[FEATURE] Speedup nonconvex collision detection (up to x30).#2818
Conversation
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bf8fc89131
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
bf8fc89 to
2043896
Compare
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2043896af8
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| if diag_a_sq > diag_b_sq: | ||
| i_ga, i_gb = i_gb, i_ga |
There was a problem hiding this comment.
Preserve type order before smooth contact refinement
The AABB-size swap here can undo the canonical type ordering established just above, so a smooth primitive (sphere/ellipsoid/capsule) may move to i_gb when it is larger than a nonconvex mesh. In this kernel we still call func_apply_smooth_refinement later, and the surrounding comment explicitly documents that running refinement on the mesh-as-A orientation sends smooth data through the type_b path with inverted correction. That means large-sphere vs small-mesh contacts can get wrong contact positions after this change, which can destabilize resting contacts and drift.
Useful? React with 👍 / 👎.
No description provided.