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/ArrayRef.h"
24#include "llvm/ADT/STLExtras.h"
25#include "llvm/ADT/SmallVector.h"
26#include "llvm/ADT/StringRef.h"
27#include "llvm/IR/DerivedTypes.h"
28#include "llvm/IR/Type.h"
29#include "llvm/Pass.h"
30#include "llvm/Support/Casting.h"
31#include "llvm/Support/ErrorHandling.h"
32#include "llvm/Support/MathExtras.h"
33#include <cassert>
34#include <cstdint>
35#include <string>
36
37// This needs to be outside of the namespace, to avoid conflict with llvm-c
38// decl.
39using LLVMTargetDataRef = struct LLVMOpaqueTargetData *;
40
41namespace llvm {
42
43class GlobalVariable;
44class LLVMContext;
45class Module;
46class StructLayout;
47class Triple;
48class Value;
49
50/// Enum used to categorize the alignment types stored by LayoutAlignElem
51enum AlignTypeEnum {
52  INVALID_ALIGN = 0,
53  INTEGER_ALIGN = 'i',
54  VECTOR_ALIGN = 'v',
55  FLOAT_ALIGN = 'f',
56  AGGREGATE_ALIGN = 'a'
57};
58
59// FIXME: Currently the DataLayout string carries a "preferred alignment"
60// for types. As the DataLayout is module/global, this should likely be
61// sunk down to an FTTI element that is queried rather than a global
62// preference.
63
64/// \brief Layout alignment element.
65///
66/// Stores the alignment data associated with a given alignment type (integer,
67/// vector, float) and type bit width.
68///
69/// \note The unusual order of elements in the structure attempts to reduce
70/// padding and make the structure slightly more cache friendly.
71struct LayoutAlignElem {
72  /// \brief Alignment type from \c AlignTypeEnum
73  unsigned AlignType : 8;
74  unsigned TypeBitWidth : 24;
75  unsigned ABIAlign : 16;
76  unsigned PrefAlign : 16;
77
78  static LayoutAlignElem get(AlignTypeEnum align_type, unsigned abi_align,
79                             unsigned pref_align, uint32_t bit_width);
80
81  bool operator==(const LayoutAlignElem &rhs) const;
82};
83
84/// \brief Layout pointer alignment element.
85///
86/// Stores the alignment data associated with a given pointer and address space.
87///
88/// \note The unusual order of elements in the structure attempts to reduce
89/// padding and make the structure slightly more cache friendly.
90struct PointerAlignElem {
91  unsigned ABIAlign;
92  unsigned PrefAlign;
93  uint32_t TypeByteWidth;
94  uint32_t AddressSpace;
95
96  /// Initializer
97  static PointerAlignElem get(uint32_t AddressSpace, unsigned ABIAlign,
98                              unsigned PrefAlign, uint32_t TypeByteWidth);
99
100  bool operator==(const PointerAlignElem &rhs) const;
101};
102
103/// \brief A parsed version of the target data layout string in and methods for
104/// querying it.
105///
106/// The target data layout string is specified *by the target* - a frontend
107/// generating LLVM IR is required to generate the right target data for the
108/// target being codegen'd to.
109class DataLayout {
110private:
111  /// Defaults to false.
112  bool BigEndian;
113
114  unsigned AllocaAddrSpace;
115  unsigned StackNaturalAlign;
116
117  enum ManglingModeT {
118    MM_None,
119    MM_ELF,
120    MM_MachO,
121    MM_WinCOFF,
122    MM_WinCOFFX86,
123    MM_Mips
124  };
125  ManglingModeT ManglingMode;
126
127  SmallVector<unsigned char, 8> LegalIntWidths;
128
129  /// \brief Primitive type alignment data. This is sorted by type and bit
130  /// width during construction.
131  using AlignmentsTy = SmallVector<LayoutAlignElem, 16>;
132  AlignmentsTy Alignments;
133
134  AlignmentsTy::const_iterator
135  findAlignmentLowerBound(AlignTypeEnum AlignType, uint32_t BitWidth) const {
136    return const_cast<DataLayout *>(this)->findAlignmentLowerBound(AlignType,
137                                                                   BitWidth);
138  }
139
140  AlignmentsTy::iterator
141  findAlignmentLowerBound(AlignTypeEnum AlignType, uint32_t BitWidth);
142
143  /// \brief The string representation used to create this DataLayout
144  std::string StringRepresentation;
145
146  using PointersTy = SmallVector<PointerAlignElem, 8>;
147  PointersTy Pointers;
148
149  PointersTy::const_iterator
150  findPointerLowerBound(uint32_t AddressSpace) const {
151    return const_cast<DataLayout *>(this)->findPointerLowerBound(AddressSpace);
152  }
153
154  PointersTy::iterator findPointerLowerBound(uint32_t AddressSpace);
155
156  // The StructType -> StructLayout map.
157  mutable void *LayoutMap = nullptr;
158
159  /// Pointers in these address spaces are non-integral, and don't have a
160  /// well-defined bitwise representation.
161  SmallVector<unsigned, 8> NonIntegralAddressSpaces;
162
163  void setAlignment(AlignTypeEnum align_type, unsigned abi_align,
164                    unsigned pref_align, uint32_t bit_width);
165  unsigned getAlignmentInfo(AlignTypeEnum align_type, uint32_t bit_width,
166                            bool ABIAlign, Type *Ty) const;
167  void setPointerAlignment(uint32_t AddrSpace, unsigned ABIAlign,
168                           unsigned PrefAlign, uint32_t TypeByteWidth);
169
170  /// Internal helper method that returns requested alignment for type.
171  unsigned getAlignment(Type *Ty, bool abi_or_pref) const;
172
173  /// Parses a target data specification string. Assert if the string is
174  /// malformed.
175  void parseSpecifier(StringRef LayoutDescription);
176
177  // Free all internal data structures.
178  void clear();
179
180public:
181  /// Constructs a DataLayout from a specification string. See reset().
182  explicit DataLayout(StringRef LayoutDescription) {
183    reset(LayoutDescription);
184  }
185
186  /// Initialize target data from properties stored in the module.
187  explicit DataLayout(const Module *M);
188
189  DataLayout(const DataLayout &DL) { *this = DL; }
190
191  ~DataLayout(); // Not virtual, do not subclass this class
192
193  DataLayout &operator=(const DataLayout &DL) {
194    clear();
195    StringRepresentation = DL.StringRepresentation;
196    BigEndian = DL.isBigEndian();
197    AllocaAddrSpace = DL.AllocaAddrSpace;
198    StackNaturalAlign = DL.StackNaturalAlign;
199    ManglingMode = DL.ManglingMode;
200    LegalIntWidths = DL.LegalIntWidths;
201    Alignments = DL.Alignments;
202    Pointers = DL.Pointers;
203    NonIntegralAddressSpaces = DL.NonIntegralAddressSpaces;
204    return *this;
205  }
206
207  bool operator==(const DataLayout &Other) const;
208  bool operator!=(const DataLayout &Other) const { return !(*this == Other); }
209
210  void init(const Module *M);
211
212  /// Parse a data layout string (with fallback to default values).
213  void reset(StringRef LayoutDescription);
214
215  /// Layout endianness...
216  bool isLittleEndian() const { return !BigEndian; }
217  bool isBigEndian() const { return BigEndian; }
218
219  /// \brief Returns the string representation of the DataLayout.
220  ///
221  /// This representation is in the same format accepted by the string
222  /// constructor above. This should not be used to compare two DataLayout as
223  /// different string can represent the same layout.
224  const std::string &getStringRepresentation() const {
225    return StringRepresentation;
226  }
227
228  /// \brief Test if the DataLayout was constructed from an empty string.
229  bool isDefault() const { return StringRepresentation.empty(); }
230
231  /// \brief Returns true if the specified type is known to be a native integer
232  /// type supported by the CPU.
233  ///
234  /// For example, i64 is not native on most 32-bit CPUs and i37 is not native
235  /// on any known one. This returns false if the integer width is not legal.
236  ///
237  /// The width is specified in bits.
238  bool isLegalInteger(uint64_t Width) const {
239    for (unsigned LegalIntWidth : LegalIntWidths)
240      if (LegalIntWidth == Width)
241        return true;
242    return false;
243  }
244
245  bool isIllegalInteger(uint64_t Width) const { return !isLegalInteger(Width); }
246
247  /// Returns true if the given alignment exceeds the natural stack alignment.
248  bool exceedsNaturalStackAlignment(unsigned Align) const {
249    return (StackNaturalAlign != 0) && (Align > StackNaturalAlign);
250  }
251
252  unsigned getStackAlignment() const { return StackNaturalAlign; }
253  unsigned getAllocaAddrSpace() const { return AllocaAddrSpace; }
254
255  bool hasMicrosoftFastStdCallMangling() const {
256    return ManglingMode == MM_WinCOFFX86;
257  }
258
259  bool hasLinkerPrivateGlobalPrefix() const { return ManglingMode == MM_MachO; }
260
261  StringRef getLinkerPrivateGlobalPrefix() const {
262    if (ManglingMode == MM_MachO)
263      return "l";
264    return "";
265  }
266
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
281  StringRef getPrivateGlobalPrefix() const {
282    switch (ManglingMode) {
283    case MM_None:
284      return "";
285    case MM_ELF:
286    case MM_WinCOFF:
287      return ".L";
288    case MM_Mips:
289      return "$";
290    case MM_MachO:
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.
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  /// Return the address spaces containing non-integral pointers.  Pointers in
327  /// this address space don't have a well-defined bitwise representation.
328  ArrayRef<unsigned> getNonIntegralAddressSpaces() const {
329    return NonIntegralAddressSpaces;
330  }
331
332  bool isNonIntegralPointerType(PointerType *PT) const {
333    ArrayRef<unsigned> NonIntegralSpaces = getNonIntegralAddressSpaces();
334    return find(NonIntegralSpaces, PT->getAddressSpace()) !=
335           NonIntegralSpaces.end();
336  }
337
338  bool isNonIntegralPointerType(Type *Ty) const {
339    auto *PTy = dyn_cast<PointerType>(Ty);
340    return PTy && isNonIntegralPointerType(PTy);
341  }
342
343  /// Layout pointer size, in bits
344  /// FIXME: The defaults need to be removed once all of
345  /// the backends/clients are updated.
346  unsigned getPointerSizeInBits(unsigned AS = 0) const {
347    return getPointerSize(AS) * 8;
348  }
349
350  /// Layout pointer size, in bits, based on the type.  If this function is
351  /// called with a pointer type, then the type size of the pointer is returned.
352  /// If this function is called with a vector of pointers, then the type size
353  /// of the pointer is returned.  This should only be called with a pointer or
354  /// vector of pointers.
355  unsigned getPointerTypeSizeInBits(Type *) const;
356
357  unsigned getPointerTypeSize(Type *Ty) const {
358    return getPointerTypeSizeInBits(Ty) / 8;
359  }
360
361  /// Size examples:
362  ///
363  /// Type        SizeInBits  StoreSizeInBits  AllocSizeInBits[*]
364  /// ----        ----------  ---------------  ---------------
365  ///  i1            1           8                8
366  ///  i8            8           8                8
367  ///  i19          19          24               32
368  ///  i32          32          32               32
369  ///  i100        100         104              128
370  ///  i128        128         128              128
371  ///  Float        32          32               32
372  ///  Double       64          64               64
373  ///  X86_FP80     80          80               96
374  ///
375  /// [*] The alloc size depends on the alignment, and thus on the target.
376  ///     These values are for x86-32 linux.
377
378  /// \brief Returns the number of bits necessary to hold the specified type.
379  ///
380  /// For example, returns 36 for i36 and 80 for x86_fp80. The type passed must
381  /// have a size (Type::isSized() must return true).
382  uint64_t getTypeSizeInBits(Type *Ty) const;
383
384  /// \brief Returns the maximum number of bytes that may be overwritten by
385  /// storing the specified type.
386  ///
387  /// For example, returns 5 for i36 and 10 for x86_fp80.
388  uint64_t getTypeStoreSize(Type *Ty) const {
389    return (getTypeSizeInBits(Ty) + 7) / 8;
390  }
391
392  /// \brief Returns the maximum number of bits that may be overwritten by
393  /// storing the specified type; always a multiple of 8.
394  ///
395  /// For example, returns 40 for i36 and 80 for x86_fp80.
396  uint64_t getTypeStoreSizeInBits(Type *Ty) const {
397    return 8 * getTypeStoreSize(Ty);
398  }
399
400  /// \brief Returns the offset in bytes between successive objects of the
401  /// specified type, including alignment padding.
402  ///
403  /// This is the amount that alloca reserves for this type. For example,
404  /// returns 12 or 16 for x86_fp80, depending on alignment.
405  uint64_t getTypeAllocSize(Type *Ty) const {
406    // Round up to the next alignment boundary.
407    return alignTo(getTypeStoreSize(Ty), getABITypeAlignment(Ty));
408  }
409
410  /// \brief Returns the offset in bits between successive objects of the
411  /// specified type, including alignment padding; always a multiple of 8.
412  ///
413  /// This is the amount that alloca reserves for this type. For example,
414  /// returns 96 or 128 for x86_fp80, depending on alignment.
415  uint64_t getTypeAllocSizeInBits(Type *Ty) const {
416    return 8 * getTypeAllocSize(Ty);
417  }
418
419  /// \brief Returns the minimum ABI-required alignment for the specified type.
420  unsigned getABITypeAlignment(Type *Ty) const;
421
422  /// \brief Returns the minimum ABI-required alignment for an integer type of
423  /// the specified bitwidth.
424  unsigned getABIIntegerTypeAlignment(unsigned BitWidth) const;
425
426  /// \brief Returns the preferred stack/global alignment for the specified
427  /// type.
428  ///
429  /// This is always at least as good as the ABI alignment.
430  unsigned getPrefTypeAlignment(Type *Ty) const;
431
432  /// \brief Returns the preferred alignment for the specified type, returned as
433  /// log2 of the value (a shift amount).
434  unsigned getPreferredTypeAlignmentShift(Type *Ty) const;
435
436  /// \brief Returns an integer type with size at least as big as that of a
437  /// pointer in the given address space.
438  IntegerType *getIntPtrType(LLVMContext &C, unsigned AddressSpace = 0) const;
439
440  /// \brief Returns an integer (vector of integer) type with size at least as
441  /// big as that of a pointer of the given pointer (vector of pointer) type.
442  Type *getIntPtrType(Type *) const;
443
444  /// \brief Returns the smallest integer type with size at least as big as
445  /// Width bits.
446  Type *getSmallestLegalIntType(LLVMContext &C, unsigned Width = 0) const;
447
448  /// \brief Returns the largest legal integer type, or null if none are set.
449  Type *getLargestLegalIntType(LLVMContext &C) const {
450    unsigned LargestSize = getLargestLegalIntTypeSizeInBits();
451    return (LargestSize == 0) ? nullptr : Type::getIntNTy(C, LargestSize);
452  }
453
454  /// \brief Returns the size of largest legal integer type size, or 0 if none
455  /// are set.
456  unsigned getLargestLegalIntTypeSizeInBits() const;
457
458  /// \brief Returns the offset from the beginning of the type for the specified
459  /// indices.
460  ///
461  /// Note that this takes the element type, not the pointer type.
462  /// This is used to implement getelementptr.
463  int64_t getIndexedOffsetInType(Type *ElemTy, ArrayRef<Value *> Indices) const;
464
465  /// \brief Returns a StructLayout object, indicating the alignment of the
466  /// struct, its size, and the offsets of its fields.
467  ///
468  /// Note that this information is lazily cached.
469  const StructLayout *getStructLayout(StructType *Ty) const;
470
471  /// \brief Returns the preferred alignment of the specified global.
472  ///
473  /// This includes an explicitly requested alignment (if the global has one).
474  unsigned getPreferredAlignment(const GlobalVariable *GV) const;
475
476  /// \brief Returns the preferred alignment of the specified global, returned
477  /// in log form.
478  ///
479  /// This includes an explicitly requested alignment (if the global has one).
480  unsigned getPreferredAlignmentLog(const GlobalVariable *GV) const;
481};
482
483inline DataLayout *unwrap(LLVMTargetDataRef P) {
484  return reinterpret_cast<DataLayout *>(P);
485}
486
487inline LLVMTargetDataRef wrap(const DataLayout *P) {
488  return reinterpret_cast<LLVMTargetDataRef>(const_cast<DataLayout *>(P));
489}
490
491/// Used to lazily calculate structure layout information for a target machine,
492/// based on the DataLayout structure.
493class StructLayout {
494  uint64_t StructSize;
495  unsigned StructAlignment;
496  unsigned IsPadded : 1;
497  unsigned NumElements : 31;
498  uint64_t MemberOffsets[1]; // variable sized array!
499
500public:
501  uint64_t getSizeInBytes() const { return StructSize; }
502
503  uint64_t getSizeInBits() const { return 8 * StructSize; }
504
505  unsigned getAlignment() const { return StructAlignment; }
506
507  /// Returns whether the struct has padding or not between its fields.
508  /// NB: Padding in nested element is not taken into account.
509  bool hasPadding() const { return IsPadded; }
510
511  /// \brief Given a valid byte offset into the structure, returns the structure
512  /// index that contains it.
513  unsigned getElementContainingOffset(uint64_t Offset) const;
514
515  uint64_t getElementOffset(unsigned Idx) const {
516    assert(Idx < NumElements && "Invalid element idx!");
517    return MemberOffsets[Idx];
518  }
519
520  uint64_t getElementOffsetInBits(unsigned Idx) const {
521    return getElementOffset(Idx) * 8;
522  }
523
524private:
525  friend class DataLayout; // Only DataLayout can create this class
526
527  StructLayout(StructType *ST, const DataLayout &DL);
528};
529
530// The implementation of this method is provided inline as it is particularly
531// well suited to constant folding when called on a specific Type subclass.
532inline uint64_t DataLayout::getTypeSizeInBits(Type *Ty) const {
533  assert(Ty->isSized() && "Cannot getTypeInfo() on a type that is unsized!");
534  switch (Ty->getTypeID()) {
535  case Type::LabelTyID:
536    return getPointerSizeInBits(0);
537  case Type::PointerTyID:
538    return getPointerSizeInBits(Ty->getPointerAddressSpace());
539  case Type::ArrayTyID: {
540    ArrayType *ATy = cast<ArrayType>(Ty);
541    return ATy->getNumElements() *
542           getTypeAllocSizeInBits(ATy->getElementType());
543  }
544  case Type::StructTyID:
545    // Get the layout annotation... which is lazily created on demand.
546    return getStructLayout(cast<StructType>(Ty))->getSizeInBits();
547  case Type::IntegerTyID:
548    return Ty->getIntegerBitWidth();
549  case Type::HalfTyID:
550    return 16;
551  case Type::FloatTyID:
552    return 32;
553  case Type::DoubleTyID:
554  case Type::X86_MMXTyID:
555    return 64;
556  case Type::PPC_FP128TyID:
557  case Type::FP128TyID:
558    return 128;
559  // In memory objects this is always aligned to a higher boundary, but
560  // only 80 bits contain information.
561  case Type::X86_FP80TyID:
562    return 80;
563  case Type::VectorTyID: {
564    VectorType *VTy = cast<VectorType>(Ty);
565    return VTy->getNumElements() * getTypeSizeInBits(VTy->getElementType());
566  }
567  default:
568    llvm_unreachable("DataLayout::getTypeSizeInBits(): Unsupported type");
569  }
570}
571
572} // end namespace llvm
573
574#endif // LLVM_IR_DATALAYOUT_H
575