Hello,
while continuing my work on my deferred shader engine I descided to implement a shadowmapping feature.

Here is the process i do for each light that casts shadows:
1. render the scene from light point of view to create depth-shadowmap
2. store modelview and projection matrix
3. draw light to lightmap using g-buffer and shadowmap
3.1 build transformation matrix that goes from cameraspace to light-view-clipping space by using this matrix:
transformation-matrix = light_projection_matrix * light_modelview_matrix * camera_modelview_matrix_inverse
3.2 In the shader: read fragment position from gbuffer, calculate shadowmap_texture_coordinate with:
shadowmap_texture_coordinate = transformation-matrix * gbuffer_pos
3.3 do the is-in-shadow check stuff

I have several problems:
1. The shadowmap_texture_coordinate does not seem to have an orthographic projection. I fixed this by applying this to the shadowmap_texture_coordinate (sc):
sc.xy = sc.xy / sc.z
This is not correct in any way I think but I have no idea what else could cause the projection-bug.

2. When I read from the shadowmap I get values in [0,1]. How do I transform these to compare them with the distance of a fragment from the light-position since those are given in engine-units.

I hope someone will understand what I am doing sometimes I'm not sure my self

Wow thanks that worked!
Now one step further: Is there a more efficient way of rendering a cubemap than rendering all 6 directions seperately ? I found something that uses a geometry shader to render to all 6 faces of the cube at one time but that looks to complicated for me.

ok, I guess I will be fine with my cube map for now ... with a little bit of optimization it will only draw visible objects for each side of the cube and I share one FBO for the shadow map. Thanks for your help!