1 //===--------- llvm/DataLayout.h - Data size & alignment info ---*- C++ -*-===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file defines layout properties related to datatype size/offset/alignment
11 // information.  It uses lazy annotations to cache information about how
12 // structure types are laid out and used.
13 //
14 // This structure should be created once, filled in if the defaults are not
15 // correct and then passed around by const&.  None of the members functions
16 // require modification to the object.
17 //
18 //===----------------------------------------------------------------------===//
19 
20 #ifndef LLVM_IR_DATALAYOUT_H
21 #define LLVM_IR_DATALAYOUT_H
22 
23 #include "llvm/ADT/DenseMap.h"
24 #include "llvm/ADT/SmallVector.h"
25 #include "llvm/IR/DerivedTypes.h"
26 #include "llvm/IR/Type.h"
27 #include "llvm/Pass.h"
28 #include "llvm/Support/DataTypes.h"
29 
30 // This needs to be outside of the namespace, to avoid conflict with llvm-c
31 // decl.
32 typedef struct LLVMOpaqueTargetData *LLVMTargetDataRef;
33 
34 namespace llvm {
35 
36 class Value;
37 class Type;
38 class IntegerType;
39 class StructType;
40 class StructLayout;
41 class Triple;
42 class GlobalVariable;
43 class LLVMContext;
44 template<typename T>
45 class ArrayRef;
46 
47 /// Enum used to categorize the alignment types stored by LayoutAlignElem
48 enum AlignTypeEnum {
49   INVALID_ALIGN = 0,
50   INTEGER_ALIGN = 'i',
51   VECTOR_ALIGN = 'v',
52   FLOAT_ALIGN = 'f',
53   AGGREGATE_ALIGN = 'a'
54 };
55 
56 // FIXME: Currently the DataLayout string carries a "preferred alignment"
57 // for types. As the DataLayout is module/global, this should likely be
58 // sunk down to an FTTI element that is queried rather than a global
59 // preference.
60 
61 /// \brief Layout alignment element.
62 ///
63 /// Stores the alignment data associated with a given alignment type (integer,
64 /// vector, float) and type bit width.
65 ///
66 /// \note The unusual order of elements in the structure attempts to reduce
67 /// padding and make the structure slightly more cache friendly.
68 struct LayoutAlignElem {
69   /// \brief Alignment type from \c AlignTypeEnum
70   unsigned AlignType : 8;
71   unsigned TypeBitWidth : 24;
72   unsigned ABIAlign : 16;
73   unsigned PrefAlign : 16;
74 
75   static LayoutAlignElem get(AlignTypeEnum align_type, unsigned abi_align,
76                              unsigned pref_align, uint32_t bit_width);
77   bool operator==(const LayoutAlignElem &rhs) const;
78 };
79 
80 /// \brief Layout pointer alignment element.
81 ///
82 /// Stores the alignment data associated with a given pointer and address space.
83 ///
84 /// \note The unusual order of elements in the structure attempts to reduce
85 /// padding and make the structure slightly more cache friendly.
86 struct PointerAlignElem {
87   unsigned ABIAlign;
88   unsigned PrefAlign;
89   uint32_t TypeByteWidth;
90   uint32_t AddressSpace;
91 
92   /// Initializer
93   static PointerAlignElem get(uint32_t AddressSpace, unsigned ABIAlign,
94                               unsigned PrefAlign, uint32_t TypeByteWidth);
95   bool operator==(const PointerAlignElem &rhs) const;
96 };
97 
98 /// \brief A parsed version of the target data layout string in and methods for
99 /// querying it.
100 ///
101 /// The target data layout string is specified *by the target* - a frontend
102 /// generating LLVM IR is required to generate the right target data for the
103 /// target being codegen'd to.
104 class DataLayout {
105 private:
106   /// Defaults to false.
107   bool BigEndian;
108 
109   unsigned StackNaturalAlign;
110 
111   enum ManglingModeT {
112     MM_None,
113     MM_ELF,
114     MM_MachO,
115     MM_WinCOFF,
116     MM_WinCOFFX86,
117     MM_Mips
118   };
119   ManglingModeT ManglingMode;
120 
121   SmallVector<unsigned char, 8> LegalIntWidths;
122 
123   /// \brief Primitive type alignment data.
124   SmallVector<LayoutAlignElem, 16> Alignments;
125 
126   /// \brief The string representation used to create this DataLayout
127   std::string StringRepresentation;
128 
129   typedef SmallVector<PointerAlignElem, 8> PointersTy;
130   PointersTy Pointers;
131 
132   PointersTy::const_iterator
findPointerLowerBound(uint32_t AddressSpace)133   findPointerLowerBound(uint32_t AddressSpace) const {
134     return const_cast<DataLayout *>(this)->findPointerLowerBound(AddressSpace);
135   }
136 
137   PointersTy::iterator findPointerLowerBound(uint32_t AddressSpace);
138 
139   /// This member is a signal that a requested alignment type and bit width were
140   /// not found in the SmallVector.
141   static const LayoutAlignElem InvalidAlignmentElem;
142 
143   /// This member is a signal that a requested pointer type and bit width were
144   /// not found in the DenseSet.
145   static const PointerAlignElem InvalidPointerElem;
146 
147   // The StructType -> StructLayout map.
148   mutable void *LayoutMap;
149 
150   void setAlignment(AlignTypeEnum align_type, unsigned abi_align,
151                     unsigned pref_align, uint32_t bit_width);
152   unsigned getAlignmentInfo(AlignTypeEnum align_type, uint32_t bit_width,
153                             bool ABIAlign, Type *Ty) const;
154   void setPointerAlignment(uint32_t AddrSpace, unsigned ABIAlign,
155                            unsigned PrefAlign, uint32_t TypeByteWidth);
156 
157   /// Internal helper method that returns requested alignment for type.
158   unsigned getAlignment(Type *Ty, bool abi_or_pref) const;
159 
160   /// \brief Valid alignment predicate.
161   ///
162   /// Predicate that tests a LayoutAlignElem reference returned by get() against
163   /// InvalidAlignmentElem.
validAlignment(const LayoutAlignElem & align)164   bool validAlignment(const LayoutAlignElem &align) const {
165     return &align != &InvalidAlignmentElem;
166   }
167 
168   /// \brief Valid pointer predicate.
169   ///
170   /// Predicate that tests a PointerAlignElem reference returned by get()
171   /// against \c InvalidPointerElem.
validPointer(const PointerAlignElem & align)172   bool validPointer(const PointerAlignElem &align) const {
173     return &align != &InvalidPointerElem;
174   }
175 
176   /// Parses a target data specification string. Assert if the string is
177   /// malformed.
178   void parseSpecifier(StringRef LayoutDescription);
179 
180   // Free all internal data structures.
181   void clear();
182 
183 public:
184   /// Constructs a DataLayout from a specification string. See reset().
DataLayout(StringRef LayoutDescription)185   explicit DataLayout(StringRef LayoutDescription) : LayoutMap(nullptr) {
186     reset(LayoutDescription);
187   }
188 
189   /// Initialize target data from properties stored in the module.
190   explicit DataLayout(const Module *M);
191 
192   void init(const Module *M);
193 
DataLayout(const DataLayout & DL)194   DataLayout(const DataLayout &DL) : LayoutMap(nullptr) { *this = DL; }
195 
196   DataLayout &operator=(const DataLayout &DL) {
197     clear();
198     StringRepresentation = DL.StringRepresentation;
199     BigEndian = DL.isBigEndian();
200     StackNaturalAlign = DL.StackNaturalAlign;
201     ManglingMode = DL.ManglingMode;
202     LegalIntWidths = DL.LegalIntWidths;
203     Alignments = DL.Alignments;
204     Pointers = DL.Pointers;
205     return *this;
206   }
207 
208   bool operator==(const DataLayout &Other) const;
209   bool operator!=(const DataLayout &Other) const { return !(*this == Other); }
210 
211   ~DataLayout(); // Not virtual, do not subclass this class
212 
213   /// Parse a data layout string (with fallback to default values).
214   void reset(StringRef LayoutDescription);
215 
216   /// Layout endianness...
isLittleEndian()217   bool isLittleEndian() const { return !BigEndian; }
isBigEndian()218   bool isBigEndian() const { return BigEndian; }
219 
220   /// \brief Returns the string representation of the DataLayout.
221   ///
222   /// This representation is in the same format accepted by the string
223   /// constructor above. This should not be used to compare two DataLayout as
224   /// different string can represent the same layout.
getStringRepresentation()225   const std::string &getStringRepresentation() const {
226     return StringRepresentation;
227   }
228 
229   /// \brief Test if the DataLayout was constructed from an empty string.
isDefault()230   bool isDefault() const { return StringRepresentation.empty(); }
231 
232   /// \brief Returns true if the specified type is known to be a native integer
233   /// type supported by the CPU.
234   ///
235   /// For example, i64 is not native on most 32-bit CPUs and i37 is not native
236   /// on any known one. This returns false if the integer width is not legal.
237   ///
238   /// The width is specified in bits.
isLegalInteger(unsigned Width)239   bool isLegalInteger(unsigned Width) const {
240     for (unsigned LegalIntWidth : LegalIntWidths)
241       if (LegalIntWidth == Width)
242         return true;
243     return false;
244   }
245 
isIllegalInteger(unsigned Width)246   bool isIllegalInteger(unsigned Width) const { return !isLegalInteger(Width); }
247 
248   /// Returns true if the given alignment exceeds the natural stack alignment.
exceedsNaturalStackAlignment(unsigned Align)249   bool exceedsNaturalStackAlignment(unsigned Align) const {
250     return (StackNaturalAlign != 0) && (Align > StackNaturalAlign);
251   }
252 
getStackAlignment()253   unsigned getStackAlignment() const { return StackNaturalAlign; }
254 
hasMicrosoftFastStdCallMangling()255   bool hasMicrosoftFastStdCallMangling() const {
256     return ManglingMode == MM_WinCOFFX86;
257   }
258 
hasLinkerPrivateGlobalPrefix()259   bool hasLinkerPrivateGlobalPrefix() const { return ManglingMode == MM_MachO; }
260 
getLinkerPrivateGlobalPrefix()261   const char *getLinkerPrivateGlobalPrefix() const {
262     if (ManglingMode == MM_MachO)
263       return "l";
264     return "";
265   }
266 
getGlobalPrefix()267   char getGlobalPrefix() const {
268     switch (ManglingMode) {
269     case MM_None:
270     case MM_ELF:
271     case MM_Mips:
272     case MM_WinCOFF:
273       return '\0';
274     case MM_MachO:
275     case MM_WinCOFFX86:
276       return '_';
277     }
278     llvm_unreachable("invalid mangling mode");
279   }
280 
getPrivateGlobalPrefix()281   const char *getPrivateGlobalPrefix() const {
282     switch (ManglingMode) {
283     case MM_None:
284       return "";
285     case MM_ELF:
286       return ".L";
287     case MM_Mips:
288       return "$";
289     case MM_MachO:
290     case MM_WinCOFF:
291     case MM_WinCOFFX86:
292       return "L";
293     }
294     llvm_unreachable("invalid mangling mode");
295   }
296 
297   static const char *getManglingComponent(const Triple &T);
298 
299   /// \brief Returns true if the specified type fits in a native integer type
300   /// supported by the CPU.
301   ///
302   /// For example, if the CPU only supports i32 as a native integer type, then
303   /// i27 fits in a legal integer type but i45 does not.
fitsInLegalInteger(unsigned Width)304   bool fitsInLegalInteger(unsigned Width) const {
305     for (unsigned LegalIntWidth : LegalIntWidths)
306       if (Width <= LegalIntWidth)
307         return true;
308     return false;
309   }
310 
311   /// Layout pointer alignment
312   /// FIXME: The defaults need to be removed once all of
313   /// the backends/clients are updated.
314   unsigned getPointerABIAlignment(unsigned AS = 0) const;
315 
316   /// Return target's alignment for stack-based pointers
317   /// FIXME: The defaults need to be removed once all of
318   /// the backends/clients are updated.
319   unsigned getPointerPrefAlignment(unsigned AS = 0) const;
320 
321   /// Layout pointer size
322   /// FIXME: The defaults need to be removed once all of
323   /// the backends/clients are updated.
324   unsigned getPointerSize(unsigned AS = 0) const;
325 
326   /// Layout pointer size, in bits
327   /// FIXME: The defaults need to be removed once all of
328   /// the backends/clients are updated.
329   unsigned getPointerSizeInBits(unsigned AS = 0) const {
330     return getPointerSize(AS) * 8;
331   }
332 
333   /// Layout pointer size, in bits, based on the type.  If this function is
334   /// called with a pointer type, then the type size of the pointer is returned.
335   /// If this function is called with a vector of pointers, then the type size
336   /// of the pointer is returned.  This should only be called with a pointer or
337   /// vector of pointers.
338   unsigned getPointerTypeSizeInBits(Type *) const;
339 
getPointerTypeSize(Type * Ty)340   unsigned getPointerTypeSize(Type *Ty) const {
341     return getPointerTypeSizeInBits(Ty) / 8;
342   }
343 
344   /// Size examples:
345   ///
346   /// Type        SizeInBits  StoreSizeInBits  AllocSizeInBits[*]
347   /// ----        ----------  ---------------  ---------------
348   ///  i1            1           8                8
349   ///  i8            8           8                8
350   ///  i19          19          24               32
351   ///  i32          32          32               32
352   ///  i100        100         104              128
353   ///  i128        128         128              128
354   ///  Float        32          32               32
355   ///  Double       64          64               64
356   ///  X86_FP80     80          80               96
357   ///
358   /// [*] The alloc size depends on the alignment, and thus on the target.
359   ///     These values are for x86-32 linux.
360 
361   /// \brief Returns the number of bits necessary to hold the specified type.
362   ///
363   /// For example, returns 36 for i36 and 80 for x86_fp80. The type passed must
364   /// have a size (Type::isSized() must return true).
365   uint64_t getTypeSizeInBits(Type *Ty) const;
366 
367   /// \brief Returns the maximum number of bytes that may be overwritten by
368   /// storing the specified type.
369   ///
370   /// For example, returns 5 for i36 and 10 for x86_fp80.
getTypeStoreSize(Type * Ty)371   uint64_t getTypeStoreSize(Type *Ty) const {
372     return (getTypeSizeInBits(Ty) + 7) / 8;
373   }
374 
375   /// \brief Returns the maximum number of bits that may be overwritten by
376   /// storing the specified type; always a multiple of 8.
377   ///
378   /// For example, returns 40 for i36 and 80 for x86_fp80.
getTypeStoreSizeInBits(Type * Ty)379   uint64_t getTypeStoreSizeInBits(Type *Ty) const {
380     return 8 * getTypeStoreSize(Ty);
381   }
382 
383   /// \brief Returns the offset in bytes between successive objects of the
384   /// specified type, including alignment padding.
385   ///
386   /// This is the amount that alloca reserves for this type. For example,
387   /// returns 12 or 16 for x86_fp80, depending on alignment.
getTypeAllocSize(Type * Ty)388   uint64_t getTypeAllocSize(Type *Ty) const {
389     // Round up to the next alignment boundary.
390     return RoundUpToAlignment(getTypeStoreSize(Ty), getABITypeAlignment(Ty));
391   }
392 
393   /// \brief Returns the offset in bits between successive objects of the
394   /// specified type, including alignment padding; always a multiple of 8.
395   ///
396   /// This is the amount that alloca reserves for this type. For example,
397   /// returns 96 or 128 for x86_fp80, depending on alignment.
getTypeAllocSizeInBits(Type * Ty)398   uint64_t getTypeAllocSizeInBits(Type *Ty) const {
399     return 8 * getTypeAllocSize(Ty);
400   }
401 
402   /// \brief Returns the minimum ABI-required alignment for the specified type.
403   unsigned getABITypeAlignment(Type *Ty) const;
404 
405   /// \brief Returns the minimum ABI-required alignment for an integer type of
406   /// the specified bitwidth.
407   unsigned getABIIntegerTypeAlignment(unsigned BitWidth) const;
408 
409   /// \brief Returns the preferred stack/global alignment for the specified
410   /// type.
411   ///
412   /// This is always at least as good as the ABI alignment.
413   unsigned getPrefTypeAlignment(Type *Ty) const;
414 
415   /// \brief Returns the preferred alignment for the specified type, returned as
416   /// log2 of the value (a shift amount).
417   unsigned getPreferredTypeAlignmentShift(Type *Ty) const;
418 
419   /// \brief Returns an integer type with size at least as big as that of a
420   /// pointer in the given address space.
421   IntegerType *getIntPtrType(LLVMContext &C, unsigned AddressSpace = 0) const;
422 
423   /// \brief Returns an integer (vector of integer) type with size at least as
424   /// big as that of a pointer of the given pointer (vector of pointer) type.
425   Type *getIntPtrType(Type *) const;
426 
427   /// \brief Returns the smallest integer type with size at least as big as
428   /// Width bits.
429   Type *getSmallestLegalIntType(LLVMContext &C, unsigned Width = 0) const;
430 
431   /// \brief Returns the largest legal integer type, or null if none are set.
getLargestLegalIntType(LLVMContext & C)432   Type *getLargestLegalIntType(LLVMContext &C) const {
433     unsigned LargestSize = getLargestLegalIntTypeSize();
434     return (LargestSize == 0) ? nullptr : Type::getIntNTy(C, LargestSize);
435   }
436 
437   /// \brief Returns the size of largest legal integer type size, or 0 if none
438   /// are set.
439   unsigned getLargestLegalIntTypeSize() const;
440 
441   /// \brief Returns the offset from the beginning of the type for the specified
442   /// indices.
443   ///
444   /// This is used to implement getelementptr.
445   uint64_t getIndexedOffset(Type *Ty, ArrayRef<Value *> Indices) const;
446 
447   /// \brief Returns a StructLayout object, indicating the alignment of the
448   /// struct, its size, and the offsets of its fields.
449   ///
450   /// Note that this information is lazily cached.
451   const StructLayout *getStructLayout(StructType *Ty) const;
452 
453   /// \brief Returns the preferred alignment of the specified global.
454   ///
455   /// This includes an explicitly requested alignment (if the global has one).
456   unsigned getPreferredAlignment(const GlobalVariable *GV) const;
457 
458   /// \brief Returns the preferred alignment of the specified global, returned
459   /// in log form.
460   ///
461   /// This includes an explicitly requested alignment (if the global has one).
462   unsigned getPreferredAlignmentLog(const GlobalVariable *GV) const;
463 };
464 
unwrap(LLVMTargetDataRef P)465 inline DataLayout *unwrap(LLVMTargetDataRef P) {
466   return reinterpret_cast<DataLayout *>(P);
467 }
468 
wrap(const DataLayout * P)469 inline LLVMTargetDataRef wrap(const DataLayout *P) {
470   return reinterpret_cast<LLVMTargetDataRef>(const_cast<DataLayout *>(P));
471 }
472 
473 /// Used to lazily calculate structure layout information for a target machine,
474 /// based on the DataLayout structure.
475 class StructLayout {
476   uint64_t StructSize;
477   unsigned StructAlignment;
478   bool IsPadded : 1;
479   unsigned NumElements : 31;
480   uint64_t MemberOffsets[1]; // variable sized array!
481 public:
getSizeInBytes()482   uint64_t getSizeInBytes() const { return StructSize; }
483 
getSizeInBits()484   uint64_t getSizeInBits() const { return 8 * StructSize; }
485 
getAlignment()486   unsigned getAlignment() const { return StructAlignment; }
487 
488   /// Returns whether the struct has padding or not between its fields.
489   /// NB: Padding in nested element is not taken into account.
hasPadding()490   bool hasPadding() const { return IsPadded; }
491 
492   /// \brief Given a valid byte offset into the structure, returns the structure
493   /// index that contains it.
494   unsigned getElementContainingOffset(uint64_t Offset) const;
495 
getElementOffset(unsigned Idx)496   uint64_t getElementOffset(unsigned Idx) const {
497     assert(Idx < NumElements && "Invalid element idx!");
498     return MemberOffsets[Idx];
499   }
500 
getElementOffsetInBits(unsigned Idx)501   uint64_t getElementOffsetInBits(unsigned Idx) const {
502     return getElementOffset(Idx) * 8;
503   }
504 
505 private:
506   friend class DataLayout; // Only DataLayout can create this class
507   StructLayout(StructType *ST, const DataLayout &DL);
508 };
509 
510 // The implementation of this method is provided inline as it is particularly
511 // well suited to constant folding when called on a specific Type subclass.
getTypeSizeInBits(Type * Ty)512 inline uint64_t DataLayout::getTypeSizeInBits(Type *Ty) const {
513   assert(Ty->isSized() && "Cannot getTypeInfo() on a type that is unsized!");
514   switch (Ty->getTypeID()) {
515   case Type::LabelTyID:
516     return getPointerSizeInBits(0);
517   case Type::PointerTyID:
518     return getPointerSizeInBits(Ty->getPointerAddressSpace());
519   case Type::ArrayTyID: {
520     ArrayType *ATy = cast<ArrayType>(Ty);
521     return ATy->getNumElements() *
522            getTypeAllocSizeInBits(ATy->getElementType());
523   }
524   case Type::StructTyID:
525     // Get the layout annotation... which is lazily created on demand.
526     return getStructLayout(cast<StructType>(Ty))->getSizeInBits();
527   case Type::IntegerTyID:
528     return Ty->getIntegerBitWidth();
529   case Type::HalfTyID:
530     return 16;
531   case Type::FloatTyID:
532     return 32;
533   case Type::DoubleTyID:
534   case Type::X86_MMXTyID:
535     return 64;
536   case Type::PPC_FP128TyID:
537   case Type::FP128TyID:
538     return 128;
539   // In memory objects this is always aligned to a higher boundary, but
540   // only 80 bits contain information.
541   case Type::X86_FP80TyID:
542     return 80;
543   case Type::VectorTyID: {
544     VectorType *VTy = cast<VectorType>(Ty);
545     return VTy->getNumElements() * getTypeSizeInBits(VTy->getElementType());
546   }
547   default:
548     llvm_unreachable("DataLayout::getTypeSizeInBits(): Unsupported type");
549   }
550 }
551 
552 } // End llvm namespace
553 
554 #endif
555