AI WebGPU Lab Benchmark

Atomics and Memory Benchmark

`bench-atomics-and-memory` compares deterministic histogram, scatter-accumulation, and reduction kernels under one atomics-heavy benchmark path.

Benchmark Controls

Run the suite to compare contention-heavy kernels and capture one schema-aligned atomics result.

Kernel Matrix

Activity Log

    Contention Preview

    Metrics

    Environment

    Fixture

    Schema-Aligned Result Draft

    {
  "status": "pending"
}

    Benchmark Notes

    • Each case uses a fixed working set, shared memory budget, and atomic pass count.
    • The suite records one aggregate result plus per-case contention and memory bandwidth comparisons.
    • Real WGSL atomics and memory kernels should preserve this result contract before replacing the deterministic harness.