//
//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "assembler"
#include "llvm/MC/MCAssembler.h"
-#include "llvm/MC/MCSectionMachO.h"
-#include "llvm/Target/TargetMachOWriterInfo.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/SmallString.h"
-#include "llvm/ADT/StringMap.h"
+#include "llvm/MC/MCAsmLayout.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCObjectWriter.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/MC/MCValue.h"
+#include "llvm/ADT/OwningPtr.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/Twine.h"
+#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetRegistry.h"
+#include "llvm/Target/TargetAsmBackend.h"
+
+#include <vector>
using namespace llvm;
-class MachObjectWriter;
-
-static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
- MachObjectWriter &MOW);
-
-class MachObjectWriter {
- // See <mach-o/loader.h>.
- enum {
- Header_Magic32 = 0xFEEDFACE,
- Header_Magic64 = 0xFEEDFACF
- };
-
- static const unsigned Header32Size = 28;
- static const unsigned Header64Size = 32;
- static const unsigned SegmentLoadCommand32Size = 56;
- static const unsigned Section32Size = 68;
- static const unsigned SymtabLoadCommandSize = 24;
- static const unsigned DysymtabLoadCommandSize = 80;
- static const unsigned Nlist32Size = 12;
-
- enum HeaderFileType {
- HFT_Object = 0x1
- };
-
- enum LoadCommandType {
- LCT_Segment = 0x1,
- LCT_Symtab = 0x2,
- LCT_Dysymtab = 0xb
- };
-
- // See <mach-o/nlist.h>.
- enum SymbolTypeType {
- STT_Undefined = 0x00,
- STT_Absolute = 0x02,
- STT_Section = 0x0e
- };
-
- enum SymbolTypeFlags {
- // If any of these bits are set, then the entry is a stab entry number (see
- // <mach-o/stab.h>. Otherwise the other masks apply.
- STF_StabsEntryMask = 0xe0,
-
- STF_TypeMask = 0x0e,
- STF_External = 0x01,
- STF_PrivateExtern = 0x10
- };
-
- /// MachSymbolData - Helper struct for containing some precomputed information
- /// on symbols.
- struct MachSymbolData {
- MCSymbolData *SymbolData;
- uint64_t StringIndex;
- uint8_t SectionIndex;
-
- // Support lexicographic sorting.
- bool operator<(const MachSymbolData &RHS) const {
- const std::string &Name = SymbolData->getSymbol().getName();
- return Name < RHS.SymbolData->getSymbol().getName();
- }
- };
+namespace {
+namespace stats {
+STATISTIC(EmittedFragments, "Number of emitted assembler fragments");
+STATISTIC(EvaluateFixup, "Number of evaluated fixups");
+STATISTIC(FragmentLayouts, "Number of fragment layouts");
+STATISTIC(ObjectBytes, "Number of emitted object file bytes");
+STATISTIC(RelaxationSteps, "Number of assembler layout and relaxation steps");
+STATISTIC(RelaxedInstructions, "Number of relaxed instructions");
+STATISTIC(SectionLayouts, "Number of section layouts");
+}
+}
- raw_ostream &OS;
- bool IsLSB;
+// FIXME FIXME FIXME: There are number of places in this file where we convert
+// what is a 64-bit assembler value used for computation into a value in the
+// object file, which may truncate it. We should detect that truncation where
+// invalid and report errors back.
-public:
- MachObjectWriter(raw_ostream &_OS, bool _IsLSB = true)
- : OS(_OS), IsLSB(_IsLSB) {
- }
+/* *** */
- /// @name Helper Methods
- /// @{
+MCAsmLayout::MCAsmLayout(MCAssembler &Asm)
+ : Assembler(Asm), LastValidFragment(0)
+ {
+ // Compute the section layout order. Virtual sections must go last.
+ for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
+ if (!Asm.getBackend().isVirtualSection(it->getSection()))
+ SectionOrder.push_back(&*it);
+ for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
+ if (Asm.getBackend().isVirtualSection(it->getSection()))
+ SectionOrder.push_back(&*it);
+}
- void Write8(uint8_t Value) {
- OS << char(Value);
- }
+bool MCAsmLayout::isSectionUpToDate(const MCSectionData *SD) const {
+ // The first section is always up-to-date.
+ unsigned Index = SD->getLayoutOrder();
+ if (!Index)
+ return true;
- void Write16(uint16_t Value) {
- if (IsLSB) {
- Write8(uint8_t(Value >> 0));
- Write8(uint8_t(Value >> 8));
- } else {
- Write8(uint8_t(Value >> 8));
- Write8(uint8_t(Value >> 0));
- }
- }
+ // Otherwise, sections are always implicitly computed when the preceeding
+ // fragment is layed out.
+ const MCSectionData *Prev = getSectionOrder()[Index - 1];
+ return isFragmentUpToDate(&(Prev->getFragmentList().back()));
+}
- void Write32(uint32_t Value) {
- if (IsLSB) {
- Write16(uint16_t(Value >> 0));
- Write16(uint16_t(Value >> 16));
- } else {
- Write16(uint16_t(Value >> 16));
- Write16(uint16_t(Value >> 0));
- }
- }
+bool MCAsmLayout::isFragmentUpToDate(const MCFragment *F) const {
+ return (LastValidFragment &&
+ F->getLayoutOrder() <= LastValidFragment->getLayoutOrder());
+}
- void Write64(uint64_t Value) {
- if (IsLSB) {
- Write32(uint32_t(Value >> 0));
- Write32(uint32_t(Value >> 32));
- } else {
- Write32(uint32_t(Value >> 32));
- Write32(uint32_t(Value >> 0));
+void MCAsmLayout::UpdateForSlide(MCFragment *F, int SlideAmount) {
+ // If this fragment wasn't already up-to-date, we don't need to do anything.
+ if (!isFragmentUpToDate(F))
+ return;
+
+ // Otherwise, reset the last valid fragment to the predecessor of the
+ // invalidated fragment.
+ LastValidFragment = F->getPrevNode();
+ if (!LastValidFragment) {
+ unsigned Index = F->getParent()->getLayoutOrder();
+ if (Index != 0) {
+ MCSectionData *Prev = getSectionOrder()[Index - 1];
+ LastValidFragment = &(Prev->getFragmentList().back());
}
}
+}
- void WriteZeros(unsigned N) {
- const char Zeros[16] = { 0 };
-
- for (unsigned i = 0, e = N / 16; i != e; ++i)
- OS << StringRef(Zeros, 16);
-
- OS << StringRef(Zeros, N % 16);
- }
+void MCAsmLayout::EnsureValid(const MCFragment *F) const {
+ // Advance the layout position until the fragment is up-to-date.
+ while (!isFragmentUpToDate(F)) {
+ // Advance to the next fragment.
+ MCFragment *Cur = LastValidFragment;
+ if (Cur)
+ Cur = Cur->getNextNode();
+ if (!Cur) {
+ unsigned NextIndex = 0;
+ if (LastValidFragment)
+ NextIndex = LastValidFragment->getParent()->getLayoutOrder() + 1;
+ Cur = SectionOrder[NextIndex]->begin();
+ }
- void WriteString(const StringRef &Str, unsigned ZeroFillSize = 0) {
- OS << Str;
- if (ZeroFillSize)
- WriteZeros(ZeroFillSize - Str.size());
+ const_cast<MCAsmLayout*>(this)->LayoutFragment(Cur);
}
+}
- /// @}
-
- void WriteHeader32(unsigned NumLoadCommands, unsigned LoadCommandsSize) {
- // struct mach_header (28 bytes)
-
- uint64_t Start = OS.tell();
- (void) Start;
-
- Write32(Header_Magic32);
-
- // FIXME: Support cputype.
- Write32(TargetMachOWriterInfo::HDR_CPU_TYPE_I386);
-
- // FIXME: Support cpusubtype.
- Write32(TargetMachOWriterInfo::HDR_CPU_SUBTYPE_I386_ALL);
-
- Write32(HFT_Object);
-
- // Object files have a single load command, the segment.
- Write32(NumLoadCommands);
- Write32(LoadCommandsSize);
- Write32(0); // Flags
-
- assert(OS.tell() - Start == Header32Size);
- }
-
- /// WriteSegmentLoadCommand32 - Write a 32-bit segment load command.
- ///
- /// \arg NumSections - The number of sections in this segment.
- /// \arg SectionDataSize - The total size of the sections.
- void WriteSegmentLoadCommand32(unsigned NumSections,
- uint64_t SectionDataStartOffset,
- uint64_t SectionDataSize) {
- // struct segment_command (56 bytes)
-
- uint64_t Start = OS.tell();
- (void) Start;
-
- Write32(LCT_Segment);
- Write32(SegmentLoadCommand32Size + NumSections * Section32Size);
-
- WriteString("", 16);
- Write32(0); // vmaddr
- Write32(SectionDataSize); // vmsize
- Write32(SectionDataStartOffset); // file offset
- Write32(SectionDataSize); // file size
- Write32(0x7); // maxprot
- Write32(0x7); // initprot
- Write32(NumSections);
- Write32(0); // flags
-
- assert(OS.tell() - Start == SegmentLoadCommand32Size);
- }
-
- void WriteSection32(const MCSectionData &SD, uint64_t FileOffset) {
- // struct section (68 bytes)
-
- uint64_t Start = OS.tell();
- (void) Start;
-
- // FIXME: cast<> support!
- const MCSectionMachO &Section =
- static_cast<const MCSectionMachO&>(SD.getSection());
- WriteString(Section.getSectionName(), 16);
- WriteString(Section.getSegmentName(), 16);
- Write32(0); // address
- Write32(SD.getFileSize()); // size
- Write32(FileOffset);
-
- assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
- Write32(Log2_32(SD.getAlignment()));
- Write32(0); // file offset of relocation entries
- Write32(0); // number of relocation entrions
- Write32(Section.getTypeAndAttributes());
- Write32(0); // reserved1
- Write32(Section.getStubSize()); // reserved2
-
- assert(OS.tell() - Start == Section32Size);
- }
-
- void WriteSymtabLoadCommand(uint32_t SymbolOffset, uint32_t NumSymbols,
- uint32_t StringTableOffset,
- uint32_t StringTableSize) {
- // struct symtab_command (24 bytes)
-
- uint64_t Start = OS.tell();
- (void) Start;
-
- Write32(LCT_Symtab);
- Write32(SymtabLoadCommandSize);
- Write32(SymbolOffset);
- Write32(NumSymbols);
- Write32(StringTableOffset);
- Write32(StringTableSize);
-
- assert(OS.tell() - Start == SymtabLoadCommandSize);
- }
-
- void WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
- uint32_t NumLocalSymbols,
- uint32_t FirstExternalSymbol,
- uint32_t NumExternalSymbols,
- uint32_t FirstUndefinedSymbol,
- uint32_t NumUndefinedSymbols,
- uint32_t IndirectSymbolOffset,
- uint32_t NumIndirectSymbols) {
- // struct dysymtab_command (80 bytes)
-
- uint64_t Start = OS.tell();
- (void) Start;
-
- Write32(LCT_Dysymtab);
- Write32(DysymtabLoadCommandSize);
- Write32(FirstLocalSymbol);
- Write32(NumLocalSymbols);
- Write32(FirstExternalSymbol);
- Write32(NumExternalSymbols);
- Write32(FirstUndefinedSymbol);
- Write32(NumUndefinedSymbols);
- Write32(0); // tocoff
- Write32(0); // ntoc
- Write32(0); // modtaboff
- Write32(0); // nmodtab
- Write32(0); // extrefsymoff
- Write32(0); // nextrefsyms
- Write32(IndirectSymbolOffset);
- Write32(NumIndirectSymbols);
- Write32(0); // extreloff
- Write32(0); // nextrel
- Write32(0); // locreloff
- Write32(0); // nlocrel
-
- assert(OS.tell() - Start == DysymtabLoadCommandSize);
- }
-
- void WriteNlist32(MachSymbolData &MSD) {
- MCSymbol &Symbol = MSD.SymbolData->getSymbol();
- uint8_t Type = 0;
-
- // Set the N_TYPE bits. See <mach-o/nlist.h>.
- //
- // FIXME: Are the prebound or indirect fields possible here?
- if (Symbol.isUndefined())
- Type = STT_Undefined;
- else if (Symbol.isAbsolute())
- Type = STT_Absolute;
- else
- Type = STT_Section;
-
- // FIXME: Set STAB bits.
-
- // FIXME: Set private external bit.
-
- // Set external bit.
- if (MSD.SymbolData->isExternal())
- Type |= STF_External;
-
- // struct nlist (12 bytes)
-
- Write32(MSD.StringIndex);
- Write8(Type);
- Write8(MSD.SectionIndex);
- Write16(0); // FIXME: Desc
- Write32(0); // FIXME: Value
- }
-
- /// ComputeSymbolTable - Compute the symbol table data
- ///
- /// \param StringTable [out] - The string table data.
- /// \param StringIndexMap [out] - Map from symbol names to offsets in the
- /// string table.
-
- void ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable,
- std::vector<MachSymbolData> &LocalSymbolData,
- std::vector<MachSymbolData> &ExternalSymbolData,
- std::vector<MachSymbolData> &UndefinedSymbolData) {
- // Build section lookup table.
- DenseMap<const MCSection*, uint8_t> SectionIndexMap;
- unsigned Index = 1;
- for (MCAssembler::iterator it = Asm.begin(),
- ie = Asm.end(); it != ie; ++it, ++Index)
- SectionIndexMap[&it->getSection()] = Index;
- assert(Index <= 256 && "Too many sections!");
-
- // Index 0 is always the empty string.
- StringMap<uint64_t> StringIndexMap;
- StringTable += '\x00';
-
- // Build the symbol arrays and the string table, but only for non-local
- // symbols.
- //
- // The particular order that we collect the symbols and create the string
- // table, then sort the symbols is chosen to match 'as'. Even though it
- // doesn't matter for correctness, this is important for letting us diff .o
- // files.
- for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
- ie = Asm.symbol_end(); it != ie; ++it) {
- MCSymbol &Symbol = it->getSymbol();
-
- if (!it->isExternal())
- continue;
-
- uint64_t &Entry = StringIndexMap[Symbol.getName()];
- if (!Entry) {
- Entry = StringTable.size();
- StringTable += Symbol.getName();
- StringTable += '\x00';
- }
-
- MachSymbolData MSD;
- MSD.SymbolData = it;
- MSD.StringIndex = Entry;
-
- if (Symbol.isUndefined()) {
- MSD.SectionIndex = 0;
- UndefinedSymbolData.push_back(MSD);
- } else if (Symbol.isAbsolute()) {
- MSD.SectionIndex = 0;
- ExternalSymbolData.push_back(MSD);
- } else {
- MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
- assert(MSD.SectionIndex && "Invalid section index!");
- ExternalSymbolData.push_back(MSD);
- }
- }
-
- // Now add the data for local symbols.
- for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
- ie = Asm.symbol_end(); it != ie; ++it) {
- MCSymbol &Symbol = it->getSymbol();
+void MCAsmLayout::FragmentReplaced(MCFragment *Src, MCFragment *Dst) {
+ if (LastValidFragment == Src)
+ LastValidFragment = Dst;
- if (it->isExternal())
- continue;
+ Dst->Offset = Src->Offset;
+ Dst->EffectiveSize = Src->EffectiveSize;
+}
- uint64_t &Entry = StringIndexMap[Symbol.getName()];
- if (!Entry) {
- Entry = StringTable.size();
- StringTable += Symbol.getName();
- StringTable += '\x00';
- }
+uint64_t MCAsmLayout::getFragmentAddress(const MCFragment *F) const {
+ assert(F->getParent() && "Missing section()!");
+ return getSectionAddress(F->getParent()) + getFragmentOffset(F);
+}
- MachSymbolData MSD;
- MSD.SymbolData = it;
- MSD.StringIndex = Entry;
-
- assert(!Symbol.isUndefined() && "Local symbol can not be undefined!");
- if (Symbol.isAbsolute()) {
- MSD.SectionIndex = 0;
- LocalSymbolData.push_back(MSD);
- } else {
- MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
- assert(MSD.SectionIndex && "Invalid section index!");
- LocalSymbolData.push_back(MSD);
- }
- }
+uint64_t MCAsmLayout::getFragmentEffectiveSize(const MCFragment *F) const {
+ EnsureValid(F);
+ assert(F->EffectiveSize != ~UINT64_C(0) && "Address not set!");
+ return F->EffectiveSize;
+}
- // External and undefined symbols are required to be in lexicographic order.
- std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
- std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
-
- // The string table is padded to a multiple of 4.
- //
- // FIXME: Check to see if this varies per arch.
- while (StringTable.size() % 4)
- StringTable += '\x00';
- }
-
- void WriteObject(MCAssembler &Asm) {
- unsigned NumSections = Asm.size();
-
- // Compute symbol table information.
- SmallString<256> StringTable;
- std::vector<MachSymbolData> LocalSymbolData;
- std::vector<MachSymbolData> ExternalSymbolData;
- std::vector<MachSymbolData> UndefinedSymbolData;
- unsigned NumSymbols = Asm.symbol_size();
-
- // No symbol table command is written if there are no symbols.
- if (NumSymbols)
- ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData,
- UndefinedSymbolData);
-
- // Compute the file offsets for all the sections in advance, so that we can
- // write things out in order.
- SmallVector<uint64_t, 16> SectionFileOffsets;
- SectionFileOffsets.resize(NumSections);
-
- // The section data starts after the header, the segment load command (and
- // section headers) and the symbol table.
- unsigned NumLoadCommands = 1;
- uint64_t LoadCommandsSize =
- SegmentLoadCommand32Size + NumSections * Section32Size;
-
- // Add the symbol table load command sizes, if used.
- if (NumSymbols) {
- NumLoadCommands += 2;
- LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize;
- }
+uint64_t MCAsmLayout::getFragmentOffset(const MCFragment *F) const {
+ EnsureValid(F);
+ assert(F->Offset != ~UINT64_C(0) && "Address not set!");
+ return F->Offset;
+}
- uint64_t FileOffset = Header32Size + LoadCommandsSize;
- uint64_t SectionDataStartOffset = FileOffset;
- uint64_t SectionDataSize = 0;
- unsigned Index = 0;
- for (MCAssembler::iterator it = Asm.begin(),
- ie = Asm.end(); it != ie; ++it, ++Index) {
- SectionFileOffsets[Index] = FileOffset;
- FileOffset += it->getFileSize();
- SectionDataSize += it->getFileSize();
- }
+uint64_t MCAsmLayout::getSymbolAddress(const MCSymbolData *SD) const {
+ assert(SD->getFragment() && "Invalid getAddress() on undefined symbol!");
+ return getFragmentAddress(SD->getFragment()) + SD->getOffset();
+}
- // Write the prolog, starting with the header and load command...
- WriteHeader32(NumLoadCommands, LoadCommandsSize);
- WriteSegmentLoadCommand32(NumSections, SectionDataStartOffset,
- SectionDataSize);
-
- // ... and then the section headers.
- Index = 0;
- for (MCAssembler::iterator it = Asm.begin(),
- ie = Asm.end(); it != ie; ++it, ++Index)
- WriteSection32(*it, SectionFileOffsets[Index]);
-
- // Write the symbol table load command, if used.
- if (NumSymbols) {
- // The string table is written after all the section data.
- uint64_t SymbolTableOffset = SectionDataStartOffset + SectionDataSize;
- uint64_t StringTableOffset =
- SymbolTableOffset + NumSymbols * Nlist32Size;
- WriteSymtabLoadCommand(SymbolTableOffset, NumSymbols,
- StringTableOffset, StringTable.size());
-
- unsigned FirstLocalSymbol = 0;
- unsigned NumLocalSymbols = LocalSymbolData.size();
- unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
- unsigned NumExternalSymbols = ExternalSymbolData.size();
- unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
- unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
- // FIXME: Get correct symbol indices and counts for indirect symbols.
- unsigned IndirectSymbolOffset = 0;
- unsigned NumIndirectSymbols = 0;
- WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
- FirstExternalSymbol, NumExternalSymbols,
- FirstUndefinedSymbol, NumUndefinedSymbols,
- IndirectSymbolOffset, NumIndirectSymbols);
- }
+uint64_t MCAsmLayout::getSectionAddress(const MCSectionData *SD) const {
+ EnsureValid(SD->begin());
+ assert(SD->Address != ~UINT64_C(0) && "Address not set!");
+ return SD->Address;
+}
- // Write the actual section data.
- for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
- WriteFileData(OS, *it, *this);
+uint64_t MCAsmLayout::getSectionAddressSize(const MCSectionData *SD) const {
+ // The size is the last fragment's end offset.
+ const MCFragment &F = SD->getFragmentList().back();
+ return getFragmentOffset(&F) + getFragmentEffectiveSize(&F);
+}
- // Write the symbol table data, if used.
- if (NumSymbols) {
- // FIXME: Check that offsets match computed ones.
+uint64_t MCAsmLayout::getSectionFileSize(const MCSectionData *SD) const {
+ // Virtual sections have no file size.
+ if (getAssembler().getBackend().isVirtualSection(SD->getSection()))
+ return 0;
- // FIXME: Some of these are ordered by name to help the linker.
+ // Otherwise, the file size is the same as the address space size.
+ return getSectionAddressSize(SD);
+}
- // Write the symbol table entries.
- for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
- WriteNlist32(LocalSymbolData[i]);
- for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
- WriteNlist32(ExternalSymbolData[i]);
- for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
- WriteNlist32(UndefinedSymbolData[i]);
+uint64_t MCAsmLayout::getSectionSize(const MCSectionData *SD) const {
+ // The logical size is the address space size minus any tail padding.
+ uint64_t Size = getSectionAddressSize(SD);
+ const MCAlignFragment *AF =
+ dyn_cast<MCAlignFragment>(&(SD->getFragmentList().back()));
+ if (AF && AF->hasOnlyAlignAddress())
+ Size -= getFragmentEffectiveSize(AF);
- // Write the string table.
- OS << StringTable.str();
- }
- }
-};
+ return Size;
+}
/* *** */
MCFragment::MCFragment() : Kind(FragmentType(~0)) {
}
-MCFragment::MCFragment(FragmentType _Kind, MCSectionData *SD)
- : Kind(_Kind),
- FileSize(~UINT64_C(0))
+MCFragment::MCFragment(FragmentType _Kind, MCSectionData *_Parent)
+ : Kind(_Kind), Parent(_Parent), Atom(0), EffectiveSize(~UINT64_C(0))
{
- if (SD)
- SD->getFragmentList().push_back(this);
-}
-
-MCFragment::~MCFragment() {
+ if (Parent)
+ Parent->getFragmentList().push_back(this);
}
/* *** */
-MCSectionData::MCSectionData() : Section(*(MCSection*)0) {}
+MCSectionData::MCSectionData() : Section(0) {}
MCSectionData::MCSectionData(const MCSection &_Section, MCAssembler *A)
- : Section(_Section),
+ : Section(&_Section),
Alignment(1),
- FileSize(~UINT64_C(0))
+ Address(~UINT64_C(0)),
+ HasInstructions(false)
{
if (A)
A->getSectionList().push_back(this);
/* *** */
-MCSymbolData::MCSymbolData() : Symbol(*(MCSymbol*)0) {}
+MCSymbolData::MCSymbolData() : Symbol(0) {}
-MCSymbolData::MCSymbolData(MCSymbol &_Symbol, MCFragment *_Fragment,
+MCSymbolData::MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment,
uint64_t _Offset, MCAssembler *A)
- : Symbol(_Symbol), Fragment(_Fragment), Offset(_Offset),
- IsExternal(false)
+ : Symbol(&_Symbol), Fragment(_Fragment), Offset(_Offset),
+ IsExternal(false), IsPrivateExtern(false),
+ CommonSize(0), CommonAlign(0), Flags(0), Index(0)
{
if (A)
A->getSymbolList().push_back(this);
/* *** */
-MCAssembler::MCAssembler(raw_ostream &_OS) : OS(_OS) {}
+MCAssembler::MCAssembler(MCContext &_Context, TargetAsmBackend &_Backend,
+ MCCodeEmitter &_Emitter, raw_ostream &_OS)
+ : Context(_Context), Backend(_Backend), Emitter(_Emitter),
+ OS(_OS), RelaxAll(false), SubsectionsViaSymbols(false)
+{
+}
MCAssembler::~MCAssembler() {
}
-void MCAssembler::LayoutSection(MCSectionData &SD) {
- uint64_t Offset = 0;
+static bool isScatteredFixupFullyResolvedSimple(const MCAssembler &Asm,
+ const MCFixup &Fixup,
+ const MCValue Target,
+ const MCSection *BaseSection) {
+ // The effective fixup address is
+ // addr(atom(A)) + offset(A)
+ // - addr(atom(B)) - offset(B)
+ // - addr(<base symbol>) + <fixup offset from base symbol>
+ // and the offsets are not relocatable, so the fixup is fully resolved when
+ // addr(atom(A)) - addr(atom(B)) - addr(<base symbol>)) == 0.
+ //
+ // The simple (Darwin, except on x86_64) way of dealing with this was to
+ // assume that any reference to a temporary symbol *must* be a temporary
+ // symbol in the same atom, unless the sections differ. Therefore, any PCrel
+ // relocation to a temporary symbol (in the same section) is fully
+ // resolved. This also works in conjunction with absolutized .set, which
+ // requires the compiler to use .set to absolutize the differences between
+ // symbols which the compiler knows to be assembly time constants, so we don't
+ // need to worry about considering symbol differences fully resolved.
+
+ // Non-relative fixups are only resolved if constant.
+ if (!BaseSection)
+ return Target.isAbsolute();
+
+ // Otherwise, relative fixups are only resolved if not a difference and the
+ // target is a temporary in the same section.
+ if (Target.isAbsolute() || Target.getSymB())
+ return false;
+
+ const MCSymbol *A = &Target.getSymA()->getSymbol();
+ if (!A->isTemporary() || !A->isInSection() ||
+ &A->getSection() != BaseSection)
+ return false;
+
+ return true;
+}
- for (MCSectionData::iterator it = SD.begin(), ie = SD.end(); it != ie; ++it) {
- MCFragment &F = *it;
+static bool isScatteredFixupFullyResolved(const MCAssembler &Asm,
+ const MCAsmLayout &Layout,
+ const MCFixup &Fixup,
+ const MCValue Target,
+ const MCSymbolData *BaseSymbol) {
+ // The effective fixup address is
+ // addr(atom(A)) + offset(A)
+ // - addr(atom(B)) - offset(B)
+ // - addr(BaseSymbol) + <fixup offset from base symbol>
+ // and the offsets are not relocatable, so the fixup is fully resolved when
+ // addr(atom(A)) - addr(atom(B)) - addr(BaseSymbol) == 0.
+ //
+ // Note that "false" is almost always conservatively correct (it means we emit
+ // a relocation which is unnecessary), except when it would force us to emit a
+ // relocation which the target cannot encode.
+
+ const MCSymbolData *A_Base = 0, *B_Base = 0;
+ if (const MCSymbolRefExpr *A = Target.getSymA()) {
+ // Modified symbol references cannot be resolved.
+ if (A->getKind() != MCSymbolRefExpr::VK_None)
+ return false;
+
+ A_Base = Asm.getAtom(Layout, &Asm.getSymbolData(A->getSymbol()));
+ if (!A_Base)
+ return false;
+ }
- F.setOffset(Offset);
+ if (const MCSymbolRefExpr *B = Target.getSymB()) {
+ // Modified symbol references cannot be resolved.
+ if (B->getKind() != MCSymbolRefExpr::VK_None)
+ return false;
- // Evaluate fragment size.
- switch (F.getKind()) {
- case MCFragment::FT_Align: {
- MCAlignFragment &AF = cast<MCAlignFragment>(F);
-
- uint64_t AlignedOffset = RoundUpToAlignment(Offset, AF.getAlignment());
- uint64_t PaddingBytes = AlignedOffset - Offset;
+ B_Base = Asm.getAtom(Layout, &Asm.getSymbolData(B->getSymbol()));
+ if (!B_Base)
+ return false;
+ }
- if (PaddingBytes > AF.getMaxBytesToEmit())
- AF.setFileSize(0);
- else
- AF.setFileSize(PaddingBytes);
- break;
- }
+ // If there is no base, A and B have to be the same atom for this fixup to be
+ // fully resolved.
+ if (!BaseSymbol)
+ return A_Base == B_Base;
- case MCFragment::FT_Data:
- case MCFragment::FT_Fill:
- F.setFileSize(F.getMaxFileSize());
- break;
+ // Otherwise, B must be missing and A must be the base.
+ return !B_Base && BaseSymbol == A_Base;
+}
- case MCFragment::FT_Org: {
- MCOrgFragment &OF = cast<MCOrgFragment>(F);
+bool MCAssembler::isSymbolLinkerVisible(const MCSymbol &Symbol) const {
+ // Non-temporary labels should always be visible to the linker.
+ if (!Symbol.isTemporary())
+ return true;
- if (!OF.getOffset().isAbsolute())
- llvm_unreachable("FIXME: Not yet implemented!");
- uint64_t OrgOffset = OF.getOffset().getConstant();
+ // Absolute temporary labels are never visible.
+ if (!Symbol.isInSection())
+ return false;
- // FIXME: We need a way to communicate this error.
- if (OrgOffset < Offset)
- llvm_report_error("invalid .org offset '" + Twine(OrgOffset) +
- "' (section offset '" + Twine(Offset) + "'");
-
- F.setFileSize(OrgOffset - Offset);
- break;
- }
+ // Otherwise, check if the section requires symbols even for temporary labels.
+ return getBackend().doesSectionRequireSymbols(Symbol.getSection());
+}
+
+const MCSymbolData *MCAssembler::getAtom(const MCAsmLayout &Layout,
+ const MCSymbolData *SD) const {
+ // Linker visible symbols define atoms.
+ if (isSymbolLinkerVisible(SD->getSymbol()))
+ return SD;
+
+ // Absolute and undefined symbols have no defining atom.
+ if (!SD->getFragment())
+ return 0;
+
+ // Non-linker visible symbols in sections which can't be atomized have no
+ // defining atom.
+ if (!getBackend().isSectionAtomizable(
+ SD->getFragment()->getParent()->getSection()))
+ return 0;
+
+ // Otherwise, return the atom for the containing fragment.
+ return SD->getFragment()->getAtom();
+}
+
+bool MCAssembler::EvaluateFixup(const MCAsmLayout &Layout,
+ const MCFixup &Fixup, const MCFragment *DF,
+ MCValue &Target, uint64_t &Value) const {
+ ++stats::EvaluateFixup;
+
+ if (!Fixup.getValue()->EvaluateAsRelocatable(Target, &Layout))
+ report_fatal_error("expected relocatable expression");
+
+ // FIXME: How do non-scattered symbols work in ELF? I presume the linker
+ // doesn't support small relocations, but then under what criteria does the
+ // assembler allow symbol differences?
+
+ Value = Target.getConstant();
+
+ bool IsPCRel = Emitter.getFixupKindInfo(
+ Fixup.getKind()).Flags & MCFixupKindInfo::FKF_IsPCRel;
+ bool IsResolved = true;
+ if (const MCSymbolRefExpr *A = Target.getSymA()) {
+ if (A->getSymbol().isDefined())
+ Value += Layout.getSymbolAddress(&getSymbolData(A->getSymbol()));
+ else
+ IsResolved = false;
+ }
+ if (const MCSymbolRefExpr *B = Target.getSymB()) {
+ if (B->getSymbol().isDefined())
+ Value -= Layout.getSymbolAddress(&getSymbolData(B->getSymbol()));
+ else
+ IsResolved = false;
+ }
+
+ // If we are using scattered symbols, determine whether this value is actually
+ // resolved; scattering may cause atoms to move.
+ if (IsResolved && getBackend().hasScatteredSymbols()) {
+ if (getBackend().hasReliableSymbolDifference()) {
+ // If this is a PCrel relocation, find the base atom (identified by its
+ // symbol) that the fixup value is relative to.
+ const MCSymbolData *BaseSymbol = 0;
+ if (IsPCRel) {
+ BaseSymbol = DF->getAtom();
+ if (!BaseSymbol)
+ IsResolved = false;
+ }
+
+ if (IsResolved)
+ IsResolved = isScatteredFixupFullyResolved(*this, Layout, Fixup, Target,
+ BaseSymbol);
+ } else {
+ const MCSection *BaseSection = 0;
+ if (IsPCRel)
+ BaseSection = &DF->getParent()->getSection();
+
+ IsResolved = isScatteredFixupFullyResolvedSimple(*this, Fixup, Target,
+ BaseSection);
}
+ }
+
+ if (IsPCRel)
+ Value -= Layout.getFragmentAddress(DF) + Fixup.getOffset();
+
+ return IsResolved;
+}
+
+uint64_t MCAssembler::ComputeFragmentSize(MCAsmLayout &Layout,
+ const MCFragment &F,
+ uint64_t SectionAddress,
+ uint64_t FragmentOffset) const {
+ switch (F.getKind()) {
+ case MCFragment::FT_Data:
+ return cast<MCDataFragment>(F).getContents().size();
+ case MCFragment::FT_Fill:
+ return cast<MCFillFragment>(F).getSize();
+ case MCFragment::FT_Inst:
+ return cast<MCInstFragment>(F).getInstSize();
+
+ case MCFragment::FT_Align: {
+ const MCAlignFragment &AF = cast<MCAlignFragment>(F);
- Offset += F.getFileSize();
+ assert((!AF.hasOnlyAlignAddress() || !AF.getNextNode()) &&
+ "Invalid OnlyAlignAddress bit, not the last fragment!");
+
+ uint64_t Size = OffsetToAlignment(SectionAddress + FragmentOffset,
+ AF.getAlignment());
+
+ // Honor MaxBytesToEmit.
+ if (Size > AF.getMaxBytesToEmit())
+ return 0;
+
+ return Size;
}
- // FIXME: Pad section?
- SD.setFileSize(Offset);
+ case MCFragment::FT_Org: {
+ const MCOrgFragment &OF = cast<MCOrgFragment>(F);
+
+ // FIXME: We should compute this sooner, we don't want to recurse here, and
+ // we would like to be more functional.
+ int64_t TargetLocation;
+ if (!OF.getOffset().EvaluateAsAbsolute(TargetLocation, &Layout))
+ report_fatal_error("expected assembly-time absolute expression");
+
+ // FIXME: We need a way to communicate this error.
+ int64_t Offset = TargetLocation - FragmentOffset;
+ if (Offset < 0)
+ report_fatal_error("invalid .org offset '" + Twine(TargetLocation) +
+ "' (at offset '" + Twine(FragmentOffset) + "'");
+
+ return Offset;
+ }
+ }
+
+ assert(0 && "invalid fragment kind");
+ return 0;
}
-/// WriteFileData - Write the \arg F data to the output file.
-static void WriteFileData(raw_ostream &OS, const MCFragment &F,
- MachObjectWriter &MOW) {
- uint64_t Start = OS.tell();
+void MCAsmLayout::LayoutFile() {
+ // Initialize the first section and set the valid fragment layout point. All
+ // actual layout computations are done lazily.
+ LastValidFragment = 0;
+ if (!getSectionOrder().empty())
+ getSectionOrder().front()->Address = 0;
+}
+
+void MCAsmLayout::LayoutFragment(MCFragment *F) {
+ MCFragment *Prev = F->getPrevNode();
+
+ // We should never try to recompute something which is up-to-date.
+ assert(!isFragmentUpToDate(F) && "Attempt to recompute up-to-date fragment!");
+ // We should never try to compute the fragment layout if the section isn't
+ // up-to-date.
+ assert(isSectionUpToDate(F->getParent()) &&
+ "Attempt to compute fragment before it's section!");
+ // We should never try to compute the fragment layout if it's predecessor
+ // isn't up-to-date.
+ assert((!Prev || isFragmentUpToDate(Prev)) &&
+ "Attempt to compute fragment before it's predecessor!");
+
+ ++stats::FragmentLayouts;
+
+ // Compute the fragment start address.
+ uint64_t StartAddress = F->getParent()->Address;
+ uint64_t Address = StartAddress;
+ if (Prev)
+ Address += Prev->Offset + Prev->EffectiveSize;
+
+ // Compute fragment offset and size.
+ F->Offset = Address - StartAddress;
+ F->EffectiveSize = getAssembler().ComputeFragmentSize(*this, *F, StartAddress,
+ F->Offset);
+ LastValidFragment = F;
+
+ // If this is the last fragment in a section, update the next section address.
+ if (!F->getNextNode()) {
+ unsigned NextIndex = F->getParent()->getLayoutOrder() + 1;
+ if (NextIndex != getSectionOrder().size())
+ LayoutSection(getSectionOrder()[NextIndex]);
+ }
+}
+
+void MCAsmLayout::LayoutSection(MCSectionData *SD) {
+ unsigned SectionOrderIndex = SD->getLayoutOrder();
+
+ ++stats::SectionLayouts;
+
+ // Compute the section start address.
+ uint64_t StartAddress = 0;
+ if (SectionOrderIndex) {
+ MCSectionData *Prev = getSectionOrder()[SectionOrderIndex - 1];
+ StartAddress = getSectionAddress(Prev) + getSectionAddressSize(Prev);
+ }
+
+ // Honor the section alignment requirements.
+ StartAddress = RoundUpToAlignment(StartAddress, SD->getAlignment());
+
+ // Set the section address.
+ SD->Address = StartAddress;
+}
+
+/// WriteFragmentData - Write the \arg F data to the output file.
+static void WriteFragmentData(const MCAssembler &Asm, const MCAsmLayout &Layout,
+ const MCFragment &F, MCObjectWriter *OW) {
+ uint64_t Start = OW->getStream().tell();
(void) Start;
-
+
+ ++stats::EmittedFragments;
+
// FIXME: Embed in fragments instead?
+ uint64_t FragmentSize = Layout.getFragmentEffectiveSize(&F);
switch (F.getKind()) {
case MCFragment::FT_Align: {
MCAlignFragment &AF = cast<MCAlignFragment>(F);
- uint64_t Count = AF.getFileSize() / AF.getValueSize();
+ uint64_t Count = FragmentSize / AF.getValueSize();
+
+ assert(AF.getValueSize() && "Invalid virtual align in concrete fragment!");
// FIXME: This error shouldn't actually occur (the front end should emit
// multiple .align directives to enforce the semantics it wants), but is
// severe enough that we want to report it. How to handle this?
- if (Count * AF.getValueSize() != AF.getFileSize())
- llvm_report_error("undefined .align directive, value size '" +
- Twine(AF.getValueSize()) +
+ if (Count * AF.getValueSize() != FragmentSize)
+ report_fatal_error("undefined .align directive, value size '" +
+ Twine(AF.getValueSize()) +
"' is not a divisor of padding size '" +
- Twine(AF.getFileSize()) + "'");
+ Twine(FragmentSize) + "'");
+
+ // See if we are aligning with nops, and if so do that first to try to fill
+ // the Count bytes. Then if that did not fill any bytes or there are any
+ // bytes left to fill use the the Value and ValueSize to fill the rest.
+ // If we are aligning with nops, ask that target to emit the right data.
+ if (AF.hasEmitNops()) {
+ if (!Asm.getBackend().WriteNopData(Count, OW))
+ report_fatal_error("unable to write nop sequence of " +
+ Twine(Count) + " bytes");
+ break;
+ }
+ // Otherwise, write out in multiples of the value size.
for (uint64_t i = 0; i != Count; ++i) {
switch (AF.getValueSize()) {
default:
assert(0 && "Invalid size!");
- case 1: MOW.Write8 (uint8_t (AF.getValue())); break;
- case 2: MOW.Write16(uint16_t(AF.getValue())); break;
- case 4: MOW.Write32(uint32_t(AF.getValue())); break;
- case 8: MOW.Write64(uint64_t(AF.getValue())); break;
+ case 1: OW->Write8 (uint8_t (AF.getValue())); break;
+ case 2: OW->Write16(uint16_t(AF.getValue())); break;
+ case 4: OW->Write32(uint32_t(AF.getValue())); break;
+ case 8: OW->Write64(uint64_t(AF.getValue())); break;
}
}
break;
}
- case MCFragment::FT_Data:
- OS << cast<MCDataFragment>(F).getContents().str();
+ case MCFragment::FT_Data: {
+ MCDataFragment &DF = cast<MCDataFragment>(F);
+ assert(FragmentSize == DF.getContents().size() && "Invalid size!");
+ OW->WriteBytes(DF.getContents().str());
break;
+ }
case MCFragment::FT_Fill: {
MCFillFragment &FF = cast<MCFillFragment>(F);
- if (!FF.getValue().isAbsolute())
- llvm_unreachable("FIXME: Not yet implemented!");
- int64_t Value = FF.getValue().getConstant();
+ assert(FF.getValueSize() && "Invalid virtual align in concrete fragment!");
- for (uint64_t i = 0, e = FF.getCount(); i != e; ++i) {
+ for (uint64_t i = 0, e = FF.getSize() / FF.getValueSize(); i != e; ++i) {
switch (FF.getValueSize()) {
default:
assert(0 && "Invalid size!");
- case 1: MOW.Write8 (uint8_t (Value)); break;
- case 2: MOW.Write16(uint16_t(Value)); break;
- case 4: MOW.Write32(uint32_t(Value)); break;
- case 8: MOW.Write64(uint64_t(Value)); break;
+ case 1: OW->Write8 (uint8_t (FF.getValue())); break;
+ case 2: OW->Write16(uint16_t(FF.getValue())); break;
+ case 4: OW->Write32(uint32_t(FF.getValue())); break;
+ case 8: OW->Write64(uint64_t(FF.getValue())); break;
}
}
break;
}
-
+
+ case MCFragment::FT_Inst:
+ llvm_unreachable("unexpected inst fragment after lowering");
+ break;
+
case MCFragment::FT_Org: {
MCOrgFragment &OF = cast<MCOrgFragment>(F);
- for (uint64_t i = 0, e = OF.getFileSize(); i != e; ++i)
- MOW.Write8(uint8_t(OF.getValue()));
+ for (uint64_t i = 0, e = FragmentSize; i != e; ++i)
+ OW->Write8(uint8_t(OF.getValue()));
break;
}
}
- assert(OS.tell() - Start == F.getFileSize());
+ assert(OW->getStream().tell() - Start == FragmentSize);
}
-/// WriteFileData - Write the \arg SD data to the output file.
-static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
- MachObjectWriter &MOW) {
- uint64_t Start = OS.tell();
+void MCAssembler::WriteSectionData(const MCSectionData *SD,
+ const MCAsmLayout &Layout,
+ MCObjectWriter *OW) const {
+ // Ignore virtual sections.
+ if (getBackend().isVirtualSection(SD->getSection())) {
+ assert(Layout.getSectionFileSize(SD) == 0 && "Invalid size for section!");
+
+ // Check that contents are only things legal inside a virtual section.
+ for (MCSectionData::const_iterator it = SD->begin(),
+ ie = SD->end(); it != ie; ++it) {
+ switch (it->getKind()) {
+ default:
+ assert(0 && "Invalid fragment in virtual section!");
+ case MCFragment::FT_Align:
+ assert(!cast<MCAlignFragment>(it)->getValueSize() &&
+ "Invalid align in virtual section!");
+ break;
+ case MCFragment::FT_Fill:
+ assert(!cast<MCFillFragment>(it)->getValueSize() &&
+ "Invalid fill in virtual section!");
+ break;
+ }
+ }
+
+ return;
+ }
+
+ uint64_t Start = OW->getStream().tell();
(void) Start;
-
- for (MCSectionData::const_iterator it = SD.begin(),
- ie = SD.end(); it != ie; ++it)
- WriteFileData(OS, *it, MOW);
- assert(OS.tell() - Start == SD.getFileSize());
+ for (MCSectionData::const_iterator it = SD->begin(),
+ ie = SD->end(); it != ie; ++it)
+ WriteFragmentData(*this, Layout, *it, OW);
+
+ assert(OW->getStream().tell() - Start == Layout.getSectionFileSize(SD));
}
void MCAssembler::Finish() {
- // Layout the sections and fragments.
- for (iterator it = begin(), ie = end(); it != ie; ++it)
- LayoutSection(*it);
+ DEBUG_WITH_TYPE("mc-dump", {
+ llvm::errs() << "assembler backend - pre-layout\n--\n";
+ dump(); });
+
+ // Create the layout object.
+ MCAsmLayout Layout(*this);
+
+ // Insert additional align fragments for concrete sections to explicitly pad
+ // the previous section to match their alignment requirements. This is for
+ // 'gas' compatibility, it shouldn't strictly be necessary.
+ //
+ // FIXME: This may be Mach-O specific.
+ for (unsigned i = 1, e = Layout.getSectionOrder().size(); i < e; ++i) {
+ MCSectionData *SD = Layout.getSectionOrder()[i];
+
+ // Ignore sections without alignment requirements.
+ unsigned Align = SD->getAlignment();
+ if (Align <= 1)
+ continue;
+
+ // Ignore virtual sections, they don't cause file size modifications.
+ if (getBackend().isVirtualSection(SD->getSection()))
+ continue;
+
+ // Otherwise, create a new align fragment at the end of the previous
+ // section.
+ MCAlignFragment *AF = new MCAlignFragment(Align, 0, 1, Align,
+ Layout.getSectionOrder()[i - 1]);
+ AF->setOnlyAlignAddress(true);
+ }
+
+ // Create dummy fragments and assign section ordinals.
+ unsigned SectionIndex = 0;
+ for (MCAssembler::iterator it = begin(), ie = end(); it != ie; ++it) {
+ // Create dummy fragments to eliminate any empty sections, this simplifies
+ // layout.
+ if (it->getFragmentList().empty())
+ new MCFillFragment(0, 1, 0, it);
+
+ it->setOrdinal(SectionIndex++);
+ }
+
+ // Assign layout order indices to sections and fragments.
+ unsigned FragmentIndex = 0;
+ for (unsigned i = 0, e = Layout.getSectionOrder().size(); i != e; ++i) {
+ MCSectionData *SD = Layout.getSectionOrder()[i];
+ SD->setLayoutOrder(i);
+
+ for (MCSectionData::iterator it2 = SD->begin(),
+ ie2 = SD->end(); it2 != ie2; ++it2)
+ it2->setLayoutOrder(FragmentIndex++);
+ }
+
+ // Layout until everything fits.
+ while (LayoutOnce(Layout))
+ continue;
+
+ DEBUG_WITH_TYPE("mc-dump", {
+ llvm::errs() << "assembler backend - post-relaxation\n--\n";
+ dump(); });
+
+ // Finalize the layout, including fragment lowering.
+ FinishLayout(Layout);
+
+ DEBUG_WITH_TYPE("mc-dump", {
+ llvm::errs() << "assembler backend - final-layout\n--\n";
+ dump(); });
+
+ uint64_t StartOffset = OS.tell();
+ llvm::OwningPtr<MCObjectWriter> Writer(getBackend().createObjectWriter(OS));
+ if (!Writer)
+ report_fatal_error("unable to create object writer!");
+
+ // Allow the object writer a chance to perform post-layout binding (for
+ // example, to set the index fields in the symbol data).
+ Writer->ExecutePostLayoutBinding(*this);
+
+ // Evaluate and apply the fixups, generating relocation entries as necessary.
+ for (MCAssembler::iterator it = begin(), ie = end(); it != ie; ++it) {
+ for (MCSectionData::iterator it2 = it->begin(),
+ ie2 = it->end(); it2 != ie2; ++it2) {
+ MCDataFragment *DF = dyn_cast<MCDataFragment>(it2);
+ if (!DF)
+ continue;
+
+ for (MCDataFragment::fixup_iterator it3 = DF->fixup_begin(),
+ ie3 = DF->fixup_end(); it3 != ie3; ++it3) {
+ MCFixup &Fixup = *it3;
+
+ // Evaluate the fixup.
+ MCValue Target;
+ uint64_t FixedValue;
+ if (!EvaluateFixup(Layout, Fixup, DF, Target, FixedValue)) {
+ // The fixup was unresolved, we need a relocation. Inform the object
+ // writer of the relocation, and give it an opportunity to adjust the
+ // fixup value if need be.
+ Writer->RecordRelocation(*this, Layout, DF, Fixup, Target,FixedValue);
+ }
+
+ getBackend().ApplyFixup(Fixup, *DF, FixedValue);
+ }
+ }
+ }
// Write the object file.
- MachObjectWriter MOW(OS);
- MOW.WriteObject(*this);
+ Writer->WriteObject(*this, Layout);
+
+ stats::ObjectBytes += OS.tell() - StartOffset;
+}
+
+bool MCAssembler::FixupNeedsRelaxation(const MCFixup &Fixup,
+ const MCFragment *DF,
+ const MCAsmLayout &Layout) const {
+ if (getRelaxAll())
+ return true;
+
+ // If we cannot resolve the fixup value, it requires relaxation.
+ MCValue Target;
+ uint64_t Value;
+ if (!EvaluateFixup(Layout, Fixup, DF, Target, Value))
+ return true;
+
+ // Otherwise, relax if the value is too big for a (signed) i8.
+ //
+ // FIXME: This is target dependent!
+ return int64_t(Value) != int64_t(int8_t(Value));
+}
+
+bool MCAssembler::FragmentNeedsRelaxation(const MCInstFragment *IF,
+ const MCAsmLayout &Layout) const {
+ // If this inst doesn't ever need relaxation, ignore it. This occurs when we
+ // are intentionally pushing out inst fragments, or because we relaxed a
+ // previous instruction to one that doesn't need relaxation.
+ if (!getBackend().MayNeedRelaxation(IF->getInst()))
+ return false;
+
+ for (MCInstFragment::const_fixup_iterator it = IF->fixup_begin(),
+ ie = IF->fixup_end(); it != ie; ++it)
+ if (FixupNeedsRelaxation(*it, IF, Layout))
+ return true;
+
+ return false;
+}
- OS.flush();
+bool MCAssembler::LayoutOnce(MCAsmLayout &Layout) {
+ ++stats::RelaxationSteps;
+
+ // Layout the sections in order.
+ Layout.LayoutFile();
+
+ // Scan for fragments that need relaxation.
+ bool WasRelaxed = false;
+ for (iterator it = begin(), ie = end(); it != ie; ++it) {
+ MCSectionData &SD = *it;
+
+ for (MCSectionData::iterator it2 = SD.begin(),
+ ie2 = SD.end(); it2 != ie2; ++it2) {
+ // Check if this is an instruction fragment that needs relaxation.
+ MCInstFragment *IF = dyn_cast<MCInstFragment>(it2);
+ if (!IF || !FragmentNeedsRelaxation(IF, Layout))
+ continue;
+
+ ++stats::RelaxedInstructions;
+
+ // FIXME-PERF: We could immediately lower out instructions if we can tell
+ // they are fully resolved, to avoid retesting on later passes.
+
+ // Relax the fragment.
+
+ MCInst Relaxed;
+ getBackend().RelaxInstruction(IF->getInst(), Relaxed);
+
+ // Encode the new instruction.
+ //
+ // FIXME-PERF: If it matters, we could let the target do this. It can
+ // probably do so more efficiently in many cases.
+ SmallVector<MCFixup, 4> Fixups;
+ SmallString<256> Code;
+ raw_svector_ostream VecOS(Code);
+ getEmitter().EncodeInstruction(Relaxed, VecOS, Fixups);
+ VecOS.flush();
+
+ // Update the instruction fragment.
+ int SlideAmount = Code.size() - IF->getInstSize();
+ IF->setInst(Relaxed);
+ IF->getCode() = Code;
+ IF->getFixups().clear();
+ // FIXME: Eliminate copy.
+ for (unsigned i = 0, e = Fixups.size(); i != e; ++i)
+ IF->getFixups().push_back(Fixups[i]);
+
+ // Update the layout, and remember that we relaxed.
+ Layout.UpdateForSlide(IF, SlideAmount);
+ WasRelaxed = true;
+ }
+ }
+
+ return WasRelaxed;
+}
+
+void MCAssembler::FinishLayout(MCAsmLayout &Layout) {
+ // Lower out any instruction fragments, to simplify the fixup application and
+ // output.
+ //
+ // FIXME-PERF: We don't have to do this, but the assumption is that it is
+ // cheap (we will mostly end up eliminating fragments and appending on to data
+ // fragments), so the extra complexity downstream isn't worth it. Evaluate
+ // this assumption.
+ for (iterator it = begin(), ie = end(); it != ie; ++it) {
+ MCSectionData &SD = *it;
+
+ for (MCSectionData::iterator it2 = SD.begin(),
+ ie2 = SD.end(); it2 != ie2; ++it2) {
+ MCInstFragment *IF = dyn_cast<MCInstFragment>(it2);
+ if (!IF)
+ continue;
+
+ // Create a new data fragment for the instruction.
+ //
+ // FIXME-PERF: Reuse previous data fragment if possible.
+ MCDataFragment *DF = new MCDataFragment();
+ SD.getFragmentList().insert(it2, DF);
+
+ // Update the data fragments layout data.
+ DF->setParent(IF->getParent());
+ DF->setAtom(IF->getAtom());
+ DF->setLayoutOrder(IF->getLayoutOrder());
+ Layout.FragmentReplaced(IF, DF);
+
+ // Copy in the data and the fixups.
+ DF->getContents().append(IF->getCode().begin(), IF->getCode().end());
+ for (unsigned i = 0, e = IF->getFixups().size(); i != e; ++i)
+ DF->getFixups().push_back(IF->getFixups()[i]);
+
+ // Delete the instruction fragment and update the iterator.
+ SD.getFragmentList().erase(IF);
+ it2 = DF;
+ }
+ }
+}
+
+// Debugging methods
+
+namespace llvm {
+
+raw_ostream &operator<<(raw_ostream &OS, const MCFixup &AF) {
+ OS << "<MCFixup" << " Offset:" << AF.getOffset()
+ << " Value:" << *AF.getValue()
+ << " Kind:" << AF.getKind() << ">";
+ return OS;
+}
+
+}
+
+void MCFragment::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<";
+ switch (getKind()) {
+ case MCFragment::FT_Align: OS << "MCAlignFragment"; break;
+ case MCFragment::FT_Data: OS << "MCDataFragment"; break;
+ case MCFragment::FT_Fill: OS << "MCFillFragment"; break;
+ case MCFragment::FT_Inst: OS << "MCInstFragment"; break;
+ case MCFragment::FT_Org: OS << "MCOrgFragment"; break;
+ }
+
+ OS << "<MCFragment " << (void*) this << " LayoutOrder:" << LayoutOrder
+ << " Offset:" << Offset << " EffectiveSize:" << EffectiveSize << ">";
+
+ switch (getKind()) {
+ case MCFragment::FT_Align: {
+ const MCAlignFragment *AF = cast<MCAlignFragment>(this);
+ if (AF->hasEmitNops())
+ OS << " (emit nops)";
+ if (AF->hasOnlyAlignAddress())
+ OS << " (only align section)";
+ OS << "\n ";
+ OS << " Alignment:" << AF->getAlignment()
+ << " Value:" << AF->getValue() << " ValueSize:" << AF->getValueSize()
+ << " MaxBytesToEmit:" << AF->getMaxBytesToEmit() << ">";
+ break;
+ }
+ case MCFragment::FT_Data: {
+ const MCDataFragment *DF = cast<MCDataFragment>(this);
+ OS << "\n ";
+ OS << " Contents:[";
+ const SmallVectorImpl<char> &Contents = DF->getContents();
+ for (unsigned i = 0, e = Contents.size(); i != e; ++i) {
+ if (i) OS << ",";
+ OS << hexdigit((Contents[i] >> 4) & 0xF) << hexdigit(Contents[i] & 0xF);
+ }
+ OS << "] (" << Contents.size() << " bytes)";
+
+ if (!DF->getFixups().empty()) {
+ OS << ",\n ";
+ OS << " Fixups:[";
+ for (MCDataFragment::const_fixup_iterator it = DF->fixup_begin(),
+ ie = DF->fixup_end(); it != ie; ++it) {
+ if (it != DF->fixup_begin()) OS << ",\n ";
+ OS << *it;
+ }
+ OS << "]";
+ }
+ break;
+ }
+ case MCFragment::FT_Fill: {
+ const MCFillFragment *FF = cast<MCFillFragment>(this);
+ OS << " Value:" << FF->getValue() << " ValueSize:" << FF->getValueSize()
+ << " Size:" << FF->getSize();
+ break;
+ }
+ case MCFragment::FT_Inst: {
+ const MCInstFragment *IF = cast<MCInstFragment>(this);
+ OS << "\n ";
+ OS << " Inst:";
+ IF->getInst().dump_pretty(OS);
+ break;
+ }
+ case MCFragment::FT_Org: {
+ const MCOrgFragment *OF = cast<MCOrgFragment>(this);
+ OS << "\n ";
+ OS << " Offset:" << OF->getOffset() << " Value:" << OF->getValue();
+ break;
+ }
+ }
+ OS << ">";
+}
+
+void MCSectionData::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCSectionData";
+ OS << " Alignment:" << getAlignment() << " Address:" << Address
+ << " Fragments:[\n ";
+ for (iterator it = begin(), ie = end(); it != ie; ++it) {
+ if (it != begin()) OS << ",\n ";
+ it->dump();
+ }
+ OS << "]>";
+}
+
+void MCSymbolData::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCSymbolData Symbol:" << getSymbol()
+ << " Fragment:" << getFragment() << " Offset:" << getOffset()
+ << " Flags:" << getFlags() << " Index:" << getIndex();
+ if (isCommon())
+ OS << " (common, size:" << getCommonSize()
+ << " align: " << getCommonAlignment() << ")";
+ if (isExternal())
+ OS << " (external)";
+ if (isPrivateExtern())
+ OS << " (private extern)";
+ OS << ">";
+}
+
+void MCAssembler::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCAssembler\n";
+ OS << " Sections:[\n ";
+ for (iterator it = begin(), ie = end(); it != ie; ++it) {
+ if (it != begin()) OS << ",\n ";
+ it->dump();
+ }
+ OS << "],\n";
+ OS << " Symbols:[";
+
+ for (symbol_iterator it = symbol_begin(), ie = symbol_end(); it != ie; ++it) {
+ if (it != symbol_begin()) OS << ",\n ";
+ it->dump();
+ }
+ OS << "]>\n";
}