comparison of the current level
(b) start node comparison at the leafs instead the root,
because this will enhance early returns
Bug fix in Safety calculation: Actually call ::Safety on the TGeoShape **not** on the TGeoNode. Otherwise the coordinate transformation is done twice which leads to wrong safety values. Consistency fix in ::FindNextBoundary: Apply the `IsMatchinsState` check also in the case of non-voxelised navigation to be consistent with the voxelized treatment
This commit provides an (optional) optimization to reduce the cost of safety evaluations associated with TGeoParallelWorld. The approach consists in the introduction of additional cache-state in TGeoNavigator. This state remembers the last location for which TGeoParallelWorld::Safety was evaluated, together with the corresponding value returned. Since TGeoNavigator follows the evolution of tracks step-by-step, we have some sort of history-locality. This means that the cache-state has a high-probability of being relevant for multiple track locations in a row. The optimization proposed here is orthogonal to other improvements. It adds on top of algorithmic improvements within TGeoParallelWorld (bounding volume hierarchies, see separate PR), gaining a few extra percent in speed. By default, the new feature is not enabled by default for backward compatibility. Once fully verified, we can make it the default mode. That said, in ALICE simulations, the caching did not modify stepping behaviour and yielded identical results compared to not doing the caching.
…utines This commit provides a rewrite of key functions of TGeoParallelWorld, achieving: (a) faster initialization (b) faster execution of the Safety function (from ~O(N) to ~O(1)) (c) similar execution of FindNode/FindBoundary functions (~log(N)) (d) less memory consumption (better memory scalability) The development for this commit was motivated from a use case in ALICE, in which the parallel world "scene" can be very large (~100K volumes). In this case, TGeoVoxelFinder takes very long to construct and consumes a very large amount of memory (GBs). In addition, the evaluation of the Safety function dominates the Geant simulation time. The improvements in this commit are mainly achieved through: * The use of a boundary volume hierarchy (BVH) as the base acceleration entity, replacing TGeoVoxelFinder. BVH are the standard in industry/computer-graphics, for what concerns ray-object intersection tasks. The BVH is constructed from axis-aligned bounding boxes and employed in the FindBoundary/FindNode implementations * The use of a 3D voxel grid (TGeoVoxelGrid) structure, able to store properties "local" or in the vicinity of a cartesian coordinate P. This structure allows to reduce the (typical) algorithmic complexity for "Safety" queries to ~O(1) (with a constant factor determined by the voxel size). Filling of the 3D voxel grid cache for Safety is done on-the-fly (using the BVH once). Ideas for these improvements come from prior work in related libraries such as VecGeom. ----- Implementation details: * The implementation is, for now (until fully tested), provided in a backward compatible manner: - By default, nothing changes - Users have to activate the BVH mode via TGeoParallelWorld::SetAccelerationMode - Users may hence compare the 2 modes or choose the best depending on complexity and needs * Functions for Safety, FindNode, FindBoundary dispatch to some internal implementation. This causes an extra lookup/jump, which can be removed once BVH is fully validated * For the BVH, a well known open source implementation is included in header-only form. The headers are copied from https://github.com/madmann91/bvh commit 66e445b92f68801a6dd8ef943fe3038976ecb4ff. Some minor patching has been done in order to achieve compilation for C++17 (see README). * A new class, TGeoVoxelGrid is provided for the cartesian VoxelGrid container. ---- Performance examples: In a test with the ALICE simulation framework including the ITS + TPC detectors with 48240 volumes on the parallel world, we see * initialization time goes from TGeoVoxelFinder: 10s ---> BVH: 40ms * Geant simulation time: 10s --> 2s * memory usage: 3GB --> 1GB Hence, this PR will make a big difference for the ALICE simulation program. It was verified, that identical results (number of hits, steps, etc) are obtained when going from TGeoVoxelFinder --> BVH+GRID. ---- Outlook: Similar techniques could be applied to ordinary TGeoNavigator routines.
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.
cherry-picking some recent TGeo improvements for immediate use in ALICE.
relevant PRs are here
root-project#16442
root-project#16470
root-project#16425