perf-analyzer

Performance benchmark analysis extracts metrics from profiling outputs to pinpoint bottlenecks. perf-analyzer automatically parses CUDA, CPU, and memory test results across CSV, JSON, and text formats, computes statistics like mean and percentiles, detects regressions, and delivers optimization insights to guide where to focus improvement efforts.

Yes. perf-analyzer tracks metrics over time across different configurations, calculates speedups between runs, and identifies performance regressions. Feed it benchmark outputs from multiple CUDA or CPU test sessions and it generates comparative reports highlighting which configurations perform best.

Performance analysis supports CSV, JSON, and plain-text benchmark outputs. perf-analyzer parses CUDA GFLOPS/TFLOPS metrics, CPU timings, and memory test results, handling diverse profiling tool output formats so you can consolidate results regardless of which benchmarking tool generated them.

Regression detection compares historical benchmark metrics to identify performance drops. perf-analyzer tracks time-series data across runs, calculates statistical measures like standard deviation and percentiles, and flags when new results fall below baseline thresholds, enabling early detection of optimization failures.

Yes. perf-analyzer integrates with reporting components to export analysis results to xlsx and pptx formats. After parsing benchmarks, computing metrics, and identifying bottlenecks, it generates structured reports and visualizations summarizing optimization insights and recommendations.

Benchmark analysis computes statistical metrics including mean, median, standard deviation, and percentiles from performance data. perf-analyzer also estimates speedups between runs, detects bottlenecks in CUDA and CPU workloads, and delivers actionable insights on which components to optimize.