src/jsr166e/StripedAdder.java

/*
 * Written by Doug Lea with assistance from members of JCP JSR-166
 * Expert Group and released to the public domain, as explained at
 * http://creativecommons.org/publicdomain/zero/1.0/
 */

package jsr166e;
import java.util.Arrays;
import java.util.Random;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.atomic.AtomicLong;
import java.io.IOException;
import java.io.Serializable;
import java.io.ObjectInputStream;
import java.io.ObjectOutputStream;

/**
 * A set of variables that together maintain a sum.  When updates
 * (method {@link #add}) are contended across threads, this set of
 * adder variables may grow dynamically to reduce contention. Method
 * {@link #sum} returns the current combined total across these
 * adders. This value is <em>NOT</em> an atomic snapshot (concurrent
 * updates may occur while the sum is being calculated), and so cannot
 * be used alone for fine-grained synchronization control.
 *
 * <p> This class may be applicable when many threads frequently
 * update a common sum that is used for purposes such as collecting
 * statistics. In this case, performance may be significantly faster
 * than using a shared {@link AtomicLong}, at the expense of using
 * much more space.  On the other hand, if it is known that only one
 * thread can ever update the sum, performance may be significantly
 * slower than just updating a local variable.
 *
 * <p>A StripedAdder may optionally be constructed with a given
 * expected contention level; i.e., the number of threads that are
 * expected to concurrently update the sum. Supplying an accurate
 * value may improve performance by reducing the need for dynamic
 * adjustment.
 *
 * @author Doug Lea
 */
public class StripedAdder implements Serializable {
    private static final long serialVersionUID = 7249069246863182397L;

    /*
     * A StripedAdder maintains a table of Atomic long variables. The
     * table is indexed by per-thread hash codes.
     *
     * By default, the table is lazily initialized, to minimize
     * footprint until adders are used. On first use, the table is set
     * to size DEFAULT_INITIAL_SIZE (currently 8). Table size is
     * bounded by the number of CPUS (if larger than the default
     * size).
     *
     * Per-thread hash codes are initialized to random values.
     * Collisions are indicated by failed CASes when performing an add
     * operation (see method retryAdd). Upon a collision, if the table
     * size is less than the capacity, it is doubled in size unless
     * some other thread holds lock. If a hashed slot is empty, and
     * lock is available, a new Adder is created. Otherwise, if the
     * slot exists, a CAS is tried.  Retries proceed by "double
     * hashing", using a secondary hash (Marsaglia XorShift) to try to
     * find a free slot.
     *
     * The table size is capped because, when there are more threads
     * than CPUs, supposing that each thread were bound to a CPU,
     * there would exist a perfect hash function mapping threads to
     * slots that eliminates collisions. When we reach capacity, we
     * search for this mapping by randomly varying the hash codes of
     * colliding threads.  Because search is random, and failures only
     * become known via CAS failures, convergence will be slow, and
     * because threads are typically not bound to CPUS forever, may
     * not occur at all. However, despite these limitations, observed
     * contention is typically low in these cases.
     *
     * Table entries are of class Adder; a form of AtomicLong padded
     * to reduce cache contention on most processors. Padding is
     * overkill for most Atomics because they are usually irregularly
     * scattered in memory and thus don't interfere much with each
     * other. But Atomic objects residing in arrays will tend to be
     * placed adjacent to each other, and so will most often share
     * cache lines without this precaution.  Adders are by default
     * constructed upon first use, which further improves per-thread
     * locality and helps reduce footprint.
     *
     * A single spinlock is used for resizing the table as well as
     * populating slots with new Adders. Upon lock contention, threads
     * try other slots rather than blocking. After initialization, at
     * least one slot exists, so retries will eventually find a
     * candidate Adder.  During these retries, there is increased
     * contention and reduced locality, which is still better than
     * alternatives.
     */

    /**
     * Padded version of AtomicLong
     */
    static final class Adder extends AtomicLong {
        long p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pa, pb, pc, pd, pe;
        Adder(long x) { super(x); }
    }

    private static final int NCPU = Runtime.getRuntime().availableProcessors();

    /**
     * Table bounds. DEFAULT_INITIAL_SIZE is the table size set upon
     * first use under default constructor, and must be a power of
     * two. There is not much point in making size a lot smaller than
     * that of Adders though.  CAP is the maximum allowed table size.
     */
    private static final int DEFAULT_INITIAL_SIZE = 8;
    private static final int CAP = Math.max(NCPU, DEFAULT_INITIAL_SIZE);

    /**
     * Holder for the thread-local hash code. The code is initially
     * random, but may be set to a different value upon collisions.
     */
    static final class HashCode {
        static final Random rng = new Random();
        int code;
        HashCode() {
            int h = rng.nextInt();
            code = (h == 0) ? 1 : h; // ensure nonzero
        }
    }

    /**
     * The corresponding ThreadLocal class
     */
    static final class ThreadHashCode extends ThreadLocal<HashCode> {
        public HashCode initialValue() { return new HashCode(); }
    }

    /**
     * Static per-thread hash codes. Shared across all StripedAdders
     * to reduce ThreadLocal pollution and because adjustments due to
     * collisions in one table are likely to be appropriate for
     * others.
     */
    static final ThreadHashCode threadHashCode = new ThreadHashCode();

    /**
     * Table of adders. Size is power of two, grows to be at most CAP.
     */
    private transient volatile Adder[] adders;

    /**
     * Serves as a lock when resizing and/or creating Adders.  There
     * is no need for a blocking lock: Except during initialization
     * races, when busy, other threads try other slots. However,
     * during (double-checked) initializations, we use the
     * "synchronized" lock on this object.
     */
    private final AtomicInteger mutex;

    /**
     * Creates a new adder with zero sum.
     */
    public StripedAdder() {
        this.mutex = new AtomicInteger();
        // remaining initialization on first call to add.
    }

    /**
     * Creates a new adder with zero sum, and with stripes presized
     * for the given expected contention level.
     *
     * @param expectedContention the expected number of threads that
     * will concurrently update the sum.
     */
    public StripedAdder(int expectedContention) {
        int cap = (expectedContention < CAP) ? expectedContention : CAP;
        int size = 1;
        while (size < cap)
            size <<= 1;
        Adder[] as = new Adder[size];
        for (int i = 0; i < size; ++i)
            as[i] = new Adder(0);
        this.adders = as;
        this.mutex = new AtomicInteger();
    }

    /**
     * Adds the given value.
     *
     * @param x the value to add
     */
    public void add(long x) {
        Adder[] as; Adder a; int n; long v; // locals to hold volatile reads
        HashCode hc = threadHashCode.get();
        int h = hc.code;
        if ((as = adders) == null || (n = as.length) < 1 ||
            (a = as[(n - 1) & h]) == null ||
            !a.compareAndSet(v = a.get(), v + x))
            retryAdd(x, hc);
    }

    /**
     * Handle cases of add involving initialization, resizing,
     * creating new Adders, and/or contention. See above for
     * explanation.
     */
    private void retryAdd(long x, HashCode hc) {
        int h = hc.code;
        final AtomicInteger mutex = this.mutex;
        int collisions = 1 - mutex.get(); // first guess: collides if not locked
        for (;;) {
            Adder[] as; Adder a; long v; int k, n;
            while ((as = adders) == null || (n = as.length) < 1) {
                synchronized(mutex) {                // Try to initialize
                    if (adders == null) {
                        Adder[] rs = new Adder[DEFAULT_INITIAL_SIZE];
                        rs[h & (DEFAULT_INITIAL_SIZE - 1)] = new Adder(0);
                        adders = rs;
                    }
                }
                collisions = 0;
            }

            if ((a = as[k = (n - 1) & h]) == null) { // Try to add slot
                if (mutex.get() == 0 && mutex.compareAndSet(0, 1)) {
                    try {
                        if (adders == as && as[k] == null)
                            a = as[k] = new Adder(x);
                    } finally {
                        mutex.set(0);
                    }
                    if (a != null)
                        break;
                }
                collisions = 0;
            }
            else if (collisions != 0 && n < CAP &&   // Try to expand table
                     mutex.get() == 0 && mutex.compareAndSet(0, 1)) {
                try {
                    if (adders == as) {
                        Adder[] rs = new Adder[n << 1];
                        for (int i = 0; i < n; ++i)
                            rs[i] = as[i];
                        adders = rs;
                    }
                } finally {
                    mutex.set(0);
                }
                collisions = 0;
            }
            else if (a.compareAndSet(v = a.get(), v + x))
                break;
            else
                collisions = 1;
            h ^= h << 13;                            // Rehash
            h ^= h >>> 17;
            h ^= h << 5;
        }
        hc.code = h;
    }

    /**
     * Returns an estimate of the current sum.  The result is
     * calculated by summing multiple variables, so may not be
     * accurate if updates occur concurrently with this method.
     *
     * @return the estimated sum
     */
    public long sum() {
        long sum = 0L;
        Adder[] as = adders;
        if (as != null) {
            int n = as.length;
            for (int i = 0; i < n; ++i) {
                Adder a = as[i];
                if (a != null)
                    sum += a.get();
            }
        }
        return sum;
    }

    /**
     * Resets each of the variables to zero. This is effective in
     * fully resetting the sum only if there are no concurrent
     * updates.
     */
    public void reset() {
        Adder[] as = adders;
        if (as != null) {
            int n = as.length;
            for (int i = 0; i < n; ++i) {
                Adder a = as[i];
                if (a != null)
                    a.set(0L);
            }
        }
    }

    /**
     * Equivalent to {@code add(1)}.
     */
    public void increment() {
        add(1L);
    }

    /**
     * Equivalent to {@code add(-1)}.
     */
    public void decrement() {
        add(-1L);
    }

    /**
     * Equivalent to {@link #sum} followed by {@link #reset}.
     *
     * @return the estimated sum
     */
    public long sumAndReset() {
        long sum = 0L;
        Adder[] as = adders;
        if (as != null) {
            int n = as.length;
            for (int i = 0; i < n; ++i) {
                Adder a = as[i];
                if (a != null) {
                    sum += a.get();
                    a.set(0L);
                }
            }
        }
        return sum;
    }

    private void writeObject(java.io.ObjectOutputStream s)
        throws java.io.IOException {
        s.defaultWriteObject();
        s.writeLong(sum());
    }

    private void readObject(ObjectInputStream s)
        throws IOException, ClassNotFoundException {
        s.defaultReadObject();
        mutex.set(0);
        add(s.readLong());
    }

}
Revision:	1.6
Committed:	Tue Jul 26 17:16:36 2011 UTC (12 years, 10 months ago) by dl
Branch:	MAIN
Changes since 1.5:	+102 -86 lines
Log Message:	Performance improvements
#	User	Rev	Content
1	dl	1.1	/*
2			* Written by Doug Lea with assistance from members of JCP JSR-166
3			* Expert Group and released to the public domain, as explained at
4			* http://creativecommons.org/publicdomain/zero/1.0/
5			*/
6
7			package jsr166e;
8			import java.util.Arrays;
9			import java.util.Random;
10			import java.util.concurrent.atomic.AtomicInteger;
11			import java.util.concurrent.atomic.AtomicLong;
12			import java.io.IOException;
13			import java.io.Serializable;
14			import java.io.ObjectInputStream;
15			import java.io.ObjectOutputStream;
16
17			/**
18			* A set of variables that together maintain a sum. When updates
19	dl	1.3	* (method {@link #add}) are contended across threads, this set of
20			* adder variables may grow dynamically to reduce contention. Method
21			* {@link #sum} returns the current combined total across these
22			* adders. This value is <em>NOT</em> an atomic snapshot (concurrent
23			* updates may occur while the sum is being calculated), and so cannot
24			* be used alone for fine-grained synchronization control.
25	jsr166	1.2	*
26	dl	1.1	* <p> This class may be applicable when many threads frequently
27			* update a common sum that is used for purposes such as collecting
28			* statistics. In this case, performance may be significantly faster
29			* than using a shared {@link AtomicLong}, at the expense of using
30	dl	1.4	* much more space. On the other hand, if it is known that only one
31			* thread can ever update the sum, performance may be significantly
32			* slower than just updating a local variable.
33	dl	1.1	*
34	dl	1.3	* <p>A StripedAdder may optionally be constructed with a given
35			* expected contention level; i.e., the number of threads that are
36			* expected to concurrently update the sum. Supplying an accurate
37			* value may improve performance by reducing the need for dynamic
38			* adjustment.
39			*
40	jsr166	1.2	* @author Doug Lea
41	dl	1.1	*/
42			public class StripedAdder implements Serializable {
43			private static final long serialVersionUID = 7249069246863182397L;
44
45			/*
46	dl	1.4	* A StripedAdder maintains a table of Atomic long variables. The
47	dl	1.6	* table is indexed by per-thread hash codes.
48	dl	1.3	*
49	dl	1.6	* By default, the table is lazily initialized, to minimize
50			* footprint until adders are used. On first use, the table is set
51			* to size DEFAULT_INITIAL_SIZE (currently 8). Table size is
52			* bounded by the number of CPUS (if larger than the default
53			* size).
54			*
55			* Per-thread hash codes are initialized to random values.
56			* Collisions are indicated by failed CASes when performing an add
57			* operation (see method retryAdd). Upon a collision, if the table
58			* size is less than the capacity, it is doubled in size unless
59			* some other thread holds lock. If a hashed slot is empty, and
60			* lock is available, a new Adder is created. Otherwise, if the
61			* slot exists, a CAS is tried. Retries proceed by "double
62			* hashing", using a secondary hash (Marsaglia XorShift) to try to
63			* find a free slot.
64			*
65			* The table size is capped because, when there are more threads
66			* than CPUs, supposing that each thread were bound to a CPU,
67			* there would exist a perfect hash function mapping threads to
68			* slots that eliminates collisions. When we reach capacity, we
69			* search for this mapping by randomly varying the hash codes of
70			* colliding threads. Because search is random, and failures only
71			* become known via CAS failures, convergence will be slow, and
72			* because threads are typically not bound to CPUS forever, may
73			* not occur at all. However, despite these limitations, observed
74			* contention is typically low in these cases.
75	dl	1.3	*
76			* Table entries are of class Adder; a form of AtomicLong padded
77			* to reduce cache contention on most processors. Padding is
78	dl	1.6	* overkill for most Atomics because they are usually irregularly
79			* scattered in memory and thus don't interfere much with each
80			* other. But Atomic objects residing in arrays will tend to be
81			* placed adjacent to each other, and so will most often share
82			* cache lines without this precaution. Adders are by default
83	dl	1.4	* constructed upon first use, which further improves per-thread
84	dl	1.6	* locality and helps reduce footprint.
85	dl	1.3	*
86			* A single spinlock is used for resizing the table as well as
87	dl	1.1	* populating slots with new Adders. Upon lock contention, threads
88	dl	1.4	* try other slots rather than blocking. After initialization, at
89			* least one slot exists, so retries will eventually find a
90	dl	1.6	* candidate Adder. During these retries, there is increased
91	dl	1.3	* contention and reduced locality, which is still better than
92			* alternatives.
93	dl	1.1	*/
94
95	jsr166	1.2	/**
96	dl	1.6	* Padded version of AtomicLong
97	dl	1.1	*/
98	dl	1.6	static final class Adder extends AtomicLong {
99			long p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pa, pb, pc, pd, pe;
100			Adder(long x) { super(x); }
101			}
102	dl	1.1
103	dl	1.6	private static final int NCPU = Runtime.getRuntime().availableProcessors();
104	dl	1.5
105			/**
106	dl	1.6	* Table bounds. DEFAULT_INITIAL_SIZE is the table size set upon
107			* first use under default constructor, and must be a power of
108			* two. There is not much point in making size a lot smaller than
109			* that of Adders though. CAP is the maximum allowed table size.
110	dl	1.1	*/
111	dl	1.6	private static final int DEFAULT_INITIAL_SIZE = 8;
112			private static final int CAP = Math.max(NCPU, DEFAULT_INITIAL_SIZE);
113	dl	1.1
114	jsr166	1.2	/**
115	dl	1.6	* Holder for the thread-local hash code. The code is initially
116			* random, but may be set to a different value upon collisions.
117	dl	1.1	*/
118			static final class HashCode {
119	dl	1.6	static final Random rng = new Random();
120	dl	1.1	int code;
121	dl	1.6	HashCode() {
122			int h = rng.nextInt();
123			code = (h == 0) ? 1 : h; // ensure nonzero
124			}
125	dl	1.1	}
126
127			/**
128			* The corresponding ThreadLocal class
129			*/
130			static final class ThreadHashCode extends ThreadLocal<HashCode> {
131	dl	1.6	public HashCode initialValue() { return new HashCode(); }
132	dl	1.1	}
133
134			/**
135			* Static per-thread hash codes. Shared across all StripedAdders
136	dl	1.6	* to reduce ThreadLocal pollution and because adjustments due to
137			* collisions in one table are likely to be appropriate for
138			* others.
139	dl	1.1	*/
140			static final ThreadHashCode threadHashCode = new ThreadHashCode();
141
142			/**
143	dl	1.6	* Table of adders. Size is power of two, grows to be at most CAP.
144	dl	1.1	*/
145			private transient volatile Adder[] adders;
146
147			/**
148			* Serves as a lock when resizing and/or creating Adders. There
149	dl	1.5	* is no need for a blocking lock: Except during initialization
150	dl	1.6	* races, when busy, other threads try other slots. However,
151			* during (double-checked) initializations, we use the
152			* "synchronized" lock on this object.
153	dl	1.1	*/
154			private final AtomicInteger mutex;
155
156			/**
157	dl	1.3	* Creates a new adder with zero sum.
158	dl	1.1	*/
159			public StripedAdder() {
160	dl	1.4	this.mutex = new AtomicInteger();
161			// remaining initialization on first call to add.
162	dl	1.3	}
163
164			/**
165			* Creates a new adder with zero sum, and with stripes presized
166			* for the given expected contention level.
167			*
168			* @param expectedContention the expected number of threads that
169			* will concurrently update the sum.
170			*/
171			public StripedAdder(int expectedContention) {
172	dl	1.6	int cap = (expectedContention < CAP) ? expectedContention : CAP;
173			int size = 1;
174			while (size < cap)
175			size <<= 1;
176	dl	1.3	Adder[] as = new Adder[size];
177	dl	1.4	for (int i = 0; i < size; ++i)
178			as[i] = new Adder(0);
179	dl	1.1	this.adders = as;
180			this.mutex = new AtomicInteger();
181			}
182
183			/**
184			* Adds the given value.
185			*
186			* @param x the value to add
187			*/
188			public void add(long x) {
189	dl	1.4	Adder[] as; Adder a; int n; long v; // locals to hold volatile reads
190	dl	1.1	HashCode hc = threadHashCode.get();
191	dl	1.6	int h = hc.code;
192	dl	1.4	if ((as = adders) == null \|\| (n = as.length) < 1 \|\|
193	dl	1.6	(a = as[(n - 1) & h]) == null \|\|
194	dl	1.4	!a.compareAndSet(v = a.get(), v + x))
195			retryAdd(x, hc);
196			}
197
198			/**
199			* Handle cases of add involving initialization, resizing,
200	dl	1.6	* creating new Adders, and/or contention. See above for
201			* explanation.
202	dl	1.4	*/
203			private void retryAdd(long x, HashCode hc) {
204			int h = hc.code;
205			final AtomicInteger mutex = this.mutex;
206	dl	1.6	int collisions = 1 - mutex.get(); // first guess: collides if not locked
207			for (;;) {
208			Adder[] as; Adder a; long v; int k, n;
209			while ((as = adders) == null \|\| (n = as.length) < 1) {
210			synchronized(mutex) { // Try to initialize
211			if (adders == null) {
212			Adder[] rs = new Adder[DEFAULT_INITIAL_SIZE];
213			rs[h & (DEFAULT_INITIAL_SIZE - 1)] = new Adder(0);
214			adders = rs;
215			}
216			}
217			collisions = 0;
218			}
219
220			if ((a = as[k = (n - 1) & h]) == null) { // Try to add slot
221	dl	1.5	if (mutex.get() == 0 && mutex.compareAndSet(0, 1)) {
222			try {
223	dl	1.6	if (adders == as && as[k] == null)
224			a = as[k] = new Adder(x);
225	dl	1.5	} finally {
226			mutex.set(0);
227			}
228	dl	1.6	if (a != null)
229			break;
230	dl	1.1	}
231	dl	1.6	collisions = 0;
232	dl	1.5	}
233	dl	1.6	else if (collisions != 0 && n < CAP && // Try to expand table
234	dl	1.5	mutex.get() == 0 && mutex.compareAndSet(0, 1)) {
235			try {
236			if (adders == as) {
237	dl	1.6	Adder[] rs = new Adder[n << 1];
238			for (int i = 0; i < n; ++i)
239			rs[i] = as[i];
240			adders = rs;
241	dl	1.1	}
242	dl	1.5	} finally {
243			mutex.set(0);
244	dl	1.1	}
245	dl	1.6	collisions = 0;
246	dl	1.5	}
247	dl	1.6	else if (a.compareAndSet(v = a.get(), v + x))
248			break;
249			else
250			collisions = 1;
251			h ^= h << 13; // Rehash
252			h ^= h >>> 17;
253			h ^= h << 5;
254	dl	1.1	}
255	dl	1.5	hc.code = h;
256	dl	1.1	}
257
258			/**
259			* Returns an estimate of the current sum. The result is
260			* calculated by summing multiple variables, so may not be
261			* accurate if updates occur concurrently with this method.
262	jsr166	1.2	*
263	dl	1.1	* @return the estimated sum
264			*/
265			public long sum() {
266	dl	1.4	long sum = 0L;
267	dl	1.1	Adder[] as = adders;
268	dl	1.4	if (as != null) {
269			int n = as.length;
270			for (int i = 0; i < n; ++i) {
271			Adder a = as[i];
272			if (a != null)
273			sum += a.get();
274			}
275	dl	1.1	}
276			return sum;
277			}
278
279			/**
280			* Resets each of the variables to zero. This is effective in
281			* fully resetting the sum only if there are no concurrent
282			* updates.
283			*/
284			public void reset() {
285			Adder[] as = adders;
286	dl	1.4	if (as != null) {
287			int n = as.length;
288			for (int i = 0; i < n; ++i) {
289			Adder a = as[i];
290			if (a != null)
291			a.set(0L);
292			}
293	dl	1.1	}
294			}
295
296			/**
297			* Equivalent to {@code add(1)}.
298			*/
299			public void increment() {
300			add(1L);
301			}
302
303			/**
304			* Equivalent to {@code add(-1)}.
305			*/
306			public void decrement() {
307			add(-1L);
308			}
309
310			/**
311			* Equivalent to {@link #sum} followed by {@link #reset}.
312			*
313			* @return the estimated sum
314			*/
315			public long sumAndReset() {
316	dl	1.4	long sum = 0L;
317	dl	1.1	Adder[] as = adders;
318	dl	1.4	if (as != null) {
319			int n = as.length;
320			for (int i = 0; i < n; ++i) {
321			Adder a = as[i];
322			if (a != null) {
323			sum += a.get();
324			a.set(0L);
325			}
326	dl	1.1	}
327			}
328			return sum;
329			}
330
331			private void writeObject(java.io.ObjectOutputStream s)
332			throws java.io.IOException {
333			s.defaultWriteObject();
334			s.writeLong(sum());
335			}
336
337			private void readObject(ObjectInputStream s)
338			throws IOException, ClassNotFoundException {
339			s.defaultReadObject();
340			mutex.set(0);
341	dl	1.4	add(s.readLong());
342	dl	1.1	}
343
344			}