But I can’t make PCSS look the same as they look in nvidia sample, so any suggestions and corrections are welcome.
To make things a bit different from jhonny.goransson’s shadow maps example I use view-space z value for shadow map and deferred shading (to keep shading in a separate pass).
Just added a branch with perspective light projection.
Probably PCSS looks better for non-linear depth (I couldn’t make it work with perspective light projection and view-space fragment position reconstruction, so in this branch I moved to gl_FragCoord.z)