[OPENMP] Introduced type trait "__builtin_omp_required_simd_align" for default simd alignment.

[ABataev]

Summary

Adds type trait "__builtin_omp_required_simd_align" after discussions here http://reviews.llvm.org/D9894

[rjmccall]

include/clang/AST/ASTContext.h

85

Just make this a separate query on ASTContext. TypeInfo computation is important to a lot of different clients, basically none of which care about this.

I don't think it's important to cache.

include/clang/Basic/TokenKinds.def

507

This is a very general name to give this, and putting it in the unreserved namespace is problematic. I think that, given some time and coordination, we could design a more general "what's the largest supported vector type for this underlying type" query, and that would justify a name like this. But the existing vector attributes always talk about, well, "vector" instead of "simd"; and anyway, it's not obvious that the answer to that general query would necessarily dictate the answer to this OpenMP-specific one.

So let's design this as a more targeted feature around the needs of OpenMP. In that context, "simd" is fine because it's exactly the name of the OpenMP feature. How about "builtin_openmp_simd_align"? Or even "builtin_openmp_required_simd_align"?

lib/AST/ItaniumMangle.cpp

3024

Copy-paste error.

lib/Basic/Targets.cpp

2979

Aren't we saying that this is 64 for AVX512?

lib/CodeGen/CGExprScalar.cpp

2044

This needs to produce a size_t, not an int32_t. You can use CGM::getSize.

lib/Sema/SemaExpr.cpp

3522

This should be a hard error, not just an extension warning.

3829

Hmm. Thinking more about it, using this trait with an expression instead of a type seems pretty misleading. Let's make this a hard error for now. We can continue to use the UETT parsing logic and AST representation.

[ABataev]

John, thank you for the review!

include/clang/AST/ASTContext.h

85

Ok, will separate it.

include/clang/Basic/TokenKinds.def

507

Ok, turned it to __builtin_omp_required_simd_align. Actually, I thought about such kind of name, but it was too long, so I changed it to a shorter version. :)

lib/AST/ItaniumMangle.cpp

3024

Fixed, thanks

lib/Basic/Targets.cpp

2979

There were no "avx512" defined when I prepared a patch. Will be added

lib/CodeGen/CGExprScalar.cpp

2044

Thanks, fixed

lib/Sema/SemaExpr.cpp

3522

Will be removed it and expressions won't be allowed as arguments

3829

Ok

[ABataev]

Fixed after review.

[rjmccall]

"omp" instead of "openmp" makes a lot sense, since that's what's used in the pragma.

include/clang/Basic/DiagnosticSemaKinds.td

4688

You lost the 0 here.

include/clang/Basic/TargetInfo.h

397

Please embellish this comment to make it clear that it's sometimes type-specific.

It would be great if we could move the type-specific logic to TargetInfo, but I don't know of a way to do that without either (1) making a lot of assumptions about what kinds of types are interesting to the target or (2) creating a cyclic dependency between lib/Basic and lib/AST. Ugh. We can stick with what we have right now.

lib/AST/ASTContext.cpp

1791

I think this query should probably apply to the underlying type, not to the pointer type. That is, if I want to know the expected alignment of an array of doubles, I should use

__builtin_omp_required_simd_align(double)

not

__builtin_omp_required_simd_align(double*)

Unless PPC64's rule really is about arrays of pointers to double?

lib/CodeGen/CGExprScalar.cpp

2045

I assume the type of the argument here is always a pointer, which means you'll need to drill down to the element type.

lib/Parse/ParseExpr.cpp

1046

These comments are supposed to be grammar specifications; whoever implemented vec_step apparently missed that.

What you want is:

unary-expression: '__builtin_omp_required_simd_align' '(' type-name ')'

(since we don't actually allow the expression form semantically)

[ABataev]

Updated after review

[rjmccall]

One more tweak, and then LGTM.

include/clang/Basic/TypeTraits.h

96

Please rename this to follow the actual name of the trait, i.e. UETT_OpenMPRequiredSimdAlign.

include/clang/AST/ASTContext.h

namespace clang {
 
   struct TypeInfo {
     uint64_t Width;
     unsigned Align;
     bool AlignIsRequired : 1;
     TypeInfo() : Width(0), Align(0), AlignIsRequired(false) {}
     TypeInfo(uint64_t Width, unsigned Align, bool AlignIsRequired)
         : Width(Width), Align(Align), AlignIsRequired(AlignIsRequired) {}
   };
 
 /// \brief Holds long-lived AST nodes (such as types and decls) that can be
 /// referred to throughout the semantic analysis of a file.
 class ASTContext : public RefCountedBase<ASTContext> {
   ASTContext &this_() { return *this; }
 
   mutable SmallVector<Type *, 0> Types;
   mutable llvm::FoldingSet<ExtQuals> ExtQualNodes;
public:
   /// \brief Return the APFloat 'semantics' for the specified scalar floating
   /// point type.
   const llvm::fltSemantics &getFloatTypeSemantics(QualType T) const;
 
   /// \brief Get the size and alignment of the specified complete type in bits.
   TypeInfo getTypeInfo(const Type *T) const;
   TypeInfo getTypeInfo(QualType T) const { return getTypeInfo(T.getTypePtr()); }
 
+  /// \brief Get default simd alignment of the specified complete type in bits.
+  unsigned getOpenMPDefaultSimdAlign(QualType T) const;
+
   /// \brief Return the size of the specified (complete) type \p T, in bits.
   uint64_t getTypeSize(QualType T) const { return getTypeInfo(T).Width; }
   uint64_t getTypeSize(const Type *T) const { return getTypeInfo(T).Width; }
 
   /// \brief Return the size of the character type, in bits.
   uint64_t getCharWidth() const {
     return getTypeSize(CharTy);
   }

include/clang/Basic/DiagnosticSemaKinds.td

def err_func_def_incomplete_result : Error<
   "incomplete result type %0 in function definition">;
 def err_atomic_specifier_bad_type : Error<
   "_Atomic cannot be applied to "
   "%select{incomplete |array |function |reference |atomic |qualified |}0type "
   "%1 %select{||||||which is not trivially copyable}0">;
 
 // Expressions.
 def ext_sizeof_alignof_function_type : Extension<
-  "invalid application of '%select{sizeof|alignof|vec_step}0' to a "
+  "invalid application of '%select{sizeof|alignof|vec_step|__builtin_omp_required_simd_align} to a "
   "function type">, InGroup<PointerArith>;
 def ext_sizeof_alignof_void_type : Extension<
-  "invalid application of '%select{sizeof|alignof|vec_step}0' to a void "
+  "invalid application of '%select{sizeof|alignof|vec_step|__builtin_omp_required_simd_align}0' to a void "
   "type">, InGroup<PointerArith>;
 def err_opencl_sizeof_alignof_type : Error<
-  "invalid application of '%select{sizeof|alignof|vec_step}0' to a void type">;
+  "invalid application of '%select{sizeof|alignof|vec_step|__builtin_omp_required_simd_align}0' to a void type">;
 def err_sizeof_alignof_incomplete_type : Error<
-  "invalid application of '%select{sizeof|alignof|vec_step}0' to an "
+  "invalid application of '%select{sizeof|alignof|vec_step|__builtin_omp_required_simd_align}0' to an "
   "incomplete type %1">;
 def err_sizeof_alignof_function_type : Error<
-  "invalid application of '%select{sizeof|alignof|vec_step}0' to a "
+  "invalid application of '%select{sizeof|alignof|vec_step|__builtin_omp_required_simd_align}0' to a "
   "function type">;
+def err_openmp_default_simd_align_expr : Error<
+  "invalid application of '__builtin_omp_required_simd_align' to an expression, only type is allowed">;
 def err_sizeof_alignof_bitfield : Error<
   "invalid application of '%select{sizeof|alignof}0' to bit-field">;
 def err_alignof_member_of_incomplete_type : Error<
   "invalid application of 'alignof' to a field of a class still being defined">;
 def err_vecstep_non_scalar_vector_type : Error<
   "'vec_step' requires built-in scalar or vector type, %0 invalid">;
 def err_offsetof_incomplete_type : Error<
   "offsetof of incomplete type %0">;

include/clang/Basic/TargetInfo.h

protected:
   unsigned char LargeArrayMinWidth, LargeArrayAlign;
   unsigned char LongWidth, LongAlign;
   unsigned char LongLongWidth, LongLongAlign;
   unsigned char SuitableAlign;
   unsigned char DefaultAlignForAttributeAligned;
   unsigned char MinGlobalAlign;
   unsigned char MaxAtomicPromoteWidth, MaxAtomicInlineWidth;
   unsigned short MaxVectorAlign;
+  unsigned short SimdDefaultAlign;
   const char *DescriptionString;
   const char *UserLabelPrefix;
   const char *MCountName;
   const llvm::fltSemantics *HalfFormat, *FloatFormat, *DoubleFormat,
     *LongDoubleFormat;
   unsigned char RegParmMax, SSERegParmMax;
   TargetCXXABI TheCXXABI;
   const LangAS::Map *AddrSpaceMap;
virtual bool hasBuiltinAtomic(uint64_t AtomicSizeInBits,
     return AtomicSizeInBits <= AlignmentInBits &&
            AtomicSizeInBits <= getMaxAtomicInlineWidth() &&
            (AtomicSizeInBits <= getCharWidth() ||
             llvm::isPowerOf2_64(AtomicSizeInBits / getCharWidth()));
   }
 
   /// \brief Return the maximum vector alignment supported for the given target.
   unsigned getMaxVectorAlign() const { return MaxVectorAlign; }
+  /// \brief Return default simd alignment for the given target.
+  unsigned getSimdDefaultAlign() const { return SimdDefaultAlign; }
 
   /// \brief Return the size of intmax_t and uintmax_t for this target, in bits.
   unsigned getIntMaxTWidth() const {
     return getTypeWidth(IntMaxType);
   }
 
   // Return the size of unwind_word for this target.
   unsigned getUnwindWordWidth() const { return getPointerWidth(0); }

include/clang/Basic/TokenKinds.def

 KEYWORD(__read_write                , KEYOPENCL)
 ALIAS("read_only", __read_only      , KEYOPENCL)
 ALIAS("write_only", __write_only    , KEYOPENCL)
 ALIAS("read_write", __read_write    , KEYOPENCL)
 // OpenCL builtins
 KEYWORD(__builtin_astype            , KEYOPENCL)
 KEYWORD(vec_step                    , KEYOPENCL|KEYALTIVEC)
 
+// OpenMP Type Traits
+KEYWORD(__builtin_omp_required_simd_align, KEYALL)
+
 // Borland Extensions.
 KEYWORD(__pascal                    , KEYALL)
 
 // Altivec Extension.
 KEYWORD(__vector                    , KEYALTIVEC)
 KEYWORD(__pixel                     , KEYALTIVEC)
 KEYWORD(__bool                      , KEYALTIVEC)
 

include/clang/Basic/TypeTraits.h

enum ArrayTypeTrait {
     ATT_ArrayRank,
     ATT_ArrayExtent
   };
 
   /// \brief Names for the "expression or type" traits.
   enum UnaryExprOrTypeTrait {
     UETT_SizeOf,
     UETT_AlignOf,
-    UETT_VecStep
+    UETT_VecStep,
+    UETT_OpenMPDefaultSimdAlign,
   };
 }
 
 #endif

lib/AST/ASTContext.cpp

#include "clang/AST/TypeNodes.def"
   }
 
   }
 
   assert(llvm::isPowerOf2_32(Align) && "Alignment must be power of 2");
   return TypeInfo(Width, Align, AlignIsRequired);
 }
 
+unsigned ASTContext::getOpenMPDefaultSimdAlign(QualType T) const {
+  unsigned SimdAlign = getTargetInfo().getSimdDefaultAlign();
+  // Target ppc64 with QPX: simd default alignment for pointer to double is 32.
+  if ((getTargetInfo().getTriple().getArch() == llvm::Triple::ppc64 ||
+       getTargetInfo().getTriple().getArch() == llvm::Triple::ppc64le) &&
+      getTargetInfo().getABI() == "elfv1-qpx" && T->isAnyPointerType() &&
+      T->getPointeeType()->isSpecificBuiltinType(BuiltinType::Double))
+    SimdAlign = 256;
+  return SimdAlign;
+}
+
 /// toCharUnitsFromBits - Convert a size in bits to a size in characters.
 CharUnits ASTContext::toCharUnitsFromBits(int64_t BitSize) const {
   return CharUnits::fromQuantity(BitSize / getCharWidth());
 }
 
 /// toBits - Convert a size in characters to a size in characters.
 int64_t ASTContext::toBits(CharUnits CharSize) const {
   return CharSize.getQuantity() * getCharWidth();

lib/AST/ASTDumper.cpp

case UETT_SizeOf:
     OS << " sizeof";
     break;
   case UETT_AlignOf:
     OS << " alignof";
     break;
   case UETT_VecStep:
     OS << " vec_step";
     break;
+  case UETT_OpenMPDefaultSimdAlign:
+    OS << " __builtin_omp_required_simd_align";
+    break;
   }
   if (Node->isArgumentType())
     dumpType(Node->getArgumentType());
 }
 
 void ASTDumper::VisitMemberExpr(const MemberExpr *Node) {
   VisitExpr(Node);
   OS << " " << (Node->isArrow() ? "->" : ".") << *Node->getMemberDecl();

lib/AST/ExprConstant.cpp

case UETT_SizeOf: {
     if (const ReferenceType *Ref = SrcTy->getAs<ReferenceType>())
       SrcTy = Ref->getPointeeType();
 
     CharUnits Sizeof;
     if (!HandleSizeof(Info, E->getExprLoc(), SrcTy, Sizeof))
       return false;
     return Success(Sizeof, E);
   }
+  case UETT_OpenMPDefaultSimdAlign:
+    return Success(
+        Info.Ctx.toCharUnitsFromBits(Info.Ctx.getOpenMPDefaultSimdAlign(
+                                         E->isArgumentType()
+                                             ? E->getArgumentType()
+                                             : E->getArgumentExpr()->getType()))
+            .getQuantity(),
+        E);
   }
 
   llvm_unreachable("unknown expr/type trait");
 }
 
 bool IntExprEvaluator::VisitOffsetOfExpr(const OffsetOfExpr *OOE) {
   CharUnits Result;
   unsigned n = OOE->getNumComponents();

lib/AST/ItaniumMangle.cpp

case Expr::UnaryExprOrTypeTraitExprClass: {
     
     switch(SAE->getKind()) {
     case UETT_SizeOf:
       Out << 's';
       break;
     case UETT_AlignOf:
       Out << 'a';
       break;
-    case UETT_VecStep:
+    case UETT_VecStep: {
       DiagnosticsEngine &Diags = Context.getDiags();
       unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
                                      "cannot yet mangle vec_step expression");
       Diags.Report(DiagID);
       return;
     }
+    case UETT_OpenMPDefaultSimdAlign:
+      DiagnosticsEngine &Diags = Context.getDiags();
+      unsigned DiagID = Diags.getCustomDiagID(
+          DiagnosticsEngine::Error,
+          "cannot yet mangle __builtin_omp_required_simd_align expression");
+      Diags.Report(DiagID);
+      return;
+    }
     if (SAE->isArgumentType()) {
       Out << 't';
       mangleType(SAE->getArgumentType());
     } else {
       Out << 'z';
       mangleExpression(SAE->getArgumentExpr());
     }
     break;

lib/AST/StmtPrinter.cpp

case UETT_AlignOf:
     else if (Policy.LangOpts.C11)
       OS << "_Alignof";
     else
       OS << "__alignof";
     break;
   case UETT_VecStep:
     OS << "vec_step";
     break;
+  case UETT_OpenMPDefaultSimdAlign:
+    OS << "__builtin_omp_required_simd_align";
+    break;
   }
   if (Node->isArgumentType()) {
     OS << '(';
     Node->getArgumentType().print(OS, Policy);
     OS << ')';
   } else {
     OS << " ";
     PrintExpr(Node->getArgumentExpr());

lib/Basic/TargetInfo.cpp

TargetInfo::TargetInfo(const llvm::Triple &T) : TargetOpts(), Triple(T) {
   DoubleWidth = 64;
   DoubleAlign = 64;
   LongDoubleWidth = 64;
   LongDoubleAlign = 64;
   LargeArrayMinWidth = 0;
   LargeArrayAlign = 0;
   MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 0;
   MaxVectorAlign = 0;
+  SimdDefaultAlign = 0;
   SizeType = UnsignedLong;
   PtrDiffType = SignedLong;
   IntMaxType = SignedLongLong;
   IntPtrType = SignedLong;
   WCharType = SignedInt;
   WIntType = SignedInt;
   Char16Type = UnsignedShort;
   Char32Type = UnsignedInt;

lib/Basic/Targets.cpp

protected:
   std::string ABI;
 
 public:
   PPCTargetInfo(const llvm::Triple &Triple)
     : TargetInfo(Triple), HasVSX(false), HasP8Vector(false),
       HasP8Crypto(false), HasDirectMove(false), HasQPX(false), HasHTM(false),
       HasBPERMD(false), HasExtDiv(false) {
     BigEndian = (Triple.getArch() != llvm::Triple::ppc64le);
+    SimdDefaultAlign = 128;
     LongDoubleWidth = LongDoubleAlign = 128;
     LongDoubleFormat = &llvm::APFloat::PPCDoubleDouble;
   }
 
   /// \brief Flags for architecture specific defines.
   typedef enum {
     ArchDefineNone  = 0,
     ArchDefineName  = 1 << 0, // <name> is substituted for arch name.
bool X86TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
   // Additionally, if SSE is enabled and mmx is not explicitly disabled,
   // then enable MMX.
   std::vector<std::string>::iterator it;
   it = std::find(Features.begin(), Features.end(), "-mmx");
   if (it != Features.end())
     Features.erase(it);
   else if (SSELevel > NoSSE)
     MMX3DNowLevel = std::max(MMX3DNowLevel, MMX);
+
+  SimdDefaultAlign =
+      (getABI() == "avx512") ? 512 : (getABI() == "avx") ? 256 : 128;
   return true;
 }
 
 /// X86TargetInfo::getTargetDefines - Return the set of the X86-specific macro
 /// definitions for this particular subtarget.
 void X86TargetInfo::getTargetDefines(const LangOptions &Opts,
                                      MacroBuilder &Builder) const {
   // Target identification.

lib/CodeGen/CGExprScalar.cpp

 //
 //===----------------------------------------------------------------------===//
 
 #include "CodeGenFunction.h"
 #include "CGCXXABI.h"
 #include "CGDebugInfo.h"
 #include "CGObjCRuntime.h"
 #include "CodeGenModule.h"
+#include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/DeclObjC.h"
 #include "clang/AST/RecordLayout.h"
 #include "clang/AST/StmtVisitor.h"
 #include "clang/Basic/TargetInfo.h"
 #include "clang/Frontend/CodeGenOptions.h"
 #include "llvm/IR/CFG.h"
 #include "llvm/IR/Constants.h"
if (const VariableArrayType *VAT =
 
       // Scale the number of non-VLA elements by the non-VLA element size.
       CharUnits eltSize = CGF.getContext().getTypeSizeInChars(eltType);
       if (!eltSize.isOne())
         size = CGF.Builder.CreateNUWMul(CGF.CGM.getSize(eltSize), numElts);
 
       return size;
     }
+  } else if (E->getKind() == UETT_OpenMPDefaultSimdAlign) {
+    auto Alignment =
+        CGF.getContext()
+            .toCharUnitsFromBits(CGF.getContext().getOpenMPDefaultSimdAlign(
+                E->getTypeOfArgument()))
+            .getQuantity();
+    return llvm::ConstantInt::get(CGF.SizeTy, Alignment);
   }
 
   // If this isn't sizeof(vla), the result must be constant; use the constant
   // folding logic so we don't have to duplicate it here.
   return Builder.getInt(E->EvaluateKnownConstInt(CGF.getContext()));
 }
 
 Value *ScalarExprEmitter::VisitUnaryReal(const UnaryOperator *E) {

lib/CodeGen/CGStmtOpenMP.cpp

for (auto &&I = D.getClausesOfKind(OMPC_aligned); I; ++I) {
     }
     for (auto E : Clause->varlists()) {
       unsigned Alignment = ClauseAlignment;
       if (Alignment == 0) {
         // OpenMP [2.8.1, Description]
         // If no optional parameter is specified, implementation-defined default
         // alignments for SIMD instructions on the target platforms are assumed.
         Alignment =
-            CGF.CGM.getTargetCodeGenInfo().getOpenMPSimdDefaultAlignment(
-                E->getType());
+            CGF.getContext()
+                .toCharUnitsFromBits(
+                    CGF.getContext().getOpenMPDefaultSimdAlign(E->getType()))
+                .getQuantity();
       }
       assert((Alignment == 0 || llvm::isPowerOf2_32(Alignment)) &&
              "alignment is not power of 2");
       if (Alignment != 0) {
         llvm::Value *PtrValue = CGF.EmitScalarExpr(E);
         CGF.EmitAlignmentAssumption(PtrValue, Alignment);
       }
     }

lib/CodeGen/TargetInfo.h

public:
   virtual void getDependentLibraryOption(llvm::StringRef Lib,
                                          llvm::SmallString<24> &Opt) const;
 
   /// Gets the linker options necessary to detect object file mismatches on
   /// this platform.
   virtual void getDetectMismatchOption(llvm::StringRef Name,
                                        llvm::StringRef Value,
                                        llvm::SmallString<32> &Opt) const {}
-
-  /// Gets the target-specific default alignment used when an 'aligned' clause
-  /// is used with a 'simd' OpenMP directive without specifying a specific
-  /// alignment.
-  virtual unsigned getOpenMPSimdDefaultAlignment(QualType Type) const {
-    return 0;
-  }
 };
 }
 
 #endif

lib/CodeGen/TargetInfo.cpp

public:
   bool isHomogeneousAggregateSmallEnough(const Type *Ty,
                                          uint64_t NumMembers) const override {
     // FIXME: Assumes vectorcall is in use.
     return isX86VectorCallAggregateSmallEnough(NumMembers);
   }
 };
 
 class X86_64TargetCodeGenInfo : public TargetCodeGenInfo {
-  X86AVXABILevel AVXLevel;
 public:
   X86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, X86AVXABILevel AVXLevel)
-      : TargetCodeGenInfo(new X86_64ABIInfo(CGT, AVXLevel)),
-        AVXLevel(AVXLevel) {}
+      : TargetCodeGenInfo(new X86_64ABIInfo(CGT, AVXLevel)) {}
 
   const X86_64ABIInfo &getABIInfo() const {
     return static_cast<const X86_64ABIInfo&>(TargetCodeGenInfo::getABIInfo());
   }
 
   int getDwarfEHStackPointer(CodeGen::CodeGenModule &CGM) const override {
     return 7;
   }
if (getABIInfo().has64BitPointers())
             ('T' << 24);
     else
       Sig = (0xeb << 0) |  // jmp rel8
             (0x06 << 8) |  //           .+0x08
             ('F' << 16) |
             ('T' << 24);
     return llvm::ConstantInt::get(CGM.Int32Ty, Sig);
   }
-
-  unsigned getOpenMPSimdDefaultAlignment(QualType) const override {
-    return getNativeVectorSizeForAVXABI(AVXLevel) / 8;
-  }
 };
 
 class PS4TargetCodeGenInfo : public X86_64TargetCodeGenInfo {
 public:
   PS4TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, X86AVXABILevel AVXLevel)
     : X86_64TargetCodeGenInfo(CGT, AVXLevel) {}
 
   void getDependentLibraryOption(llvm::StringRef Lib,
void WinX86_32TargetCodeGenInfo::setTargetAttributes(const Decl *D,
                                                      llvm::GlobalValue *GV,
                                             CodeGen::CodeGenModule &CGM) const {
   X86_32TargetCodeGenInfo::setTargetAttributes(D, GV, CGM);
 
   addStackProbeSizeTargetAttribute(D, GV, CGM);
 }
 
 class WinX86_64TargetCodeGenInfo : public TargetCodeGenInfo {
-  X86AVXABILevel AVXLevel;
 public:
   WinX86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT,
                              X86AVXABILevel AVXLevel)
-      : TargetCodeGenInfo(new WinX86_64ABIInfo(CGT)), AVXLevel(AVXLevel) {}
+      : TargetCodeGenInfo(new WinX86_64ABIInfo(CGT)) {}
 
   void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                            CodeGen::CodeGenModule &CGM) const override;
 
   int getDwarfEHStackPointer(CodeGen::CodeGenModule &CGM) const override {
     return 7;
   }
 
void getDependentLibraryOption(llvm::StringRef Lib,
     Opt += qualifyWindowsLibrary(Lib);
   }
 
   void getDetectMismatchOption(llvm::StringRef Name,
                                llvm::StringRef Value,
                                llvm::SmallString<32> &Opt) const override {
     Opt = "/FAILIFMISMATCH:\"" + Name.str() + "=" + Value.str() + "\"";
   }
-
-  unsigned getOpenMPSimdDefaultAlignment(QualType) const override {
-    return getNativeVectorSizeForAVXABI(AVXLevel) / 8;
-  }
 };
 
 void WinX86_64TargetCodeGenInfo::setTargetAttributes(const Decl *D,
                                                      llvm::GlobalValue *GV,
                                             CodeGen::CodeGenModule &CGM) const {
   TargetCodeGenInfo::setTargetAttributes(D, GV, CGM);
 
   addStackProbeSizeTargetAttribute(D, GV, CGM);
public:
 
   int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
     // This is recovered from gcc output.
     return 1; // r1 is the dedicated stack pointer
   }
 
   bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                llvm::Value *Address) const override;
-
-  unsigned getOpenMPSimdDefaultAlignment(QualType) const override {
-    return 16; // Natural alignment for Altivec vectors.
-  }
 };
 
 }
 
 llvm::Value *PPC32_SVR4_ABIInfo::EmitVAArg(llvm::Value *VAListAddr,
                                            QualType Ty,
                                            CodeGenFunction &CGF) const {
   if (const ComplexType *CTy = Ty->getAs<ComplexType>()) {
void computeInfo(CGFunctionInfo &FI) const override {
     }
   }
 
   llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
                          CodeGenFunction &CGF) const override;
 };
 
 class PPC64_SVR4_TargetCodeGenInfo : public TargetCodeGenInfo {
-  bool HasQPX;
 
 public:
   PPC64_SVR4_TargetCodeGenInfo(CodeGenTypes &CGT,
                                PPC64_SVR4_ABIInfo::ABIKind Kind, bool HasQPX)
-    : TargetCodeGenInfo(new PPC64_SVR4_ABIInfo(CGT, Kind, HasQPX)),
-      HasQPX(HasQPX) {}
+      : TargetCodeGenInfo(new PPC64_SVR4_ABIInfo(CGT, Kind, HasQPX)) {}
 
   int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
     // This is recovered from gcc output.
     return 1; // r1 is the dedicated stack pointer
   }
 
   bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                llvm::Value *Address) const override;
-
-  unsigned getOpenMPSimdDefaultAlignment(QualType QT) const override {
-    if (HasQPX)
-      if (const PointerType *PT = QT->getAs<PointerType>())
-        if (PT->getPointeeType()->isSpecificBuiltinType(BuiltinType::Double))
-          return 32; // Natural alignment for QPX doubles.
-
-    return 16; // Natural alignment for Altivec and VSX vectors.
-  }
 };
 
 class PPC64TargetCodeGenInfo : public DefaultTargetCodeGenInfo {
 public:
   PPC64TargetCodeGenInfo(CodeGenTypes &CGT) : DefaultTargetCodeGenInfo(CGT) {}
 
   int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
     // This is recovered from gcc output.
     return 1; // r1 is the dedicated stack pointer
   }
 
   bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                llvm::Value *Address) const override;
-
-  unsigned getOpenMPSimdDefaultAlignment(QualType) const override {
-    return 16; // Natural alignment for Altivec vectors.
-  }
 };
 
 }
 
 // Return true if the ABI requires Ty to be passed sign- or zero-
 // extended to 64 bits.
 bool
 PPC64_SVR4_ABIInfo::isPromotableTypeForABI(QualType Ty) const {

lib/Parse/ParseExpr.cpp

if (!getLangOpts().C11)
       Diag(Tok, diag::ext_c11_alignment) << Tok.getName();
     // fallthrough
   case tok::kw_alignof:    // unary-expression: 'alignof' '(' type-id ')'
   case tok::kw___alignof:  // unary-expression: '__alignof' unary-expression
                            // unary-expression: '__alignof' '(' type-name ')'
   case tok::kw_sizeof:     // unary-expression: 'sizeof' unary-expression
                            // unary-expression: 'sizeof' '(' type-name ')'
   case tok::kw_vec_step:   // unary-expression: OpenCL 'vec_step' expression
+  // unary-expression: OpenMP 'default simd alignment' expression
+  case tok::kw___builtin_omp_required_simd_align:
     return ParseUnaryExprOrTypeTraitExpression();
   case tok::ampamp: {      // unary-expression: '&&' identifier
     SourceLocation AmpAmpLoc = ConsumeToken();
     if (Tok.isNot(tok::identifier))
       return ExprError(Diag(Tok, diag::err_expected) << tok::identifier);
 
     if (getCurScope()->getFnParent() == nullptr)
       return ExprError(Diag(Tok, diag::err_address_of_label_outside_fn));

 ///           vec_step ( type-name )
 /// \endverbatim
 ExprResult
 Parser::ParseExprAfterUnaryExprOrTypeTrait(const Token &OpTok,
                                            bool &isCastExpr,
                                            ParsedType &CastTy,
                                            SourceRange &CastRange) {
 
-  assert(OpTok.isOneOf(tok::kw_typeof,  tok::kw_sizeof,   tok::kw___alignof,
-                       tok::kw_alignof, tok::kw__Alignof, tok::kw_vec_step) &&
+  assert(OpTok.isOneOf(tok::kw_typeof, tok::kw_sizeof, tok::kw___alignof,
+                       tok::kw_alignof, tok::kw__Alignof, tok::kw_vec_step,
+                       tok::kw___builtin_omp_required_simd_align) &&
          "Not a typeof/sizeof/alignof/vec_step expression!");
 
   ExprResult Operand;
 
   // If the operand doesn't start with an '(', it must be an expression.
   if (Tok.isNot(tok::l_paren)) {
     // If construct allows a form without parenthesis, user may forget to put
     // pathenthesis around type name.

 /// [C++11] 'sizeof' '...' '(' identifier ')'
 /// [GNU]   '__alignof' unary-expression
 /// [GNU]   '__alignof' '(' type-name ')'
 /// [C11]   '_Alignof' '(' type-name ')'
 /// [C++11] 'alignof' '(' type-id ')'
 /// \endverbatim
 ExprResult Parser::ParseUnaryExprOrTypeTraitExpression() {
   assert(Tok.isOneOf(tok::kw_sizeof, tok::kw___alignof, tok::kw_alignof,
-                     tok::kw__Alignof, tok::kw_vec_step) &&
+                     tok::kw__Alignof, tok::kw_vec_step,
+                     tok::kw___builtin_omp_required_simd_align) &&
          "Not a sizeof/alignof/vec_step expression!");
   Token OpTok = Tok;
   ConsumeToken();
 
   // [C++11] 'sizeof' '...' '(' identifier ')'
   if (Tok.is(tok::ellipsis) && OpTok.is(tok::kw_sizeof)) {
     SourceLocation EllipsisLoc = ConsumeToken();
     SourceLocation LParenLoc, RParenLoc;
ExprResult Operand = ParseExprAfterUnaryExprOrTypeTrait(OpTok,
                                                           CastTy,
                                                           CastRange);
 
   UnaryExprOrTypeTrait ExprKind = UETT_SizeOf;
   if (OpTok.isOneOf(tok::kw_alignof, tok::kw___alignof, tok::kw__Alignof))
     ExprKind = UETT_AlignOf;
   else if (OpTok.is(tok::kw_vec_step))
     ExprKind = UETT_VecStep;
+  else if (OpTok.is(tok::kw___builtin_omp_required_simd_align))
+    ExprKind = UETT_OpenMPDefaultSimdAlign;
 
   if (isCastExpr)
     return Actions.ActOnUnaryExprOrTypeTraitExpr(OpTok.getLocation(),
                                                  ExprKind,
                                                  /*isType=*/true,
                                                  CastTy.getAsOpaquePtr(),
                                                  CastRange);
 

lib/Sema/SemaExpr.cpp

static bool CheckExtensionTraitOperandType(Sema &S, QualType T,
                                            UnaryExprOrTypeTrait TraitKind) {
   // Invalid types must be hard errors for SFINAE in C++.
   if (S.LangOpts.CPlusPlus)
     return true;
 
   // C99 6.5.3.4p1:
   if (T->isFunctionType() &&
       (TraitKind == UETT_SizeOf || TraitKind == UETT_AlignOf)) {
     // sizeof(function)/alignof(function) is allowed as an extension.
     S.Diag(Loc, diag::ext_sizeof_alignof_function_type)
       << TraitKind << ArgRange;
     return false;
   }
 
   // Allow sizeof(void)/alignof(void) as an extension, unless in OpenCL where
   // this is an error (OpenCL v1.1 s6.3.k)
   if (T->isVoidType()) {
bool Sema::CheckUnaryExprOrTypeTraitOperand(QualType ExprType,
   //     When alignof is applied to a reference type, the result
   //     shall be the alignment of the referenced type.
   if (const ReferenceType *Ref = ExprType->getAs<ReferenceType>())
     ExprType = Ref->getPointeeType();
 
   // C11 6.5.3.4/3, C++11 [expr.alignof]p3:
   //   When alignof or _Alignof is applied to an array type, the result
   //   is the alignment of the element type.
-  if (ExprKind == UETT_AlignOf)
+  if (ExprKind == UETT_AlignOf || ExprKind == UETT_OpenMPDefaultSimdAlign)
     ExprType = Context.getBaseElementType(ExprType);
 
   if (ExprKind == UETT_VecStep)
     return CheckVecStepTraitOperandType(*this, ExprType, OpLoc, ExprRange);
 
   // Whitelist some types as extensions
   if (!CheckExtensionTraitOperandType(*this, ExprType, OpLoc, ExprRange,
                                       ExprKind))
Sema::CreateUnaryExprOrTypeTraitExpr(Expr *E, SourceLocation OpLoc,
   // Verify that the operand is valid.
   bool isInvalid = false;
   if (E->isTypeDependent()) {
     // Delay type-checking for type-dependent expressions.
   } else if (ExprKind == UETT_AlignOf) {
     isInvalid = CheckAlignOfExpr(*this, E);
   } else if (ExprKind == UETT_VecStep) {
     isInvalid = CheckVecStepExpr(E);
+  } else if (ExprKind == UETT_OpenMPDefaultSimdAlign) {
+      Diag(E->getExprLoc(), diag::err_openmp_default_simd_align_expr);
+      isInvalid = true;
   } else if (E->refersToBitField()) {  // C99 6.5.3.4p1.
     Diag(E->getExprLoc(), diag::err_sizeof_alignof_bitfield) << 0;
     isInvalid = true;
   } else {
     isInvalid = CheckUnaryExprOrTypeTraitOperand(E, UETT_SizeOf);
   }
 
   if (isInvalid)

test/CodeGen/openmp_default_simd_align.c

+// RUN: %clang_cc1 -triple x86_64-unknown-unknown -O1 -emit-llvm -o - %s | FileCheck %s
+
+enum e0 { E0 };
+struct s0 {
+  enum e0         a:31;
+};
+
+int f0() {
+  return __builtin_omp_required_simd_align(struct s0);
+  // CHECK: ret i32 16
+}

test/SemaCXX/openmp_default_simd_align.cpp

+// RUN: %clang_cc1 -std=c++11 -fsyntax-only -triple x86_64-unknown-unknown -verify %s
+
+struct S0 {
+  int x;
+  static const int test0 = __builtin_omp_required_simd_align(x); // expected-error {{invalid application of '__builtin_omp_required_simd_align' to an expression, only type is allowed}}
+  static const int test1 = __builtin_omp_required_simd_align(decltype(S0::x));
+  auto test2() -> char(&)[__builtin_omp_required_simd_align(decltype(x))];
+};
+
+struct S1; // expected-note 6 {{forward declaration}}
+extern S1 s1;
+const int test3 = __builtin_omp_required_simd_align(decltype(s1)); // expected-error {{invalid application of '__builtin_omp_required_simd_align' to an incomplete type 'decltype(s1)' (aka 'S1')}}
+
+struct S2 {
+  S2();
+  S1 &s;
+  int x;
+
+  int test4 = __builtin_omp_required_simd_align(decltype(x)); // ok
+  int test5 = __builtin_omp_required_simd_align(decltype(s)); // expected-error {{invalid application of '__builtin_omp_required_simd_align' to an incomplete type 'S1'}}
+};
+
+const int test6 = __builtin_omp_required_simd_align(decltype(S2::x));
+const int test7 = __builtin_omp_required_simd_align(decltype(S2::s)); // expected-error {{invalid application of '__builtin_omp_required_simd_align' to an incomplete type 'S1'}}
+
+// Arguably, these should fail like the S1 cases do: the alignment of
+// 's2.x' should depend on the alignment of both x-within-S2 and
+// s2-within-S3 and thus require 'S3' to be complete.  If we start
+// doing the appropriate recursive walk to do that, we should make
+// sure that these cases don't explode.
+struct S3 {
+  S2 s2;
+
+  static const int test8 = __builtin_omp_required_simd_align(decltype(s2.x));
+  static const int test9 = __builtin_omp_required_simd_align(decltype(s2.s)); // expected-error {{invalid application of '__builtin_omp_required_simd_align' to an incomplete type 'S1'}}
+  auto test10() -> char(&)[__builtin_omp_required_simd_align(decltype(s2.x))];
+  static const int test11 = __builtin_omp_required_simd_align(decltype(S3::s2.x));
+  static const int test12 = __builtin_omp_required_simd_align(decltype(S3::s2.s)); // expected-error {{invalid application of '__builtin_omp_required_simd_align' to an incomplete type 'S1'}}
+  auto test13() -> char(&)[__builtin_omp_required_simd_align(decltype(s2.x))];
+};
+
+// Same reasoning as S3.
+struct S4 {
+  union {
+      int x;
+    };
+  static const int test0 = __builtin_omp_required_simd_align(decltype(x));
+  static const int test1 = __builtin_omp_required_simd_align(decltype(S0::x));
+  auto test2() -> char(&)[__builtin_omp_required_simd_align(decltype(x))];
+};
+
+// Regression test for asking for the alignment of a field within an invalid
+// record.
+struct S5 {
+  S1 s;  // expected-error {{incomplete type}}
+  int x;
+};
+const int test8 = __builtin_omp_required_simd_align(decltype(S5::x));
+
+long long int test14[2];
+
+static_assert(__builtin_omp_required_simd_align(decltype(test14)) == 16, "foo");
+
+static_assert(__builtin_omp_required_simd_align(int[2]) == __builtin_omp_required_simd_align(int), ""); // ok
+
+namespace __builtin_omp_required_simd_align_array_expr {
+  alignas(32) extern int n[2];
+  static_assert(__builtin_omp_required_simd_align(decltype(n)) == 16, "");
+
+  template<int> struct S {
+      static int a[];
+    };
+  template<int N> int S<N>::a[N];
+  static_assert(__builtin_omp_required_simd_align(decltype(S<1>::a)) == __builtin_omp_required_simd_align(int), "");
+  static_assert(__builtin_omp_required_simd_align(decltype(S<1128>::a)) == __builtin_omp_required_simd_align(int), "");
+}
+
+template <typename T> void n(T) {
+  alignas(T) int T1;
+  char k[__builtin_omp_required_simd_align(decltype(T1))];
+  static_assert(sizeof(k) == __builtin_omp_required_simd_align(long long), "");
+}
+template void n(long long);