luma.gl glossary

This glossary gives luma.gl terms one canonical definition. Reference pages link here instead of redefining the same concept differently in every module.

GPU resources and execution

Term	Meaning
Resource	A GPU-backed object such as a buffer, texture, sampler, shader, or pipeline. Resources have an explicit owner and lifetime even when another object borrows them.
Ownership	Responsibility for destroying an object. A borrowing renderer may use a resource without owning it.
Binding	A connection between a shader-visible name or slot and a buffer, texture, sampler, or other resource.
Layout	The declared memory or binding structure that lets application data agree with shader inputs.
Pipeline	Compiled render or compute state: shaders plus fixed configuration such as vertex layouts, attachment formats, and depth or blend state.
Pass	A group of render or compute commands encoded against compatible attachments and pipeline state.
Command encoder	A CPU-side recorder that collects GPU commands before submission.
Submission	Sending a finished command buffer to the GPU queue. Encoding describes work; submission schedules it.
GPU residency	Data being available in GPU-accessible memory without requiring a CPU download or re-upload for the next operation.

Scheduling and synchronization

Term	Meaning
Data hazard	A read/write conflict where one operation depends on data another operation reads or writes. A command graph uses declared accesses to order hazardous operations safely.
Dependency	An ordering relationship between operations. It may follow from resource access or be declared explicitly for a semantic reason.
Indirect work	A draw or dispatch whose count or arguments are written into a GPU buffer, allowing GPU results to control later GPU work without CPU readback.
Transient resource	Scratch storage whose logical lifetime is limited to part of a compiled workflow and can therefore share physical memory with non-overlapping resources.
Aliasing	Reusing the same physical allocation for distinct logical resources whose live ranges do not overlap. Writable overlap at the same time is rejected.
Readback	Copying a GPU result into CPU-visible memory. Readback can introduce synchronization and should be limited to small, explicitly requested results.

Engine and shaders

Term	Meaning
Redraw	A request to produce another frame because view, data, animation, or asynchronous GPU state changed. An unchanged view should remain idle.
Shader module	Reusable WGSL/GLSL source plus its declared dependencies and shader-facing inputs.
Shader hook	A named extension point owned and called by base shader source.
Injection	Source inserted at a hook or standard assembly location without copying the complete base shader.
Shader plugin	Optional reusable behavior that can contribute modules, defines, vertex inputs, and injections to a model or computation.

Related pages

• How luma.gl fits together
• Core GPU programming
• Shader assembly
• GPU scheduling concepts