界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
数据与研究 · SKILL.md

binary-size

binary-size

【数据与研究】Measure, compare, and reduce ExecuTorch binary size using size-optimized builds, stripped binaries, bloaty/nm analysis, and safe reporting constraints.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: binary-size description: "Measure, compare, and reduce ExecuTorch binary size using size-optimized builds, stripped binaries, bloaty/nm analysis, and safe reporting constraints." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

binary-size

Use this sub-skill when the user asks about ExecuTorch binary size, stripped runtime size, size_test, size_test_all_ops, bloaty, linker symbols, EXECUTORCH_OPTIMIZE_SIZE, or size-regression workflows.

Workflow

  1. Establish a baseline from a clean source checkout and release/size-optimized build.
  2. Measure stripped binaries, not only unstripped debug outputs.
  3. Use symbol/section analysis to identify .text, .rodata, static initializers, logging strings, and template bloat.
  4. Make one logical change at a time and record before/after numbers.
  5. Run relevant correctness tests; do not trade functionality or latency for unverified size savings.

References and Helper

python scripts/compare_binary_sizes.py --before old.bin --after new.bin
我的收藏