Re: [Flightgear-devel] New texture draw method

[Julian F. <jul...@ne...>]

Norman Vine wrote:
> 
> To exploit this means we need to determine the furthest
> point in each tile from the tile's center point and then
> scale it to be 2**23

No.  You are thinking of integer (or maybe fixed-point) arithmetic.  We are using floating-point arithmetic, which does the scaling automatically so that there are 23 bits of precision whatever the magnitude of numbers.  That's what the 8-bit exponent field does.

Anyway, as I see it the point is that in the final conversion from a floating-point screen position (of a vertex) to an integer screen position, whatever the precision, there can be an off-by-one rounding error if the floating-point number is not IDENTICAL.  (The precision determines how often this occurs.  With 23-bit precision it happens often enough for many people to notice it quite easily.)

If the graphics card is doing "anti-aliasing" then it does not do that final conversion to an integer pixel position.  Instead it calculates a partial intensity, either directly from the floating-point position or by sub-sampling the pixel (drawing the polygon onto a 2x2 or 3x3 or 4x4 array of sub-pixels and then averaging them).

Maybe this is why some people see the aliasing artifact and others don't.  On my GeForce 256 DDR/DVI at 1152x864 resolution I had not noticed the artifacts but when I took a look just now, I could see the white dotted lines flickering occasionally along the edges of runways and along tile boundaries.  I can choose among:

- Allow applications to control the antialiasing mode
- Manually select the antialiasing mode: Off (no antialiasing)
- Manually select the antialiasing mode: 2x
- Manually select the antialiasing mode: 4x

and I can see the difference.  If I look at a tile boundary with anti-aliasing off, the artifact looks something like the picture that someone posted (http://gwmostreetnet.kircpe.cableone.net/tmp/fgfs-screen-001.png), flickering because it changes completely from frame to frame, sometimes with more or fewer dots, often disappearing entirely for a fraction of a second.  With "2x" (which I guess means 2x2 sub-sampling) I get fewer dots and with "4x", even fewer.  But some are still there.  It does not smooth them out in the way that I was hoping it would.

- Julian