Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

JIT: Compute fgCalledCount earlier #110693

Closed
wants to merge 1 commit into from

Conversation

jakobbotsch
Copy link
Member

Move the computation of fgCalledCount back to happen after profile incorporation, when profile weights are consistent and we can be more sure that querying the entry BB's weights is meaningful.

@dotnet-issue-labeler dotnet-issue-labeler bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Dec 13, 2024
Copy link
Contributor

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Copy link
Contributor

Draft Pull Request was automatically closed for 30 days of inactivity. Please let us know if you'd like to reopen it.

amanasifkhalid added a commit that referenced this pull request Feb 1, 2025
Part of #107749. Follow-up to #111971 and #110693. For methods without profile data, ensure the default call count is available throughout compilation (this had no diffs for me locally). For methods with profile data, compute the call count after synthesis runs to ensure it is available early, and reasonably accurate.

I'm only seeing diffs in OSR methods locally, due to the logic in `fgFixEntryFlowForOSR` (which runs right after profile incorporation) no longer affecting `fgCalledCount`. This method guesses that the loop iterates about 100x the method call count, and scales the method entry block's weight down accordingly. This gives the impression later on that `fgCalledCount` is much lower than what we calculated using `fgEntryBB`.

The actual diffs seem to manifest largely in LSRA, which uses `fgCalledCount` to normalize block weights, though there are a few other phases that use `BasicBlock::getBBWeight` in lieu of the raw weight as well. I think we ought to consolidate our block weight strategy at some point, especially if we have newfound faith in `fgCalledCount`. For example, instead of this check in if conversion:
```
if (m_startBlock->getBBWeight(m_comp) > BB_UNITY_WEIGHT * 1.05)
```

Perhaps we could do:
```
if (m_startBlock->bbWeight > fgCalledCount * 1.05)
```

But that's for another PR.
@github-actions github-actions bot locked and limited conversation to collaborators Feb 12, 2025
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant