WebGLState: Added caching for blend equation

[oguzeroglu]

In this PR a cache mechanism is implemented for Blend equations inside the WebGLState. I found this to be necessary since three.js was making redundant WebGL calls for such functions:

After the implementation I observed that such WebGL calls are eliminated thus performance increase is expected for scenes with many materials.

[Mugen87]

Can you please setup your PR on the latest dev branch? Besides, please use hard tabs instead of soft tabs.

[Mugen87]

I'm confused. This line should already ensure no unnecessary updates, right?

[oguzeroglu]

I was confused as well but it turns out that the currentBlending variable is shared between different materials so it did not ensure the blending equation caching (that's why I use the currentMaterial variable actually). Maybe it was supposed to serve another purpose that I'm not aware of, so I kept it just in case.

[Mugen87]

I don't think the change is correct. Caching WebGL State information is something different than caching uniforms. The WebGL state is always related to a draw call, it can't retain information that is related to a material (or shader program). In other words: You always have to set the blending equation or function if the current ones do not match.

[oguzeroglu]

Yes I believe you are right about that. I believe that the reason of the redundant calls was that the blending equation function is acually the same (gl.FUNC_ADD) for different blending modes and that probably was the reason why this PR seemed to work. Maybe there should be different caches for blending equations and functions to prevent the redundant calls. Any thoughts?

For example:

if (NOT the last blending equation was gl.FUNC_ADD){
gl.blendEquationSeparate( gl.FUNC_ADD, gl.FUNC_ADD );
}

The line if ( blending !== currentBlending || premultipliedAlpha !== currentPremultipledAlpha) is not preventing gl.BlendEquation calls. I believe it was designed for gl.blendFuncSeparate calls.

I'm closing this PR as this adresses the problem in a wrong way :) I'll try non material based caching for blend equations as I mentionned above.

[Mugen87]

Thanks for investigating this! 👍