target map comes from a depth map of a previous stage.
It is called "dual" because it only uses 2 accumalative passes (2 stages @ 1 pass).

store a gradient or gamma multiplier in the alpha channel;
this effects all pixels in the view rather than the rgb values that are only multiplied by the lit pixels.

For aliasing, expect a version 2.0 soon with a linear search solution for both the loss of z-buffer precision and limitation of pixel space.

Also,
I sort of released this code just to "prove" that I am working and not dead.
I think this is my first contribution. I'll update it with something more special soon enough (Slin). But I agree, I think i made a nice start but I will work on the quality. Thanks for the inspiration.

Please post a working project.

Thanks for this!

The problem you get with your method is that when you sample the shadowmap near the (0,0) coordinates you will get less smoothiness then at the point where you sample the shadowmap near the (1,1) coordinates.
if you want to make the blur independent of the viewing distance you should do something like this:
float4(IN.Depth.x + 0.001/viewDistance,IN.Depth.y + 0.001/viewDistance,IN.Depth.z,IN.Depth.w)

Thanks Chris B,
this was one of my first shadow mappers and your absolutley right.

I fixed whis quite a while ago, sorry i didn't reply faster.
-Mike-