[COFF, ARM64] Align global symbol by size for ARM64 MSVC ABI

[TomTan]

Summary

According to alignment section in below ARM64 ABI document, MSVC could increase alignment of global data based on its total size. Clang doesn't do this. Compile the same symbol into different alignments by Clang and MSVC could cause link error because some instruction encodings, like 64-bit LDR/STR with immediate, require the target to be 8 bytes aligned, and linker could choose code stream with such LDR/STR instruction from MSVC and 4 bytes aligned data from Clang into final image, which actually cannot be linked together (see https://bugs.llvm.org/show_bug.cgi?id=41506 for more details).

https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=vs-2019#alignment

[efriedma]

I was going to ask if local variables are also supposed to be aligned this way... but I guess there's no way for standard C++ code to tell without explicitly making weird alignment assumptions, so let's not worry about that.

This should do the right thing, as far as I can tell.

It probably would make sense to refactor the arm64-specific code into a TargetInfo API... or at least the triple check. Not sure what that would look like; maybe virtual uint64_t getMinGlobalAlign(uint64_t TypeSize)? Or maybe should just be "bool useMicrosoftGlobalAlign()" and keep the main logic here, if the rule is the same for multiple Windows targets.

Needs a testcase in test/CodeGenCXX/ demonstrating the alignment of the emitted global in various cases.

[riccibruno]

Since this is target-specific, is it right to put this here ?

[TomTan]

Added test cases and also merged this alignment adjustment to getMinGlobalAlign in MicrosoftARM64TargetInfo.

[efriedma]

LGTM

include/clang/Basic/TargetInfo.h

public:
   /// Return the default alignment for __attribute__((aligned)) on
   /// this target, to be used if no alignment value is specified.
   unsigned getDefaultAlignForAttributeAligned() const {
     return DefaultAlignForAttributeAligned;
   }
 
   /// getMinGlobalAlign - Return the minimum alignment of a global variable,
   /// unless its alignment is explicitly reduced via attributes.
-  unsigned getMinGlobalAlign() const { return MinGlobalAlign; }
+  virtual unsigned getMinGlobalAlign (uint64_t) const {
+    return MinGlobalAlign;
+  }
 
   /// Return the largest alignment for which a suitably-sized allocation with
   /// '::operator new(size_t)' is guaranteed to produce a correctly-aligned
   /// pointer.
   unsigned getNewAlign() const {
     return NewAlign ? NewAlign : std::max(LongDoubleAlign, LongLongAlign);
   }
 

lib/AST/ASTContext.cpp

else if (!BaseT->isIncompleteType()) {
                    MinWidth <= getTypeSize(cast<ConstantArrayType>(arrayType)))
             Align = std::max(Align, Target->getLargeArrayAlign());
         }
       }
       Align = std::max(Align, getPreferredTypeAlign(T.getTypePtr()));
       if (BaseT.getQualifiers().hasUnaligned())
         Align = Target->getCharWidth();
       if (const auto *VD = dyn_cast<VarDecl>(D)) {
-        if (VD->hasGlobalStorage() && !ForAlignof)
-          Align = std::max(Align, getTargetInfo().getMinGlobalAlign());
+        if (VD->hasGlobalStorage() && !ForAlignof) {
+          uint64_t TypeSize = getTypeSize(T.getTypePtr());
+          Align = std::max(Align, getTargetInfo().getMinGlobalAlign(TypeSize));
+        }
       }
     }
 
     // Fields can be subject to extra alignment constraints, like if
     // the field is packed, the struct is packed, or the struct has a
     // a max-field-alignment constraint (#pragma pack).  So calculate
     // the actual alignment of the field within the struct, and then
     // (as we're expected to) constrain that by the alignment of the type.

 /// value is specified.
 unsigned ASTContext::getTargetDefaultAlignForAttributeAligned() const {
   return getTargetInfo().getDefaultAlignForAttributeAligned();
 }
 
 /// getAlignOfGlobalVar - Return the alignment in bits that should be given
 /// to a global variable of the specified type.
 unsigned ASTContext::getAlignOfGlobalVar(QualType T) const {
-  return std::max(getTypeAlign(T), getTargetInfo().getMinGlobalAlign());
+  uint64_t TypeSize = getTypeSize(T.getTypePtr());
+  return std::max(getTypeAlign(T), getTargetInfo().getMinGlobalAlign(TypeSize));
 }
 
 /// getAlignOfGlobalVarInChars - Return the alignment in characters that
 /// should be given to a global variable of the specified type.
 CharUnits ASTContext::getAlignOfGlobalVarInChars(QualType T) const {
   return toCharUnitsFromBits(getAlignOfGlobalVar(T));
 }
 

lib/Basic/Targets/AArch64.h

MicrosoftARM64TargetInfo(const llvm::Triple &Triple,
                            const TargetOptions &Opts);
 
   void getVisualStudioDefines(const LangOptions &Opts,
                               MacroBuilder &Builder) const;
   void getTargetDefines(const LangOptions &Opts,
                         MacroBuilder &Builder) const override;
   TargetInfo::CallingConvKind
   getCallingConvKind(bool ClangABICompat4) const override;
+
+  unsigned getMinGlobalAlign(uint64_t TypeSize) const override;
 };
 
 // ARM64 MinGW target
 class LLVM_LIBRARY_VISIBILITY MinGWARM64TargetInfo
     : public WindowsARM64TargetInfo {
 public:
   MinGWARM64TargetInfo(const llvm::Triple &Triple, const TargetOptions &Opts);
 };

lib/Basic/Targets/AArch64.cpp

void MicrosoftARM64TargetInfo::getTargetDefines(const LangOptions &Opts,
   getVisualStudioDefines(Opts, Builder);
 }
 
 TargetInfo::CallingConvKind
 MicrosoftARM64TargetInfo::getCallingConvKind(bool ClangABICompat4) const {
   return CCK_MicrosoftWin64;
 }
 
+unsigned MicrosoftARM64TargetInfo::getMinGlobalAlign(uint64_t TypeSize) const {
+  unsigned Align = WindowsARM64TargetInfo::getMinGlobalAlign(TypeSize);
+
+  // MSVC does size based alignment for arm64 based on alignment section in
+  // below document, replicate that to keep alignment consistent with object
+  // files compiled by MSVC.
+  // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions
+  if (TypeSize >= 512) {              // TypeSize >= 64 bytes
+    Align = std::max(Align, 128u);    // align type at least 16 bytes
+  } else if (TypeSize >= 64) {        // TypeSize >= 8 bytes
+    Align = std::max(Align, 64u);     // align type at least 8 butes
+  } else if (TypeSize >= 16) {        // TypeSize >= 2 bytes
+    Align = std::max(Align, 32u);     // align type at least 4 bytes
+  }
+  return Align;
+}
+
 MinGWARM64TargetInfo::MinGWARM64TargetInfo(const llvm::Triple &Triple,
                                            const TargetOptions &Opts)
     : WindowsARM64TargetInfo(Triple, Opts) {
   TheCXXABI.set(TargetCXXABI::GenericAArch64);
 }
 
 DarwinAArch64TargetInfo::DarwinAArch64TargetInfo(const llvm::Triple &Triple,
                                                  const TargetOptions &Opts)

lib/Basic/Targets/NVPTX.cpp

NVPTXTargetInfo::NVPTXTargetInfo(const llvm::Triple &Triple,
   FloatWidth = HostTarget->getFloatWidth();
   FloatAlign = HostTarget->getFloatAlign();
   DoubleWidth = HostTarget->getDoubleWidth();
   DoubleAlign = HostTarget->getDoubleAlign();
   LongWidth = HostTarget->getLongWidth();
   LongAlign = HostTarget->getLongAlign();
   LongLongWidth = HostTarget->getLongLongWidth();
   LongLongAlign = HostTarget->getLongLongAlign();
-  MinGlobalAlign = HostTarget->getMinGlobalAlign();
+  MinGlobalAlign = HostTarget->getMinGlobalAlign(/* TypeSize = */ 0);
   NewAlign = HostTarget->getNewAlign();
   DefaultAlignForAttributeAligned =
       HostTarget->getDefaultAlignForAttributeAligned();
   SizeType = HostTarget->getSizeType();
   IntMaxType = HostTarget->getIntMaxType();
   PtrDiffType = HostTarget->getPtrDiffType(/* AddrSpace = */ 0);
   IntPtrType = HostTarget->getIntPtrType();
   WCharType = HostTarget->getWCharType();

test/CodeGen/arm64-microsoft-struct-align.cpp

+// RUN: %clang_cc1 -triple aarch64-windows -ffreestanding -emit-llvm -O0 \
+// RUN: -x c++ -o - %s | FileCheck %s
+
+struct size1 { char str[1]; };
+struct size2 { char str[2]; };
+struct size7 { char str[4]; };
+struct size8 { char str[8]; };
+struct size63 { char str[63]; };
+struct size64 { char str[64]; };
+
+struct size1 s1;
+// CHECK: @"?s1@@3Usize1@@A" = dso_local global %struct.size1 zeroinitializer, align 1
+
+struct size2 s2;
+// CHECK: @"?s2@@3Usize2@@A" = dso_local global %struct.size2 zeroinitializer, align 4
+
+struct size7 s7;
+// CHECK: @"?s7@@3Usize7@@A" = dso_local global %struct.size7 zeroinitializer, align 4
+
+struct size8 s8;
+// CHECK: @"?s8@@3Usize8@@A" = dso_local global %struct.size8 zeroinitializer, align 8
+
+struct size63 s63;
+// CHECK: @"?s63@@3Usize63@@A" = dso_local global %struct.size63 zeroinitializer, align 8
+
+struct size64 s64;
+// CHECK: @"?s64@@3Usize64@@A" = dso_local global %struct.size64 zeroinitializer, align 16