• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2014 the V8 project authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
4 
5 #ifndef V8_BASE_BITS_H_
6 #define V8_BASE_BITS_H_
7 
8 #include <stdint.h>
9 #include <type_traits>
10 
11 #include "src/base/base-export.h"
12 #include "src/base/macros.h"
13 #if V8_CC_MSVC
14 #include <intrin.h>
15 #endif
16 #if V8_OS_WIN32
17 #include "src/base/win32-headers.h"
18 #endif
19 
20 namespace v8 {
21 namespace base {
22 namespace bits {
23 
24 // CountPopulation(value) returns the number of bits set in |value|.
25 template <typename T>
26 constexpr inline
27     typename std::enable_if<std::is_unsigned<T>::value && sizeof(T) <= 8,
28                             unsigned>::type
CountPopulation(T value)29     CountPopulation(T value) {
30   STATIC_ASSERT(sizeof(T) <= 8);
31 #if V8_HAS_BUILTIN_POPCOUNT
32   return sizeof(T) == 8 ? __builtin_popcountll(static_cast<uint64_t>(value))
33                         : __builtin_popcount(static_cast<uint32_t>(value));
34 #else
35   // Fall back to divide-and-conquer popcount (see "Hacker's Delight" by Henry
36   // S. Warren, Jr.), chapter 5-1.
37   constexpr uint64_t mask[] = {0x5555555555555555, 0x3333333333333333,
38                                0x0f0f0f0f0f0f0f0f};
39   // Start with 64 buckets of 1 bits, holding values from [0,1].
40   value = ((value >> 1) & mask[0]) + (value & mask[0]);
41   // Having 32 buckets of 2 bits, holding values from [0,2] now.
42   value = ((value >> 2) & mask[1]) + (value & mask[1]);
43   // Having 16 buckets of 4 bits, holding values from [0,4] now.
44   value = ((value >> 4) & mask[2]) + (value & mask[2]);
45   // Having 8 buckets of 8 bits, holding values from [0,8] now.
46   // From this point on, the buckets are bigger than the number of bits
47   // required to hold the values, and the buckets are bigger the maximum
48   // result, so there's no need to mask value anymore, since there's no
49   // more risk of overflow between buckets.
50   if (sizeof(T) > 1) value = (value >> (sizeof(T) > 1 ? 8 : 0)) + value;
51   // Having 4 buckets of 16 bits, holding values from [0,16] now.
52   if (sizeof(T) > 2) value = (value >> (sizeof(T) > 2 ? 16 : 0)) + value;
53   // Having 2 buckets of 32 bits, holding values from [0,32] now.
54   if (sizeof(T) > 4) value = (value >> (sizeof(T) > 4 ? 32 : 0)) + value;
55   // Having 1 buckets of 64 bits, holding values from [0,64] now.
56   return static_cast<unsigned>(value & 0xff);
57 #endif
58 }
59 
60 // ReverseBits(value) returns |value| in reverse bit order.
61 template <typename T>
ReverseBits(T value)62 T ReverseBits(T value) {
63   STATIC_ASSERT((sizeof(value) == 1) || (sizeof(value) == 2) ||
64                 (sizeof(value) == 4) || (sizeof(value) == 8));
65   T result = 0;
66   for (unsigned i = 0; i < (sizeof(value) * 8); i++) {
67     result = (result << 1) | (value & 1);
68     value >>= 1;
69   }
70   return result;
71 }
72 
73 // CountLeadingZeros(value) returns the number of zero bits following the most
74 // significant 1 bit in |value| if |value| is non-zero, otherwise it returns
75 // {sizeof(T) * 8}.
76 template <typename T, unsigned bits = sizeof(T) * 8>
77 inline constexpr
78     typename std::enable_if<std::is_unsigned<T>::value && sizeof(T) <= 8,
79                             unsigned>::type
CountLeadingZeros(T value)80     CountLeadingZeros(T value) {
81   static_assert(bits > 0, "invalid instantiation");
82 #if V8_HAS_BUILTIN_CLZ
83   return value == 0
84              ? bits
85              : bits == 64
86                    ? __builtin_clzll(static_cast<uint64_t>(value))
87                    : __builtin_clz(static_cast<uint32_t>(value)) - (32 - bits);
88 #else
89   // Binary search algorithm taken from "Hacker's Delight" (by Henry S. Warren,
90   // Jr.), figures 5-11 and 5-12.
91   if (bits == 1) return static_cast<unsigned>(value) ^ 1;
92   T upper_half = value >> (bits / 2);
93   T next_value = upper_half != 0 ? upper_half : value;
94   unsigned add = upper_half != 0 ? 0 : bits / 2;
95   constexpr unsigned next_bits = bits == 1 ? 1 : bits / 2;
96   return CountLeadingZeros<T, next_bits>(next_value) + add;
97 #endif
98 }
99 
CountLeadingZeros32(uint32_t value)100 inline constexpr unsigned CountLeadingZeros32(uint32_t value) {
101   return CountLeadingZeros(value);
102 }
CountLeadingZeros64(uint64_t value)103 inline constexpr unsigned CountLeadingZeros64(uint64_t value) {
104   return CountLeadingZeros(value);
105 }
106 
107 // CountTrailingZeros(value) returns the number of zero bits preceding the
108 // least significant 1 bit in |value| if |value| is non-zero, otherwise it
109 // returns {sizeof(T) * 8}.
110 // See CountTrailingZerosNonZero for an optimized version for the case that
111 // |value| is guaranteed to be non-zero.
112 template <typename T, unsigned bits = sizeof(T) * 8>
113 inline constexpr
114     typename std::enable_if<std::is_integral<T>::value && sizeof(T) <= 8,
115                             unsigned>::type
CountTrailingZeros(T value)116     CountTrailingZeros(T value) {
117 #if V8_HAS_BUILTIN_CTZ
118   return value == 0 ? bits
119                     : bits == 64 ? __builtin_ctzll(static_cast<uint64_t>(value))
120                                  : __builtin_ctz(static_cast<uint32_t>(value));
121 #else
122   // Fall back to popcount (see "Hacker's Delight" by Henry S. Warren, Jr.),
123   // chapter 5-4. On x64, since is faster than counting in a loop and faster
124   // than doing binary search.
125   using U = typename std::make_unsigned<T>::type;
126   U u = value;
127   return CountPopulation(static_cast<U>(~u & (u - 1u)));
128 #endif
129 }
130 
CountTrailingZeros32(uint32_t value)131 inline constexpr unsigned CountTrailingZeros32(uint32_t value) {
132   return CountTrailingZeros(value);
133 }
CountTrailingZeros64(uint64_t value)134 inline constexpr unsigned CountTrailingZeros64(uint64_t value) {
135   return CountTrailingZeros(value);
136 }
137 
138 // CountTrailingZerosNonZero(value) returns the number of zero bits preceding
139 // the least significant 1 bit in |value| if |value| is non-zero, otherwise the
140 // behavior is undefined.
141 // See CountTrailingZeros for an alternative version that allows |value| == 0.
142 template <typename T, unsigned bits = sizeof(T) * 8>
143 inline constexpr
144     typename std::enable_if<std::is_integral<T>::value && sizeof(T) <= 8,
145                             unsigned>::type
CountTrailingZerosNonZero(T value)146     CountTrailingZerosNonZero(T value) {
147   DCHECK_NE(0, value);
148 #if V8_HAS_BUILTIN_CTZ
149   return bits == 64 ? __builtin_ctzll(static_cast<uint64_t>(value))
150                     : __builtin_ctz(static_cast<uint32_t>(value));
151 #else
152   return CountTrailingZeros<T, bits>(value);
153 #endif
154 }
155 
156 // Returns true iff |value| is a power of 2.
157 template <typename T,
158           typename = typename std::enable_if<std::is_integral<T>::value ||
159                                              std::is_enum<T>::value>::type>
IsPowerOfTwo(T value)160 constexpr inline bool IsPowerOfTwo(T value) {
161   return value > 0 && (value & (value - 1)) == 0;
162 }
163 
164 // Identical to {CountTrailingZeros}, but only works for powers of 2.
165 template <typename T,
166           typename = typename std::enable_if<std::is_integral<T>::value>::type>
WhichPowerOfTwo(T value)167 inline constexpr int WhichPowerOfTwo(T value) {
168   DCHECK(IsPowerOfTwo(value));
169 #if V8_HAS_BUILTIN_CTZ
170   STATIC_ASSERT(sizeof(T) <= 8);
171   return sizeof(T) == 8 ? __builtin_ctzll(static_cast<uint64_t>(value))
172                         : __builtin_ctz(static_cast<uint32_t>(value));
173 #else
174   // Fall back to popcount (see "Hacker's Delight" by Henry S. Warren, Jr.),
175   // chapter 5-4. On x64, since is faster than counting in a loop and faster
176   // than doing binary search.
177   using U = typename std::make_unsigned<T>::type;
178   U u = value;
179   return CountPopulation(static_cast<U>(u - 1));
180 #endif
181 }
182 
183 // RoundUpToPowerOfTwo32(value) returns the smallest power of two which is
184 // greater than or equal to |value|. If you pass in a |value| that is already a
185 // power of two, it is returned as is. |value| must be less than or equal to
186 // 0x80000000u. Uses computation based on leading zeros if we have compiler
187 // support for that. Falls back to the implementation from "Hacker's Delight" by
188 // Henry S. Warren, Jr., figure 3-3, page 48, where the function is called clp2.
189 V8_BASE_EXPORT uint32_t RoundUpToPowerOfTwo32(uint32_t value);
190 // Same for 64 bit integers. |value| must be <= 2^63
191 V8_BASE_EXPORT uint64_t RoundUpToPowerOfTwo64(uint64_t value);
192 // Same for size_t integers.
RoundUpToPowerOfTwo(size_t value)193 inline size_t RoundUpToPowerOfTwo(size_t value) {
194   if (sizeof(size_t) == sizeof(uint64_t)) {
195     return RoundUpToPowerOfTwo64(value);
196   } else {
197     // Without windows.h included this line triggers a truncation warning on
198     // 64-bit builds. Presumably windows.h disables the relevant warning.
199     return RoundUpToPowerOfTwo32(static_cast<uint32_t>(value));
200   }
201 }
202 
203 // RoundDownToPowerOfTwo32(value) returns the greatest power of two which is
204 // less than or equal to |value|. If you pass in a |value| that is already a
205 // power of two, it is returned as is.
RoundDownToPowerOfTwo32(uint32_t value)206 inline uint32_t RoundDownToPowerOfTwo32(uint32_t value) {
207   if (value > 0x80000000u) return 0x80000000u;
208   uint32_t result = RoundUpToPowerOfTwo32(value);
209   if (result > value) result >>= 1;
210   return result;
211 }
212 
213 
214 // Precondition: 0 <= shift < 32
RotateRight32(uint32_t value,uint32_t shift)215 inline constexpr uint32_t RotateRight32(uint32_t value, uint32_t shift) {
216   return (value >> shift) | (value << ((32 - shift) & 31));
217 }
218 
219 // Precondition: 0 <= shift < 32
RotateLeft32(uint32_t value,uint32_t shift)220 inline constexpr uint32_t RotateLeft32(uint32_t value, uint32_t shift) {
221   return (value << shift) | (value >> ((32 - shift) & 31));
222 }
223 
224 // Precondition: 0 <= shift < 64
RotateRight64(uint64_t value,uint64_t shift)225 inline constexpr uint64_t RotateRight64(uint64_t value, uint64_t shift) {
226   return (value >> shift) | (value << ((64 - shift) & 63));
227 }
228 
229 // Precondition: 0 <= shift < 64
RotateLeft64(uint64_t value,uint64_t shift)230 inline constexpr uint64_t RotateLeft64(uint64_t value, uint64_t shift) {
231   return (value << shift) | (value >> ((64 - shift) & 63));
232 }
233 
234 // SignedAddOverflow32(lhs,rhs,val) performs a signed summation of |lhs| and
235 // |rhs| and stores the result into the variable pointed to by |val| and
236 // returns true if the signed summation resulted in an overflow.
SignedAddOverflow32(int32_t lhs,int32_t rhs,int32_t * val)237 inline bool SignedAddOverflow32(int32_t lhs, int32_t rhs, int32_t* val) {
238 #if V8_HAS_BUILTIN_SADD_OVERFLOW
239   return __builtin_sadd_overflow(lhs, rhs, val);
240 #else
241   uint32_t res = static_cast<uint32_t>(lhs) + static_cast<uint32_t>(rhs);
242   *val = bit_cast<int32_t>(res);
243   return ((res ^ lhs) & (res ^ rhs) & (1U << 31)) != 0;
244 #endif
245 }
246 
247 
248 // SignedSubOverflow32(lhs,rhs,val) performs a signed subtraction of |lhs| and
249 // |rhs| and stores the result into the variable pointed to by |val| and
250 // returns true if the signed subtraction resulted in an overflow.
SignedSubOverflow32(int32_t lhs,int32_t rhs,int32_t * val)251 inline bool SignedSubOverflow32(int32_t lhs, int32_t rhs, int32_t* val) {
252 #if V8_HAS_BUILTIN_SSUB_OVERFLOW
253   return __builtin_ssub_overflow(lhs, rhs, val);
254 #else
255   uint32_t res = static_cast<uint32_t>(lhs) - static_cast<uint32_t>(rhs);
256   *val = bit_cast<int32_t>(res);
257   return ((res ^ lhs) & (res ^ ~rhs) & (1U << 31)) != 0;
258 #endif
259 }
260 
261 // SignedMulOverflow32(lhs,rhs,val) performs a signed multiplication of |lhs|
262 // and |rhs| and stores the result into the variable pointed to by |val| and
263 // returns true if the signed multiplication resulted in an overflow.
264 V8_BASE_EXPORT bool SignedMulOverflow32(int32_t lhs, int32_t rhs, int32_t* val);
265 
266 // SignedAddOverflow64(lhs,rhs,val) performs a signed summation of |lhs| and
267 // |rhs| and stores the result into the variable pointed to by |val| and
268 // returns true if the signed summation resulted in an overflow.
SignedAddOverflow64(int64_t lhs,int64_t rhs,int64_t * val)269 inline bool SignedAddOverflow64(int64_t lhs, int64_t rhs, int64_t* val) {
270   uint64_t res = static_cast<uint64_t>(lhs) + static_cast<uint64_t>(rhs);
271   *val = bit_cast<int64_t>(res);
272   return ((res ^ lhs) & (res ^ rhs) & (1ULL << 63)) != 0;
273 }
274 
275 
276 // SignedSubOverflow64(lhs,rhs,val) performs a signed subtraction of |lhs| and
277 // |rhs| and stores the result into the variable pointed to by |val| and
278 // returns true if the signed subtraction resulted in an overflow.
SignedSubOverflow64(int64_t lhs,int64_t rhs,int64_t * val)279 inline bool SignedSubOverflow64(int64_t lhs, int64_t rhs, int64_t* val) {
280   uint64_t res = static_cast<uint64_t>(lhs) - static_cast<uint64_t>(rhs);
281   *val = bit_cast<int64_t>(res);
282   return ((res ^ lhs) & (res ^ ~rhs) & (1ULL << 63)) != 0;
283 }
284 
285 // SignedMulHigh32(lhs, rhs) multiplies two signed 32-bit values |lhs| and
286 // |rhs|, extracts the most significant 32 bits of the result, and returns
287 // those.
288 V8_BASE_EXPORT int32_t SignedMulHigh32(int32_t lhs, int32_t rhs);
289 
290 // SignedMulHighAndAdd32(lhs, rhs, acc) multiplies two signed 32-bit values
291 // |lhs| and |rhs|, extracts the most significant 32 bits of the result, and
292 // adds the accumulate value |acc|.
293 V8_BASE_EXPORT int32_t SignedMulHighAndAdd32(int32_t lhs, int32_t rhs,
294                                              int32_t acc);
295 
296 // SignedDiv32(lhs, rhs) divides |lhs| by |rhs| and returns the quotient
297 // truncated to int32. If |rhs| is zero, then zero is returned. If |lhs|
298 // is minint and |rhs| is -1, it returns minint.
299 V8_BASE_EXPORT int32_t SignedDiv32(int32_t lhs, int32_t rhs);
300 
301 // SignedMod32(lhs, rhs) divides |lhs| by |rhs| and returns the remainder
302 // truncated to int32. If either |rhs| is zero or |lhs| is minint and |rhs|
303 // is -1, it returns zero.
304 V8_BASE_EXPORT int32_t SignedMod32(int32_t lhs, int32_t rhs);
305 
306 // UnsignedAddOverflow32(lhs,rhs,val) performs an unsigned summation of |lhs|
307 // and |rhs| and stores the result into the variable pointed to by |val| and
308 // returns true if the unsigned summation resulted in an overflow.
UnsignedAddOverflow32(uint32_t lhs,uint32_t rhs,uint32_t * val)309 inline bool UnsignedAddOverflow32(uint32_t lhs, uint32_t rhs, uint32_t* val) {
310 #if V8_HAS_BUILTIN_SADD_OVERFLOW
311   return __builtin_uadd_overflow(lhs, rhs, val);
312 #else
313   *val = lhs + rhs;
314   return *val < (lhs | rhs);
315 #endif
316 }
317 
318 
319 // UnsignedDiv32(lhs, rhs) divides |lhs| by |rhs| and returns the quotient
320 // truncated to uint32. If |rhs| is zero, then zero is returned.
UnsignedDiv32(uint32_t lhs,uint32_t rhs)321 inline uint32_t UnsignedDiv32(uint32_t lhs, uint32_t rhs) {
322   return rhs ? lhs / rhs : 0u;
323 }
324 
325 
326 // UnsignedMod32(lhs, rhs) divides |lhs| by |rhs| and returns the remainder
327 // truncated to uint32. If |rhs| is zero, then zero is returned.
UnsignedMod32(uint32_t lhs,uint32_t rhs)328 inline uint32_t UnsignedMod32(uint32_t lhs, uint32_t rhs) {
329   return rhs ? lhs % rhs : 0u;
330 }
331 
332 // Wraparound integer arithmetic without undefined behavior.
333 
WraparoundAdd32(int32_t lhs,int32_t rhs)334 inline int32_t WraparoundAdd32(int32_t lhs, int32_t rhs) {
335   return static_cast<int32_t>(static_cast<uint32_t>(lhs) +
336                               static_cast<uint32_t>(rhs));
337 }
338 
WraparoundNeg32(int32_t x)339 inline int32_t WraparoundNeg32(int32_t x) {
340   return static_cast<int32_t>(-static_cast<uint32_t>(x));
341 }
342 
343 // SignedSaturatedAdd64(lhs, rhs) adds |lhs| and |rhs|,
344 // checks and returns the result.
345 V8_BASE_EXPORT int64_t SignedSaturatedAdd64(int64_t lhs, int64_t rhs);
346 
347 // SignedSaturatedSub64(lhs, rhs) subtracts |lhs| by |rhs|,
348 // checks and returns the result.
349 V8_BASE_EXPORT int64_t SignedSaturatedSub64(int64_t lhs, int64_t rhs);
350 
351 }  // namespace bits
352 }  // namespace base
353 }  // namespace v8
354 
355 #endif  // V8_BASE_BITS_H_
356