1 //===- MCAssembler.h - Object File Generation -------------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #ifndef LLVM_MC_MCASSEMBLER_H
11 #define LLVM_MC_MCASSEMBLER_H
13 #include "llvm/ADT/DenseMap.h"
14 #include "llvm/ADT/SmallString.h"
15 #include "llvm/ADT/ilist.h"
16 #include "llvm/ADT/ilist_node.h"
17 #include "llvm/Support/Casting.h"
18 #include "llvm/MC/MCFixup.h"
19 #include "llvm/MC/MCInst.h"
20 #include "llvm/System/DataTypes.h"
21 #include <vector> // FIXME: Shouldn't be needed.
37 class TargetAsmBackend;
39 /// MCAsmFixup - Represent a fixed size region of bytes inside some fragment
40 /// which needs to be rewritten. This region will either be rewritten by the
41 /// assembler or cause a relocation entry to be generated.
43 // FIXME: This should probably just be merged with MCFixup.
46 /// Offset - The offset inside the fragment which needs to be rewritten.
49 /// Value - The expression to eventually write into the fragment.
52 /// Kind - The fixup kind.
56 MCAsmFixup(uint64_t _Offset, const MCExpr &_Value, MCFixupKind _Kind)
57 : Offset(_Offset), Value(&_Value), Kind(_Kind) {}
60 class MCFragment : public ilist_node<MCFragment> {
61 friend class MCAsmLayout;
63 MCFragment(const MCFragment&); // DO NOT IMPLEMENT
64 void operator=(const MCFragment&); // DO NOT IMPLEMENT
79 /// Parent - The data for the section this fragment is in.
80 MCSectionData *Parent;
82 /// Atom - The atom this fragment is in, as represented by it's defining
83 /// symbol. Atom's are only used by backends which set
84 /// \see MCAsmBackend::hasReliableSymbolDifference().
87 /// @name Assembler Backend Data
90 // FIXME: This could all be kept private to the assembler implementation.
92 /// Offset - The offset of this fragment in its section. This is ~0 until
96 /// EffectiveSize - The compute size of this section. This is ~0 until
98 uint64_t EffectiveSize;
100 /// Ordinal - The global index of this fragment. This is the index across all
101 /// sections, not just the parent section.
107 MCFragment(FragmentType _Kind, MCSectionData *_Parent = 0);
110 // Only for sentinel.
112 virtual ~MCFragment();
114 FragmentType getKind() const { return Kind; }
116 MCSectionData *getParent() const { return Parent; }
117 void setParent(MCSectionData *Value) { Parent = Value; }
119 MCSymbolData *getAtom() const { return Atom; }
120 void setAtom(MCSymbolData *Value) { Atom = Value; }
122 unsigned getOrdinal() const { return Ordinal; }
123 void setOrdinal(unsigned Value) { Ordinal = Value; }
125 static bool classof(const MCFragment *O) { return true; }
130 class MCDataFragment : public MCFragment {
131 SmallString<32> Contents;
133 /// Fixups - The list of fixups in this fragment.
134 std::vector<MCAsmFixup> Fixups;
137 typedef std::vector<MCAsmFixup>::const_iterator const_fixup_iterator;
138 typedef std::vector<MCAsmFixup>::iterator fixup_iterator;
141 MCDataFragment(MCSectionData *SD = 0) : MCFragment(FT_Data, SD) {}
146 SmallString<32> &getContents() { return Contents; }
147 const SmallString<32> &getContents() const { return Contents; }
150 /// @name Fixup Access
153 void addFixup(MCAsmFixup Fixup) {
154 // Enforce invariant that fixups are in offset order.
155 assert((Fixups.empty() || Fixup.Offset > Fixups.back().Offset) &&
156 "Fixups must be added in order!");
157 Fixups.push_back(Fixup);
160 std::vector<MCAsmFixup> &getFixups() { return Fixups; }
161 const std::vector<MCAsmFixup> &getFixups() const { return Fixups; }
163 fixup_iterator fixup_begin() { return Fixups.begin(); }
164 const_fixup_iterator fixup_begin() const { return Fixups.begin(); }
166 fixup_iterator fixup_end() {return Fixups.end();}
167 const_fixup_iterator fixup_end() const {return Fixups.end();}
169 size_t fixup_size() const { return Fixups.size(); }
173 static bool classof(const MCFragment *F) {
174 return F->getKind() == MCFragment::FT_Data;
176 static bool classof(const MCDataFragment *) { return true; }
181 // FIXME: This current incarnation of MCInstFragment doesn't make much sense, as
182 // it is almost entirely a duplicate of MCDataFragment. If we decide to stick
183 // with this approach (as opposed to making MCInstFragment a very light weight
184 // object with just the MCInst and a code size, then we should just change
185 // MCDataFragment to have an optional MCInst at its end.
186 class MCInstFragment : public MCFragment {
187 /// Inst - The instruction this is a fragment for.
190 /// InstSize - The size of the currently encoded instruction.
193 /// Fixups - The list of fixups in this fragment.
194 SmallVector<MCAsmFixup, 1> Fixups;
197 typedef SmallVectorImpl<MCAsmFixup>::const_iterator const_fixup_iterator;
198 typedef SmallVectorImpl<MCAsmFixup>::iterator fixup_iterator;
201 MCInstFragment(MCInst _Inst, MCSectionData *SD = 0)
202 : MCFragment(FT_Inst, SD), Inst(_Inst) {
208 SmallVectorImpl<char> &getCode() { return Code; }
209 const SmallVectorImpl<char> &getCode() const { return Code; }
211 unsigned getInstSize() const { return Code.size(); }
213 MCInst &getInst() { return Inst; }
214 const MCInst &getInst() const { return Inst; }
216 void setInst(MCInst Value) { Inst = Value; }
219 /// @name Fixup Access
222 SmallVectorImpl<MCAsmFixup> &getFixups() { return Fixups; }
223 const SmallVectorImpl<MCAsmFixup> &getFixups() const { return Fixups; }
225 fixup_iterator fixup_begin() { return Fixups.begin(); }
226 const_fixup_iterator fixup_begin() const { return Fixups.begin(); }
228 fixup_iterator fixup_end() {return Fixups.end();}
229 const_fixup_iterator fixup_end() const {return Fixups.end();}
231 size_t fixup_size() const { return Fixups.size(); }
235 static bool classof(const MCFragment *F) {
236 return F->getKind() == MCFragment::FT_Inst;
238 static bool classof(const MCInstFragment *) { return true; }
243 class MCAlignFragment : public MCFragment {
244 /// Alignment - The alignment to ensure, in bytes.
247 /// Value - Value to use for filling padding bytes.
250 /// ValueSize - The size of the integer (in bytes) of \arg Value.
253 /// MaxBytesToEmit - The maximum number of bytes to emit; if the alignment
254 /// cannot be satisfied in this width then this fragment is ignored.
255 unsigned MaxBytesToEmit;
257 /// EmitNops - true when aligning code and optimal nops to be used for
262 MCAlignFragment(unsigned _Alignment, int64_t _Value, unsigned _ValueSize,
263 unsigned _MaxBytesToEmit, bool _EmitNops,
264 MCSectionData *SD = 0)
265 : MCFragment(FT_Align, SD), Alignment(_Alignment),
266 Value(_Value),ValueSize(_ValueSize),
267 MaxBytesToEmit(_MaxBytesToEmit), EmitNops(_EmitNops) {}
272 unsigned getAlignment() const { return Alignment; }
274 int64_t getValue() const { return Value; }
276 unsigned getValueSize() const { return ValueSize; }
278 unsigned getMaxBytesToEmit() const { return MaxBytesToEmit; }
280 unsigned getEmitNops() const { return EmitNops; }
284 static bool classof(const MCFragment *F) {
285 return F->getKind() == MCFragment::FT_Align;
287 static bool classof(const MCAlignFragment *) { return true; }
292 class MCFillFragment : public MCFragment {
293 /// Value - Value to use for filling bytes.
296 /// ValueSize - The size (in bytes) of \arg Value to use when filling, or 0 if
297 /// this is a virtual fill fragment.
300 /// Size - The number of bytes to insert.
304 MCFillFragment(int64_t _Value, unsigned _ValueSize, uint64_t _Size,
305 MCSectionData *SD = 0)
306 : MCFragment(FT_Fill, SD),
307 Value(_Value), ValueSize(_ValueSize), Size(_Size) {
308 assert((!ValueSize || (Size % ValueSize) == 0) &&
309 "Fill size must be a multiple of the value size!");
315 int64_t getValue() const { return Value; }
317 unsigned getValueSize() const { return ValueSize; }
319 uint64_t getSize() const { return Size; }
323 static bool classof(const MCFragment *F) {
324 return F->getKind() == MCFragment::FT_Fill;
326 static bool classof(const MCFillFragment *) { return true; }
331 class MCOrgFragment : public MCFragment {
332 /// Offset - The offset this fragment should start at.
333 const MCExpr *Offset;
335 /// Value - Value to use for filling bytes.
339 MCOrgFragment(const MCExpr &_Offset, int8_t _Value, MCSectionData *SD = 0)
340 : MCFragment(FT_Org, SD),
341 Offset(&_Offset), Value(_Value) {}
346 const MCExpr &getOffset() const { return *Offset; }
348 uint8_t getValue() const { return Value; }
352 static bool classof(const MCFragment *F) {
353 return F->getKind() == MCFragment::FT_Org;
355 static bool classof(const MCOrgFragment *) { return true; }
360 /// MCZeroFillFragment - Represent data which has a fixed size and alignment,
361 /// but requires no physical space in the object file.
362 class MCZeroFillFragment : public MCFragment {
363 /// Size - The size of this fragment.
367 MCZeroFillFragment(uint64_t _Size, MCSectionData *SD = 0)
368 : MCFragment(FT_ZeroFill, SD), Size(_Size) {}
373 uint64_t getSize() const { return Size; }
377 static bool classof(const MCFragment *F) {
378 return F->getKind() == MCFragment::FT_ZeroFill;
380 static bool classof(const MCZeroFillFragment *) { return true; }
385 // FIXME: Should this be a separate class, or just merged into MCSection? Since
386 // we anticipate the fast path being through an MCAssembler, the only reason to
387 // keep it out is for API abstraction.
388 class MCSectionData : public ilist_node<MCSectionData> {
389 friend class MCAsmLayout;
391 MCSectionData(const MCSectionData&); // DO NOT IMPLEMENT
392 void operator=(const MCSectionData&); // DO NOT IMPLEMENT
395 typedef iplist<MCFragment> FragmentListType;
397 typedef FragmentListType::const_iterator const_iterator;
398 typedef FragmentListType::iterator iterator;
400 typedef FragmentListType::const_reverse_iterator const_reverse_iterator;
401 typedef FragmentListType::reverse_iterator reverse_iterator;
404 iplist<MCFragment> Fragments;
405 const MCSection *Section;
407 /// Ordinal - The section index in the assemblers section list.
410 /// Alignment - The maximum alignment seen in this section.
413 /// @name Assembler Backend Data
416 // FIXME: This could all be kept private to the assembler implementation.
418 /// Address - The computed address of this section. This is ~0 until
422 /// Size - The content size of this section. This is ~0 until initialized.
425 /// FileSize - The size of this section in the object file. This is ~0 until
429 /// HasInstructions - Whether this section has had instructions emitted into
431 unsigned HasInstructions : 1;
436 // Only for use as sentinel.
438 MCSectionData(const MCSection &Section, MCAssembler *A = 0);
440 const MCSection &getSection() const { return *Section; }
442 unsigned getAlignment() const { return Alignment; }
443 void setAlignment(unsigned Value) { Alignment = Value; }
445 bool hasInstructions() const { return HasInstructions; }
446 void setHasInstructions(bool Value) { HasInstructions = Value; }
448 unsigned getOrdinal() const { return Ordinal; }
449 void setOrdinal(unsigned Value) { Ordinal = Value; }
451 /// @name Fragment Access
454 const FragmentListType &getFragmentList() const { return Fragments; }
455 FragmentListType &getFragmentList() { return Fragments; }
457 iterator begin() { return Fragments.begin(); }
458 const_iterator begin() const { return Fragments.begin(); }
460 iterator end() { return Fragments.end(); }
461 const_iterator end() const { return Fragments.end(); }
463 reverse_iterator rbegin() { return Fragments.rbegin(); }
464 const_reverse_iterator rbegin() const { return Fragments.rbegin(); }
466 reverse_iterator rend() { return Fragments.rend(); }
467 const_reverse_iterator rend() const { return Fragments.rend(); }
469 size_t size() const { return Fragments.size(); }
471 bool empty() const { return Fragments.empty(); }
478 // FIXME: Same concerns as with SectionData.
479 class MCSymbolData : public ilist_node<MCSymbolData> {
481 const MCSymbol *Symbol;
483 /// Fragment - The fragment this symbol's value is relative to, if any.
484 MCFragment *Fragment;
486 /// Offset - The offset to apply to the fragment address to form this symbol's
490 /// IsExternal - True if this symbol is visible outside this translation
492 unsigned IsExternal : 1;
494 /// IsPrivateExtern - True if this symbol is private extern.
495 unsigned IsPrivateExtern : 1;
497 /// CommonSize - The size of the symbol, if it is 'common', or 0.
499 // FIXME: Pack this in with other fields? We could put it in offset, since a
500 // common symbol can never get a definition.
503 /// CommonAlign - The alignment of the symbol, if it is 'common'.
505 // FIXME: Pack this in with other fields?
506 unsigned CommonAlign;
508 /// Flags - The Flags field is used by object file implementations to store
509 /// additional per symbol information which is not easily classified.
512 /// Index - Index field, for use by the object file implementation.
516 // Only for use as sentinel.
518 MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment, uint64_t _Offset,
524 const MCSymbol &getSymbol() const { return *Symbol; }
526 MCFragment *getFragment() const { return Fragment; }
527 void setFragment(MCFragment *Value) { Fragment = Value; }
529 uint64_t getOffset() const { return Offset; }
530 void setOffset(uint64_t Value) { Offset = Value; }
533 /// @name Symbol Attributes
536 bool isExternal() const { return IsExternal; }
537 void setExternal(bool Value) { IsExternal = Value; }
539 bool isPrivateExtern() const { return IsPrivateExtern; }
540 void setPrivateExtern(bool Value) { IsPrivateExtern = Value; }
542 /// isCommon - Is this a 'common' symbol.
543 bool isCommon() const { return CommonSize != 0; }
545 /// setCommon - Mark this symbol as being 'common'.
547 /// \param Size - The size of the symbol.
548 /// \param Align - The alignment of the symbol.
549 void setCommon(uint64_t Size, unsigned Align) {
554 /// getCommonSize - Return the size of a 'common' symbol.
555 uint64_t getCommonSize() const {
556 assert(isCommon() && "Not a 'common' symbol!");
560 /// getCommonAlignment - Return the alignment of a 'common' symbol.
561 unsigned getCommonAlignment() const {
562 assert(isCommon() && "Not a 'common' symbol!");
566 /// getFlags - Get the (implementation defined) symbol flags.
567 uint32_t getFlags() const { return Flags; }
569 /// setFlags - Set the (implementation defined) symbol flags.
570 void setFlags(uint32_t Value) { Flags = Value; }
572 /// modifyFlags - Modify the flags via a mask
573 void modifyFlags(uint32_t Value, uint32_t Mask) {
574 Flags = (Flags & ~Mask) | Value;
577 /// getIndex - Get the (implementation defined) index.
578 uint64_t getIndex() const { return Index; }
580 /// setIndex - Set the (implementation defined) index.
581 void setIndex(uint64_t Value) { Index = Value; }
588 // FIXME: This really doesn't belong here. See comments below.
589 struct IndirectSymbolData {
591 MCSectionData *SectionData;
595 friend class MCAsmLayout;
598 typedef iplist<MCSectionData> SectionDataListType;
599 typedef iplist<MCSymbolData> SymbolDataListType;
601 typedef SectionDataListType::const_iterator const_iterator;
602 typedef SectionDataListType::iterator iterator;
604 typedef SymbolDataListType::const_iterator const_symbol_iterator;
605 typedef SymbolDataListType::iterator symbol_iterator;
607 typedef std::vector<IndirectSymbolData>::const_iterator
608 const_indirect_symbol_iterator;
609 typedef std::vector<IndirectSymbolData>::iterator indirect_symbol_iterator;
612 MCAssembler(const MCAssembler&); // DO NOT IMPLEMENT
613 void operator=(const MCAssembler&); // DO NOT IMPLEMENT
617 TargetAsmBackend &Backend;
619 MCCodeEmitter &Emitter;
623 iplist<MCSectionData> Sections;
625 iplist<MCSymbolData> Symbols;
627 /// The map of sections to their associated assembler backend data.
629 // FIXME: Avoid this indirection?
630 DenseMap<const MCSection*, MCSectionData*> SectionMap;
632 /// The map of symbols to their associated assembler backend data.
634 // FIXME: Avoid this indirection?
635 DenseMap<const MCSymbol*, MCSymbolData*> SymbolMap;
637 std::vector<IndirectSymbolData> IndirectSymbols;
639 unsigned RelaxAll : 1;
640 unsigned SubsectionsViaSymbols : 1;
643 /// Evaluate a fixup to a relocatable expression and the value which should be
644 /// placed into the fixup.
646 /// \param Layout The layout to use for evaluation.
647 /// \param Fixup The fixup to evaluate.
648 /// \param DF The fragment the fixup is inside.
649 /// \param Target [out] On return, the relocatable expression the fixup
651 /// \param Value [out] On return, the value of the fixup as currently layed
653 /// \return Whether the fixup value was fully resolved. This is true if the
654 /// \arg Value result is fixed, otherwise the value may change due to
656 bool EvaluateFixup(const MCAsmLayout &Layout,
657 const MCAsmFixup &Fixup, const MCFragment *DF,
658 MCValue &Target, uint64_t &Value) const;
660 /// Check whether a fixup can be satisfied, or whether it needs to be relaxed
661 /// (increased in size, in order to hold its value correctly).
662 bool FixupNeedsRelaxation(const MCAsmFixup &Fixup, const MCFragment *DF,
663 const MCAsmLayout &Layout) const;
665 /// Check whether the given fragment needs relaxation.
666 bool FragmentNeedsRelaxation(const MCInstFragment *IF,
667 const MCAsmLayout &Layout) const;
669 /// LayoutFragment - Performs layout of the given \arg Fragment; assuming that
670 /// the previous fragment has already been layed out correctly, and the parent
671 /// section has been initialized.
672 void LayoutFragment(MCAsmLayout &Layout, MCFragment &Fragment);
674 /// LayoutSection - Performs layout of the section referenced by the given
675 /// \arg SectionOrderIndex. The layout assumes that the previous section has
676 /// already been layed out correctly.
677 void LayoutSection(MCAsmLayout &Layout, unsigned SectionOrderIndex);
679 /// LayoutOnce - Perform one layout iteration and return true if any offsets
681 bool LayoutOnce(MCAsmLayout &Layout);
683 /// FinishLayout - Finalize a layout, including fragment lowering.
684 void FinishLayout(MCAsmLayout &Layout);
687 /// Find the symbol which defines the atom containing the given symbol, or
688 /// null if there is no such symbol.
689 const MCSymbolData *getAtom(const MCAsmLayout &Layout,
690 const MCSymbolData *Symbol) const;
692 /// Check whether a particular symbol is visible to the linker and is required
693 /// in the symbol table, or whether it can be discarded by the assembler. This
694 /// also effects whether the assembler treats the label as potentially
695 /// defining a separate atom.
696 bool isSymbolLinkerVisible(const MCSymbolData *SD) const;
698 /// Emit the section contents using the given object writer.
700 // FIXME: Should MCAssembler always have a reference to the object writer?
701 void WriteSectionData(const MCSectionData *Section, const MCAsmLayout &Layout,
702 MCObjectWriter *OW) const;
705 /// Construct a new assembler instance.
707 /// \arg OS - The stream to output to.
709 // FIXME: How are we going to parameterize this? Two obvious options are stay
710 // concrete and require clients to pass in a target like object. The other
711 // option is to make this abstract, and have targets provide concrete
712 // implementations as we do with AsmParser.
713 MCAssembler(MCContext &_Context, TargetAsmBackend &_Backend,
714 MCCodeEmitter &_Emitter, raw_ostream &OS);
717 MCContext &getContext() const { return Context; }
719 TargetAsmBackend &getBackend() const { return Backend; }
721 MCCodeEmitter &getEmitter() const { return Emitter; }
723 /// Finish - Do final processing and write the object to the output stream.
726 // FIXME: This does not belong here.
727 bool getSubsectionsViaSymbols() const {
728 return SubsectionsViaSymbols;
730 void setSubsectionsViaSymbols(bool Value) {
731 SubsectionsViaSymbols = Value;
734 bool getRelaxAll() const { return RelaxAll; }
735 void setRelaxAll(bool Value) { RelaxAll = Value; }
737 /// @name Section List Access
740 const SectionDataListType &getSectionList() const { return Sections; }
741 SectionDataListType &getSectionList() { return Sections; }
743 iterator begin() { return Sections.begin(); }
744 const_iterator begin() const { return Sections.begin(); }
746 iterator end() { return Sections.end(); }
747 const_iterator end() const { return Sections.end(); }
749 size_t size() const { return Sections.size(); }
752 /// @name Symbol List Access
755 const SymbolDataListType &getSymbolList() const { return Symbols; }
756 SymbolDataListType &getSymbolList() { return Symbols; }
758 symbol_iterator symbol_begin() { return Symbols.begin(); }
759 const_symbol_iterator symbol_begin() const { return Symbols.begin(); }
761 symbol_iterator symbol_end() { return Symbols.end(); }
762 const_symbol_iterator symbol_end() const { return Symbols.end(); }
764 size_t symbol_size() const { return Symbols.size(); }
767 /// @name Indirect Symbol List Access
770 // FIXME: This is a total hack, this should not be here. Once things are
771 // factored so that the streamer has direct access to the .o writer, it can
773 std::vector<IndirectSymbolData> &getIndirectSymbols() {
774 return IndirectSymbols;
777 indirect_symbol_iterator indirect_symbol_begin() {
778 return IndirectSymbols.begin();
780 const_indirect_symbol_iterator indirect_symbol_begin() const {
781 return IndirectSymbols.begin();
784 indirect_symbol_iterator indirect_symbol_end() {
785 return IndirectSymbols.end();
787 const_indirect_symbol_iterator indirect_symbol_end() const {
788 return IndirectSymbols.end();
791 size_t indirect_symbol_size() const { return IndirectSymbols.size(); }
794 /// @name Backend Data Access
797 MCSectionData &getSectionData(const MCSection &Section) const {
798 MCSectionData *Entry = SectionMap.lookup(&Section);
799 assert(Entry && "Missing section data!");
803 MCSectionData &getOrCreateSectionData(const MCSection &Section,
805 MCSectionData *&Entry = SectionMap[&Section];
807 if (Created) *Created = !Entry;
809 Entry = new MCSectionData(Section, this);
814 MCSymbolData &getSymbolData(const MCSymbol &Symbol) const {
815 MCSymbolData *Entry = SymbolMap.lookup(&Symbol);
816 assert(Entry && "Missing symbol data!");
820 MCSymbolData &getOrCreateSymbolData(const MCSymbol &Symbol,
822 MCSymbolData *&Entry = SymbolMap[&Symbol];
824 if (Created) *Created = !Entry;
826 Entry = new MCSymbolData(Symbol, 0, 0, this);
836 } // end namespace llvm