Hi! Please look at these pictures (they're from Planetside 2)

As you may notice, the glow of the sun and the bloom of the surfaces has this tint that looks to depend on the direction. I'm not terribly sure whats going on.

I added (for teh lulz) a chromatic aberration shader on the bloom pass:

[attachment=28455:bloom.jpg]

But it doesn't quite looks the same. Besides, from the Planetside 2 pics you can see the tinting doesn't always follow the same direction (sometimes its reddish to blueish from right to left, sometimes from top to bottom, and so on).

Looks like the bloom buffer is tapped 2 times with a centered radial UV shift and each tap is tinted with a purple and orange tint and added together.

So the two proposals are:

1. To shift bloom result based on fragment to center vector.
2. To do it in the directional light pass, shifting the lighting itself before the bloom pass.

Both make sense, I'll try with the simplest (1) one and see what I get. Thanks a lot !

EDIT: BTW I'm not terribly sure how would you detect where the ray hits the light disc? Some intersection function?

EDIT: BTW I'm not terribly sure how would you detect where the ray hits the light disc? Some intersection function?

Yes, with an intersection function. A disc is always planar. You have to chose a point on a plane to test against anyway, so you can just do:

t = distFromPlane(discCenter, discNormal, rayOrigin, rayDirection)
hitPos = rayOrigin + rayDirection * t;
if (length(hitPos -  discCenter) >= discRadius)
    discardOrSomething;


distFromPlane(discCenter, discNormal, rayOrigin, rayDirection):
    if (dot(rayDirection, discNormal) >= 0)
        return 0;
    return dot(discCenter - rayOrigin) / dot(rayDirection, discNormal);

The issue is, I'm not terribly sure where to put the disc center. All I got is a light direction for the "sun", not a position. All of this ties in with something else I have yet to implement, rendering the glowy sun disc, so no idea how to do that.

Your chromatic aberration does look very close to the Planetside pics, except, it looks like you're doing something like:
offset = float2(1,1) * pixelSize
uv1 = uv + offset
uv2 = uv - offset

And they're doing something like:
offset = normalize(uv*2-1) * pixelSize
uv1 = uv + offset
uv2 = uv - offset

(btw this code is what I think Ryokeen was suggesting too)