[clang] [TBAA] Only emit pointer tbaa metedata for record types. (PR #116991)
Florian Hahn via cfe-commits
cfe-commits at lists.llvm.org
Thu Nov 21 10:48:05 PST 2024
https://github.com/fhahn updated https://github.com/llvm/llvm-project/pull/116991
>From 3a8c157ca484db54a3e1d1ff8061d7b76ce46834 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 20 Nov 2024 15:31:54 +0000
Subject: [PATCH 1/5] [TBAA] Add run-line with -fpointer-tbaa to cwg158
---
clang/test/CXX/drs/cwg158.cpp | 26 +++++++++++++++++++++-----
1 file changed, 21 insertions(+), 5 deletions(-)
diff --git a/clang/test/CXX/drs/cwg158.cpp b/clang/test/CXX/drs/cwg158.cpp
index 9301c790297e9d..6e18156d9ccc6f 100644
--- a/clang/test/CXX/drs/cwg158.cpp
+++ b/clang/test/CXX/drs/cwg158.cpp
@@ -1,12 +1,14 @@
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++98 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck %s
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++11 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck %s
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++14 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck %s
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++98 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,DEFAULT %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++11 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,DEFAULT %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++14 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,DEFAULT %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,DEFAULT %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -pointer-tbaa -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,POINTER-TBAA %s
// cwg158: yes
// CHECK-LABEL: define {{.*}} @_Z1f
const int *f(const int * const *p, int **q) {
+ // CHECK: load ptr, ptr %p.addr
// CHECK: load ptr, {{.*}}, !tbaa ![[INTPTR_TBAA:[^,]*]]
const int *x = *p;
// CHECK: store ptr null, {{.*}}, !tbaa ![[INTPTR_TBAA]]
@@ -18,10 +20,24 @@ struct A {};
// CHECK-LABEL: define {{.*}} @_Z1g
const int *(A::*const *g(const int *(A::* const **p)[3], int *(A::***q)[3]))[3] {
+ // CHECK: load ptr, ptr %p.addr
// CHECK: load ptr, {{.*}}, !tbaa ![[MEMPTR_TBAA:[^,]*]]
const int *(A::*const *x)[3] = *p;
- // CHECK: store ptr null, {{.*}}, !tbaa ![[MEMPTR_TBAA]]
+ // DEFAULT: store ptr null, {{.*}}, !tbaa ![[MEMPTR_TBAA]]
+ // POINTER-TBAA-NOT: store ptr null, {{.*}}, !tbaa ![[MEMPTR_TBAA]]
*q = 0;
return x;
}
+// CHECK-LABEL: define {{.*}} @_Z1h
+const int * h(const int * (*p)[10], int *(*q)[9]) {
+ // CHECK: load ptr, ptr %p.addr, align 8, !tbaa [[PTRARRAY_TBAA:!.+]]
+ const int * x = *p[0];
+
+ // CHECK: load ptr, ptr %q.addr, align 8, !tbaa [[PTRARRAY_TBAA]]
+ *q[0] = 0;
+ return x;
+}
+
+// POINTER-TBAA: [[PTRARRAY_TBAA]] = !{[[PTRARRAY_TY:!.+]], [[PTRARRAY_TY]], i64 0}
+// POINTER-TBAA: [[PTRARRAY_TY:!.+]] = !{!"p1 _ZTSPi", !4, i64 0}
>From 776f4279467f479ff3ee5b89c00b5f8e210e987a Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 20 Nov 2024 15:37:37 +0000
Subject: [PATCH 2/5] [TBAA] Only emit pointer tbaa metedata for record types.
Be conservative if the type isn't a record type. Handling other types may
require stripping const-qualifiers inside the type, e.g. MemberPointerType.
Without this, we assign different tags to the accesses for p an q in the
second test in cwg158.
---
clang/lib/CodeGen/CodeGenTBAA.cpp | 6 ++++++
clang/test/CXX/drs/cwg158.cpp | 13 ++++++-------
2 files changed, 12 insertions(+), 7 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenTBAA.cpp b/clang/lib/CodeGen/CodeGenTBAA.cpp
index c31579e8323174..73aae6b0db8d90 100644
--- a/clang/lib/CodeGen/CodeGenTBAA.cpp
+++ b/clang/lib/CodeGen/CodeGenTBAA.cpp
@@ -230,6 +230,12 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
->getString();
TyName = Name;
} else {
+ // Be conservative if the type isn't a record type. Handling other types
+ // may require stripping const-qualifiers inside the type, e.g.
+ // MemberPointerType.
+ if (!Ty->isRecordType())
+ return AnyPtr;
+
// For non-builtin types use the mangled name of the canonical type.
llvm::raw_svector_ostream TyOut(TyName);
MangleCtx->mangleCanonicalTypeName(QualType(Ty, 0), TyOut);
diff --git a/clang/test/CXX/drs/cwg158.cpp b/clang/test/CXX/drs/cwg158.cpp
index 6e18156d9ccc6f..cee0b506509015 100644
--- a/clang/test/CXX/drs/cwg158.cpp
+++ b/clang/test/CXX/drs/cwg158.cpp
@@ -1,8 +1,8 @@
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++98 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,DEFAULT %s
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++11 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,DEFAULT %s
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++14 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,DEFAULT %s
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,DEFAULT %s
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -pointer-tbaa -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,POINTER-TBAA %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++98 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++11 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++14 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -pointer-tbaa -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
// cwg158: yes
@@ -23,8 +23,7 @@ const int *(A::*const *g(const int *(A::* const **p)[3], int *(A::***q)[3]))[3]
// CHECK: load ptr, ptr %p.addr
// CHECK: load ptr, {{.*}}, !tbaa ![[MEMPTR_TBAA:[^,]*]]
const int *(A::*const *x)[3] = *p;
- // DEFAULT: store ptr null, {{.*}}, !tbaa ![[MEMPTR_TBAA]]
- // POINTER-TBAA-NOT: store ptr null, {{.*}}, !tbaa ![[MEMPTR_TBAA]]
+ // CHECK: store ptr null, {{.*}}, !tbaa ![[MEMPTR_TBAA]]
*q = 0;
return x;
}
>From e0076d68d416357f66a9dc25c1452de30e6ff76a Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 20 Nov 2024 22:41:15 +0000
Subject: [PATCH 3/5] !fixup loop through array types
---
clang/lib/CodeGen/CodeGenTBAA.cpp | 15 ++++++++-------
clang/test/CXX/drs/cwg158.cpp | 2 +-
2 files changed, 9 insertions(+), 8 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenTBAA.cpp b/clang/lib/CodeGen/CodeGenTBAA.cpp
index 73aae6b0db8d90..e40addbe2bea7e 100644
--- a/clang/lib/CodeGen/CodeGenTBAA.cpp
+++ b/clang/lib/CodeGen/CodeGenTBAA.cpp
@@ -206,12 +206,14 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
if (!CodeGenOpts.PointerTBAA)
return AnyPtr;
// Compute the depth of the pointer and generate a tag of the form "p<depth>
- // <base type tag>".
+ // <base type tag>". Look through pointer and array types to determine the
+ // base type.
unsigned PtrDepth = 0;
do {
PtrDepth++;
- Ty = Ty->getPointeeType().getTypePtr();
- } while (Ty->isPointerType());
+ Ty = Ty->isPointerType() ? Ty->getPointeeType().getTypePtr()
+ : Ty->getArrayElementTypeNoTypeQual();
+ } while (Ty->isPointerType() || Ty->isArrayType());
Ty = Context.getBaseElementType(QualType(Ty, 0)).getTypePtr();
assert(!isa<VariableArrayType>(Ty));
// When the underlying type is a builtin type, we compute the pointee type
@@ -230,10 +232,9 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
->getString();
TyName = Name;
} else {
- // Be conservative if the type isn't a record type. Handling other types
- // may require stripping const-qualifiers inside the type, e.g.
- // MemberPointerType.
- if (!Ty->isRecordType())
+ // Be conservative if the type a MemberPointerType. Those would require
+ // stripping const-qualifiers inside the type.
+ if (Ty->isMemberPointerType())
return AnyPtr;
// For non-builtin types use the mangled name of the canonical type.
diff --git a/clang/test/CXX/drs/cwg158.cpp b/clang/test/CXX/drs/cwg158.cpp
index cee0b506509015..a990aea2db2ecc 100644
--- a/clang/test/CXX/drs/cwg158.cpp
+++ b/clang/test/CXX/drs/cwg158.cpp
@@ -39,4 +39,4 @@ const int * h(const int * (*p)[10], int *(*q)[9]) {
}
// POINTER-TBAA: [[PTRARRAY_TBAA]] = !{[[PTRARRAY_TY:!.+]], [[PTRARRAY_TY]], i64 0}
-// POINTER-TBAA: [[PTRARRAY_TY:!.+]] = !{!"p1 _ZTSPi", !4, i64 0}
+// POINTER-TBAA: [[PTRARRAY_TY]] = !{!"p3 int", !4, i64 0}
>From 65755fed015b9a58137f300aee040d4f1a3869f1 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Thu, 21 Nov 2024 14:11:03 +0000
Subject: [PATCH 4/5] !fixup address comments
---
clang/lib/CodeGen/CodeGenTBAA.cpp | 30 +++++++++++++++++++++---------
1 file changed, 21 insertions(+), 9 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenTBAA.cpp b/clang/lib/CodeGen/CodeGenTBAA.cpp
index e40addbe2bea7e..72e79bf94a05c5 100644
--- a/clang/lib/CodeGen/CodeGenTBAA.cpp
+++ b/clang/lib/CodeGen/CodeGenTBAA.cpp
@@ -205,16 +205,27 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
llvm::MDNode *AnyPtr = createScalarTypeNode("any pointer", getChar(), Size);
if (!CodeGenOpts.PointerTBAA)
return AnyPtr;
- // Compute the depth of the pointer and generate a tag of the form "p<depth>
- // <base type tag>". Look through pointer and array types to determine the
- // base type.
+ // C++ [basic.lval]p11 permits objects to accessed through an l-value of
+ // similar type. Two types are similar under C++ [conv.qual]p2 if the
+ // decomposition of the types into pointers, member pointers, and arrays has
+ // the same structure when ignoring cv-qualifiers at each level of the
+ // decomposition. Meanwhile, C makes T(*)[] and T(*)[N] compatible, which
+ // would really complicate any attempt to distinguish pointers to arrays by
+ // their bounds. It's simpler, and much easier to explain to users, to
+ // simply treat all pointers to arrays as pointers to their element type for
+ // aliasing purposes. So when creating a TBAA tag for a pointer type, we
+ // recursively ignore both qualifiers and array types when decomposing the
+ // pointee type. The only meaningful remaining structure is the number of
+ // pointer types we encountered along the way, so we just produce the tag
+ // "p<depth> <base type tag>". If we do find a member pointer type, for now
+ // we just conservatively bail out with AnyPtr (below) rather than trying to
+ // create a tag that honors the similar-type rules while still
+ // distinguishing different kinds of member pointer.
unsigned PtrDepth = 0;
do {
PtrDepth++;
- Ty = Ty->isPointerType() ? Ty->getPointeeType().getTypePtr()
- : Ty->getArrayElementTypeNoTypeQual();
+ Ty = Ty->getPointeeType()->getBaseElementTypeUnsafe();
} while (Ty->isPointerType() || Ty->isArrayType());
- Ty = Context.getBaseElementType(QualType(Ty, 0)).getTypePtr();
assert(!isa<VariableArrayType>(Ty));
// When the underlying type is a builtin type, we compute the pointee type
// string recursively, which is implicitly more forgiving than the standards
@@ -232,9 +243,10 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
->getString();
TyName = Name;
} else {
- // Be conservative if the type a MemberPointerType. Those would require
- // stripping const-qualifiers inside the type.
- if (Ty->isMemberPointerType())
+ // Be conservative if the type isn't a RecordType. We are specifically
+ // required to do this for member pointers until we implement the
+ // similar-types rule.
+ if (!Ty->isRecordType())
return AnyPtr;
// For non-builtin types use the mangled name of the canonical type.
>From 2372e815a1e1d453b6435dbc0a40d40dc6491c78 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Thu, 21 Nov 2024 18:47:19 +0000
Subject: [PATCH 5/5] !fixup address latest comments, thanks!
---
clang/lib/CodeGen/CodeGenTBAA.cpp | 4 ++--
clang/test/CXX/drs/cwg158.cpp | 5 +++--
2 files changed, 5 insertions(+), 4 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenTBAA.cpp b/clang/lib/CodeGen/CodeGenTBAA.cpp
index 72e79bf94a05c5..4bcb541156bd23 100644
--- a/clang/lib/CodeGen/CodeGenTBAA.cpp
+++ b/clang/lib/CodeGen/CodeGenTBAA.cpp
@@ -225,7 +225,7 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
do {
PtrDepth++;
Ty = Ty->getPointeeType()->getBaseElementTypeUnsafe();
- } while (Ty->isPointerType() || Ty->isArrayType());
+ } while (Ty->isPointerType());
assert(!isa<VariableArrayType>(Ty));
// When the underlying type is a builtin type, we compute the pointee type
// string recursively, which is implicitly more forgiving than the standards
@@ -243,7 +243,7 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
->getString();
TyName = Name;
} else {
- // Be conservative if the type isn't a RecordType. We are specifically
+ // Be conservative if the type isn't a RecordType. We are specifically
// required to do this for member pointers until we implement the
// similar-types rule.
if (!Ty->isRecordType())
diff --git a/clang/test/CXX/drs/cwg158.cpp b/clang/test/CXX/drs/cwg158.cpp
index a990aea2db2ecc..2a744382647773 100644
--- a/clang/test/CXX/drs/cwg158.cpp
+++ b/clang/test/CXX/drs/cwg158.cpp
@@ -2,7 +2,7 @@
// RUN: %clang_cc1 -triple x86_64-linux -std=c++11 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
// RUN: %clang_cc1 -triple x86_64-linux -std=c++14 %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
-// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -pointer-tbaa -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK %s
+// RUN: %clang_cc1 -triple x86_64-linux -std=c++1z %s -O3 -pointer-tbaa -disable-llvm-passes -pedantic-errors -emit-llvm -o - | FileCheck --check-prefixes=CHECK,POINTER-TBAA %s
// cwg158: yes
@@ -39,4 +39,5 @@ const int * h(const int * (*p)[10], int *(*q)[9]) {
}
// POINTER-TBAA: [[PTRARRAY_TBAA]] = !{[[PTRARRAY_TY:!.+]], [[PTRARRAY_TY]], i64 0}
-// POINTER-TBAA: [[PTRARRAY_TY]] = !{!"p3 int", !4, i64 0}
+// POINTER-TBAA: [[PTRARRAY_TY]] = !{!"p2 int", [[ANYPTR:!.+]], i64 0}
+// POINTER-TBAA: [[ANYPTR]] = !{!"any pointer"
More information about the cfe-commits
mailing list