16 |
|
|
17 |
|
/** |
18 |
|
* A set of variables that together maintain a sum. When updates |
19 |
< |
* (method {@link #add}) are contended across threads, the set of |
20 |
< |
* adders may grow to reduce contention. Method {@link #sum} returns |
21 |
< |
* the current combined total across these adders. This value is |
22 |
< |
* <em>NOT</em> an atomic snapshot (concurrent updates may occur while |
23 |
< |
* the sum is being calculated), and so cannot be used alone for |
24 |
< |
* fine-grained synchronization control. |
19 |
> |
* (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 |
|
* |
26 |
|
* <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 |
< |
* significantly more space. On the other hand, if it is known that |
31 |
< |
* only one thread can ever update the sum, performance may be |
32 |
< |
* significantly slower than just updating a local variable. |
30 |
> |
* 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 |
> |
* |
34 |
> |
* <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 |
|
* @author Doug Lea |
41 |
|
*/ |
43 |
|
private static final long serialVersionUID = 7249069246863182397L; |
44 |
|
|
45 |
|
/* |
46 |
< |
* Overview: We maintain a table of AtomicLongs (padded to reduce |
47 |
< |
* false sharing). The table is indexed by per-thread hash codes |
48 |
< |
* that are initialized as random values. The table doubles in |
49 |
< |
* size upon contention (as indicated by failed CASes when |
50 |
< |
* performing add()), but is capped at the nearest power of two >= |
51 |
< |
* #cpus: At that point, contention should be infrequent if each |
52 |
< |
* thread has a unique index; so we instead adjust hash codes to |
53 |
< |
* new random values upon contention rather than expanding. A |
54 |
< |
* single spinlock is used for resizing the table as well as |
46 |
> |
* A StripedAdder maintains a table of Atomic long variables. The |
47 |
> |
* table is indexed by per-thread hash codes that are initialized |
48 |
> |
* to random values. |
49 |
> |
* |
50 |
> |
* The table doubles in size upon contention (as indicated by |
51 |
> |
* failed CASes when performing add()), but is capped at the |
52 |
> |
* nearest power of two >= #CPUS. This reflects the idea that, |
53 |
> |
* when there are more threads than CPUs, then if each thread were |
54 |
> |
* bound to a CPU, there would exist a perfect hash function |
55 |
> |
* mapping threads to slots that eliminates collisions. When we |
56 |
> |
* reach capacity, we search for this mapping by randomly varying |
57 |
> |
* the hash codes of colliding threads. Because search is random, |
58 |
> |
* and failures only become known via CAS failures, convergence |
59 |
> |
* will be slow, and because threads are typically not bound to |
60 |
> |
* CPUS forever, may not occur at all. However, despite these |
61 |
> |
* limitations, observed contention is typically low in these |
62 |
> |
* cases. |
63 |
> |
* |
64 |
> |
* Table entries are of class Adder; a form of AtomicLong padded |
65 |
> |
* to reduce cache contention on most processors. Padding is |
66 |
> |
* overkill for most Atomics because they are most often |
67 |
> |
* irregularly scattered in memory and thus don't interfere much |
68 |
> |
* with each other. But Atomic objects residing in arrays will |
69 |
> |
* tend to be placed adjacent to each other, and so will most |
70 |
> |
* often share cache lines without this precaution. Adders are |
71 |
> |
* constructed upon first use, which further improves per-thread |
72 |
> |
* locality and helps reduce (an already large) footprint. |
73 |
> |
* |
74 |
> |
* A single spinlock is used for resizing the table as well as |
75 |
|
* populating slots with new Adders. Upon lock contention, threads |
76 |
< |
* just try other slots rather than blocking. We guarantee that at |
76 |
> |
* try other slots rather than blocking. After initialization, at |
77 |
|
* least one slot exists, so retries will eventually find a |
78 |
< |
* candidate Adder. |
78 |
> |
* candidate Adder. During these retries, there is increased |
79 |
> |
* contention and reduced locality, which is still better than |
80 |
> |
* alternatives. |
81 |
|
*/ |
82 |
|
|
83 |
|
/** |
86 |
|
static final int NCPU = Runtime.getRuntime().availableProcessors(); |
87 |
|
|
88 |
|
/** |
89 |
< |
* Version of AtomicLong padded to avoid sharing cache |
62 |
< |
* lines on most processors |
89 |
> |
* Padded version of AtomicLong |
90 |
|
*/ |
91 |
|
static final class Adder extends AtomicLong { |
92 |
|
long p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pa, pb, pc, pd; |
94 |
|
} |
95 |
|
|
96 |
|
/** |
97 |
< |
* Holder for the thread-local hash code. |
97 |
> |
* Holder for the thread-local hash code. The code starts off with |
98 |
> |
* a given random value, but may be set to a different value upon |
99 |
> |
* collisions in retryAdd. |
100 |
|
*/ |
101 |
|
static final class HashCode { |
102 |
|
int code; |
122 |
|
static final ThreadHashCode threadHashCode = new ThreadHashCode(); |
123 |
|
|
124 |
|
/** |
125 |
< |
* Table of adders. Initially of size 2; grows to be at most NCPU. |
125 |
> |
* Table of adders. Minimum size 2. Size grows to be at most NCPU. |
126 |
|
*/ |
127 |
|
private transient volatile Adder[] adders; |
128 |
|
|
134 |
|
private final AtomicInteger mutex; |
135 |
|
|
136 |
|
/** |
137 |
< |
* Marsaglia XorShift for rehashing on collisions |
137 |
> |
* Creates a new adder with zero sum. |
138 |
|
*/ |
139 |
< |
private static int xorShift(int r) { |
140 |
< |
r ^= r << 13; |
141 |
< |
r ^= r >>> 17; |
113 |
< |
return r ^ (r << 5); |
139 |
> |
public StripedAdder() { |
140 |
> |
this.mutex = new AtomicInteger(); |
141 |
> |
// remaining initialization on first call to add. |
142 |
|
} |
143 |
|
|
144 |
|
/** |
145 |
< |
* Creates a new adder with initially zero sum. |
145 |
> |
* Creates a new adder with zero sum, and with stripes presized |
146 |
> |
* for the given expected contention level. |
147 |
> |
* |
148 |
> |
* @param expectedContention the expected number of threads that |
149 |
> |
* will concurrently update the sum. |
150 |
|
*/ |
151 |
< |
public StripedAdder() { |
152 |
< |
Adder[] as = new Adder[2]; |
153 |
< |
as[0] = new Adder(0); // ensure at least one available adder |
151 |
> |
public StripedAdder(int expectedContention) { |
152 |
> |
int cap = (expectedContention < NCPU) ? expectedContention : NCPU; |
153 |
> |
int size = 2; |
154 |
> |
while (size < cap) |
155 |
> |
size <<= 1; |
156 |
> |
Adder[] as = new Adder[size]; |
157 |
> |
for (int i = 0; i < size; ++i) |
158 |
> |
as[i] = new Adder(0); |
159 |
|
this.adders = as; |
160 |
|
this.mutex = new AtomicInteger(); |
161 |
|
} |
166 |
|
* @param x the value to add |
167 |
|
*/ |
168 |
|
public void add(long x) { |
169 |
+ |
Adder[] as; Adder a; int n; long v; // locals to hold volatile reads |
170 |
|
HashCode hc = threadHashCode.get(); |
171 |
< |
for (int h = hc.code;;) { |
172 |
< |
Adder[] as = adders; |
173 |
< |
int n = as.length; |
174 |
< |
Adder a = as[h & (n - 1)]; |
175 |
< |
if (a != null) { |
176 |
< |
long v = a.get(); |
177 |
< |
if (a.compareAndSet(v, v + x)) |
178 |
< |
break; |
179 |
< |
if (n >= NCPU) { // Collision when table at max |
180 |
< |
h = hc.code = xorShift(h); // change code |
181 |
< |
continue; |
171 |
> |
if ((as = adders) == null || (n = as.length) < 1 || |
172 |
> |
(a = as[hc.code & (n - 1)]) == null || |
173 |
> |
!a.compareAndSet(v = a.get(), v + x)) |
174 |
> |
retryAdd(x, hc); |
175 |
> |
} |
176 |
> |
|
177 |
> |
/** |
178 |
> |
* Handle cases of add involving initialization, resizing, |
179 |
> |
* creating new Adders, and/or contention. |
180 |
> |
*/ |
181 |
> |
private void retryAdd(long x, HashCode hc) { |
182 |
> |
int h = hc.code; |
183 |
> |
final AtomicInteger mutex = this.mutex; |
184 |
> |
AtomicInteger lock = null; // nonnull when held |
185 |
> |
try { |
186 |
> |
for (;;) { |
187 |
> |
Adder[] as; Adder a; long v; int n, k; // locals for volatiles |
188 |
> |
boolean needLock = true; |
189 |
> |
if ((as = adders) == null || (n = as.length) < 1) { |
190 |
> |
if (lock != null) // default-initialize |
191 |
> |
adders = new Adder[2]; |
192 |
|
} |
193 |
< |
} |
194 |
< |
final AtomicInteger mutex = this.mutex; |
195 |
< |
if (mutex.get() != 0) |
196 |
< |
h = xorShift(h); // Try elsewhere |
149 |
< |
else if (mutex.compareAndSet(0, 1)) { |
150 |
< |
boolean created = false; |
151 |
< |
try { |
152 |
< |
Adder[] rs = adders; |
153 |
< |
if (a != null && rs == as) // Resize table |
154 |
< |
rs = adders = Arrays.copyOf(as, as.length << 1); |
155 |
< |
int j = h & (rs.length - 1); |
156 |
< |
if (rs[j] == null) { // Create adder |
157 |
< |
rs[j] = new Adder(x); |
158 |
< |
created = true; |
193 |
> |
else if ((a = as[k = h & (n - 1)]) == null) { |
194 |
> |
if (lock != null) { // attach new adder |
195 |
> |
as[k] = new Adder(x); |
196 |
> |
break; |
197 |
|
} |
160 |
– |
} finally { |
161 |
– |
mutex.set(0); |
198 |
|
} |
199 |
< |
if (created) { |
164 |
< |
hc.code = h; // Use this adder next time |
199 |
> |
else if (a.compareAndSet(v = a.get(), v + x)) |
200 |
|
break; |
201 |
+ |
else if (n >= NCPU) // cannot expand |
202 |
+ |
needLock = false; |
203 |
+ |
else if (lock != null) // expand table |
204 |
+ |
adders = Arrays.copyOf(as, n << 1); |
205 |
+ |
|
206 |
+ |
if (lock == null) { |
207 |
+ |
if (needLock && mutex.get() == 0 && |
208 |
+ |
mutex.compareAndSet(0, 1)) |
209 |
+ |
lock = mutex; |
210 |
+ |
else { // try elsewhere |
211 |
+ |
h ^= h << 13; // Marsaglia XorShift |
212 |
+ |
h ^= h >>> 17; |
213 |
+ |
h ^= h << 5; |
214 |
+ |
} |
215 |
|
} |
216 |
|
} |
217 |
+ |
} finally { |
218 |
+ |
if (lock != null) |
219 |
+ |
lock.set(0); |
220 |
|
} |
221 |
+ |
if (hc.code != h) // avoid unneeded writes |
222 |
+ |
hc.code = h; |
223 |
|
} |
224 |
|
|
225 |
|
/** |
230 |
|
* @return the estimated sum |
231 |
|
*/ |
232 |
|
public long sum() { |
233 |
< |
long sum = 0; |
233 |
> |
long sum = 0L; |
234 |
|
Adder[] as = adders; |
235 |
< |
int n = as.length; |
236 |
< |
for (int i = 0; i < n; ++i) { |
237 |
< |
Adder a = as[i]; |
238 |
< |
if (a != null) |
239 |
< |
sum += a.get(); |
235 |
> |
if (as != null) { |
236 |
> |
int n = as.length; |
237 |
> |
for (int i = 0; i < n; ++i) { |
238 |
> |
Adder a = as[i]; |
239 |
> |
if (a != null) |
240 |
> |
sum += a.get(); |
241 |
> |
} |
242 |
|
} |
243 |
|
return sum; |
244 |
|
} |
250 |
|
*/ |
251 |
|
public void reset() { |
252 |
|
Adder[] as = adders; |
253 |
< |
int n = as.length; |
254 |
< |
for (int i = 0; i < n; ++i) { |
255 |
< |
Adder a = as[i]; |
256 |
< |
if (a != null) |
257 |
< |
a.set(0L); |
253 |
> |
if (as != null) { |
254 |
> |
int n = as.length; |
255 |
> |
for (int i = 0; i < n; ++i) { |
256 |
> |
Adder a = as[i]; |
257 |
> |
if (a != null) |
258 |
> |
a.set(0L); |
259 |
> |
} |
260 |
|
} |
261 |
|
} |
262 |
|
|
280 |
|
* @return the estimated sum |
281 |
|
*/ |
282 |
|
public long sumAndReset() { |
283 |
< |
long sum = 0; |
283 |
> |
long sum = 0L; |
284 |
|
Adder[] as = adders; |
285 |
< |
int n = as.length; |
286 |
< |
for (int i = 0; i < n; ++i) { |
287 |
< |
Adder a = as[i]; |
288 |
< |
if (a != null) { |
289 |
< |
sum += a.get(); |
290 |
< |
a.set(0L); |
285 |
> |
if (as != null) { |
286 |
> |
int n = as.length; |
287 |
> |
for (int i = 0; i < n; ++i) { |
288 |
> |
Adder a = as[i]; |
289 |
> |
if (a != null) { |
290 |
> |
sum += a.get(); |
291 |
> |
a.set(0L); |
292 |
> |
} |
293 |
|
} |
294 |
|
} |
295 |
|
return sum; |
304 |
|
private void readObject(ObjectInputStream s) |
305 |
|
throws IOException, ClassNotFoundException { |
306 |
|
s.defaultReadObject(); |
247 |
– |
long c = s.readLong(); |
248 |
– |
Adder[] as = new Adder[2]; |
249 |
– |
as[0] = new Adder(c); |
250 |
– |
this.adders = as; |
307 |
|
mutex.set(0); |
308 |
+ |
add(s.readLong()); |
309 |
|
} |
310 |
|
|
311 |
|
} |
255 |
– |
|
256 |
– |
|