Merged
Conversation
mvaligursky
added
performance
Contributor
There was a problem hiding this comment.
Pull request overview
This PR introduces a compact work buffer format for GSplat rendering that reduces memory and bandwidth usage by 37.5% (from 32 to 20 bytes per splat) compared to the existing format. The compact format uses advanced encoding techniques including 11+11+10 bit RGB color (supporting [0,4] range), half-angle quaternion projection for rotation (3 components instead of 4), and log-encoded scale values covering e^-12 to e^9 range. The format is runtime-switchable via the scene.gsplat.dataFormat property, with GSPLATDATA_COMPACT as the new default and GSPLATDATA_LARGE available for full precision when needed.
The PR also unifies the rendering code path by using instanced draws for all splats (both octree and non-octree), eliminating the zero-fill code path that was previously needed for non-LOD splats. This simplification removes redundant uniforms (uActiveSplats, uViewportWidth) and ensures consistent behavior across all rendering scenarios.
Changes:
- Added compact 20 bytes/splat format with specialized encoding (half-angle quaternion, log-encoded scale, packed RGB)
- Created format-specific read/write shader interfaces for both GLSL and WGSL
- Unified instanced draw path for all splats, removing conditional rendering logic
- Added runtime format switching capability with automatic work buffer recreation
- Updated viewer and lod-streaming examples with compact format toggle (defaulting to enabled)
Reviewed changes
Copilot reviewed 22 out of 22 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| src/scene/constants.js | Added GSPLATDATA_COMPACT and GSPLATDATA_LARGE constants for format selection |
| src/scene/gsplat-unified/gsplat-params.js | Implemented format creation and runtime switching logic, defaulting to compact format |
| src/scene/gsplat-unified/gsplat-manager.js | Added handleFormatChange() to detect and handle format changes, ensuring work buffer recreation |
| src/scene/gsplat-unified/gsplat-work-buffer-render-pass.js | Removed conditional intervals logic, unified to always use instanced draws |
| src/scene/gsplat-unified/gsplat-renderer.js | Extracted setOrderData() method for reuse after work buffer changes |
| src/scene/gsplat-unified/gsplat-info.js | Synthesizes full-range interval when none exist to ensure unified instanced path |
| src/scene/gsplat/gsplat-format.js | Added setWriteCode/getWriteCode methods for format-specific write encoding |
| src/scene/gsplat/gsplat-resource-base.js | Integrated format-specific write code injection, removed useIntervals parameter |
| src/scene/shader-lib/wgsl/chunks/gsplat/vert/formats/containerCompactRead.js | WGSL read functions for compact format with half-angle quaternion and log-scale decoding |
| src/scene/shader-lib/wgsl/chunks/gsplat/frag/formats/containerCompactWrite.js | WGSL write functions for compact format with specialized encoding |
| src/scene/shader-lib/wgsl/chunks/gsplat/frag/formats/containerPackedWrite.js | WGSL write functions for large format, extracted from copy shader |
| src/scene/shader-lib/wgsl/chunks/gsplat/frag/gsplatCopyToWorkbuffer.js | Unified shader logic, removed zero-fill path, uses format-specific writeSplat() |
| src/scene/shader-lib/glsl/chunks/gsplat/vert/formats/containerCompactRead.js | GLSL read functions for compact format (parallel to WGSL version) |
| src/scene/shader-lib/glsl/chunks/gsplat/frag/formats/containerCompactWrite.js | GLSL write functions for compact format (parallel to WGSL version) |
| src/scene/shader-lib/glsl/chunks/gsplat/frag/formats/containerPackedWrite.js | GLSL write functions for large format, extracted from copy shader |
| src/scene/shader-lib/glsl/chunks/gsplat/frag/gsplatCopyToWorkbuffer.js | Unified shader logic (parallel to WGSL version) |
| src/scene/shader-lib/wgsl/collections/gsplat-chunks-wgsl.js | Removed containerPackedRead from collection (now imported directly where needed) |
| src/scene/shader-lib/glsl/collections/gsplat-chunks-glsl.js | Removed containerPackedRead from collection (now imported directly where needed) |
| examples/src/examples/gaussian-splatting/viewer.example.mjs | Added compact toggle control, defaults to enabled |
| examples/src/examples/gaussian-splatting/viewer.controls.mjs | Added UI control for compact format toggle |
| examples/src/examples/gaussian-splatting/lod-streaming.example.mjs | Added compact toggle control, defaults to enabled |
| examples/src/examples/gaussian-splatting/lod-streaming.controls.mjs | Added UI control for compact format toggle |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
mvaligursky
pushed a commit
that referenced
this pull request
Feb 26, 2026
setLayout synthesized a full-range interval [0, activeSplats) directly into this.intervals when no intervals existed (fully-loaded octree). This mutation leaked into GSplatIntervalCompaction.uploadIntervals, which saw intervals.length > 0 and fell into the wrong branch — mapping all splats to a single boundsIndex and losing per-node culling granularity. The result was incorrect GPU frustum culling when all octree nodes were loaded. Move the synthesis into updateSubDraws as a local variable so this.intervals is never mutated and the interval compaction correctly falls back to placementIntervals for per-node bounds mapping. Fixes regression from #8480. Made-with: Cursor
mvaligursky
added a commit
that referenced
this pull request
Feb 26, 2026
setLayout synthesized a full-range interval [0, activeSplats) directly into this.intervals when no intervals existed (fully-loaded octree). This mutation leaked into GSplatIntervalCompaction.uploadIntervals, which saw intervals.length > 0 and fell into the wrong branch — mapping all splats to a single boundsIndex and losing per-node culling granularity. The result was incorrect GPU frustum culling when all octree nodes were loaded. Move the synthesis into updateSubDraws as a local variable so this.intervals is never mutated and the interval compaction correctly falls back to placementIntervals for per-node bounds mapping. Fixes regression from #8480. Made-with: Cursor Co-authored-by: Martin Valigursky <mvaligursky@snapchat.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a compact 20 bytes/splat work buffer format (
GSPLATDATA_COMPACT) as the new default,reducing memory and bandwidth by 37.5% compared to the existing 32 bytes/splat format.
(f32 position + half-angle quaternion), R32U scale/alpha (3x8-bit log-encoded scale + 8-bit alpha)
scene.gsplat.dataFormatpropertyzero-fill code path and redundant uniforms
writeSplat()shader interface decoupling the copy shader fromindividual format encodings