[clang] [llvm] [clang][CodeGen] Global constructors/destructors are globals (PR #93914)
Alex Voicu via cfe-commits
cfe-commits at lists.llvm.org
Mon Jun 3 07:46:16 PDT 2024
https://github.com/AlexVlx updated https://github.com/llvm/llvm-project/pull/93914
>From 34b9646a38c6bdd0419dd9d2eb1bc847c4d16596 Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Tue, 28 May 2024 20:09:15 +0100
Subject: [PATCH 01/10] Fix `emitUsed` to place `used` globals in the Global
AS. Update tests accordingly.
---
clang/lib/CodeGen/CodeGenModule.cpp | 11 ++++++-----
clang/test/CodeGen/2005-12-04-AttributeUsed.c | 2 ++
clang/test/CodeGen/attr-retain.c | 4 +++-
clang/test/CodeGenCUDA/llvm-used.cu | 7 +++++--
clang/test/CodeGenCXX/attr-retain.cpp | 2 ++
...static-member-variable-explicit-specialization.cpp | 6 +++++-
.../test/OpenMP/amdgcn_target_global_constructor.cpp | 4 ++--
7 files changed, 25 insertions(+), 11 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index e4774a587707a..30a18b1c5e7a8 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2047,9 +2047,9 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
llvm::Type *CtorPFTy = llvm::PointerType::get(CtorFTy,
TheModule.getDataLayout().getProgramAddressSpace());
- // Get the type of a ctor entry, { i32, void ()*, i8* }.
+ // Get the type of a ctor entry, { i32, program void ()*, global i8* }.
llvm::StructType *CtorStructTy = llvm::StructType::get(
- Int32Ty, CtorPFTy, VoidPtrTy);
+ Int32Ty, CtorPFTy, GlobalsInt8PtrTy);
// Construct the constructor and destructor arrays.
ConstantInitBuilder builder(*this);
@@ -2061,7 +2061,7 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
if (I.AssociatedData)
ctor.add(I.AssociatedData);
else
- ctor.addNullPointer(VoidPtrTy);
+ ctor.addNullPointer(GlobalsInt8PtrTy);
ctor.finishAndAddTo(ctors);
}
@@ -2928,12 +2928,13 @@ static void emitUsed(CodeGenModule &CGM, StringRef Name,
for (unsigned i = 0, e = List.size(); i != e; ++i) {
UsedArray[i] =
llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
- cast<llvm::Constant>(&*List[i]), CGM.Int8PtrTy);
+ cast<llvm::Constant>(&*List[i]), CGM.GlobalsInt8PtrTy);
}
if (UsedArray.empty())
return;
- llvm::ArrayType *ATy = llvm::ArrayType::get(CGM.Int8PtrTy, UsedArray.size());
+ llvm::ArrayType *ATy = llvm::ArrayType::get(CGM.GlobalsInt8PtrTy,
+ UsedArray.size());
auto *GV = new llvm::GlobalVariable(
CGM.getModule(), ATy, false, llvm::GlobalValue::AppendingLinkage,
diff --git a/clang/test/CodeGen/2005-12-04-AttributeUsed.c b/clang/test/CodeGen/2005-12-04-AttributeUsed.c
index bd232831fe1a6..c5089d9bda2c6 100644
--- a/clang/test/CodeGen/2005-12-04-AttributeUsed.c
+++ b/clang/test/CodeGen/2005-12-04-AttributeUsed.c
@@ -1,6 +1,8 @@
// RUN: %clang_cc1 %s -triple x86_64-apple-darwin -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 %s -triple amdgcn-amd-amdhsa -emit-llvm -o - | FileCheck %s --check-prefix=GLOBALAS
// CHECK: @llvm.used = appending global [2 x ptr] [ptr @foo, ptr @X], section "llvm.metadata"
+// GLOBALAS: @llvm.compiler.used = appending addrspace(1) global [2 x ptr addrspace(1)] [ptr addrspace(1) addrspacecast (ptr @foo to ptr addrspace(1)), ptr addrspace(1) @X], section "llvm.metadata"
int X __attribute__((used));
int Y;
diff --git a/clang/test/CodeGen/attr-retain.c b/clang/test/CodeGen/attr-retain.c
index fc41be76f9d0a..e86f6ae8ed06c 100644
--- a/clang/test/CodeGen/attr-retain.c
+++ b/clang/test/CodeGen/attr-retain.c
@@ -1,4 +1,5 @@
// RUN: %clang_cc1 -emit-llvm -triple x86_64 %s -o - | FileCheck %s
+// RUN: %clang_cc1 -emit-llvm -triple amdgcn-amd-amdhsa %s -o - | FileCheck %s --check-prefix=GLOBALAS
/// Set !retain regardless of the target. The backend will lower !retain to
/// SHF_GNU_RETAIN on ELF and ignore the metadata for other binary formats.
@@ -11,7 +12,8 @@
// CHECK: @llvm.used = appending global [8 x ptr] [ptr @c0, ptr @foo.l0, ptr @f0, ptr @f2, ptr @g0, ptr @g1, ptr @g3, ptr @g4], section "llvm.metadata"
// CHECK: @llvm.compiler.used = appending global [3 x ptr] [ptr @f2, ptr @g3, ptr @g4], section "llvm.metadata"
-
+// GLOBALAS: @llvm.used = appending addrspace(1) global [8 x ptr addrspace(1)] [ptr addrspace(1) addrspacecast (ptr addrspace(4) @c0 to ptr addrspace(1)), ptr addrspace(1) @foo.l0, ptr addrspace(1) addrspacecast (ptr @f0 to ptr addrspace(1)), ptr addrspace(1) addrspacecast (ptr @f2 to ptr addrspace(1)), ptr addrspace(1) @g0, ptr addrspace(1) @g1, ptr addrspace(1) @g3, ptr addrspace(1) @g4], section "llvm.metadata"
+// GLOBALAS: @llvm.compiler.used = appending addrspace(1) global [3 x ptr addrspace(1)] [ptr addrspace(1) addrspacecast (ptr @f2 to ptr addrspace(1)), ptr addrspace(1) @g3, ptr addrspace(1) @g4], section "llvm.metadata"
const int c0 __attribute__((retain)) = 42;
void foo(void) {
diff --git a/clang/test/CodeGenCUDA/llvm-used.cu b/clang/test/CodeGenCUDA/llvm-used.cu
index c39111dd48036..7782048b52836 100644
--- a/clang/test/CodeGenCUDA/llvm-used.cu
+++ b/clang/test/CodeGenCUDA/llvm-used.cu
@@ -1,8 +1,11 @@
// RUN: %clang_cc1 -emit-llvm %s -o - -fcuda-is-device -triple nvptx64-unknown-unknown | FileCheck %s
+// RUN: %clang_cc1 -emit-llvm %s -o - -fcuda-is-device -triple amdgcn-amd-amdhsa | FileCheck %s --check-prefix=NOASCAST
-// Make sure we emit the proper addrspacecast for llvm.used. PR22383 exposed an
-// issue where we were generating a bitcast instead of an addrspacecast.
+// Make sure we emit the proper addrspacecast for llvm.used iff necessary.
+// PR22383 exposed an issue where we were generating a bitcast instead of an
+// addrspacecast.
// CHECK: @llvm.compiler.used = appending global [1 x ptr] [ptr addrspacecast (ptr addrspace(1) @a to ptr)], section "llvm.metadata"
+// NOASCAST: @llvm.compiler.used = appending addrspace(1) global [1 x ptr addrspace(1)] [ptr addrspace(1) @a], section "llvm.metadata"
__attribute__((device)) __attribute__((__used__)) int a[] = {};
diff --git a/clang/test/CodeGenCXX/attr-retain.cpp b/clang/test/CodeGenCXX/attr-retain.cpp
index 3a56576d81632..73c649ae972bf 100644
--- a/clang/test/CodeGenCXX/attr-retain.cpp
+++ b/clang/test/CodeGenCXX/attr-retain.cpp
@@ -1,6 +1,8 @@
// RUN: %clang_cc1 -emit-llvm -triple x86_64 -Werror %s -o - | FileCheck %s
+// RUN: %clang_cc1 -emit-llvm -triple amdgcn-amd-amdhsa -Werror %s -o - | FileCheck %s --check-prefix=GLOBALAS
// CHECK: @llvm.used = appending global [7 x ptr]
+// GLOBALAS: @llvm.used = appending addrspace(1) global [7 x ptr addrspace(1)]
// CHECK-SAME: @_ZN2X0C2Ev
// CHECK-SAME: @_ZN2X0C1Ev
// CHECK-SAME: @_ZN2X0D2Ev
diff --git a/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp b/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
index 46c4c4d391769..04d29cf1cea51 100644
--- a/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
+++ b/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
@@ -1,6 +1,8 @@
// RUN: %clang_cc1 %s -std=c++1y -triple=x86_64-pc-linux -emit-llvm -o - | FileCheck --check-prefix=ELF --check-prefix=ALL %s
// RUN: %clang_cc1 %s -std=c++1y -triple=x86_64-apple-darwin -emit-llvm -o - | FileCheck --check-prefix=MACHO --check-prefix=ALL %s
// RUN: %clang_cc1 %s -std=c++1y -triple=x86_64-pc-linux -emit-llvm -fdeclspec -DSELECTANY -o - | FileCheck --check-prefix=ELF-SELECTANY %s
+// RUN: %clang_cc1 %s -std=c++1y -triple=amdgcn-amd-amdhsa -emit-llvm -o - | FileCheck --check-prefix=GLOBALAS-ELF %s
+// RUN: %clang_cc1 %s -std=c++1y -triple=amdgcn-amd-amdhsa -emit-llvm -fdeclspec -DSELECTANY -o - | FileCheck --check-prefix=GLOBALAS-ELF-SELECTANY %s
#ifdef SELECTANY
struct S {
@@ -12,6 +14,7 @@ int f();
// ELF-SELECTANY: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__cxx_global_var_init, ptr @selectany }]
// ELF-SELECTANY: @llvm.used = appending global [1 x ptr] [ptr @selectany]
+// GLOBALAS-ELF-SELECTANY: @llvm.used = appending addrspace(1) global [1 x ptr addrspace(1)] [ptr addrspace(1) @selectany]
int __declspec(selectany) selectany = f();
#else
@@ -73,12 +76,13 @@ template<> int A<bool>::a = 10;
// MACHO: { i32, ptr, ptr } { i32 65535, ptr @[[unordered22:[^,]*]], ptr null },
// ELF: { i32, ptr, ptr } { i32 65535, ptr @[[unordered21:[^,]*]], ptr @_ZN4Fib2ILi5EE1aE },
-// MACHO: { i32, ptr, ptr } { i32 65535, ptr @[[unordered21:[^,]*]], ptr null },
+// MACHO: { i32, ptr, ptr } { i32 65535, ptr @[[unordered21:[^,]*]], ptr null },
// ALL: { i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__sub_I_static_member_variable_explicit_specialization.cpp, ptr null }]
/// llvm.used ensures SHT_INIT_ARRAY in a section group cannot be GCed.
// ELF: @llvm.used = appending global [14 x ptr] [ptr @_ZN1AIsE1aE, ptr @_Z1xIsE, ptr @_ZN2ns1aIiE1iE, ptr @_ZN2ns1b1iIiEE, ptr @_ZN1AIvE1aE, ptr @_Z1xIcE, ptr @_ZN3FibILi5EE1aE, ptr @_ZN3FibILi3EE1aE, ptr @_ZN3FibILi2EE1aE, ptr @_ZN3FibILi4EE1aE, ptr @_ZN4Fib2ILi5EE1aE, ptr @_ZN4Fib2ILi4EE1aE, ptr @_ZN4Fib2ILi2EE1aE, ptr @_ZN4Fib2ILi3EE1aE]
+// GLOBALAS-ELF: @llvm.used = appending addrspace(1) global [14 x ptr addrspace(1)] [ptr addrspace(1) @_ZN1AIsE1aE, ptr addrspace(1) @_Z1xIsE, ptr addrspace(1) @_ZN2ns1aIiE1iE, ptr addrspace(1) @_ZN2ns1b1iIiEE, ptr addrspace(1) @_ZN1AIvE1aE, ptr addrspace(1) @_Z1xIcE, ptr addrspace(1) @_ZN3FibILi5EE1aE, ptr addrspace(1) @_ZN3FibILi3EE1aE, ptr addrspace(1) @_ZN3FibILi2EE1aE, ptr addrspace(1) @_ZN3FibILi4EE1aE, ptr addrspace(1) @_ZN4Fib2ILi5EE1aE, ptr addrspace(1) @_ZN4Fib2ILi4EE1aE, ptr addrspace(1) @_ZN4Fib2ILi2EE1aE, ptr addrspace(1) @_ZN4Fib2ILi3EE1aE], section "llvm.metadata"
template int A<short>::a; // Unordered
int b = foo();
diff --git a/clang/test/OpenMP/amdgcn_target_global_constructor.cpp b/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
index 80a5067b357a2..cc9dd56615dc6 100644
--- a/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
+++ b/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
@@ -21,8 +21,8 @@ S A;
#endif
//.
-// CHECK: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__sub_I_amdgcn_target_global_constructor.cpp, ptr null }]
-// CHECK: @llvm.global_dtors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__dtor_A, ptr null }]
+// CHECK: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @_GLOBAL__sub_I_amdgcn_target_global_constructor.cpp, ptr addrspace(1) null }]
+// CHECK: @llvm.global_dtors = appending global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @__dtor_A, ptr addrspace(1) null }]
//.
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
>From 90cd705b63cbdf1939b61b07a11d94abbcdea3ec Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Tue, 28 May 2024 20:30:00 +0100
Subject: [PATCH 02/10] Fix formatting.
---
clang/lib/CodeGen/CodeGenModule.cpp | 15 +++++++--------
1 file changed, 7 insertions(+), 8 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 30a18b1c5e7a8..397c611b3ea33 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2048,8 +2048,8 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
TheModule.getDataLayout().getProgramAddressSpace());
// Get the type of a ctor entry, { i32, program void ()*, global i8* }.
- llvm::StructType *CtorStructTy = llvm::StructType::get(
- Int32Ty, CtorPFTy, GlobalsInt8PtrTy);
+ llvm::StructType *CtorStructTy =
+ llvm::StructType::get(Int32Ty, CtorPFTy, GlobalsInt8PtrTy);
// Construct the constructor and destructor arrays.
ConstantInitBuilder builder(*this);
@@ -2068,7 +2068,7 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
auto list =
ctors.finishAndCreateGlobal(GlobalName, getPointerAlign(),
/*constant*/ false,
- llvm::GlobalValue::AppendingLinkage);
+ llvm::GlobalValue::AppendingLinkage, );
// The LTO linker doesn't seem to like it when we set an alignment
// on appending variables. Take it off as a workaround.
@@ -2926,15 +2926,14 @@ static void emitUsed(CodeGenModule &CGM, StringRef Name,
SmallVector<llvm::Constant*, 8> UsedArray;
UsedArray.resize(List.size());
for (unsigned i = 0, e = List.size(); i != e; ++i) {
- UsedArray[i] =
- llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
- cast<llvm::Constant>(&*List[i]), CGM.GlobalsInt8PtrTy);
+ UsedArray[i] = llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
+ cast<llvm::Constant>(&*List[i]), CGM.GlobalsInt8PtrTy);
}
if (UsedArray.empty())
return;
- llvm::ArrayType *ATy = llvm::ArrayType::get(CGM.GlobalsInt8PtrTy,
- UsedArray.size());
+ llvm::ArrayType *ATy =
+ llvm::ArrayType::get(CGM.GlobalsInt8PtrTy, UsedArray.size());
auto *GV = new llvm::GlobalVariable(
CGM.getModule(), ATy, false, llvm::GlobalValue::AppendingLinkage,
>From 780f541ed427ad847ec80dd5dd6cc602ef9a01da Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Tue, 28 May 2024 20:44:53 +0100
Subject: [PATCH 03/10] Delete accidental noise.
---
clang/lib/CodeGen/CodeGenModule.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 397c611b3ea33..4346ea7b1b6ad 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2068,7 +2068,7 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
auto list =
ctors.finishAndCreateGlobal(GlobalName, getPointerAlign(),
/*constant*/ false,
- llvm::GlobalValue::AppendingLinkage, );
+ llvm::GlobalValue::AppendingLinkage);
// The LTO linker doesn't seem to like it when we set an alignment
// on appending variables. Take it off as a workaround.
>From e5b604f2b839705d951cd3739210571c70653970 Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Thu, 30 May 2024 15:28:42 +0100
Subject: [PATCH 04/10] Remove `used`/`compiler.used` related changes.
---
clang/lib/CodeGen/CodeGenModule.cpp | 5 ++---
clang/test/CodeGen/2005-12-04-AttributeUsed.c | 2 --
clang/test/CodeGen/asan-constructor.c | 1 +
clang/test/CodeGen/attr-retain.c | 3 ---
clang/test/CodeGenCUDA/llvm-used.cu | 7 ++-----
clang/test/CodeGenCXX/attr-retain.cpp | 2 --
.../static-member-variable-explicit-specialization.cpp | 4 ++--
7 files changed, 7 insertions(+), 17 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 473a8a251a58f..fc08932d9ae7a 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2927,13 +2927,12 @@ static void emitUsed(CodeGenModule &CGM, StringRef Name,
UsedArray.resize(List.size());
for (unsigned i = 0, e = List.size(); i != e; ++i) {
UsedArray[i] = llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
- cast<llvm::Constant>(&*List[i]), CGM.GlobalsInt8PtrTy);
+ cast<llvm::Constant>(&*List[i]), CGM.Int8PtrTy);
}
if (UsedArray.empty())
return;
- llvm::ArrayType *ATy =
- llvm::ArrayType::get(CGM.GlobalsInt8PtrTy, UsedArray.size());
+ llvm::ArrayType *ATy = llvm::ArrayType::get(CGM.Int8PtrTy, UsedArray.size());
auto *GV = new llvm::GlobalVariable(
CGM.getModule(), ATy, false, llvm::GlobalValue::AppendingLinkage,
diff --git a/clang/test/CodeGen/2005-12-04-AttributeUsed.c b/clang/test/CodeGen/2005-12-04-AttributeUsed.c
index c5089d9bda2c6..bd232831fe1a6 100644
--- a/clang/test/CodeGen/2005-12-04-AttributeUsed.c
+++ b/clang/test/CodeGen/2005-12-04-AttributeUsed.c
@@ -1,8 +1,6 @@
// RUN: %clang_cc1 %s -triple x86_64-apple-darwin -emit-llvm -o - | FileCheck %s
-// RUN: %clang_cc1 %s -triple amdgcn-amd-amdhsa -emit-llvm -o - | FileCheck %s --check-prefix=GLOBALAS
// CHECK: @llvm.used = appending global [2 x ptr] [ptr @foo, ptr @X], section "llvm.metadata"
-// GLOBALAS: @llvm.compiler.used = appending addrspace(1) global [2 x ptr addrspace(1)] [ptr addrspace(1) addrspacecast (ptr @foo to ptr addrspace(1)), ptr addrspace(1) @X], section "llvm.metadata"
int X __attribute__((used));
int Y;
diff --git a/clang/test/CodeGen/asan-constructor.c b/clang/test/CodeGen/asan-constructor.c
index 762bf0778565d..280c550a75fb3 100644
--- a/clang/test/CodeGen/asan-constructor.c
+++ b/clang/test/CodeGen/asan-constructor.c
@@ -1,5 +1,6 @@
// RUN: %clang_cc1 -fsanitize=address -emit-llvm -o - %s | FileCheck %s
// RUN: %clang_cc1 -triple wasm32-unknown-emscripten -fsanitize=address -emit-llvm -o - %s | FileCheck %s --check-prefix=EMSCRIPTEN
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fsanitize=address -emit-llvm -o - %s | FileCheck %s
// CHECK: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @asan.module_ctor, ptr null }]
// EMSCRIPTEN: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 50, ptr @asan.module_ctor, ptr null }]
diff --git a/clang/test/CodeGen/attr-retain.c b/clang/test/CodeGen/attr-retain.c
index e86f6ae8ed06c..6a9869c00392d 100644
--- a/clang/test/CodeGen/attr-retain.c
+++ b/clang/test/CodeGen/attr-retain.c
@@ -1,5 +1,4 @@
// RUN: %clang_cc1 -emit-llvm -triple x86_64 %s -o - | FileCheck %s
-// RUN: %clang_cc1 -emit-llvm -triple amdgcn-amd-amdhsa %s -o - | FileCheck %s --check-prefix=GLOBALAS
/// Set !retain regardless of the target. The backend will lower !retain to
/// SHF_GNU_RETAIN on ELF and ignore the metadata for other binary formats.
@@ -12,8 +11,6 @@
// CHECK: @llvm.used = appending global [8 x ptr] [ptr @c0, ptr @foo.l0, ptr @f0, ptr @f2, ptr @g0, ptr @g1, ptr @g3, ptr @g4], section "llvm.metadata"
// CHECK: @llvm.compiler.used = appending global [3 x ptr] [ptr @f2, ptr @g3, ptr @g4], section "llvm.metadata"
-// GLOBALAS: @llvm.used = appending addrspace(1) global [8 x ptr addrspace(1)] [ptr addrspace(1) addrspacecast (ptr addrspace(4) @c0 to ptr addrspace(1)), ptr addrspace(1) @foo.l0, ptr addrspace(1) addrspacecast (ptr @f0 to ptr addrspace(1)), ptr addrspace(1) addrspacecast (ptr @f2 to ptr addrspace(1)), ptr addrspace(1) @g0, ptr addrspace(1) @g1, ptr addrspace(1) @g3, ptr addrspace(1) @g4], section "llvm.metadata"
-// GLOBALAS: @llvm.compiler.used = appending addrspace(1) global [3 x ptr addrspace(1)] [ptr addrspace(1) addrspacecast (ptr @f2 to ptr addrspace(1)), ptr addrspace(1) @g3, ptr addrspace(1) @g4], section "llvm.metadata"
const int c0 __attribute__((retain)) = 42;
void foo(void) {
diff --git a/clang/test/CodeGenCUDA/llvm-used.cu b/clang/test/CodeGenCUDA/llvm-used.cu
index 7782048b52836..9f22dabde3c2d 100644
--- a/clang/test/CodeGenCUDA/llvm-used.cu
+++ b/clang/test/CodeGenCUDA/llvm-used.cu
@@ -1,11 +1,8 @@
// RUN: %clang_cc1 -emit-llvm %s -o - -fcuda-is-device -triple nvptx64-unknown-unknown | FileCheck %s
-// RUN: %clang_cc1 -emit-llvm %s -o - -fcuda-is-device -triple amdgcn-amd-amdhsa | FileCheck %s --check-prefix=NOASCAST
-// Make sure we emit the proper addrspacecast for llvm.used iff necessary.
-// PR22383 exposed an issue where we were generating a bitcast instead of an
-// addrspacecast.
+// Make sure we emit the proper addrspacecast for llvm.used. PR22383 exposed an
+// issue where we were generating a bitcast instead of an addrspacecast.
// CHECK: @llvm.compiler.used = appending global [1 x ptr] [ptr addrspacecast (ptr addrspace(1) @a to ptr)], section "llvm.metadata"
-// NOASCAST: @llvm.compiler.used = appending addrspace(1) global [1 x ptr addrspace(1)] [ptr addrspace(1) @a], section "llvm.metadata"
__attribute__((device)) __attribute__((__used__)) int a[] = {};
diff --git a/clang/test/CodeGenCXX/attr-retain.cpp b/clang/test/CodeGenCXX/attr-retain.cpp
index 73c649ae972bf..3a56576d81632 100644
--- a/clang/test/CodeGenCXX/attr-retain.cpp
+++ b/clang/test/CodeGenCXX/attr-retain.cpp
@@ -1,8 +1,6 @@
// RUN: %clang_cc1 -emit-llvm -triple x86_64 -Werror %s -o - | FileCheck %s
-// RUN: %clang_cc1 -emit-llvm -triple amdgcn-amd-amdhsa -Werror %s -o - | FileCheck %s --check-prefix=GLOBALAS
// CHECK: @llvm.used = appending global [7 x ptr]
-// GLOBALAS: @llvm.used = appending addrspace(1) global [7 x ptr addrspace(1)]
// CHECK-SAME: @_ZN2X0C2Ev
// CHECK-SAME: @_ZN2X0C1Ev
// CHECK-SAME: @_ZN2X0D2Ev
diff --git a/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp b/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
index 04d29cf1cea51..839e528bf2a9f 100644
--- a/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
+++ b/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
@@ -14,7 +14,7 @@ int f();
// ELF-SELECTANY: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__cxx_global_var_init, ptr @selectany }]
// ELF-SELECTANY: @llvm.used = appending global [1 x ptr] [ptr @selectany]
-// GLOBALAS-ELF-SELECTANY: @llvm.used = appending addrspace(1) global [1 x ptr addrspace(1)] [ptr addrspace(1) @selectany]
+// GLOBALAS-ELF-SELECTANY: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__cxx_global_var_init, ptr @selectany }]
int __declspec(selectany) selectany = f();
#else
@@ -33,6 +33,7 @@ template<> int A<char>::a;
template<> int A<bool>::a = 10;
// ALL: @llvm.global_ctors = appending global [16 x { i32, ptr, ptr }]
+// GLOBALAS-ELF: @llvm.global_ctors = appending global [16 x { i32, ptr, ptr }]
// ELF: [{ i32, ptr, ptr } { i32 65535, ptr @[[unordered:[^,]*]], ptr @_ZN1AIsE1aE },
// MACHO: [{ i32, ptr, ptr } { i32 65535, ptr @[[unordered:[^,]*]], ptr null },
@@ -82,7 +83,6 @@ template<> int A<bool>::a = 10;
/// llvm.used ensures SHT_INIT_ARRAY in a section group cannot be GCed.
// ELF: @llvm.used = appending global [14 x ptr] [ptr @_ZN1AIsE1aE, ptr @_Z1xIsE, ptr @_ZN2ns1aIiE1iE, ptr @_ZN2ns1b1iIiEE, ptr @_ZN1AIvE1aE, ptr @_Z1xIcE, ptr @_ZN3FibILi5EE1aE, ptr @_ZN3FibILi3EE1aE, ptr @_ZN3FibILi2EE1aE, ptr @_ZN3FibILi4EE1aE, ptr @_ZN4Fib2ILi5EE1aE, ptr @_ZN4Fib2ILi4EE1aE, ptr @_ZN4Fib2ILi2EE1aE, ptr @_ZN4Fib2ILi3EE1aE]
-// GLOBALAS-ELF: @llvm.used = appending addrspace(1) global [14 x ptr addrspace(1)] [ptr addrspace(1) @_ZN1AIsE1aE, ptr addrspace(1) @_Z1xIsE, ptr addrspace(1) @_ZN2ns1aIiE1iE, ptr addrspace(1) @_ZN2ns1b1iIiEE, ptr addrspace(1) @_ZN1AIvE1aE, ptr addrspace(1) @_Z1xIcE, ptr addrspace(1) @_ZN3FibILi5EE1aE, ptr addrspace(1) @_ZN3FibILi3EE1aE, ptr addrspace(1) @_ZN3FibILi2EE1aE, ptr addrspace(1) @_ZN3FibILi4EE1aE, ptr addrspace(1) @_ZN4Fib2ILi5EE1aE, ptr addrspace(1) @_ZN4Fib2ILi4EE1aE, ptr addrspace(1) @_ZN4Fib2ILi2EE1aE, ptr addrspace(1) @_ZN4Fib2ILi3EE1aE], section "llvm.metadata"
template int A<short>::a; // Unordered
int b = foo();
>From 2aa32c55c1cd16c7d1213fb8a47b3ee61a534408 Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Thu, 30 May 2024 15:30:10 +0100
Subject: [PATCH 05/10] Revert unrelated formatting change.
---
clang/lib/CodeGen/CodeGenModule.cpp | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index fc08932d9ae7a..a22f1eb2b8065 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2926,8 +2926,8 @@ static void emitUsed(CodeGenModule &CGM, StringRef Name,
SmallVector<llvm::Constant*, 8> UsedArray;
UsedArray.resize(List.size());
for (unsigned i = 0, e = List.size(); i != e; ++i) {
- UsedArray[i] = llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
- cast<llvm::Constant>(&*List[i]), CGM.Int8PtrTy);
+ llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
+ cast<llvm::Constant>(&*List[i]), CGM.Int8PtrTy);
}
if (UsedArray.empty())
>From 7985d43a97251b28ae092f1a90bbf5a726836096 Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Thu, 30 May 2024 15:38:21 +0100
Subject: [PATCH 06/10] Fix typo.
---
clang/lib/CodeGen/CodeGenModule.cpp | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index a22f1eb2b8065..c45c5cab9f48c 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2926,8 +2926,9 @@ static void emitUsed(CodeGenModule &CGM, StringRef Name,
SmallVector<llvm::Constant*, 8> UsedArray;
UsedArray.resize(List.size());
for (unsigned i = 0, e = List.size(); i != e; ++i) {
- llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
- cast<llvm::Constant>(&*List[i]), CGM.Int8PtrTy);
+ UsedArray[i] =
+ llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
+ cast<llvm::Constant>(&*List[i]), CGM.Int8PtrTy);
}
if (UsedArray.empty())
>From 17e41275180ab8f5f1dec7d872e53efa0a40c8eb Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Fri, 31 May 2024 02:59:41 +0100
Subject: [PATCH 07/10] Fix the AS for the `global_*tors` arrays.
---
clang/lib/CodeGen/CodeGenModule.cpp | 3 ++-
clang/test/CodeGen/asan-constructor.c | 3 ++-
.../cxx0x-initializer-stdinitializerlist.cpp | 3 ++-
...atic-member-variable-explicit-specialization.cpp | 4 ++--
.../OpenMP/amdgcn_target_global_constructor.cpp | 4 ++--
llvm/lib/Transforms/Utils/ModuleUtils.cpp | 13 ++++++-------
.../llvm-reduce/remove-ifunc-program-addrspace.ll | 4 ++--
7 files changed, 18 insertions(+), 16 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index c45c5cab9f48c..b59c5c52783e9 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2068,7 +2068,8 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
auto list =
ctors.finishAndCreateGlobal(GlobalName, getPointerAlign(),
/*constant*/ false,
- llvm::GlobalValue::AppendingLinkage);
+ llvm::GlobalValue::AppendingLinkage,
+ GlobalsInt8PtrTy->getAddressSpace());
// The LTO linker doesn't seem to like it when we set an alignment
// on appending variables. Take it off as a workaround.
diff --git a/clang/test/CodeGen/asan-constructor.c b/clang/test/CodeGen/asan-constructor.c
index 280c550a75fb3..41212d45a5c63 100644
--- a/clang/test/CodeGen/asan-constructor.c
+++ b/clang/test/CodeGen/asan-constructor.c
@@ -1,6 +1,7 @@
// RUN: %clang_cc1 -fsanitize=address -emit-llvm -o - %s | FileCheck %s
// RUN: %clang_cc1 -triple wasm32-unknown-emscripten -fsanitize=address -emit-llvm -o - %s | FileCheck %s --check-prefix=EMSCRIPTEN
-// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fsanitize=address -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fsanitize=address -emit-llvm -o - %s | FileCheck %s --check-prefix=GLOBALAS
// CHECK: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @asan.module_ctor, ptr null }]
// EMSCRIPTEN: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 50, ptr @asan.module_ctor, ptr null }]
+// GLOBALAS: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 1, ptr @asan.module_ctor, ptr addrspace(1) null }]
diff --git a/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp b/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp
index aa2f078a5fb0c..2e22192e8b388 100644
--- a/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp
+++ b/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp
@@ -90,7 +90,8 @@ std::initializer_list<int> thread_local x = {1, 2, 3, 4};
// AMDGCN: @[[REFTMP1:.*]] = private addrspace(4) constant [2 x i32] [i32 42, i32 43], align 4
// AMDGCN: @[[REFTMP2:.*]] = private addrspace(4) constant [3 x %{{.*}}] [%{{.*}} { i32 1 }, %{{.*}} { i32 2 }, %{{.*}} { i32 3 }], align 4
-// CHECK: appending global
+// X86: appending global
+// AMDGCN: appending addrspace(1) global
// thread_local initializer:
// X86-LABEL: define internal void @__cxx_global_var_init
diff --git a/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp b/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
index 839e528bf2a9f..58055130cd708 100644
--- a/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
+++ b/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
@@ -14,7 +14,7 @@ int f();
// ELF-SELECTANY: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__cxx_global_var_init, ptr @selectany }]
// ELF-SELECTANY: @llvm.used = appending global [1 x ptr] [ptr @selectany]
-// GLOBALAS-ELF-SELECTANY: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__cxx_global_var_init, ptr @selectany }]
+// GLOBALAS-ELF-SELECTANY: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @__cxx_global_var_init, ptr addrspace(1) @selectany }]
int __declspec(selectany) selectany = f();
#else
@@ -33,7 +33,7 @@ template<> int A<char>::a;
template<> int A<bool>::a = 10;
// ALL: @llvm.global_ctors = appending global [16 x { i32, ptr, ptr }]
-// GLOBALAS-ELF: @llvm.global_ctors = appending global [16 x { i32, ptr, ptr }]
+// GLOBALAS-ELF: @llvm.global_ctors = appending addrspace(1) global [16 x { i32, ptr, ptr addrspace(1) }]
// ELF: [{ i32, ptr, ptr } { i32 65535, ptr @[[unordered:[^,]*]], ptr @_ZN1AIsE1aE },
// MACHO: [{ i32, ptr, ptr } { i32 65535, ptr @[[unordered:[^,]*]], ptr null },
diff --git a/clang/test/OpenMP/amdgcn_target_global_constructor.cpp b/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
index cc9dd56615dc6..fe414b59588c0 100644
--- a/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
+++ b/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
@@ -21,8 +21,8 @@ S A;
#endif
//.
-// CHECK: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @_GLOBAL__sub_I_amdgcn_target_global_constructor.cpp, ptr addrspace(1) null }]
-// CHECK: @llvm.global_dtors = appending global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @__dtor_A, ptr addrspace(1) null }]
+// CHECK: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @_GLOBAL__sub_I_amdgcn_target_global_constructor.cpp, ptr addrspace(1) null }]
+// CHECK: @llvm.global_dtors = appending addrspace(1) global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @__dtor_A, ptr addrspace(1) null }]
//.
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
diff --git a/llvm/lib/Transforms/Utils/ModuleUtils.cpp b/llvm/lib/Transforms/Utils/ModuleUtils.cpp
index 122279160cc7e..7151cc869bc95 100644
--- a/llvm/lib/Transforms/Utils/ModuleUtils.cpp
+++ b/llvm/lib/Transforms/Utils/ModuleUtils.cpp
@@ -30,6 +30,8 @@ static void appendToGlobalArray(StringRef ArrayName, Module &M, Function *F,
int Priority, Constant *Data) {
IRBuilder<> IRB(M.getContext());
FunctionType *FnTy = FunctionType::get(IRB.getVoidTy(), false);
+ PointerType *GlobalsInt8PtrTy = PointerType::get(
+ M.getContext(), M.getDataLayout().getDefaultGlobalsAddressSpace());
// Get the current set of static global constructors and add the new ctor
// to the list.
@@ -47,15 +49,15 @@ static void appendToGlobalArray(StringRef ArrayName, Module &M, Function *F,
} else {
EltTy = StructType::get(IRB.getInt32Ty(),
PointerType::get(FnTy, F->getAddressSpace()),
- IRB.getPtrTy());
+ GlobalsInt8PtrTy);
}
// Build a 3 field global_ctor entry. We don't take a comdat key.
Constant *CSVals[3];
CSVals[0] = IRB.getInt32(Priority);
CSVals[1] = F;
- CSVals[2] = Data ? ConstantExpr::getPointerCast(Data, IRB.getPtrTy())
- : Constant::getNullValue(IRB.getPtrTy());
+ CSVals[2] = Data ? ConstantExpr::getPointerCast(Data, GlobalsInt8PtrTy)
+ : Constant::getNullValue(GlobalsInt8PtrTy);
Constant *RuntimeCtorInit =
ConstantStruct::get(EltTy, ArrayRef(CSVals, EltTy->getNumElements()));
@@ -437,12 +439,9 @@ bool llvm::lowerGlobalIFuncUsersAsGlobalCtor(
InitBuilder.CreateRetVoid();
- PointerType *ConstantDataTy = PointerType::get(Ctx, 0);
-
// TODO: Is this the right priority? Probably should be before any other
// constructors?
const int Priority = 10;
- appendToGlobalCtors(M, NewCtor, Priority,
- ConstantPointerNull::get(ConstantDataTy));
+ appendToGlobalCtors(M, NewCtor, Priority, nullptr);
return UnhandledUsers;
}
diff --git a/llvm/test/tools/llvm-reduce/remove-ifunc-program-addrspace.ll b/llvm/test/tools/llvm-reduce/remove-ifunc-program-addrspace.ll
index e275d61764b21..7ddab0b6a815e 100644
--- a/llvm/test/tools/llvm-reduce/remove-ifunc-program-addrspace.ll
+++ b/llvm/test/tools/llvm-reduce/remove-ifunc-program-addrspace.ll
@@ -15,8 +15,8 @@ define void @existing_ctor() addrspace(1) {
; CHECK-FINAL: [[TABLE:@[0-9]+]] = internal addrspace(2) global [6 x ptr addrspace(1)] poison, align 8
-; CHECK-FINAL: @llvm.global_ctors = appending addrspace(2) global [2 x { i32, ptr addrspace(1), ptr }] [{ i32, ptr addrspace(1), ptr } { i32 0, ptr addrspace(1) @existing_ctor, ptr null }, { i32, ptr addrspace(1), ptr } { i32 10, ptr addrspace(1) [[TABLE_CTOR:@[0-9]+]], ptr null }]
- at llvm.global_ctors = appending global [1 x { i32, ptr addrspace(1), ptr }] [{ i32, ptr addrspace(1), ptr } { i32 0, ptr addrspace(1) @existing_ctor, ptr null }]
+; CHECK-FINAL: @llvm.global_ctors = appending addrspace(2) global [2 x { i32, ptr addrspace(1), ptr addrspace(2) }] [{ i32, ptr addrspace(1), ptr addrspace(2) } { i32 0, ptr addrspace(1) @existing_ctor, ptr addrspace(2) null }, { i32, ptr addrspace(1), ptr addrspace(2) } { i32 10, ptr addrspace(1) [[TABLE_CTOR:@[0-9]+]], ptr addrspace(2) null }]
+ at llvm.global_ctors = appending global [1 x { i32, ptr addrspace(1), ptr addrspace(2) }] [{ i32, ptr addrspace(1), ptr addrspace(2) } { i32 0, ptr addrspace(1) @existing_ctor, ptr addrspace(2) null }]
>From d3136cdbf1a71dcb8fe7236e007d2fb9a8a1575a Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Fri, 31 May 2024 03:26:31 +0100
Subject: [PATCH 08/10] Fix formatting.
---
clang/lib/CodeGen/CodeGenModule.cpp | 9 ++++-----
1 file changed, 4 insertions(+), 5 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index b59c5c52783e9..c12a2f4719c5e 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2065,11 +2065,10 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
ctor.finishAndAddTo(ctors);
}
- auto list =
- ctors.finishAndCreateGlobal(GlobalName, getPointerAlign(),
- /*constant*/ false,
- llvm::GlobalValue::AppendingLinkage,
- GlobalsInt8PtrTy->getAddressSpace());
+ auto list = ctors.finishAndCreateGlobal(GlobalName, getPointerAlign(),
+ /*constant*/ false,
+ llvm::GlobalValue::AppendingLinkage,
+ GlobalsInt8PtrTy->getAddressSpace());
// The LTO linker doesn't seem to like it when we set an alignment
// on appending variables. Take it off as a workaround.
>From 1f42f8264f8f7b0a394e2362f8f59da55eb37637 Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Mon, 3 Jun 2024 15:35:24 +0100
Subject: [PATCH 09/10] Switch to unconditional Unqual use.
---
clang/lib/CodeGen/CodeGenModule.cpp | 17 +++++--
clang/test/CodeGen/asan-constructor.c | 2 +-
clang/test/CodeGen/attr-retain.c | 1 +
clang/test/CodeGenCUDA/llvm-used.cu | 2 +-
...ember-variable-explicit-specialization.cpp | 4 +-
.../amdgcn_target_global_constructor.cpp | 4 +-
llvm/docs/LangRef.rst | 48 +++++++++++--------
llvm/lib/Transforms/Utils/ModuleUtils.cpp | 11 +++--
.../remove-ifunc-program-addrspace.ll | 4 +-
9 files changed, 54 insertions(+), 39 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index c12a2f4719c5e..34f32af968b12 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2046,10 +2046,16 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
llvm::FunctionType* CtorFTy = llvm::FunctionType::get(VoidTy, false);
llvm::Type *CtorPFTy = llvm::PointerType::get(CtorFTy,
TheModule.getDataLayout().getProgramAddressSpace());
-
- // Get the type of a ctor entry, { i32, program void ()*, global i8* }.
+ llvm::PointerType *AssocDataPtrTy =
+ llvm::PointerType::getUnqual(getLLVMContext());
+
+ // Get the type of a ctor entry, { i32, program void ()*, i8* }.
+ // Note that we unconditionally emit an unqualified pointer to the associated
+ // data - this is intentional as this is a fake global, serving only as a
+ // lifetime extension hook; this must match the type we use in the llvm.used
+ // and llvm.compiler.used arrays.
llvm::StructType *CtorStructTy =
- llvm::StructType::get(Int32Ty, CtorPFTy, GlobalsInt8PtrTy);
+ llvm::StructType::get(Int32Ty, CtorPFTy, AssocDataPtrTy);
// Construct the constructor and destructor arrays.
ConstantInitBuilder builder(*this);
@@ -2059,9 +2065,10 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
ctor.addInt(Int32Ty, I.Priority);
ctor.add(I.Initializer);
if (I.AssociatedData)
- ctor.add(I.AssociatedData);
+ ctor.add(llvm::ConstantExpr::getPointerCast(I.AssociatedData,
+ AssocDataPtrTy));
else
- ctor.addNullPointer(GlobalsInt8PtrTy);
+ ctor.addNullPointer(AssocDataPtrTy);
ctor.finishAndAddTo(ctors);
}
diff --git a/clang/test/CodeGen/asan-constructor.c b/clang/test/CodeGen/asan-constructor.c
index 41212d45a5c63..85bd59a4c75b2 100644
--- a/clang/test/CodeGen/asan-constructor.c
+++ b/clang/test/CodeGen/asan-constructor.c
@@ -4,4 +4,4 @@
// CHECK: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @asan.module_ctor, ptr null }]
// EMSCRIPTEN: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 50, ptr @asan.module_ctor, ptr null }]
-// GLOBALAS: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 1, ptr @asan.module_ctor, ptr addrspace(1) null }]
+// GLOBALAS: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @asan.module_ctor, ptr null }]
diff --git a/clang/test/CodeGen/attr-retain.c b/clang/test/CodeGen/attr-retain.c
index 6a9869c00392d..fc41be76f9d0a 100644
--- a/clang/test/CodeGen/attr-retain.c
+++ b/clang/test/CodeGen/attr-retain.c
@@ -11,6 +11,7 @@
// CHECK: @llvm.used = appending global [8 x ptr] [ptr @c0, ptr @foo.l0, ptr @f0, ptr @f2, ptr @g0, ptr @g1, ptr @g3, ptr @g4], section "llvm.metadata"
// CHECK: @llvm.compiler.used = appending global [3 x ptr] [ptr @f2, ptr @g3, ptr @g4], section "llvm.metadata"
+
const int c0 __attribute__((retain)) = 42;
void foo(void) {
diff --git a/clang/test/CodeGenCUDA/llvm-used.cu b/clang/test/CodeGenCUDA/llvm-used.cu
index 9f22dabde3c2d..c39111dd48036 100644
--- a/clang/test/CodeGenCUDA/llvm-used.cu
+++ b/clang/test/CodeGenCUDA/llvm-used.cu
@@ -1,7 +1,7 @@
// RUN: %clang_cc1 -emit-llvm %s -o - -fcuda-is-device -triple nvptx64-unknown-unknown | FileCheck %s
-// Make sure we emit the proper addrspacecast for llvm.used. PR22383 exposed an
+// Make sure we emit the proper addrspacecast for llvm.used. PR22383 exposed an
// issue where we were generating a bitcast instead of an addrspacecast.
// CHECK: @llvm.compiler.used = appending global [1 x ptr] [ptr addrspacecast (ptr addrspace(1) @a to ptr)], section "llvm.metadata"
diff --git a/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp b/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
index 58055130cd708..2753aaf9f5134 100644
--- a/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
+++ b/clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp
@@ -14,7 +14,7 @@ int f();
// ELF-SELECTANY: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__cxx_global_var_init, ptr @selectany }]
// ELF-SELECTANY: @llvm.used = appending global [1 x ptr] [ptr @selectany]
-// GLOBALAS-ELF-SELECTANY: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @__cxx_global_var_init, ptr addrspace(1) @selectany }]
+// GLOBALAS-ELF-SELECTANY: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__cxx_global_var_init, ptr addrspacecast (ptr addrspace(1) @selectany to ptr) }]
int __declspec(selectany) selectany = f();
#else
@@ -33,7 +33,7 @@ template<> int A<char>::a;
template<> int A<bool>::a = 10;
// ALL: @llvm.global_ctors = appending global [16 x { i32, ptr, ptr }]
-// GLOBALAS-ELF: @llvm.global_ctors = appending addrspace(1) global [16 x { i32, ptr, ptr addrspace(1) }]
+// GLOBALAS-ELF: @llvm.global_ctors = appending addrspace(1) global [16 x { i32, ptr, ptr }]
// ELF: [{ i32, ptr, ptr } { i32 65535, ptr @[[unordered:[^,]*]], ptr @_ZN1AIsE1aE },
// MACHO: [{ i32, ptr, ptr } { i32 65535, ptr @[[unordered:[^,]*]], ptr null },
diff --git a/clang/test/OpenMP/amdgcn_target_global_constructor.cpp b/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
index fe414b59588c0..5d4dec6848970 100644
--- a/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
+++ b/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
@@ -21,8 +21,8 @@ S A;
#endif
//.
-// CHECK: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @_GLOBAL__sub_I_amdgcn_target_global_constructor.cpp, ptr addrspace(1) null }]
-// CHECK: @llvm.global_dtors = appending addrspace(1) global [1 x { i32, ptr, ptr addrspace(1) }] [{ i32, ptr, ptr addrspace(1) } { i32 65535, ptr @__dtor_A, ptr addrspace(1) null }]
+// CHECK: @llvm.global_ctors = appending addrspace(1) global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__sub_I_amdgcn_target_global_constructor.cpp, ptr null }]
+// CHECK: @llvm.global_dtors = appending addrspace(1) global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @__dtor_A, ptr null }]
//.
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index c58f7f7140e47..816f5a7af0de0 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -354,7 +354,7 @@ added in the future:
not be used lightly but only for specific situations such as an
alternative to the *register pinning* performance technique often
used when implementing functional programming languages. At the
- moment only X86, AArch64, and RISCV support this convention. The
+ moment only X86, AArch64, and RISCV support this convention. The
following limitations exist:
- On *X86-32* only up to 4 bit type parameters are supported. No
@@ -685,10 +685,10 @@ implementation defined, the optimizer can't do the latter. The former is
challenging as many commonly expected properties, such as
``ptrtoint(v)-ptrtoint(v) == 0``, don't hold for non-integral types.
Similar restrictions apply to intrinsics that might examine the pointer bits,
-such as :ref:`llvm.ptrmask<int_ptrmask>`.
+such as :ref:`llvm.ptrmask<int_ptrmask>`.
The alignment information provided by the frontend for a non-integral pointer
-(typically using attributes or metadata) must be valid for every possible
+(typically using attributes or metadata) must be valid for every possible
representation of the pointer.
.. _globalvars:
@@ -1649,10 +1649,10 @@ Currently, only the following parameter attributes are defined:
- Both ``a`` and ``b`` are constants.
- The range is allowed to wrap.
- The range should not represent the full or empty set. That is, ``a!=b``.
-
- This attribute may only be applied to parameters or return values with integer
+
+ This attribute may only be applied to parameters or return values with integer
or vector of integer types.
-
+
For vector-typed parameters, the range is applied element-wise.
.. _gc:
@@ -8699,7 +8699,10 @@ functions with the same priority is not defined.
If the third field is non-null, and points to a global variable
or function, the initializer function will only run if the associated
-data from the current module is not discarded.
+data from the current module is not discarded. An unqualified pointer is used
+unconditionally to underline the ephemeral nature of this field - it is
+essentially a lifetime hook meant for internal compiler usage, just like the
+elements of the ``@llvm.used`` and ``@llvm.compiler.used`` arrays.
On ELF the referenced global variable or function must be in a comdat.
.. _llvmglobaldtors:
@@ -8720,7 +8723,10 @@ order of functions with the same priority is not defined.
If the third field is non-null, and points to a global variable
or function, the destructor function will only run if the associated
-data from the current module is not discarded.
+data from the current module is not discarded. An unqualified pointer is used
+unconditionally to underline the ephemeral nature of this field - it is
+essentially a lifetime hook meant for internal compiler usage, just like the
+elements of the ``@llvm.used`` and ``@llvm.compiler.used`` arrays.
On ELF the referenced global variable or function must be in a comdat.
Instruction Reference
@@ -14254,7 +14260,7 @@ Arguments:
""""""""""
The first 4 arguments are similar to ``llvm.instrprof.increment``. The indexing
is specific to callsites, meaning callsites are indexed from 0, independent from
-the indexes used by the other intrinsics (such as
+the indexes used by the other intrinsics (such as
``llvm.instrprof.increment[.step]``).
The last argument is the called value of the callsite this intrinsic precedes.
@@ -14268,7 +14274,7 @@ a buffer LLVM can use to perform counter increments (i.e. the lowering of
``llvm.instrprof.increment[.step]``. The address range following the counter
buffer, ``<num-counters>`` x ``sizeof(ptr)`` - sized, is expected to contain
pointers to contexts of functions called from this function ("subcontexts").
-LLVM does not dereference into that memory region, just calculates GEPs.
+LLVM does not dereference into that memory region, just calculates GEPs.
The lowering of ``llvm.instrprof.callsite`` consists of:
@@ -14888,8 +14894,8 @@ integer bit width or any vector of integer elements.
Overview:
"""""""""
-Return ``-1`` if ``%a`` is signed less than ``%b``, ``0`` if they are equal, and
-``1`` if ``%a`` is signed greater than ``%b``. Vector intrinsics operate on a per-element basis.
+Return ``-1`` if ``%a`` is signed less than ``%b``, ``0`` if they are equal, and
+``1`` if ``%a`` is signed greater than ``%b``. Vector intrinsics operate on a per-element basis.
Arguments:
""""""""""
@@ -14917,8 +14923,8 @@ integer bit width or any vector of integer elements.
Overview:
"""""""""
-Return ``-1`` if ``%a`` is unsigned less than ``%b``, ``0`` if they are equal, and
-``1`` if ``%a`` is unsigned greater than ``%b``. Vector intrinsics operate on a per-element basis.
+Return ``-1`` if ``%a`` is unsigned less than ``%b``, ``0`` if they are equal, and
+``1`` if ``%a`` is unsigned greater than ``%b``. Vector intrinsics operate on a per-element basis.
Arguments:
""""""""""
@@ -20980,9 +20986,9 @@ Semantics:
""""""""""
The '``llvm.vp.minimum``' intrinsic performs floating-point minimum (:ref:`minimum <i_minimum>`)
-of the first and second vector operand on each enabled lane, the result being
+of the first and second vector operand on each enabled lane, the result being
NaN if either operand is a NaN. -0.0 is considered to be less than +0.0 for this
-intrinsic. The result on disabled lanes is a :ref:`poison value <poisonvalues>`.
+intrinsic. The result on disabled lanes is a :ref:`poison value <poisonvalues>`.
The operation is performed in the default floating-point environment.
Examples:
@@ -21030,9 +21036,9 @@ Semantics:
""""""""""
The '``llvm.vp.maximum``' intrinsic performs floating-point maximum (:ref:`maximum <i_maximum>`)
-of the first and second vector operand on each enabled lane, the result being
+of the first and second vector operand on each enabled lane, the result being
NaN if either operand is a NaN. -0.0 is considered to be less than +0.0 for this
-intrinsic. The result on disabled lanes is a :ref:`poison value <poisonvalues>`.
+intrinsic. The result on disabled lanes is a :ref:`poison value <poisonvalues>`.
The operation is performed in the default floating-point environment.
Examples:
@@ -28309,7 +28315,7 @@ Semantics:
""""""""""
The intrinsic ``@llvm.allow.ubsan.check()`` returns either ``true`` or
-``false``, depending on compiler options.
+``false``, depending on compiler options.
For each evaluation of a call to this intrinsic, the program must be valid and
correct both if it returns ``true`` and if it returns ``false``.
@@ -28368,13 +28374,13 @@ Semantics:
""""""""""
The intrinsic ``@llvm.allow.runtime.check()`` returns either ``true`` or
-``false``, depending on compiler options.
+``false``, depending on compiler options.
For each evaluation of a call to this intrinsic, the program must be valid and
correct both if it returns ``true`` and if it returns ``false``.
When used in a branch condition, it allows us to choose between
-two alternative correct solutions for the same problem.
+two alternative correct solutions for the same problem.
If the intrinsic is evaluated as ``true``, program should execute a guarded
check. If the intrinsic is evaluated as ``false``, the program should avoid any
diff --git a/llvm/lib/Transforms/Utils/ModuleUtils.cpp b/llvm/lib/Transforms/Utils/ModuleUtils.cpp
index 7151cc869bc95..f17d3edbf0e81 100644
--- a/llvm/lib/Transforms/Utils/ModuleUtils.cpp
+++ b/llvm/lib/Transforms/Utils/ModuleUtils.cpp
@@ -30,8 +30,9 @@ static void appendToGlobalArray(StringRef ArrayName, Module &M, Function *F,
int Priority, Constant *Data) {
IRBuilder<> IRB(M.getContext());
FunctionType *FnTy = FunctionType::get(IRB.getVoidTy(), false);
- PointerType *GlobalsInt8PtrTy = PointerType::get(
- M.getContext(), M.getDataLayout().getDefaultGlobalsAddressSpace());
+ // The pointer to associated data MUST be emitted as an unqualified ptr, see
+ // the comment in CodeGenModule::EmitCtorList().
+ PointerType *AssocDataPtrTy = PointerType::getUnqual(M.getContext());
// Get the current set of static global constructors and add the new ctor
// to the list.
@@ -49,15 +50,15 @@ static void appendToGlobalArray(StringRef ArrayName, Module &M, Function *F,
} else {
EltTy = StructType::get(IRB.getInt32Ty(),
PointerType::get(FnTy, F->getAddressSpace()),
- GlobalsInt8PtrTy);
+ AssocDataPtrTy);
}
// Build a 3 field global_ctor entry. We don't take a comdat key.
Constant *CSVals[3];
CSVals[0] = IRB.getInt32(Priority);
CSVals[1] = F;
- CSVals[2] = Data ? ConstantExpr::getPointerCast(Data, GlobalsInt8PtrTy)
- : Constant::getNullValue(GlobalsInt8PtrTy);
+ CSVals[2] = Data ? ConstantExpr::getPointerCast(Data, AssocDataPtrTy)
+ : Constant::getNullValue(AssocDataPtrTy);
Constant *RuntimeCtorInit =
ConstantStruct::get(EltTy, ArrayRef(CSVals, EltTy->getNumElements()));
diff --git a/llvm/test/tools/llvm-reduce/remove-ifunc-program-addrspace.ll b/llvm/test/tools/llvm-reduce/remove-ifunc-program-addrspace.ll
index 7ddab0b6a815e..9573c55030242 100644
--- a/llvm/test/tools/llvm-reduce/remove-ifunc-program-addrspace.ll
+++ b/llvm/test/tools/llvm-reduce/remove-ifunc-program-addrspace.ll
@@ -15,8 +15,8 @@ define void @existing_ctor() addrspace(1) {
; CHECK-FINAL: [[TABLE:@[0-9]+]] = internal addrspace(2) global [6 x ptr addrspace(1)] poison, align 8
-; CHECK-FINAL: @llvm.global_ctors = appending addrspace(2) global [2 x { i32, ptr addrspace(1), ptr addrspace(2) }] [{ i32, ptr addrspace(1), ptr addrspace(2) } { i32 0, ptr addrspace(1) @existing_ctor, ptr addrspace(2) null }, { i32, ptr addrspace(1), ptr addrspace(2) } { i32 10, ptr addrspace(1) [[TABLE_CTOR:@[0-9]+]], ptr addrspace(2) null }]
- at llvm.global_ctors = appending global [1 x { i32, ptr addrspace(1), ptr addrspace(2) }] [{ i32, ptr addrspace(1), ptr addrspace(2) } { i32 0, ptr addrspace(1) @existing_ctor, ptr addrspace(2) null }]
+; CHECK-FINAL: @llvm.global_ctors = appending addrspace(2) global [2 x { i32, ptr addrspace(1), ptr }] [{ i32, ptr addrspace(1), ptr } { i32 0, ptr addrspace(1) @existing_ctor, ptr null }, { i32, ptr addrspace(1), ptr } { i32 10, ptr addrspace(1) [[TABLE_CTOR:@[0-9]+]], ptr null }]
+ at llvm.global_ctors = appending addrspace(2) global [1 x { i32, ptr addrspace(1), ptr }] [{ i32, ptr addrspace(1), ptr } { i32 0, ptr addrspace(1) @existing_ctor, ptr null }]
>From f89c5bf2e886b2181b4ebb789a33230ed4dcfbee Mon Sep 17 00:00:00 2001
From: Alex Voicu <alexandru.voicu at amd.com>
Date: Mon, 3 Jun 2024 15:46:03 +0100
Subject: [PATCH 10/10] Fix formatting.
---
clang/lib/CodeGen/CodeGenModule.cpp | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 34f32af968b12..33019d03161b4 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -2065,8 +2065,8 @@ void CodeGenModule::EmitCtorList(CtorList &Fns, const char *GlobalName) {
ctor.addInt(Int32Ty, I.Priority);
ctor.add(I.Initializer);
if (I.AssociatedData)
- ctor.add(llvm::ConstantExpr::getPointerCast(I.AssociatedData,
- AssocDataPtrTy));
+ ctor.add(
+ llvm::ConstantExpr::getPointerCast(I.AssociatedData, AssocDataPtrTy));
else
ctor.addNullPointer(AssocDataPtrTy);
ctor.finishAndAddTo(ctors);
More information about the cfe-commits
mailing list