imla-perfetto-feedback

name	imla_perfetto_feedback
description	Capture and analyze Imla Android renderer smoke traces with the repo-local Perfetto config, SQL queries, and Android Performance Analyzer workflow.

Imla Perfetto Feedback

Use this skill before and after renderer, blur, GL threading, capture, or Compose modifier changes that need runtime feedback beyond screenshots.

Tool

The repo-local tool is:

tools/imla-perfetto-feedback

It wraps the checked-in Perfetto smoke config, named SQL queries, screenshot capture, and optional Android Performance Analyzer opening. The extensionless shell wrapper is the stable entrypoint; the .py file is the implementation detail.

Commands

Analyze an existing trace:

tools/imla-perfetto-feedback analyze diagnostics/apa/traces/imla-smoke-2026-05-23.perfetto-trace

Capture a new smoke trace on the active emulator, exercise the demo, pull a screenshot, analyze it, and open it in APA:

tools/imla-perfetto-feedback smoke --device emulator-5580 --open-apa

Capture without scripted input:

tools/imla-perfetto-feedback capture --device emulator-5580 --no-exercise

Open a trace in Android Performance Analyzer:

tools/imla-perfetto-feedback open diagnostics/apa/traces/<trace>.perfetto-trace

Trace Anchors

Start investigation from these labels when present:

• ImlaSceneCoordinator#commitRootCapture
• ImlaSceneCoordinator#refreshSlotGeometry
• RenderableRootLayer#captureLayer
• GraphicsLayerTexture#captureGraphicsLayer
• GraphicsLayerTexture#captureCanvas
• SceneLayerRepository#captureSlotContent
• SceneMaskRepository#captureSlotMasks
• SceneGlRenderer#render
• Renderer2D#endScene
• Renderer2D#uploadVertices
• vboSetData
• glBufferData

Key threads are the main app thread, Android RenderThread, and Imla GLUiLayerRender.

Workflow

1. Build and install the app if code changed: ./gradlew -q :app:assembleDebug then install or launch via ADB.
2. Run tools/imla-perfetto-feedback smoke --device <serial>.
3. Read the generated analysis directory next to the trace before editing performance-sensitive code.
4. Open the trace in APA when thread timing, frame timeline, or visual trace navigation is needed.
5. Use the installed official skills perfetto-trace-analysis and perfetto-sql for deeper AI-assisted trace questions; this repo tool provides the repeatable capture/query baseline.

The scene counter summary is written to 60_scene_counters.txt when scene counters are present.