GPU A5

@luma.gl/gpgpu/gpu-a5 brings the same GPU-resident cell-center workflow to A5 indexes. Its API is intentionally identical to the H3 adapter, allowing graph builders to change the grid family without changing buffer layouts or downstream coordinate consumers.

import {GPUA5CellProjection} from '@luma.gl/gpgpu/gpu-a5';

new GPUA5CellProjection({
  cells,
  output: longitudeLatitudes,
  validity,
  projection: 'lnglat'
}).addToGraph(graph);

Try it on this device

Choose A5 below to measure the public A5 adapter. The source set spans representative resolutions and reflected topology; it is repeated to the chosen array size before upload.

Live DGGS cell-center projection

Decode a repeated, globally distributed set of valid H3 or A5 indexes with the public command-graph primitive. Correctness checks, upload, compilation, and readback stay outside measured submissions.

Ready to decode 16,384 H3 indexes into normalized vectors.

GridH3A5OutputUnit vectorLongitude / latitudeCells16,38465,536262,1441,048,576

Mobile quality · smallest workload

Shared by design

GPUA5CellProjection subclasses the public GPUDGGSCellProjection contributor. Buffer validation, word ordering, output formats, validity masks, resource declarations, workload estimates, bounded dispatch, and benchmark timing are shared with H3. Only the index validation and topology-to-sphere functions differ.

Property	A5 behavior
cells	Packed low-word/high-word uint32x2 by default
projection: 'lnglat'	Writes float32x2 longitude/latitude degrees
projection: 'unit-vector'	Writes normalized float32x3 Cartesian centers
validity	Optional uint32 mask; reserved zero and malformed cells write 0

Resolution and topology

The shader deserializes the A5 resolution and topology bits from the split index, reconstructs the cell's spherical center, and handles resolution-zero centers explicitly. The unit-vector path is a natural choice for globe rendering and spatial tests; longitude/latitude is available for labels, interchange, and downstream projections.

tip

One planner for both grids.

If an application selects A5 or H3 from data metadata, instantiate GPUDGGSCellProjection with a runtime family value. If the grid is fixed, prefer GPUA5CellProjection: its import communicates intent and prevents accidentally selecting another family.

Composition

A5 output is source-aligned and graph-resident. Common continuations include:

• passing unit vectors directly to a globe model;
• compacting rows with the validity mask;
• computing dot products or angular thresholds without converting back to degrees;
• explicitly reading a small selected subset through a readback ring.

No implicit upload or readback occurs. The caller retains ownership of every buffer and controls when the compiled graph is submitted.

Current scope

The module decodes A5 cell centers. Boundary generation, neighborhood traversal, hierarchy transforms, and coordinate-to-cell encoding are not part of this first API. They can reuse the shared split-64-bit and command-graph infrastructure while defining their own capacity and precision contracts.