[OpenCL] Add intel_reqd_sub_group_size attribute support

[pxli168]

Summary

Add intel_reqd_sub_group_size attribute support as intel extension cl_intel_required_subgroup_size from
https://www.khronos.org/registry/OpenCL/extensions/intel/cl_intel_required_subgroup_size.txt

[bader]

LGTM.
Just a few minor issues.

lib/CodeGen/CodeGenFunction.cpp

699–702

It would be nice if you update the comment for EmitOpenCLKernelMetadata function declaration in CodeGenFunction.
I think we should make it more generic and avoid implementation details like listing exact attribute names.

lib/Sema/SemaDeclAttr.cpp

2877–2878

Please, fix indentation.

2889

Please, write your code to fit within 80 columns.

[pxli168]

Fix some minor issues from bader.

[bader]

lib/CodeGen/CodeGenFunction.cpp

650

Xuili, sorry for confusion.
I was talking about the declaration in CodeGenFunction.h file. https://github.com/llvm-mirror/clang/blob/master/lib/CodeGen/CodeGenFunction.h#L1397

 
/// Add a kernel metadata node to the named metadata node 'opencl.kernels'.
/// In the kernel metadata node, reference the kernel function and metadata 
/// nodes for its optional attribute qualifiers (OpenCL 1.1 6.7.2):
/// - A node for the vec_type_hint(<type>) qualifier contains string
///   "vec_type_hint", an undefined value of the <type> data type,
///   and a Boolean that is true if the <type> is integer and signed.
/// - A node for the work_group_size_hint(X,Y,Z) qualifier contains string 
///   "work_group_size_hint", and three 32-bit integers X, Y and Z.
/// - A node for the reqd_work_group_size(X,Y,Z) qualifier contains string 
///   "reqd_work_group_size", and three 32-bit integers X, Y and Z.

As you see this comment explicitly lists all the metadata added to 'opencl.kernels', which seems to be unnecessary.
Instead of adding new comment about 'intel_reqd_sub_group_size' metadata, I think it would be better just leave the fist line of the comment and remove the rest.

[pxli168]

lib/CodeGen/CodeGenFunction.cpp

650

OK. I will change it in next patch. And I think we no longer add these metadatas to the metadata node 'opencl.kernels'. We have change these to be function metadatas, this comment is out of date.

[pxli168]

Refine some comments.

[Anastasia]

lib/CodeGen/CodeGenFunction.cpp

701

Variable name should start in upper case.

[pxli168]

lib/CodeGen/CodeGenFunction.cpp

701

I think I should also change the whole function here. I just do some copy paste and don't notice this detail.

[pxli168]

Fix some varibale up case problems.

[Anastasia]

lib/Sema/SemaDeclAttr.cpp

2886

Could you add this warning to testing as well please.

[pxli168]

Add duplicate warning test case.

[Anastasia]

LGTM! Thanks!

[pxli168]

I found that there is a newly added patch about attribute that will fail the test, and I am working on it.
https://reviews.llvm.org/D30009
https://reviews.llvm.org/D32176

[pxli168]

Add attribute test.

[pxli168]

@Anastasia is this test added OK?

[Anastasia]

Sure! LGTM!

[pxli168]

Fix attribute test case.

include/clang/Basic/Attr.td

 }
 
 def OpenCLUnrollHint : InheritableAttr {
   let Spellings = [GNU<"opencl_unroll_hint">];
   let Args = [UnsignedArgument<"UnrollHint">];
   let Documentation = [OpenCLUnrollHintDocs];
 }
 
+def OpenCLIntelReqdSubGroupSize: InheritableAttr {
+  let Spellings = [GNU<"intel_reqd_sub_group_size">];
+  let Args = [UnsignedArgument<"SubGroupSize">];
+  let Subjects = SubjectList<[Function], ErrorDiag>;
+  let Documentation = [OpenCLIntelReqdSubGroupSizeDocs];
+}
+
 // This attribute is both a type attribute, and a declaration attribute (for
 // parameter variables).
 def OpenCLAccess : Attr {
   let Spellings = [Keyword<"__read_only">, Keyword<"read_only">,
                    Keyword<"__write_only">, Keyword<"write_only">,
                    Keyword<"__read_write">, Keyword<"read_write">];
   let Subjects = SubjectList<[ParmVar, TypedefName], ErrorDiag,
                              "ExpectedParameterOrTypedef">;

include/clang/Basic/AttrDocs.td

 (for, while and do loops) can be unrolled. This attribute qualifier can be
 used to specify full unrolling or partial unrolling by a specified amount.
 This is a compiler hint and the compiler may ignore this directive. See
 `OpenCL v2.0 <https://www.khronos.org/registry/cl/specs/opencl-2.0.pdf>`_
 s6.11.5 for details.
   }];
 }
 
+def OpenCLIntelReqdSubGroupSizeDocs : Documentation {
+  let Category = DocCatStmt;
+  let Heading = "__attribute__((intel_reqd_sub_group_size))";
+  let Content = [{
+The optional attribute intel_reqd_sub_group_size can be used to indicate that
+the kernel must be compiled and executed with the specified subgroup size. When
+this attribute is present, get_max_sub_group_size() is guaranteed to return the
+specified integer value. This is important for the correctness of many subgroup
+algorithms, and in some cases may be used by the compiler to generate more optimal
+code. See `cl_intel_required_subgroup_size
+<https://www.khronos.org/registry/OpenCL/extensions/intel/cl_intel_required_subgroup_size.txt>`
+for details.
+  }];
+}
+
 def OpenCLAccessDocs : Documentation {
   let Category = DocCatStmt;
   let Heading = "__read_only, __write_only, __read_write (read_only, write_only, read_write)";
   let Content = [{
 The access qualifiers must be used with image object arguments or pipe arguments
 to declare if they are being read or written by a kernel or function.
 
 The read_only/__read_only, write_only/__write_only and read_write/__read_write

lib/CodeGen/CodeGenFunction.h

private:
 
   llvm::BasicBlock *TerminateLandingPad;
   llvm::BasicBlock *TerminateHandler;
   llvm::BasicBlock *TrapBB;
 
   /// True if we need emit the life-time markers.
   const bool ShouldEmitLifetimeMarkers;
 
-  /// Add a kernel metadata node to the named metadata node 'opencl.kernels'.
-  /// In the kernel metadata node, reference the kernel function and metadata 
-  /// nodes for its optional attribute qualifiers (OpenCL 1.1 6.7.2):
-  /// - A node for the vec_type_hint(<type>) qualifier contains string
-  ///   "vec_type_hint", an undefined value of the <type> data type,
-  ///   and a Boolean that is true if the <type> is integer and signed.
-  /// - A node for the work_group_size_hint(X,Y,Z) qualifier contains string 
-  ///   "work_group_size_hint", and three 32-bit integers X, Y and Z.
-  /// - A node for the reqd_work_group_size(X,Y,Z) qualifier contains string 
-  ///   "reqd_work_group_size", and three 32-bit integers X, Y and Z.
+  /// Add OpenCL kernel arg metadata and the kernel attribute meatadata to
+  /// the function metadata.
   void EmitOpenCLKernelMetadata(const FunctionDecl *FD, 
                                 llvm::Function *Fn);
 
 public:
   CodeGenFunction(CodeGenModule &cgm, bool suppressNewContext=false);
   ~CodeGenFunction();
 
   CodeGenTypes &getTypes() const { return CGM.getTypes(); }

lib/CodeGen/CodeGenFunction.cpp

Fn->setMetadata("kernel_arg_base_type",
                   llvm::MDNode::get(Context, argBaseTypeNames));
   Fn->setMetadata("kernel_arg_type_qual",
                   llvm::MDNode::get(Context, argTypeQuals));
   if (CGM.getCodeGenOpts().EmitOpenCLArgMetadata)
     Fn->setMetadata("kernel_arg_name",
                     llvm::MDNode::get(Context, argNames));
 }
 
 void CodeGenFunction::EmitOpenCLKernelMetadata(const FunctionDecl *FD,
                                                llvm::Function *Fn)
 {
   if (!FD->hasAttr<OpenCLKernelAttr>())
     return;
 
   llvm::LLVMContext &Context = getLLVMContext();
 
   GenOpenCLArgMetadata(FD, Fn, CGM, Context, Builder, getContext());
 
   if (const VecTypeHintAttr *A = FD->getAttr<VecTypeHintAttr>()) {
-    QualType hintQTy = A->getTypeHint();
-    const ExtVectorType *hintEltQTy = hintQTy->getAs<ExtVectorType>();
-    bool isSignedInteger =
-        hintQTy->isSignedIntegerType() ||
-        (hintEltQTy && hintEltQTy->getElementType()->isSignedIntegerType());
-    llvm::Metadata *attrMDArgs[] = {
+    QualType HintQTy = A->getTypeHint();
+    const ExtVectorType *HintEltQTy = HintQTy->getAs<ExtVectorType>();
+    bool IsSignedInteger =
+        HintQTy->isSignedIntegerType() ||
+        (HintEltQTy && HintEltQTy->getElementType()->isSignedIntegerType());
+    llvm::Metadata *AttrMDArgs[] = {
         llvm::ConstantAsMetadata::get(llvm::UndefValue::get(
             CGM.getTypes().ConvertType(A->getTypeHint()))),
         llvm::ConstantAsMetadata::get(llvm::ConstantInt::get(
             llvm::IntegerType::get(Context, 32),
-            llvm::APInt(32, (uint64_t)(isSignedInteger ? 1 : 0))))};
-    Fn->setMetadata("vec_type_hint", llvm::MDNode::get(Context, attrMDArgs));
+            llvm::APInt(32, (uint64_t)(IsSignedInteger ? 1 : 0))))};
+    Fn->setMetadata("vec_type_hint", llvm::MDNode::get(Context, AttrMDArgs));
   }
 
   if (const WorkGroupSizeHintAttr *A = FD->getAttr<WorkGroupSizeHintAttr>()) {
-    llvm::Metadata *attrMDArgs[] = {
+    llvm::Metadata *AttrMDArgs[] = {
         llvm::ConstantAsMetadata::get(Builder.getInt32(A->getXDim())),
         llvm::ConstantAsMetadata::get(Builder.getInt32(A->getYDim())),
         llvm::ConstantAsMetadata::get(Builder.getInt32(A->getZDim()))};
-    Fn->setMetadata("work_group_size_hint", llvm::MDNode::get(Context, attrMDArgs));
+    Fn->setMetadata("work_group_size_hint", llvm::MDNode::get(Context, AttrMDArgs));
   }
 
   if (const ReqdWorkGroupSizeAttr *A = FD->getAttr<ReqdWorkGroupSizeAttr>()) {
-    llvm::Metadata *attrMDArgs[] = {
+    llvm::Metadata *AttrMDArgs[] = {
         llvm::ConstantAsMetadata::get(Builder.getInt32(A->getXDim())),
         llvm::ConstantAsMetadata::get(Builder.getInt32(A->getYDim())),
         llvm::ConstantAsMetadata::get(Builder.getInt32(A->getZDim()))};
-    Fn->setMetadata("reqd_work_group_size", llvm::MDNode::get(Context, attrMDArgs));
+    Fn->setMetadata("reqd_work_group_size", llvm::MDNode::get(Context, AttrMDArgs));
+  }
+
+  if (const OpenCLIntelReqdSubGroupSizeAttr *A =
+          FD->getAttr<OpenCLIntelReqdSubGroupSizeAttr>()) {
+    llvm::Metadata *AttrMDArgs[] = {
+        llvm::ConstantAsMetadata::get(Builder.getInt32(A->getSubGroupSize()))};
+    Fn->setMetadata("intel_reqd_sub_group_size",
+                    llvm::MDNode::get(Context, AttrMDArgs));
   }
 }
 
 /// Determine whether the function F ends with a return stmt.
 static bool endsWithReturn(const Decl* F) {
   const Stmt *Body = nullptr;
   if (auto *FD = dyn_cast_or_null<FunctionDecl>(F))
     Body = FD->getBody();
   else if (auto *OMD = dyn_cast_or_null<ObjCMethodDecl>(F))
     Body = OMD->getBody();
 
   if (auto *CS = dyn_cast_or_null<CompoundStmt>(Body)) {
     auto LastStmt = CS->body_rbegin();
     if (LastStmt != CS->body_rend())

lib/Sema/SemaDeclAttr.cpp

if (Existing && !(Existing->getXDim() == WGSize[0] &&
                     Existing->getZDim() == WGSize[2]))
     S.Diag(Attr.getLoc(), diag::warn_duplicate_attribute) << Attr.getName();
 
   D->addAttr(::new (S.Context) WorkGroupAttr(Attr.getRange(), S.Context,
                                              WGSize[0], WGSize[1], WGSize[2],
                                        Attr.getAttributeSpellingListIndex()));
 }
 
+// Handles intel_reqd_sub_group_size.
+static void handleSubGroupSize(Sema &S, Decl *D, const AttributeList &Attr) {
+  uint32_t SGSize;
+  const Expr *E = Attr.getArgAsExpr(0);
+  if (!checkUInt32Argument(S, Attr, E, SGSize))
+    return;
+  if (SGSize == 0) {
+    S.Diag(Attr.getLoc(), diag::err_attribute_argument_is_zero)
+        << Attr.getName() << E->getSourceRange();
+    return;
+  }
+
+  OpenCLIntelReqdSubGroupSizeAttr *Existing =
+      D->getAttr<OpenCLIntelReqdSubGroupSizeAttr>();
+  if (Existing && Existing->getSubGroupSize() != SGSize)
+    S.Diag(Attr.getLoc(), diag::warn_duplicate_attribute) << Attr.getName();
+
+  D->addAttr(::new (S.Context) OpenCLIntelReqdSubGroupSizeAttr(
+      Attr.getRange(), S.Context, SGSize,
+      Attr.getAttributeSpellingListIndex()));
+}
+
 static void handleVecTypeHint(Sema &S, Decl *D, const AttributeList &Attr) {
   if (!Attr.hasParsedType()) {
     S.Diag(Attr.getLoc(), diag::err_attribute_wrong_number_arguments)
       << Attr.getName() << 1;
     return;
   }
 
   TypeSourceInfo *ParmTSI = nullptr;
case AttributeList::AT_CFReturnsRetained:
     handleNSReturnsRetainedAttr(S, D, Attr);
     break;
   case AttributeList::AT_WorkGroupSizeHint:
     handleWorkGroupSize<WorkGroupSizeHintAttr>(S, D, Attr);
     break;
   case AttributeList::AT_ReqdWorkGroupSize:
     handleWorkGroupSize<ReqdWorkGroupSizeAttr>(S, D, Attr);
     break;
+  case AttributeList::AT_OpenCLIntelReqdSubGroupSize:
+    handleSubGroupSize(S, D, Attr);
+    break;
   case AttributeList::AT_VecTypeHint:
     handleVecTypeHint(S, D, Attr);
     break;
   case AttributeList::AT_RequireConstantInit:
     handleSimpleAttribute<RequireConstantInitAttr>(S, D, Attr);
     break;
   case AttributeList::AT_InitPriority:
     handleInitPriorityAttr(S, D, Attr);
if (!D->hasAttr<OpenCLKernelAttr>()) {
     } else if (Attr *A = D->getAttr<AMDGPUNumSGPRAttr>()) {
       Diag(D->getLocation(), diag::err_attribute_wrong_decl_type)
         << A << ExpectedKernelFunction;
       D->setInvalidDecl();
     } else if (Attr *A = D->getAttr<AMDGPUNumVGPRAttr>()) {
       Diag(D->getLocation(), diag::err_attribute_wrong_decl_type)
         << A << ExpectedKernelFunction;
       D->setInvalidDecl();
+    } else if (Attr *A = D->getAttr<OpenCLIntelReqdSubGroupSizeAttr>()) {
+      Diag(D->getLocation(), diag::err_opencl_kernel_attr) << A;
+      D->setInvalidDecl();
     }
   }
 }
 
 // Helper for delayed proccessing TransparentUnion attribute.
 void Sema::ProcessDeclAttributeDelayed(Decl *D, const AttributeList *AttrList) {
   for (const AttributeList *Attr = AttrList; Attr; Attr = Attr->getNext())
     if (Attr->getKind() == AttributeList::AT_TransparentUnion) {

test/CodeGenOpenCL/kernel-attributes.cl

 // RUN: %clang_cc1 -emit-llvm -o - %s | FileCheck %s
 
 typedef unsigned int uint4 __attribute__((ext_vector_type(4)));
 
 kernel  __attribute__((vec_type_hint(int))) __attribute__((reqd_work_group_size(1,2,4))) void kernel1(int a) {}
 // CHECK: define void @kernel1(i32 {{[^%]*}}%a) {{[^{]+}} !vec_type_hint ![[MD1:[0-9]+]] !reqd_work_group_size ![[MD2:[0-9]+]]
 
 kernel __attribute__((vec_type_hint(uint4))) __attribute__((work_group_size_hint(8,16,32))) void kernel2(int a) {}
 // CHECK: define void @kernel2(i32 {{[^%]*}}%a) {{[^{]+}} !vec_type_hint ![[MD3:[0-9]+]] !work_group_size_hint ![[MD4:[0-9]+]]
 
+kernel __attribute__((intel_reqd_sub_group_size(8))) void kernel3(int a) {}
+// CHECK: define void @kernel3(i32 {{[^%]*}}%a) {{[^{]+}} !intel_reqd_sub_group_size ![[MD5:[0-9]+]]
+
 // CHECK: [[MD1]] = !{i32 undef, i32 1}
 // CHECK: [[MD2]] = !{i32 1, i32 2, i32 4}
 // CHECK: [[MD3]] = !{<4 x i32> undef, i32 0}
 // CHECK: [[MD4]] = !{i32 8, i32 16, i32 32}
+// CHECK: [[MD5]] = !{i32 8}

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

 // RUN: clang-tblgen -gen-clang-test-pragma-attribute-supported-attributes -I%src_include_dir %src_include_dir/clang/Basic/Attr.td -o - | FileCheck %s
 
 // The number of supported attributes should never go down!
 
-// CHECK: #pragma clang attribute supports 58 attributes:
+// CHECK: #pragma clang attribute supports 59 attributes:
 // CHECK-NEXT: AMDGPUFlatWorkGroupSize (SubjectMatchRule_function)
 // CHECK-NEXT: AMDGPUNumSGPR (SubjectMatchRule_function)
 // CHECK-NEXT: AMDGPUNumVGPR (SubjectMatchRule_function)
 // CHECK-NEXT: AMDGPUWavesPerEU (SubjectMatchRule_function)
 // CHECK-NEXT: AVRSignal (SubjectMatchRule_function)
 // CHECK-NEXT: AbiTag (SubjectMatchRule_record_not_is_union, SubjectMatchRule_variable, SubjectMatchRule_function, SubjectMatchRule_namespace)
 // CHECK-NEXT: AlignValue (SubjectMatchRule_variable, SubjectMatchRule_type_alias)
 // CHECK-NEXT: AllocSize (SubjectMatchRule_function)

 // CHECK-NEXT: NoSplitStack (SubjectMatchRule_function)
 // CHECK-NEXT: NotTailCalled (SubjectMatchRule_function)
 // CHECK-NEXT: ObjCBoxable (SubjectMatchRule_record)
 // CHECK-NEXT: ObjCMethodFamily (SubjectMatchRule_objc_method)
 // CHECK-NEXT: ObjCRequiresSuper (SubjectMatchRule_objc_method)
 // CHECK-NEXT: ObjCRuntimeName (SubjectMatchRule_objc_interface, SubjectMatchRule_objc_protocol)
 // CHECK-NEXT: ObjCRuntimeVisible (SubjectMatchRule_objc_interface)
 // CHECK-NEXT: ObjCSubclassingRestricted (SubjectMatchRule_objc_interface)
+// CHECK-NEXT: OpenCLIntelReqdSubGroupSize (SubjectMatchRule_function)
 // CHECK-NEXT: OpenCLNoSVM (SubjectMatchRule_variable)
 // CHECK-NEXT: OptimizeNone (SubjectMatchRule_function, SubjectMatchRule_objc_method)
 // CHECK-NEXT: Overloadable (SubjectMatchRule_function)
 // CHECK-NEXT: ParamTypestate (SubjectMatchRule_variable_is_parameter)
 // CHECK-NEXT: PassObjectSize (SubjectMatchRule_variable_is_parameter)
 // CHECK-NEXT: RenderScriptKernel (SubjectMatchRule_function)
 // CHECK-NEXT: RequireConstantInit (SubjectMatchRule_variable_is_global)
 // CHECK-NEXT: ReturnTypestate (SubjectMatchRule_function, SubjectMatchRule_variable_is_parameter)

test/SemaOpenCL/invalid-kernel-attrs.cl

 
 void f_kernel_image2d_t( kernel image2d_t image ) { // expected-error {{'kernel' attribute only applies to functions}}
   int __kernel x; // expected-error {{'__kernel' attribute only applies to functions}}
 }
 
 kernel __attribute__((reqd_work_group_size(1,2,0))) void kernel11(){} // expected-error {{'reqd_work_group_size' attribute must be greater than 0}}
 kernel __attribute__((reqd_work_group_size(1,0,2))) void kernel12(){} // expected-error {{'reqd_work_group_size' attribute must be greater than 0}}
 kernel __attribute__((reqd_work_group_size(0,1,2))) void kernel13(){} // expected-error {{'reqd_work_group_size' attribute must be greater than 0}}
+
+__attribute__((intel_reqd_sub_group_size(8))) void kernel14(){} // expected-error {{attribute 'intel_reqd_sub_group_size' can only be applied to a kernel}}
+kernel __attribute__((intel_reqd_sub_group_size(0))) void kernel15(){} // expected-error {{'intel_reqd_sub_group_size' attribute must be greater than 0}}
+kernel __attribute__((intel_reqd_sub_group_size(8))) __attribute__((intel_reqd_sub_group_size(16))) void kernel16() {}  //expected-warning{{attribute 'intel_reqd_sub_group_size' is already applied with different parameters}}