[SYCL RTC] Workaround data race related to the VFS containing toolchain #20360
+62
−5
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
aelovikov-intel
requested review from
a team and
cperkinsintel
as code owners
aelovikov-intel
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
aelovikov-intel
added
the
run-perf-tests
|
Unfortunately, hits compilation times pretty hard... Before (locally):
After (same system):
Despite that, I think we still should proceed with this PR because stability is more important. I'll just have to follow-up on the performance later. |
aelovikov-intel
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
aelovikov-intel
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
This regressed due to intel#19924 but, apparently, we didn't have proper tests in place. I'm not sure what's causing this exactly, but having each compilation create its unique `ToolchainFS` instead of all of them using the same `llvm::IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> SYCLToolchain::ToolchainFS` somehow results in the test (added in this PR) passing consistently.
aelovikov-intel
force-pushed
the
rtc-race
branch
from
5340413 to
3c8a50b
Compare
aelovikov-intel
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
aelovikov-intel
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
aelovikov-intel
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
cperkinsintel
approved these changes
Oct 15, 2025
On the test I'm adding here and on other tests I'm adding to verify future |
aelovikov-intel
added a commit
to aelovikov-intel/llvm
that referenced
this pull request
Oct 16, 2025
intel#19924 essentially made it `static` but that caused data races that were later fixed by intel#20360 changing each use of it to re-create this in-memory FS (essentially, "removing" `static`), incurring significant performance costs. This PR addresses the issue by "adding" `thread_local` instead of "removing" `static` allowing us to have both no crashes due to data races and minimal overhead. No tests added as the one from intel#20360 is verifying this.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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 regressed due to #19924 but, apparently, we didn't have proper tests in place. I'm not sure what's causing this exactly, but having each compilation create its unique
ToolchainFSinstead of all of them using the samellvm::IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> SYCLToolchain::ToolchainFSsomehow results in the test (added in this PR) passing consistently.