/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* AtomicHashMap --
*
* A high-performance concurrent hash map with int32_t or int64_t keys. Supports
* insert, find(key), findAt(index), erase(key), size, and more. Memory cannot
* be freed or reclaimed by erase. Can grow to a maximum of about 18 times the
* initial capacity, but performance degrades linearly with growth. Can also be
* used as an object store with unique 32-bit references directly into the
* internal storage (retrieved with iterator::getIndex()).
*
* Advantages:
* - High-performance (~2-4x tbb::concurrent_hash_map in heavily
* multi-threaded environments).
* - Efficient memory usage if initial capacity is not over estimated
* (especially for small keys and values).
* - Good fragmentation properties (only allocates in large slabs which can
* be reused with clear() and never move).
* - Can generate unique, long-lived 32-bit references for efficient lookup
* (see findAt()).
*
* Disadvantages:
* - Keys must be native int32_t or int64_t, or explicitly converted.
* - Must be able to specify unique empty, locked, and erased keys
* - Performance degrades linearly as size grows beyond initialization
* capacity.
* - Max size limit of ~18x initial size (dependent on max load factor).
* - Memory is not freed or reclaimed by erase.
*
* Usage and Operation Details:
* Simple performance/memory tradeoff with maxLoadFactor. Higher load factors
* give better memory utilization but probe lengths increase, reducing
* performance.
*
* Implementation and Performance Details:
* AHArray is a fixed size contiguous block of value_type cells. When
* writing a cell, the key is locked while the rest of the record is
* written. Once done, the cell is unlocked by setting the key. find()
* is completely wait-free and doesn't require any non-relaxed atomic
* operations. AHA cannot grow beyond initialization capacity, but is
* faster because of reduced data indirection.
*
* AHMap is a wrapper around AHArray sub-maps that allows growth and provides
* an interface closer to the STL UnorderedAssociativeContainer concept. These
* sub-maps are allocated on the fly and are processed in series, so the more
* there are (from growing past initial capacity), the worse the performance.
*
* Insert returns false if there is a key collision and throws if the max size
* of the map is exceeded.
*
* Benchmark performance with 8 simultaneous threads processing 1 million
* unique <int64_t, int64_t> entries on a 4-core, 2.5 GHz machine:
*
* Load Factor Mem Efficiency usec/Insert usec/Find
* 50% 50% 0.19 0.05
* 85% 85% 0.20 0.06
* 90% 90% 0.23 0.08
* 95% 95% 0.27 0.10
*
* See folly/tests/AtomicHashMapTest.cpp for more benchmarks.
*/
#pragma once
#define FOLLY_ATOMICHASHMAP_H_
#include <atomic>
#include <functional>
#include <stdexcept>
#include <folly/AtomicHashArray.h>
#include <folly/CPortability.h>
#include <folly/Likely.h>
#include <folly/ThreadCachedInt.h>
#include <folly/container/Foreach.h>
#include <folly/hash/Hash.h>
namespace folly {
/*
* AtomicHashMap provides an interface somewhat similar to the
* UnorderedAssociativeContainer concept in C++. This does not
* exactly match this concept (or even the basic Container concept),
* because of some restrictions imposed by our datastructure.
*
* Specific differences (there are quite a few):
*
* - Efficiently thread safe for inserts (main point of this stuff),
* wait-free for lookups.
*
* - You can erase from this container, but the cell containing the key will
* not be free or reclaimed.
*
* - You can erase everything by calling clear() (and you must guarantee only
* one thread can be using the container to do that).
*
* - We aren't DefaultConstructible, CopyConstructible, Assignable, or
* EqualityComparable. (Most of these are probably not something
* you actually want to do with this anyway.)
*
* - We don't support the various bucket functions, rehash(),
* reserve(), or equal_range(). Also no constructors taking
* iterators, although this could change.
*
* - Several insertion functions, notably operator[], are not
* implemented. It is a little too easy to misuse these functions
* with this container, where part of the point is that when an
* insertion happens for a new key, it will atomically have the
* desired value.
*
* - The map has no templated insert() taking an iterator range, but
* we do provide an insert(key, value). The latter seems more
* frequently useful for this container (to avoid sprinkling
* make_pair everywhere), and providing both can lead to some gross
* template error messages.
*
* - The Allocator must not be stateful (a new instance will be spun up for
* each allocation), and its allocate() method must take a raw number of
* bytes.
*
* - KeyT must be a 32 bit or 64 bit atomic integer type, and you must
* define special 'locked' and 'empty' key values in the ctor
*
* - We don't take the Hash function object as an instance in the
* constructor.
*
*/
// Thrown when insertion fails due to running out of space for
// submaps.
struct FOLLY_EXPORT AtomicHashMapFullError : std::runtime_error {
explicit AtomicHashMapFullError()
: std::runtime_error("AtomicHashMap is full") {}
};
template <
class KeyT,
class ValueT,
class HashFcn,
class EqualFcn,
class Allocator,
class ProbeFcn,
class KeyConvertFcn>
class AtomicHashMap {
typedef AtomicHashArray<
KeyT,
ValueT,
HashFcn,
EqualFcn,
Allocator,
ProbeFcn,
KeyConvertFcn>
SubMap;
public:
typedef KeyT key_type;
typedef ValueT mapped_type;
typedef std::pair<const KeyT, ValueT> value_type;
typedef HashFcn hasher;
typedef EqualFcn key_equal;
typedef KeyConvertFcn key_convert;
typedef value_type* pointer;
typedef value_type& reference;
typedef const value_type& const_reference;
typedef std::ptrdiff_t difference_type;
typedef std::size_t size_type;
typedef typename SubMap::Config Config;
template <class ContT, class IterVal, class SubIt>
struct ahm_iterator;
typedef ahm_iterator<
const AtomicHashMap,
const value_type,
typename SubMap::const_iterator>
const_iterator;
typedef ahm_iterator<AtomicHashMap, value_type, typename SubMap::iterator>
iterator;
public:
const float kGrowthFrac_; // How much to grow when we run out of capacity.
// The constructor takes a finalSizeEst which is the optimal
// number of elements to maximize space utilization and performance,
// and a Config object to specify more advanced options.
explicit AtomicHashMap(size_t finalSizeEst, const Config& c = Config());
AtomicHashMap(const AtomicHashMap&) = delete;
AtomicHashMap& operator=(const AtomicHashMap&) = delete;
~AtomicHashMap() {
const unsigned int numMaps =
numMapsAllocated_.load(std::memory_order_relaxed);
FOR_EACH_RANGE (i, 0, numMaps) {
SubMap* thisMap = subMaps_[i].load(std::memory_order_relaxed);
DCHECK(thisMap);
SubMap::destroy(thisMap);
}
}
key_equal key_eq() const { return key_equal(); }
hasher hash_function() const { return hasher(); }
/*
* insert --
*
* Returns a pair with iterator to the element at r.first and
* success. Retrieve the index with ret.first.getIndex().
*
* Does not overwrite on key collision, but returns an iterator to
* the existing element (since this could due to a race with
* another thread, it is often important to check this return
* value).
*
* Allocates new sub maps as the existing ones become full. If
* all sub maps are full, no element is inserted, and
* AtomicHashMapFullError is thrown.
*/
std::pair<iterator, bool> insert(const value_type& r) {
return emplace(r.first, r.second);
}
std::pair<iterator, bool> insert(key_type k, const mapped_type& v) {
return emplace(k, v);
}
std::pair<iterator, bool> insert(value_type&& r) {
return emplace(r.first, std::move(r.second));
}
std::pair<iterator, bool> insert(key_type k, mapped_type&& v) {
return emplace(k, std::move(v));
}
/*
* emplace --
*
* Same contract as insert(), but performs in-place construction
* of the value type using the specified arguments.
*
* Also, like find(), this method optionally allows 'key_in' to have a type
* different from that stored in the table; see find(). If and only if no
* equal key is already present, this method converts 'key_in' to a key of
* type KeyT using the provided LookupKeyToKeyFcn.
*/
template <
typename LookupKeyT = key_type,
typename LookupHashFcn = hasher,
typename LookupEqualFcn = key_equal,
typename LookupKeyToKeyFcn = key_convert,
typename... ArgTs>
std::pair<iterator, bool> emplace(LookupKeyT k, ArgTs&&... vCtorArg);
/*
* find --
*
* Returns the iterator to the element if found, otherwise end().
*
* As an optional feature, the type of the key to look up (LookupKeyT) is
* allowed to be different from the type of keys actually stored (KeyT).
*
* This enables use cases where materializing the key is costly and usually
* redundant, e.g., canonicalizing/interning a set of strings and being able
* to look up by StringPiece. To use this feature, LookupHashFcn must take
* a LookupKeyT, and LookupEqualFcn must take KeyT and LookupKeyT as first
* and second parameter, respectively.
*
* See folly/test/ArrayHashMapTest.cpp for sample usage.
*/
template <
typename LookupKeyT = key_type,
typename LookupHashFcn = hasher,
typename LookupEqualFcn = key_equal>
iterator find(LookupKeyT k);
template <
typename LookupKeyT = key_type,
typename LookupHashFcn = hasher,
typename LookupEqualFcn = key_equal>
const_iterator find(LookupKeyT k) const;
/*
* erase --
*
* Erases key k from the map
*
* Returns 1 iff the key is found and erased, and 0 otherwise.
*/
size_type erase(key_type k);
/*
* clear --
*
* Wipes all keys and values from primary map and destroys all secondary
* maps. Primary map remains allocated and thus the memory can be reused
* in place. Not thread safe.
*
*/
void clear();
/*
* size --
*
* Returns the exact size of the map. Note this is not as cheap as typical
* size() implementations because, for each AtomicHashArray in this AHM, we
* need to grab a lock and accumulate the values from all the thread local
* counters. See folly/ThreadCachedInt.h for more details.
*/
size_t size() const;
bool empty() const { return size() == 0; }
size_type count(key_type k) const { return find(k) == end() ? 0 : 1; }
/*
* findAt --
*
* Returns an iterator into the map.
*
* idx should only be an unmodified value returned by calling getIndex() on
* a valid iterator returned by find() or insert(). If idx is invalid you
* have a bug and the process aborts.
*/
iterator findAt(uint32_t idx) {
SimpleRetT ret = findAtInternal(idx);
DCHECK_LT(ret.i, numSubMaps());
return iterator(
this,
ret.i,
subMaps_[ret.i].load(std::memory_order_relaxed)->makeIter(ret.j));
}
const_iterator findAt(uint32_t idx) const {
return const_cast<AtomicHashMap*>(this)->findAt(idx);
}
// Total capacity - summation of capacities of all submaps.
size_t capacity() const;
// Number of new insertions until current submaps are all at max load factor.
size_t spaceRemaining() const;
void setEntryCountThreadCacheSize(int32_t newSize) {
const int numMaps = numMapsAllocated_.load(std::memory_order_acquire);
for (int i = 0; i < numMaps; ++i) {
SubMap* map = subMaps_[i].load(std::memory_order_relaxed);
map->setEntryCountThreadCacheSize(newSize);
}
}
// Number of sub maps allocated so far to implement this map. The more there
// are, the worse the performance.
int numSubMaps() const {
return numMapsAllocated_.load(std::memory_order_acquire);
}
iterator begin() {
iterator it(this, 0, subMaps_[0].load(std::memory_order_relaxed)->begin());
it.checkAdvanceToNextSubmap();
return it;
}
const_iterator begin() const {
const_iterator it(
this, 0, subMaps_[0].load(std::memory_order_relaxed)->begin());
it.checkAdvanceToNextSubmap();
return it;
}
iterator end() { return iterator(); }
const_iterator end() const { return const_iterator(); }
/* Advanced functions for direct access: */
inline uint32_t recToIdx(const value_type& r, bool mayInsert = true) {
SimpleRetT ret =
mayInsert ? insertInternal(r.first, r.second) : findInternal(r.first);
return encodeIndex(ret.i, ret.j);
}
inline uint32_t recToIdx(value_type&& r, bool mayInsert = true) {
SimpleRetT ret = mayInsert ? insertInternal(r.first, std::move(r.second))
: findInternal(r.first);
return encodeIndex(ret.i, ret.j);
}
inline uint32_t recToIdx(
key_type k, const mapped_type& v, bool mayInsert = true) {
SimpleRetT ret = mayInsert ? insertInternal(k, v) : findInternal(k);
return encodeIndex(ret.i, ret.j);
}
inline uint32_t recToIdx(key_type k, mapped_type&& v, bool mayInsert = true) {
SimpleRetT ret =
mayInsert ? insertInternal(k, std::move(v)) : findInternal(k);
return encodeIndex(ret.i, ret.j);
}
inline uint32_t keyToIdx(const KeyT k, bool mayInsert = false) {
return recToIdx(value_type(k), mayInsert);
}
inline const value_type& idxToRec(uint32_t idx) const {
SimpleRetT ret = findAtInternal(idx);
return subMaps_[ret.i].load(std::memory_order_relaxed)->idxToRec(ret.j);
}
/* Private data and helper functions... */
private:
// This limits primary submap size to 2^31 ~= 2 billion, secondary submap
// size to 2^(32 - kNumSubMapBits_ - 1) = 2^27 ~= 130 million, and num subMaps
// to 2^kNumSubMapBits_ = 16.
static const uint32_t kNumSubMapBits_ = 4;
static const uint32_t kSecondaryMapBit_ = 1u << 31; // Highest bit
static const uint32_t kSubMapIndexShift_ = 32 - kNumSubMapBits_ - 1;
static const uint32_t kSubMapIndexMask_ = (1 << kSubMapIndexShift_) - 1;
static const uint32_t kNumSubMaps_ = 1 << kNumSubMapBits_;
static const uintptr_t kLockedPtr_ = 0x88ULL << 48; // invalid pointer
struct SimpleRetT {
uint32_t i;
size_t j;
bool success;
SimpleRetT(uint32_t ii, size_t jj, bool s) : i(ii), j(jj), success(s) {}
SimpleRetT() = default;
};
template <
typename LookupKeyT = key_type,
typename LookupHashFcn = hasher,
typename LookupEqualFcn = key_equal,
typename LookupKeyToKeyFcn = key_convert,
typename... ArgTs>
SimpleRetT insertInternal(LookupKeyT key, ArgTs&&... value);
template <
typename LookupKeyT = key_type,
typename LookupHashFcn = hasher,
typename LookupEqualFcn = key_equal>
SimpleRetT findInternal(const LookupKeyT k) const;
SimpleRetT findAtInternal(uint32_t idx) const;
std::atomic<SubMap*> subMaps_[kNumSubMaps_];
std::atomic<uint32_t> numMapsAllocated_;
inline bool tryLockMap(unsigned int idx) {
SubMap* val = nullptr;
return subMaps_[idx].compare_exchange_strong(
val, (SubMap*)kLockedPtr_, std::memory_order_acquire);
}
static inline uint32_t encodeIndex(uint32_t subMap, uint32_t subMapIdx);
}; // AtomicHashMap
template <
class KeyT,
class ValueT,
class HashFcn = std::hash<KeyT>,
class EqualFcn = std::equal_to<KeyT>,
class Allocator = std::allocator<char>>
using QuadraticProbingAtomicHashMap = AtomicHashMap<
KeyT,
ValueT,
HashFcn,
EqualFcn,
Allocator,
AtomicHashArrayQuadraticProbeFcn>;
} // namespace folly
#include <folly/AtomicHashMap-inl.h>
Codebase Browser
folly