Assert with PX_ENABLE_ASSERTS and ABP_SIMD_OVERLAP

[Sknowman]

Im hitting an assert in

void BoxManager::prepareData(RadixSortBuffered& /rs/, ABP_Object* PX_RESTRICT objects, PxU32 objectsCapacity, ABP_MM& memoryManager, PxU64 contextID)

file BpBroadPhaseABP.cpp
line 1473
at PX_ASSERT(boxesYZ[i].mMinY==-aabb.mMinY);
debug build
physX 5.4.0
https://github.com/NVIDIA-Omniverse/PhysX/blob/44667a2576f8145bb972c81cab255824455fc5c7/physx/source/lowlevelaabb/src/BpBroadPhaseABP.cpp#L1473

in my case, the computed bounds mMinY value seems to be off by a significant amount, about 6.0 units (scenes scale would be considered big by normal standards I think, but was never a problem in previous versions of physx).

If I go back up the stack prepareData() is called on the kinematic list of boxes from void ABP::Region_prepareOverlaps() line 3414
https://github.com/NVIDIA-Omniverse/PhysX/blob/44667a2576f8145bb972c81cab255824455fc5c7/physx/source/lowlevelaabb/src/BpBroadPhaseABP.cpp#L3414
on the 'mKBM' member.

It happens when first load a scene of static actors, I then add a bunch of kinematic actors. Then shortly after I change the scene by removing all the static actors, and replace them with new ones, upon removing all the static actors the scene contains original kinematics (possibly dynamics but i dont think so) and new statics, on the very next series of calls to .........

// <-- before simulate(), remove static actors here, kinematics remain -->
scene->simulate(step); // completes fine
scene->fetchResults(true); // <-- assertion here
// <-- scene with only kinematics here, I dont get this far

..... I get the assertion. This was never an issue before physx 5.4.0ish that I know of, maybe I didnt notice, maybe its a 5.0.0+ thing and this isnt supported anymore but I didnt get anything like this in 4.0-.

Ive looked through the migration guide, Im not sure what the process for updating a scene is in 5.0