SuperAPI AI Directory

py-perf-analyzer

Coding & development

View repository

Updated 11-06 16:23

About this skill

Perf Profiler

Python performance analysis tool, integrating cProfile, line_profiler, and tracemalloc, with one-click location of CPU and memory bottlenecks.

Features

  • CPU analysis (cProfile): Counts each function's call count, cumulative time, and self time, locating hotspot functions
  • Memory analysis (tracemalloc): Tracks memory allocation hotspots, peak memory, and sources of memory growth
  • Line-by-line analysis (line_profiler): Time and hit count accurate to each line of code, suitable for in-depth optimization
  • Combined analysis:all mode executes once and collects CPU + memory data simultaneously, reducing the overhead of repeated runs

Dependencies

Dependency Type Purpose
Python 3.7+ Required Runtime environment
cProfile / pstats Built-in CPU performance analysis
tracemalloc Built-in Memory tracking
line_profiler Optional Line-by-line analysis (pip install line_profiler)

Usage

python scripts/perf_profile.py <脚本路径> [脚本参数...] [选项]

Parameters

Parameter Description Default value
script Path to the Python script to analyze (required) -
script_args Arguments passed to the target script None
--mode Analysis mode:cpu / memory / line / all cpu
--top Show Top N results 20
--sort CPU analysis sorting:cumulative / tottime / calls cumulative
--output Output JSON report to file Terminal output only
--function Target function names for line-by-line analysis (comma-separated) Auto-discover
--threshold Only show functions whose percentage exceeds this value (%) 0

Examples

# CPU 分析(默认模式)
python scripts/perf_profile.py my_script.py

# 内存分析
python scripts/perf_profile.py my_script.py --mode memory

# 逐行分析指定函数
python scripts/perf_profile.py my_script.py --mode line --function compute,process_data

# 全量分析(CPU + 内存),只看 Top 10
python scripts/perf_profile.py my_script.py --mode all --top 10

# 传递参数给目标脚本,输出 JSON 报告
python scripts/perf_profile.py my_script.py --output report.json -- --input data.csv --output result.csv

# 只关注占比 > 5% 的函数
python scripts/perf_profile.py my_script.py --threshold 5

Output description

CPU analysis report

============================================================
  CPU 性能分析报告 (cProfile)
============================================================

  总执行时间: 2.3456 秒
  总函数调用: 1,234,567 次
  分析函数数: 89 个

  Top 5 耗时函数:
  --------------------------------------------------------
  排名   占比   累计(s)    自身(s)      调用  函数
  --------------------------------------------------------
  1     45.2%    1.0605    0.8234     1000  compute.py:23:matrix_multiply
  2     22.1%    0.5183    0.5183    50000  utils.py:45:normalize
  3     12.3%    0.2885    0.1200      500  io.py:12:read_batch
  ...

  🔍 主要瓶颈: matrix_multiply (45.2% 时间)
     位置: compute.py:23

Memory analysis report

============================================================
  内存分析报告 (tracemalloc)
============================================================

  峰值内存: 128.5 MB
  当前内存: 64.2 MB

  Top 5 内存分配:
  --------------------------------------------------------
  排名       大小     数量  位置
  --------------------------------------------------------
  1      45.2 MB    10000  data_loader.py:78
  2      22.1 MB     5000  transform.py:45
  ...

  内存增长热点:
  --------------------------------------------------------
  1     +32.0 MB  (+8000)  data_loader.py:78
  ...

JSON report

Use the --output parameter to export a complete structured JSON report, containing detailed data for all analysis dimensions, for subsequent processing or integration into a CI pipeline.