Quote:Original post by Matt Aufderheide
"soft" just implies non-hard-edges that's all. All soft-shadows are meant to somehow simulate a penumbra with more or less physical accuracy...
That's entirely my point - VSM is not primarily targeted at softening edges at all! Being able to cheaply blur the shadow map is a useful side-effect, but the main goal is to be able to properly mipmap and aniso filter shadow maps.
Quote:Original post by Matt Aufderheide
screen space is a remarkaby efficient place to do filtering.
No not at all! Screen space is a convenient place to do bluring and other such fun, but *not* filtering. Indeed you *can't* do filtering in screen space, which is why I always bring up the point that VSM is a shadow filtering method, not an edge softening/bluring whatever... it is not comparable in the slightest to screen-space bluring.
Quote:Original post by Matt Aufderheide
A rasterizer is all a big fake anyway.
Nonsense - it produces 100% physically accurate results to a specific set of ray queries. It's just an efficient way of building a "perfect" projected uniform grid accelerator for primary ray queries. In terms of shadows it's a bit more of a "hack" in that you're querying a lossy accelerator in typical implementations, but "deep" shadow maps or irregular rasterization can solve that. There's nothing inherently wrong with rasterization - indeed it is particularly useful for primary and shadow rays. But sometimes you need more complicated queries which just means you often need more complicated data structures and query mechanisms.