LLVM  mainline
MCAsmInfo.h
Go to the documentation of this file.
00001 //===-- llvm/MC/MCAsmInfo.h - Asm info --------------------------*- C++ -*-===//
00002 //
00003 //                     The LLVM Compiler Infrastructure
00004 //
00005 // This file is distributed under the University of Illinois Open Source
00006 // License. See LICENSE.TXT for details.
00007 //
00008 //===----------------------------------------------------------------------===//
00009 //
00010 // This file contains a class to be used as the basis for target specific
00011 // asm writers.  This class primarily takes care of global printing constants,
00012 // which are used in very similar ways across all targets.
00013 //
00014 //===----------------------------------------------------------------------===//
00015 
00016 #ifndef LLVM_MC_MCASMINFO_H
00017 #define LLVM_MC_MCASMINFO_H
00018 
00019 #include "llvm/MC/MCDirectives.h"
00020 #include "llvm/MC/MCDwarf.h"
00021 #include <cassert>
00022 #include <vector>
00023 
00024 namespace llvm {
00025 class MCExpr;
00026 class MCSection;
00027 class MCStreamer;
00028 class MCSymbol;
00029 class MCContext;
00030 
00031 namespace WinEH {
00032 enum class EncodingType {
00033   Invalid, /// Invalid
00034   Alpha,   /// Windows Alpha
00035   Alpha64, /// Windows AXP64
00036   ARM,     /// Windows NT (Windows on ARM)
00037   CE,      /// Windows CE ARM, PowerPC, SH3, SH4
00038   Itanium, /// Windows x64, Windows Itanium (IA-64)
00039   X86,     /// Windows x86, uses no CFI, just EH tables
00040   MIPS = Alpha,
00041 };
00042 }
00043 
00044 enum class ExceptionHandling {
00045   None,     /// No exception support
00046   DwarfCFI, /// DWARF-like instruction based exceptions
00047   SjLj,     /// setjmp/longjmp based exceptions
00048   ARM,      /// ARM EHABI
00049   WinEH,    /// Windows Exception Handling
00050 };
00051 
00052 namespace LCOMM {
00053 enum LCOMMType { NoAlignment, ByteAlignment, Log2Alignment };
00054 }
00055 
00056 /// This class is intended to be used as a base class for asm
00057 /// properties and features specific to the target.
00058 class MCAsmInfo {
00059 protected:
00060   //===------------------------------------------------------------------===//
00061   // Properties to be set by the target writer, used to configure asm printer.
00062   //
00063 
00064   /// Pointer size in bytes.  Default is 4.
00065   unsigned PointerSize;
00066 
00067   /// Size of the stack slot reserved for callee-saved registers, in bytes.
00068   /// Default is same as pointer size.
00069   unsigned CalleeSaveStackSlotSize;
00070 
00071   /// True if target is little endian.  Default is true.
00072   bool IsLittleEndian;
00073 
00074   /// True if target stack grow up.  Default is false.
00075   bool StackGrowsUp;
00076 
00077   /// True if this target has the MachO .subsections_via_symbols directive.
00078   /// Default is false.
00079   bool HasSubsectionsViaSymbols;
00080 
00081   /// True if this is a MachO target that supports the macho-specific .zerofill
00082   /// directive for emitting BSS Symbols.  Default is false.
00083   bool HasMachoZeroFillDirective;
00084 
00085   /// True if this is a MachO target that supports the macho-specific .tbss
00086   /// directive for emitting thread local BSS Symbols.  Default is false.
00087   bool HasMachoTBSSDirective;
00088 
00089   /// True if the compiler should emit a ".reference .constructors_used" or
00090   /// ".reference .destructors_used" directive after the static ctor/dtor
00091   /// list.  This directive is only emitted in Static relocation model.  Default
00092   /// is false.
00093   bool HasStaticCtorDtorReferenceInStaticMode;
00094 
00095   /// This is the maximum possible length of an instruction, which is needed to
00096   /// compute the size of an inline asm.  Defaults to 4.
00097   unsigned MaxInstLength;
00098 
00099   /// Every possible instruction length is a multiple of this value.  Factored
00100   /// out in .debug_frame and .debug_line.  Defaults to 1.
00101   unsigned MinInstAlignment;
00102 
00103   /// The '$' token, when not referencing an identifier or constant, refers to
00104   /// the current PC.  Defaults to false.
00105   bool DollarIsPC;
00106 
00107   /// This string, if specified, is used to separate instructions from each
00108   /// other when on the same line.  Defaults to ';'
00109   const char *SeparatorString;
00110 
00111   /// This indicates the comment character used by the assembler.  Defaults to
00112   /// "#"
00113   const char *CommentString;
00114 
00115   /// This is appended to emitted labels.  Defaults to ":"
00116   const char *LabelSuffix;
00117 
00118   // Print the EH begin symbol with an assignment. Defaults to false.
00119   bool UseAssignmentForEHBegin;
00120 
00121   // Do we need to create a local symbol for .size?
00122   bool NeedsLocalForSize;
00123 
00124   /// This prefix is used for globals like constant pool entries that are
00125   /// completely private to the .s file and should not have names in the .o
00126   /// file.  Defaults to "L"
00127   const char *PrivateGlobalPrefix;
00128 
00129   /// This prefix is used for labels for basic blocks. Defaults to the same as
00130   /// PrivateGlobalPrefix.
00131   const char *PrivateLabelPrefix;
00132 
00133   /// This prefix is used for symbols that should be passed through the
00134   /// assembler but be removed by the linker.  This is 'l' on Darwin, currently
00135   /// used for some ObjC metadata.  The default of "" meast that for this system
00136   /// a plain private symbol should be used.  Defaults to "".
00137   const char *LinkerPrivateGlobalPrefix;
00138 
00139   /// If these are nonempty, they contain a directive to emit before and after
00140   /// an inline assembly statement.  Defaults to "#APP\n", "#NO_APP\n"
00141   const char *InlineAsmStart;
00142   const char *InlineAsmEnd;
00143 
00144   /// These are assembly directives that tells the assembler to interpret the
00145   /// following instructions differently.  Defaults to ".code16", ".code32",
00146   /// ".code64".
00147   const char *Code16Directive;
00148   const char *Code32Directive;
00149   const char *Code64Directive;
00150 
00151   /// Which dialect of an assembler variant to use.  Defaults to 0
00152   unsigned AssemblerDialect;
00153 
00154   /// This is true if the assembler allows @ characters in symbol names.
00155   /// Defaults to false.
00156   bool AllowAtInName;
00157 
00158   /// This is true if data region markers should be printed as
00159   /// ".data_region/.end_data_region" directives. If false, use "$d/$a" labels
00160   /// instead.
00161   bool UseDataRegionDirectives;
00162 
00163   //===--- Data Emission Directives -------------------------------------===//
00164 
00165   /// This should be set to the directive used to get some number of zero bytes
00166   /// emitted to the current section.  Common cases are "\t.zero\t" and
00167   /// "\t.space\t".  If this is set to null, the Data*bitsDirective's will be
00168   /// used to emit zero bytes.  Defaults to "\t.zero\t"
00169   const char *ZeroDirective;
00170 
00171   /// This directive allows emission of an ascii string with the standard C
00172   /// escape characters embedded into it.  Defaults to "\t.ascii\t"
00173   const char *AsciiDirective;
00174 
00175   /// If not null, this allows for special handling of zero terminated strings
00176   /// on this target.  This is commonly supported as ".asciz".  If a target
00177   /// doesn't support this, it can be set to null.  Defaults to "\t.asciz\t"
00178   const char *AscizDirective;
00179 
00180   /// These directives are used to output some unit of integer data to the
00181   /// current section.  If a data directive is set to null, smaller data
00182   /// directives will be used to emit the large sizes.  Defaults to "\t.byte\t",
00183   /// "\t.short\t", "\t.long\t", "\t.quad\t"
00184   const char *Data8bitsDirective;
00185   const char *Data16bitsDirective;
00186   const char *Data32bitsDirective;
00187   const char *Data64bitsDirective;
00188 
00189   /// If non-null, a directive that is used to emit a word which should be
00190   /// relocated as a 64-bit GP-relative offset, e.g. .gpdword on Mips.  Defaults
00191   /// to NULL.
00192   const char *GPRel64Directive;
00193 
00194   /// If non-null, a directive that is used to emit a word which should be
00195   /// relocated as a 32-bit GP-relative offset, e.g. .gpword on Mips or .gprel32
00196   /// on Alpha.  Defaults to NULL.
00197   const char *GPRel32Directive;
00198 
00199   /// This is true if this target uses "Sun Style" syntax for section switching
00200   /// ("#alloc,#write" etc) instead of the normal ELF syntax (,"a,w") in
00201   /// .section directives.  Defaults to false.
00202   bool SunStyleELFSectionSwitchSyntax;
00203 
00204   /// This is true if this target uses ELF '.section' directive before the
00205   /// '.bss' one. It's used for PPC/Linux which doesn't support the '.bss'
00206   /// directive only.  Defaults to false.
00207   bool UsesELFSectionDirectiveForBSS;
00208 
00209   bool NeedsDwarfSectionOffsetDirective;
00210 
00211   //===--- Alignment Information ----------------------------------------===//
00212 
00213   /// If this is true (the default) then the asmprinter emits ".align N"
00214   /// directives, where N is the number of bytes to align to.  Otherwise, it
00215   /// emits ".align log2(N)", e.g. 3 to align to an 8 byte boundary.  Defaults
00216   /// to true.
00217   bool AlignmentIsInBytes;
00218 
00219   /// If non-zero, this is used to fill the executable space created as the
00220   /// result of a alignment directive.  Defaults to 0
00221   unsigned TextAlignFillValue;
00222 
00223   //===--- Global Variable Emission Directives --------------------------===//
00224 
00225   /// This is the directive used to declare a global entity. Defaults to
00226   /// ".globl".
00227   const char *GlobalDirective;
00228 
00229   /// True if the expression
00230   ///   .long f - g
00231   /// uses a relocation but it can be suppressed by writing
00232   ///   a = f - g
00233   ///   .long a
00234   bool SetDirectiveSuppressesReloc;
00235 
00236   /// False if the assembler requires that we use
00237   /// \code
00238   ///   Lc = a - b
00239   ///   .long Lc
00240   /// \endcode
00241   //
00242   /// instead of
00243   //
00244   /// \code
00245   ///   .long a - b
00246   /// \endcode
00247   ///
00248   ///  Defaults to true.
00249   bool HasAggressiveSymbolFolding;
00250 
00251   /// True is .comm's and .lcomms optional alignment is to be specified in bytes
00252   /// instead of log2(n).  Defaults to true.
00253   bool COMMDirectiveAlignmentIsInBytes;
00254 
00255   /// Describes if the .lcomm directive for the target supports an alignment
00256   /// argument and how it is interpreted.  Defaults to NoAlignment.
00257   LCOMM::LCOMMType LCOMMDirectiveAlignmentType;
00258 
00259   // True if the target allows .align directives on functions. This is true for
00260   // most targets, so defaults to true.
00261   bool HasFunctionAlignment;
00262 
00263   /// True if the target has .type and .size directives, this is true for most
00264   /// ELF targets.  Defaults to true.
00265   bool HasDotTypeDotSizeDirective;
00266 
00267   /// True if the target has a single parameter .file directive, this is true
00268   /// for ELF targets.  Defaults to true.
00269   bool HasSingleParameterDotFile;
00270 
00271   /// True if the target has a .ident directive, this is true for ELF targets.
00272   /// Defaults to false.
00273   bool HasIdentDirective;
00274 
00275   /// True if this target supports the MachO .no_dead_strip directive.  Defaults
00276   /// to false.
00277   bool HasNoDeadStrip;
00278 
00279   /// Used to declare a global as being a weak symbol. Defaults to ".weak".
00280   const char *WeakDirective;
00281 
00282   /// This directive, if non-null, is used to declare a global as being a weak
00283   /// undefined symbol.  Defaults to NULL.
00284   const char *WeakRefDirective;
00285 
00286   /// True if we have a directive to declare a global as being a weak defined
00287   /// symbol.  Defaults to false.
00288   bool HasWeakDefDirective;
00289 
00290   /// True if we have a directive to declare a global as being a weak defined
00291   /// symbol that can be hidden (unexported).  Defaults to false.
00292   bool HasWeakDefCanBeHiddenDirective;
00293 
00294   /// True if we have a .linkonce directive.  This is used on cygwin/mingw.
00295   /// Defaults to false.
00296   bool HasLinkOnceDirective;
00297 
00298   /// This attribute, if not MCSA_Invalid, is used to declare a symbol as having
00299   /// hidden visibility.  Defaults to MCSA_Hidden.
00300   MCSymbolAttr HiddenVisibilityAttr;
00301 
00302   /// This attribute, if not MCSA_Invalid, is used to declare an undefined
00303   /// symbol as having hidden visibility. Defaults to MCSA_Hidden.
00304   MCSymbolAttr HiddenDeclarationVisibilityAttr;
00305 
00306   /// This attribute, if not MCSA_Invalid, is used to declare a symbol as having
00307   /// protected visibility.  Defaults to MCSA_Protected
00308   MCSymbolAttr ProtectedVisibilityAttr;
00309 
00310   //===--- Dwarf Emission Directives -----------------------------------===//
00311 
00312   /// True if target supports emission of debugging information.  Defaults to
00313   /// false.
00314   bool SupportsDebugInformation;
00315 
00316   /// Exception handling format for the target.  Defaults to None.
00317   ExceptionHandling ExceptionsType;
00318 
00319   /// Windows exception handling data (.pdata) encoding.  Defaults to Invalid.
00320   WinEH::EncodingType WinEHEncodingType;
00321 
00322   /// True if Dwarf2 output generally uses relocations for references to other
00323   /// .debug_* sections.
00324   bool DwarfUsesRelocationsAcrossSections;
00325 
00326   /// True if DWARF FDE symbol reference relocations should be replaced by an
00327   /// absolute difference.
00328   bool DwarfFDESymbolsUseAbsDiff;
00329 
00330   /// True if dwarf register numbers are printed instead of symbolic register
00331   /// names in .cfi_* directives.  Defaults to false.
00332   bool DwarfRegNumForCFI;
00333 
00334   /// True if target uses parens to indicate the symbol variant instead of @.
00335   /// For example, foo(plt) instead of foo@plt.  Defaults to false.
00336   bool UseParensForSymbolVariant;
00337 
00338   //===--- Prologue State ----------------------------------------------===//
00339 
00340   std::vector<MCCFIInstruction> InitialFrameState;
00341 
00342   //===--- Integrated Assembler Information ----------------------------===//
00343 
00344   /// Should we use the integrated assembler?
00345   /// The integrated assembler should be enabled by default (by the
00346   /// constructors) when failing to parse a valid piece of assembly (inline
00347   /// or otherwise) is considered a bug. It may then be overridden after
00348   /// construction (see LLVMTargetMachine::initAsmInfo()).
00349   bool UseIntegratedAssembler;
00350 
00351   /// Compress DWARF debug sections. Defaults to false.
00352   bool CompressDebugSections;
00353 
00354   /// True if the integrated assembler should interpret 'a >> b' constant
00355   /// expressions as logical rather than arithmetic.
00356   bool UseLogicalShr;
00357 
00358 public:
00359   explicit MCAsmInfo();
00360   virtual ~MCAsmInfo();
00361 
00362   /// Get the pointer size in bytes.
00363   unsigned getPointerSize() const { return PointerSize; }
00364 
00365   /// Get the callee-saved register stack slot
00366   /// size in bytes.
00367   unsigned getCalleeSaveStackSlotSize() const {
00368     return CalleeSaveStackSlotSize;
00369   }
00370 
00371   /// True if the target is little endian.
00372   bool isLittleEndian() const { return IsLittleEndian; }
00373 
00374   /// True if target stack grow up.
00375   bool isStackGrowthDirectionUp() const { return StackGrowsUp; }
00376 
00377   bool hasSubsectionsViaSymbols() const { return HasSubsectionsViaSymbols; }
00378 
00379   // Data directive accessors.
00380 
00381   const char *getData8bitsDirective() const { return Data8bitsDirective; }
00382   const char *getData16bitsDirective() const { return Data16bitsDirective; }
00383   const char *getData32bitsDirective() const { return Data32bitsDirective; }
00384   const char *getData64bitsDirective() const { return Data64bitsDirective; }
00385   const char *getGPRel64Directive() const { return GPRel64Directive; }
00386   const char *getGPRel32Directive() const { return GPRel32Directive; }
00387 
00388   /// Targets can implement this method to specify a section to switch to if the
00389   /// translation unit doesn't have any trampolines that require an executable
00390   /// stack.
00391   virtual MCSection *getNonexecutableStackSection(MCContext &Ctx) const {
00392     return nullptr;
00393   }
00394 
00395   /// \brief True if the section is atomized using the symbols in it.
00396   /// This is false if the section is not atomized at all (most ELF sections) or
00397   /// if it is atomized based on its contents (MachO' __TEXT,__cstring for
00398   /// example).
00399   virtual bool isSectionAtomizableBySymbols(const MCSection &Section) const;
00400 
00401   virtual const MCExpr *getExprForPersonalitySymbol(const MCSymbol *Sym,
00402                                                     unsigned Encoding,
00403                                                     MCStreamer &Streamer) const;
00404 
00405   virtual const MCExpr *getExprForFDESymbol(const MCSymbol *Sym,
00406                                             unsigned Encoding,
00407                                             MCStreamer &Streamer) const;
00408 
00409   bool usesSunStyleELFSectionSwitchSyntax() const {
00410     return SunStyleELFSectionSwitchSyntax;
00411   }
00412 
00413   bool usesELFSectionDirectiveForBSS() const {
00414     return UsesELFSectionDirectiveForBSS;
00415   }
00416 
00417   bool needsDwarfSectionOffsetDirective() const {
00418     return NeedsDwarfSectionOffsetDirective;
00419   }
00420 
00421   // Accessors.
00422 
00423   bool hasMachoZeroFillDirective() const { return HasMachoZeroFillDirective; }
00424   bool hasMachoTBSSDirective() const { return HasMachoTBSSDirective; }
00425   bool hasStaticCtorDtorReferenceInStaticMode() const {
00426     return HasStaticCtorDtorReferenceInStaticMode;
00427   }
00428   unsigned getMaxInstLength() const { return MaxInstLength; }
00429   unsigned getMinInstAlignment() const { return MinInstAlignment; }
00430   bool getDollarIsPC() const { return DollarIsPC; }
00431   const char *getSeparatorString() const { return SeparatorString; }
00432 
00433   /// This indicates the column (zero-based) at which asm comments should be
00434   /// printed.
00435   unsigned getCommentColumn() const { return 40; }
00436 
00437   const char *getCommentString() const { return CommentString; }
00438   const char *getLabelSuffix() const { return LabelSuffix; }
00439 
00440   bool useAssignmentForEHBegin() const { return UseAssignmentForEHBegin; }
00441   bool needsLocalForSize() const { return NeedsLocalForSize; }
00442   const char *getPrivateGlobalPrefix() const { return PrivateGlobalPrefix; }
00443   const char *getPrivateLabelPrefix() const { return PrivateLabelPrefix; }
00444   bool hasLinkerPrivateGlobalPrefix() const {
00445     return LinkerPrivateGlobalPrefix[0] != '\0';
00446   }
00447   const char *getLinkerPrivateGlobalPrefix() const {
00448     if (hasLinkerPrivateGlobalPrefix())
00449       return LinkerPrivateGlobalPrefix;
00450     return getPrivateGlobalPrefix();
00451   }
00452   const char *getInlineAsmStart() const { return InlineAsmStart; }
00453   const char *getInlineAsmEnd() const { return InlineAsmEnd; }
00454   const char *getCode16Directive() const { return Code16Directive; }
00455   const char *getCode32Directive() const { return Code32Directive; }
00456   const char *getCode64Directive() const { return Code64Directive; }
00457   unsigned getAssemblerDialect() const { return AssemblerDialect; }
00458   bool doesAllowAtInName() const { return AllowAtInName; }
00459   bool doesSupportDataRegionDirectives() const {
00460     return UseDataRegionDirectives;
00461   }
00462   const char *getZeroDirective() const { return ZeroDirective; }
00463   const char *getAsciiDirective() const { return AsciiDirective; }
00464   const char *getAscizDirective() const { return AscizDirective; }
00465   bool getAlignmentIsInBytes() const { return AlignmentIsInBytes; }
00466   unsigned getTextAlignFillValue() const { return TextAlignFillValue; }
00467   const char *getGlobalDirective() const { return GlobalDirective; }
00468   bool doesSetDirectiveSuppressesReloc() const {
00469     return SetDirectiveSuppressesReloc;
00470   }
00471   bool hasAggressiveSymbolFolding() const { return HasAggressiveSymbolFolding; }
00472   bool getCOMMDirectiveAlignmentIsInBytes() const {
00473     return COMMDirectiveAlignmentIsInBytes;
00474   }
00475   LCOMM::LCOMMType getLCOMMDirectiveAlignmentType() const {
00476     return LCOMMDirectiveAlignmentType;
00477   }
00478   bool hasFunctionAlignment() const { return HasFunctionAlignment; }
00479   bool hasDotTypeDotSizeDirective() const { return HasDotTypeDotSizeDirective; }
00480   bool hasSingleParameterDotFile() const { return HasSingleParameterDotFile; }
00481   bool hasIdentDirective() const { return HasIdentDirective; }
00482   bool hasNoDeadStrip() const { return HasNoDeadStrip; }
00483   const char *getWeakDirective() const { return WeakDirective; }
00484   const char *getWeakRefDirective() const { return WeakRefDirective; }
00485   bool hasWeakDefDirective() const { return HasWeakDefDirective; }
00486   bool hasWeakDefCanBeHiddenDirective() const {
00487     return HasWeakDefCanBeHiddenDirective;
00488   }
00489   bool hasLinkOnceDirective() const { return HasLinkOnceDirective; }
00490 
00491   MCSymbolAttr getHiddenVisibilityAttr() const { return HiddenVisibilityAttr; }
00492   MCSymbolAttr getHiddenDeclarationVisibilityAttr() const {
00493     return HiddenDeclarationVisibilityAttr;
00494   }
00495   MCSymbolAttr getProtectedVisibilityAttr() const {
00496     return ProtectedVisibilityAttr;
00497   }
00498   bool doesSupportDebugInformation() const { return SupportsDebugInformation; }
00499   bool doesSupportExceptionHandling() const {
00500     return ExceptionsType != ExceptionHandling::None;
00501   }
00502   ExceptionHandling getExceptionHandlingType() const { return ExceptionsType; }
00503   WinEH::EncodingType getWinEHEncodingType() const { return WinEHEncodingType; }
00504 
00505   /// Returns true if the exception handling method for the platform uses call
00506   /// frame information to unwind.
00507   bool usesCFIForEH() const {
00508     return (ExceptionsType == ExceptionHandling::DwarfCFI ||
00509             ExceptionsType == ExceptionHandling::ARM || usesWindowsCFI());
00510   }
00511 
00512   bool usesWindowsCFI() const {
00513     return ExceptionsType == ExceptionHandling::WinEH &&
00514            (WinEHEncodingType != WinEH::EncodingType::Invalid &&
00515             WinEHEncodingType != WinEH::EncodingType::X86);
00516   }
00517 
00518   bool doesDwarfUseRelocationsAcrossSections() const {
00519     return DwarfUsesRelocationsAcrossSections;
00520   }
00521   bool doDwarfFDESymbolsUseAbsDiff() const { return DwarfFDESymbolsUseAbsDiff; }
00522   bool useDwarfRegNumForCFI() const { return DwarfRegNumForCFI; }
00523   bool useParensForSymbolVariant() const { return UseParensForSymbolVariant; }
00524 
00525   void addInitialFrameState(const MCCFIInstruction &Inst) {
00526     InitialFrameState.push_back(Inst);
00527   }
00528 
00529   const std::vector<MCCFIInstruction> &getInitialFrameState() const {
00530     return InitialFrameState;
00531   }
00532 
00533   /// Return true if assembly (inline or otherwise) should be parsed.
00534   bool useIntegratedAssembler() const { return UseIntegratedAssembler; }
00535 
00536   /// Set whether assembly (inline or otherwise) should be parsed.
00537   virtual void setUseIntegratedAssembler(bool Value) {
00538     UseIntegratedAssembler = Value;
00539   }
00540 
00541   bool compressDebugSections() const { return CompressDebugSections; }
00542 
00543   void setCompressDebugSections(bool CompressDebugSections) {
00544     this->CompressDebugSections = CompressDebugSections;
00545   }
00546 
00547   bool shouldUseLogicalShr() const { return UseLogicalShr; }
00548 };
00549 }
00550 
00551 #endif