Patch by Anton Lokhmotov to add OpenCL work group size attributes.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@159965 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/Attr.td b/include/clang/Basic/Attr.td
index 468d623eb7a896217a47882c3e9b54f62744d8a2..99180e450e65a5b9e170dc212039c4de2b82b907 100644 (file)
--- a/include/clang/Basic/Attr.td
+++ b/include/clang/Basic/Attr.td
@@ -578,6 +578,13 @@ def ReqdWorkGroupSize : InheritableAttr {
               UnsignedArgument<"ZDim">];
 }
 
+def WorkGroupSizeHint :  InheritableAttr {
+  let Spellings = [GNU<"work_group_size_hint">];
+  let Args = [UnsignedArgument<"XDim">, 
+              UnsignedArgument<"YDim">,
+              UnsignedArgument<"ZDim">];
+}
+
 def InitPriority : InheritableAttr {
   let Spellings = [GNU<"init_priority">];
   let Args = [UnsignedArgument<"Priority">];

diff --git a/include/clang/Basic/DiagnosticSemaKinds.td b/include/clang/Basic/DiagnosticSemaKinds.td
index 92bedbae3f58d6d9818af261291873790daf93d4..52641b86d69293f66cff6e993cea8d0402e46148 100644 (file)
--- a/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/include/clang/Basic/DiagnosticSemaKinds.td
@@ -5523,6 +5523,9 @@ def warn_collection_expr_type : Warning<
 def err_invalid_conversion_between_ext_vectors : Error<
   "invalid conversion between ext-vector type %0 and %1">;
 
+def warn_duplicate_attribute : Warning<
+  "attribute %0 is already applied with different parameters">;
+
 // Type
 def ext_invalid_sign_spec : Extension<"'%0' cannot be signed or unsigned">;
 def warn_receiver_forward_class : Warning<

diff --git a/lib/CodeGen/CodeGenFunction.cpp b/lib/CodeGen/CodeGenFunction.cpp
index 68593946f6ddaa8a87f4d96d0ec6281d0b62c631..c40b42193f9054a7d313fac76cf3247e45e78765 100644 (file)
--- a/lib/CodeGen/CodeGenFunction.cpp
+++ b/lib/CodeGen/CodeGenFunction.cpp
@@ -252,6 +252,51 @@ void CodeGenFunction::EmitMCountInstrumentation() {
   Builder.CreateCall(MCountFn);
 }
 
+void CodeGenFunction::EmitOpenCLKernelMetadata(const FunctionDecl *FD, 
+                                               llvm::Function *Fn)
+{
+  if (!FD->hasAttr<OpenCLKernelAttr>())
+    return;
+
+  llvm::LLVMContext &Context = getLLVMContext();
+
+  llvm::SmallVector <llvm::Value*, 5> kernelMDArgs;
+  kernelMDArgs.push_back(Fn);
+
+  if (FD->hasAttr<WorkGroupSizeHintAttr>()) {
+    llvm::SmallVector <llvm::Value*, 5> attrMDArgs;
+    attrMDArgs.push_back(llvm::MDString::get(Context, "work_group_size_hint"));
+    WorkGroupSizeHintAttr *attr = FD->getAttr<WorkGroupSizeHintAttr>();
+    llvm::Type *iTy = llvm::IntegerType::get(Context, 32);
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getXDim())));
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getYDim())));
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getZDim())));
+    kernelMDArgs.push_back(llvm::MDNode::get(Context, attrMDArgs));
+  }
+
+  if (FD->hasAttr<ReqdWorkGroupSizeAttr>()) {
+    llvm::SmallVector <llvm::Value*, 5> attrMDArgs;
+    attrMDArgs.push_back(llvm::MDString::get(Context, "reqd_work_group_size"));
+    ReqdWorkGroupSizeAttr *attr = FD->getAttr<ReqdWorkGroupSizeAttr>();
+    llvm::Type *iTy = llvm::IntegerType::get(Context, 32);
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getXDim())));
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getYDim())));
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getZDim())));
+    kernelMDArgs.push_back(llvm::MDNode::get(Context, attrMDArgs));
+  }
+
+  llvm::MDNode *kernelMDNode = llvm::MDNode::get(Context, kernelMDArgs);
+  llvm::NamedMDNode *OpenCLKernelMetadata =
+    CGM.getModule().getOrInsertNamedMetadata("opencl.kernels");
+  OpenCLKernelMetadata->addOperand(kernelMDNode);
+}
+
 void CodeGenFunction::StartFunction(GlobalDecl GD, QualType RetTy,
                                     llvm::Function *Fn,
                                     const CGFunctionInfo &FnInfo,
@@ -280,14 +325,7 @@ void CodeGenFunction::StartFunction(GlobalDecl GD, QualType RetTy,
   if (getContext().getLangOpts().OpenCL) {
     // Add metadata for a kernel function.
     if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D))
-      if (FD->hasAttr<OpenCLKernelAttr>()) {
-        llvm::LLVMContext &Context = getLLVMContext();
-        llvm::NamedMDNode *OpenCLMetadata = 
-          CGM.getModule().getOrInsertNamedMetadata("opencl.kernels");
-          
-        llvm::Value *Op = Fn;
-        OpenCLMetadata->addOperand(llvm::MDNode::get(Context, Op));
-      }
+      EmitOpenCLKernelMetadata(FD, Fn);
   }
 
   llvm::BasicBlock *EntryBB = createBasicBlock("entry", CurFn);

diff --git a/lib/CodeGen/CodeGenFunction.h b/lib/CodeGen/CodeGenFunction.h
index a3d1acf5917f2935acc671e343ef64abcada2e87..b737a9ed7ab739d5a404967875ae36849af0a81a 100644 (file)
--- a/lib/CodeGen/CodeGenFunction.h
+++ b/lib/CodeGen/CodeGenFunction.h
@@ -1197,6 +1197,16 @@ private:
   llvm::BasicBlock *TerminateHandler;
   llvm::BasicBlock *TrapBB;
 
+  /// 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 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.
+  void EmitOpenCLKernelMetadata(const FunctionDecl *FD, 
+                                llvm::Function *Fn);
+
 public:
   CodeGenFunction(CodeGenModule &cgm, bool suppressNewContext=false);
   ~CodeGenFunction();

diff --git a/lib/Sema/SemaDeclAttr.cpp b/lib/Sema/SemaDeclAttr.cpp
index 744d18bfd9fbda8fdc569e26b1852d3307b321b0..65fceaa93add73b707763cb82420b429f0a59c9b 100644 (file)
--- a/lib/Sema/SemaDeclAttr.cpp
+++ b/lib/Sema/SemaDeclAttr.cpp
@@ -2355,11 +2355,14 @@ static void handleWeakImportAttr(Sema &S, Decl *D, const AttributeList &Attr) {
   D->addAttr(::new (S.Context) WeakImportAttr(Attr.getRange(), S.Context));
 }
 
-static void handleReqdWorkGroupSize(Sema &S, Decl *D,
-                                    const AttributeList &Attr) {
+// Handles reqd_work_group_size and work_group_size_hint.
+static void handleWorkGroupSize(Sema &S, Decl *D,
+                               const AttributeList &Attr) {
+  assert(Attr.getKind() == AttributeList::AT_ReqdWorkGroupSize 
+      || Attr.getKind() == AttributeList::AT_WorkGroupSizeHint);
+
   // Attribute has 3 arguments.
-  if (!checkAttributeNumArgs(S, Attr, 3))
-    return;
+  if (!checkAttributeNumArgs(S, Attr, 3)) return;
 
   unsigned WGSize[3];
   for (unsigned i = 0; i < 3; ++i) {
@@ -2368,14 +2371,42 @@ static void handleReqdWorkGroupSize(Sema &S, Decl *D,
     if (E->isTypeDependent() || E->isValueDependent() ||
         !E->isIntegerConstantExpr(ArgNum, S.Context)) {
       S.Diag(Attr.getLoc(), diag::err_attribute_argument_not_int)
-        << "reqd_work_group_size" << E->getSourceRange();
+        << Attr.getName()->getName() << E->getSourceRange();
       return;
     }
     WGSize[i] = (unsigned) ArgNum.getZExtValue();
   }
-  D->addAttr(::new (S.Context) ReqdWorkGroupSizeAttr(Attr.getRange(), S.Context,
-                                                     WGSize[0], WGSize[1],
-                                                     WGSize[2]));
+
+  if (Attr.getKind() == AttributeList::AT_ReqdWorkGroupSize
+    && D->hasAttr<ReqdWorkGroupSizeAttr>()) {
+      ReqdWorkGroupSizeAttr *A = D->getAttr<ReqdWorkGroupSizeAttr>();
+      if (!(A->getXDim() == WGSize[0] &&
+            A->getYDim() == WGSize[1] &&
+            A->getZDim() == WGSize[2])) {
+        S.Diag(Attr.getLoc(), diag::warn_duplicate_attribute) <<
+          Attr.getName();
+      }
+  }
+
+  if (Attr.getKind() == AttributeList::AT_WorkGroupSizeHint
+    && D->hasAttr<WorkGroupSizeHintAttr>()) {
+      WorkGroupSizeHintAttr *A = D->getAttr<WorkGroupSizeHintAttr>();
+      if (!(A->getXDim() == WGSize[0] &&
+            A->getYDim() == WGSize[1] &&
+            A->getZDim() == WGSize[2])) {
+        S.Diag(Attr.getLoc(), diag::warn_duplicate_attribute) <<
+          Attr.getName();
+      }
+  }
+
+  if (Attr.getKind() == AttributeList::AT_ReqdWorkGroupSize)
+    D->addAttr(::new (S.Context)
+                 ReqdWorkGroupSizeAttr(Attr.getRange(), S.Context,
+                                       WGSize[0], WGSize[1], WGSize[2]));
+  else
+    D->addAttr(::new (S.Context)
+                 WorkGroupSizeHintAttr(Attr.getRange(), S.Context,
+                                       WGSize[0], WGSize[1], WGSize[2]));
 }
 
 SectionAttr *Sema::mergeSectionAttr(Decl *D, SourceRange Range,
@@ -4043,8 +4074,9 @@ static void ProcessInheritableDeclAttr(Sema &S, Scope *scope, Decl *D,
   case AttributeList::AT_CFReturnsRetained:
     handleNSReturnsRetainedAttr(S, D, Attr); break;
 
+  case AttributeList::AT_WorkGroupSizeHint:
   case AttributeList::AT_ReqdWorkGroupSize:
-    handleReqdWorkGroupSize(S, D, Attr); break;
+    handleWorkGroupSize(S, D, Attr); break;
 
   case AttributeList::AT_InitPriority: 
       handleInitPriorityAttr(S, D, Attr); break;

diff --git a/test/CodeGenOpenCL/kernel-attributes.cl b/test/CodeGenOpenCL/kernel-attributes.cl
new file mode 100644 (file)
index 0000000..014285f
--- /dev/null
+++ b/test/CodeGenOpenCL/kernel-attributes.cl
@@ -0,0 +1,14 @@
+// RUN: %clang_cc1 -emit-llvm -O0 -o - %s | FileCheck %s
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+
+kernel __attribute__((reqd_work_group_size(1,2,4))) void kernel1(int a) {}
+
+kernel __attribute__((work_group_size_hint(8,16,32))) void kernel2(float4 a) {}
+
+// CHECK: opencl.kernels = !{[[MDNODE0:![0-9]+]], [[MDNODE3:![0-9]+]]}
+
+// CHECK: [[MDNODE0]] = metadata !{void (i32)* @kernel1, metadata [[MDNODE2:![0-9]+]]}
+// CHECK: [[MDNODE2]] = metadata !{metadata !"reqd_work_group_size", i32 1, i32 2, i32 4}
+// CHECK: [[MDNODE3]] = metadata !{void (<4 x float>)* @kernel2, metadata [[MDNODE5:![0-9]+]]}
+// CHECK: [[MDNODE5]] = metadata !{metadata !"work_group_size_hint", i32 8, i32 16, i32 32}