[clang][SYCL] Add sycl_external attribute and restrict emitting device code (#140282)

This patch is part of the upstreaming effort for supporting SYCL
language front end.
It makes the following changes:
1. Adds sycl_external attribute for functions with external linkage,
which is intended for use to implement the SYCL_EXTERNAL macro as
specified by the SYCL 2020 specification
2. Adds checks to avoid emitting device code when sycl_external and
sycl_kernel_entry_point attributes are not enabled
3. Fixes test failures caused by the above changes

This patch is missing diagnostics for the following diagnostics listed
in the SYCL 2020 specification's section 5.10.1, which will be addressed
in a subsequent PR:
Functions that are declared using SYCL_EXTERNAL have the following
additional restrictions beyond those imposed on other device functions:
1. If the SYCL backend does not support the generic address space then
the function cannot use raw pointers as parameter or return types.
Explicit pointer classes must be used instead;
2. The function cannot call group::parallel_for_work_item;
3. The function cannot be called from a parallel_for_work_group scope.

In addition to that, the subsequent PR will also implement diagnostics
for inline functions including virtual functions defined as inline.

---------

Co-authored-by: Mariya Podchishchaeva <[email protected]>

Author: schittir

clang/include/clang/Basic/Attr.td

@@ -1632,6 +1632,13 @@ def DeviceKernel : DeclOrTypeAttr {
 }];
 }
 
+def SYCLExternal : InheritableAttr {
+  let Spellings = [CXX11<"clang", "sycl_external">];
+  let Subjects = SubjectList<[Function], ErrorDiag>;
+  let LangOpts = [SYCLHost, SYCLDevice];
+  let Documentation = [SYCLExternalDocs];
+}
+
 def SYCLKernelEntryPoint : InheritableAttr {
   let Spellings = [CXX11<"clang", "sycl_kernel_entry_point">];
   let Args = [

clang/include/clang/Basic/AttrDocs.td

@@ -476,6 +476,47 @@ The SYCL kernel in the previous code sample meets these expectations.
   }];
 }
 
+def SYCLExternalDocs : Documentation {
+  let Category = DocCatFunction;
+  let Heading = "sycl_external";
+  let Content = [{
+The ``sycl_external`` attribute indicates that a function defined in another
+translation unit may be called by a device function defined in the current
+translation unit or, if defined in the current translation unit, the function
+may be called by device functions defined in other translation units.
+The attribute is intended for use in the implementation of the ``SYCL_EXTERNAL``
+macro as specified in section 5.10.1, "SYCL functions and member functions
+linkage", of the SYCL 2020 specification.
+
+The attribute only appertains to functions and only those that meet the
+following requirements:
+
+* Has external linkage
+* Is not explicitly defined as deleted (the function may be an explicitly
+  defaulted function that is defined as deleted)
+
+The attribute shall be present on the first declaration of a function and
+may optionally be present on subsequent declarations.
+
+When compiling for a SYCL device target that does not support the generic
+address space, the function shall not specify a raw pointer or reference type
+as the return type or as a parameter type.
+See section 5.10, "SYCL offline linking", of the SYCL 2020 specification.
+The following examples demonstrate the use of this attribute:
+
+.. code-block:: c++
+
+  [[clang::sycl_external]] void Foo(); // Ok.
+
+  [[clang::sycl_external]] void Bar() { /* ... */ } // Ok.
+
+  [[clang::sycl_external]] extern void Baz(); // Ok.
+
+  [[clang::sycl_external]] static void Quux() { /* ... */ } // error:  Quux() has internal linkage.
+
+  }];
+}
+
 def SYCLKernelEntryPointDocs : Documentation {
   let Category = DocCatFunction;
   let Content = [{

clang/include/clang/Basic/DiagnosticGroups.td

@@ -652,6 +652,7 @@ def NonNull : DiagGroup<"nonnull">;
 def NonPODVarargs : DiagGroup<"non-pod-varargs">;
 def ClassVarargs : DiagGroup<"class-varargs", [NonPODVarargs]>;
 def : DiagGroup<"nonportable-cfstrings">;
+def NonPortableSYCL : DiagGroup<"nonportable-sycl">;
 def NonVirtualDtor : DiagGroup<"non-virtual-dtor">;
 def GNUNullPointerArithmetic : DiagGroup<"gnu-null-pointer-arithmetic">;
 def NullPointerArithmetic

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -12958,6 +12958,17 @@ def err_sycl_special_type_num_init_method : Error<
   "types with 'sycl_special_class' attribute must have one and only one '__init' "
   "method defined">;
 
+// SYCL external attribute diagnostics
+def err_sycl_external_invalid_linkage : Error<
+  "%0 can only be applied to functions with external linkage">;
+def err_sycl_external_invalid_main : Error<
+  "%0 cannot be applied to the 'main' function">;
+def err_sycl_external_invalid_deleted_function : Error<
+  "%0 cannot be applied to an explicitly deleted function">;
+def warn_sycl_external_missing_on_first_decl : Warning<
+  "%0 attribute does not appear on the first declaration">,
+  InGroup<NonPortableSYCL>;
+
 // SYCL kernel entry point diagnostics
 def err_sycl_entry_point_invalid : Error<
   "the %0 attribute cannot be applied to a"
@@ -12972,7 +12983,7 @@ def err_sycl_kernel_name_conflict : Error<
   "the %0 kernel name argument conflicts with a previous declaration">;
 def warn_sycl_kernel_name_not_a_class_type : Warning<
   "%0 is not a valid SYCL kernel name type; a non-union class type is required">,
-  InGroup<DiagGroup<"nonportable-sycl">>, DefaultError;
+  InGroup<NonPortableSYCL>, DefaultError;
 def warn_sycl_entry_point_redundant_declaration : Warning<
   "redundant %0 attribute">, InGroup<RedundantAttribute>;
 def err_sycl_entry_point_after_definition : Error<

clang/include/clang/Sema/SemaSYCL.h

@@ -64,6 +64,7 @@ class SemaSYCL : public SemaBase {
   void handleKernelAttr(Decl *D, const ParsedAttr &AL);
   void handleKernelEntryPointAttr(Decl *D, const ParsedAttr &AL);
 
+  void CheckSYCLExternalFunctionDecl(FunctionDecl *FD);
   void CheckSYCLEntryPointFunctionDecl(FunctionDecl *FD);
   StmtResult BuildSYCLKernelCallStmt(FunctionDecl *FD, CompoundStmt *Body);
 };

clang/lib/AST/ASTContext.cpp

@@ -13127,6 +13127,14 @@ bool ASTContext::DeclMustBeEmitted(const Decl *D) {
   if (D->hasAttr<WeakRefAttr>())
     return false;
 
+  // SYCL device compilation requires that functions defined with the
+  // sycl_kernel_entry_point or sycl_external attributes be emitted. All
+  // other entities are emitted only if they are used by a function
+  // defined with one of those attributes.
+  if (LangOpts.SYCLIsDevice)
+    return isa<FunctionDecl>(D) && (D->hasAttr<SYCLKernelEntryPointAttr>() ||
+                                    D->hasAttr<SYCLExternalAttr>());
+
   // Aliases and used decls are required.
   if (D->hasAttr<AliasAttr>() || D->hasAttr<UsedAttr>())
     return true;
@@ -13136,15 +13144,6 @@ bool ASTContext::DeclMustBeEmitted(const Decl *D) {
     if (!FD->doesThisDeclarationHaveABody())
       return FD->doesDeclarationForceExternallyVisibleDefinition();
 
-    // Function definitions with the sycl_kernel_entry_point attribute are
-    // required during device compilation so that SYCL kernel caller offload
-    // entry points are emitted.
-    if (LangOpts.SYCLIsDevice && FD->hasAttr<SYCLKernelEntryPointAttr>())
-      return true;
-
-    // FIXME: Functions declared with SYCL_EXTERNAL are required during
-    // device compilation.
-
     // Constructors and destructors are required.
     if (FD->hasAttr<ConstructorAttr>() || FD->hasAttr<DestructorAttr>())
       return true;

clang/lib/Sema/SemaDecl.cpp

@@ -3115,6 +3115,10 @@ static void checkNewAttributesAfterDef(Sema &S, Decl *New, const Decl *Old) {
       cast<SYCLKernelEntryPointAttr>(NewAttribute)->setInvalidAttr();
       ++I;
       continue;
+    } else if (isa<SYCLExternalAttr>(NewAttribute)) {
+      // SYCLExternalAttr may be added after a definition.
+      ++I;
+      continue;
     }
 
     S.Diag(NewAttribute->getLocation(),
@@ -4140,6 +4144,18 @@ bool Sema::MergeFunctionDecl(FunctionDecl *New, NamedDecl *&OldD, Scope *S,
            diag::note_carries_dependency_missing_first_decl) << 0/*Function*/;
     }
 
+    // SYCL 2020 section 5.10.1, "SYCL functions and member functions linkage":
+    //   When a function is declared with SYCL_EXTERNAL, that macro must be
+    //   used on the first declaration of that function in the translation unit.
+    //   Redeclarations of the function in the same translation unit may
+    //   optionally use SYCL_EXTERNAL, but this is not required.
+    const SYCLExternalAttr *SEA = New->getAttr<SYCLExternalAttr>();
+    if (SEA && !Old->hasAttr<SYCLExternalAttr>()) {
+      Diag(SEA->getLocation(), diag::warn_sycl_external_missing_on_first_decl)
+          << SEA;
+      Diag(Old->getLocation(), diag::note_previous_declaration);
+    }
+
     // (C++98 8.3.5p3):
     //   All declarations for a function shall agree exactly in both the
     //   return type and the parameter-type-list.
@@ -12325,6 +12341,9 @@ bool Sema::CheckFunctionDeclaration(Scope *S, FunctionDecl *NewFD,
   if (NewFD->hasAttr<SYCLKernelEntryPointAttr>())
     SYCL().CheckSYCLEntryPointFunctionDecl(NewFD);
 
+  if (NewFD->hasAttr<SYCLExternalAttr>())
+    SYCL().CheckSYCLExternalFunctionDecl(NewFD);
+
   // Semantic checking for this function declaration (in isolation).
 
   if (getLangOpts().CPlusPlus) {
@@ -12513,6 +12532,13 @@ void Sema::CheckMain(FunctionDecl *FD, const DeclSpec &DS) {
     return;
   }
 
+  if (FD->hasAttr<SYCLExternalAttr>()) {
+    Diag(FD->getLocation(), diag::err_sycl_external_invalid_main)
+        << FD->getAttr<SYCLExternalAttr>();
+    FD->setInvalidDecl();
+    return;
+  }
+
   // Functions named main in hlsl are default entries, but don't have specific
   // signatures they are required to conform to.
   if (getLangOpts().HLSL)
@@ -16351,6 +16377,14 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body, bool IsInstantiation,
     }
   }
 
+  if (FD && !FD->isInvalidDecl() && FD->hasAttr<SYCLExternalAttr>()) {
+    SYCLExternalAttr *SEAttr = FD->getAttr<SYCLExternalAttr>();
+    if (FD->isDeletedAsWritten())
+      Diag(SEAttr->getLocation(),
+           diag::err_sycl_external_invalid_deleted_function)
+          << SEAttr;
+  }
+
   {
     // Do not call PopExpressionEvaluationContext() if it is a lambda because
     // one is already popped when finishing the lambda in BuildLambdaExpr().

clang/lib/Sema/SemaDeclAttr.cpp

@@ -7061,6 +7061,9 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
   case ParsedAttr::AT_EnumExtensibility:
     handleEnumExtensibilityAttr(S, D, AL);
     break;
+  case ParsedAttr::AT_SYCLExternal:
+    handleSimpleAttribute<SYCLExternalAttr>(S, D, AL);
+    break;
   case ParsedAttr::AT_SYCLKernelEntryPoint:
     S.SYCL().handleKernelEntryPointAttr(D, AL);
     break;

clang/lib/Sema/SemaSYCL.cpp

@@ -250,6 +250,23 @@ static bool CheckSYCLKernelName(Sema &S, SourceLocation Loc,
   return false;
 }
 
+void SemaSYCL::CheckSYCLExternalFunctionDecl(FunctionDecl *FD) {
+  const auto *SEAttr = FD->getAttr<SYCLExternalAttr>();
+  assert(SEAttr && "Missing sycl_external attribute");
+  if (!FD->isInvalidDecl() && !FD->isTemplated()) {
+    if (!FD->isExternallyVisible())
+      if (!FD->isFunctionTemplateSpecialization() ||
+          FD->getTemplateSpecializationInfo()->isExplicitSpecialization())
+        Diag(SEAttr->getLocation(), diag::err_sycl_external_invalid_linkage)
+            << SEAttr;
+  }
+  if (FD->isDeletedAsWritten()) {
+    Diag(SEAttr->getLocation(),
+         diag::err_sycl_external_invalid_deleted_function)
+        << SEAttr;
+  }
+}
+
 void SemaSYCL::CheckSYCLEntryPointFunctionDecl(FunctionDecl *FD) {
   // Ensure that all attributes present on the declaration are consistent
   // and warn about any redundant ones.

clang/test/CodeGenSPIRV/Builtins/generic_cast_to_ptr_explicit.c

@@ -1,33 +1,36 @@
-// RUN: %clang_cc1 -O1 -triple spirv64 -fsycl-is-device %s -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 -O1 -triple spirv64 -fsycl-is-device -x c++ %s -emit-llvm -o - | FileCheck %s
 // RUN: %clang_cc1 -O1 -triple spirv64 -cl-std=CL3.0 -x cl %s -emit-llvm -o - | FileCheck %s
 // RUN: %clang_cc1 -O1 -triple spirv32 -cl-std=CL3.0 -x cl %s -emit-llvm -o - | FileCheck %s
 
-// CHECK: spir_func noundef ptr @test_cast_to_private(
-// CHECK-SAME: ptr addrspace(4) noundef readnone [[P:%.*]]
+#ifdef __SYCL_DEVICE_ONLY__
+#define SYCL_EXTERNAL [[clang::sycl_external]]
+#else
+#define SYCL_EXTERNAL
+#endif
+
+// CHECK: spir_func noundef ptr @{{.*}}test_cast_to_private{{.*}}(ptr addrspace(4) noundef readnone [[P:%.*]]
 // CHECK-NEXT:  [[ENTRY:.*:]]
 // CHECK-NEXT:    [[SPV_CAST:%.*]] = tail call noundef ptr @llvm.spv.generic.cast.to.ptr.explicit.p0(ptr addrspace(4) %p)
 // CHECK-NEXT:    ret ptr [[SPV_CAST]]
 //
-__attribute__((opencl_private)) int* test_cast_to_private(int* p) {
+SYCL_EXTERNAL __attribute__((opencl_private)) int* test_cast_to_private(int* p) {
     return __builtin_spirv_generic_cast_to_ptr_explicit(p, 7);
 }
 
-// CHECK: spir_func noundef ptr addrspace(1) @test_cast_to_global(
-// CHECK-SAME: ptr addrspace(4) noundef readnone [[P:%.*]]
+// CHECK: spir_func noundef ptr addrspace(1) @{{.*}}test_cast_to_global{{.*}}(ptr addrspace(4) noundef readnone [[P:%.*]]
 // CHECK-NEXT:  [[ENTRY:.*:]]
 // CHECK-NEXT:    [[SPV_CAST:%.*]] = tail call noundef ptr addrspace(1) @llvm.spv.generic.cast.to.ptr.explicit.p1(ptr addrspace(4) %p)
 // CHECK-NEXT:    ret ptr addrspace(1) [[SPV_CAST]]
 //
-__attribute__((opencl_global)) int* test_cast_to_global(int* p) {
+SYCL_EXTERNAL __attribute__((opencl_global)) int* test_cast_to_global(int* p) {
     return __builtin_spirv_generic_cast_to_ptr_explicit(p, 5);
 }
 
-// CHECK: spir_func noundef ptr addrspace(3) @test_cast_to_local(
-// CHECK-SAME: ptr addrspace(4) noundef readnone [[P:%.*]]
+// CHECK: spir_func noundef ptr addrspace(3) @{{.*}}test_cast_to_local{{.*}}(ptr addrspace(4) noundef readnone [[P:%.*]]
 // CHECK-NEXT:  [[ENTRY:.*:]]
 // CHECK-NEXT:    [[SPV_CAST:%.*]] = tail call noundef ptr addrspace(3) @llvm.spv.generic.cast.to.ptr.explicit.p3(ptr addrspace(4) %p)
 // CHECK-NEXT:    ret ptr addrspace(3) [[SPV_CAST]]
 //
-__attribute__((opencl_local)) int* test_cast_to_local(int* p) {
+SYCL_EXTERNAL __attribute__((opencl_local)) int* test_cast_to_local(int* p) {
     return __builtin_spirv_generic_cast_to_ptr_explicit(p, 4);
 }