gh-119258: Eliminate Type Guards in Tier 2 Optimizer with Watcher #119365
Conversation
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
saulshanabrook
requested review from
Fidget-Spinner,
markshannon and
gvanrossum
as code owners
|
Looks like a print statement was left in there somewhere. I can try rerunning the benchmarks once you remove it! |
|
Hey @brandtbucher I may have removed it yesterday but just after you started the benchmark. I looked through the current diff and didn't see anymore. Thanks! Sorry about that. |
to match `typ_version_tag` suggested by @brandtbucher
suggested by @brandtbucher
suggested by @brandtbucher
suggested by @brandtbucher
As per comment by @brandtbucher: > One that performs the mutation partway through execution like this. However, maybe change Foo.attr to 2 so we know it keeps running, but can see the effect of the changed value? I'd also make xxx just another class Bar, since I'm worried that our function watcher branch might invalidate this or something weird since it's just a function.
|
Thank you @brandtbucher for the review! I resolved most of the comments and left a few responses for ones I was unclear about. |
|
(You can’t assert it like that, since everything inside the assert is removed in debug builds.) |
|
Heh, now you’ll get an unused variable warning for It’s probably fine to just omit the assert. If you want to keep it, adding a |
|
I am not sure if the JIT aarch64 Linux failure is real or spurious: |
|
JIT aarch64 errors happen on the other PRs as well. So it's not this PR. Will merge this PR in 24 hours or so. |
|
Sweet! Were the benchmarks ever re-run on comparing this, or is it not worth it b/c it's probably similar the first time? |
The benchmarks the second time reported no speedup. But the noise range of a bench run is like 1%, so for all we know this couldve sped things up 0.5% but we cant tell. |
|
Congrats @saulshanabrook ! |
Fidget-Spinner
added a commit
that referenced
this pull request
Jun 20, 2024
noahbkim
pushed a commit
to hudson-trading/cpython
that referenced
this pull request
Jul 11, 2024
…er (pythonGH-119365) Co-authored-by: parmeggiani <[email protected]> Co-authored-by: dpdani <[email protected]> Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com> Co-authored-by: Brandt Bucher <[email protected]> Co-authored-by: Ken Jin <[email protected]>
estyxx
pushed a commit
to estyxx/cpython
that referenced
this pull request
Jul 17, 2024
…er (pythonGH-119365) Co-authored-by: parmeggiani <[email protected]> Co-authored-by: dpdani <[email protected]> Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com> Co-authored-by: Brandt Bucher <[email protected]> Co-authored-by: Ken Jin <[email protected]>
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.
This PR replaces #119259 to more accurately remove
_GUARD_TYPE_VERSION.With help and guidance from @brandtbucher, I added a global type version watcher to the interpreter, so that we can look at types by their versions.
Then, whenever we see a
_GUARD_TYPE_VERSIONwe can add a watcher to that type, and if it isn't changed, the next time we see that type version we can omit the guard.We should see if the performance is improved compared to the other PR, or at least see if it removes more guard type versions.