---

name: perf-analyzer description: Analyzes performance benchmarks from CUDA, CPU, memory tests. Parses output, identifies bottlenecks, tracks metrics over time, generates optimization insights.

Performance Analyzer Skill

Purpose

Analyzes performance benchmarks and profiling data from CUDA, CPU, memory, and system tests. Automatically parses benchmark output, identifies performance bottlenecks, tracks metrics over time, and generates actionable optimization insights.

Capabilities

Benchmark Parsing

• CUDA kernel benchmarks (GFLOPS, TFLOPS, execution time)
• CPU benchmarks (GFLOPS, multi-core performance)
• Memory bandwidth tests
• Custom benchmark formats (CSV, JSON, text output)

Analysis Features

• Performance regression detection
• Bottleneck identification
• Comparison across runs/configurations
• Statistical analysis (mean, std dev, percentiles)
• Speedup calculations
• Efficiency metrics

Tracking

• Historical performance data
• Trend analysis over time
• Configuration impact analysis

Usage

The skill automatically loads when you:

• Ask to analyze benchmark results
• Request performance comparisons
• Need to identify bottlenecks
• Want to track performance over time

Files

• parsers.py - Parsers for various benchmark formats
• analyzer.py - Core analysis logic and metrics calculation
• visualizer.py - Data visualization and charting

Example Workflows

1. Single Benchmark Analysis

   • Parse benchmark output
   • Calculate key metrics
   • Identify bottlenecks
   • Generate summary

2. Comparative Analysis

   • Parse multiple benchmark runs
   • Compare performance across configurations
   • Calculate speedups and improvements
   • Highlight regressions

3. Historical Tracking

   • Store results over time
   • Detect performance trends
   • Alert on regressions
   • Track optimization impact

Integration

Works seamlessly with:

• tech-report skill for generating professional reports
• xlsx skill for detailed data tables
• pptx skill for presentation-ready visualizations