[AArch64] Add support for __builtin_ms_va_list on aarch64

[mstorsjo]

Summary

Move builtins from the x86 specific scope into the global scope. Their use is still limited to x86_64 and aarch64 though.

This allows wine on aarch64 to properly handle variadic functions.

This depends on D34474 for llvm.

[mstorsjo]

Did a minor adjustment, to make sure a plain va_arg produces the right kind of output, when targeting windows directly (while the rest of the patch mostly is about supporting the MS ABI while targeting a different OS).

[mstorsjo]

Split the patch into two; one for generic handling of va_arg for win/arm64, and one for using the same functionality while targeting other OSes (i.e. for wine) via the __builtin_ms_* intrinsics.

[mstorsjo]

Added test cases, a minor tweak to semantic error handling, removed the RFC tag, removed the reference to the clang prerequisite that now is committed.

[rnk]

include/clang/Basic/BuiltinsAArch64.def

65

I strongly suspect that Microsoft will never adopt a varargs calling convention that uses a complex, non-char* va_list.

I'm starting to think we should move this to the generic builtin list and make it available everywhere. The semantics are that you can only use __builtin_ms_va_start in ms_abi functions. It always produces a char*-style va_list.

include/clang/Basic/Specifiers.h

239

I think we might prefer to make this non-x86_64 specific. I suspect that this pattern will arise again on ARM32 if anyone goes back there in seriousness. We'll probably want sysv_abi as well as ms_abi, and all the logic should be independent of the ISA: ms_abi is a no-op when the target OS is already Windows, and sysv_abi is a no-op when the target OS isn't Windows.

lib/CodeGen/CGBuiltin.cpp

5276

If we made __builtin_ms_va_start generic, that would also eliminate this duplicate code.

lib/CodeGen/CGCall.cpp

223–224

Unifying the ms_abi CCs would remove the need for this.

lib/Sema/SemaChecking.cpp

3625

Oh dear, how can we keep this simple. I think unifying the CC's would improve things.

lib/Sema/SemaDeclAttr.cpp

4283

Ditto

[mstorsjo]

include/clang/Basic/BuiltinsAArch64.def

65

Yes, that seems probable.

I'm a little weary about making this available anywhere though, since it is coupled with the x86_64/aarch64 specific calling convention for lowering va_start, and there we only support it specifically on those two arches.

include/clang/Basic/Specifiers.h

239

FWIW, we already support ARM32 for windows, and there, varargs are identical to other platforms.

Extending this to all platforms probably also makes sense (although I'm a little weary about how it would work for signaling down to the LLVM IR).

[rnk]

include/clang/Basic/BuiltinsAArch64.def

65

We should have Sema checking that rejects it when not targeting x64 or aarch64. Ultimately the codegen is shared and simple: just call llvm.va.start.

[mstorsjo]

Initial attempt at unifying the use of __builtin_ms_va_list and calling conventions between x86_64 and aarch64. This goes with the current (un-unified) version of D34474.

I only hooked up attribute((ms_abi)) on aarch64, I didn't hook up sysv_abi yet, because it's not obvious what it should map to, since aarch64 has got two other calling conventions in addition to the win64 one; standard AAPCS (for ELF systems) and the darwin modification of AAPCS. Since that's not directly needed for the wine usecase, I'm omitting that for now.

[mstorsjo]

Updated for the new version of D34474.

[mstorsjo]

Reid, with D34474 approved, does this look like what you had in mind?

[rnk]

Thanks, looks great!

include/clang-c/Index.h

enum CXCallingConv {
   CXCallingConv_X86RegCall = 8,
   CXCallingConv_IntelOclBicc = 9,
   CXCallingConv_X86_64Win64 = 10,
   CXCallingConv_X86_64SysV = 11,
   CXCallingConv_X86VectorCall = 12,
   CXCallingConv_Swift = 13,
   CXCallingConv_PreserveMost = 14,
   CXCallingConv_PreserveAll = 15,
+  CXCallingConv_AArch64Win64 = 16,
 
   CXCallingConv_Invalid = 100,
   CXCallingConv_Unexposed = 200
 };
 
 /**
  * \brief The type of an element in the abstract syntax tree.
  *

include/clang/Basic/BuiltinsAArch64.def

 // System Registers
 BUILTIN(__builtin_arm_rsr, "UicC*", "nc")
 BUILTIN(__builtin_arm_rsr64, "LUicC*", "nc")
 BUILTIN(__builtin_arm_rsrp, "v*cC*", "nc")
 BUILTIN(__builtin_arm_wsr, "vcC*Ui", "nc")
 BUILTIN(__builtin_arm_wsr64, "vcC*LUi", "nc")
 BUILTIN(__builtin_arm_wsrp, "vcC*vC*", "nc")
 
+// Win64-compatible va_list functions
+BUILTIN(__builtin_ms_va_start, "vc*&.", "nt")
+BUILTIN(__builtin_ms_va_end, "vc*&", "n")
+BUILTIN(__builtin_ms_va_copy, "vc*&c*&", "n")
+
 #undef BUILTIN

include/clang/Basic/DiagnosticSemaKinds.td

 def err_va_start_outside_function : Error<
   "'va_start' cannot be used outside a function">;
 def err_va_start_fixed_function : Error<
   "'va_start' used in function with fixed args">;
 def err_va_start_used_in_wrong_abi_function : Error<
   "'va_start' used in %select{System V|Win64}0 ABI function">;
 def err_ms_va_start_used_in_sysv_function : Error<
   "'__builtin_ms_va_start' used in System V ABI function">;
+def err_ms_va_start_used_in_aapcs_function : Error<
+  "'__builtin_ms_va_start' used in AAPCS ABI function">;
 def warn_second_arg_of_va_start_not_last_named_param : Warning<
   "second argument to 'va_start' is not the last named parameter">,
   InGroup<Varargs>;
 def warn_va_start_type_is_undefined : Warning<
   "passing %select{an object that undergoes default argument promotion|"
   "an object of reference type|a parameter declared with the 'register' "
   "keyword}0 to 'va_start' has undefined behavior">, InGroup<Varargs>;
 def err_first_argument_to_va_arg_not_of_type_va_list : Error<

include/clang/Basic/Specifiers.h

#include "clang/Basic/OpenCLImageTypes.def"
   /// \brief CallingConv - Specifies the calling convention that a function uses.
   enum CallingConv {
     CC_C,           // __attribute__((cdecl))
     CC_X86StdCall,  // __attribute__((stdcall))
     CC_X86FastCall, // __attribute__((fastcall))
     CC_X86ThisCall, // __attribute__((thiscall))
     CC_X86VectorCall, // __attribute__((vectorcall))
     CC_X86Pascal,   // __attribute__((pascal))
     CC_X86_64Win64, // __attribute__((ms_abi))
     CC_X86_64SysV,  // __attribute__((sysv_abi))
     CC_X86RegCall, // __attribute__((regcall))
     CC_AAPCS,       // __attribute__((pcs("aapcs")))
     CC_AAPCS_VFP,   // __attribute__((pcs("aapcs-vfp")))
     CC_IntelOclBicc, // __attribute__((intel_ocl_bicc))
     CC_SpirFunction, // default for OpenCL functions on SPIR target
     CC_OpenCLKernel, // inferred for OpenCL kernels
     CC_Swift,        // __attribute__((swiftcall))
     CC_PreserveMost, // __attribute__((preserve_most))
     CC_PreserveAll,  // __attribute__((preserve_all))
+    CC_AArch64Win64, // __attribute__((ms_abi))
   };
 
   /// \brief Checks whether the given calling convention supports variadic
   /// calls. Unprototyped calls also use the variadic call rules.
   inline bool supportsVariadicCall(CallingConv CC) {
     switch (CC) {
     case CC_X86StdCall:
     case CC_X86FastCall:

lib/AST/ItaniumMangle.cpp

StringRef CXXNameMangler::getCallingConvQualifierName(CallingConv CC) {
   case CC_X86RegCall:
   case CC_AAPCS:
   case CC_AAPCS_VFP:
   case CC_IntelOclBicc:
   case CC_SpirFunction:
   case CC_OpenCLKernel:
   case CC_PreserveMost:
   case CC_PreserveAll:
+  case CC_AArch64Win64:
     // FIXME: we should be mangling all of the above.
     return "";
 
   case CC_Swift:
     return "swiftcall";
   }
   llvm_unreachable("bad calling convention");
 }

lib/AST/MicrosoftMangle.cpp

void MicrosoftCXXNameMangler::mangleCallingConvention(CallingConv CC) {
   // that they could be in a DLL and somebody from another module could call
   // them.)
 
   switch (CC) {
     default:
       llvm_unreachable("Unsupported CC for mangling");
     case CC_X86_64Win64:
     case CC_X86_64SysV:
+    case CC_AArch64Win64:
     case CC_C: Out << 'A'; break;
     case CC_X86Pascal: Out << 'C'; break;
     case CC_X86ThisCall: Out << 'E'; break;
     case CC_X86StdCall: Out << 'G'; break;
     case CC_X86FastCall: Out << 'I'; break;
     case CC_X86VectorCall: Out << 'Q'; break;
     case CC_X86RegCall: Out << 'w'; break;
   }

lib/AST/Type.cpp

StringRef FunctionType::getNameForCallConv(CallingConv CC) {
   case CC_AAPCS: return "aapcs";
   case CC_AAPCS_VFP: return "aapcs-vfp";
   case CC_IntelOclBicc: return "intel_ocl_bicc";
   case CC_SpirFunction: return "spir_function";
   case CC_OpenCLKernel: return "opencl_kernel";
   case CC_Swift: return "swiftcall";
   case CC_PreserveMost: return "preserve_most";
   case CC_PreserveAll: return "preserve_all";
+  case CC_AArch64Win64: return "ms_abi";
   }
 
   llvm_unreachable("Invalid calling convention.");
 }
 
 FunctionProtoType::FunctionProtoType(QualType result, ArrayRef<QualType> params,
                                      QualType canonical,
                                      const ExtProtoInfo &epi)

lib/AST/TypePrinter.cpp

case CC_Swift:
       OS << " __attribute__((swiftcall))";
       break;
     case CC_PreserveMost:
       OS << " __attribute__((preserve_most))";
       break;
     case CC_PreserveAll:
       OS << " __attribute__((preserve_all))";
       break;
+    case CC_AArch64Win64:
+      OS << " __attribute__((ms_abi))";
+      break;
     }
   }
 
   if (Info.getNoReturn())
     OS << " __attribute__((noreturn))";
   if (Info.getRegParm())
     OS << " __attribute__((regparm ("
        << Info.getRegParm() << ")))";

lib/Basic/Targets.cpp

AArch64TargetInfo(const llvm::Triple &Triple, const TargetOptions &Opts)
     LongWidth = LongAlign = PointerWidth = PointerAlign = 64;
     MaxVectorAlign = 128;
     MaxAtomicInlineWidth = 128;
     MaxAtomicPromoteWidth = 128;
 
     LongDoubleWidth = LongDoubleAlign = SuitableAlign = 128;
     LongDoubleFormat = &llvm::APFloat::IEEEquad();
 
+    // Make __builtin_ms_va_list available.
+    HasBuiltinMSVaList = true;
+
     // {} in inline assembly are neon specifiers, not assembly variant
     // specifiers.
     NoAsmVariants = true;
 
     // AAPCS gives rules for bitfields. 7.1.7 says: "The container type
     // contributes to the alignment of the containing aggregate in the same way
     // a plain (non bit-field) member of that type would, without exception for
     // zero-sized or anonymous bit-fields."
public:
 
   CallingConvCheckResult checkCallingConvention(CallingConv CC) const override {
     switch (CC) {
     case CC_C:
     case CC_Swift:
     case CC_PreserveMost:
     case CC_PreserveAll:
     case CC_OpenCLKernel:
+    case CC_AArch64Win64:
       return CCCR_OK;
     default:
       return CCCR_Warning;
     }
   }
 
   bool isCLZForZeroUndef() const override { return false; }
 

lib/CodeGen/CGBuiltin.cpp

Value *CodeGenFunction::vectorWrapScalar16(Value *Op) {
   Value *V = UndefValue::get(VTy);
   llvm::Constant *CI = ConstantInt::get(SizeTy, 0);
   Op = Builder.CreateInsertElement(V, Op, CI);
   return Op;
 }
 
 Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
                                                const CallExpr *E) {
+  if (BuiltinID == AArch64::BI__builtin_ms_va_start ||
+      BuiltinID == AArch64::BI__builtin_ms_va_end)
+    return EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(),
+                          BuiltinID == AArch64::BI__builtin_ms_va_start);
+  if (BuiltinID == AArch64::BI__builtin_ms_va_copy) {
+    // Lower this manually. We can't reliably determine whether or not any
+    // given va_copy() is for a Win64 va_list from the calling convention
+    // alone, because it's legal to do this from an AAPCS ABI function.
+    // With opaque pointer types, we won't have enough information in LLVM
+    // IR to determine this from the argument types, either. Best to do it
+    // now, while we have enough information.
+    Address DestAddr = EmitMSVAListRef(E->getArg(0));
+    Address SrcAddr = EmitMSVAListRef(E->getArg(1));
+
+    llvm::Type *BPP = Int8PtrPtrTy;
+
+    DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"),
+                       DestAddr.getAlignment());
+    SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"),
+                      SrcAddr.getAlignment());
+
+    Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val");
+    return Builder.CreateStore(ArgPtr, DestAddr);
+  }
+
   unsigned HintID = static_cast<unsigned>(-1);
   switch (BuiltinID) {
   default: break;
   case AArch64::BI__builtin_arm_nop:
     HintID = 0;
     break;
   case AArch64::BI__builtin_arm_yield:
     HintID = 1;

lib/CodeGen/CGCall.cpp

unsigned CodeGenTypes::ClangCallConvToLLVMCallConv(CallingConv CC) {
   case CC_X86Pascal: return llvm::CallingConv::C;
   // TODO: Add support for __vectorcall to LLVM.
   case CC_X86VectorCall: return llvm::CallingConv::X86_VectorCall;
   case CC_SpirFunction: return llvm::CallingConv::SPIR_FUNC;
   case CC_OpenCLKernel: return CGM.getTargetCodeGenInfo().getOpenCLKernelCallingConv();
   case CC_PreserveMost: return llvm::CallingConv::PreserveMost;
   case CC_PreserveAll: return llvm::CallingConv::PreserveAll;
   case CC_Swift: return llvm::CallingConv::Swift;
+  case CC_AArch64Win64: return llvm::CallingConv::AArch64_Win64;
   }
 }
 
 /// Derives the 'this' type for codegen purposes, i.e. ignoring method
 /// qualification.
 /// FIXME: address space qualification?
 static CanQualType GetThisType(ASTContext &Context, const CXXRecordDecl *RD) {
   QualType RecTy = Context.getTagDeclType(RD)->getCanonicalTypeInternal();

 const CGFunctionInfo &
 CodeGenTypes::arrangeFreeFunctionType(CanQual<FunctionProtoType> FTP,
                                       const FunctionDecl *FD) {
   SmallVector<CanQualType, 16> argTypes;
   return ::arrangeLLVMFunctionInfo(*this, /*instanceMethod=*/false, argTypes,
                                    FTP, FD);
 }
 
-static CallingConv getCallingConventionForDecl(const Decl *D, bool IsWindows) {
+static CallingConv getCallingConventionForDecl(const Decl *D, bool IsWindows,
+                                               llvm::Triple::ArchType Arch) {
   // Set the appropriate calling convention for the Function.
   if (D->hasAttr<StdCallAttr>())
     return CC_X86StdCall;
 
   if (D->hasAttr<FastCallAttr>())
     return CC_X86FastCall;
 
   if (D->hasAttr<RegCallAttr>())
static CallingConv getCallingConventionForDecl(const Decl *D, bool IsWindows,
 
   if (PcsAttr *PCS = D->getAttr<PcsAttr>())
     return (PCS->getPCS() == PcsAttr::AAPCS ? CC_AAPCS : CC_AAPCS_VFP);
 
   if (D->hasAttr<IntelOclBiccAttr>())
     return CC_IntelOclBicc;
 
   if (D->hasAttr<MSABIAttr>())
-    return IsWindows ? CC_C : CC_X86_64Win64;
+    return IsWindows ? CC_C : Arch == llvm::Triple::aarch64 ? CC_AArch64Win64
+                                                            : CC_X86_64Win64;
 
   if (D->hasAttr<SysVABIAttr>())
     return IsWindows ? CC_X86_64SysV : CC_C;
 
   if (D->hasAttr<PreserveMostAttr>())
     return CC_PreserveMost;
 
   if (D->hasAttr<PreserveAllAttr>())
CodeGenTypes::arrangeObjCMessageSendSignature(const ObjCMethodDecl *MD,
   argTys.push_back(Context.getCanonicalParamType(receiverType));
   argTys.push_back(Context.getCanonicalParamType(Context.getObjCSelType()));
   // FIXME: Kill copy?
   for (const auto *I : MD->parameters()) {
     argTys.push_back(Context.getCanonicalParamType(I->getType()));
   }
 
   FunctionType::ExtInfo einfo;
-  bool IsWindows = getContext().getTargetInfo().getTriple().isOSWindows();
-  einfo = einfo.withCallingConv(getCallingConventionForDecl(MD, IsWindows));
+  const llvm::Triple &TT = getContext().getTargetInfo().getTriple();
+  bool IsWindows = TT.isOSWindows();
+  llvm::Triple::ArchType Arch = TT.getArch();
+  einfo =
+      einfo.withCallingConv(getCallingConventionForDecl(MD, IsWindows, Arch));
 
   if (getContext().getLangOpts().ObjCAutoRefCount &&
       MD->hasAttr<NSReturnsRetainedAttr>())
     einfo = einfo.withProducesResult(true);
 
   RequiredArgs required =
     (MD->isVariadic() ? RequiredArgs(argTys.size()) : RequiredArgs::All);
 

lib/CodeGen/CGDebugInfo.cpp

static unsigned getDwarfCC(CallingConv CC) {
   case CC_AAPCS_VFP:
   case CC_IntelOclBicc:
   case CC_SpirFunction:
   case CC_OpenCLKernel:
   case CC_Swift:
   case CC_PreserveMost:
   case CC_PreserveAll:
   case CC_X86RegCall:
+  case CC_AArch64Win64:
     return 0;
   }
   return 0;
 }
 
 llvm::DIType *CGDebugInfo::CreateType(const FunctionType *Ty,
                                       llvm::DIFile *Unit) {
   SmallVector<llvm::Metadata *, 16> EltTys;

lib/Sema/SemaChecking.cpp

bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   }
 
   // FIXME: VFP Intrinsics should error if VFP not present.
   return SemaBuiltinConstantArgRange(TheCall, i, l, u + l);
 }
 
 bool Sema::CheckAArch64BuiltinFunctionCall(unsigned BuiltinID,
                                          CallExpr *TheCall) {
+  if (BuiltinID == AArch64::BI__builtin_ms_va_start)
+    return SemaBuiltinVAStart(BuiltinID, TheCall);
+
   if (BuiltinID == AArch64::BI__builtin_arm_ldrex ||
       BuiltinID == AArch64::BI__builtin_arm_ldaex ||
       BuiltinID == AArch64::BI__builtin_arm_strex ||
       BuiltinID == AArch64::BI__builtin_arm_stlex) {
     return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall, 128);
   }
 
   if (BuiltinID == AArch64::BI__builtin_arm_prefetch) {
ExprResult Sema::CheckOSLogFormatStringArg(Expr *Arg) {
   InitializedEntity Entity =
       InitializedEntity::InitializeParameter(Context, ResultTy, false);
   Result = PerformCopyInitialization(Entity, SourceLocation(), Result);
   return Result;
 }
 
 /// Check that the user is calling the appropriate va_start builtin for the
 /// target and calling convention.
 static bool checkVAStartABI(Sema &S, unsigned BuiltinID, Expr *Fn) {
   const llvm::Triple &TT = S.Context.getTargetInfo().getTriple();
   bool IsX64 = TT.getArch() == llvm::Triple::x86_64;
+  bool IsAArch64 = TT.getArch() == llvm::Triple::aarch64;
   bool IsWindows = TT.isOSWindows();
-  bool IsMSVAStart = BuiltinID == X86::BI__builtin_ms_va_start;
-  if (IsX64) {
-    clang::CallingConv CC = CC_C;
-    if (const FunctionDecl *FD = S.getCurFunctionDecl())
-      CC = FD->getType()->getAs<FunctionType>()->getCallConv();
+  bool IsMSVAStart = BuiltinID == X86::BI__builtin_ms_va_start ||
+                     BuiltinID == AArch64::BI__builtin_ms_va_start;
+
+  if (IsX64) {
     if (IsMSVAStart) {
       // Don't allow this in System V ABI functions.
       if (CC == CC_X86_64SysV || (!IsWindows && CC != CC_X86_64Win64))
         return S.Diag(Fn->getLocStart(),
                       diag::err_ms_va_start_used_in_sysv_function);
     } else {
       // On x86-64 Unix, don't allow this in Win64 ABI functions.
       // On x64 Windows, don't allow this in System V ABI functions.
       // (Yes, that means there's no corresponding way to support variadic
       // System V ABI functions on Windows.)
       if ((IsWindows && CC == CC_X86_64SysV) ||
           (!IsWindows && CC == CC_X86_64Win64))
         return S.Diag(Fn->getLocStart(),
                       diag::err_va_start_used_in_wrong_abi_function)
                << !IsWindows;
     }
     return false;
   }
 
+  if (IsAArch64) {
+    if (IsMSVAStart) {
+      // Don't allow this in AAPCS functions.
+      if (!IsWindows && CC != CC_AArch64Win64)
+        return S.Diag(Fn->getLocStart(),
+                      diag::err_ms_va_start_used_in_aapcs_function);
+    } else {
+      // On aarch64 Unix, don't allow this in Win64 ABI functions.
+      if (!IsWindows && CC == CC_AArch64Win64)
+        return S.Diag(Fn->getLocStart(),
+                      diag::err_va_start_used_in_wrong_abi_function)
+               << !IsWindows;
+    }
+    return false;
+  }
+
   if (IsMSVAStart)
     return S.Diag(Fn->getLocStart(), diag::err_x86_builtin_64_only);
   return false;
 }
 
 static bool checkVAStartIsInVariadicFunction(Sema &S, Expr *Fn,
                                              ParmVarDecl **LastParam = nullptr) {
   // Determine whether the current function, block, or obj-c method is variadic

lib/Sema/SemaDeclAttr.cpp

bool Sema::CheckCallingConvAttr(const AttributeList &attr, CallingConv &CC,
   case AttributeList::AT_FastCall: CC = CC_X86FastCall; break;
   case AttributeList::AT_StdCall: CC = CC_X86StdCall; break;
   case AttributeList::AT_ThisCall: CC = CC_X86ThisCall; break;
   case AttributeList::AT_Pascal: CC = CC_X86Pascal; break;
   case AttributeList::AT_SwiftCall: CC = CC_Swift; break;
   case AttributeList::AT_VectorCall: CC = CC_X86VectorCall; break;
   case AttributeList::AT_RegCall: CC = CC_X86RegCall; break;
   case AttributeList::AT_MSABI:
-    CC = Context.getTargetInfo().getTriple().isOSWindows() ? CC_C :
-                                                             CC_X86_64Win64;
+
+    CC = Context.getTargetInfo().getTriple().isOSWindows()
+             ? CC_C
+             : Context.getTargetInfo().getTriple().getArch() ==
+                       llvm::Triple::aarch64
+                   ? CC_AArch64Win64
+                   : CC_X86_64Win64;
     break;
   case AttributeList::AT_SysVABI:
     CC = Context.getTargetInfo().getTriple().isOSWindows() ? CC_X86_64SysV :
                                                              CC_C;
     break;
   case AttributeList::AT_Pcs: {
     StringRef StrRef;
     if (!checkStringLiteralArgumentAttr(attr, 0, StrRef)) {

test/CodeGen/ms_abi_aarch64.c

+// RUN: %clang_cc1 -triple aarch64-linux-gnu -emit-llvm < %s | FileCheck -check-prefix=LINUX %s
+// RUN: %clang_cc1 -triple aarch64-pc-win32 -emit-llvm < %s | FileCheck -check-prefix=WIN64 %s
+
+void __attribute__((ms_abi)) f1(void);
+void f2(void);
+void f3(void) {
+  // LINUX-LABEL: define void @f3()
+  // WIN64-LABEL: define void @f3()
+  f1();
+  // LINUX: call aarch64_win64cc void @f1()
+  // WIN64: call void @f1()
+  f2();
+  // LINUX: call void @f2()
+  // WIN64: call void @f2()
+}
+// LINUX: declare aarch64_win64cc void @f1()
+// LINUX: declare void @f2()
+// WIN64: declare void @f1()
+// WIN64: declare void @f2()
+
+// Win64 ABI varargs
+void __attribute__((ms_abi)) f4(int a, ...) {
+  // LINUX-LABEL: define aarch64_win64cc void @f4
+  // WIN64-LABEL: define void @f4
+  __builtin_ms_va_list ap;
+  __builtin_ms_va_start(ap, a);
+  // LINUX: %[[AP:.*]] = alloca i8*
+  // LINUX: call void @llvm.va_start
+  // WIN64: %[[AP:.*]] = alloca i8*
+  // WIN64: call void @llvm.va_start
+  int b = __builtin_va_arg(ap, int);
+  // LINUX: %[[AP_CUR:.*]] = load i8*, i8** %[[AP]]
+  // LINUX-NEXT: %[[AP_NEXT:.*]] = getelementptr inbounds i8, i8* %[[AP_CUR]], i64 8
+  // LINUX-NEXT: store i8* %[[AP_NEXT]], i8** %[[AP]]
+  // LINUX-NEXT: bitcast i8* %[[AP_CUR]] to i32*
+  // WIN64: %[[AP_CUR:.*]] = load i8*, i8** %[[AP]]
+  // WIN64-NEXT: %[[AP_NEXT:.*]] = getelementptr inbounds i8, i8* %[[AP_CUR]], i64 8
+  // WIN64-NEXT: store i8* %[[AP_NEXT]], i8** %[[AP]]
+  // WIN64-NEXT: bitcast i8* %[[AP_CUR]] to i32*
+  __builtin_ms_va_list ap2;
+  __builtin_ms_va_copy(ap2, ap);
+  // LINUX: %[[AP_VAL:.*]] = load i8*, i8** %[[AP]]
+  // LINUX-NEXT: store i8* %[[AP_VAL]], i8** %[[AP2:.*]]
+  // WIN64: %[[AP_VAL:.*]] = load i8*, i8** %[[AP]]
+  // WIN64-NEXT: store i8* %[[AP_VAL]], i8** %[[AP2:.*]]
+  __builtin_ms_va_end(ap);
+  // LINUX: call void @llvm.va_end
+  // WIN64: call void @llvm.va_end
+}
+
+// Let's verify that normal va_lists work right on Win64, too.
+void f5(int a, ...) {
+  // WIN64-LABEL: define void @f5
+  __builtin_va_list ap;
+  __builtin_va_start(ap, a);
+  // WIN64: %[[AP:.*]] = alloca i8*
+  // WIN64: call void @llvm.va_start
+  int b = __builtin_va_arg(ap, int);
+  // WIN64: %[[AP_CUR:.*]] = load i8*, i8** %[[AP]]
+  // WIN64-NEXT: %[[AP_NEXT:.*]] = getelementptr inbounds i8, i8* %[[AP_CUR]], i64 8
+  // WIN64-NEXT: store i8* %[[AP_NEXT]], i8** %[[AP]]
+  // WIN64-NEXT: bitcast i8* %[[AP_CUR]] to i32*
+  __builtin_va_list ap2;
+  __builtin_va_copy(ap2, ap);
+  // WIN64: call void @llvm.va_copy
+  __builtin_va_end(ap);
+  // WIN64: call void @llvm.va_end
+}

test/Sema/varargs-aarch64.c

+// RUN: %clang_cc1 -fsyntax-only -verify %s -triple aarch64-linux-gnu
+
+void f1(int a, ...) {
+  __builtin_ms_va_list ap;
+  __builtin_ms_va_start(ap, a); // expected-error {{'__builtin_ms_va_start' used in AAPCS ABI function}}
+}
+
+void __attribute__((ms_abi)) f2(int a, ...) {
+  __builtin_va_list ap;
+  __builtin_va_start(ap, a); // expected-error {{'va_start' used in Win64 ABI function}}
+}

tools/libclang/CXType.cpp

switch (FD->getCallConv()) {
       TCALLINGCONV(X86_64Win64);
       TCALLINGCONV(X86_64SysV);
       TCALLINGCONV(AAPCS);
       TCALLINGCONV(AAPCS_VFP);
       TCALLINGCONV(IntelOclBicc);
       TCALLINGCONV(Swift);
       TCALLINGCONV(PreserveMost);
       TCALLINGCONV(PreserveAll);
+      TCALLINGCONV(AArch64Win64);
     case CC_SpirFunction: return CXCallingConv_Unexposed;
     case CC_OpenCLKernel: return CXCallingConv_Unexposed;
       break;
     }
 #undef TCALLINGCONV
   }
   
   return CXCallingConv_Invalid;