[OpenMP] Make Ctor / Dtor functions have external visibility

The default construction of constructor functions by LLVM tends to make
them have internal linkage. When we call a ctor / dtor function in the
target region we are actually creating a kernel that is called at
registration. Because the ctor is a kernel we need to make sure it's
externally visible so we can actually call it. This prevented AMDGPU
from correctly using constructors while NVPTX could use them simply
because it ignored internal visibility.

Reviewed By: JonChesterfield

Differential Revision: https://reviews.llvm.org/D122504

diff --git a/clang/lib/CodeGen/CGDeclCXX.cpp b/clang/lib/CodeGen/CGDeclCXX.cpp
index 7b880c1354e190f18edae0c86961e93097e1e608..52d5b27eb01cb4c412378ae5b8aed3f29d3ad872 100644 (file)
--- a/clang/lib/CodeGen/CGDeclCXX.cpp
+++ b/clang/lib/CodeGen/CGDeclCXX.cpp
@@ -425,9 +425,8 @@ void CodeGenFunction::EmitCXXGuardedInitBranch(llvm::Value *NeedsInit,
 
 llvm::Function *CodeGenModule::CreateGlobalInitOrCleanUpFunction(
     llvm::FunctionType *FTy, const Twine &Name, const CGFunctionInfo &FI,
-    SourceLocation Loc, bool TLS) {
-  llvm::Function *Fn = llvm::Function::Create(
-      FTy, llvm::GlobalValue::InternalLinkage, Name, &getModule());
+    SourceLocation Loc, bool TLS, llvm::GlobalVariable::LinkageTypes Linkage) {
+  llvm::Function *Fn = llvm::Function::Create(FTy, Linkage, Name, &getModule());
 
   if (!getLangOpts().AppleKext && !TLS) {
     // Set the section if needed.
@@ -435,7 +434,8 @@ llvm::Function *CodeGenModule::CreateGlobalInitOrCleanUpFunction(
       Fn->setSection(Section);
   }
 
-  SetInternalFunctionAttributes(GlobalDecl(), Fn, FI);
+  if (Linkage == llvm::GlobalVariable::InternalLinkage)
+    SetInternalFunctionAttributes(GlobalDecl(), Fn, FI);
 
   Fn->setCallingConv(getRuntimeCC());
 

diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 84527eb884cfd5f52e7c7c95ff79d6142ad1d6b7..eeefa608fb7ee54590b8a77ace6ca43c6541ea37 100644 (file)
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -1926,7 +1926,8 @@ bool CGOpenMPRuntime::emitDeclareTargetVarDefinition(const VarDecl *VD,
       const CGFunctionInfo &FI = CGM.getTypes().arrangeNullaryFunction();
       llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
       llvm::Function *Fn = CGM.CreateGlobalInitOrCleanUpFunction(
-          FTy, Twine(Buffer, "_ctor"), FI, Loc);
+          FTy, Twine(Buffer, "_ctor"), FI, Loc, false,
+          llvm::GlobalValue::WeakODRLinkage);
       auto NL = ApplyDebugLocation::CreateEmpty(CtorCGF);
       CtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI,
                             FunctionArgList(), Loc, Loc);
@@ -1944,7 +1945,6 @@ bool CGOpenMPRuntime::emitDeclareTargetVarDefinition(const VarDecl *VD,
       CtorCGF.FinishFunction();
       Ctor = Fn;
       ID = llvm::ConstantExpr::getBitCast(Fn, CGM.Int8PtrTy);
-      CGM.addUsedGlobal(cast<llvm::GlobalValue>(Ctor));
     } else {
       Ctor = new llvm::GlobalVariable(
           CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,
@@ -1970,7 +1970,8 @@ bool CGOpenMPRuntime::emitDeclareTargetVarDefinition(const VarDecl *VD,
       const CGFunctionInfo &FI = CGM.getTypes().arrangeNullaryFunction();
       llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
       llvm::Function *Fn = CGM.CreateGlobalInitOrCleanUpFunction(
-          FTy, Twine(Buffer, "_dtor"), FI, Loc);
+          FTy, Twine(Buffer, "_dtor"), FI, Loc, false,
+          llvm::GlobalValue::WeakODRLinkage);
       auto NL = ApplyDebugLocation::CreateEmpty(DtorCGF);
       DtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI,
                             FunctionArgList(), Loc, Loc);
@@ -1989,7 +1990,6 @@ bool CGOpenMPRuntime::emitDeclareTargetVarDefinition(const VarDecl *VD,
       DtorCGF.FinishFunction();
       Dtor = Fn;
       ID = llvm::ConstantExpr::getBitCast(Fn, CGM.Int8PtrTy);
-      CGM.addUsedGlobal(cast<llvm::GlobalValue>(Dtor));
     } else {
       Dtor = new llvm::GlobalVariable(
           CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,

diff --git a/clang/lib/CodeGen/CodeGenModule.h b/clang/lib/CodeGen/CodeGenModule.h
index 1fcd5d4d808a1cab51b3cce98c7b01af86900604..b04393e38cf7499df3f69c21435460205c9d06de 100644 (file)
--- a/clang/lib/CodeGen/CodeGenModule.h
+++ b/clang/lib/CodeGen/CodeGenModule.h
@@ -826,7 +826,9 @@ public:
 
   llvm::Function *CreateGlobalInitOrCleanUpFunction(
       llvm::FunctionType *ty, const Twine &name, const CGFunctionInfo &FI,
-      SourceLocation Loc = SourceLocation(), bool TLS = false);
+      SourceLocation Loc = SourceLocation(), bool TLS = false,
+      llvm::GlobalVariable::LinkageTypes Linkage =
+          llvm::GlobalVariable::InternalLinkage);
 
   /// Return the AST address space of the underlying global variable for D, as
   /// determined by its declaration. Normally this is the same as the address

diff --git a/clang/test/OpenMP/amdgcn_target_global_constructor.cpp b/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
index 830497a661850b13af4af21dbb71b4dd6f24af8a..35bebaafdf54475fbdcb697ad571c81fff0ce182 100644 (file)
--- a/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
+++ b/clang/test/OpenMP/amdgcn_target_global_constructor.cpp
@@ -26,9 +26,8 @@ S A;
 // CHECK: @__omp_rtl_assume_threads_oversubscription = weak_odr hidden addrspace(1) constant i32 0
 // CHECK: @__omp_rtl_assume_no_thread_state = weak_odr hidden addrspace(1) constant i32 0
 // CHECK: @A = addrspace(1) global %struct.S zeroinitializer, align 4
-// CHECK: @llvm.used = appending addrspace(1) global [2 x i8*] [i8* bitcast (void ()* @__omp_offloading_{{.*}}_ctor to i8*), i8* bitcast (void ()* @__omp_offloading_{{.*}}_dtor to i8*)], section "llvm.metadata"
 //.
-// CHECK-LABEL: define {{[^@]+}}@__omp_offloading_{{.*}}_ctor
+// CHECK-LABEL: define {{[^@]+}}@__omp_offloading__fd02_613a0d56_A_l19_ctor
 // CHECK-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    call void @_ZN1SC1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) addrspacecast ([[STRUCT_S:%.*]] addrspace(1)* @A to %struct.S*)) #[[ATTR3:[0-9]+]]
@@ -46,7 +45,7 @@ S A;
 // CHECK-NEXT:    ret void
 //
 //
-// CHECK-LABEL: define {{[^@]+}}@__omp_offloading_{{.*}}_dtor
+// CHECK-LABEL: define {{[^@]+}}@__omp_offloading__fd02_613a0d56_A_l19_dtor
 // CHECK-SAME: () #[[ATTR0]] {
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    call void @_ZN1SD1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) addrspacecast ([[STRUCT_S:%.*]] addrspace(1)* @A to %struct.S*)) #[[ATTR4:[0-9]+]]
@@ -87,19 +86,19 @@ S A;
 // CHECK-NEXT:    ret void
 //
 //.
-// CHECK: attributes #0 = { convergent
-// CHECK: attributes #1 = { convergent
-// CHECK: attributes #2 = { convergent
-// CHECK: attributes #3 = { convergent
-// CHECK: attributes #4 = { convergent
+// CHECK: attributes #0 = { nounwind "kernel" "min-legal-vector-width"="0" }
+// CHECK: attributes #1 = { convergent noinline nounwind optnone "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+// CHECK: attributes #2 = { convergent "frame-pointer"="none" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+// CHECK: attributes #3 = { convergent }
+// CHECK: attributes #4 = { convergent nounwind }
 //.
-// CHECK-DAG: !{i32 0, {{.*}}, !"__omp_offloading_{{.*}}_ctor", i32 19, i32 1}
-// CHECK-DAG: !{i32 0, {{.*}}, !"__omp_offloading_{{.*}}_dtor", i32 19, i32 2}
-// CHECK-DAG: !{i32 1, !"A", i32 0, i32 0}
-// CHECK-DAG: !{void ()* @__omp_offloading_{{.*}}_ctor, !"kernel", i32 1}
-// CHECK-DAG: !{void ()* @__omp_offloading_{{.*}}_dtor, !"kernel", i32 1}
-// CHECK-DAG: !{i32 1, !"wchar_size", i32 4}
-// CHECK-DAG: !{i32 7, !"openmp", i32 50}
-// CHECK-DAG: !{i32 7, !"openmp-device", i32 50}
-// CHECK-DAG: !{!"clang version
+// CHECK: !0 = !{i32 0, i32 64770, i32 1631194454, !"__omp_offloading__fd02_613a0d56_A_l19_ctor", i32 19, i32 1}
+// CHECK: !1 = !{i32 0, i32 64770, i32 1631194454, !"__omp_offloading__fd02_613a0d56_A_l19_dtor", i32 19, i32 2}
+// CHECK: !2 = !{i32 1, !"A", i32 0, i32 0}
+// CHECK: !3 = !{void ()* @__omp_offloading__fd02_613a0d56_A_l19_ctor, !"kernel", i32 1}
+// CHECK: !4 = !{void ()* @__omp_offloading__fd02_613a0d56_A_l19_dtor, !"kernel", i32 1}
+// CHECK: !5 = !{i32 1, !"wchar_size", i32 4}
+// CHECK: !6 = !{i32 7, !"openmp", i32 50}
+// CHECK: !7 = !{i32 7, !"openmp-device", i32 50}
+// CHECK: !8 = !{!"clang version 15.0.0"}
 //.

diff --git a/clang/test/OpenMP/declare_target_codegen.cpp b/clang/test/OpenMP/declare_target_codegen.cpp
index 8a52d55b69f81718270d33afc62dd3ef94240a8d..f64b607da8b6b50ea1d8550008a5bdcdc8cb3fbf 100644 (file)
--- a/clang/test/OpenMP/declare_target_codegen.cpp
+++ b/clang/test/OpenMP/declare_target_codegen.cpp
@@ -46,7 +46,6 @@
 // CHECK-DAG: [[STAT:@.+stat]] = internal global %struct.S zeroinitializer,
 // CHECK-DAG: [[STAT_REF:@.+]] = internal constant %struct.S* [[STAT]]
 // CHECK-DAG: @out_decl_target ={{ protected | }}global i32 0,
-// CHECK-DAG: @llvm.used = appending global [2 x i8*] [i8* bitcast (void ()* @__omp_offloading__{{.+}}_globals_l[[@LINE+84]]_ctor to i8*), i8* bitcast (void ()* @__omp_offloading__{{.+}}_stat_l[[@LINE+85]]_ctor to i8*)],
 // CHECK-DAG: @llvm.compiler.used = appending global [1 x i8*] [i8* bitcast (%struct.S** [[STAT_REF]] to i8*)],
 
 // CHECK-DAG: define {{.*}}i32 @{{.*}}{{foo|bar|baz2|baz3|FA|f_method}}{{.*}}()

diff --git a/clang/test/OpenMP/nvptx_declare_target_var_ctor_dtor_codegen.cpp b/clang/test/OpenMP/nvptx_declare_target_var_ctor_dtor_codegen.cpp
index 45b8372bf64ec37a826af172c4a5e86672a8fce4..7f632c25780fd66bf5a391733a2169cac67fa839 100644 (file)
--- a/clang/test/OpenMP/nvptx_declare_target_var_ctor_dtor_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_declare_target_var_ctor_dtor_codegen.cpp
@@ -44,7 +44,7 @@ int caz() { return 0; }
 static int c = foo() + bar() + baz();
 #pragma omp declare target (c)
 // HOST-DAG: @[[C_CTOR:__omp_offloading__.+_c_l44_ctor]] = private constant i8 0
-// DEVICE-DAG: define internal void [[C_CTOR:@__omp_offloading__.+_c_l44_ctor]]()
+// DEVICE-DAG: define weak_odr void [[C_CTOR:@__omp_offloading__.+_c_l44_ctor]]()
 // DEVICE-DAG: call noundef i32 [[FOO]]()
 // DEVICE-DAG: call noundef i32 [[BAR]]()
 // DEVICE-DAG: call noundef i32 [[BAZ]]()
@@ -61,14 +61,14 @@ struct S {
 S cd = doo() + car() + caz() + baz();
 #pragma omp end declare target
 // HOST-DAG: @[[CD_CTOR:__omp_offloading__.+_cd_l61_ctor]] = private constant i8 0
-// DEVICE-DAG: define internal void [[CD_CTOR:@__omp_offloading__.+_cd_l61_ctor]]()
+// DEVICE-DAG: define weak_odr void [[CD_CTOR:@__omp_offloading__.+_cd_l61_ctor]]()
 // DEVICE-DAG: call noundef i32 [[DOO]]()
 // DEVICE-DAG: call noundef i32 [[CAR]]()
 // DEVICE-DAG: call noundef i32 [[CAZ]]()
 // DEVICE-DAG: ret void
 
 // HOST-DAG: @[[CD_DTOR:__omp_offloading__.+_cd_l61_dtor]] = private constant i8 0
-// DEVICE-DAG: define internal void [[CD_DTOR:@__omp_offloading__.+_cd_l61_dtor]]()
+// DEVICE-DAG: define weak_odr void [[CD_DTOR:@__omp_offloading__.+_cd_l61_dtor]]()
 // DEVICE-DAG: call void
 // DEVICE-DAG: ret void