Webgpu Compute Shaders

WebGPU & Compute Shaders

Core Mechanics

WebGPU is the modern successor to WebGL. It unlocks the massive parallel processing power of the GPU directly in the browser, not just for drawing triangles, but for general-purpose compute (GPGPU) like Physics simulations or Machine Learning.

1. Compute Shaders (WGSL)

Unlike Vertex or Fragment shaders which are tied to the rendering pipeline, a Compute Shader is just a program executed thousands of times simultaneously by the GPU. Written in WebGPU Shading Language (WGSL).

2. Workgroups & Shared Memory

GPU threads are organized into Workgroups. Threads within the same Workgroup can share a special, ultra-fast memory space (var<workgroup>) and synchronize with each other using barriers (workgroupBarrier()). This is critical for complex algorithms like Matrix Multiplication.

Compute Shader Architecture Map

%%{init: {"theme": "default", "flowchart": {"useMaxWidth": true}}}%%
flowchart TD
    subgraph JS ["JavaScript (CPU)"]
        A["Allocate Float32Array"]
        B["Write to GPU Buffer"]
        C["Dispatch Compute Pipeline"]
    end
    
    subgraph GPU_Hardware ["GPU Parallel Execution"]
        subgraph Workgroup ["Workgroup (e.g., 64 Threads)"]
            D["Invocation 1"]
            E["Invocation 2"]
            F["Invocation 64"]
            G["Shared Memory (Ultra-fast)"]
        end
    end
    
    subgraph Result ["Readback"]
        H["Copy Buffer to CPU"]
    end
    
    A --> B
    B --> C
    C -->|"Spawn Threads"| D
    C -->|"Spawn Threads"| E
    C -->|"Spawn Threads"| F
    D <--> G
    E <--> G
    F <--> G
    Workgroup -->|"Barrier Sync"| H