Precision issue in index computation in Gu::HeightField::computeCellCoordinates (PxReal to PxU32)

[Nodli]

Library and Version

PhysX v5.1.0
Release 104.0

Issue

Hi,

There is a precision issue in the computation of the vertex index in Gu::HeightField::computeCellCoordinates.

PhysX/physx/source/geomutils/src/hf/GuHeightField.cpp

Line 729 in 8c2791a

const PxU32 vertexIndex = PxU32(PxU32(x) * (mData.nbColumns) + PxU32(z));

mData.nbColumns is of type PxReal so the index computation is done in PxReal and only then cast to PxU32. This leads to problems when x, z and nbColumns have large values (~4k x 4k).
A suggested fix is to either use mData.columns like the line below or cast mData.nbColumns directly using PxU32(mData.nbColumns).

A pull request for the same issue was submitted for PhysX 4.1 ( NVIDIAGameWorks/PhysX#574 ) but the modifications that were applied for PhysX 5.1 (the final cast to PxU32) do not fix the problem.

[PierreTerdiman]

I changed these float members back to integers, so the issue should be fixed for good now. Should appear in a future release.

[Nodli]

Thank you, i'm closing this issue.