Put large common blocks into .lbss for the medium and large code models

llvm/include/llvm/BinaryFormat/ELF.h

@@ -589,6 +589,11 @@ enum {
   SHN_MIPS_SUNDEFINED = 0xff04 // Undefined symbols for global data area
 };
 
+// x86-64 speciifc section index
+enum {
+  SHN_X86_64_LCOMMON = 0xff02, // Large FORTRAN COMMON variables
+};
+
 // ELF Relocation types for Mips
 enum {
 #include "ELFRelocs/Mips.def"

llvm/include/llvm/MC/MCELFStreamer.h

@@ -56,6 +56,8 @@ class MCELFStreamer : public MCObjectStreamer {
   bool emitSymbolAttribute(MCSymbol *Symbol, MCSymbolAttr Attribute) override;
   void emitCommonSymbol(MCSymbol *Symbol, uint64_t Size,
                         Align ByteAlignment) override;
+  void emitLargeCommonSymbol(MCSymbol *Symbol, uint64_t Size,
+                             Align ByteAlignment) override;
 
   void emitELFSize(MCSymbol *Symbol, const MCExpr *Value) override;
   void emitELFSymverDirective(const MCSymbol *OriginalSym, StringRef Name,

llvm/include/llvm/MC/MCStreamer.h

@@ -677,6 +677,14 @@ class LLVM_ABI MCStreamer {
   virtual void emitCommonSymbol(MCSymbol *Symbol, uint64_t Size,
                                 Align ByteAlignment) = 0;
 
+  /// Emit a large common (.largecomm) symbol.
+  ///
+  /// \param Symbol - The common symbol to emit.
+  /// \param Size - The size of the common symbol.
+  /// \param ByteAlignment - The alignment of the common symbol in bytes.
+  virtual void emitLargeCommonSymbol(MCSymbol *Symbol, uint64_t Size,
+                                     Align ByteAlignment);
+
   /// Emit a local common (.lcomm) symbol.
   ///
   /// \param Symbol - The common symbol to emit.

llvm/include/llvm/MC/MCSymbolELF.h

@@ -48,6 +48,9 @@ class MCSymbolELF : public MCSymbol {
   void setMemtag(bool Tagged);
   bool isMemtag() const;
 
+  void setIsLargeCommon();
+  bool isLargeCommon() const;
+
 private:
   void setIsBindingSet() const;
 };

llvm/include/llvm/Object/ELFTypes.h

@@ -252,6 +252,8 @@ struct Elf_Sym_Impl : Elf_Sym_Base<ELFT> {
     return getType() == ELF::STT_COMMON || st_shndx == ELF::SHN_COMMON;
   }
 
+  bool isLargeCommon() const { return st_shndx == ELF::SHN_X86_64_LCOMMON; }
+
   bool isDefined() const { return !isUndefined(); }
 
   bool isProcessorSpecific() const {

llvm/include/llvm/Target/TargetLoweringObjectFile.h

@@ -305,6 +305,12 @@ class LLVM_ABI TargetLoweringObjectFile : public MCObjectFileInfo {
     return nullptr;
   }
 
+  virtual MCSection *LargeSectionForCommon(const GlobalObject *GO,
+                                           SectionKind Kind,
+                                           const TargetMachine &TM) const {
+    return nullptr;
+  }
+
 protected:
   virtual MCSection *SelectSectionForGlobal(const GlobalObject *GO,
                                             SectionKind Kind,

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

@@ -782,11 +782,18 @@ void AsmPrinter::emitGlobalVariable(const GlobalVariable *GV) {
     OutContext.reportError(SMLoc(), "symbol '" + Twine(GVSym->getName()) +
                                         "' is already defined");
 
+  SectionKind GVKind = TargetLoweringObjectFile::getKindForGlobal(GV, TM);
+
+  // Place the large common variables in the .lbss section for the medium and
+  // large code models on x86_64. (AMD64 ABI, 9.2.5 COMMON blocks)
+  MCSection *LargeCommonSec =
+      getObjFileLowering().LargeSectionForCommon(GV, GVKind, TM);
+  if (LargeCommonSec)
+    OutStreamer->switchSection(LargeCommonSec);
+
   if (MAI->hasDotTypeDotSizeDirective())
     OutStreamer->emitSymbolAttribute(EmittedSym, MCSA_ELF_TypeObject);
 
-  SectionKind GVKind = TargetLoweringObjectFile::getKindForGlobal(GV, TM);
-
   const DataLayout &DL = GV->getDataLayout();
   uint64_t Size = DL.getTypeAllocSize(GV->getValueType());
 
@@ -802,10 +809,12 @@ void AsmPrinter::emitGlobalVariable(const GlobalVariable *GV) {
   if (GVKind.isCommon()) {
     if (Size == 0) Size = 1;   // .comm Foo, 0 is undefined, avoid it.
     // .comm _foo, 42, 4
-    OutStreamer->emitCommonSymbol(GVSym, Size, Alignment);
+    if (LargeCommonSec)
+      OutStreamer->emitLargeCommonSymbol(GVSym, Size, Alignment);
+    else
+      OutStreamer->emitCommonSymbol(GVSym, Size, Alignment);
     return;
   }
-
   // Determine to which section this global should be emitted.
   MCSection *TheSection = getObjFileLowering().SectionForGlobal(GV, GVKind, TM);
 

llvm/lib/MC/ELFObjectWriter.cpp

@@ -27,6 +27,7 @@
 #include "llvm/MC/MCELFObjectWriter.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCFixup.h"
+#include "llvm/MC/MCObjectFileInfo.h"
 #include "llvm/MC/MCObjectWriter.h"
 #include "llvm/MC/MCSection.h"
 #include "llvm/MC/MCSectionELF.h"
@@ -544,7 +545,10 @@ void ELFWriter::computeSymbolTable(const RevGroupMapTy &RevGroupMap) {
       auto Shndx = Symbol.getIndex();
       if (!Shndx) {
         assert(!Local);
-        Shndx = ELF::SHN_COMMON;
+        if (Symbol.isLargeCommon())
+          Shndx = ELF::SHN_X86_64_LCOMMON;
+        else
+          Shndx = ELF::SHN_COMMON;
       }
       MSD.SectionIndex = Shndx;
     } else if (Symbol.isUndefined()) {

llvm/lib/MC/MCAsmStreamer.cpp

@@ -228,7 +228,8 @@ class MCAsmStreamer final : public MCStreamer {
   void emitELFSize(MCSymbol *Symbol, const MCExpr *Value) override;
   void emitCommonSymbol(MCSymbol *Symbol, uint64_t Size,
                         Align ByteAlignment) override;
-
+  void emitLargeCommonSymbol(MCSymbol *Symbol, uint64_t Size,
+                             Align ByteAlignment) override;
   /// Emit a local common (.lcomm) symbol.
   ///
   /// @param Symbol - The common symbol to emit.
@@ -1075,6 +1076,19 @@ void MCAsmStreamer::emitCommonSymbol(MCSymbol *Symbol, uint64_t Size,
   }
 }
 
+void MCAsmStreamer::emitLargeCommonSymbol(MCSymbol *Symbol, uint64_t Size,
+                                          Align ByteAlignment) {
+  OS << "\t.largecomm\t";
+  Symbol->print(OS, MAI);
+  OS << ',' << Size;
+
+  if (MAI->getCOMMDirectiveAlignmentIsInBytes())
+    OS << ',' << ByteAlignment.value();
+  else
+    OS << ',' << Log2(ByteAlignment);
+  EmitEOL();
+}
+
 void MCAsmStreamer::emitLocalCommonSymbol(MCSymbol *Symbol, uint64_t Size,
                                           Align ByteAlign) {
   OS << "\t.lcomm\t";

llvm/lib/MC/MCELFStreamer.cpp

@@ -244,6 +244,23 @@ bool MCELFStreamer::emitSymbolAttribute(MCSymbol *S, MCSymbolAttr Attribute) {
   return true;
 }
 
+void MCELFStreamer::emitLargeCommonSymbol(MCSymbol *S, uint64_t Size,
+                                          Align ByteAlignment) {
+  auto *Symbol = static_cast<MCSymbolELF *>(S);
+  getAssembler().registerSymbol(*Symbol);
+
+  if (!Symbol->isBindingSet())
+    Symbol->setBinding(ELF::STB_GLOBAL);
+
+  Symbol->setType(ELF::STT_OBJECT);
+
+  if (Symbol->declareCommon(Size, ByteAlignment))
+    report_fatal_error(Twine("Symbol: ") + Symbol->getName() +
+                       " redeclared as different type");
+  Symbol->setIsLargeCommon();
+  Symbol->setSize(MCConstantExpr::create(Size, getContext()));
+}
+
 void MCELFStreamer::emitCommonSymbol(MCSymbol *S, uint64_t Size,
                                      Align ByteAlignment) {
   auto *Symbol = static_cast<MCSymbolELF *>(S);

llvm/lib/MC/MCParser/ELFAsmParser.cpp

@@ -77,6 +77,7 @@ class ELFAsmParser : public MCAsmParserExtension {
       &ELFAsmParser::parseDirectiveSymbolAttribute>(".hidden");
     addDirectiveHandler<&ELFAsmParser::parseDirectiveSubsection>(".subsection");
     addDirectiveHandler<&ELFAsmParser::parseDirectiveCGProfile>(".cg_profile");
+    addDirectiveHandler<&ELFAsmParser::parseDirectiveLargecomm>(".largecomm");
   }
 
   // FIXME: Part of this logic is duplicated in the MCELFStreamer. What is
@@ -126,6 +127,7 @@ class ELFAsmParser : public MCAsmParserExtension {
   bool parseDirectiveSymbolAttribute(StringRef, SMLoc);
   bool parseDirectiveSubsection(StringRef, SMLoc);
   bool parseDirectiveCGProfile(StringRef, SMLoc);
+  bool parseDirectiveLargecomm(StringRef, SMLoc);
 
 private:
   bool parseSectionName(StringRef &SectionName);
@@ -886,6 +888,54 @@ bool ELFAsmParser::parseDirectiveCGProfile(StringRef S, SMLoc Loc) {
   return MCAsmParserExtension::parseDirectiveCGProfile(S, Loc);
 }
 
+bool ELFAsmParser::parseDirectiveLargecomm(StringRef, SMLoc Loc) {
+  if (getParser().checkForValidSection())
+    return true;
+
+  MCSymbol *Sym;
+  if (getParser().parseSymbol(Sym))
+    return TokError("expected identifier in directive");
+
+  if (getLexer().isNot(AsmToken::Comma))
+    return TokError("expected a comma");
+  Lex();
+
+  int64_t Size;
+  SMLoc SizeLoc = getLexer().getLoc();
+  if (getParser().parseAbsoluteExpression(Size))
+    return true;
+
+  int64_t Pow2Alignment = 0;
+  SMLoc Pow2AlignmentLoc;
+  if (getLexer().is(AsmToken::Comma)) {
+    Lex();
+    Pow2AlignmentLoc = getLexer().getLoc();
+    if (getParser().parseAbsoluteExpression(Pow2Alignment))
+      return true;
+
+    // If this target takes alignments in bytes (not log) validate and convert.
+    if (getLexer().getMAI().getCOMMDirectiveAlignmentIsInBytes()) {
+      if (!isPowerOf2_64(Pow2Alignment))
+        return Error(Pow2AlignmentLoc, "alignment must be a power of 2");
+      Pow2Alignment = Log2_64(Pow2Alignment);
+    }
+  }
+
+  if (parseEOL())
+    return true;
+
+  if (Size < 0)
+    return Error(SizeLoc, "size must be non-negative");
+
+  Sym->redefineIfPossible();
+  if (!Sym->isUndefined())
+    return Error(Loc, "invalid symbol redefinition");
+
+  getStreamer().emitLargeCommonSymbol(Sym, Size, Align(1ULL << Pow2Alignment));
+
+  return false;
+}
+
 namespace llvm {
 
 MCAsmParserExtension *createELFAsmParser() {

llvm/lib/MC/MCStreamer.cpp

@@ -1309,6 +1309,8 @@ void MCStreamer::emitELFSymverDirective(const MCSymbol *OriginalSym,
                                         StringRef Name, bool KeepOriginalSym) {}
 void MCStreamer::emitLocalCommonSymbol(MCSymbol *Symbol, uint64_t Size,
                                        Align ByteAlignment) {}
+void MCStreamer::emitLargeCommonSymbol(MCSymbol *Symbol, uint64_t Size,
+                                       Align ByteAlignment) {}
 void MCStreamer::emitZerofill(MCSection *, MCSymbol *, uint64_t, Align, SMLoc) {
 }
 void MCStreamer::emitTBSSSymbol(MCSection *Section, MCSymbol *Symbol,

llvm/lib/MC/MCSymbolELF.cpp

@@ -37,6 +37,9 @@ enum {
 
   // One bit.
   ELF_IsMemoryTagged_Shift = 13,
+
+  // One bit.
+  ELF_LargeCommon_Shift = 14,
 };
 }
 
@@ -206,4 +209,13 @@ void MCSymbolELF::setMemtag(bool Tagged) {
   else
     setFlags(OtherFlags);
 }
+
+bool MCSymbolELF::isLargeCommon() const {
+  return getFlags() & (0x1 << ELF_LargeCommon_Shift);
+}
+
+void MCSymbolELF::setIsLargeCommon() {
+  uint32_t OtherFlags = getFlags() & ~(0x1 << ELF_LargeCommon_Shift);
+  setFlags(OtherFlags | (1 << ELF_LargeCommon_Shift));
+}
 }

llvm/lib/Target/X86/X86TargetObjectFile.cpp

@@ -8,8 +8,11 @@
 
 #include "X86TargetObjectFile.h"
 #include "MCTargetDesc/X86MCAsmInfo.h"
+#include "X86TargetMachine.h"
 #include "llvm/BinaryFormat/Dwarf.h"
+#include "llvm/BinaryFormat/ELF.h"
 #include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCSectionELF.h"
 #include "llvm/MC/MCValue.h"
 #include "llvm/Target/TargetMachine.h"
 
@@ -71,3 +74,13 @@ const MCExpr *X86_64ELFTargetObjectFile::getIndirectSymViaGOTPCRel(
   const MCExpr *Off = MCConstantExpr::create(FinalOffset, getContext());
   return MCBinaryExpr::createAdd(Res, Off, getContext());
 }
+
+MCSection *X86_64ELFTargetObjectFile::LargeSectionForCommon(
+    const GlobalObject *GV, SectionKind Kind, const TargetMachine &TM) const {
+  auto &X86_TM = static_cast<const X86TargetMachine &>(TM);
+  if (GV && Kind.isCommon() && TM.isLargeGlobalValue(GV) && !X86_TM.isJIT()) {
+    unsigned Flags = ELF::SHF_ALLOC | ELF::SHF_WRITE | ELF::SHF_X86_64_LARGE;
+    return getContext().getELFSection(".lbss", ELF::SHT_NOBITS, Flags);
+  }
+  return nullptr;
+}

llvm/lib/Target/X86/X86TargetObjectFile.h

@@ -56,6 +56,9 @@ namespace llvm {
                               const MCValue &MV, int64_t Offset,
                               MachineModuleInfo *MMI,
                               MCStreamer &Streamer) const override;
+
+    MCSection *LargeSectionForCommon(const GlobalObject *GO, SectionKind Kind,
+                                     const TargetMachine &TM) const override;
   };
 
 } // end namespace llvm

llvm/test/MC/X86/largecomm.ll

@@ -0,0 +1,50 @@
+; RUN: llc -filetype=asm -code-model=medium %s --large-data-threshold=65636 -o - | FileCheck %s --check-prefix=CHECKASM-MEDIUM
+; RUN: llc -filetype=asm -code-model=large %s -o - | FileCheck %s --check-prefix=CHECKASM-LARGE
+
+; CHECKASM-MEDIUM:       .section	.lbss,"awl",@nobits
+; CHECKASM-MEDIUM-NEXT:	 .type	__BLNK__,@object                # @__BLNK__
+; CHECKASM-MEDIUM-NEXT:  .largecomm	__BLNK__,48394093832,8
+; CHECKASM-MEDIUM-NEXT:  .type	ccc_,@object                    # @ccc_
+; CHECKASM-MEDIUM-NEXT:  .comm	ccc_,8,8
+
+; CHECKASM-LARGE:       .section	.lbss,"awl",@nobits
+; CHECKASM-LARGE-NEXT:	.type	__BLNK__,@object                # @__BLNK__
+; CHECKASM-LARGE-NEXT:  .largecomm	__BLNK__,48394093832,8
+; CHECKASM-LARGE-NEXT:  .type	ccc_,@object                    # @ccc_
+; CHECKASM-LARGE-NEXT:  .largecomm	ccc_,8,8
+
+source_filename = "FIRModule"
+target triple = "x86_64-unknown-linux-gnu"
+
+@__BLNK__ = common global [48394093832 x i8] zeroinitializer, align 8
+@ccc_ = common global [8 x i8] zeroinitializer, align 8
+@_QFECn1 = internal constant i32 77777
+@_QFECn2 = internal constant i32 77777
+
+define void @_QQmain() #0 {
+  store double 1.000000e+00, ptr @ccc_, align 8
+  store double 2.000000e+00, ptr getelementptr inbounds nuw (i8, ptr @__BLNK__, i64 61600176), align 8
+  ret void
+}
+
+declare void @_FortranAProgramStart(i32, ptr, ptr, ptr) #1
+
+declare void @_FortranAProgramEndStatement() #1
+
+define i32 @main(i32 %0, ptr %1, ptr %2) #0 {
+  call void @_FortranAProgramStart(i32 %0, ptr %1, ptr %2, ptr null)
+  call void @_QQmain()
+  call void @_FortranAProgramEndStatement()
+  ret i32 0
+}
+
+attributes #0 = { "frame-pointer"="all" "target-cpu"="x86-64" }
+attributes #1 = { "frame-pointer"="all" }
+
+!llvm.ident = !{!0}
+!llvm.module.flags = !{!1, !2, !3}
+
+!0 = !{!"flang version 22.0.0 (https://github.com/llvm/llvm-project.git e1afe25356b8d2ee14f5f88bdb6c2a1526ed14ef)"}
+!1 = !{i32 2, !"Debug Info Version", i32 3}
+!2 = !{i32 8, !"PIC Level", i32 2}
+!3 = !{i32 7, !"PIE Level", i32 2}