Skip to content

Use instead of restoring side table for Loop#705

Merged
ia0 merged 6 commits into
google:dev/fast-interpfrom
zhouwfang:use-side-table-for-loop-dup
Dec 13, 2024
Merged

Use instead of restoring side table for Loop#705
ia0 merged 6 commits into
google:dev/fast-interpfrom
zhouwfang:use-side-table-for-loop-dup

Conversation

@zhouwfang

Copy link
Copy Markdown
Contributor

On my Linux Docker container:

CoreMark results of this PR: 37.991516 (in 18.695s), 36.7287 (in 19.258s)

CoreMark results of #690: 37.101162 (in 19.133s), 37.39716 (in 18.996s)

Look very similar. Not sure about nordic.

CoreMark results of main: 28.791477 (in 18.09s), 28.814291 (in 17.734s)

So there does seem some minor performance improvement with delta_ip and delta_stp.

#46

Comment thread crates/interpreter/src/exec.rs Outdated
Comment thread crates/interpreter/src/valid.rs Outdated
Comment thread crates/interpreter/src/valid.rs Outdated
@ia0

ia0 commented Dec 12, 2024

Copy link
Copy Markdown
Member

On nordic I get:

  • This pull request (with the suggested changes): 0.13723445 (in 146.037s) and 6660 bytes of RAM
  • This development branch: 0.13704263 (in 146.247s) and 7588 of RAM
  • The main branch: 0.09060761 (in 221.04s) and 5416 of RAM

So we should merge this PR, not only because it uses less RAM, but also because it simplifies the code.

@zhouwfang
zhouwfang requested a review from ia0 December 12, 2024 18:48
@zhouwfang

Copy link
Copy Markdown
Contributor Author

On nordic I get:

  • This pull request (with the suggested changes): 0.13723445 (in 146.037s) and 6660 bytes of RAM
  • This development branch: 0.13704263 (in 146.247s) and 7588 of RAM
  • The main branch: 0.09060761 (in 221.04s) and 5416 of RAM

So we should merge this PR, not only because it uses less RAM, but also because it simplifies the code.

How do you measure RAM?

@ia0
ia0 merged commit de94f8c into google:dev/fast-interp Dec 13, 2024
@ia0

ia0 commented Dec 13, 2024

Copy link
Copy Markdown
Member

How do you measure RAM?

This is specific to Nordic:

println!("Peak RAM usage: {}", USAGE.load(Ordering::Relaxed));

@zhouwfang
zhouwfang deleted the use-side-table-for-loop-dup branch December 13, 2024 16:31
ia0 pushed a commit that referenced this pull request Dec 24, 2024
It improves the performance by removing the costly
`last_frame_values_cnt()`.

On linux, `CoreMark result: 39.795715 (in 17.853s)`. cf. CoreMark result
was about 37 in 19s in #705.

#46

---------

Co-authored-by: Zhou Fang <[email protected]>
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.

2 participants