"Stretching" shadow z coordinate to make the most of non-linear depth buffer?

[JoshKlint]

Non-linear depth textures provide no advantage to shadowmaps. They provide greater precision closer to the light source, but lower precision far away from the source.

To get around this, I am thinking about bringing the z coordinate of my shadow vertices closer to the camera. The goal is to get an even distribution of precision across the entire shadowmap, without manually writing to gl_FragDepth, so we can retain early Z discard.

[mhagain]

For cubemap shadowmaps I just don't bother with depth; instead I store a linear distance using a color buffer only (and using a minimum blend operation to simulate "depth") which works fine and makes the biasing easier when it comes to comparison time.