[Game Programming] Data Locality Pattern

One-line pattern summary

A performance-oriented pattern that places frequently accessed data in contiguous memory to improve cache efficiency.

Typical Unity use cases

• When updating a large number of objects such as bullets or particles every frame.
• When CPU bottlenecks need to be reduced.

Parts (roles)

• Hot Data: frequently accessed values
• Contiguous Storage: sequential arrays
• Loop: simple repetitive processing

Unity example (C#)

The code below is a simplified Unity example based on the scenario described above.

using UnityEngine;

public struct ProjectileState
{
    public Vector3 Position;
    public Vector3 Velocity;
}

public static class ProjectileSimulation
{
    public static void Simulate(ProjectileState[] projectileStates, float deltaTime)
    {
        for (int projectileIndex = 0; projectileIndex < projectileStates.Length; projectileIndex++)
        {
            projectileStates[projectileIndex].Position += projectileStates[projectileIndex].Velocity * deltaTime;
        }
    }
}

Advantages

• Sequential memory access reduces cache misses and improves throughput for large-scale calculations.
• It becomes especially powerful when combined with data-oriented systems such as Burst and Jobs.

Things to watch out for

• If the structure is optimized too heavily for performance, readability and domain clarity can suffer.
• Mistakes in array synchronization and index management can easily cause data mismatch bugs.

Interaction diagram

This shows the flow where contiguous memory blocks are processed sequentially to improve cache efficiency.

flowchart LR

  frame["Frame Start"]
  chunk_loop["for (i = 0..N)"]
  cache_hit["Cache-friendly access"]
  result["Bulk update complete"]

  subgraph soa["SoA Buffers"]
    positions["positions[]"]
    velocities["velocities[]"]
    hp["hp[]"]
  end

  frame --> chunk_loop
  chunk_loop -->|read i| positions
  chunk_loop -->|read i| velocities
  chunk_loop -->|read i| hp
  positions --> cache_hit
  velocities --> cache_hit
  hp --> cache_hit
  cache_hit --> result