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   std::string getStringRepresentation() const { return StringRepresentation; }
226 
227   /// \brief Test if the DataLayout was constructed from an empty string.
isDefault()228   bool isDefault() const { return StringRepresentation.empty(); }
229 
230   /// \brief Returns true if the specified type is known to be a native integer
231   /// type supported by the CPU.
232   ///
233   /// For example, i64 is not native on most 32-bit CPUs and i37 is not native
234   /// on any known one. This returns false if the integer width is not legal.
235   ///
236   /// The width is specified in bits.
isLegalInteger(unsigned Width)237   bool isLegalInteger(unsigned Width) const {
238     for (unsigned LegalIntWidth : LegalIntWidths)
239       if (LegalIntWidth == Width)
240         return true;
241     return false;
242   }
243 
isIllegalInteger(unsigned Width)244   bool isIllegalInteger(unsigned Width) const { return !isLegalInteger(Width); }
245 
246   /// Returns true if the given alignment exceeds the natural stack alignment.
exceedsNaturalStackAlignment(unsigned Align)247   bool exceedsNaturalStackAlignment(unsigned Align) const {
248     return (StackNaturalAlign != 0) && (Align > StackNaturalAlign);
249   }
250 
getStackAlignment()251   unsigned getStackAlignment() const { return StackNaturalAlign; }
252 
hasMicrosoftFastStdCallMangling()253   bool hasMicrosoftFastStdCallMangling() const {
254     return ManglingMode == MM_WinCOFFX86;
255   }
256 
hasLinkerPrivateGlobalPrefix()257   bool hasLinkerPrivateGlobalPrefix() const { return ManglingMode == MM_MachO; }
258 
getLinkerPrivateGlobalPrefix()259   const char *getLinkerPrivateGlobalPrefix() const {
260     if (ManglingMode == MM_MachO)
261       return "l";
262     return "";
263   }
264 
getGlobalPrefix()265   char getGlobalPrefix() const {
266     switch (ManglingMode) {
267     case MM_None:
268     case MM_ELF:
269     case MM_Mips:
270     case MM_WinCOFF:
271       return '\0';
272     case MM_MachO:
273     case MM_WinCOFFX86:
274       return '_';
275     }
276     llvm_unreachable("invalid mangling mode");
277   }
278 
getPrivateGlobalPrefix()279   const char *getPrivateGlobalPrefix() const {
280     switch (ManglingMode) {
281     case MM_None:
282       return "";
283     case MM_ELF:
284       return ".L";
285     case MM_Mips:
286       return "$";
287     case MM_MachO:
288     case MM_WinCOFF:
289     case MM_WinCOFFX86:
290       return "L";
291     }
292     llvm_unreachable("invalid mangling mode");
293   }
294 
295   static const char *getManglingComponent(const Triple &T);
296 
297   /// \brief Returns true if the specified type fits in a native integer type
298   /// supported by the CPU.
299   ///
300   /// For example, if the CPU only supports i32 as a native integer type, then
301   /// i27 fits in a legal integer type but i45 does not.
fitsInLegalInteger(unsigned Width)302   bool fitsInLegalInteger(unsigned Width) const {
303     for (unsigned LegalIntWidth : LegalIntWidths)
304       if (Width <= LegalIntWidth)
305         return true;
306     return false;
307   }
308 
309   /// Layout pointer alignment
310   /// FIXME: The defaults need to be removed once all of
311   /// the backends/clients are updated.
312   unsigned getPointerABIAlignment(unsigned AS = 0) const;
313 
314   /// Return target's alignment for stack-based pointers
315   /// FIXME: The defaults need to be removed once all of
316   /// the backends/clients are updated.
317   unsigned getPointerPrefAlignment(unsigned AS = 0) const;
318 
319   /// Layout pointer size
320   /// FIXME: The defaults need to be removed once all of
321   /// the backends/clients are updated.
322   unsigned getPointerSize(unsigned AS = 0) const;
323 
324   /// Layout pointer size, in bits
325   /// FIXME: The defaults need to be removed once all of
326   /// the backends/clients are updated.
327   unsigned getPointerSizeInBits(unsigned AS = 0) const {
328     return getPointerSize(AS) * 8;
329   }
330 
331   /// Layout pointer size, in bits, based on the type.  If this function is
332   /// called with a pointer type, then the type size of the pointer is returned.
333   /// If this function is called with a vector of pointers, then the type size
334   /// of the pointer is returned.  This should only be called with a pointer or
335   /// vector of pointers.
336   unsigned getPointerTypeSizeInBits(Type *) const;
337 
getPointerTypeSize(Type * Ty)338   unsigned getPointerTypeSize(Type *Ty) const {
339     return getPointerTypeSizeInBits(Ty) / 8;
340   }
341 
342   /// Size examples:
343   ///
344   /// Type        SizeInBits  StoreSizeInBits  AllocSizeInBits[*]
345   /// ----        ----------  ---------------  ---------------
346   ///  i1            1           8                8
347   ///  i8            8           8                8
348   ///  i19          19          24               32
349   ///  i32          32          32               32
350   ///  i100        100         104              128
351   ///  i128        128         128              128
352   ///  Float        32          32               32
353   ///  Double       64          64               64
354   ///  X86_FP80     80          80               96
355   ///
356   /// [*] The alloc size depends on the alignment, and thus on the target.
357   ///     These values are for x86-32 linux.
358 
359   /// \brief Returns the number of bits necessary to hold the specified type.
360   ///
361   /// For example, returns 36 for i36 and 80 for x86_fp80. The type passed must
362   /// have a size (Type::isSized() must return true).
363   uint64_t getTypeSizeInBits(Type *Ty) const;
364 
365   /// \brief Returns the maximum number of bytes that may be overwritten by
366   /// storing the specified type.
367   ///
368   /// For example, returns 5 for i36 and 10 for x86_fp80.
getTypeStoreSize(Type * Ty)369   uint64_t getTypeStoreSize(Type *Ty) const {
370     return (getTypeSizeInBits(Ty) + 7) / 8;
371   }
372 
373   /// \brief Returns the maximum number of bits that may be overwritten by
374   /// storing the specified type; always a multiple of 8.
375   ///
376   /// For example, returns 40 for i36 and 80 for x86_fp80.
getTypeStoreSizeInBits(Type * Ty)377   uint64_t getTypeStoreSizeInBits(Type *Ty) const {
378     return 8 * getTypeStoreSize(Ty);
379   }
380 
381   /// \brief Returns the offset in bytes between successive objects of the
382   /// specified type, including alignment padding.
383   ///
384   /// This is the amount that alloca reserves for this type. For example,
385   /// returns 12 or 16 for x86_fp80, depending on alignment.
getTypeAllocSize(Type * Ty)386   uint64_t getTypeAllocSize(Type *Ty) const {
387     // Round up to the next alignment boundary.
388     return RoundUpToAlignment(getTypeStoreSize(Ty), getABITypeAlignment(Ty));
389   }
390 
391   /// \brief Returns the offset in bits between successive objects of the
392   /// specified type, including alignment padding; always a multiple of 8.
393   ///
394   /// This is the amount that alloca reserves for this type. For example,
395   /// returns 96 or 128 for x86_fp80, depending on alignment.
getTypeAllocSizeInBits(Type * Ty)396   uint64_t getTypeAllocSizeInBits(Type *Ty) const {
397     return 8 * getTypeAllocSize(Ty);
398   }
399 
400   /// \brief Returns the minimum ABI-required alignment for the specified type.
401   unsigned getABITypeAlignment(Type *Ty) const;
402 
403   /// \brief Returns the minimum ABI-required alignment for an integer type of
404   /// the specified bitwidth.
405   unsigned getABIIntegerTypeAlignment(unsigned BitWidth) const;
406 
407   /// \brief Returns the preferred stack/global alignment for the specified
408   /// type.
409   ///
410   /// This is always at least as good as the ABI alignment.
411   unsigned getPrefTypeAlignment(Type *Ty) const;
412 
413   /// \brief Returns the preferred alignment for the specified type, returned as
414   /// log2 of the value (a shift amount).
415   unsigned getPreferredTypeAlignmentShift(Type *Ty) const;
416 
417   /// \brief Returns an integer type with size at least as big as that of a
418   /// pointer in the given address space.
419   IntegerType *getIntPtrType(LLVMContext &C, unsigned AddressSpace = 0) const;
420 
421   /// \brief Returns an integer (vector of integer) type with size at least as
422   /// big as that of a pointer of the given pointer (vector of pointer) type.
423   Type *getIntPtrType(Type *) const;
424 
425   /// \brief Returns the smallest integer type with size at least as big as
426   /// Width bits.
427   Type *getSmallestLegalIntType(LLVMContext &C, unsigned Width = 0) const;
428 
429   /// \brief Returns the largest legal integer type, or null if none are set.
getLargestLegalIntType(LLVMContext & C)430   Type *getLargestLegalIntType(LLVMContext &C) const {
431     unsigned LargestSize = getLargestLegalIntTypeSize();
432     return (LargestSize == 0) ? nullptr : Type::getIntNTy(C, LargestSize);
433   }
434 
435   /// \brief Returns the size of largest legal integer type size, or 0 if none
436   /// are set.
437   unsigned getLargestLegalIntTypeSize() const;
438 
439   /// \brief Returns the offset from the beginning of the type for the specified
440   /// indices.
441   ///
442   /// This is used to implement getelementptr.
443   uint64_t getIndexedOffset(Type *Ty, ArrayRef<Value *> Indices) const;
444 
445   /// \brief Returns a StructLayout object, indicating the alignment of the
446   /// struct, its size, and the offsets of its fields.
447   ///
448   /// Note that this information is lazily cached.
449   const StructLayout *getStructLayout(StructType *Ty) const;
450 
451   /// \brief Returns the preferred alignment of the specified global.
452   ///
453   /// This includes an explicitly requested alignment (if the global has one).
454   unsigned getPreferredAlignment(const GlobalVariable *GV) const;
455 
456   /// \brief Returns the preferred alignment of the specified global, returned
457   /// in log form.
458   ///
459   /// This includes an explicitly requested alignment (if the global has one).
460   unsigned getPreferredAlignmentLog(const GlobalVariable *GV) const;
461 };
462 
unwrap(LLVMTargetDataRef P)463 inline DataLayout *unwrap(LLVMTargetDataRef P) {
464   return reinterpret_cast<DataLayout *>(P);
465 }
466 
wrap(const DataLayout * P)467 inline LLVMTargetDataRef wrap(const DataLayout *P) {
468   return reinterpret_cast<LLVMTargetDataRef>(const_cast<DataLayout *>(P));
469 }
470 
471 /// Used to lazily calculate structure layout information for a target machine,
472 /// based on the DataLayout structure.
473 class StructLayout {
474   uint64_t StructSize;
475   unsigned StructAlignment;
476   unsigned NumElements;
477   uint64_t MemberOffsets[1]; // variable sized array!
478 public:
getSizeInBytes()479   uint64_t getSizeInBytes() const { return StructSize; }
480 
getSizeInBits()481   uint64_t getSizeInBits() const { return 8 * StructSize; }
482 
getAlignment()483   unsigned getAlignment() const { return StructAlignment; }
484 
485   /// \brief Given a valid byte offset into the structure, returns the structure
486   /// index that contains it.
487   unsigned getElementContainingOffset(uint64_t Offset) const;
488 
getElementOffset(unsigned Idx)489   uint64_t getElementOffset(unsigned Idx) const {
490     assert(Idx < NumElements && "Invalid element idx!");
491     return MemberOffsets[Idx];
492   }
493 
getElementOffsetInBits(unsigned Idx)494   uint64_t getElementOffsetInBits(unsigned Idx) const {
495     return getElementOffset(Idx) * 8;
496   }
497 
498 private:
499   friend class DataLayout; // Only DataLayout can create this class
500   StructLayout(StructType *ST, const DataLayout &DL);
501 };
502 
503 // The implementation of this method is provided inline as it is particularly
504 // well suited to constant folding when called on a specific Type subclass.
getTypeSizeInBits(Type * Ty)505 inline uint64_t DataLayout::getTypeSizeInBits(Type *Ty) const {
506   assert(Ty->isSized() && "Cannot getTypeInfo() on a type that is unsized!");
507   switch (Ty->getTypeID()) {
508   case Type::LabelTyID:
509     return getPointerSizeInBits(0);
510   case Type::PointerTyID:
511     return getPointerSizeInBits(Ty->getPointerAddressSpace());
512   case Type::ArrayTyID: {
513     ArrayType *ATy = cast<ArrayType>(Ty);
514     return ATy->getNumElements() *
515            getTypeAllocSizeInBits(ATy->getElementType());
516   }
517   case Type::StructTyID:
518     // Get the layout annotation... which is lazily created on demand.
519     return getStructLayout(cast<StructType>(Ty))->getSizeInBits();
520   case Type::IntegerTyID:
521     return Ty->getIntegerBitWidth();
522   case Type::HalfTyID:
523     return 16;
524   case Type::FloatTyID:
525     return 32;
526   case Type::DoubleTyID:
527   case Type::X86_MMXTyID:
528     return 64;
529   case Type::PPC_FP128TyID:
530   case Type::FP128TyID:
531     return 128;
532   // In memory objects this is always aligned to a higher boundary, but
533   // only 80 bits contain information.
534   case Type::X86_FP80TyID:
535     return 80;
536   case Type::VectorTyID: {
537     VectorType *VTy = cast<VectorType>(Ty);
538     return VTy->getNumElements() * getTypeSizeInBits(VTy->getElementType());
539   }
540   default:
541     llvm_unreachable("DataLayout::getTypeSizeInBits(): Unsupported type");
542   }
543 }
544 
545 } // End llvm namespace
546 
547 #endif
548