Here gains are a bit smaller but still quite noticeable on slow machines: 20 fps => ~25-27 fps

It runs 60+ fps on my Nvidia Quadro 2000M but it suffers a lot on ATI 3650M at ~20 fps (these changes bumped it up to ~30 fps).

Added corresponding example. Tweaked bit previous bump map example.

- Add parameter for vertex skimming
- Minimize square root calculations

I find dark background intimidating. Specially for newcomers.

Specular map works like it used to for normal map shader. It is assumed to be a grayscale texture.

For Phong specular map value modulates specular term strength and environment map reflection strength.

For Lambert and Basic materials specular map affects just environment map reflection strength (as these have no specular term).

Still not very sure how environment mapping and specular term should interact with each other.

Testing editing via GitHub interface

This is using Prettify highlighter: http://code.google.com/p/google-code-prettify/

Added "threejs.css" for doing custom styling (if somebody would like to do this). See for example here:

http://google-code-prettify.googlecode.com/svn/trunk/styles/index.html
http://code.google.com/p/google-code-prettify/source/browse/trunk/styles/sunburst.css

Fixes #2318

Changed multiply to use temporary variables in case a or b is this.

It requires OES_standard_derivatives extension and benefits a lot from having EXT_texture_filter_anisotropic extension (set high anisotropy for bump texture).

Added new LeePerry example and modified materials2 WebGL example to show this in action.

Still quite a few things not solved:

- why there are seams at LeePerry model? (I suspect mesh discontinuities? this method is supposed to have less seams problems than normal mapping :S)
- it will crash without color map due to missing UVs, need to solve better UV initialization / sharing of shader parameters between chunks
- didn't check impact on environment mapping yet
- maybe also Lambert could be extended
- it should be somehow possible to precompute derivative maps and use this instead of "dHdxy_fwd()" method

isZero actually checks the squared length against the "small epsilon".
0.0001 = 0.01² so the function will return true for a length < 0.01.

By the way if there is a good reason for a so limited accuracy it should be given here in the doc (or maybe even in the code).