[Clang] Undef attribute for global variables

Summary:
[Clang] Attribute to allow defining undef global variables

Initializing global variables is very cheap on hosted implementations. The
C semantics of zero initializing globals work very well there. It is not
necessarily cheap on freestanding implementations. Where there is no loader
available, code must be emitted near the start point to write the appropriate
values into memory.

At present, external variables can be declared in C++ and definitions provided
in assembly (or IR) to achive this effect. This patch provides an attribute in
order to remove this reason for writing assembly for performance sensitive
freestanding implementations.

A close analogue in tree is LDS memory for amdgcn, where the kernel is
responsible for initializing the memory after it starts executing on the gpu.
Uninitalized variables in LDS are observably cheaper than zero initialized.

Patch is loosely based on the cuda __shared__ and opencl __local variable
implementation which also produces undef global variables.

Reviewers: kcc, rjmccall, rsmith, glider, vitalybuka, pcc, eugenis, vlad.tsyrklevich, jdoerfert, gregrodgers, jfb, aaron.ballman

Reviewed By: rjmccall, aaron.ballman

Subscribers: Anastasia, aaron.ballman, davidb, Quuxplusone, dexonsmith, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D74361

clang/include/clang/Basic/Attr.td

@@ -3313,6 +3313,12 @@ def Uninitialized : InheritableAttr {
   let Documentation = [UninitializedDocs];
 }
 
+def LoaderUninitialized : Attr {
+  let Spellings = [Clang<"loader_uninitialized">];
+  let Subjects = SubjectList<[GlobalVar]>;
+  let Documentation = [LoaderUninitializedDocs];
+}
+
 def ObjCExternallyRetained : InheritableAttr {
   let LangOpts = [ObjCAutoRefCount];
   let Spellings = [Clang<"objc_externally_retained">];

clang/include/clang/Basic/AttrDocs.td

@@ -4358,6 +4358,29 @@ it rather documents the programmer's intent.
   }];
 }
 
+def LoaderUninitializedDocs : Documentation {
+  let Category = DocCatVariable;
+  let Content = [{
+The ``loader_uninitialized`` attribute can be placed on global variables to
+indicate that the variable does not need to be zero initialized by the loader.
+On most targets, zero-initialization does not incur any additional cost.
+For example, most general purpose operating systems deliberately ensure
+that all memory is properly initialized in order to avoid leaking privileged
+information from the kernel or other programs. However, some targets
+do not make this guarantee, and on these targets, avoiding an unnecessary
+zero-initialization can have a significant impact on load times and/or code
+size.
+
+A declaration with this attribute is a non-tentative definition just as if it
+provided an initializer. Variables with this attribute are considered to be
+uninitialized in the same sense as a local variable, and the programs must
+write to them before reading from them. If the variable's type is a C++ class
+type with a non-trivial default constructor, or an array thereof, this attribute
+only suppresses the static zero-initialization of the variable, not the dynamic
+initialization provided by executing the default constructor.
+  }];
+}
+
 def CallbackDocs : Documentation {
   let Category = DocCatFunction;
   let Content = [{

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -5333,6 +5333,17 @@ def ext_aggregate_init_not_constant : Extension<
   "initializer for aggregate is not a compile-time constant">, InGroup<C99>;
 def err_local_cant_init : Error<
   "'__local' variable cannot have an initializer">;
+def err_loader_uninitialized_cant_init
+    : Error<"variable with 'loader_uninitialized' attribute cannot have an "
+            "initializer">;
+def err_loader_uninitialized_trivial_ctor
+    : Error<"variable with 'loader_uninitialized' attribute must have a "
+            "trivial default constructor">;
+def err_loader_uninitialized_redeclaration
+    : Error<"redeclaration cannot add 'loader_uninitialized' attribute">;
+def err_loader_uninitialized_extern_decl
+    : Error<"variable %0 cannot be declared both 'extern' and with the "
+            "'loader_uninitialized' attribute">;
 def err_block_extern_cant_init : Error<
   "'extern' variable cannot have an initializer">;
 def warn_extern_init : Warning<"'extern' variable has an initializer">,

clang/lib/AST/DeclBase.cpp

@@ -454,14 +454,17 @@ ExternalSourceSymbolAttr *Decl::getExternalSourceSymbolAttr() const {
 }
 
 bool Decl::hasDefiningAttr() const {
-  return hasAttr<AliasAttr>() || hasAttr<IFuncAttr>();
+  return hasAttr<AliasAttr>() || hasAttr<IFuncAttr>() ||
+         hasAttr<LoaderUninitializedAttr>();
 }
 
 const Attr *Decl::getDefiningAttr() const {
   if (auto *AA = getAttr<AliasAttr>())
     return AA;
   if (auto *IFA = getAttr<IFuncAttr>())
     return IFA;
+  if (auto *NZA = getAttr<LoaderUninitializedAttr>())
+    return NZA;
   return nullptr;
 }
 

clang/lib/CodeGen/CGDecl.cpp

@@ -249,7 +249,7 @@ llvm::Constant *CodeGenModule::getOrCreateStaticVarDecl(
   // variables cannot have an initializer.
   llvm::Constant *Init = nullptr;
   if (Ty.getAddressSpace() == LangAS::opencl_local ||
-      D.hasAttr<CUDASharedAttr>())
+      D.hasAttr<CUDASharedAttr>() || D.hasAttr<LoaderUninitializedAttr>())
     Init = llvm::UndefValue::get(LTy);
   else
     Init = EmitNullConstant(Ty);

clang/lib/CodeGen/CodeGenModule.cpp

@@ -3952,6 +3952,8 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
   if (getLangOpts().CUDA &&
       (IsCUDASharedVar || IsCUDAShadowVar || IsHIPPinnedShadowVar))
     Init = llvm::UndefValue::get(getTypes().ConvertType(ASTTy));
+  else if (D->hasAttr<LoaderUninitializedAttr>())
+    Init = llvm::UndefValue::get(getTypes().ConvertType(ASTTy));
   else if (!InitExpr) {
     // This is a tentative definition; tentative definitions are
     // implicitly initialized with { 0 }.

clang/lib/Sema/SemaDecl.cpp

@@ -2712,6 +2712,18 @@ static void checkNewAttributesAfterDef(Sema &S, Decl *New, const Decl *Old) {
         --E;
         continue;
       }
+    } else if (isa<LoaderUninitializedAttr>(NewAttribute)) {
+      // If there is a C definition followed by a redeclaration with this
+      // attribute then there are two different definitions. In C++, prefer the
+      // standard diagnostics.
+      if (!S.getLangOpts().CPlusPlus) {
+        S.Diag(NewAttribute->getLocation(),
+               diag::err_loader_uninitialized_redeclaration);
+        S.Diag(Def->getLocation(), diag::note_previous_definition);
+        NewAttributes.erase(NewAttributes.begin() + I);
+        --E;
+        continue;
+      }
     } else if (isa<SelectAnyAttr>(NewAttribute) &&
                cast<VarDecl>(New)->isInline() &&
                !cast<VarDecl>(New)->isInlineSpecified()) {
@@ -11915,6 +11927,13 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init, bool DirectInit) {
     return;
   }
 
+  // The LoaderUninitialized attribute acts as a definition (of undef).
+  if (VDecl->hasAttr<LoaderUninitializedAttr>()) {
+    Diag(VDecl->getLocation(), diag::err_loader_uninitialized_cant_init);
+    VDecl->setInvalidDecl();
+    return;
+  }
+
   // Get the decls type and save a reference for later, since
   // CheckInitializerTypes may change it.
   QualType DclT = VDecl->getType(), SavT = DclT;
@@ -12328,6 +12347,22 @@ void Sema::ActOnUninitializedDecl(Decl *RealDecl) {
       return;
     }
 
+    if (!Var->isInvalidDecl() && RealDecl->hasAttr<LoaderUninitializedAttr>()) {
+      if (CXXRecordDecl *RD = Var->getType()->getAsCXXRecordDecl()) {
+        if (!RD->hasTrivialDefaultConstructor()) {
+          Diag(Var->getLocation(), diag::err_loader_uninitialized_trivial_ctor);
+          Var->setInvalidDecl();
+          return;
+        }
+      }
+      if (Var->getStorageClass() == SC_Extern) {
+        Diag(Var->getLocation(), diag::err_loader_uninitialized_extern_decl)
+            << Var;
+        Var->setInvalidDecl();
+        return;
+      }
+    }
+
     VarDecl::DefinitionKind DefKind = Var->isThisDeclarationADefinition();
     if (!Var->isInvalidDecl() && DefKind != VarDecl::DeclarationOnly &&
         Var->getType().hasNonTrivialToPrimitiveDefaultInitializeCUnion())

clang/lib/Sema/SemaDeclAttr.cpp

@@ -7467,6 +7467,10 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
     handleUninitializedAttr(S, D, AL);
     break;
 
+  case ParsedAttr::AT_LoaderUninitialized:
+    handleSimpleAttribute<LoaderUninitializedAttr>(S, D, AL);
+    break;
+
   case ParsedAttr::AT_ObjCExternallyRetained:
     handleObjCExternallyRetainedAttr(S, D, AL);
     break;

clang/test/CodeGen/attr-loader-uninitialized.c

@@ -0,0 +1,24 @@
+// RUN: %clang_cc1 -emit-llvm %s -o - | FileCheck %s
+
+// CHECK: @tentative_attr_first = global i32 undef, align 4
+int tentative_attr_first __attribute__((loader_uninitialized));
+int tentative_attr_first;
+
+// CHECK: @tentative_attr_second = global i32 undef, align 4
+int tentative_attr_second;
+int tentative_attr_second __attribute__((loader_uninitialized));
+
+// CHECK: @array = global [16 x float] undef, align 16
+float array[16] __attribute__((loader_uninitialized));
+
+typedef struct
+{
+  int x;
+  float y;
+} s;
+
+// CHECK: @i = global %struct.s undef, align 4
+s i __attribute__((loader_uninitialized));
+
+// CHECK: @private_extern_ok = hidden global i32 undef, align 4
+__private_extern__ int private_extern_ok __attribute__((loader_uninitialized));

clang/test/CodeGenCXX/attr-loader-uninitialized.cpp

@@ -0,0 +1,29 @@
+// RUN: %clang_cc1 -emit-llvm -o - %s | FileCheck %s
+
+// CHECK: @defn = global i32 undef
+int defn  [[clang::loader_uninitialized]];
+
+// CHECK: @_ZL11defn_static = internal global i32 undef
+static int defn_static [[clang::loader_uninitialized]] __attribute__((used));
+
+// CHECK: @_ZZ4funcvE4data = internal global i32 undef
+int* func(void)
+{
+  static int data [[clang::loader_uninitialized]];
+  return &data;
+}
+
+class trivial
+{
+  float x;
+};
+
+// CHECK: @ut = global %class.trivial undef
+trivial ut [[clang::loader_uninitialized]];
+
+// CHECK: @arr = global [32 x double] undef, align 16
+double arr[32] __attribute__((loader_uninitialized));
+
+// Defining as arr2[] [[clang..]] raises the error: attribute cannot be applied to types
+// CHECK: @arr2 = global [4 x double] undef, align 16
+double arr2 [[clang::loader_uninitialized]] [4];

clang/test/Misc/pragma-attribute-supported-attributes-list.test

@@ -65,6 +65,7 @@
 // CHECK-NEXT: InitPriority (SubjectMatchRule_variable)
 // CHECK-NEXT: InternalLinkage (SubjectMatchRule_variable, SubjectMatchRule_function, SubjectMatchRule_record)
 // CHECK-NEXT: LTOVisibilityPublic (SubjectMatchRule_record)
+// CHECK-NEXT: LoaderUninitialized (SubjectMatchRule_variable_is_global)
 // CHECK-NEXT: Lockable (SubjectMatchRule_record)
 // CHECK-NEXT: MIGServerRoutine (SubjectMatchRule_function, SubjectMatchRule_objc_method, SubjectMatchRule_block)
 // CHECK-NEXT: MSStruct (SubjectMatchRule_record)

clang/test/Sema/attr-loader-uninitialized.c

@@ -0,0 +1,37 @@
+// RUN: %clang_cc1 %s -verify -fsyntax-only
+// See also attr-loader-uninitialized.cpp
+
+int good __attribute__((loader_uninitialized));
+static int local_ok __attribute__((loader_uninitialized));
+int hidden_ok __attribute__((visibility("hidden"))) __attribute__((loader_uninitialized));
+
+const int can_still_be_const __attribute__((loader_uninitialized));
+
+extern int external_rejected __attribute__((loader_uninitialized));
+// expected-error@-1 {{variable 'external_rejected' cannot be declared both 'extern' and with the 'loader_uninitialized' attribute}}
+
+int noargs __attribute__((loader_uninitialized(0)));
+// expected-error@-1 {{'loader_uninitialized' attribute takes no arguments}}
+
+int init_rejected __attribute__((loader_uninitialized)) = 42;
+// expected-error@-1 {{variable with 'loader_uninitialized' attribute cannot have an initializer}}
+
+int declaration_then_uninit_ok;
+int declaration_then_uninit_ok __attribute__((loader_uninitialized));
+
+int definition_then_uninit_rejected = 0;
+int definition_then_uninit_rejected __attribute__((loader_uninitialized));
+// expected-error@-1 {{redeclaration cannot add 'loader_uninitialized' attribute}}
+// expected-note@-3 {{previous definition is here}}
+
+int tentative_repeated_ok __attribute__((loader_uninitialized));
+int tentative_repeated_ok __attribute__((loader_uninitialized));
+
+__private_extern__ int private_extern_can_be_initialised = 10;
+__private_extern__ int therefore_uninit_private_extern_ok __attribute__((loader_uninitialized));
+
+__private_extern__ int initialized_private_extern_rejected __attribute__((loader_uninitialized)) = 5;
+// expected-error@-1 {{variable with 'loader_uninitialized' attribute cannot have an initializer}}
+
+extern __attribute__((visibility("hidden"))) int extern_hidden __attribute__((loader_uninitialized));
+// expected-error@-1 {{variable 'extern_hidden' cannot be declared both 'extern' and with the 'loader_uninitialized' attribute}}

clang/test/Sema/attr-loader-uninitialized.cpp

@@ -0,0 +1,60 @@
+// RUN: %clang_cc1 %s -verify -fsyntax-only
+
+int good __attribute__((loader_uninitialized));
+static int local_ok __attribute__((loader_uninitialized));
+int hidden_ok __attribute__((visibility("hidden"))) __attribute__((loader_uninitialized));
+
+const int still_cant_be_const __attribute__((loader_uninitialized));
+// expected-error@-1 {{default initialization of an object of const type}}
+extern int external_rejected __attribute__((loader_uninitialized));
+// expected-error@-1 {{variable 'external_rejected' cannot be declared both 'extern' and with the 'loader_uninitialized' attribute}}
+
+int noargs __attribute__((loader_uninitialized(0)));
+// expected-error@-1 {{'loader_uninitialized' attribute takes no arguments}}
+
+int init_rejected __attribute__((loader_uninitialized)) = 42;
+// expected-error@-1 {{variable with 'loader_uninitialized' attribute cannot have an initializer}}
+
+void func() __attribute__((loader_uninitialized))
+// expected-warning@-1 {{'loader_uninitialized' attribute only applies to global variables}}
+{
+  int local __attribute__((loader_uninitialized));
+  // expected-warning@-1 {{'loader_uninitialized' attribute only applies to global variables}}
+
+  static int sl __attribute__((loader_uninitialized));
+}
+
+struct s {
+  __attribute__((loader_uninitialized)) int field;
+  // expected-warning@-1 {{'loader_uninitialized' attribute only applies to global variables}}
+
+  static __attribute__((loader_uninitialized)) int sfield;
+
+} __attribute__((loader_uninitialized));
+// expected-warning@-1 {{'loader_uninitialized' attribute only applies to global variables}}
+
+int redef_attr_first __attribute__((loader_uninitialized));
+int redef_attr_first;
+// expected-error@-1 {{redefinition of 'redef_attr_first'}}
+// expected-note@-3 {{previous definition is here}}
+
+int redef_attr_second;
+int redef_attr_second __attribute__((loader_uninitialized));
+// expected-warning@-1 {{attribute declaration must precede definition}}
+// expected-note@-3 {{previous definition is here}}
+// expected-error@-3 {{redefinition of 'redef_attr_second'}}
+// expected-note@-5 {{previous definition is here}}
+
+struct trivial {};
+
+trivial default_ok __attribute__((loader_uninitialized));
+trivial value_rejected  __attribute__((loader_uninitialized)) {};
+// expected-error@-1 {{variable with 'loader_uninitialized' attribute cannot have an initializer}}
+
+struct nontrivial
+{
+  nontrivial() {}
+};
+
+nontrivial needs_trivial_ctor __attribute__((loader_uninitialized));
+// expected-error@-1 {{variable with 'loader_uninitialized' attribute must have a trivial default constructor}}