AGENT SKILL

scientific-rust-performance-benchmarking

Design or run an explicitly budgeted scientific Rust benchmark comparing equivalent numerical workloads with reproducible timing, memory and scaling evidence.

License 许可
MIT
最近更新
2026-10-30
Stars
★ 0
Forks
0

01 前言 Front Matter

name
scientific-rust-performance-benchmarking
description
Design or run an explicitly budgeted scientific Rust benchmark comparing equivalent numerical workloads with reproducible timing, memory and scaling evidence.

03 SKILL.md 正文

Read Rust performance and reproducibility. Establish the performance question, baseline revision, workload, accuracy contract and resource budget. Confirm correctness equivalence before interpreting speed differences. Inspect existing benchmark targets/tools and actual project commands; do not install a framework or assume a nightly harness.

Define inputs, optimized profile, target/features, precision, thread counts, warmup, repetitions, noise controls and reported statistic. Specify which costs are included (setup, transfer, I/O, kernel, full run) and relevant memory/scaling metrics. Account for dead-code elimination, caches and nested thread pools. Record hardware and software context without environment/credential dumps.

Prefer a small representative experiment before an expensive run. Execute only authorized workloads within declared wall-time/memory/disk/worker limits using existing project/OS controls. Pack activation and a runner timeout are not authorization or resource isolation. Missing hardware, references or budget block claims that depend on them; do not submit cluster jobs or download large data automatically.

Use the profile's scientific-benchmark command for repeatable runs. Return baseline/candidate commands and revisions, accuracy evidence, sample/variability statistics, memory/resource observations, artifact locations, limitations and the scope of any supported claim. For design-only requests return a protocol, not invented timings. Unmeasured optimizations remain hypotheses; do not cherry-pick samples or silently change tolerances/workloads.

04 同仓库更多 Skills