Repository navigation
Conversation
…r boot The interpreter's decode cache keeps 64 physical pages, direct-mapped on a hash of the page number. Which text pages collide depends on where an SMP IRIX boot placed the program's, libc's and its own text, so the same build ran a guest awk 64-67 s in some Tezro boots and 74-110 s in others, with the re-decode share at 3-8% of host samples in the slow ones. With 256 pages the variance went away. The note records the measurement and the rule: compare interpreter builds over several boots and check the decode share in a profile before calling a difference a regression. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
clanker: note that decode-cache conflicts change interpreter speed per boot
The interpreter's decode cache keeps 64 physical pages, direct-mapped on
a hash of the page number. Which text pages collide depends on where an
SMP IRIX boot placed the program's, libc's and its own text, so the same
build ran a guest awk 64-67 s in some Tezro boots and 74-110 s in others,
with the re-decode share at 3-8% of host samples in the slow ones. With
256 pages the variance went away. The note records the measurement and
the rule: compare interpreter builds over several boots and check the
decode share in a profile before calling a difference a regression.
Docs only: an agent note in
clanker/and itsINDEX.mdline. No code change.🤖 Generated with Claude Code