Skip to content

meirborowski/Lodestone / .claude/skills/run-benchmark

run-benchmark

Verified

Measure Lodestone's renderer with its benchmark against the recorded baseline, or deliberately record a new baseline. Use after changes that could affect rendering performance or draw calls, when the "Benchmark draw calls stay within the baseline" test fails, and before finishing renderer work.

FileKindSize
SKILL.mdskill md2.2 KB
---
name: run-benchmark
description: Measure Lodestone's renderer with its benchmark against the recorded baseline, or deliberately record a new baseline. Use after changes that could affect rendering performance or draw calls, when the "Benchmark draw calls stay within the baseline" test fails, and before finishing renderer work.
---

# Run the Renderer Benchmark

`LodestoneBenchmark` (`Tests/Benchmark`) renders a small city that turns on every renderer feature, along a fixed camera flight, and reports CPU and GPU frame times, per-pass GPU times, draw calls and triangles (see `docs/Testing.md#performance`). The baseline is `Tests/Benchmark/Baseline.json`: its draw calls are gated on every platform, and its timings are the reference GPU's.

## Measure
1. Build Release - Debug timings mean nothing: `cmake --build --preset release --target LodestoneBenchmark`
2. Run it on the GPU, at 1920x1080:
   ```sh
   build/release/Tests/LodestoneBenchmark --baseline Tests/Benchmark/Baseline.json --min-fps 60
   ```
   It prints the report as JSON (`--output <file>` also writes it), fails if draw calls rose above the baseline's or the frame rate is under `--min-fps`, and compares timings with the baseline's when the baseline was recorded on the same GPU at the same size - warning about slowdowns over 10%
3. Laptops: plug in power and close heavy programs; numbers vary by a few percent between runs. Run twice before believing a regression

## When draw calls rose
The test `Render.Benchmark draw calls stay within the baseline` runs the benchmark's probe frames on lavapipe. Draw calls don't depend on the GPU, so a rise is a real change: culling broken, extra passes, more pipelines per material. Find the cause first. Only if the rise is meant (a new pass, a richer benchmark scene), record a new baseline.

## Record a new baseline
Only on purpose, and on a real GPU:
```sh
build/release/Tests/LodestoneBenchmark --frames 300 --record Tests/Benchmark/Baseline.json
```
Commit it with a message saying why the baseline changed, the GPU it was recorded on, and the old and new draw calls and frame times. Changing the benchmark scene (`BenchmarkScene.cpp`) or its camera flight always needs a new baseline.
finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 322kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 322kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 322kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 198kUpdated Claude CodeCodexVerifiedInstall