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.
38 class TargetAsmBackend;
40 class MCFragment : public ilist_node<MCFragment> {
41 friend class MCAsmLayout;
43 MCFragment(const MCFragment&); // DO NOT IMPLEMENT
44 void operator=(const MCFragment&); // DO NOT IMPLEMENT
58 /// Parent - The data for the section this fragment is in.
59 MCSectionData *Parent;
61 /// Atom - The atom this fragment is in, as represented by it's defining
62 /// symbol. Atom's are only used by backends which set
63 /// \see MCAsmBackend::hasReliableSymbolDifference().
66 /// @name Assembler Backend Data
69 // FIXME: This could all be kept private to the assembler implementation.
71 /// Offset - The offset of this fragment in its section. This is ~0 until
75 /// EffectiveSize - The compute size of this section. This is ~0 until
77 uint64_t EffectiveSize;
79 /// LayoutOrder - The global layout order of this fragment. This is the index
80 /// across all fragments in the file, not just within the section.
86 MCFragment(FragmentType _Kind, MCSectionData *_Parent = 0);
91 virtual ~MCFragment();
93 FragmentType getKind() const { return Kind; }
95 MCSectionData *getParent() const { return Parent; }
96 void setParent(MCSectionData *Value) { Parent = Value; }
98 MCSymbolData *getAtom() const { return Atom; }
99 void setAtom(MCSymbolData *Value) { Atom = Value; }
101 unsigned getLayoutOrder() const { return LayoutOrder; }
102 void setLayoutOrder(unsigned Value) { LayoutOrder = Value; }
104 static bool classof(const MCFragment *O) { return true; }
109 class MCDataFragment : public MCFragment {
110 SmallString<32> Contents;
112 /// Fixups - The list of fixups in this fragment.
113 std::vector<MCFixup> Fixups;
116 typedef std::vector<MCFixup>::const_iterator const_fixup_iterator;
117 typedef std::vector<MCFixup>::iterator fixup_iterator;
120 MCDataFragment(MCSectionData *SD = 0) : MCFragment(FT_Data, SD) {}
125 SmallString<32> &getContents() { return Contents; }
126 const SmallString<32> &getContents() const { return Contents; }
129 /// @name Fixup Access
132 void addFixup(MCFixup Fixup) {
133 // Enforce invariant that fixups are in offset order.
134 assert((Fixups.empty() || Fixup.getOffset() > Fixups.back().getOffset()) &&
135 "Fixups must be added in order!");
136 Fixups.push_back(Fixup);
139 std::vector<MCFixup> &getFixups() { return Fixups; }
140 const std::vector<MCFixup> &getFixups() const { return Fixups; }
142 fixup_iterator fixup_begin() { return Fixups.begin(); }
143 const_fixup_iterator fixup_begin() const { return Fixups.begin(); }
145 fixup_iterator fixup_end() {return Fixups.end();}
146 const_fixup_iterator fixup_end() const {return Fixups.end();}
148 size_t fixup_size() const { return Fixups.size(); }
152 static bool classof(const MCFragment *F) {
153 return F->getKind() == MCFragment::FT_Data;
155 static bool classof(const MCDataFragment *) { return true; }
158 // FIXME: This current incarnation of MCInstFragment doesn't make much sense, as
159 // it is almost entirely a duplicate of MCDataFragment. If we decide to stick
160 // with this approach (as opposed to making MCInstFragment a very light weight
161 // object with just the MCInst and a code size, then we should just change
162 // MCDataFragment to have an optional MCInst at its end.
163 class MCInstFragment : public MCFragment {
164 /// Inst - The instruction this is a fragment for.
167 /// Code - Binary data for the currently encoded instruction.
170 /// Fixups - The list of fixups in this fragment.
171 SmallVector<MCFixup, 1> Fixups;
174 typedef SmallVectorImpl<MCFixup>::const_iterator const_fixup_iterator;
175 typedef SmallVectorImpl<MCFixup>::iterator fixup_iterator;
178 MCInstFragment(MCInst _Inst, MCSectionData *SD = 0)
179 : MCFragment(FT_Inst, SD), Inst(_Inst) {
185 SmallVectorImpl<char> &getCode() { return Code; }
186 const SmallVectorImpl<char> &getCode() const { return Code; }
188 unsigned getInstSize() const { return Code.size(); }
190 MCInst &getInst() { return Inst; }
191 const MCInst &getInst() const { return Inst; }
193 void setInst(MCInst Value) { Inst = Value; }
196 /// @name Fixup Access
199 SmallVectorImpl<MCFixup> &getFixups() { return Fixups; }
200 const SmallVectorImpl<MCFixup> &getFixups() const { return Fixups; }
202 fixup_iterator fixup_begin() { return Fixups.begin(); }
203 const_fixup_iterator fixup_begin() const { return Fixups.begin(); }
205 fixup_iterator fixup_end() {return Fixups.end();}
206 const_fixup_iterator fixup_end() const {return Fixups.end();}
208 size_t fixup_size() const { return Fixups.size(); }
212 static bool classof(const MCFragment *F) {
213 return F->getKind() == MCFragment::FT_Inst;
215 static bool classof(const MCInstFragment *) { return true; }
218 class MCAlignFragment : public MCFragment {
219 /// Alignment - The alignment to ensure, in bytes.
222 /// Value - Value to use for filling padding bytes.
225 /// ValueSize - The size of the integer (in bytes) of \arg Value.
228 /// MaxBytesToEmit - The maximum number of bytes to emit; if the alignment
229 /// cannot be satisfied in this width then this fragment is ignored.
230 unsigned MaxBytesToEmit;
232 /// EmitNops - Flag to indicate that (optimal) NOPs should be emitted instead
233 /// of using the provided value. The exact interpretation of this flag is
234 /// target dependent.
237 /// OnlyAlignAddress - Flag to indicate that this align is only used to adjust
238 /// the address space size of a section and that it should not be included as
239 /// part of the section size. This flag can only be used on the last fragment
241 bool OnlyAlignAddress : 1;
244 MCAlignFragment(unsigned _Alignment, int64_t _Value, unsigned _ValueSize,
245 unsigned _MaxBytesToEmit, MCSectionData *SD = 0)
246 : MCFragment(FT_Align, SD), Alignment(_Alignment),
247 Value(_Value),ValueSize(_ValueSize),
248 MaxBytesToEmit(_MaxBytesToEmit), EmitNops(false),
249 OnlyAlignAddress(false) {}
254 unsigned getAlignment() const { return Alignment; }
256 int64_t getValue() const { return Value; }
258 unsigned getValueSize() const { return ValueSize; }
260 unsigned getMaxBytesToEmit() const { return MaxBytesToEmit; }
262 bool hasEmitNops() const { return EmitNops; }
263 void setEmitNops(bool Value) { EmitNops = Value; }
265 bool hasOnlyAlignAddress() const { return OnlyAlignAddress; }
266 void setOnlyAlignAddress(bool Value) { OnlyAlignAddress = Value; }
270 static bool classof(const MCFragment *F) {
271 return F->getKind() == MCFragment::FT_Align;
273 static bool classof(const MCAlignFragment *) { return true; }
276 class MCFillFragment : public MCFragment {
277 /// Value - Value to use for filling bytes.
280 /// ValueSize - The size (in bytes) of \arg Value to use when filling, or 0 if
281 /// this is a virtual fill fragment.
284 /// Size - The number of bytes to insert.
288 MCFillFragment(int64_t _Value, unsigned _ValueSize, uint64_t _Size,
289 MCSectionData *SD = 0)
290 : MCFragment(FT_Fill, SD),
291 Value(_Value), ValueSize(_ValueSize), Size(_Size) {
292 assert((!ValueSize || (Size % ValueSize) == 0) &&
293 "Fill size must be a multiple of the value size!");
299 int64_t getValue() const { return Value; }
301 unsigned getValueSize() const { return ValueSize; }
303 uint64_t getSize() const { return Size; }
307 static bool classof(const MCFragment *F) {
308 return F->getKind() == MCFragment::FT_Fill;
310 static bool classof(const MCFillFragment *) { return true; }
313 class MCOrgFragment : public MCFragment {
314 /// Offset - The offset this fragment should start at.
315 const MCExpr *Offset;
317 /// Value - Value to use for filling bytes.
321 MCOrgFragment(const MCExpr &_Offset, int8_t _Value, MCSectionData *SD = 0)
322 : MCFragment(FT_Org, SD),
323 Offset(&_Offset), Value(_Value) {}
328 const MCExpr &getOffset() const { return *Offset; }
330 uint8_t getValue() const { return Value; }
334 static bool classof(const MCFragment *F) {
335 return F->getKind() == MCFragment::FT_Org;
337 static bool classof(const MCOrgFragment *) { return true; }
340 // FIXME: Should this be a separate class, or just merged into MCSection? Since
341 // we anticipate the fast path being through an MCAssembler, the only reason to
342 // keep it out is for API abstraction.
343 class MCSectionData : public ilist_node<MCSectionData> {
344 friend class MCAsmLayout;
346 MCSectionData(const MCSectionData&); // DO NOT IMPLEMENT
347 void operator=(const MCSectionData&); // DO NOT IMPLEMENT
350 typedef iplist<MCFragment> FragmentListType;
352 typedef FragmentListType::const_iterator const_iterator;
353 typedef FragmentListType::iterator iterator;
355 typedef FragmentListType::const_reverse_iterator const_reverse_iterator;
356 typedef FragmentListType::reverse_iterator reverse_iterator;
359 FragmentListType Fragments;
360 const MCSection *Section;
362 /// Ordinal - The section index in the assemblers section list.
365 /// LayoutOrder - The index of this section in the layout order.
366 unsigned LayoutOrder;
368 /// Alignment - The maximum alignment seen in this section.
371 /// @name Assembler Backend Data
374 // FIXME: This could all be kept private to the assembler implementation.
376 /// Address - The computed address of this section. This is ~0 until
380 /// HasInstructions - Whether this section has had instructions emitted into
382 unsigned HasInstructions : 1;
387 // Only for use as sentinel.
389 MCSectionData(const MCSection &Section, MCAssembler *A = 0);
391 const MCSection &getSection() const { return *Section; }
393 unsigned getAlignment() const { return Alignment; }
394 void setAlignment(unsigned Value) { Alignment = Value; }
396 bool hasInstructions() const { return HasInstructions; }
397 void setHasInstructions(bool Value) { HasInstructions = Value; }
399 unsigned getOrdinal() const { return Ordinal; }
400 void setOrdinal(unsigned Value) { Ordinal = Value; }
402 unsigned getLayoutOrder() const { return LayoutOrder; }
403 void setLayoutOrder(unsigned Value) { LayoutOrder = Value; }
405 /// @name Fragment Access
408 const FragmentListType &getFragmentList() const { return Fragments; }
409 FragmentListType &getFragmentList() { return Fragments; }
411 iterator begin() { return Fragments.begin(); }
412 const_iterator begin() const { return Fragments.begin(); }
414 iterator end() { return Fragments.end(); }
415 const_iterator end() const { return Fragments.end(); }
417 reverse_iterator rbegin() { return Fragments.rbegin(); }
418 const_reverse_iterator rbegin() const { return Fragments.rbegin(); }
420 reverse_iterator rend() { return Fragments.rend(); }
421 const_reverse_iterator rend() const { return Fragments.rend(); }
423 size_t size() const { return Fragments.size(); }
425 bool empty() const { return Fragments.empty(); }
432 // FIXME: Same concerns as with SectionData.
433 class MCSymbolData : public ilist_node<MCSymbolData> {
435 const MCSymbol *Symbol;
437 /// Fragment - The fragment this symbol's value is relative to, if any.
438 MCFragment *Fragment;
440 /// Offset - The offset to apply to the fragment address to form this symbol's
444 /// IsExternal - True if this symbol is visible outside this translation
446 unsigned IsExternal : 1;
448 /// IsPrivateExtern - True if this symbol is private extern.
449 unsigned IsPrivateExtern : 1;
451 /// CommonSize - The size of the symbol, if it is 'common', or 0.
453 // FIXME: Pack this in with other fields? We could put it in offset, since a
454 // common symbol can never get a definition.
457 /// SymbolSize - An expression describing how to calculate the size of
458 /// a symbol. If a symbol has no size this field will be NULL.
459 const MCExpr *SymbolSize;
461 /// CommonAlign - The alignment of the symbol, if it is 'common'.
463 // FIXME: Pack this in with other fields?
464 unsigned CommonAlign;
466 /// Flags - The Flags field is used by object file implementations to store
467 /// additional per symbol information which is not easily classified.
470 /// Index - Index field, for use by the object file implementation.
474 // Only for use as sentinel.
476 MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment, uint64_t _Offset,
482 const MCSymbol &getSymbol() const { return *Symbol; }
484 MCFragment *getFragment() const { return Fragment; }
485 void setFragment(MCFragment *Value) { Fragment = Value; }
487 uint64_t getOffset() const { return Offset; }
488 void setOffset(uint64_t Value) { Offset = Value; }
491 /// @name Symbol Attributes
494 bool isExternal() const { return IsExternal; }
495 void setExternal(bool Value) { IsExternal = Value; }
497 bool isPrivateExtern() const { return IsPrivateExtern; }
498 void setPrivateExtern(bool Value) { IsPrivateExtern = Value; }
500 /// isCommon - Is this a 'common' symbol.
501 bool isCommon() const { return CommonSize != 0; }
503 /// setCommon - Mark this symbol as being 'common'.
505 /// \param Size - The size of the symbol.
506 /// \param Align - The alignment of the symbol.
507 void setCommon(uint64_t Size, unsigned Align) {
512 /// getCommonSize - Return the size of a 'common' symbol.
513 uint64_t getCommonSize() const {
514 assert(isCommon() && "Not a 'common' symbol!");
518 void setSize(const MCExpr *SS) {
522 const MCExpr *getSize() const {
527 /// getCommonAlignment - Return the alignment of a 'common' symbol.
528 unsigned getCommonAlignment() const {
529 assert(isCommon() && "Not a 'common' symbol!");
533 /// getFlags - Get the (implementation defined) symbol flags.
534 uint32_t getFlags() const { return Flags; }
536 /// setFlags - Set the (implementation defined) symbol flags.
537 void setFlags(uint32_t Value) { Flags = Value; }
539 /// modifyFlags - Modify the flags via a mask
540 void modifyFlags(uint32_t Value, uint32_t Mask) {
541 Flags = (Flags & ~Mask) | Value;
544 /// getIndex - Get the (implementation defined) index.
545 uint64_t getIndex() const { return Index; }
547 /// setIndex - Set the (implementation defined) index.
548 void setIndex(uint64_t Value) { Index = Value; }
555 // FIXME: This really doesn't belong here. See comments below.
556 struct IndirectSymbolData {
558 MCSectionData *SectionData;
562 friend class MCAsmLayout;
565 typedef iplist<MCSectionData> SectionDataListType;
566 typedef iplist<MCSymbolData> SymbolDataListType;
568 typedef SectionDataListType::const_iterator const_iterator;
569 typedef SectionDataListType::iterator iterator;
571 typedef SymbolDataListType::const_iterator const_symbol_iterator;
572 typedef SymbolDataListType::iterator symbol_iterator;
574 typedef std::vector<IndirectSymbolData>::const_iterator
575 const_indirect_symbol_iterator;
576 typedef std::vector<IndirectSymbolData>::iterator indirect_symbol_iterator;
579 MCAssembler(const MCAssembler&); // DO NOT IMPLEMENT
580 void operator=(const MCAssembler&); // DO NOT IMPLEMENT
584 TargetAsmBackend &Backend;
586 MCCodeEmitter &Emitter;
590 iplist<MCSectionData> Sections;
592 iplist<MCSymbolData> Symbols;
594 /// The map of sections to their associated assembler backend data.
596 // FIXME: Avoid this indirection?
597 DenseMap<const MCSection*, MCSectionData*> SectionMap;
599 /// The map of symbols to their associated assembler backend data.
601 // FIXME: Avoid this indirection?
602 DenseMap<const MCSymbol*, MCSymbolData*> SymbolMap;
604 std::vector<IndirectSymbolData> IndirectSymbols;
606 unsigned RelaxAll : 1;
607 unsigned SubsectionsViaSymbols : 1;
608 unsigned PadSectionToAlignment : 1;
611 /// Evaluate a fixup to a relocatable expression and the value which should be
612 /// placed into the fixup.
614 /// \param Layout The layout to use for evaluation.
615 /// \param Fixup The fixup to evaluate.
616 /// \param DF The fragment the fixup is inside.
617 /// \param Target [out] On return, the relocatable expression the fixup
619 /// \param Value [out] On return, the value of the fixup as currently layed
621 /// \return Whether the fixup value was fully resolved. This is true if the
622 /// \arg Value result is fixed, otherwise the value may change due to
624 bool EvaluateFixup(const MCAsmLayout &Layout,
625 const MCFixup &Fixup, const MCFragment *DF,
626 MCValue &Target, uint64_t &Value) const;
628 /// Check whether a fixup can be satisfied, or whether it needs to be relaxed
629 /// (increased in size, in order to hold its value correctly).
630 bool FixupNeedsRelaxation(const MCFixup &Fixup, const MCFragment *DF,
631 const MCAsmLayout &Layout) const;
633 /// Check whether the given fragment needs relaxation.
634 bool FragmentNeedsRelaxation(const MCInstFragment *IF,
635 const MCAsmLayout &Layout) const;
637 /// Compute the effective fragment size assuming it is layed out at the given
638 /// \arg SectionAddress and \arg FragmentOffset.
639 uint64_t ComputeFragmentSize(MCAsmLayout &Layout, const MCFragment &F,
640 uint64_t SectionAddress,
641 uint64_t FragmentOffset) const;
643 /// LayoutOnce - Perform one layout iteration and return true if any offsets
645 bool LayoutOnce(MCAsmLayout &Layout);
647 /// FinishLayout - Finalize a layout, including fragment lowering.
648 void FinishLayout(MCAsmLayout &Layout);
651 /// Find the symbol which defines the atom containing the given symbol, or
652 /// null if there is no such symbol.
653 const MCSymbolData *getAtom(const MCAsmLayout &Layout,
654 const MCSymbolData *Symbol) const;
656 /// Check whether a particular symbol is visible to the linker and is required
657 /// in the symbol table, or whether it can be discarded by the assembler. This
658 /// also effects whether the assembler treats the label as potentially
659 /// defining a separate atom.
660 bool isSymbolLinkerVisible(const MCSymbol &SD) const;
662 /// Emit the section contents using the given object writer.
664 // FIXME: Should MCAssembler always have a reference to the object writer?
665 void WriteSectionData(const MCSectionData *Section, const MCAsmLayout &Layout,
666 MCObjectWriter *OW) const;
668 void AddSectionToTheEnd(MCSectionData &SD, MCAsmLayout &Layout);
671 /// Construct a new assembler instance.
673 /// \arg OS - The stream to output to.
675 // FIXME: How are we going to parameterize this? Two obvious options are stay
676 // concrete and require clients to pass in a target like object. The other
677 // option is to make this abstract, and have targets provide concrete
678 // implementations as we do with AsmParser.
679 MCAssembler(MCContext &_Context, TargetAsmBackend &_Backend,
680 MCCodeEmitter &_Emitter, bool _PadSectionToAlignment,
684 MCContext &getContext() const { return Context; }
686 TargetAsmBackend &getBackend() const { return Backend; }
688 MCCodeEmitter &getEmitter() const { return Emitter; }
690 /// Finish - Do final processing and write the object to the output stream.
691 /// \arg Writer is used for custom object writer (as the MCJIT does),
692 /// if not specified it is automatically created from backend.
693 void Finish(MCObjectWriter *Writer = 0);
695 // FIXME: This does not belong here.
696 bool getSubsectionsViaSymbols() const {
697 return SubsectionsViaSymbols;
699 void setSubsectionsViaSymbols(bool Value) {
700 SubsectionsViaSymbols = Value;
703 bool getRelaxAll() const { return RelaxAll; }
704 void setRelaxAll(bool Value) { RelaxAll = Value; }
706 /// @name Section List Access
709 const SectionDataListType &getSectionList() const { return Sections; }
710 SectionDataListType &getSectionList() { return Sections; }
712 iterator begin() { return Sections.begin(); }
713 const_iterator begin() const { return Sections.begin(); }
715 iterator end() { return Sections.end(); }
716 const_iterator end() const { return Sections.end(); }
718 size_t size() const { return Sections.size(); }
721 /// @name Symbol List Access
724 const SymbolDataListType &getSymbolList() const { return Symbols; }
725 SymbolDataListType &getSymbolList() { return Symbols; }
727 symbol_iterator symbol_begin() { return Symbols.begin(); }
728 const_symbol_iterator symbol_begin() const { return Symbols.begin(); }
730 symbol_iterator symbol_end() { return Symbols.end(); }
731 const_symbol_iterator symbol_end() const { return Symbols.end(); }
733 size_t symbol_size() const { return Symbols.size(); }
736 /// @name Indirect Symbol List Access
739 // FIXME: This is a total hack, this should not be here. Once things are
740 // factored so that the streamer has direct access to the .o writer, it can
742 std::vector<IndirectSymbolData> &getIndirectSymbols() {
743 return IndirectSymbols;
746 indirect_symbol_iterator indirect_symbol_begin() {
747 return IndirectSymbols.begin();
749 const_indirect_symbol_iterator indirect_symbol_begin() const {
750 return IndirectSymbols.begin();
753 indirect_symbol_iterator indirect_symbol_end() {
754 return IndirectSymbols.end();
756 const_indirect_symbol_iterator indirect_symbol_end() const {
757 return IndirectSymbols.end();
760 size_t indirect_symbol_size() const { return IndirectSymbols.size(); }
763 /// @name Backend Data Access
766 MCSectionData &getSectionData(const MCSection &Section) const {
767 MCSectionData *Entry = SectionMap.lookup(&Section);
768 assert(Entry && "Missing section data!");
772 MCSectionData &getOrCreateSectionData(const MCSection &Section,
774 MCSectionData *&Entry = SectionMap[&Section];
776 if (Created) *Created = !Entry;
778 Entry = new MCSectionData(Section, this);
783 MCSymbolData &getSymbolData(const MCSymbol &Symbol) const {
784 MCSymbolData *Entry = SymbolMap.lookup(&Symbol);
785 assert(Entry && "Missing symbol data!");
789 MCSymbolData &getOrCreateSymbolData(const MCSymbol &Symbol,
791 MCSymbolData *&Entry = SymbolMap[&Symbol];
793 if (Created) *Created = !Entry;
795 Entry = new MCSymbolData(Symbol, 0, 0, this);
805 } // end namespace llvm