| 276 |
|
/** Interface describing a function mapping two ints to an int */ |
| 277 |
|
public interface IntByIntToInt { int apply(int a, int b); } |
| 278 |
|
|
| 279 |
+ |
|
| 280 |
|
/* |
| 281 |
|
* Overview: |
| 282 |
|
* |
| 382 |
|
* The table is resized when occupancy exceeds a percentage |
| 383 |
|
* threshold (nominally, 0.75, but see below). Any thread |
| 384 |
|
* noticing an overfull bin may assist in resizing after the |
| 385 |
< |
* initiating thread allocates and sets up the replacement |
| 386 |
< |
* array. However, rather than stalling, these other threads may |
| 387 |
< |
* proceed with insertions etc. The use of TreeBins shields us |
| 388 |
< |
* from the worst case effects of overfilling while resizes are in |
| 385 |
> |
* initiating thread allocates and sets up the replacement array. |
| 386 |
> |
* However, rather than stalling, these other threads may proceed |
| 387 |
> |
* with insertions etc. The use of TreeBins shields us from the |
| 388 |
> |
* worst case effects of overfilling while resizes are in |
| 389 |
|
* progress. Resizing proceeds by transferring bins, one by one, |
| 390 |
< |
* from the table to the next table. To enable concurrency, the |
| 391 |
< |
* next table must be (incrementally) prefilled with place-holders |
| 392 |
< |
* serving as reverse forwarders to the old table. Because we are |
| 393 |
< |
* using power-of-two expansion, the elements from each bin must |
| 394 |
< |
* either stay at same index, or move with a power of two |
| 395 |
< |
* offset. We eliminate unnecessary node creation by catching |
| 396 |
< |
* cases where old nodes can be reused because their next fields |
| 397 |
< |
* won't change. On average, only about one-sixth of them need |
| 398 |
< |
* cloning when a table doubles. The nodes they replace will be |
| 399 |
< |
* garbage collectable as soon as they are no longer referenced by |
| 400 |
< |
* any reader thread that may be in the midst of concurrently |
| 401 |
< |
* traversing table. Upon transfer, the old table bin contains |
| 402 |
< |
* only a special forwarding node (with hash field "MOVED") that |
| 403 |
< |
* contains the next table as its key. On encountering a |
| 404 |
< |
* forwarding node, access and update operations restart, using |
| 404 |
< |
* the new table. |
| 390 |
> |
* from the table to the next table. However, threads claim small |
| 391 |
> |
* blocks of indices to transfer (via field transferIndex) before |
| 392 |
> |
* doing so, reducing contention. Because we are using |
| 393 |
> |
* power-of-two expansion, the elements from each bin must either |
| 394 |
> |
* stay at same index, or move with a power of two offset. We |
| 395 |
> |
* eliminate unnecessary node creation by catching cases where old |
| 396 |
> |
* nodes can be reused because their next fields won't change. On |
| 397 |
> |
* average, only about one-sixth of them need cloning when a table |
| 398 |
> |
* doubles. The nodes they replace will be garbage collectable as |
| 399 |
> |
* soon as they are no longer referenced by any reader thread that |
| 400 |
> |
* may be in the midst of concurrently traversing table. Upon |
| 401 |
> |
* transfer, the old table bin contains only a special forwarding |
| 402 |
> |
* node (with hash field "MOVED") that contains the next table as |
| 403 |
> |
* its key. On encountering a forwarding node, access and update |
| 404 |
> |
* operations restart, using the new table. |
| 405 |
|
* |
| 406 |
|
* Each bin transfer requires its bin lock, which can stall |
| 407 |
|
* waiting for locks while resizing. However, because other |
| 409 |
|
* locks, average aggregate waits become shorter as resizing |
| 410 |
|
* progresses. The transfer operation must also ensure that all |
| 411 |
|
* accessible bins in both the old and new table are usable by any |
| 412 |
< |
* traversal. This is arranged by proceeding from the last bin |
| 413 |
< |
* (table.length - 1) up towards the first. Upon seeing a |
| 414 |
< |
* forwarding node, traversals (see class Traverser) arrange to |
| 415 |
< |
* move to the new table without revisiting nodes. However, to |
| 416 |
< |
* ensure that no intervening nodes are skipped, bin splitting can |
| 417 |
< |
* only begin after the associated reverse-forwarders are in |
| 418 |
< |
* place. |
| 412 |
> |
* traversal. This is arranged in part by proceeding from the |
| 413 |
> |
* last bin (table.length - 1) up towards the first. Upon seeing |
| 414 |
> |
* a forwarding node, traversals (see class Traverser) arrange to |
| 415 |
> |
* move to the new table without revisiting nodes. To ensure that |
| 416 |
> |
* no intervening nodes are skipped even when moved out of order, |
| 417 |
> |
* a stack (see class TableStack) is created on first encounter of |
| 418 |
> |
* a forwarding node during a traversal, to maintain its place if |
| 419 |
> |
* later processing the current table. The need for these |
| 420 |
> |
* save/restore mechanics is relatively rare, but when one |
| 421 |
> |
* forwarding node is encountered, typically many more will be. |
| 422 |
> |
* So Traversers use a simple caching scheme to avoid creating so |
| 423 |
> |
* many new TableStack nodes. (Thanks to Peter Levart for |
| 424 |
> |
* suggesting use of a stack here.) |
| 425 |
|
* |
| 426 |
|
* The traversal scheme also applies to partial traversals of |
| 427 |
|
* ranges of bins (via an alternate Traverser constructor) |
| 790 |
|
private transient volatile int transferIndex; |
| 791 |
|
|
| 792 |
|
/** |
| 787 |
– |
* The least available table index to split while resizing. |
| 788 |
– |
*/ |
| 789 |
– |
private transient volatile int transferOrigin; |
| 790 |
– |
|
| 791 |
– |
/** |
| 793 |
|
* Spinlock (locked via CAS) used when resizing and/or creating CounterCells. |
| 794 |
|
*/ |
| 795 |
|
private transient volatile int cellsBusy; |
| 2253 |
|
while (s >= (long)(sc = sizeCtl) && (tab = table) != null && |
| 2254 |
|
tab.length < MAXIMUM_CAPACITY) { |
| 2255 |
|
if (sc < 0) { |
| 2256 |
< |
if (sc == -1 || transferIndex <= transferOrigin || |
| 2256 |
> |
if (sc == -1 || transferIndex <= 0 || |
| 2257 |
|
(nt = nextTable) == null) |
| 2258 |
|
break; |
| 2259 |
|
if (U.compareAndSwapInt(this, SIZECTL, sc, sc - 1)) |
| 2273 |
|
Node<K,V>[] nextTab; int sc; |
| 2274 |
|
if ((f instanceof ForwardingNode) && |
| 2275 |
|
(nextTab = ((ForwardingNode<K,V>)f).nextTable) != null) { |
| 2276 |
< |
if (nextTab == nextTable && tab == table && |
| 2277 |
< |
transferIndex > transferOrigin && (sc = sizeCtl) < -1 && |
| 2278 |
< |
U.compareAndSwapInt(this, SIZECTL, sc, sc - 1)) |
| 2279 |
< |
transfer(tab, nextTab); |
| 2276 |
> |
while (transferIndex > 0 && nextTab == nextTable && |
| 2277 |
> |
(sc = sizeCtl) < -1) { |
| 2278 |
> |
if (U.compareAndSwapInt(this, SIZECTL, sc, sc - 1)) { |
| 2279 |
> |
transfer(tab, nextTab); |
| 2280 |
> |
break; |
| 2281 |
> |
} |
| 2282 |
> |
} |
| 2283 |
|
return nextTab; |
| 2284 |
|
} |
| 2285 |
|
return table; |
| 2330 |
|
if (nextTab == null) { // initiating |
| 2331 |
|
try { |
| 2332 |
|
@SuppressWarnings("unchecked") |
| 2333 |
< |
Node<K,V>[] nt = (Node<K,V>[])new Node<?,?>[n << 1]; |
| 2333 |
> |
Node<K,V>[] nt = (Node<K,V>[])new Node<?,?>[n << 1]; |
| 2334 |
|
nextTab = nt; |
| 2335 |
|
} catch (Throwable ex) { // try to cope with OOME |
| 2336 |
|
sizeCtl = Integer.MAX_VALUE; |
| 2337 |
|
return; |
| 2338 |
|
} |
| 2339 |
|
nextTable = nextTab; |
| 2336 |
– |
transferOrigin = n; |
| 2340 |
|
transferIndex = n; |
| 2338 |
– |
ForwardingNode<K,V> rev = new ForwardingNode<K,V>(tab); |
| 2339 |
– |
for (int k = n; k > 0;) { // progressively reveal ready slots |
| 2340 |
– |
int nextk = (k > stride) ? k - stride : 0; |
| 2341 |
– |
for (int m = nextk; m < k; ++m) |
| 2342 |
– |
nextTab[m] = rev; |
| 2343 |
– |
for (int m = n + nextk; m < n + k; ++m) |
| 2344 |
– |
nextTab[m] = rev; |
| 2345 |
– |
U.putOrderedInt(this, TRANSFERORIGIN, k = nextk); |
| 2346 |
– |
} |
| 2341 |
|
} |
| 2342 |
|
int nextn = nextTab.length; |
| 2343 |
|
ForwardingNode<K,V> fwd = new ForwardingNode<K,V>(nextTab); |
| 2344 |
|
boolean advance = true; |
| 2345 |
|
boolean finishing = false; // to ensure sweep before committing nextTab |
| 2346 |
|
for (int i = 0, bound = 0;;) { |
| 2347 |
< |
int nextIndex, nextBound, fh; Node<K,V> f; |
| 2347 |
> |
Node<K,V> f; int fh; |
| 2348 |
|
while (advance) { |
| 2349 |
+ |
int nextIndex, nextBound; |
| 2350 |
|
if (--i >= bound || finishing) |
| 2351 |
|
advance = false; |
| 2352 |
< |
else if ((nextIndex = transferIndex) <= transferOrigin) { |
| 2352 |
> |
else if ((nextIndex = transferIndex) <= 0) { |
| 2353 |
|
i = -1; |
| 2354 |
|
advance = false; |
| 2355 |
|
} |
| 2363 |
|
} |
| 2364 |
|
} |
| 2365 |
|
if (i < 0 || i >= n || i + n >= nextn) { |
| 2366 |
+ |
int sc; |
| 2367 |
|
if (finishing) { |
| 2368 |
|
nextTable = null; |
| 2369 |
|
table = nextTab; |
| 2370 |
|
sizeCtl = (n << 1) - (n >>> 1); |
| 2371 |
|
return; |
| 2372 |
|
} |
| 2373 |
< |
for (int sc;;) { |
| 2374 |
< |
if (U.compareAndSwapInt(this, SIZECTL, sc = sizeCtl, ++sc)) { |
| 2375 |
< |
if (sc != -1) |
| 2376 |
< |
return; |
| 2377 |
< |
finishing = advance = true; |
| 2382 |
< |
i = n; // recheck before commit |
| 2383 |
< |
break; |
| 2384 |
< |
} |
| 2385 |
< |
} |
| 2386 |
< |
} |
| 2387 |
< |
else if ((f = tabAt(tab, i)) == null) { |
| 2388 |
< |
if (casTabAt(tab, i, null, fwd)) { |
| 2389 |
< |
setTabAt(nextTab, i, null); |
| 2390 |
< |
setTabAt(nextTab, i + n, null); |
| 2391 |
< |
advance = true; |
| 2373 |
> |
if (U.compareAndSwapInt(this, SIZECTL, sc = sizeCtl, ++sc)) { |
| 2374 |
> |
if (sc != -1) |
| 2375 |
> |
return; |
| 2376 |
> |
finishing = advance = true; |
| 2377 |
> |
i = n; // recheck before commit |
| 2378 |
|
} |
| 2379 |
|
} |
| 2380 |
+ |
else if ((f = tabAt(tab, i)) == null) |
| 2381 |
+ |
advance = casTabAt(tab, i, null, fwd); |
| 2382 |
|
else if ((fh = f.hash) == MOVED) |
| 2383 |
|
advance = true; // already processed |
| 2384 |
|
else { |
| 3116 |
|
/* ----------------Table Traversal -------------- */ |
| 3117 |
|
|
| 3118 |
|
/** |
| 3119 |
+ |
* Records the table, its length, and current traversal index for a |
| 3120 |
+ |
* traverser that must process a region of a forwarded table before |
| 3121 |
+ |
* proceeding with current table. |
| 3122 |
+ |
*/ |
| 3123 |
+ |
static final class TableStack<K,V> { |
| 3124 |
+ |
int length; |
| 3125 |
+ |
int index; |
| 3126 |
+ |
Node<K,V>[] tab; |
| 3127 |
+ |
TableStack<K,V> next; |
| 3128 |
+ |
} |
| 3129 |
+ |
|
| 3130 |
+ |
/** |
| 3131 |
|
* Encapsulates traversal for methods such as containsValue; also |
| 3132 |
|
* serves as a base class for other iterators and spliterators. |
| 3133 |
|
* |
| 3151 |
|
static class Traverser<K,V> { |
| 3152 |
|
Node<K,V>[] tab; // current table; updated if resized |
| 3153 |
|
Node<K,V> next; // the next entry to use |
| 3154 |
+ |
TableStack<K,V> stack, spare; // to save/restore on ForwardingNodes |
| 3155 |
|
int index; // index of bin to use next |
| 3156 |
|
int baseIndex; // current index of initial table |
| 3157 |
|
int baseLimit; // index bound for initial table |
| 3173 |
|
if ((e = next) != null) |
| 3174 |
|
e = e.next; |
| 3175 |
|
for (;;) { |
| 3176 |
< |
Node<K,V>[] t; int i, n; K ek; // must use locals in checks |
| 3176 |
> |
Node<K,V>[] t; int i, n; // must use locals in checks |
| 3177 |
|
if (e != null) |
| 3178 |
|
return next = e; |
| 3179 |
|
if (baseIndex >= baseLimit || (t = tab) == null || |
| 3180 |
|
(n = t.length) <= (i = index) || i < 0) |
| 3181 |
|
return next = null; |
| 3182 |
< |
if ((e = tabAt(t, index)) != null && e.hash < 0) { |
| 3182 |
> |
if ((e = tabAt(t, i)) != null && e.hash < 0) { |
| 3183 |
|
if (e instanceof ForwardingNode) { |
| 3184 |
|
tab = ((ForwardingNode<K,V>)e).nextTable; |
| 3185 |
|
e = null; |
| 3186 |
+ |
pushState(t, i, n); |
| 3187 |
|
continue; |
| 3188 |
|
} |
| 3189 |
|
else if (e instanceof TreeBin) |
| 3191 |
|
else |
| 3192 |
|
e = null; |
| 3193 |
|
} |
| 3194 |
< |
if ((index += baseSize) >= n) |
| 3195 |
< |
index = ++baseIndex; // visit upper slots if present |
| 3194 |
> |
if (stack != null) |
| 3195 |
> |
recoverState(n); |
| 3196 |
> |
else if ((index = i + baseSize) >= n) |
| 3197 |
> |
index = ++baseIndex; // visit upper slots if present |
| 3198 |
> |
} |
| 3199 |
> |
} |
| 3200 |
> |
|
| 3201 |
> |
/** |
| 3202 |
> |
* Saves traversal state upon encountering a forwarding node. |
| 3203 |
> |
*/ |
| 3204 |
> |
private void pushState(Node<K,V>[] t, int i, int n) { |
| 3205 |
> |
TableStack<K,V> s = spare; // reuse if possible |
| 3206 |
> |
if (s != null) |
| 3207 |
> |
spare = s.next; |
| 3208 |
> |
else |
| 3209 |
> |
s = new TableStack<K,V>(); |
| 3210 |
> |
s.tab = t; |
| 3211 |
> |
s.length = n; |
| 3212 |
> |
s.index = i; |
| 3213 |
> |
s.next = stack; |
| 3214 |
> |
stack = s; |
| 3215 |
> |
} |
| 3216 |
> |
|
| 3217 |
> |
/** |
| 3218 |
> |
* Possibly pops traversal state. |
| 3219 |
> |
* |
| 3220 |
> |
* @param n length of current table |
| 3221 |
> |
*/ |
| 3222 |
> |
private void recoverState(int n) { |
| 3223 |
> |
TableStack<K,V> s; int len; |
| 3224 |
> |
while ((s = stack) != null && (index += (len = s.length)) >= n) { |
| 3225 |
> |
n = len; |
| 3226 |
> |
index = s.index; |
| 3227 |
> |
tab = s.tab; |
| 3228 |
> |
s.tab = null; |
| 3229 |
> |
TableStack<K,V> next = s.next; |
| 3230 |
> |
s.next = spare; // save for reuse |
| 3231 |
> |
stack = next; |
| 3232 |
> |
spare = s; |
| 3233 |
|
} |
| 3234 |
+ |
if (s == null && (index += baseSize) >= n) |
| 3235 |
+ |
index = ++baseIndex; |
| 3236 |
|
} |
| 3237 |
|
} |
| 3238 |
|
|
| 6200 |
|
private static final sun.misc.Unsafe U; |
| 6201 |
|
private static final long SIZECTL; |
| 6202 |
|
private static final long TRANSFERINDEX; |
| 6162 |
– |
private static final long TRANSFERORIGIN; |
| 6203 |
|
private static final long BASECOUNT; |
| 6204 |
|
private static final long CELLSBUSY; |
| 6205 |
|
private static final long CELLVALUE; |
| 6214 |
|
(k.getDeclaredField("sizeCtl")); |
| 6215 |
|
TRANSFERINDEX = U.objectFieldOffset |
| 6216 |
|
(k.getDeclaredField("transferIndex")); |
| 6177 |
– |
TRANSFERORIGIN = U.objectFieldOffset |
| 6178 |
– |
(k.getDeclaredField("transferOrigin")); |
| 6217 |
|
BASECOUNT = U.objectFieldOffset |
| 6218 |
|
(k.getDeclaredField("baseCount")); |
| 6219 |
|
CELLSBUSY = U.objectFieldOffset |
