• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //===- subzero/src/IceUtils.h - Utility functions ---------------*- C++ -*-===//
2 //
3 //                        The Subzero Code Generator
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 ///
10 /// \file
11 /// \brief Defines some utility functions.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #ifndef SUBZERO_SRC_ICEUTILS_H
16 #define SUBZERO_SRC_ICEUTILS_H
17 
18 #include <climits>
19 #include <cmath> // std::signbit()
20 
21 namespace Ice {
22 namespace Utils {
23 
24 /// Allows copying from types of unrelated sizes. This method was introduced to
25 /// enable the strict aliasing optimizations of GCC 4.4. Basically, GCC
26 /// mindlessly relies on obscure details in the C++ standard that make
27 /// reinterpret_cast virtually useless.
bitCopy(const S & Source)28 template <typename D, typename S> inline D bitCopy(const S &Source) {
29   static_assert(sizeof(D) <= sizeof(S),
30                 "bitCopy between incompatible type widths");
31   static_assert(!std::is_pointer<S>::value, "");
32   D Destination;
33   // This use of memcpy is safe: source and destination cannot overlap.
34   memcpy(&Destination, reinterpret_cast<const void *>(&Source), sizeof(D));
35   return Destination;
36 }
37 
38 /// Check whether an N-bit two's-complement representation can hold value.
IsInt(int N,T value)39 template <typename T> inline bool IsInt(int N, T value) {
40   assert((0 < N) &&
41          (static_cast<unsigned int>(N) < (CHAR_BIT * sizeof(value))));
42   T limit = static_cast<T>(1) << (N - 1);
43   return (-limit <= value) && (value < limit);
44 }
45 
IsUint(int N,T value)46 template <typename T> inline bool IsUint(int N, T value) {
47   assert((0 < N) &&
48          (static_cast<unsigned int>(N) < (CHAR_BIT * sizeof(value))));
49   T limit = static_cast<T>(1) << N;
50   return (0 <= value) && (value < limit);
51 }
52 
53 /// Check whether the magnitude of value fits in N bits, i.e., whether an
54 /// (N+1)-bit sign-magnitude representation can hold value.
IsAbsoluteUint(int N,T Value)55 template <typename T> inline bool IsAbsoluteUint(int N, T Value) {
56   assert((0 < N) &&
57          (static_cast<unsigned int>(N) < (CHAR_BIT * sizeof(Value))));
58   if (Value < 0)
59     Value = -Value;
60   return IsUint(N, Value);
61 }
62 
63 /// Return true if the addition X + Y will cause integer overflow for integers
64 /// of type T.
WouldOverflowAdd(T X,T Y)65 template <typename T> inline bool WouldOverflowAdd(T X, T Y) {
66   return ((X > 0 && Y > 0 && (X > std::numeric_limits<T>::max() - Y)) ||
67           (X < 0 && Y < 0 && (X < std::numeric_limits<T>::min() - Y)));
68 }
69 
70 /// Adds x to y and stores the result in sum. Returns true if the addition
71 /// overflowed.
add_overflow(uint32_t x,uint32_t y,uint32_t * sum)72 inline bool add_overflow(uint32_t x, uint32_t y, uint32_t *sum) {
73   static_assert(std::is_same<uint32_t, unsigned>::value, "Must match type");
74 #if __has_builtin(__builtin_uadd_overflow)
75   return __builtin_uadd_overflow(x, y, sum);
76 #else
77   *sum = x + y;
78   return WouldOverflowAdd(x, y);
79 #endif
80 }
81 
82 /// Return true if X is already aligned by N, where N is a power of 2.
IsAligned(T X,intptr_t N)83 template <typename T> inline bool IsAligned(T X, intptr_t N) {
84   assert(llvm::isPowerOf2_64(N));
85   return (X & (N - 1)) == 0;
86 }
87 
88 /// Return Value adjusted to the next highest multiple of Alignment.
applyAlignment(uint32_t Value,uint32_t Alignment)89 inline uint32_t applyAlignment(uint32_t Value, uint32_t Alignment) {
90   assert(llvm::isPowerOf2_32(Alignment));
91   return (Value + Alignment - 1) & -Alignment;
92 }
93 
94 /// Return amount which must be added to adjust Pos to the next highest
95 /// multiple of Align.
OffsetToAlignment(uint64_t Pos,uint64_t Align)96 inline uint64_t OffsetToAlignment(uint64_t Pos, uint64_t Align) {
97   assert(llvm::isPowerOf2_64(Align));
98   uint64_t Mod = Pos & (Align - 1);
99   if (Mod == 0)
100     return 0;
101   return Align - Mod;
102 }
103 
104 /// Rotate the value bit pattern to the left by shift bits.
105 /// Precondition: 0 <= shift < 32
rotateLeft32(uint32_t value,uint32_t shift)106 inline uint32_t rotateLeft32(uint32_t value, uint32_t shift) {
107   if (shift == 0)
108     return value;
109   return (value << shift) | (value >> (32 - shift));
110 }
111 
112 /// Rotate the value bit pattern to the right by shift bits.
rotateRight32(uint32_t value,uint32_t shift)113 inline uint32_t rotateRight32(uint32_t value, uint32_t shift) {
114   if (shift == 0)
115     return value;
116   return (value >> shift) | (value << (32 - shift));
117 }
118 
119 /// Returns true if Val is +0.0. It requires T to be a floating point type.
isPositiveZero(T Val)120 template <typename T> bool isPositiveZero(T Val) {
121   static_assert(std::is_floating_point<T>::value,
122                 "Input type must be floating point");
123   return Val == 0 && !std::signbit(Val);
124 }
125 
126 /// Resize a vector (or other suitable container) to a particular size, and also
127 /// reserve possibly a larger size to avoid repeatedly recopying as the
128 /// container grows.  It uses a strategy of doubling capacity up to a certain
129 /// point, after which it bumps the capacity by a fixed amount.
130 template <typename Container>
131 inline void reserveAndResize(Container &V, uint32_t Size,
132                              uint32_t ChunkSizeBits = 10) {
133 #if __has_builtin(__builtin_clz)
134   // Don't call reserve() if Size==0.
135   if (Size > 0) {
136     uint32_t Mask;
137     if (Size <= (1 << ChunkSizeBits)) {
138       // For smaller sizes, reserve the smallest power of 2 greater than or
139       // equal to Size.
140       Mask =
141           ((1 << (CHAR_BIT * sizeof(uint32_t) - __builtin_clz(Size))) - 1) - 1;
142     } else {
143       // For larger sizes, round up to the smallest multiple of 1<<ChunkSizeBits
144       // greater than or equal to Size.
145       Mask = (1 << ChunkSizeBits) - 1;
146     }
147     V.reserve((Size + Mask) & ~Mask);
148   }
149 #endif
150   V.resize(Size);
151 }
152 
153 /// An RAII class to ensure that a boolean flag is restored to its previous
154 /// value upon function exit.
155 ///
156 /// Used in places like RandomizationPoolingPause and generating target helper
157 /// calls.
158 class BoolFlagSaver {
159   BoolFlagSaver() = delete;
160   BoolFlagSaver(const BoolFlagSaver &) = delete;
161   BoolFlagSaver &operator=(const BoolFlagSaver &) = delete;
162 
163 public:
BoolFlagSaver(bool & F,bool NewValue)164   BoolFlagSaver(bool &F, bool NewValue) : OldValue(F), Flag(F) { F = NewValue; }
~BoolFlagSaver()165   ~BoolFlagSaver() { Flag = OldValue; }
166 
167 private:
168   const bool OldValue;
169   bool &Flag;
170 };
171 
172 } // end of namespace Utils
173 } // end of namespace Ice
174 
175 #endif // SUBZERO_SRC_ICEUTILS_H
176