benches

benches (benches/) is the Criterion-based benchmark crate.

Purpose

• Track the performance of hot paths in wgpu and wgpu-core: command-encoder recording, render-pass setup, shader compilation, ...
• Catch performance regressions in CI by running each benchmark for a single iteration.
• Provide a place for maintainers to drill into a perf issue.

Directory layout

benches/
├── Cargo.toml
├── README.md
└── benches/
    └── wgpu-benchmark/...

Run them

cargo bench                            # full Criterion run (slow)
cargo nextest run --bench wgpu-benchmark  # smoke run, one iteration each (CI does this)

docs/testing.md notes that benchmarks should avoid expensive setup outside the Criterion benchmark functions — otherwise listing tests becomes slow and CI's smoke run drags.

Hot paths benchmarked

The benchmark file names describe what's measured: computepass.rs for compute-pass encoding, renderpass.rs for render-pass encoding, shader.rs for shader compilation, resource_creation.rs for buffer/texture creation, etc.

Integration points

• Above: developers tracking performance.
• Below: wgpu (which transitively pulls in everything else).

When to add a benchmark

When you optimize a hot path, add (or extend) a benchmark in benches/benches/wgpu-benchmark/ so the gain is measurable. Future regressions on the same path will show up in the Criterion comparison.