[DevTSAN] Sync thread clock to global state before kernel exit #18529
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
We need to sync each thread's clock back to global state before the thread exit to avoid false alarms in some reduction cases.
There was a problem hiding this comment.
Pull Request Overview
This PR ensures that each thread synchronizes its clock back to the global state before kernel exit to avoid false alarms, with changes applied to instrumentation logic and sanitizer driver arguments.
- Updated the SPIR-V instrumentation routine to properly handle function entry/exit instrumentation.
- Introduced a new __tsan_func_exit() in the device sanitizer runtime to sync clocks.
- Adjusted sanitizer argument handling for device offloading by removing tsan function entry/exit flags.
Reviewed Changes
Copilot reviewed 3 out of 8 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| llvm/lib/Transforms/Instrumentation/ThreadSanitizer.cpp | Updated API call and instrumentation logic to force exit instrumentation based on SPIR-V conditions. |
| libdevice/sanitizer/tsan_rtl.cpp | Added __tsan_func_exit() to sync thread clocks to global state. |
| clang/lib/Driver/SanitizerArgs.cpp | Removed tsan function entry/exit flag for device offloading. |
Files not reviewed (5)
- llvm/test/Instrumentation/ThreadSanitizer/SPIRV/basic.ll: Language not supported
- llvm/test/Instrumentation/ThreadSanitizer/SPIRV/cleanup_private_shadow.ll: Language not supported
- llvm/test/Instrumentation/ThreadSanitizer/SPIRV/instrument_barrier.ll: Language not supported
- llvm/test/Instrumentation/ThreadSanitizer/SPIRV/instrument_device_global.ll: Language not supported
- llvm/test/Instrumentation/ThreadSanitizer/SPIRV/kernel_metadata.ll: Language not supported
Comments suppressed due to low confidence (2)
clang/lib/Driver/SanitizerArgs.cpp:1282
- Removal of tsan function entry/exit instrumentation flags for device offloading should be double-checked to ensure it aligns with the intended behavior of synchronizing thread clocks before exit.
CmdArgs.push_back("-tsan-instrument-func-entry-exit=0");
libdevice/sanitizer/tsan_rtl.cpp:470
- Ensure that using kThreadSlotCount as an index for the global clock array is valid and within expected bounds, as this update affects thread synchronization logic.
TsanLaunchInfo->Clock[kThreadSlotCount].clk_[sid] =
yingcong-wu
approved these changes
May 19, 2025
| // The tsan function entry/exit builtins are used to record stack | ||
| // position, we don't need them in device offloading. | ||
| CmdArgs.push_back("-mllvm"); | ||
| CmdArgs.push_back("-tsan-instrument-func-entry-exit=0"); |
There was a problem hiding this comment.
Kindly remind @jinge90 for this change. Please remember to update the fortran driver.
There was a problem hiding this comment.
OK, thanks for the reminder.
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.
We need to sync each thread's clock back to global state before the thread exit to avoid false alarms in some reduction cases.