/root/doris/be/src/gutil/bits.h
Line | Count | Source (jump to first uncovered line) |
1 | | // Copyright 2002 and onwards Google Inc. |
2 | | // |
3 | | // A collection of useful (static) bit-twiddling functions. |
4 | | |
5 | | #pragma once |
6 | | |
7 | | |
8 | | #include "gutil/integral_types.h" |
9 | | // IWYU pragma: no_include <butil/macros.h> |
10 | | #include "gutil/macros.h" // IWYU pragma: keep |
11 | | |
12 | | class Bits { |
13 | | public: |
14 | | // Return the number of one bits in the given integer. |
15 | | static int CountOnesInByte(unsigned char n); |
16 | | |
17 | 0 | static int CountOnes(uint32 n) { |
18 | 0 | n -= ((n >> 1) & 0x55555555); |
19 | 0 | n = ((n >> 2) & 0x33333333) + (n & 0x33333333); |
20 | 0 | return (((n + (n >> 4)) & 0xF0F0F0F) * 0x1010101) >> 24; |
21 | 0 | } |
22 | | |
23 | | // Count bits using sideways addition [WWG'57]. See Knuth TAOCP v4 7.1.3(59) |
24 | 0 | static inline int CountOnes64(uint64 n) { |
25 | 0 | #if defined(__x86_64__) |
26 | 0 | n -= (n >> 1) & 0x5555555555555555ULL; |
27 | 0 | n = ((n >> 2) & 0x3333333333333333ULL) + (n & 0x3333333333333333ULL); |
28 | 0 | return (((n + (n >> 4)) & 0xF0F0F0F0F0F0F0FULL) * 0x101010101010101ULL) >> 56; |
29 | 0 | #else |
30 | 0 | return CountOnes(n >> 32) + CountOnes(n & 0xffffffff); |
31 | 0 | #endif |
32 | 0 | } |
33 | | |
34 | | // Count bits using popcnt instruction (available on argo machines). |
35 | | // Doesn't check if the instruction exists. |
36 | | // Please use TestCPUFeature(POPCNT) from base/cpuid/cpuid.h before using this. |
37 | 0 | static inline int CountOnes64withPopcount(uint64 n) { |
38 | 0 | #if defined(__x86_64__) && defined __GNUC__ |
39 | 0 | int64 count = 0; |
40 | 0 | asm("popcnt %1,%0" : "=r"(count) : "rm"(n) : "cc"); |
41 | 0 | return count; |
42 | 0 | #else |
43 | 0 | return CountOnes64(n); |
44 | 0 | #endif |
45 | 0 | } |
46 | | |
47 | | // Reverse the bits in the given integer. |
48 | | static uint8 ReverseBits8(uint8 n); |
49 | | static uint32 ReverseBits32(uint32 n); |
50 | | static uint64 ReverseBits64(uint64 n); |
51 | | |
52 | | // Return the number of one bits in the byte sequence. |
53 | | static int Count(const void* m, int num_bytes); |
54 | | |
55 | | // Return the number of different bits in the given byte sequences. |
56 | | // (i.e., the Hamming distance) |
57 | | static int Difference(const void* m1, const void* m2, int num_bytes); |
58 | | |
59 | | // Return the number of different bits in the given byte sequences, |
60 | | // up to a maximum. Values larger than the maximum may be returned |
61 | | // (because multiple bits are checked at a time), but the function |
62 | | // may exit early if the cap is exceeded. |
63 | | static int CappedDifference(const void* m1, const void* m2, int num_bytes, int cap); |
64 | | |
65 | | // Return floor(log2(n)) for positive integer n. Returns -1 iff n == 0. |
66 | | static int Log2Floor(uint32 n); |
67 | | static int Log2Floor64(uint64 n); |
68 | | |
69 | | // Potentially faster version of Log2Floor() that returns an |
70 | | // undefined value if n == 0 |
71 | | static int Log2FloorNonZero(uint32 n); |
72 | | static int Log2FloorNonZero64(uint64 n); |
73 | | |
74 | | // Return ceiling(log2(n)) for positive integer n. Returns -1 iff n == 0. |
75 | | static int Log2Ceiling(uint32 n); |
76 | | static int Log2Ceiling64(uint64 n); |
77 | | |
78 | | // Return the first set least / most significant bit, 0-indexed. Returns an |
79 | | // undefined value if n == 0. FindLSBSetNonZero() is similar to ffs() except |
80 | | // that it's 0-indexed, while FindMSBSetNonZero() is the same as |
81 | | // Log2FloorNonZero(). |
82 | | static int FindLSBSetNonZero(uint32 n); |
83 | | static int FindLSBSetNonZero64(uint64 n); |
84 | 0 | static int FindMSBSetNonZero(uint32 n) { return Log2FloorNonZero(n); } |
85 | 0 | static int FindMSBSetNonZero64(uint64 n) { return Log2FloorNonZero64(n); } |
86 | | |
87 | | // Portable implementations |
88 | | static int Log2Floor_Portable(uint32 n); |
89 | | static int Log2FloorNonZero_Portable(uint32 n); |
90 | | static int FindLSBSetNonZero_Portable(uint32 n); |
91 | | static int Log2Floor64_Portable(uint64 n); |
92 | | static int Log2FloorNonZero64_Portable(uint64 n); |
93 | | static int FindLSBSetNonZero64_Portable(uint64 n); |
94 | | |
95 | | // Viewing bytes as a stream of unsigned bytes, does that stream |
96 | | // contain any byte equal to c? |
97 | | template <class T> |
98 | | static bool BytesContainByte(T bytes, uint8 c); |
99 | | |
100 | | // Viewing bytes as a stream of unsigned bytes, does that stream |
101 | | // contain any byte b < c? |
102 | | template <class T> |
103 | | static bool BytesContainByteLessThan(T bytes, uint8 c); |
104 | | |
105 | | // Viewing bytes as a stream of unsigned bytes, are all elements of that |
106 | | // stream in [lo, hi]? |
107 | | template <class T> |
108 | | static bool BytesAllInRange(T bytes, uint8 lo, uint8 hi); |
109 | | |
110 | | private: |
111 | | static const char num_bits[]; |
112 | | static const unsigned char bit_reverse_table[]; |
113 | | DISALLOW_COPY_AND_ASSIGN(Bits); |
114 | | }; |
115 | | |
116 | | // A utility class for some handy bit patterns. The names l and h |
117 | | // were chosen to match Knuth Volume 4: l is 0x010101... and h is 0x808080...; |
118 | | // half_ones is ones in the lower half only. We assume sizeof(T) is 1 or even. |
119 | | template <class T> |
120 | | struct BitPattern { |
121 | | static const T half_ones = (static_cast<T>(1) << (sizeof(T) * 4)) - 1; |
122 | | static const T l = (sizeof(T) == 1) ? 1 : (half_ones / 0xff * (half_ones + 2)); |
123 | | static const T h = ~(l * 0x7f); |
124 | | }; |
125 | | |
126 | | // ------------------------------------------------------------------------ |
127 | | // Implementation details follow |
128 | | // ------------------------------------------------------------------------ |
129 | | |
130 | | // use GNU builtins where available |
131 | | #if defined(__GNUC__) && ((__GNUC__ == 3 && __GNUC_MINOR__ >= 4) || __GNUC__ >= 4) |
132 | 0 | inline int Bits::Log2Floor(uint32 n) { |
133 | 0 | return n == 0 ? -1 : 31 ^ __builtin_clz(n); |
134 | 0 | } |
135 | | |
136 | 0 | inline int Bits::Log2FloorNonZero(uint32 n) { |
137 | 0 | return 31 ^ __builtin_clz(n); |
138 | 0 | } |
139 | | |
140 | 324 | inline int Bits::FindLSBSetNonZero(uint32 n) { |
141 | 324 | return __builtin_ctz(n); |
142 | 324 | } |
143 | | |
144 | 0 | inline int Bits::Log2Floor64(uint64 n) { |
145 | 0 | return n == 0 ? -1 : 63 ^ __builtin_clzll(n); |
146 | 0 | } |
147 | | |
148 | 0 | inline int Bits::Log2FloorNonZero64(uint64 n) { |
149 | 0 | return 63 ^ __builtin_clzll(n); |
150 | 0 | } |
151 | | |
152 | 0 | inline int Bits::FindLSBSetNonZero64(uint64 n) { |
153 | 0 | return __builtin_ctzll(n); |
154 | 0 | } |
155 | | #elif defined(_MSC_VER) |
156 | | #include "gutil/bits-internal-windows.h" |
157 | | #else |
158 | | #include "gutil/bits-internal-unknown.h" |
159 | | #endif |
160 | | |
161 | 0 | inline int Bits::CountOnesInByte(unsigned char n) { |
162 | 0 | return num_bits[n]; |
163 | 0 | } |
164 | | |
165 | 0 | inline uint8 Bits::ReverseBits8(unsigned char n) { |
166 | 0 | n = ((n >> 1) & 0x55) | ((n & 0x55) << 1); |
167 | 0 | n = ((n >> 2) & 0x33) | ((n & 0x33) << 2); |
168 | 0 | return ((n >> 4) & 0x0f) | ((n & 0x0f) << 4); |
169 | 0 | } |
170 | | |
171 | 0 | inline uint32 Bits::ReverseBits32(uint32 n) { |
172 | 0 | n = ((n >> 1) & 0x55555555) | ((n & 0x55555555) << 1); |
173 | 0 | n = ((n >> 2) & 0x33333333) | ((n & 0x33333333) << 2); |
174 | 0 | n = ((n >> 4) & 0x0F0F0F0F) | ((n & 0x0F0F0F0F) << 4); |
175 | 0 | n = ((n >> 8) & 0x00FF00FF) | ((n & 0x00FF00FF) << 8); |
176 | 0 | return (n >> 16) | (n << 16); |
177 | 0 | } |
178 | | |
179 | 0 | inline uint64 Bits::ReverseBits64(uint64 n) { |
180 | 0 | #if defined(__x86_64__) |
181 | 0 | n = ((n >> 1) & 0x5555555555555555ULL) | ((n & 0x5555555555555555ULL) << 1); |
182 | 0 | n = ((n >> 2) & 0x3333333333333333ULL) | ((n & 0x3333333333333333ULL) << 2); |
183 | 0 | n = ((n >> 4) & 0x0F0F0F0F0F0F0F0FULL) | ((n & 0x0F0F0F0F0F0F0F0FULL) << 4); |
184 | 0 | n = ((n >> 8) & 0x00FF00FF00FF00FFULL) | ((n & 0x00FF00FF00FF00FFULL) << 8); |
185 | 0 | n = ((n >> 16) & 0x0000FFFF0000FFFFULL) | ((n & 0x0000FFFF0000FFFFULL) << 16); |
186 | 0 | return (n >> 32) | (n << 32); |
187 | 0 | #else |
188 | 0 | return ReverseBits32(n >> 32) | (static_cast<uint64>(ReverseBits32(n & 0xffffffff)) << 32); |
189 | 0 | #endif |
190 | 0 | } |
191 | | |
192 | 0 | inline int Bits::Log2FloorNonZero_Portable(uint32 n) { |
193 | 0 | // Just use the common routine |
194 | 0 | return Log2Floor(n); |
195 | 0 | } |
196 | | |
197 | | // Log2Floor64() is defined in terms of Log2Floor32(), Log2FloorNonZero32() |
198 | 0 | inline int Bits::Log2Floor64_Portable(uint64 n) { |
199 | 0 | const uint32 topbits = static_cast<uint32>(n >> 32); |
200 | 0 | if (topbits == 0) { |
201 | 0 | // Top bits are zero, so scan in bottom bits |
202 | 0 | return Log2Floor(static_cast<uint32>(n)); |
203 | 0 | } else { |
204 | 0 | return 32 + Log2FloorNonZero(topbits); |
205 | 0 | } |
206 | 0 | } |
207 | | |
208 | | // Log2FloorNonZero64() is defined in terms of Log2FloorNonZero32() |
209 | 0 | inline int Bits::Log2FloorNonZero64_Portable(uint64 n) { |
210 | 0 | const uint32 topbits = static_cast<uint32>(n >> 32); |
211 | 0 | if (topbits == 0) { |
212 | 0 | // Top bits are zero, so scan in bottom bits |
213 | 0 | return Log2FloorNonZero(static_cast<uint32>(n)); |
214 | 0 | } else { |
215 | 0 | return 32 + Log2FloorNonZero(topbits); |
216 | 0 | } |
217 | 0 | } |
218 | | |
219 | | // FindLSBSetNonZero64() is defined in terms of FindLSBSetNonZero() |
220 | 0 | inline int Bits::FindLSBSetNonZero64_Portable(uint64 n) { |
221 | 0 | const uint32 bottombits = static_cast<uint32>(n); |
222 | 0 | if (bottombits == 0) { |
223 | 0 | // Bottom bits are zero, so scan in top bits |
224 | 0 | return 32 + FindLSBSetNonZero(static_cast<uint32>(n >> 32)); |
225 | 0 | } else { |
226 | 0 | return FindLSBSetNonZero(bottombits); |
227 | 0 | } |
228 | 0 | } |
229 | | |
230 | | template <class T> |
231 | | bool Bits::BytesContainByteLessThan(T bytes, uint8 c) { |
232 | | T l = BitPattern<T>::l; |
233 | | T h = BitPattern<T>::h; |
234 | | // The c <= 0x80 code is straight out of Knuth Volume 4. |
235 | | // Usually c will be manifestly constant. |
236 | | return c <= 0x80 ? ((h & (bytes - l * c) & ~bytes) != 0) |
237 | | : ((((bytes - l * c) | (bytes ^ h)) & h) != 0); |
238 | | } |
239 | | |
240 | | template <class T> |
241 | | bool Bits::BytesContainByte(T bytes, uint8 c) { |
242 | | // Usually c will be manifestly constant. |
243 | | return Bits::BytesContainByteLessThan<T>(bytes ^ (c * BitPattern<T>::l), 1); |
244 | | } |
245 | | |
246 | | template <class T> |
247 | | bool Bits::BytesAllInRange(T bytes, uint8 lo, uint8 hi) { |
248 | | T l = BitPattern<T>::l; |
249 | | T h = BitPattern<T>::h; |
250 | | // In the common case, lo and hi are manifest constants. |
251 | | if (lo > hi) { |
252 | | return false; |
253 | | } |
254 | | if (hi - lo < 128) { |
255 | | T x = bytes - l * lo; |
256 | | T y = bytes + l * (127 - hi); |
257 | | return ((x | y) & h) == 0; |
258 | | } |
259 | | return !Bits::BytesContainByteLessThan(bytes + (255 - hi) * l, lo + (255 - hi)); |
260 | | } |