precision problems

OpenGL OpenGL: Basic Coding
1

Hello all,

I need to display objects with very big coordinates in my application.
When I try to display these objects, some visualization problems appear. When I move the camera, the geometry coordinates seems to change at each frame like each point is moving randomly.
It seems like precision problems.
I use only double precision values so it is a bit strange.
I use a lot of matrix transformations so maybe the problem comes from matrix operations. Matrix values are also defined with double precision.
Has someone already had this problem?

Thanks a lot for your answers.

Emmanuel Quételard
Igeoss.

2

Double precision is not supported by hardware, so it all gets converted to floats anyway… I don’t have a solution for you though, only some value ideas. I would recalculate the positions relative to the camera on-the-fly so that they fit the single precision when drawing.

3

If you are using matrix inversion, try taking tha nature of the matrices into account.

e.g. the inverse of modelviewmatrix without skewing/scaling:



Modelview = | R3x3 t3x1 |
            |   0    1  |

Inverse   = | R3X3_transpose -R3X3_transpose*t3x1 |
            |        0               1            |

You can make small adjustments for it to support scaling.

You can also write closed form inverse math formulas for the inverse of projection matrices (ortho/perspective). They will most likely be more accurate than the generic operations used to invert an arbitrary 4x4 matrix because projection matrices are sparse.