Concurrent Hopscotch HashMap : HashMap « Collections Data Structure « Java
Concurrent Hopscotch HashMap
////////////////////////////////////////////////////////////////////////////////
// ConcurrentHopscotchHashMap Class
//
////////////////////////////////////////////////////////////////////////////////
//TERMS OF USAGE
//----------------------------------------------------------------------
//
// Permission to use, copy, modify and distribute this software and
// its documentation for any purpose is hereby granted without fee,
// provided that due acknowledgments to the authors are provided and
// this permission notice appears in all copies of the software.
// The software is provided "as is". There is no warranty of any kind.
//
//Authors:
// Maurice Herlihy
// Brown University
// and
// Nir Shavit
// Tel-Aviv University
// and
// Moran Tzafrir
// Tel-Aviv University
//
// Date: Dec 2, 2008.
//
////////////////////////////////////////////////////////////////////////////////
// Programmer : Moran Tzafrir (MoranTza@gmail.com)
//
////////////////////////////////////////////////////////////////////////////////
//package xbird.util.concurrent.map;
import java.util.concurrent.locks.ReentrantLock;
public class ConcurrentHopscotchHashMap<K, V> {
//constants -----------------------------------
static final short _NULL_DELTA_SHORT = Short.MIN_VALUE;
static final int _NULL_DELTA_INT = (int) (Short.MIN_VALUE);
static final long _NULL_DELTA_FIRST_LONG = (long) (Short.MIN_VALUE & 0xFFFFL);
static final long _NULL_DELTA_NEXT_LONG = (long) ((Short.MIN_VALUE & 0xFFFFL) << 16);
static final long _NULL_HASH_DELTA = 0x0000000080008000L;
static final int _NULL_HASH = 0;
static final int _SEGMENT_SHIFT = 0; //choosed by empirical tests
static final long _FIRST_MASK = 0x000000000000FFFFL;
static final long _NEXT_MASK = 0x00000000FFFF0000L;
static final long _HASH_MASK = 0xFFFFFFFF00000000L;
static final long _NOT_NEXT_MASK = ~(_NEXT_MASK);
static final long _NOT_FIRST_MASK = ~(_FIRST_MASK);
static final long _NOT_HASH_MASK = ~(_HASH_MASK);
static final int _NEXT_SHIFT = 16;
static final int _HASH_SHIFT = 32;
static final int _RETRIES_BEFORE_LOCK = 2;
static final int _MAX_DELTA_BUCKET = Short.MAX_VALUE;
static final int _CACHE_MASK = 4 - 1;
static final int _NOT_CACHE_MASK = ~_CACHE_MASK;
static final int _NULL_INDX = -1;
//inner classes -------------------------------
static final class Segment<K, V> extends ReentrantLock {
volatile int _timestamp;
int _bucketk_mask;
long[] _table_hash_delta;
Object[] _table_key_value;
//AtomicInteger _lock;
int _count;
public Segment(final int initialCapacity) {
init(initialCapacity);
}
private void init(final int initialCapacity) {
//_lock = new AtomicInteger();
//_lock.set(0);
_timestamp = 0;
_bucketk_mask = initialCapacity - 1;
_count = 0;
_table_hash_delta = new long[initialCapacity];
_table_key_value = new Object[initialCapacity << 1];
//create the blocks of buckets
for(int iData = 0; iData < initialCapacity; ++iData) {
_table_hash_delta[iData] = _NULL_HASH_DELTA;
}
}
/*
void lock() {
while(!_lock.compareAndSet(0, 0xFFFF)) { }
}
void unlock() {
_lock.set(0);
}
*/
@SuppressWarnings("unchecked")
static final <K, V> Segment<K, V>[] newArray(final int numSegments) {
return new Segment[numSegments];
}
boolean containsKey(final K key, final int hash) {
//go over the list and look for key
int start_timestamp = _timestamp;
int iBucket = hash & _bucketk_mask;
long data = _table_hash_delta[iBucket];
final int first_delta = (short) data;
if(0 != first_delta) {
if(_NULL_DELTA_SHORT == first_delta)
return false;
iBucket += first_delta;
data = _table_hash_delta[iBucket];
}
do {
if(hash == (data >> _HASH_SHIFT) && key.equals(_table_key_value[iBucket << 1]))
return true;
final int nextDelta = (int) data >> _NEXT_SHIFT;
if(_NULL_DELTA_INT != nextDelta) {
iBucket += nextDelta;
data = _table_hash_delta[iBucket];
continue;
} else {
final int curr_timestamp = _timestamp;
if(curr_timestamp == start_timestamp)
return false;
start_timestamp = curr_timestamp;
iBucket = hash & _bucketk_mask;
data = _table_hash_delta[iBucket];
final int first_delta2 = (short) data;
if(0 != first_delta2) {
if(_NULL_DELTA_SHORT == first_delta2)
return false;
iBucket += first_delta2;
data = _table_hash_delta[iBucket];
}
continue;
}
} while(true);
}
V get(final K key, final int hash) {
//go over the list and look for key
int start_timestamp = 0;
int iBucket = 0;
long data = 0;
boolean is_need_init = true;
do {
if(is_need_init) {
is_need_init = false;
start_timestamp = _timestamp;
iBucket = hash & _bucketk_mask;
data = _table_hash_delta[iBucket];
final int first_delta = (short) data;
if(0 != first_delta) {
if(_NULL_DELTA_SHORT == first_delta)
return null;
iBucket += first_delta;
data = _table_hash_delta[iBucket];
}
}
final int iRef;
if(hash == (data >> _HASH_SHIFT)
&& key.equals(_table_key_value[iRef = iBucket << 1])) {
final V value = (V) _table_key_value[iRef + 1];
if(_timestamp == start_timestamp)
return value;
is_need_init = true;
continue;
}
final int nextDelta = (int) data >> _NEXT_SHIFT;
if(_NULL_DELTA_INT != nextDelta) {
iBucket += nextDelta;
data = _table_hash_delta[iBucket];
continue;
} else {
if(_timestamp == start_timestamp)
return null;
is_need_init = true;
continue;
}
} while(true);
}
V put(final K key, final int hash, final V value) {
lock();
try {
//look for key in hash-map .....................................
final int i_start_bucket = hash & _bucketk_mask;
int iBucket = i_start_bucket;
long data = _table_hash_delta[i_start_bucket];
final short first_delta = (short) data;
if(_NULL_DELTA_SHORT != first_delta) {
if(0 != first_delta) {
iBucket += first_delta;
data = _table_hash_delta[iBucket];
}
do {
final int iRef;
if(hash == (data >> _HASH_SHIFT)
&& key.equals(_table_key_value[iRef = (iBucket << 1)]))
return (V) _table_key_value[iRef + 1];
final int next_delta = (int) data >> _NEXT_SHIFT;
if(_NULL_DELTA_INT == next_delta)
break;
else {
iBucket += next_delta;
data = _table_hash_delta[iBucket];
}
} while(true);
}
//try to place the key in the same cache-line ..................
final int i_start_cacheline = i_start_bucket & _NOT_CACHE_MASK;
final int i_end_cacheline = i_start_cacheline + _CACHE_MASK;
int i_free_bucket = i_start_bucket;
do {
long free_data = _table_hash_delta[i_free_bucket];
if(_NULL_HASH == (free_data >> _HASH_SHIFT)) {
//we found a free bucket at the cahce-line, so
//add the new bucket to the begining of the list
int i_ref_bucket = i_free_bucket << 1;
_table_key_value[i_ref_bucket] = key;
_table_key_value[++i_ref_bucket] = value;
free_data &= _NOT_HASH_MASK;
free_data |= ((long) hash << _HASH_SHIFT);
if(0 == first_delta) {
final long start_data = _table_hash_delta[i_start_bucket];
final int start_next = (int) start_data >> _NEXT_SHIFT;
if(_NULL_DELTA_INT != start_next) {
final long new_free_next = i_start_bucket + start_next
- i_free_bucket;
_table_hash_delta[i_free_bucket] = (free_data & _NOT_NEXT_MASK)
| ((new_free_next << _NEXT_SHIFT) & _NEXT_MASK);
} else
_table_hash_delta[i_free_bucket] = free_data;
final long new_start_next = i_free_bucket - i_start_bucket;
_table_hash_delta[i_start_bucket] = (start_data & _NOT_NEXT_MASK)
| ((new_start_next << _NEXT_SHIFT) & _NEXT_MASK);
} else {//0 != first_delta
if(_NULL_DELTA_SHORT != first_delta) {
final long new_free_next = i_start_bucket + first_delta
- i_free_bucket;
free_data &= _NOT_NEXT_MASK;
free_data |= ((new_free_next << _NEXT_SHIFT) & _NEXT_MASK);
}
final long start_data;
if(i_free_bucket != i_start_bucket) {
start_data = _table_hash_delta[i_start_bucket];
_table_hash_delta[i_free_bucket] = free_data;
} else
start_data = free_data;
final long new_start_first = i_free_bucket - i_start_bucket;
_table_hash_delta[i_start_bucket] = (start_data & _NOT_FIRST_MASK)
| (new_start_first & _FIRST_MASK);
}
++_count;
++_timestamp;
return null;
}
++i_free_bucket;
if(i_free_bucket > i_end_cacheline)
i_free_bucket = i_start_cacheline;
} while(i_start_bucket != i_free_bucket);
//place key in arbitrary free forward bucket ...................
int i_max_bucket = i_start_bucket + _MAX_DELTA_BUCKET;
if(i_max_bucket > _bucketk_mask)
i_max_bucket = _bucketk_mask;
i_free_bucket = i_end_cacheline + 1;
while(i_free_bucket <= i_max_bucket) {
long free_data = _table_hash_delta[i_free_bucket];
if(_NULL_HASH == (free_data >> _HASH_SHIFT)) {
//we found a free bucket outside of the cahce-line, so
//add the new bucket to the end of the list
int i_ref_bucket = i_free_bucket << 1;
_table_key_value[i_ref_bucket] = key;
_table_key_value[++i_ref_bucket] = value;
free_data &= _NOT_HASH_MASK;
free_data |= ((long) hash << _HASH_SHIFT);
_table_hash_delta[i_free_bucket] = free_data;
if(_NULL_DELTA_SHORT == first_delta) {
long new_start_first = (i_free_bucket - i_start_bucket) & _FIRST_MASK;
long start_data = (_table_hash_delta[i_start_bucket] & _NOT_FIRST_MASK)
| new_start_first;
_table_hash_delta[i_start_bucket] = start_data;
} else {
long new_last_next = ((i_free_bucket - iBucket) << _NEXT_SHIFT)
& _NEXT_MASK;
long last_data = (_table_hash_delta[iBucket] & _NOT_NEXT_MASK)
| new_last_next;
_table_hash_delta[iBucket] = last_data;
}
++_count;
++_timestamp;
return null;
}
i_free_bucket += 2;
}
//place key in arbitrary free backward bucket ...................
int i_min_bucket = i_start_bucket - _MAX_DELTA_BUCKET;
if(i_min_bucket < 0)
i_min_bucket = 0;
i_free_bucket = i_start_cacheline - 1;
while(i_free_bucket >= i_min_bucket) {
long free_data = _table_hash_delta[i_free_bucket];
if(_NULL_HASH == (free_data >> _HASH_SHIFT)) {
//we found a free bucket outside of the cahce-line, so
//add the new bucket to the end of the list
int i_ref_bucket = i_free_bucket << 1;
_table_key_value[i_ref_bucket] = key;
_table_key_value[++i_ref_bucket] = value;
free_data &= _NOT_HASH_MASK;
free_data |= ((long) hash << _HASH_SHIFT);
_table_hash_delta[i_free_bucket] = free_data;
if(_NULL_DELTA_SHORT == first_delta) {
long new_start_first = (i_free_bucket - i_start_bucket) & _FIRST_MASK;
long start_data = (_table_hash_delta[i_start_bucket] & _NOT_FIRST_MASK)
| new_start_first;
_table_hash_delta[i_start_bucket] = start_data;
} else {
long new_last_next = ((i_free_bucket - iBucket) << _NEXT_SHIFT)
& _NEXT_MASK;
long last_data = (_table_hash_delta[iBucket] & _NOT_NEXT_MASK)
| new_last_next;
_table_hash_delta[iBucket] = last_data;
}
++_count;
++_timestamp;
return null;
}
i_free_bucket -= 2;
}
} finally {
unlock();
}
return null;
}
private void optimize_cacheline_use(final int i_free_bucket) {
final int i_start_cacheline = i_free_bucket & _NOT_CACHE_MASK;
final int i_end_cacheline = i_start_cacheline + _CACHE_MASK;
//go over the buckets that reside in the cacheline of the free bucket
for(int i_cacheline = i_start_cacheline; i_cacheline <= i_end_cacheline; ++i_cacheline) {
//check if current bucket has keys
final long data = _table_hash_delta[i_cacheline];
final short first_delta = (short) data;
if(_NULL_DELTA_INT != first_delta) {
int last_i_relocate = _NULL_INDX;
int i_relocate = i_cacheline + first_delta;
int curr_delta = first_delta;
//go over the keys in the bucket-list
do {
//if the key reside outside the cahceline
if(curr_delta < 0 || curr_delta > _CACHE_MASK) {
//copy key, value, & hash to the free bucket
final int i_key_value = i_free_bucket << 1;
final int i_rel_key_value = i_relocate << 1;
_table_key_value[i_key_value] = _table_key_value[i_rel_key_value];
_table_key_value[i_key_value + 1] = _table_key_value[i_rel_key_value + 1];
long relocate_data = _table_hash_delta[i_relocate];
long free_data = _table_hash_delta[i_free_bucket];
free_data &= _NOT_HASH_MASK;
free_data |= (relocate_data & _HASH_MASK);
//update the next-field of the free-bucket
free_data &= _NOT_NEXT_MASK;
final int relocate_next_delta = (int) relocate_data >> _NEXT_SHIFT;
if(_NULL_DELTA_INT == relocate_next_delta) {
free_data |= _NULL_DELTA_NEXT_LONG;
} else {
final long new_next = (((i_relocate + relocate_next_delta) - i_free_bucket) & 0xFFFFL) << 16;
free_data |= new_next;
}
_table_hash_delta[i_free_bucket] = free_data;
//update the "first" or "next" field of the last
if(_NULL_INDX == last_i_relocate) {
long start_data = _table_hash_delta[i_cacheline] & _NOT_FIRST_MASK;
start_data |= ((i_free_bucket - i_cacheline) & 0xFFFFL);
_table_hash_delta[i_cacheline] = start_data;
} else {
long last_data = _table_hash_delta[last_i_relocate]
& _NOT_NEXT_MASK;
last_data |= (((i_free_bucket - last_i_relocate) & 0xFFFFL) << 16);
_table_hash_delta[last_i_relocate] = last_data;
}
//
++_timestamp;
relocate_data &= _NOT_HASH_MASK;//hash=null
relocate_data &= _NOT_NEXT_MASK;
relocate_data |= _NULL_DELTA_NEXT_LONG;//next = null
_table_hash_delta[i_relocate] = relocate_data;
_table_key_value[i_rel_key_value] = null;//key=null
_table_key_value[i_rel_key_value + 1] = null;//value=null
return;
}
final long relocate_data = _table_hash_delta[i_relocate];
final int next_delta = (int) relocate_data >> _NEXT_SHIFT;
if(_NULL_DELTA_INT == next_delta)
break;
last_i_relocate = i_relocate;
curr_delta += next_delta;
i_relocate += next_delta;
} while(true);//for on list
}//if list exists
}//for on list
}
V remove(final K key, final int hash) {
lock();
try {
//go over the list and look for key
final int i_start_bucket = hash & _bucketk_mask;
int iBucket = i_start_bucket;
long data = _table_hash_delta[iBucket];
final short first_delta = (short) data;
if(0 != first_delta) {
if(_NULL_DELTA_SHORT == first_delta)
return null;
iBucket += first_delta;
data = _table_hash_delta[iBucket];
}
int i_last_bucket = -1;
do {
final int iRef;
if(hash == (data >> _HASH_SHIFT)
&& key.equals(_table_key_value[iRef = (iBucket << 1)])) {
data &= _NOT_HASH_MASK;
final int next_delta = (int) data >> _NEXT_SHIFT;
_table_hash_delta[iBucket] = data; //hash = null
_table_key_value[iRef] = null; //key = null;
final int iRef2 = iRef + 1;
final V key_value = (V) _table_key_value[iRef2];
_table_key_value[iRef2] = null; //value = null;
if(-1 == i_last_bucket) {
long start_data = _table_hash_delta[i_start_bucket] & _NOT_FIRST_MASK;
if(_NULL_DELTA_INT == next_delta) {
start_data |= _NULL_DELTA_FIRST_LONG;
} else {
final long new_first = (first_delta + next_delta) & 0xFFFFL;
start_data |= new_first;
}
if(i_start_bucket == iBucket) {
start_data &= _NOT_NEXT_MASK;
start_data |= _NULL_DELTA_NEXT_LONG;
--_count;
++_timestamp;
_table_hash_delta[i_start_bucket] = start_data;
//return key_value;
} else
_table_hash_delta[i_start_bucket] = start_data;
} else {
long last_data = _table_hash_delta[i_last_bucket];
final int last_next_delta = (int) last_data >> _NEXT_SHIFT;
last_data &= _NOT_NEXT_MASK;
if(_NULL_DELTA_INT == next_delta) {
last_data |= _NULL_DELTA_NEXT_LONG;
} else {
final long new_next = ((last_next_delta + next_delta) & 0xFFFFL) << 16;
last_data |= new_next;
}
_table_hash_delta[i_last_bucket] = last_data;
}
if(i_start_bucket != iBucket) {
--_count;
++_timestamp;
data &= _NOT_NEXT_MASK;
data |= _NULL_DELTA_NEXT_LONG;
_table_hash_delta[iBucket] = data; //next = null
}
optimize_cacheline_use(iBucket);
return key_value;
}
final int nextDelta = (int) data >> _NEXT_SHIFT;
if(_NULL_DELTA_INT != nextDelta) {
i_last_bucket = iBucket;
iBucket += nextDelta;
data = _table_hash_delta[iBucket];
continue;
} else
return null;
} while(true);
} finally {
unlock();
}
}
void clear() {}
}
//fields --------------------------------------
final int _segment_shift;
final int _segment_mask;
final Segment<K, V>[] _segments;
//small utilities -----------------------------
private static int nearestPowerOfTwo(int value) {
int rc = 1;
while(rc < value) {
rc <<= 1;
}
return rc;
}
private static final int hash(int h) {
// Spread bits to regularize both segment and index locations,
// using variant of single-word Wang/Jenkins hash.
h += (h << 15) ^ 0xffffcd7d;
h ^= (h >>> 10);
h += (h << 3);
h ^= (h >>> 6);
h += (h << 2) + (h << 14);
return h ^ (h >>> 16);
}
private final Segment<K, V> segmentFor(int hash) {
return _segments[(hash >>> _segment_shift) & _segment_mask];
//return _segments[(hash >>> 8) & _segment_mask];
//return _segments[hash & _segment_mask];
}
//public operations ---------------------------
public ConcurrentHopscotchHashMap(final int initialCapacity, final int concurrencyLevel) {
//check for the validity of the algorithems
if(initialCapacity < 0 || concurrencyLevel <= 0 /*|| machineCachelineSize <= 0*/)
throw new IllegalArgumentException();
//set the user preference, should we force cache-line alignment
//_is_cacheline_alignment = isCachelineAlignment;
//calculate cache-line mask
//final int bucketSize = Math.max(8, 2*machinePointerSize);
//_cache_mask = ( (machineCachelineSize / bucketSize) - 1 );
//allocate the segments array
final int numSegments = nearestPowerOfTwo(concurrencyLevel);
_segment_mask = (numSegments - 1);
_segments = Segment.<K, V> newArray(numSegments);
// Find power-of-two sizes best matching arguments
int sshift = 0;
int ssize = 1;
while(ssize < numSegments) {
++sshift;
ssize <<= 1;
}
_segment_shift = 32 - sshift;
//initialize the segmens
final int initCapacity = nearestPowerOfTwo(initialCapacity);
final int segmentCapacity = initCapacity / numSegments;
for(int iSeg = 0; iSeg < numSegments; ++iSeg) {
_segments[iSeg] = new Segment<K, V>(segmentCapacity);
}
}
public boolean isEmpty() {
final Segment<K, V>[] segments = this._segments;
/*
* We keep track of per-segment "timestamp" to avoid ABA
* problems in which an element in one segment was added and
* in another removed during traversal, in which case the
* table was never actually empty at any point. Note the
* similar use of "timestamp" in the size() and containsValue()
* methods, which are the only other methods also susceptible
* to ABA problems.
*/
int[] mc = new int[segments.length];
int mcsum = 0;
for(int i = 0; i < segments.length; ++i) {
if(0 != segments[i]._count)
return false;
else
mcsum += mc[i] = segments[i]._timestamp;
}
// If mcsum happens to be zero, then we know we got a snapshot
// before any modifications at all were made. This is
// probably common enough to bother tracking.
if(mcsum != 0) {
for(int i = 0; i < segments.length; ++i) {
if(0 != segments[i]._count || mc[i] != segments[i]._timestamp)
return false;
}
}
return true;
}
public int size() {
final Segment<K, V>[] segments = this._segments;
long sum = 0;
long check = 0;
int[] mc = new int[segments.length];
// Try a few times to get accurate count. On failure due to
// continuous async changes in table, resort to locking.
for(int iTry = 0; iTry < _RETRIES_BEFORE_LOCK; ++iTry) {
check = 0;
sum = 0;
int mcsum = 0;
for(int i = 0; i < segments.length; ++i) {
sum += segments[i]._count;
mcsum += mc[i] = segments[i]._timestamp;
}
if(mcsum != 0) {
for(int i = 0; i < segments.length; ++i) {
check += segments[i]._count;
if(mc[i] != segments[i]._timestamp) {
check = -1; // force retry
break;
}
}
}
if(check == sum)
break;
}
if(check != sum) { // Resort to locking all segments
sum = 0;
for(int i = 0; i < segments.length; ++i)
segments[i].lock();
for(int i = 0; i < segments.length; ++i)
sum += segments[i]._count;
for(int i = 0; i < segments.length; ++i)
segments[i].unlock();
}
if(sum > Integer.MAX_VALUE)
return Integer.MAX_VALUE;
else
return (int) sum;
}
//contains
public boolean containsKey(final K key) {
final int hash = hash(key.hashCode());
return segmentFor(hash).containsKey(key, hash);
}
public V get(final K key) {
final int hash = hash(key.hashCode());
return segmentFor(hash).get(key, hash);
}
//add
public V put(K key, V value) {
if(value == null)
throw new NullPointerException();
final int hash = hash(key.hashCode());
return segmentFor(hash).put(key, hash, value);
}
//remove
public V remove(final K key) {
final int hash = hash(key.hashCode());
return segmentFor(hash).remove(key, hash);
}
//general
public void clear() {}
}
