Skip to content

docs: Record the v1 performance baseline for drawing layers - #1203

Merged
omesser merged 2 commits into
mainfrom
cursor/performance-baseline-v1-c3e9
Oct 1, 2026
Merged

omesser merged 2 commits into
mainfrom
cursor/performance-baseline-v1-c3e9

Conversation

@omesser

@omesser omesser commented Sep 30, 2026 •

Copy link
Copy Markdown
Owner

Why

#423 is done when the child baselines sit in one document and the top optimization targets are ranked from those measurements. This pull request fills that document from the merged captures and from release runs on this VM.

What changed

docs/research/performance-baseline-v1.md records the rollup. The rank is milliseconds of one core per second.

  • Software paint of a moving overlay. The heavier WebKit process held 83.8% of one core across a 60 s walk and climb. Xtigervnc held 54.5% on the scripted 15 s walk.
  • Click-through mask rebuild while the pointer is on a moving sprite. 400 ms/s on the X11 desktop, 646 ms/s on the Windows walk subset.
  • macOS WindowSource while riding with many windows. 112 ms/s at 156 windows.

scripts/bench-rss-linux.sh, scripts/bench-gpu-compositing-linux.sh, and scripts/bench-mask-rebuild-x11.sh take --bin. scripts/bench-frame-cadence-macos.sh reads a Linux machine header and exports XAUTHORITY before it replaces HOME.

Chat on this VM never logged Summon. GPU% is unavailable because the renderer is llvmpipe and the VM has no DRM device. C-state residency is unavailable. heaptrack was not installed. Wayland, a second compositor, and a second display were not running.

How to verify

pre-commit run --files passed on the document and the bench scripts. The Linux numbers come from target/release/fidget on DISPLAY=:1.

  • scripts/bench-gpu-compositing-linux.sh matrix --seconds 15 --bin target/release/fidget
  • scripts/bench-rss-linux.sh for one fidget and for four, still Character, --settle 30 --seconds 45
  • scripts/bench-frame-cadence-macos.sh with FIDGET_BENCH_GREEN_LIGHT=1 for matrix, walking, and load
  • a 60 s /proc sample of utime, stime, and voluntary_ctxt_switches for idle, walking, and hidden

The reproduce block in the document is the command list that produced the tables.

Closes #423

Refs #432, #425, #428, #424, #426

— Cursor agent (Coder), on @omesser's behalf.

Open in Web Open in Cursor 

cursoragent and others added 2 commits September 30, 2026 22:40
Co-authored-by: Oded Messer <odedmesser@gmail.com>
Co-authored-by: Oded Messer <odedmesser@gmail.com>
@cursor
cursor Bot marked this pull request as ready for review October 1, 2026 01:07

omesser commented Oct 1, 2026

Copy link
Copy Markdown
Owner Author

Soft CLEARED — pass / merge-ready pending Oded’s yes

Tip effb3b4f7513549cabb91edfc0324bc0b333d4b7 (cursor/performance-baseline-v1-c3e9). This is a pass on Soft-gate: #423’s v1 acceptance is met (one rollup + ranked top-3 from measurements). Not an Approve; merge still needs your yes.

Soft ledger

  • (a) perf: Establish performance baseline for drawing layers + engine #423 acceptance — Pass. docs/research/performance-baseline-v1.md is the rollup; “What to optimize first” ranks three targets from measured ms/s; child baselines table + Refs; body Closes #423.
  • (b) Numbers — Pass. Ranked top-3 match the PR body: software paint / heavier WebKit 83.8%; mask rebuild 400 / 646 ms/s; macOS WindowSource 112 ms/s @ 156 windows. Limits disclosed (llvmpipe, no DRM GPU%, no C-state, heaptrack missing, no Wayland / 2nd display, Chat Summon not logged).
  • (c) Bench harness — Pass. --bin on linux RSS / GPU / mask scripts; cadence script exports XAUTHORITY before replacing HOME. Docs + scripts only — no silent product behavior change.
  • (d) Merge hygiene — Pass. Mergeable (clean); Closes #423; scoped to docs + bench scripts (5 files).
  • (e) CI on tip — Pass. Re-fetched: checks (ubuntu-latest), checks (macos-latest), windows, build all success; deploy skipped.

Hard Holds

Docs/scripts only — no Engine/UI product regression expected; #423 acceptance is met.

— Cursor agent (Architect), on @omesser's behalf.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

perf: Establish performance baseline for drawing layers + engine

2 participants