Skip to content

[PatternMatch] Fix issue of stale reference in new m_{I,F,}Cmp matchers #98866

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

Merged
merged 1 commit into from
Jul 15, 2024

Conversation

goldsteinn
Copy link
Contributor

The new matchers don't output pred. Previously we where just creating
a value on the stack and using it as a dummy output for the matchers,
but this results in a stale reference upon return.

To fix, this patch changes the output variable to a pointer, and
passes in nullptr for the matchers that don't output pred.

…hers

The new matchers don't output pred. Previously we where just creating
a value on the stack and using it as a dummy output for the matchers,
but this results in a stale reference upon return.

To fix, this patch changes the output variable to a pointer, and
passes in `nullptr` for the matchers that don't output `pred.`
@goldsteinn goldsteinn requested review from vitalybuka and nikic July 15, 2024 07:29
@llvmbot
Copy link
Member

llvmbot commented Jul 15, 2024

@llvm/pr-subscribers-llvm-ir

Author: None (goldsteinn)

Changes

The new matchers don't output pred. Previously we where just creating
a value on the stack and using it as a dummy output for the matchers,
but this results in a stale reference upon return.

To fix, this patch changes the output variable to a pointer, and
passes in nullptr for the matchers that don't output pred.


Full diff: https://github.com/llvm/llvm-project/pull/98866.diff

2 Files Affected:

  • (modified) llvm/include/llvm/IR/PatternMatch.h (+11-10)
  • (modified) llvm/unittests/IR/PatternMatch.cpp (+2-2)
diff --git a/llvm/include/llvm/IR/PatternMatch.h b/llvm/include/llvm/IR/PatternMatch.h
index 8ae47fb556b25..bea1ad97ea09c 100644
--- a/llvm/include/llvm/IR/PatternMatch.h
+++ b/llvm/include/llvm/IR/PatternMatch.h
@@ -1550,23 +1550,27 @@ template <typename T> inline Exact_match<T> m_Exact(const T &SubPattern) {
 template <typename LHS_t, typename RHS_t, typename Class, typename PredicateTy,
           bool Commutable = false>
 struct CmpClass_match {
-  PredicateTy &Predicate;
+  PredicateTy *Predicate;
   LHS_t L;
   RHS_t R;
 
   // The evaluation order is always stable, regardless of Commutability.
   // The LHS is always matched first.
   CmpClass_match(PredicateTy &Pred, const LHS_t &LHS, const RHS_t &RHS)
-      : Predicate(Pred), L(LHS), R(RHS) {}
+      : Predicate(&Pred), L(LHS), R(RHS) {}
+  CmpClass_match(const LHS_t &LHS, const RHS_t &RHS)
+      : Predicate(nullptr), L(LHS), R(RHS) {}
 
   template <typename OpTy> bool match(OpTy *V) {
     if (auto *I = dyn_cast<Class>(V)) {
       if (L.match(I->getOperand(0)) && R.match(I->getOperand(1))) {
-        Predicate = I->getPredicate();
+        if (Predicate)
+          *Predicate = I->getPredicate();
         return true;
       } else if (Commutable && L.match(I->getOperand(1)) &&
                  R.match(I->getOperand(0))) {
-        Predicate = I->getSwappedPredicate();
+        if (Predicate)
+          *Predicate = I->getSwappedPredicate();
         return true;
       }
     }
@@ -1595,22 +1599,19 @@ m_FCmp(FCmpInst::Predicate &Pred, const LHS &L, const RHS &R) {
 template <typename LHS, typename RHS>
 inline CmpClass_match<LHS, RHS, CmpInst, CmpInst::Predicate>
 m_Cmp(const LHS &L, const RHS &R) {
-  CmpInst::Predicate Unused;
-  return CmpClass_match<LHS, RHS, CmpInst, CmpInst::Predicate>(Unused, L, R);
+  return CmpClass_match<LHS, RHS, CmpInst, CmpInst::Predicate>(L, R);
 }
 
 template <typename LHS, typename RHS>
 inline CmpClass_match<LHS, RHS, ICmpInst, ICmpInst::Predicate>
 m_ICmp(const LHS &L, const RHS &R) {
-  ICmpInst::Predicate Unused;
-  return CmpClass_match<LHS, RHS, ICmpInst, ICmpInst::Predicate>(Unused, L, R);
+  return CmpClass_match<LHS, RHS, ICmpInst, ICmpInst::Predicate>(L, R);
 }
 
 template <typename LHS, typename RHS>
 inline CmpClass_match<LHS, RHS, FCmpInst, FCmpInst::Predicate>
 m_FCmp(const LHS &L, const RHS &R) {
-  FCmpInst::Predicate Unused;
-  return CmpClass_match<LHS, RHS, FCmpInst, FCmpInst::Predicate>(Unused, L, R);
+  return CmpClass_match<LHS, RHS, FCmpInst, FCmpInst::Predicate>(L, R);
 }
 
 // Same as CmpClass, but instead of saving Pred as out output variable, match a
diff --git a/llvm/unittests/IR/PatternMatch.cpp b/llvm/unittests/IR/PatternMatch.cpp
index b82711ec244a6..309fcc93996bc 100644
--- a/llvm/unittests/IR/PatternMatch.cpp
+++ b/llvm/unittests/IR/PatternMatch.cpp
@@ -2235,7 +2235,7 @@ typedef ::testing::Types<std::tuple<Value*, Instruction*>,
     MutableConstTestTypes;
 TYPED_TEST_SUITE(MutableConstTest, MutableConstTestTypes, );
 
-TYPED_TEST(MutableConstTest, /* FIXME: UAR bug */ DISABLED_ICmp) {
+TYPED_TEST(MutableConstTest, ICmp) {
   auto &IRB = PatternMatchTest::IRB;
 
   typedef std::tuple_element_t<0, TypeParam> ValueType;
@@ -2319,7 +2319,7 @@ TYPED_TEST(MutableConstTest, /* FIXME: UAR bug */ DISABLED_ICmp) {
                    .match((InstructionType)IRB.CreateICmp(Pred, L, R)));
 }
 
-TYPED_TEST(MutableConstTest, /* FIXME: UAR bug */ DISABLED_FCmp) {
+TYPED_TEST(MutableConstTest, FCmp) {
   auto &IRB = PatternMatchTest::IRB;
 
   typedef std::tuple_element_t<0, TypeParam> ValueType;

Copy link
Contributor

@nikic nikic left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@goldsteinn goldsteinn merged commit e027017 into llvm:main Jul 15, 2024
9 checks passed
@vitalybuka
Copy link
Collaborator

vitalybuka commented Jul 15, 2024

It's not fixed, or fix incomplete: https://lab.llvm.org/buildbot/#/builders/169/builds/997

Updated: fixed URL^

@vitalybuka
Copy link
Collaborator

FYI @pcc

@goldsteinn
Copy link
Contributor Author

Hmm, not sure exactly how we still have the stale reference.

@goldsteinn
Copy link
Contributor Author

Oh, I forgot the commutative wrappers...

@goldsteinn
Copy link
Contributor Author

Sorry for all the hassle on this, I will test with asan locally before posting 3rd fix attempt.

yuxuanchen1997 pushed a commit that referenced this pull request Jul 25, 2024
…mp` matchers" (#99062)

Summary:
Reverts #98866

It's still use-after-scope.

Test Plan: 

Reviewers: 

Subscribers: 

Tasks: 

Tags: 


Differential Revision: https://phabricator.intern.facebook.com/D60251755
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants