WebGLRenderer: Stable reversed Z buffer implementation.

[eXponenta]

Fixed #29578
Related PR: #29445

Description
Fix: reset clip state when reset is called
Fix: valid depth clear value when reversed is enabled

Feat: non-persistent reversedZ state ( can be controlled via renderer.state.buffers.depth.setReversed( )))

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	692.71 171.44	692.88 171.51	+171 B +75 B
WebGPU	817.93 220.59	817.93 220.59	+0 B +0 B
WebGPU Nodes	817.44 220.46	817.44 220.46	+0 B +0 B

🌳 Bundle size after tree-shaking

Minimal build including a renderer, camera, empty scene, and dependencies.

	Before	After	Diff
WebGL	464.93 112.2	465.09 112.28	+161 B +76 B
WebGPU	538.86 145.56	538.86 145.56	+0 B +0 B
WebGPU Nodes	494.86 135.37	494.86 135.37	+0 B +0 B

[Mugen87]

LGTM! Would be great if @CodyJasonBennett could also have a look 👍 .

[eXponenta]

@CodyJasonBennett can you explain how logarithmicDepth should work with reversedZ?
Looks like it works no fully stable ( maybe for me only ).
I got flipped depth when logarithmic is enabled.

Looks like needs to change math for it also:

three.js/src/renderers/shaders/ShaderChunk/logdepthbuf_fragment.glsl.js

Line 6 in 36bac08

gl_FragDepth = vIsPerspective == 0.0 ? gl_FragCoord.z : log2( vFragDepth ) * logDepthBufFC * 0.5;

three.js/src/renderers/WebGLRenderer.js

Line 2008 in 36bac08

p_uniforms.setValue( _gl, 'logDepthBufFC',

[CodyJasonBennett]

You should not use those together. Reverse depth is a strict upgrade and doesn't disable early-z optimizations like logarithmic depth does. I'd expect something is not cleaning up if you see any change to logarithmic depth.

[eXponenta]

You should not use those together. Reverse depth is a strict upgrade and doesn't disable early-z optimizations like logarithmic depth does. I'd expect something is not cleaning up if you see any change to logarithmic depth.

But by default we can create it together, because capabilities not disable logDepth if we can use reversedZ

[WestLangley]

You should not use those together.

@Mugen87 Perhaps an API change is warranted...

THREE.StandardDepthBuffer
THREE.ReversedZDepthBuffer
THREE.LogarithmicDepthBuffer

const renderer = new THREE.WebGPURenderer( { depthBuffer: THREE.LogarithmicDepthBuffer } );

[Mugen87]

This is a great suggestion! 👍

[mrdoob]

Indeed! 👍

[CodyJasonBennett]

If EXT_clip_control was widely available, I would suggest removing logarithmic depth entirely.

[CodyJasonBennett]

Is it really worth an API change which will not be future proof when exactly reverse-z is opt-in? It is a strict improvement over logarithmic depth where EXT_clip_control is available. I'd wager it will be GA and diffuse before WebGPU, as I implemented this after sitting in the meeting for the announcement of this extension's shipping in Safari.

[prideout]

FWIW, logarithmic depth already doesn't play nicely with some features (e.g. polygonOffset) so I think Cody's proposal (API simplification rather than API enhancement) is worth considering , especially since EXT_clip_control is now supported in Safari and Chrome.

[Mugen87]

I agree it's better to just have two types of depth buffer implementation and deprecate logarithmic depth buffer in the future.

It might still be a good idea to switch to enums instead of using flags since it feels more readable. However, I'm not feeling strong about that.

[WestLangley]

Given the recent comments, I am OK with leaving the API as-is.

[prideout]

Hi @eXponenta, before this PR, reversed Z worked in my simple sandbox app without any changes other than passing the new param to the renderer constructor. However after this PR, my app does not render anything unless I add renderer.getContext().clearDepth(0) to my app as a workaround.

Could that be due to the removed line that called _gl.clearDepth?

[prideout]

@eXponenta Additionally, maybe we should also do:

const oldFunc = currentDepthFunc;
currentDepthFunc = null;
this.setFunc( oldFunc );

[eXponenta]

Oops. Yep, this one.
Because we also should upgrade depth func to inverter version.