Use half-precision types in bloom downsample/upsample WGSL shaders

[mvaligursky]

Use half-precision (f16) types for intermediate computations in the bloom downsample and upsample WGSL shaders.

Changes:

• Convert all texture sample intermediates and accumulation in the WGSL downsample shader (13-tap filter) from vec3f to half3, including BOXFILTER and PREMULTIPLY paths
• Convert all texture sample intermediates and accumulation in the WGSL upsample shader (9-tap tent filter) from vec3f to half3
• Restructure upsample accumulation in both WGSL and GLSL shaders from accumulate-then-divide to multiply-before-accumulate, preventing potential f16 overflow (intermediate values previously reached 16x the sample value, overflowing f16 at HDR values above 4094)

Performance:

• Reduces register pressure (~39 f32 registers to ~20 f16 registers in the downsample shader) improving GPU occupancy
• Enables 2x ALU throughput on GPUs with native f16 support (AMD RDNA, Apple, Qualcomm, ARM Mali)
• Falls back to f32 automatically on devices without shader-f16 support via existing half type aliases

[Copilot]

Pull request overview

Updates the bloom render-pass downsample/upsample shader chunks to use half-precision (half*) for intermediate computations in WGSL, and adjusts the upsample accumulation math to reduce risk of intermediate overflow while preserving the filter result.

Changes:

• WGSL bloom downsample: convert sampled/intermediate color values and accumulation from vec3f to half3 (including BOXFILTER + PREMULTIPLY paths).
• WGSL bloom upsample: convert sampled/intermediate color values and accumulation from vec3f to half3.
• WGSL + GLSL bloom upsample: change from “accumulate then divide” to pre-weighted accumulation (equivalent normalization, lower peak intermediate magnitude).

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
src/scene/shader-lib/wgsl/chunks/render-pass/frag/upsample.js	Uses half3 intermediates and pre-weighted accumulation; casts back to vec3f for output.
src/scene/shader-lib/wgsl/chunks/render-pass/frag/downsample.js	Uses half3 intermediates/accumulation (incl. BOXFILTER/PREMULTIPLY) and casts back to vec3f for output.
src/scene/shader-lib/glsl/chunks/render-pass/frag/upsample.js	Switches to pre-weighted accumulation constants equivalent to prior normalize-by-16 approach.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.