2 * Copyright 2013 Facebook, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 * Various low-level, bit-manipulation routines.
20 * findFirstSet(x) [constexpr]
21 * find first (least significant) bit set in a value of an integral type,
22 * 1-based (like ffs()). 0 = no bits are set (x == 0)
24 * findLastSet(x) [constexpr]
25 * find last (most significant) bit set in a value of an integral type,
26 * 1-based. 0 = no bits are set (x == 0)
27 * for x != 0, findLastSet(x) == 1 + floor(log2(x))
29 * nextPowTwo(x) [constexpr]
30 * Finds the next power of two >= x.
32 * isPowTwo(x) [constexpr]
33 * return true iff x is a power of two
36 * return the number of 1 bits in x
39 * convert between native, big, and little endian representation
40 * Endian::big(x) big <-> native
41 * Endian::little(x) little <-> native
42 * Endian::swap(x) big <-> little
45 * Wrapper around an iterator over an integral type that iterates
46 * over its underlying bits in MSb to LSb order
48 * findFirstSet(BitIterator begin, BitIterator end)
49 * return a BitIterator pointing to the first 1 bit in [begin, end), or
50 * end if all bits in [begin, end) are 0
52 * @author Tudor Bosman (tudorb@fb.com)
58 #include "folly/Portability.h"
68 #include "folly/folly-config.h"
69 #include "folly/detail/BitsDetail.h"
70 #include "folly/detail/BitIteratorDetail.h"
71 #include "folly/Likely.h"
73 #if FOLLY_HAVE_BYTESWAP_H
74 # include <byteswap.h>
81 #include <type_traits>
82 #include <boost/iterator/iterator_adaptor.hpp>
87 // Generate overloads for findFirstSet as wrappers around
88 // appropriate ffs, ffsl, ffsll gcc builtins
91 typename std::enable_if<
92 (std::is_integral<T>::value &&
93 std::is_unsigned<T>::value &&
94 sizeof(T) <= sizeof(unsigned int)),
97 return __builtin_ffs(x);
102 typename std::enable_if<
103 (std::is_integral<T>::value &&
104 std::is_unsigned<T>::value &&
105 sizeof(T) > sizeof(unsigned int) &&
106 sizeof(T) <= sizeof(unsigned long)),
109 return __builtin_ffsl(x);
114 typename std::enable_if<
115 (std::is_integral<T>::value &&
116 std::is_unsigned<T>::value &&
117 sizeof(T) > sizeof(unsigned long) &&
118 sizeof(T) <= sizeof(unsigned long long)),
121 return __builtin_ffsll(x);
126 typename std::enable_if<
127 (std::is_integral<T>::value && std::is_signed<T>::value),
130 // Note that conversion from a signed type to the corresponding unsigned
131 // type is technically implementation-defined, but will likely work
132 // on any impementation that uses two's complement.
133 return findFirstSet(static_cast<typename std::make_unsigned<T>::type>(x));
136 // findLastSet: return the 1-based index of the highest bit set
137 // for x > 0, findLastSet(x) == 1 + floor(log2(x))
140 typename std::enable_if<
141 (std::is_integral<T>::value &&
142 std::is_unsigned<T>::value &&
143 sizeof(T) <= sizeof(unsigned int)),
146 return x ? 8 * sizeof(unsigned int) - __builtin_clz(x) : 0;
151 typename std::enable_if<
152 (std::is_integral<T>::value &&
153 std::is_unsigned<T>::value &&
154 sizeof(T) > sizeof(unsigned int) &&
155 sizeof(T) <= sizeof(unsigned long)),
158 return x ? 8 * sizeof(unsigned long) - __builtin_clzl(x) : 0;
163 typename std::enable_if<
164 (std::is_integral<T>::value &&
165 std::is_unsigned<T>::value &&
166 sizeof(T) > sizeof(unsigned long) &&
167 sizeof(T) <= sizeof(unsigned long long)),
170 return x ? 8 * sizeof(unsigned long long) - __builtin_clzll(x) : 0;
175 typename std::enable_if<
176 (std::is_integral<T>::value &&
177 std::is_signed<T>::value),
180 return findLastSet(static_cast<typename std::make_unsigned<T>::type>(x));
185 typename std::enable_if<
186 std::is_integral<T>::value && std::is_unsigned<T>::value,
189 return v ? (1ul << findLastSet(v - 1)) : 1;
194 typename std::enable_if<
195 std::is_integral<T>::value && std::is_unsigned<T>::value,
198 return (v != 0) && !(v & (v - 1));
205 inline typename std::enable_if<
206 (std::is_integral<T>::value &&
207 std::is_unsigned<T>::value &&
208 sizeof(T) <= sizeof(unsigned int)),
211 return detail::popcount(x);
215 inline typename std::enable_if<
216 (std::is_integral<T>::value &&
217 std::is_unsigned<T>::value &&
218 sizeof(T) > sizeof(unsigned int) &&
219 sizeof(T) <= sizeof(unsigned long long)),
222 return detail::popcountll(x);
226 * Endianness detection and manipulation primitives.
231 struct EndianIntBase {
237 * If we have the bswap_16 macro from byteswap.h, use it; otherwise, provide our
241 # define our_bswap16 bswap_16
244 template<class Int16>
245 inline constexpr typename std::enable_if<
248 our_bswap16(Int16 x) {
249 return ((x >> 8) & 0xff) | ((x & 0xff) << 8);
253 #define FB_GEN(t, fn) \
254 template<> inline t EndianIntBase<t>::swap(t x) { return fn(x); }
256 // fn(x) expands to (x) if the second argument is empty, which is exactly
257 // what we want for [u]int8_t. Also, gcc 4.7 on Intel doesn't have
258 // __builtin_bswap16 for some reason, so we have to provide our own.
261 FB_GEN( int64_t, __builtin_bswap64)
262 FB_GEN(uint64_t, __builtin_bswap64)
263 FB_GEN( int32_t, __builtin_bswap32)
264 FB_GEN(uint32_t, __builtin_bswap32)
265 FB_GEN( int16_t, our_bswap16)
266 FB_GEN(uint16_t, our_bswap16)
270 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
273 struct EndianInt : public detail::EndianIntBase<T> {
275 static T big(T x) { return EndianInt::swap(x); }
276 static T little(T x) { return x; }
279 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
282 struct EndianInt : public detail::EndianIntBase<T> {
284 static T big(T x) { return x; }
285 static T little(T x) { return EndianInt::swap(x); }
289 # error Your machine uses a weird endianness!
290 #endif /* __BYTE_ORDER__ */
292 } // namespace detail
294 // big* convert between native and big-endian representations
295 // little* convert between native and little-endian representations
296 // swap* convert between big-endian and little-endian representations
298 // ntohs, htons == big16
299 // ntohl, htonl == big32
300 #define FB_GEN1(fn, t, sz) \
301 static t fn##sz(t x) { return fn<t>(x); } \
303 #define FB_GEN2(t, sz) \
304 FB_GEN1(swap, t, sz) \
305 FB_GEN1(big, t, sz) \
306 FB_GEN1(little, t, sz)
309 FB_GEN2(uint##sz##_t, sz) \
310 FB_GEN2(int##sz##_t, sz)
314 enum class Order : uint8_t {
319 static constexpr Order order =
320 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
322 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
325 # error Your machine uses a weird endianness!
326 #endif /* __BYTE_ORDER__ */
328 template <class T> static T swap(T x) {
329 return detail::EndianInt<T>::swap(x);
331 template <class T> static T big(T x) {
332 return detail::EndianInt<T>::big(x);
334 template <class T> static T little(T x) {
335 return detail::EndianInt<T>::little(x);
349 * Fast bit iteration facility.
353 template <class BaseIter> class BitIterator;
354 template <class BaseIter>
355 BitIterator<BaseIter> findFirstSet(BitIterator<BaseIter>,
356 BitIterator<BaseIter>);
358 * Wrapper around an iterator over an integer type that iterates
359 * over its underlying bits in LSb to MSb order.
361 * BitIterator models the same iterator concepts as the base iterator.
363 template <class BaseIter>
365 : public bititerator_detail::BitIteratorBase<BaseIter>::type {
368 * Return the number of bits in an element of the underlying iterator.
370 static size_t bitsPerBlock() {
371 return std::numeric_limits<
372 typename std::make_unsigned<
373 typename std::iterator_traits<BaseIter>::value_type
379 * Construct a BitIterator that points at a given bit offset (default 0)
382 explicit BitIterator(const BaseIter& iter, size_t bitOffset=0)
383 : bititerator_detail::BitIteratorBase<BaseIter>::type(iter),
384 bitOffset_(bitOffset) {
385 assert(bitOffset_ < bitsPerBlock());
388 size_t bitOffset() const {
392 void advanceToNextBlock() {
394 ++this->base_reference();
397 BitIterator& operator=(const BaseIter& other) {
398 this->~BitIterator();
399 new (this) BitIterator(other);
404 friend class boost::iterator_core_access;
405 friend BitIterator findFirstSet<>(BitIterator, BitIterator);
407 typedef bititerator_detail::BitReference<
408 typename std::iterator_traits<BaseIter>::reference,
409 typename std::iterator_traits<BaseIter>::value_type
412 void advanceInBlock(size_t n) {
414 assert(bitOffset_ < bitsPerBlock());
417 BitRef dereference() const {
418 return BitRef(*this->base_reference(), bitOffset_);
421 void advance(ssize_t n) {
422 size_t bpb = bitsPerBlock();
423 ssize_t blocks = n / bpb;
424 bitOffset_ += n % bpb;
425 if (bitOffset_ >= bpb) {
429 this->base_reference() += blocks;
433 if (++bitOffset_ == bitsPerBlock()) {
434 advanceToNextBlock();
439 if (bitOffset_-- == 0) {
440 bitOffset_ = bitsPerBlock() - 1;
441 --this->base_reference();
445 bool equal(const BitIterator& other) const {
446 return (bitOffset_ == other.bitOffset_ &&
447 this->base_reference() == other.base_reference());
450 ssize_t distance_to(const BitIterator& other) const {
452 (other.base_reference() - this->base_reference()) * bitsPerBlock() +
453 (other.bitOffset_ - bitOffset_);
460 * Helper function, so you can write
461 * auto bi = makeBitIterator(container.begin());
463 template <class BaseIter>
464 BitIterator<BaseIter> makeBitIterator(const BaseIter& iter) {
465 return BitIterator<BaseIter>(iter);
470 * Find first bit set in a range of bit iterators.
471 * 4.5x faster than the obvious std::find(begin, end, true);
473 template <class BaseIter>
474 BitIterator<BaseIter> findFirstSet(BitIterator<BaseIter> begin,
475 BitIterator<BaseIter> end) {
476 // shortcut to avoid ugly static_cast<>
477 static const typename BaseIter::value_type one = 1;
479 while (begin.base() != end.base()) {
480 typename BaseIter::value_type v = *begin.base();
481 // mask out the bits that don't matter (< begin.bitOffset)
482 v &= ~((one << begin.bitOffset()) - 1);
483 size_t firstSet = findFirstSet(v);
485 --firstSet; // now it's 0-based
486 assert(firstSet >= begin.bitOffset());
487 begin.advanceInBlock(firstSet - begin.bitOffset());
490 begin.advanceToNextBlock();
493 // now begin points to the same block as end
494 if (end.bitOffset() != 0) { // assume end is dereferenceable
495 typename BaseIter::value_type v = *begin.base();
496 // mask out the bits that don't matter (< begin.bitOffset)
497 v &= ~((one << begin.bitOffset()) - 1);
498 // mask out the bits that don't matter (>= end.bitOffset)
499 v &= (one << end.bitOffset()) - 1;
500 size_t firstSet = findFirstSet(v);
502 --firstSet; // now it's 0-based
503 assert(firstSet >= begin.bitOffset());
504 begin.advanceInBlock(firstSet - begin.bitOffset());
513 template <class T, class Enable=void> struct Unaligned;
516 * Representation of an unaligned value of a POD type.
521 typename std::enable_if<std::is_pod<T>::value>::type> {
522 Unaligned() = default; // uninitialized
523 /* implicit */ Unaligned(T v) : value(v) { }
525 } __attribute__((packed));
528 * Read an unaligned value of type T and return it.
531 inline T loadUnaligned(const void* p) {
532 static_assert(sizeof(Unaligned<T>) == sizeof(T), "Invalid unaligned size");
533 static_assert(alignof(Unaligned<T>) == 1, "Invalid alignment");
534 return static_cast<const Unaligned<T>*>(p)->value;
538 * Write an unaligned value of type T.
541 inline void storeUnaligned(void* p, T value) {
542 static_assert(sizeof(Unaligned<T>) == sizeof(T), "Invalid unaligned size");
543 static_assert(alignof(Unaligned<T>) == 1, "Invalid alignment");
544 new (p) Unaligned<T>(value);
549 #endif /* FOLLY_BITS_H_ */