Tracker: luma.gl v10

[ibgreen]

Themes: deck.gl WebGPU support

• Target Date: Q3 2026
• deck.gl v10 tracker
• v10 roadmap
• v9 roadmap
• Previous trackers v9.3, v9.2, v9.1 and v9.0

Dynamic Shaders

Finding the right level of support for dynamic shaders. In particular, the ability to control attribute/storagebuffer/constants

• ShaderPlugins - feat(shadertools) ShaderPlugins #2664 - Bring some of deck.gl ShaderExtension ideas into luma.gl

Batched Vector support

• Add GPUTable, GPUVector, GPUData, GPURecordBatch @ibgreen
• Add GPUTableMode that can accept a GPUArrowTable for attributes and storage buffers. @ibgreen
• Add TableBufferPlanner @ibgreen

Apache Arrow support

• Well defined mappings from common Arrow type columns to GPUVectors.
• Streaming support
• ...

GPGPU support

We want a good base set of GPU powered computations and use them when preparing buffers for rendering.

• Add GPU algorithms for important operations (fround, interleave etc) @Pessimistress
• Add lazy evaluation support for GPU operations @Pessimistress
• Review/implement follow up suggestions in feat(gpgpu): add module for generic data processing #2594

Buffer Budget Optimization

WebGPU has a strict 8 buffer limit per pipeline, and attribute budget is also limited to 16+. We'd like to keep as many buffers unallocated as possible

• Remove need for instancePickingColor attributes (no buffer, use instance_index in shaders)
• Use interleaved primitive geometries (single buffer)

Reduce deck.gl WebGPU porting friction

A goal of luma.gl v10 is to remove friction encountered when porting deck.gl layers:

• Make shader layout generation tolerant of unused supplied attributes. If a model gets a buffer for instancePickingColors but the WGSL shader does not declare it, that should be ignored cleanly, not turn into a malformed shaderLocation path.
• Give modules a stable ownership model for textures/samplers. Texture bindings should not depend on naming conventions like fooTexture and fooTextureSampler sprinkled through layers. Luma should formalize that in module metadata.
• Make WebGPU constant/noAlloc attributes behave like first-class attributes. - The base attribute system currently works well for WebGL, but on WebGPU the “constant” and noAlloc paths can leave required vertex slots unbound. Luma/deck should either:
• emulate constant attributes with small real GPU buffers automatically, or make vertex-state generation skip attributes not present in shader layout cleanly. - That would remove the matrix/picking attribute workarounds.
• Add a real bind-group allocation model for shader modules. - Right now WGSL modules are effectively hardcoding @group/@binding, which makes module composition brittle. Luma should let modules declare logical bindings and have the assembler assign final binding locations. That removes the layer-local binding hacks around simpleMesh, phongMaterial, textures, and samplers.
• Include attachment formats in the WebGPU render-pipeline cache key. The stale BGRA8/Depth24Plus vs RGBA8/Depth16 mismatch should never have reached the layer. colorAttachmentFormats and depthStencilAttachmentFormat need to participate in pipeline hashing so a model can safely render into screen and picking passes without custom no-cache workarounds.

General Features

@luma.gl/core

• Move readPixels*() to CommandEncoder and update all usage

@luma.gl/shadertools

• Implement simple WebGPU support for constant attributes (generate repeated buffers)
• Define module conventions for WGSL (more limited, cannot inject varyings etc, performance wise too ). Also see common extensions for wgsl #2246
• Fix fp64 under Apple M1 P0 - fp64 shader emulation not working under Apple M1 Max #1764

@luma.gl/core/shadertypes

• Update shader module type system to handle arrays and structs
  • Update shader types to handle structs and arrays
  • Re-implement offset generation for plain types using the new shadertypes
  • Implement offset generation for arrays and structs
• Support multiple lights in lighting module (Leverage new support for "struct / array" uniform type decls in shader modules).

@luma.gl/engine

@luma.gl/effects

@luma.gl/gltf

• PBR lighting now supported on WebGPU (in progress)
• glTF now supported on WebGPU (in progress)
• Tracker: v9 glTF support #1847

Bugs

• https://github.com/visgl/luma.gl/issues?q=is%3Aissue+is%3Aopen+milestone%3Av9.4

Docs / Website

Continued Documentation Improvements. Tracker: #2407

Tests

• Restore all luma.gl render tests (update golden images?) - P2
• Tests: Port all classic webgl test suites to core classes - P2

Deprecations

• Remove Model.updateModuleSettings.
• Remove device.setGLParameters & Co
• ...