You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
NarrowPhase.AddBatchEntries currently avoids a sweep test if maximum estimated relative motion is less than the speculative margin. This is pretty reasonable and saves a lot of time.
There could be another feature here. Have one condition for choosing whether or not to execute a sweep, and another for the speculative margin actually used to generate contacts.
This would allow an 'early out' margin that is smaller than the final contact generation speculative margin. Potentially useful for cases where you want to be able to detect secondary collisions.
This would likely require expanding the options on ContinuousDetection, which in turn could cause some cache alignment considerations in CollidableDescription. Might be worth doing, maybe? If you did this, it may also be worth biting the bullet on the expanded size and including a cached linear velocity so that the narrow phase can use pair relative linear velocities to calculate a local effective speculative margin rather than just adding the contributing margins. (Testing showed trying to load velocities from the usual otherwise-untouched cachelines was a net loss.)
Pretty low value; very corner case.
The text was updated successfully, but these errors were encountered:
NarrowPhase.AddBatchEntries
currently avoids a sweep test if maximum estimated relative motion is less than the speculative margin. This is pretty reasonable and saves a lot of time.There could be another feature here. Have one condition for choosing whether or not to execute a sweep, and another for the speculative margin actually used to generate contacts.
This would allow an 'early out' margin that is smaller than the final contact generation speculative margin. Potentially useful for cases where you want to be able to detect secondary collisions.
This would likely require expanding the options on
ContinuousDetection
, which in turn could cause some cache alignment considerations inCollidableDescription
. Might be worth doing, maybe? If you did this, it may also be worth biting the bullet on the expanded size and including a cached linear velocity so that the narrow phase can use pair relative linear velocities to calculate a local effective speculative margin rather than just adding the contributing margins. (Testing showed trying to load velocities from the usual otherwise-untouched cachelines was a net loss.)Pretty low value; very corner case.
The text was updated successfully, but these errors were encountered: