Skip to content

chore: update the compiler performance numbers on the front page - #228

Merged
magnus-madsen merged 1 commit into
masterfrom
chore/update-compiler-performance-numbers
Nov 15, 2026
Merged

magnus-madsen merged 1 commit into
masterfrom
chore/update-compiler-performance-numbers

Conversation

@magnus-madsen

Copy link
Copy Markdown
Member

Summary

Refreshes the compiler performance claims on the front page.

  • Speed-up plot: replaces speedupWithPar.png (from November 2023) with one regenerated from the current compiler. The old plot listed phases that no longer exist (Lowering, MonoDefs, MonoTypes, Tailrec, VarOffsets, JvmBackend).
  • Speed-up text: "5x – 7x on multi-core machines" becomes "up to 7.9x per phase and 4.9x for the compiler as a whole", and the machine is named: 24 threads on an AMD Ryzen 9 5900X with a 12-core CPU.
  • Raw numbers: now measured on the same Ryzen 9 5900X on OpenJDK 27 instead of the Apple M2 Pro. Frontend is 303,740 lines/sec and the entire compiler is 100,645 lines/sec, with the frontend row listed first.
  • Reproduce line: shortened to a single command, java -Xms8g -Xmx8g -jar flix.jar Xperf --par --n 500.

Where the numbers come from

  • Per-phase speed-ups are from an Xperf run comparing 1 and 24 threads. The 4.9x overall figure is the ratio of that run's median throughput on 24 threads (98,682 lines/sec) to 1 thread (20,183 lines/sec).
  • The frontend figure is the best of 500 iterations with -Xms8g -Xmx8g.
  • The entire-compiler figure is the best of 50 iterations from the speed-up run, not from the 500-iteration command shown on the page.

Test plan

  • npx astro build succeeds.
  • Checked the rendered HTML of the two sections for the new text and for spacing around inline elements.

🤖 Generated with Claude Code

Replace the parallel speed-up plot from 2023 with one regenerated from the
current compiler, and bring the surrounding text in line with it: the
machine is an AMD Ryzen 9 5900X (12 cores, 24 threads), and the speed-ups
are up to 7.9x per phase and 4.9x for the compiler as a whole.

Update the raw throughput numbers to the same machine on OpenJDK 27, list
the frontend first, and shorten the reproduce line to a single command.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@magnus-madsen
magnus-madsen merged commit 655c943 into master Nov 15, 2026
2 checks passed
@magnus-madsen
magnus-madsen deleted the chore/update-compiler-performance-numbers branch November 15, 2026 09:19
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.

1 participant