add vcpkg.json - #1
Open
MuGdxy wants to merge 2 commits into
Open
Conversation
haoxiangNtu
added a commit
to haoxiangNtu/Stiff-GIPC
that referenced
this pull request
May 6, 2026
…hybrid) Two related patches for collision-detection performance when many ABD bodies have all collisions excluded (e.g., non-gripper arm parts vs cloth): KemengHuang#2: per-body isolation flag via collision_skip_matrix diagonal - sim_engine: detect bodies with ground_skip AND all-other-bodies-excluded; set collision_skip_matrix[i][i]=1 as a "fully isolated" marker. - mlbvh kernels: at leaf-bbox build, isolated body's faces/edges get an empty bbox -> BVH overlap test always false -> traversal naturally prunes the entire isolated subtree. Self-query kernels also early-return at thread entry on the same flag. #3: indirect BVH input filter (active-subset Construct + adjusted self-query) - mlbvh: new _calcLeafBvs_indirect / _calcLeafBvs_ccd_indirect kernels write only n_active leaves; lbvh_f/_e Construct + ConstructFullCCD take the indirect path when _active_idx is set. - lbvh_e EE-self-query (which uses idx + N - 1 to locate self leaf) must use n_active as launch count, not edge_number -- otherwise threads read leaves at the default-path offset where indirect BVH didn't write. - sim_engine: build active face/edge index lists (drop primitives whose every vertex belongs to an isolated body), upload, hand off to bvh_f/_e. - Wiring deferred until after ipc.init() so bboxDiagSize2/dHat capture the full-scene BVH; second buildBVH() then transitions BVH state to indirect before buildCP runs. Measured on case_27 mobile s1_hybrid (semi-implicit, dt=0.020, 33/42 bodies fully isolated, 5 paired runs each, std<1.1ms): baseline (no exclusion API call): 368 ms median +KemengHuang#1 only (exclusion matrix used): 369 ms (no traversal effect, expected) +KemengHuang#2 (diag iso + empty bbox): 44 ms (8.45x, 5/5 runs +88%) +KemengHuang#2+#3 (indirect BVH): 41 ms (1.07x more, 5/5 runs +3-9%) -------------------------------------------- total: 9.02x median Toggles for safe rollback: BVHSKIP2=0 disable KemengHuang#2 (auto-disables #3) BVHSKIP3=0 disable #3 only default: both enabled Affects only the isolated-body fast path; case_26 (no exclusion calls) shows no measurable change (regression-tested 1091 frames, no INTERSECT). Bench reproducer in examples/bench_s1_hybrid.py. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
haoxiangNtu
added a commit
to haoxiangNtu/Stiff-GIPC
that referenced
this pull request
May 6, 2026
…trix FEM body vertices carry body_id == -1 in _bodyID by legacy semantic of tetrahedra_obj::begin_load_body. As a result, when the user calls add_collision_exclusion(abd_body, fem_global_id), the exclusion was silently ignored everywhere a kernel called _is_collision_excluded: - bodyA == -1 || bodyB == -1 -> early return false -> not excluded. This silent failure was masked in baseline because IPC barrier still recorded the contact pair from BVH detection -> no penetration occurred, and the user's exclusion declaration never had a visible effect. Once BVH-skip KemengHuang#2 (collision_skip_matrix diagonal isolation) takes the declaration at face value and removes the body's leaves from BVH, the contact pair stops being recorded, and: - IPC barrier no longer pushes the bodies apart -> penetration - _edgeTriIntersectionQuery (sanity check) sees the actual geometric crossing, prints [INTERSECT-L] and forces line search to backtrack repeatedly -> simulation stalls. Map FEM body id -1 to the last matrix row (matrix slot N-1) so the exclusion declaration is honored uniformly by: - _selfQuery_vf / _selfQuery_ee narrow-phase filters - _edgeTriIntersectionQuery sanity check - any other consumer of _is_collision_excluded Caveat: assumes a single FEM body. Multiple FEM bodies will share slot N-1 and exclusions cannot distinguish them. This also changes the semantics of KemengHuang#1 (collision matrix alone, no KemengHuang#2/#3): exclusion declarations now actually filter contact pairs. Bench: case_27 hybrid KemengHuang#1-only (no KemengHuang#2/#3) goes 369ms -> ~365ms (1% noise) since the dropped pairs are a tiny fraction of total IPC work; the saving is essentially zero relative to BVH traversal cost. Co-Authored-By: Claude Opus 4.7 (1M context) <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.
Add vcpkg.json, users don't need to manually call
vcpkg install xxx