[TEST] Add ASan/LeakSanitizer harness and fix memory bugs it found by Niz13 · Pull Request #7831 · intel/intel-xpu-backend-for-triton

Niz13 · 2026-08-24T12:51:23Z

Implements the host-side memory-leak check requested in #5029.

Harness

scripts/test-triton.sh --asan runs the host-side C++ (ctest) and MLIR lit suites under LeakSanitizer. Sets ASAN_OPTIONS=allow_user_poisoning=0 to neutralize a false use-after-poison caused by linking ASan-instrumented Triton against the non-instrumented prebuilt LLVM (does not weaken leak/redzone detection), and guards that triton-opt is actually ASan-instrumented.
scripts/asan/lsan.supp suppresses only benign LLVM/MLIR globals — nothing under third_party/intel.

Bugs found & fixed

Heap-use-after-free in StrideVersioning — the pass called forOp.erase() inside a pre-order walk(), so the walker then dereferenced the freed op. Fixed with collect-then-mutate: gather candidate loops during the walk, version them afterwards (safe — only non-nested top-level loops are candidates).
Six orphaned-ModuleOp leaks in Analysis unit-test fixtures — ModuleOp::create() returned raw and was never freed. Fixed by owning modules in a SmallVector<OwningOpRef<ModuleOp>> member declared after the MLIRContext.

Result (under LeakSanitizer)

Phase	Before	After
`ctest`	116 failing (leaks)	453/453, 0 leaks
`lit`	1 failing (UAF)	446/446, 0 UAF

🤖 Generated with Claude Code

Implements the host-side memory-leak check requested in issue #5029. - scripts/test-triton.sh gains `--asan`: runs the host-side ctest and lit suites under LeakSanitizer. Sets allow_user_poisoning=0 to neutralize the false use-after-poison from linking instrumented Triton against the non-instrumented prebuilt LLVM, guards that triton-opt is ASan-instrumented, and wires in the suppressions file. - scripts/asan/lsan.supp suppresses only benign LLVM/MLIR globals (never anything under third_party/intel). Fixes found by the harness: - StrideVersioning erased a loop from inside a pre-order walk(), so the walker dereferenced the freed op (heap-use-after-free). Switch to collect-then-mutate: gather candidate loops during the walk, version them afterwards. - Six Analysis unit-test fixtures returned a raw ModuleOp from ModuleOp::create() and never freed it. Own each module in a SmallVector<OwningOpRef<ModuleOp>> member declared after the MLIRContext. Host-side ctest (453) and lit (446) run clean under LeakSanitizer. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Address review feedback: instead of a SmallVector<OwningOpRef<ModuleOp>> member managing module lifetimes, the buildModule()/createModule() fixture helpers now return OwningOpRef<ModuleOp> so each test owns its module locally (as in OptimizePartitionWarps.cpp). Applied across all six Analysis fixtures. Still leak-free under LeakSanitizer (118 host-side tests pass). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

anmyachev · 2026-08-31T14:23:00Z

 TEST_F(AliasAnalysisTest, TwoLoadsSameArgDifferentOffsets) {
-  auto module = createModule();
+  OwningOpRef<ModuleOp> moduleRef = createModule();
+  ModuleOp module = *moduleRef;


Let's use module directly from OwningOpRef without unpacking into a separate variable. Here and elsewhere.
This way should be safer.

Otherwise LGTM!

Done, but please take a look at DPAS.h, It had to be changed for this to be uniform across.
if you prefer not to change, I can make exceptions for when it doesn't work.

Will const ModuleOp &mod work?

Copilot

Pull request overview

Adds an AddressSanitizer/LeakSanitizer testing mode to the repo’s test runner and fixes memory-safety issues uncovered by sanitizer runs (including a walk-time erase UAF in a Triton/MLIR transform and MLIR ModuleOp ownership leaks in Intel backend unit tests).

Changes:

Add scripts/test-triton.sh --asan to run host-only ctest + lit under ASan/LSan with a suppression file and instrumentation guardrails.
Fix a heap-use-after-free in TritonIntelStrideVersioning by collecting candidate scf.for loops first and versioning them after the walk.
Fix ModuleOp::create() ownership leaks in multiple Analysis unit tests by returning/holding OwningOpRef<ModuleOp>.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
third_party/intel/unittest/Analysis/TemporalReuseAnalysisTest.cpp	Switch module construction to `OwningOpRef<ModuleOp>` to prevent leaked `ModuleOp`s in test fixtures.
third_party/intel/unittest/Analysis/StrideInfoTest.cpp	Same `OwningOpRef<ModuleOp>` lifetime ownership fix for stride-analysis unit tests.
third_party/intel/unittest/Analysis/SpatialReuseAnalysisTest.cpp	Same `OwningOpRef<ModuleOp>` lifetime ownership fix for spatial-reuse unit tests.
third_party/intel/unittest/Analysis/ReuseAnalysisTest.cpp	Same `OwningOpRef<ModuleOp>` lifetime ownership fix for reuse-analysis unit tests.
third_party/intel/unittest/Analysis/DPASAnalysisTest.cpp	Same `OwningOpRef<ModuleOp>` lifetime ownership fix for DPAS-analysis unit tests.
third_party/intel/unittest/Analysis/AliasAnalysisTest.cpp	Same `OwningOpRef<ModuleOp>` lifetime ownership fix for alias-analysis unit tests.
third_party/intel/lib/Dialect/Triton/Transforms/StrideVersioning.cpp	Avoid IR mutation during a pre-order walk by collecting loops first, then versioning, preventing walk iterator UAF.
scripts/test-triton.sh	Add `--asan` option to run only host-side unit + LIT tests under ASan/LSan with env configuration and instrumentation checks.
scripts/asan/lsan.supp	Add LSan suppression list for known benign LLVM/MLIR process-lifetime allocations.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

whitneywhtsang · 2026-08-31T23:18:29Z

What's the plan to prevent regressions going forward? Without regular execution, new code could reintroduce leaks/UAFs and we wouldn't catch them until someone manually runs --asan.

januszjah · 2026-09-01T07:23:03Z

What's the plan to prevent regressions going forward? Without regular execution, new code could reintroduce leaks/UAFs and we wouldn't catch them until someone manually runs --asan.

@Niz13 we may create new issue/pr as follow-up to incorporate new feature in CI runs

… variable Address review feedback: name the OwningOpRef `module` and dereference it as `*module` at use sites, rather than unpacking into a second `ModuleOp` variable. Applied across all six Analysis test fixtures. Since OwningOpRef::operator* yields an rvalue ModuleOp, relax DPASAnalysisFactory::createDPASAnalysis to take ModuleOp by value (it never mutates the module); the handle is lightweight and existing lvalue callers still bind. Also compose ASAN_OPTIONS/LSAN_OPTIONS so the harness-critical settings are appended last (last-wins parser) and avoid a stray trailing colon when the inherited variable is empty. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Niz13 mentioned this pull request Aug 24, 2026

Check memory leaks with address sanitizer #5029

Open

anmyachev reviewed Aug 24, 2026

View reviewed changes

Comment thread third_party/intel/unittest/Analysis/TemporalReuseAnalysisTest.cpp Outdated

vlad-penkin linked an issue Aug 31, 2026 that may be closed by this pull request

Check memory leaks with address sanitizer #5029

Open

Niz13 and others added 2 commits August 31, 2026 16:45

Merge branch 'main' into niz/asan-leak-check-5029

a65e4e7

anmyachev reviewed Aug 31, 2026

View reviewed changes

etiotto requested a lite review from Copilot August 31, 2026 15:01

Copilot started reviewing on behalf of etiotto August 31, 2026 15:02 View session

etiotto requested a review from whitneywhtsang August 31, 2026 15:04

Copilot AI reviewed Aug 31, 2026

View reviewed changes

Comment thread scripts/test-triton.sh Outdated

Navigation Menu

Uh oh!

[TEST] Add ASan/LeakSanitizer harness and fix memory bugs it found - #7831

[TEST] Add ASan/LeakSanitizer harness and fix memory bugs it found#7831
Niz13 wants to merge 4 commits into
mainfrom
niz/asan-leak-check-5029

Niz13 commented Aug 24, 2026

Uh oh!

Uh oh!

anmyachev Aug 31, 2026

Uh oh!

Niz13 Sep 1, 2026

Uh oh!

anmyachev Sep 1, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

whitneywhtsang commented Aug 31, 2026

Uh oh!

januszjah commented Sep 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Uh oh!

Conversation

Niz13 commented Aug 24, 2026

Harness

Bugs found & fixed

Result (under LeakSanitizer)

Uh oh!

Uh oh!

anmyachev Aug 31, 2026

Choose a reason for hiding this comment

Uh oh!

Niz13 Sep 1, 2026

Choose a reason for hiding this comment

Uh oh!

anmyachev Sep 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

whitneywhtsang commented Aug 31, 2026

Uh oh!

januszjah commented Sep 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants