Can we reduce polyline smears on steep slopes?

[likangning93]

Polylines on terrain smear on steep slopes, since it's similar to a shadow volume:

var viewer = new Cesium.Viewer('cesiumContainer');

var positions = Cesium.Cartesian3.fromDegreesArray([
    -122.17580380403314, 46.19984918190237,
    -122.17671567204553, 46.196577275143326,
    -122.17190167749244, 46.19583668275485
]);

viewer.entities.add({
    polyline : {
        positions : positions,
        width : 32,
        material : new Cesium.PolylineGlowMaterialProperty({
            glowPower : 0.2,
            color : Cesium.Color.BLUE
        }),
        clampToGround : true
    }
});

viewer.camera.lookAt(positions[0], new Cesium.Cartesian3(0.0, 0.0, 5000.0));
viewer.camera.lookAtTransform(Cesium.Matrix4.IDENTITY);

(Note that this won't work until #6689 is merged)

We compute local terrain slopes for materials on ground primitives using the globe depth texture (this is something like screen-space partial derivatives), so I wonder if we could use these to help reduce smear.

Currently the algorithm for determining if a fragment is part of the polyline checks the distance between the fragment's position (derived from globe depth texture) and a plane, but if we had a local terrain slope we could do a ray-plane check instead of point-plane distance.

In addition to effectiveness we'd also want to evaluate the performance hit for this. Since it's much less noticeable for narrower lines, maybe it can be a toggleable option if it's too slow.

See https://github.com/AnalyticalGraphicsInc/cesium/blob/master/Source/Shaders/ShadowVolumeAppearanceFS.glsl#L28

[likangning93]

Hm. On a second reading, turns out "Deferred Vector Map Visualization" does this already. Yay! Not such unknown waters after all.

[EDIT] actually, it's a newer version of the paper