AST, Sema, Serialization: add CUDAKernelCallExpr and related semantic actions

llvm-svn: 125217

clang/include/clang/AST/ExprCXX.h

@@ -116,6 +116,35 @@ public:
   static bool classof(const CXXMemberCallExpr *) { return true; }
 };
 
+/// CUDAKernelCallExpr - Represents a call to a CUDA kernel function.
+class CUDAKernelCallExpr : public CallExpr {
+private:
+  enum { CONFIG, END_PREARG };
+
+public:
+  CUDAKernelCallExpr(ASTContext &C, Expr *fn, CallExpr *Config,
+                     Expr **args, unsigned numargs, QualType t,
+                     ExprValueKind VK, SourceLocation RP)
+    : CallExpr(C, CUDAKernelCallExprClass, fn, END_PREARG, args, numargs, t, VK,
+               RP) {
+    setConfig(Config);
+  }
+
+  CUDAKernelCallExpr(ASTContext &C, EmptyShell Empty)
+    : CallExpr(C, CUDAKernelCallExprClass, END_PREARG, Empty) { }
+
+  const CallExpr *getConfig() const {
+    return cast_or_null<CallExpr>(getPreArg(CONFIG));
+  }
+  CallExpr *getConfig() { return cast_or_null<CallExpr>(getPreArg(CONFIG)); }
+  void setConfig(CallExpr *E) { setPreArg(CONFIG, E); }
+
+  static bool classof(const Stmt *T) {
+    return T->getStmtClass() == CUDAKernelCallExprClass;
+  }
+  static bool classof(const CUDAKernelCallExpr *) { return true; }
+};
+
 /// CXXNamedCastExpr - Abstract class common to all of the C++ "named"
 /// casts, @c static_cast, @c dynamic_cast, @c reinterpret_cast, or @c
 /// const_cast.

clang/include/clang/AST/RecursiveASTVisitor.h

@@ -1846,6 +1846,7 @@ DEF_TRAVERSE_STMT(UnresolvedMemberExpr, {
 
 DEF_TRAVERSE_STMT(CXXOperatorCallExpr, { })
 DEF_TRAVERSE_STMT(OpaqueValueExpr, { })
+DEF_TRAVERSE_STMT(CUDAKernelCallExpr, { })
 
 // These operators (all of them) do not need any action except
 // iterating over the children.

clang/include/clang/Basic/StmtNodes.td

@@ -125,6 +125,9 @@ def ObjCIvarRefExpr : DStmt<Expr>;
 def ObjCPropertyRefExpr : DStmt<Expr>;
 def ObjCIsaExpr : DStmt<Expr>;
 
+// CUDA Expressions.
+def CUDAKernelCallExpr : DStmt<CallExpr>;
+
 // Clang Extensions.
 def ShuffleVectorExpr : DStmt<Expr>;
 def BlockExpr : DStmt<Expr>;

clang/include/clang/Sema/Sema.h

@@ -1177,7 +1177,8 @@ public:
                                      UnresolvedLookupExpr *ULE,
                                      SourceLocation LParenLoc,
                                      Expr **Args, unsigned NumArgs,
-                                     SourceLocation RParenLoc);
+                                     SourceLocation RParenLoc,
+                                     Expr *ExecConfig);
 
   ExprResult CreateOverloadedUnaryOp(SourceLocation OpLoc,
                                      unsigned Opc,
@@ -1913,11 +1914,16 @@ public:
   /// This provides the location of the left/right parens and a list of comma
   /// locations.
   ExprResult ActOnCallExpr(Scope *S, Expr *Fn, SourceLocation LParenLoc,
-                           MultiExprArg Args, SourceLocation RParenLoc);
+                           MultiExprArg Args, SourceLocation RParenLoc,
+                           Expr *ExecConfig = 0);
   ExprResult BuildResolvedCallExpr(Expr *Fn, NamedDecl *NDecl,
                                    SourceLocation LParenLoc,
                                    Expr **Args, unsigned NumArgs,
-                                   SourceLocation RParenLoc);
+                                   SourceLocation RParenLoc,
+                                   Expr *ExecConfig = 0);
+
+  ExprResult ActOnCUDAExecConfigExpr(Scope *S, SourceLocation LLLLoc,
+                                MultiExprArg ExecConfig, SourceLocation GGGLoc);
 
   ExprResult ActOnCastExpr(Scope *S, SourceLocation LParenLoc,
                            ParsedType Ty, SourceLocation RParenLoc,

clang/include/clang/Serialization/ASTBitCodes.h

@@ -949,7 +949,11 @@ namespace clang {
       
       EXPR_PACK_EXPANSION,        // PackExpansionExpr
       EXPR_SIZEOF_PACK,           // SizeOfPackExpr
-      EXPR_SUBST_NON_TYPE_TEMPLATE_PARM_PACK // SubstNonTypeTemplateParmPackExpr
+      EXPR_SUBST_NON_TYPE_TEMPLATE_PARM_PACK,// SubstNonTypeTemplateParmPackExpr
+
+      // CUDA
+
+      EXPR_CUDA_KERNEL_CALL       // CUDAKernelCallExpr
     };
 
     /// \brief The kinds of designators that can occur in a

clang/lib/AST/ExprClassification.cpp

@@ -238,6 +238,7 @@ static Cl::Kinds ClassifyInternal(ASTContext &Ctx, const Expr *E) {
   case Expr::CallExprClass:
   case Expr::CXXOperatorCallExprClass:
   case Expr::CXXMemberCallExprClass:
+  case Expr::CUDAKernelCallExprClass:
     return ClassifyUnnamed(Ctx, cast<CallExpr>(E)->getCallReturnType());
 
     // __builtin_choose_expr is equivalent to the chosen expression.

clang/lib/AST/ExprConstant.cpp

@@ -2599,6 +2599,7 @@ static ICEDiag CheckICE(const Expr* E, ASTContext &Ctx) {
   case Expr::AddrLabelExprClass:
   case Expr::StmtExprClass:
   case Expr::CXXMemberCallExprClass:
+  case Expr::CUDAKernelCallExprClass:
   case Expr::CXXDynamicCastExprClass:
   case Expr::CXXTypeidExprClass:
   case Expr::CXXUuidofExprClass:

clang/lib/AST/ItaniumMangle.cpp

@@ -1732,7 +1732,8 @@ void CXXNameMangler::mangleExpression(const Expr *E, unsigned Arity) {
   case Expr::BinaryTypeTraitExprClass:
   case Expr::VAArgExprClass:
   case Expr::CXXUuidofExprClass:
-  case Expr::CXXNoexceptExprClass: {
+  case Expr::CXXNoexceptExprClass:
+  case Expr::CUDAKernelCallExprClass: {
     // As bad as this diagnostic is, it's better than crashing.
     Diagnostic &Diags = Context.getDiags();
     unsigned DiagID = Diags.getCustomDiagID(Diagnostic::Error,

clang/lib/AST/StmtPrinter.cpp

@@ -961,6 +961,15 @@ void StmtPrinter::VisitCXXMemberCallExpr(CXXMemberCallExpr *Node) {
   VisitCallExpr(cast<CallExpr>(Node));
 }
 
+void StmtPrinter::VisitCUDAKernelCallExpr(CUDAKernelCallExpr *Node) {
+  PrintExpr(Node->getCallee());
+  OS << "<<<";
+  PrintCallArgs(Node->getConfig());
+  OS << ">>>(";
+  PrintCallArgs(Node);
+  OS << ")";
+}
+
 void StmtPrinter::VisitCXXNamedCastExpr(CXXNamedCastExpr *Node) {
   OS << Node->getCastName() << '<';
   OS << Node->getTypeAsWritten().getAsString(Policy) << ">(";

clang/lib/AST/StmtProfile.cpp

@@ -644,6 +644,10 @@ void StmtProfiler::VisitCXXMemberCallExpr(CXXMemberCallExpr *S) {
   VisitCallExpr(S);
 }
 
+void StmtProfiler::VisitCUDAKernelCallExpr(CUDAKernelCallExpr *S) {
+  VisitCallExpr(S);
+}
+
 void StmtProfiler::VisitCXXNamedCastExpr(CXXNamedCastExpr *S) {
   VisitExplicitCastExpr(S);
 }

clang/lib/Sema/SemaExpr.cpp

@@ -4362,7 +4362,8 @@ bool Sema::GatherArgumentsForCall(SourceLocation CallLoc,
 /// locations.
 ExprResult
 Sema::ActOnCallExpr(Scope *S, Expr *Fn, SourceLocation LParenLoc,
-                    MultiExprArg args, SourceLocation RParenLoc) {
+                    MultiExprArg args, SourceLocation RParenLoc,
+                    Expr *ExecConfig) {
   unsigned NumArgs = args.size();
 
   // Since this might be a postfix expression, get rid of ParenListExprs.
@@ -4399,10 +4400,17 @@ Sema::ActOnCallExpr(Scope *S, Expr *Fn, SourceLocation LParenLoc,
     else if (Expr::hasAnyTypeDependentArguments(Args, NumArgs))
       Dependent = true;
 
-    if (Dependent)
-      return Owned(new (Context) CallExpr(Context, Fn, Args, NumArgs,
-                                          Context.DependentTy, VK_RValue,
-                                          RParenLoc));
+    if (Dependent) {
+      if (ExecConfig) {
+        return Owned(new (Context) CUDAKernelCallExpr(
+            Context, Fn, cast<CallExpr>(ExecConfig), Args, NumArgs,
+            Context.DependentTy, VK_RValue, RParenLoc));
+      } else {
+        return Owned(new (Context) CallExpr(Context, Fn, Args, NumArgs,
+                                            Context.DependentTy, VK_RValue,
+                                            RParenLoc));
+      }
+    }
 
     // Determine whether this is a call to an object (C++ [over.call.object]).
     if (Fn->getType()->isRecordType())
@@ -4495,7 +4503,7 @@ Sema::ActOnCallExpr(Scope *S, Expr *Fn, SourceLocation LParenLoc,
   if (isa<UnresolvedLookupExpr>(NakedFn)) {
     UnresolvedLookupExpr *ULE = cast<UnresolvedLookupExpr>(NakedFn);
     return BuildOverloadedCallExpr(S, Fn, ULE, LParenLoc, Args, NumArgs,
-                                   RParenLoc);
+                                   RParenLoc, ExecConfig);
   }
 
   NamedDecl *NDecl = 0;
@@ -4506,7 +4514,23 @@ Sema::ActOnCallExpr(Scope *S, Expr *Fn, SourceLocation LParenLoc,
   if (isa<DeclRefExpr>(NakedFn))
     NDecl = cast<DeclRefExpr>(NakedFn)->getDecl();
 
-  return BuildResolvedCallExpr(Fn, NDecl, LParenLoc, Args, NumArgs, RParenLoc);
+  return BuildResolvedCallExpr(Fn, NDecl, LParenLoc, Args, NumArgs, RParenLoc,
+                               ExecConfig);
+}
+
+ExprResult
+Sema::ActOnCUDAExecConfigExpr(Scope *S, SourceLocation LLLLoc,
+                              MultiExprArg execConfig, SourceLocation GGGLoc) {
+  FunctionDecl *ConfigDecl = Context.getcudaConfigureCallDecl();
+  if (!ConfigDecl)
+    return ExprError(Diag(LLLLoc, diag::err_undeclared_var_use)
+                          << "cudaConfigureCall");
+  QualType ConfigQTy = ConfigDecl->getType();
+
+  DeclRefExpr *ConfigDR = new (Context) DeclRefExpr(
+      ConfigDecl, ConfigQTy, VK_LValue, LLLLoc);
+
+  return ActOnCallExpr(S, ConfigDR, LLLLoc, execConfig, GGGLoc, 0);
 }
 
 /// BuildResolvedCallExpr - Build a call to a resolved expression,
@@ -4519,7 +4543,8 @@ ExprResult
 Sema::BuildResolvedCallExpr(Expr *Fn, NamedDecl *NDecl,
                             SourceLocation LParenLoc,
                             Expr **Args, unsigned NumArgs,
-                            SourceLocation RParenLoc) {
+                            SourceLocation RParenLoc,
+                            Expr *Config) {
   FunctionDecl *FDecl = dyn_cast_or_null<FunctionDecl>(NDecl);
 
   // Promote the function operand.
@@ -4527,11 +4552,21 @@ Sema::BuildResolvedCallExpr(Expr *Fn, NamedDecl *NDecl,
 
   // Make the call expr early, before semantic checks.  This guarantees cleanup
   // of arguments and function on error.
-  CallExpr *TheCall = new (Context) CallExpr(Context, Fn,
-                                             Args, NumArgs,
-                                             Context.BoolTy,
-                                             VK_RValue,
-                                             RParenLoc);
+  CallExpr *TheCall;
+  if (Config) {
+    TheCall = new (Context) CUDAKernelCallExpr(Context, Fn,
+                                               cast<CallExpr>(Config),
+                                               Args, NumArgs,
+                                               Context.BoolTy,
+                                               VK_RValue,
+                                               RParenLoc);
+  } else {
+    TheCall = new (Context) CallExpr(Context, Fn,
+                                     Args, NumArgs,
+                                     Context.BoolTy,
+                                     VK_RValue,
+                                     RParenLoc);
+  }
 
   const FunctionType *FuncT;
   if (!Fn->getType()->isBlockPointerType()) {

clang/lib/Sema/SemaOverload.cpp

@@ -7426,7 +7426,8 @@ ExprResult
 Sema::BuildOverloadedCallExpr(Scope *S, Expr *Fn, UnresolvedLookupExpr *ULE,
                               SourceLocation LParenLoc,
                               Expr **Args, unsigned NumArgs,
-                              SourceLocation RParenLoc) {
+                              SourceLocation RParenLoc,
+                              Expr *ExecConfig) {
 #ifndef NDEBUG
   if (ULE->requiresADL()) {
     // To do ADL, we must have found an unqualified name.
@@ -7466,8 +7467,8 @@ Sema::BuildOverloadedCallExpr(Scope *S, Expr *Fn, UnresolvedLookupExpr *ULE,
     DiagnoseUseOfDecl(FDecl? FDecl : Best->FoundDecl.getDecl(),
                       ULE->getNameLoc());
     Fn = FixOverloadedFunctionReference(Fn, Best->FoundDecl, FDecl);
-    return BuildResolvedCallExpr(Fn, FDecl, LParenLoc, Args, NumArgs,
-                                 RParenLoc);
+    return BuildResolvedCallExpr(Fn, FDecl, LParenLoc, Args, NumArgs, RParenLoc,
+                                 ExecConfig);
   }
 
   case OR_No_Viable_Function:

clang/lib/Sema/TreeTransform.h

@@ -1348,9 +1348,10 @@ public:
   /// Subclasses may override this routine to provide different behavior.
   ExprResult RebuildCallExpr(Expr *Callee, SourceLocation LParenLoc,
                                    MultiExprArg Args,
-                                   SourceLocation RParenLoc) {
+                                   SourceLocation RParenLoc,
+                                   Expr *ExecConfig = 0) {
     return getSema().ActOnCallExpr(/*Scope=*/0, Callee, LParenLoc,
-                                   move(Args), RParenLoc);
+                                   move(Args), RParenLoc, ExecConfig);
   }
 
   /// \brief Build a new member access expression.
@@ -5919,6 +5920,39 @@ TreeTransform<Derived>::TransformCXXMemberCallExpr(CXXMemberCallExpr *E) {
   return getDerived().TransformCallExpr(E);
 }
 
+template<typename Derived>
+ExprResult
+TreeTransform<Derived>::TransformCUDAKernelCallExpr(CUDAKernelCallExpr *E) {
+  // Transform the callee.
+  ExprResult Callee = getDerived().TransformExpr(E->getCallee());
+  if (Callee.isInvalid())
+    return ExprError();
+
+  // Transform exec config.
+  ExprResult EC = getDerived().TransformCallExpr(E->getConfig());
+  if (EC.isInvalid())
+    return ExprError();
+
+  // Transform arguments.
+  bool ArgChanged = false;
+  ASTOwningVector<Expr*> Args(SemaRef);
+  if (getDerived().TransformExprs(E->getArgs(), E->getNumArgs(), true, Args, 
+                                  &ArgChanged))
+    return ExprError();
+
+  if (!getDerived().AlwaysRebuild() &&
+      Callee.get() == E->getCallee() &&
+      !ArgChanged)
+    return SemaRef.Owned(E);
+
+  // FIXME: Wrong source location information for the '('.
+  SourceLocation FakeLParenLoc
+    = ((Expr *)Callee.get())->getSourceRange().getBegin();
+  return getDerived().RebuildCallExpr(Callee.get(), FakeLParenLoc,
+                                      move_arg(Args),
+                                      E->getRParenLoc(), EC.get());
+}
+
 template<typename Derived>
 ExprResult
 TreeTransform<Derived>::TransformCXXNamedCastExpr(CXXNamedCastExpr *E) {

clang/lib/Serialization/ASTReaderStmt.cpp

@@ -182,6 +182,9 @@ namespace clang {
     void VisitSubstNonTypeTemplateParmPackExpr(
                                            SubstNonTypeTemplateParmPackExpr *E);
     void VisitOpaqueValueExpr(OpaqueValueExpr *E);
+    
+    // CUDA Expressions
+    void VisitCUDAKernelCallExpr(CUDAKernelCallExpr *E);
   };
 }
 
@@ -1323,6 +1326,15 @@ void ASTStmtReader::VisitOpaqueValueExpr(OpaqueValueExpr *E) {
   E->Loc = ReadSourceLocation(Record, Idx);
 }
 
+//===----------------------------------------------------------------------===//
+// CUDA Expressions and Statements
+//===----------------------------------------------------------------------===//
+
+void ASTStmtReader::VisitCUDAKernelCallExpr(CUDAKernelCallExpr *E) {
+  VisitCallExpr(E);
+  E->setConfig(cast<CallExpr>(Reader.ReadSubExpr()));
+}
+
 Stmt *ASTReader::ReadStmt(PerFileData &F) {
   switch (ReadingKind) {
   case Read_Decl:
@@ -1872,6 +1884,10 @@ Stmt *ASTReader::ReadStmtFromStream(PerFileData &F) {
     case EXPR_OPAQUE_VALUE:
       S = new (Context) OpaqueValueExpr(Empty);
       break;
+
+    case EXPR_CUDA_KERNEL_CALL:
+      S = new (Context) CUDAKernelCallExpr(*Context, Empty);
+      break;
     }
     
     // We hit a STMT_STOP, so we're done with this expression.

clang/lib/Serialization/ASTWriter.cpp

@@ -691,6 +691,7 @@ static void AddStmtsExprs(llvm::BitstreamWriter &Stream,
   RECORD(EXPR_PACK_EXPANSION);
   RECORD(EXPR_SIZEOF_PACK);
   RECORD(EXPR_SUBST_NON_TYPE_TEMPLATE_PARM_PACK);
+  RECORD(EXPR_CUDA_KERNEL_CALL);
 #undef RECORD
 }
 

clang/lib/Serialization/ASTWriterStmt.cpp

@@ -156,6 +156,9 @@ namespace clang {
     void VisitSubstNonTypeTemplateParmPackExpr(
                                            SubstNonTypeTemplateParmPackExpr *E);
     void VisitOpaqueValueExpr(OpaqueValueExpr *E);
+
+    // CUDA Expressions
+    void VisitCUDAKernelCallExpr(CUDAKernelCallExpr *E);
   };
 }
 
@@ -1324,6 +1327,16 @@ void ASTStmtWriter::VisitOpaqueValueExpr(OpaqueValueExpr *E) {
   Code = serialization::EXPR_OPAQUE_VALUE;
 }
 
+//===----------------------------------------------------------------------===//
+// CUDA Expressions and Statements.
+//===----------------------------------------------------------------------===//
+
+void ASTStmtWriter::VisitCUDAKernelCallExpr(CUDAKernelCallExpr *E) {
+  VisitCallExpr(E);
+  Writer.AddStmt(E->getConfig());
+  Code = serialization::EXPR_CUDA_KERNEL_CALL;
+}
+
 //===----------------------------------------------------------------------===//
 // ASTWriter Implementation
 //===----------------------------------------------------------------------===//

clang/lib/StaticAnalyzer/Checkers/ExprEngine.cpp

@@ -894,6 +894,7 @@ void ExprEngine::Visit(const Stmt* S, ExplodedNode* Pred,
     case Stmt::PredefinedExprClass:
     case Stmt::ShuffleVectorExprClass:
     case Stmt::VAArgExprClass:
+    case Stmt::CUDAKernelCallExprClass:
         // Fall through.
 
     // Cases we intentionally don't evaluate, since they don't need

clang/tools/libclang/CXCursor.cpp

@@ -187,6 +187,7 @@ CXCursor cxcursor::MakeCXCursor(Stmt *S, Decl *Parent,
   case Stmt::CallExprClass:              
   case Stmt::CXXOperatorCallExprClass:
   case Stmt::CXXMemberCallExprClass:
+  case Stmt::CUDAKernelCallExprClass:
   case Stmt::CXXConstructExprClass:  
   case Stmt::CXXTemporaryObjectExprClass:
     // FIXME: CXXUnresolvedConstructExpr