[SYCL] Add special type parameter support for free function kernels (#17789)

Add support for special types to be used as top-level arguments in free
function kernels.

---------

Co-authored-by: Mariya Podchishchaeva <mariya.podchishchaeva@intel.com>

Author: lbushi25

clang/include/clang/AST/PrettyPrinter.h

@@ -84,7 +84,8 @@ struct PrintingPolicy {
         PrintInjectedClassNameWithArguments(true), UsePreferredNames(true),
         AlwaysIncludeTypeForTemplateArgument(false),
         CleanUglifiedParameters(false), EntireContentsOfLargeArray(true),
-        UseEnumerators(true), UseHLSLTypes(LO.HLSL) {}
+        UseEnumerators(true), UseFullyQualifiedEnumerators(false),
+        UseHLSLTypes(LO.HLSL) {}
 
   /// Adjust this printing policy for cases where it's known that we're
   /// printing C++ code (for instance, if AST dumping reaches a C++-only
@@ -419,6 +420,12 @@ struct PrintingPolicy {
   LLVM_PREFERRED_TYPE(bool)
   unsigned UseEnumerators : 1;
 
+  /// Whether to print the names of enumerator non-type template parameters
+  /// with all the namespace scope qualifiers regardless of the way the user
+  /// wrote them in the source code. No effect if UseEnumerators is not set.
+  LLVM_PREFERRED_TYPE(bool)
+  unsigned UseFullyQualifiedEnumerators : 1;
+
   /// Whether or not we're printing known HLSL code and should print HLSL
   /// sugared types when possible.
   LLVM_PREFERRED_TYPE(bool)

clang/lib/AST/TemplateBase.cpp

@@ -64,7 +64,10 @@ static void printIntegral(const TemplateArgument &TemplArg, raw_ostream &Out,
         // may create a size difference between the enum value and template
         // argument value, requiring isSameValue here instead of operator==.
         if (llvm::APSInt::isSameValue(ECD->getInitVal(), Val)) {
-          ECD->printQualifiedName(Out, Policy);
+          PrintingPolicy NewPolicy = Policy;
+          if (Policy.UseFullyQualifiedEnumerators)
+            NewPolicy.PrintCanonicalTypes = true;
+          ECD->printQualifiedName(Out, NewPolicy);
           return;
         }
       }

clang/lib/Sema/SemaSYCL.cpp

@@ -1080,8 +1080,7 @@ static ParamDesc makeParamDesc(ASTContext &Ctx, StringRef Name, QualType Ty) {
 }
 
 static void unsupportedFreeFunctionParamType() {
-  llvm::report_fatal_error("Only scalars and pointers are permitted as "
-                           "free function parameters");
+  llvm::report_fatal_error("Unsupported free kernel parameter type!");
 }
 
 class MarkWIScopeFnVisitor : public RecursiveASTVisitor<MarkWIScopeFnVisitor> {
@@ -2080,13 +2079,7 @@ class SyclKernelFieldChecker : public SyclKernelFieldHandler {
   }
 
   bool handleSyclSpecialType(ParmVarDecl *PD, QualType ParamTy) final {
-    if (!SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::work_group_memory) &&
-        !SemaSYCL::isSyclType(ParamTy,
-                              SYCLTypeAttr::dynamic_work_group_memory)) {
-      Diag.Report(PD->getLocation(), diag::err_bad_kernel_param_type)
-          << ParamTy;
-      IsInvalid = true;
-    }
+    IsInvalid |= checkSyclSpecialType(ParamTy, PD->getLocation());
     return isValid();
   }
 
@@ -2238,10 +2231,7 @@ class SyclKernelUnionChecker : public SyclKernelFieldHandler {
   }
 
   bool handleSyclSpecialType(ParmVarDecl *PD, QualType ParamTy) final {
-    if (!SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::work_group_memory) &&
-        !SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::dynamic_work_group_memory))
-      unsupportedFreeFunctionParamType(); // TODO
-    return true;
+    return checkType(PD->getLocation(), ParamTy);
   }
 
   bool handleSyclSpecialType(const CXXRecordDecl *, const CXXBaseSpecifier &BS,
@@ -2830,23 +2820,34 @@ class SyclKernelDeclCreator : public SyclKernelFieldHandler {
   // kernel parameters from __init method parameters. We will use __init method
   // and kernel parameters which we build here to initialize special objects in
   // the kernel body.
-  bool handleSpecialType(FieldDecl *FD, QualType FieldTy) {
-    const auto *RecordDecl = FieldTy->getAsCXXRecordDecl();
-    assert(RecordDecl && "The type must be a RecordDecl");
+  // ParentDecl parameterizes whether we are in a free function kernel or a
+  // lambda kernel by taking the value ParmVarDecl or FieldDecl respectively.
+  template <typename ParentDecl>
+  bool handleSpecialType(ParentDecl *decl, QualType Ty) {
+    const auto *RD = Ty->getAsCXXRecordDecl();
+    assert(RD && "The type must be a RecordDecl");
     llvm::StringLiteral MethodName =
-        KernelDecl->hasAttr<SYCLSimdAttr>() && isSyclAccessorType(FieldTy)
+        KernelDecl->hasAttr<SYCLSimdAttr>() && isSyclAccessorType(Ty)
             ? InitESIMDMethodName
             : InitMethodName;
-    CXXMethodDecl *InitMethod = getMethodByName(RecordDecl, MethodName);
+    CXXMethodDecl *InitMethod = getMethodByName(RD, MethodName);
     assert(InitMethod && "The type must have the __init method");
 
     // Don't do -1 here because we count on this to be the first parameter added
     // (if any).
     size_t ParamIndex = Params.size();
     for (const ParmVarDecl *Param : InitMethod->parameters()) {
       QualType ParamTy = Param->getType();
-      addParam(FD, ParamTy.getCanonicalType());
-
+      // For lambda kernels the arguments to the OpenCL kernel are named
+      // based on the position they have as fields in the definition of the
+      // special type structure i.e __arg_field1, __arg_field2 and so on.
+      // For free function kernels the arguments are named in direct mapping
+      // with the names they have in the __init method i.e __arg_Ptr for work
+      // group memory since its init function takes a parameter with Ptr name.
+      if constexpr (std::is_same_v<ParentDecl, FieldDecl>)
+        addParam(decl, ParamTy.getCanonicalType());
+      else
+        addParam(Param, ParamTy.getCanonicalType());
       // Propagate add_ir_attributes_kernel_parameter attribute.
       if (const auto *AddIRAttr =
               Param->getAttr<SYCLAddIRAttributesKernelParameterAttr>())
@@ -2858,8 +2859,8 @@ class SyclKernelDeclCreator : public SyclKernelFieldHandler {
       // handleAccessorPropertyList. If new classes with property list are
       // added, this code needs to be refactored to call
       // handleAccessorPropertyList for each class which requires it.
-      if (ParamTy.getTypePtr()->isPointerType() && isSyclAccessorType(FieldTy))
-        handleAccessorType(FieldTy, RecordDecl, FD->getBeginLoc());
+      if (ParamTy.getTypePtr()->isPointerType() && isSyclAccessorType(Ty))
+        handleAccessorType(Ty, RD, decl->getBeginLoc());
     }
     LastParamIndex = ParamIndex;
     return true;
@@ -3026,28 +3027,7 @@ class SyclKernelDeclCreator : public SyclKernelFieldHandler {
   }
 
   bool handleSyclSpecialType(ParmVarDecl *PD, QualType ParamTy) final {
-    if (SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::work_group_memory) ||
-        SemaSYCL::isSyclType(ParamTy,
-                             SYCLTypeAttr::dynamic_work_group_memory)) {
-      const auto *RecordDecl = ParamTy->getAsCXXRecordDecl();
-      assert(RecordDecl && "The type must be a RecordDecl");
-      CXXMethodDecl *InitMethod = getMethodByName(RecordDecl, InitMethodName);
-      assert(InitMethod && "The type must have the __init method");
-      // Don't do -1 here because we count on this to be the first parameter
-      // added (if any).
-      size_t ParamIndex = Params.size();
-      for (const ParmVarDecl *Param : InitMethod->parameters()) {
-        QualType ParamTy = Param->getType();
-        addParam(Param, ParamTy.getCanonicalType());
-        // Propagate add_ir_attributes_kernel_parameter attribute.
-        if (const auto *AddIRAttr =
-                Param->getAttr<SYCLAddIRAttributesKernelParameterAttr>())
-          Params.back()->addAttr(AddIRAttr->clone(SemaSYCLRef.getASTContext()));
-      }
-      LastParamIndex = ParamIndex;
-    } else // TODO
-      unsupportedFreeFunctionParamType();
-    return true;
+    return handleSpecialType(PD, ParamTy);
   }
 
   RecordDecl *wrapField(FieldDecl *Field, QualType FieldTy) {
@@ -4540,47 +4520,48 @@ class FreeFunctionKernelBodyCreator : public SyclKernelFieldHandler {
   // TODO: Revisit this approach once https://github.com/intel/llvm/issues/16061
   // is closed.
   bool handleSyclSpecialType(ParmVarDecl *PD, QualType ParamTy) final {
-    if (SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::work_group_memory) ||
-        SemaSYCL::isSyclType(ParamTy,
-                             SYCLTypeAttr::dynamic_work_group_memory)) {
-      const auto *RecordDecl = ParamTy->getAsCXXRecordDecl();
-      AccessSpecifier DefaultConstructorAccess;
-      auto DefaultConstructor =
-          std::find_if(RecordDecl->ctor_begin(), RecordDecl->ctor_end(),
-                       [](auto it) { return it->isDefaultConstructor(); });
-      DefaultConstructorAccess = DefaultConstructor->getAccess();
-      DefaultConstructor->setAccess(AS_public);
-
-      QualType Ty = PD->getOriginalType();
-      ASTContext &Ctx = SemaSYCLRef.SemaRef.getASTContext();
-      VarDecl *WorkGroupMemoryClone = VarDecl::Create(
-          Ctx, DeclCreator.getKernelDecl(), FreeFunctionSrcLoc,
-          FreeFunctionSrcLoc, PD->getIdentifier(), PD->getType(),
-          Ctx.getTrivialTypeSourceInfo(Ty), SC_None);
-      InitializedEntity VarEntity =
-          InitializedEntity::InitializeVariable(WorkGroupMemoryClone);
-      InitializationKind InitKind =
-          InitializationKind::CreateDefault(FreeFunctionSrcLoc);
-      InitializationSequence InitSeq(SemaSYCLRef.SemaRef, VarEntity, InitKind,
-                                     std::nullopt);
-      ExprResult Init = InitSeq.Perform(SemaSYCLRef.SemaRef, VarEntity,
-                                        InitKind, std::nullopt);
-      WorkGroupMemoryClone->setInit(
-          SemaSYCLRef.SemaRef.MaybeCreateExprWithCleanups(Init.get()));
-      WorkGroupMemoryClone->setInitStyle(VarDecl::CallInit);
-      DefaultConstructor->setAccess(DefaultConstructorAccess);
-
-      Stmt *DS = new (SemaSYCLRef.getASTContext())
-          DeclStmt(DeclGroupRef(WorkGroupMemoryClone), FreeFunctionSrcLoc,
-                   FreeFunctionSrcLoc);
-      BodyStmts.push_back(DS);
-      Expr *MemberBaseExpr = SemaSYCLRef.SemaRef.BuildDeclRefExpr(
-          WorkGroupMemoryClone, Ty, VK_PRValue, FreeFunctionSrcLoc);
-      createSpecialMethodCall(RecordDecl, InitMethodName, MemberBaseExpr,
-                              BodyStmts);
-      ArgExprs.push_back(MemberBaseExpr);
-    } else // TODO
-      unsupportedFreeFunctionParamType();
+    // The code produced looks like this in the case of a work group memory
+    // parameter:
+    // void auto_generated_kernel(__local int * arg) {
+    //    work_group_memory wgm;
+    //    wgm.__init(arg);
+    //    user_kernel(some arguments..., wgm, some arguments...);
+    // }
+    const auto *RecordDecl = ParamTy->getAsCXXRecordDecl();
+    AccessSpecifier DefaultConstructorAccess;
+    auto DefaultConstructor =
+        std::find_if(RecordDecl->ctor_begin(), RecordDecl->ctor_end(),
+                     [](auto it) { return it->isDefaultConstructor(); });
+    DefaultConstructorAccess = DefaultConstructor->getAccess();
+    DefaultConstructor->setAccess(AS_public);
+
+    ASTContext &Ctx = SemaSYCLRef.SemaRef.getASTContext();
+    VarDecl *SpecialObjectClone =
+        VarDecl::Create(Ctx, DeclCreator.getKernelDecl(), FreeFunctionSrcLoc,
+                        FreeFunctionSrcLoc, PD->getIdentifier(), ParamTy,
+                        Ctx.getTrivialTypeSourceInfo(ParamTy), SC_None);
+    InitializedEntity VarEntity =
+        InitializedEntity::InitializeVariable(SpecialObjectClone);
+    InitializationKind InitKind =
+        InitializationKind::CreateDefault(FreeFunctionSrcLoc);
+    InitializationSequence InitSeq(SemaSYCLRef.SemaRef, VarEntity, InitKind,
+                                   std::nullopt);
+    ExprResult Init =
+        InitSeq.Perform(SemaSYCLRef.SemaRef, VarEntity, InitKind, std::nullopt);
+    SpecialObjectClone->setInit(
+        SemaSYCLRef.SemaRef.MaybeCreateExprWithCleanups(Init.get()));
+    SpecialObjectClone->setInitStyle(VarDecl::CallInit);
+    DefaultConstructor->setAccess(DefaultConstructorAccess);
+
+    Stmt *DS = new (SemaSYCLRef.getASTContext())
+        DeclStmt(DeclGroupRef(SpecialObjectClone), FreeFunctionSrcLoc,
+                 FreeFunctionSrcLoc);
+    BodyStmts.push_back(DS);
+    Expr *MemberBaseExpr = SemaSYCLRef.SemaRef.BuildDeclRefExpr(
+        SpecialObjectClone, ParamTy, VK_PRValue, FreeFunctionSrcLoc);
+    createSpecialMethodCall(RecordDecl, InitMethodName, MemberBaseExpr,
+                            BodyStmts);
+    ArgExprs.push_back(MemberBaseExpr);
     return true;
   }
 
@@ -4874,14 +4855,45 @@ class SyclKernelIntHeaderCreator : public SyclKernelFieldHandler {
   }
 
   bool handleSyclSpecialType(ParmVarDecl *PD, QualType ParamTy) final {
-    if (SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::work_group_memory))
+    const auto *ClassTy = ParamTy->getAsCXXRecordDecl();
+    assert(ClassTy && "Type must be a C++ record type");
+    if (isSyclAccessorType(ParamTy)) {
+      const auto *AccTy =
+          cast<ClassTemplateSpecializationDecl>(ParamTy->getAsRecordDecl());
+      assert(AccTy->getTemplateArgs().size() >= 2 &&
+             "Incorrect template args for Accessor Type");
+      int Dims = static_cast<int>(
+          AccTy->getTemplateArgs()[1].getAsIntegral().getExtValue());
+      int Info = getAccessTarget(ParamTy, AccTy) | (Dims << 11);
+      Header.addParamDesc(SYCLIntegrationHeader::kind_accessor, Info,
+                          CurOffset);
+    } else if (SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::stream)) {
+      addParam(PD, ParamTy, SYCLIntegrationHeader::kind_stream);
+    } else if (SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::work_group_memory)) {
       addParam(PD, ParamTy, SYCLIntegrationHeader::kind_work_group_memory);
-    else if (SemaSYCL::isSyclType(ParamTy,
-                                  SYCLTypeAttr::dynamic_work_group_memory))
+    } else if (SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::sampler) ||
+               SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::annotated_ptr) ||
+               SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::annotated_arg)) {
+      CXXMethodDecl *InitMethod = getMethodByName(ClassTy, InitMethodName);
+      assert(InitMethod && "type must have __init method");
+      const ParmVarDecl *InitArg = InitMethod->getParamDecl(0);
+      assert(InitArg && "Init method must have arguments");
+      QualType T = InitArg->getType();
+      SYCLIntegrationHeader::kernel_param_kind_t ParamKind =
+          SemaSYCL::isSyclType(ParamTy, SYCLTypeAttr::sampler)
+              ? SYCLIntegrationHeader::kind_sampler
+              : (T->isPointerType() ? SYCLIntegrationHeader::kind_pointer
+                                    : SYCLIntegrationHeader::kind_std_layout);
+      addParam(PD, ParamTy, ParamKind);
+    } else if (SemaSYCL::isSyclType(ParamTy,
+                                    SYCLTypeAttr::dynamic_work_group_memory))
       addParam(PD, ParamTy,
                SYCLIntegrationHeader::kind_dynamic_work_group_memory);
-    else
-      unsupportedFreeFunctionParamType(); // TODO
+
+    else {
+      llvm_unreachable(
+          "Unexpected SYCL special class when generating integration header");
+    }
     return true;
   }
 
@@ -6666,6 +6678,7 @@ void SYCLIntegrationHeader::emit(raw_ostream &O) {
   O << "#include <sycl/detail/defines_elementary.hpp>\n";
   O << "#include <sycl/detail/kernel_desc.hpp>\n";
   O << "#include <sycl/ext/oneapi/experimental/free_function_traits.hpp>\n";
+  O << "#include <sycl/access/access.hpp>\n";
   O << "\n";
 
   LangOptions LO;
@@ -6977,6 +6990,7 @@ void SYCLIntegrationHeader::emit(raw_ostream &O) {
     Policy.PolishForDeclaration = true;
     Policy.FullyQualifiedName = true;
     Policy.EnforceScopeForElaboratedTypes = true;
+    Policy.UseFullyQualifiedEnumerators = true;
 
     // Now we need to print the declaration of the kernel itself.
     // Example:

clang/test/CodeGenSYCL/Inputs/sycl.hpp

@@ -39,6 +39,9 @@ class __attribute__((sycl_special_class)) __SYCL_TYPE(sampler) sampler {
 
 public:
   void use(void) const {}
+#ifdef __SYCL_DEVICE_ONLY__
+  sampler() = default;
+#endif
 };
 
 template <int dimensions = 1>
@@ -327,6 +330,9 @@ template <typename dataT, int dimensions, access::mode accessmode,
 class __attribute__((sycl_special_class)) __SYCL_TYPE(accessor) accessor {
 
 public:
+#ifdef __SYCL_DEVICE_ONLY__               
+  accessor() = default;
+#endif
   void use(void) const {}
   template <typename... T>
   void use(T... args) {}
@@ -420,6 +426,9 @@ local_accessor: public accessor<dataT,
         dimensions, access::mode::read_write,
         access::target::local> {
 public:
+#ifdef __SYCL_DEVICE_ONLY__  
+  local_accessor() = default;
+#endif
   void use(void) const {}
   template <typename... T>
   void use(T... args) {}
@@ -487,17 +496,21 @@ int printf(const __SYCL_CONSTANT_AS char *__format, Args... args) {
 template <typename T, typename... Props>
 class __attribute__((sycl_special_class)) __SYCL_TYPE(annotated_arg) annotated_arg {
   T obj;
-  #ifdef __SYCL_DEVICE_ONLY__
+#ifdef __SYCL_DEVICE_ONLY__
     void __init(T _obj) {}
-  #endif
+public:
+    annotated_arg() = default;
+#endif
 };
 
 template <typename T, typename... Props>
 class __attribute__((sycl_special_class)) __SYCL_TYPE(annotated_ptr) annotated_ptr {
   T* obj;
-  #ifdef __SYCL_DEVICE_ONLY__
+#ifdef __SYCL_DEVICE_ONLY__
     void __init(T* _obj) {}
-  #endif
+public:
+    annotated_ptr() = default;
+#endif
 };
 
 } // namespace experimental

clang/test/CodeGenSYCL/debug-info-srcpos-kernel.cpp

@@ -29,11 +29,12 @@ int main() {
 // CHECK: call spir_func void {{.*}}6__init{{.*}} !dbg [[LINE_A0]]
 // CHECK: call spir_func void @_ZZ4mainENKUlvE_clEv{{.*}} !dbg [[LINE_B0:![0-9]+]]
 // CHECK: ret void, !dbg [[LINE_C0:![0-9]+]]
+// CHECK: !DIFile(filename: "{{.*}}debug-info-srcpos-kernel.cpp"{{.*}})
+// CHECK: [[FILE:![0-9]+]] = !DIFile(filename: "{{.*}}debug-info-srcpos-kernel.cpp"{{.*}})
 // CHECK: [[KERNEL]] = {{.*}}!DISubprogram(name: "{{.*}}19use_kernel_for_test"
-// CHECK-SAME: scope: [[FILE:![0-9]+]],
+// CHECK-SAME: scope: [[FILE]],
 // CHECK-SAME: file: [[FILE]],
 // CHECK-SAME: flags: DIFlagArtificial | DIFlagPrototyped
-// CHECK: [[FILE]] = !DIFile(filename: "{{.*}}debug-info-srcpos-kernel.cpp"{{.*}})
 // CHECK: [[LINE_A0]] = !DILocation(line: 15,{{.*}}scope: [[KERNEL]]
 // CHECK: [[LINE_B0]] = !DILocation(line: 16,{{.*}}scope: [[BLOCK:![0-9]+]]
 // CHECK: [[BLOCK]] = distinct !DILexicalBlock(scope: [[KERNEL]]