Gsplat fix

[slimbuck]

This PR updates the gaussian clamp to be relative to the viewport size, instead of fixed 1024.

Using a fixed value resulted in too-large gaussians being rendered at lower resolution and actually flooding the viewport with noise.

This update might improve performance at lower resolutions.

[Copilot]

Pull Request Overview

This PR updates the Gaussian clamp in both WGSL and GLSL vertex shaders to use the smaller viewport dimension instead of a fixed value (1024), preventing oversized gaussians at low resolutions and improving performance.

• Replace hardcoded clamp max (1024) with viewport-relative clamp
• Apply changes in both chunks-wgsl and chunks-glsl shader modules

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/scene/shader-lib/chunks-wgsl/gsplat/vert/gsplatCorner.js	Use uniform.viewport to derive max Gaussian clamp dynamically
src/scene/shader-lib/chunks-glsl/gsplat/vert/gsplatCorner.js	Use viewport uniform to derive max Gaussian clamp dynamically

Comments suppressed due to low confidence (4)

src/scene/shader-lib/chunks-wgsl/gsplat/vert/gsplatCorner.js:49

• [nitpick] The variable name vmin is ambiguous. Consider renaming it to viewportMin or minViewportDimension to clarify its purpose.

let vmin = min(uniform.viewport.x, uniform.viewport.y);

src/scene/shader-lib/chunks-wgsl/gsplat/vert/gsplatCorner.js:51

• Consider adding test cases or render tests to verify that the Gaussian radius scales correctly with different viewport sizes, ensuring this relative clamp behavior remains valid.

let l1 = 2.0 * min(sqrt(2.0 * lambda1), vmin);

src/scene/shader-lib/chunks-glsl/gsplat/vert/gsplatCorner.js:48

• [nitpick] The variable name vmin is not immediately clear. Renaming it to viewportMin or minViewportDimension would improve readability.

float vmin = min(viewport.x, viewport.y);

src/scene/shader-lib/chunks-glsl/gsplat/vert/gsplatCorner.js:50

• Add automated tests or render validations to ensure the Gaussian kernel correctly adapts when the viewport size changes, preventing regressions.

float l1 = 2.0 * min(sqrt(2.0 * lambda1), vmin);