spatial-partitioning

Use a uniform grid or spatial hash where each unit is assigned to a cell bucket based on position. Neighbor queries then check only the unit's cell and adjacent cells instead of comparing against every unit in the world.

Uniform grids are preferred for RTS games because units are typically distributed across the map and grid cells are lightweight, cache-friendly, and cheap to update. Trees add recursive overhead and complexity that is rarely justified early on.

Reuse pre-allocated query buffers instead of creating new collections per query, and update partitions incrementally as units move between cells. This keeps queries allocation-free and maintains deterministic frame performance.

Avoid adding spatial partitioning when unit counts are very small, since the overhead of maintaining cells exceeds the cost of direct comparisons. Also avoid complex tree structures prematurely before profiling shows a real bottleneck.

Update partitions incrementally by moving units between cell buckets only when they cross a cell boundary, rather than rebuilding the entire grid each frame. This minimizes per-frame work and keeps cells iteration-efficient.