ContextQMD
Back to Chromium

Chrome & V8 Autonomous Performance Optimization Loop

agents/skills/chrome-performance-optimizer/SKILL.md

154.0.8011.26.5 KB
Original Source

Chrome & V8 Autonomous Performance Optimization Loop

This skill provides an autonomous agent loop that ingests performance profiles (from web pprof links, profile IDs, Sagacity MCP, or Crossbench logs), designs macro-optimizations in Blink or V8, validates correctness locally, tests on Pinpoint hardware bots, and manages CL lifecycles based on statistical confidence.

🔁 The Optimization Loop Workflow

mermaid
graph TD
    A[1. Ingest Profile: pprof link / ID / Sagacity MCP] --> B[2. Formulate Macro Hypothesis]
    B --> C[3. Implement on Dedicated Branch]
    C --> D[4. Verify Locally: Tests & Crossbench]
    D -->|Fail| C
    D -->|Pass| E[5. Upload CL to Gerrit]
    E --> F[6. Run Pinpoint on M1: pp c -c m1 -t sp3 -r 150]
    F --> G[7. Poll & Evaluate Results: pp s]
    G -->|Stat-Significant Improvement| H[8. Propose CL & Update Benchmark Report]
    G -->|No Improvement or Regressed| I[9. Abandon CL: git cl abandon]
    H --> A
    I --> A

Step 1: Ingest & Analyze Performance Profiles

You can provide one or multiple profile sources:

  • Web pprof link(s): pprof/?id=XYZ or https://pprof.corp.google.com/?id=XYZ
  • Native pprof ID(s): id:XYZ or raw ID XYZ
  • Sagacity MCP Tools: fetch_uploaded_profile(profileKey="XYZ")
  • Local Benchmark Profiles: Crossbench CSV, Linux perf, or v8.log

Profile Analysis Commands:

  1. Top Cumulative Call Stacks (identify caller subtrees):

    bash
    vpython3 agents/skills/chrome-performance-optimizer/scripts/analyze_profile.py "pprof/?id=XYZ" --mode=cum --nodecount=30

  2. Top Flat Functions (identify hot leaf loops):

    bash
    vpython3 agents/skills/chrome-performance-optimizer/scripts/analyze_profile.py "pprof/?id=XYZ" --mode=flat --nodecount=30

  3. Inspect Callers & Callees for a Specific Symbol:

    bash
    vpython3 agents/skills/chrome-performance-optimizer/scripts/analyze_profile.py "pprof/?id=XYZ" --mode=peek --symbol="*HasOwnProperty*"

  4. Compare Two Profiles (Diff Mode):

    bash
    vpython3 agents/skills/chrome-performance-optimizer/scripts/analyze_profile.py "pprof/?id=EXP_ID" --base="pprof/?id=BASE_ID" --mode=cum

  5. Classify the Bottleneck Pattern: Consult Macro-Optimization Patterns for proven solutions:

    • DOM / Layout: FlatTree iteration, slot distribution, style recalc tree walks.
    • Canvas 2D: API call dispatch overhead, SkPathBuilder allocations, disconnected strokes.
    • V8 ICs: StubCache evictions, megamorphic dispatch thrashing.
    • Compiler: Maglev/Turboshaft loop unrolling for dense high-order array callbacks.
    • consider other places based on the profile provided.

Step 2: Formulate Hypothesis & Create Isolated Branch

  1. Create a dedicated branch for the optimization:
    bash
    # For Blink / Chromium root changes:
git checkout -b perf_<feature_name> origin/main

# For V8 engine submodule changes:
git -C v8 checkout -b perf_<feature_name> origin/main
  2. Implement the macro-optimization cleanly, adhering to codebase conventions.

Step 3: Local Verification & Correctness Testing

Always verify correctness before uploading to avoid wasting Pinpoint bot resources:

  1. Unit Tests:

    bash
    # For Blink changes:
autoninja -C out/release blink_unittests
./out/release/blink_unittests --gtest_filter="<RelevantTestPattern>"

# For V8 changes:
autoninja -C out/release v8:d8
./out/release/d8 v8/test/mjsunit/mjsunit.js <path_to_test.js>

  2. Web Tests (Layout / Rendering / Canvas):

    bash
    autoninja -C out/release content_shell
./third_party/blink/tools/run_web_tests.py -t release <path_to_web_test.html>

  3. Crossbench Benchmark Smoke Test:

    bash
    autoninja -C out/release chrome chromedriver
./third_party/crossbench/cb.py speedometer_3.1 --browser=out/release/chrome --driver-path=out/release/chromedriver --stories=<TargetStory> --headless

Step 4: Submit CL to Gerrit

  1. Commit all modified files with descriptive rationale:
    bash
    git commit -m "[<Subsystem>] <Title>

<Detailed architectural explanation and expected benchmark impact>

TAG=agy
CONV=<conversation_id>"
  2. Upload the CL to Gerrit:
    bash
    git cl upload -m "Performance optimization for Speedometer 3" --cq-dry-run
  3. Retrieve the Gerrit Issue ID:
    bash
    git cl issue

Step 5: Run Pinpoint Try Job on M1 Hardware

Launch a 150-iteration try job on Apple Silicon M1 bots:

bash
pp c -c m1 -t sp3 -r 150
  • -c m1: Target M1 hardware bot.
  • -t sp3: Target Speedometer 3 benchmark template.
  • -r 150: 150 repetitions per variant for high statistical confidence.

Step 6: Evaluate Results & Autonomous Decision

  1. Inspect results once the job completes:

    bash
    vpython3 agents/skills/chrome-performance-optimizer/scripts/pinpoint_evaluator.py --action evaluate --job-id <JOB_ID>

    Or directly view the comparison table:

    bash
    pp s <JOB_ID>

  2. Decision Rules:

    • ✅ Statistically Significant Improvement ($p < 0.05$):
      • Keep the CL.
      • Add the Pinpoint benchmark results to the CL description:
        bash
        git cl upload -m "Add Pinpoint M1 benchmark results (+X.X% improvement)"
      • Propose the change to the user and reviewers.
    • ❌ Neutral or Regressed:
      • Abandon the CL immediately:
        bash
        git cl abandon -m "Pinpoint try job (150 iterations on M1) showed no statistically significant speedup."
      • Switch back to origin/main and iterate to the next candidate profile/bottleneck.

References & Utilities