/*
* 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.
*/
package org.jctools.queues.atomic;
import org.jctools.queues.MessagePassingQueue;
import org.jctools.queues.QueueProgressIndicators;
import org.jctools.util.Pow2;
import org.jctools.util.RangeUtil;
import java.util.AbstractQueue;
import java.util.Iterator;
import java.util.concurrent.atomic.AtomicLong;
import java.util.concurrent.atomic.AtomicReferenceArray;
import static org.jctools.util.JvmInfo.CPUs;
abstract class BaseMpscLinkedAtomicArrayQueuePad1<E> extends AbstractQueue<E> {
long p01, p02, p03, p04, p05, p06, p07;
long p10, p11, p12, p13, p14, p15, p16, p17;
}
abstract class BaseMpscLinkedAtomicArrayQueueProducerFields<E> extends BaseMpscLinkedAtomicArrayQueuePad1<E> {
protected AtomicLong producerIndex;
}
abstract class BaseMpscLinkedAtomicArrayQueuePad2<E> extends BaseMpscLinkedAtomicArrayQueueProducerFields<E> {
long p01, p02, p03, p04, p05, p06, p07;
long p10, p11, p12, p13, p14, p15, p16, p17;
}
abstract class BaseMpscLinkedAtomicArrayQueueConsumerFields<E> extends BaseMpscLinkedAtomicArrayQueuePad2<E> {
protected long consumerMask;
protected AtomicReferenceArray<E> consumerBuffer;
protected AtomicLong consumerIndex;
}
abstract class BaseMpscLinkedAtomicArrayQueuePad3<E> extends BaseMpscLinkedAtomicArrayQueueConsumerFields<E> {
long p0, p1, p2, p3, p4, p5, p6, p7;
long p10, p11, p12, p13, p14, p15, p16, p17;
}
abstract class BaseMpscLinkedAtomicArrayQueueColdProducerFields<E> extends BaseMpscLinkedAtomicArrayQueuePad3<E> {
// todo evaluate volatile
protected volatile AtomicLong producerLimit;
protected long producerMask;
protected AtomicReferenceArray<E> producerBuffer;
}
/**
* An MPSC array queue which starts at <i>initialCapacity</i> and grows to <i>maxCapacity</i> in linked chunks
* of the initial size. The queue grows only when the current buffer is full and elements are not copied on
* resize, instead a link to the new buffer is stored in the old buffer for the consumer to follow.<br>
*
* @param <E>
*/
public abstract class BaseMpscLinkedAtomicArrayQueue<E> extends BaseMpscLinkedAtomicArrayQueueColdProducerFields<E>
implements MessagePassingQueue<E>, QueueProgressIndicators {
// No post padding here, subclasses must add
// todo remove duplication
private final static int RECOMMENDED_OFFER_BATCH = CPUs * 4;
private final static Object JUMP = new Object();
/**
* @param initialCapacity the queue initial capacity. If chunk size is fixed this will be the chunk size.
* Must be 2 or more.
*/
public BaseMpscLinkedAtomicArrayQueue(final int initialCapacity) {
RangeUtil.checkGreaterThanOrEqual(initialCapacity, 2, "initialCapacity");
this.consumerIndex = new AtomicLong();
this.producerIndex = new AtomicLong();
this.producerLimit = new AtomicLong();
int p2capacity = Pow2.roundToPowerOfTwo(initialCapacity);
// leave lower bit of mask clear
long mask = (p2capacity - 1) << 1;
// need extra element to point at next array
AtomicReferenceArray<E> buffer = new AtomicReferenceArray<E>(p2capacity + 1);
producerBuffer = buffer;
producerMask = mask;
consumerBuffer = buffer;
consumerMask = mask;
soProducerLimit(mask); // we know it's all empty to start with
}
@Override
public final Iterator<E> iterator() {
throw new UnsupportedOperationException();
}
@Override
public String toString() {
return this.getClass().getName();
}
@Override
public boolean offer(final E e) {
if (null == e) {
throw new NullPointerException();
}
long mask;
AtomicReferenceArray<E> buffer;
long pIndex;
while (true) {
long producerLimit = lvProducerLimit();
pIndex = lvProducerIndex();
// lower bit is indicative of resize, if we see it we spin until it's cleared
if ((pIndex & 1) == 1) {
continue;
}
// pIndex is even (lower bit is 0) -> actual index is (pIndex >> 1)
// mask/buffer may get changed by resizing -> only use for array access after successful CAS.
mask = this.producerMask;
buffer = this.producerBuffer;
// a successful CAS ties the ordering, lv(pIndex)-[mask/buffer]->cas(pIndex)
// assumption behind this optimization is that queue is almost always empty or near empty
if (producerLimit <= pIndex) {
int result = offerSlowPath(mask, pIndex, producerLimit);
switch (result) {
case 0:
break;
case 1:
continue;
case 2:
return false;
case 3:
resize(mask, buffer, pIndex, e);
return true;
}
}
if (casProducerIndex(pIndex, pIndex + 2)) {
break;
}
}
// INDEX visible before ELEMENT, consistent with consumer expectation
final int offset = modifiedCalcElementOffset(pIndex, mask);
buffer.lazySet(offset, e);
return true;
}
/**
* We do not inline resize into this method because we do not resize on fill.
*/
private int offerSlowPath(long mask, long pIndex, long producerLimit) {
int result;
final long cIndex = lvConsumerIndex();
long bufferCapacity = getCurrentBufferCapacity(mask);
result = 0;// 0 - goto pIndex CAS
if (cIndex + bufferCapacity > pIndex) {
if (!casProducerLimit(producerLimit, cIndex + bufferCapacity)) {
result = 1;// retry from top
}
}
// full and cannot grow
else if (availableInQueue(pIndex, cIndex) <= 0) {
result = 2;// -> return false;
}
// grab index for resize -> set lower bit
else if (casProducerIndex(pIndex, pIndex + 1)) {
result = 3;// -> resize
}
else {
result = 1;// failed resize attempt, retry from top
}
return result;
}
/**
* @return available elements in queue * 2
*/
protected abstract long availableInQueue(long pIndex, final long cIndex);
/**
* This method assumes index is actually (index << 1) because lower bit is used for resize hence the >> 1
*/
private static int modifiedCalcElementOffset(long index, long mask) {
return (int) (index & mask) >> 1;
}
/**
* {@inheritDoc}
* <p>
* This implementation is correct for single consumer thread use only.
*/
@SuppressWarnings("unchecked")
@Override
public E poll() {
final AtomicReferenceArray<E> buffer = consumerBuffer;
final long index = consumerIndex.get();
final long mask = consumerMask;
final int offset = modifiedCalcElementOffset(index, mask);
Object e = buffer.get(offset);
if (e == null) {
if (index != lvProducerIndex()) {
// poll() == null iff queue is empty, null element is not strong enough indicator, so we must
// check the producer index. If the queue is indeed not empty we spin until element is
// visible.
do {
e = buffer.get(offset);
} while (e == null);
}
else {
return null;
}
}
if (e == JUMP) {
final AtomicReferenceArray<E> nextBuffer = getNextBuffer(buffer, mask);
return newBufferPoll(nextBuffer, index);
}
buffer.lazySet(offset, null);
soConsumerIndex(index + 2);
return (E) e;
}
/**
* {@inheritDoc}
* <p>
* This implementation is correct for single consumer thread use only.
*/
@SuppressWarnings("unchecked")
@Override
public E peek() {
final AtomicReferenceArray<E> buffer = consumerBuffer;
final long index = consumerIndex.get();
final long mask = consumerMask;
final int offset = modifiedCalcElementOffset(index, mask);
Object e = buffer.get(offset);
if (e == null && index != lvProducerIndex()) {
// peek() == null iff queue is empty, null element is not strong enough indicator, so we must
// check the producer index. If the queue is indeed not empty we spin until element is visible.
while ((e = buffer.get(offset)) == null)
;
}
if (e == JUMP) {
return newBufferPeek(getNextBuffer(buffer, mask), index);
}
return (E) e;
}
@SuppressWarnings("unchecked")
private AtomicReferenceArray<E> getNextBuffer(final AtomicReferenceArray<E> buffer, final long mask) {
final int nextArrayOffset = nextArrayOffset(mask);
final AtomicReferenceArray<E> nextBuffer = (AtomicReferenceArray<E>) buffer.get(nextArrayOffset);
buffer.lazySet(nextArrayOffset, null);
return nextBuffer;
}
private int nextArrayOffset(final long mask) {
return modifiedCalcElementOffset(mask + 2, Long.MAX_VALUE);
}
private E newBufferPoll(AtomicReferenceArray<E> nextBuffer, final long index) {
final int offsetInNew = newBufferAndOffset(nextBuffer, index);
final E n = nextBuffer.get(offsetInNew);
if (n == null) {
throw new IllegalStateException("new buffer must have at least one element");
}
nextBuffer.lazySet(offsetInNew, null);
soConsumerIndex(index + 2);
return n;
}
private E newBufferPeek(AtomicReferenceArray<E> nextBuffer, final long index) {
final int offsetInNew = newBufferAndOffset(nextBuffer, index);
final E n = nextBuffer.get(offsetInNew);
if (null == n) {
throw new IllegalStateException("new buffer must have at least one element");
}
return n;
}
private int newBufferAndOffset(AtomicReferenceArray<E> nextBuffer, final long index) {
consumerBuffer = nextBuffer;
consumerMask = (nextBuffer.length() - 2) << 1;
final int offsetInNew = modifiedCalcElementOffset(index, consumerMask);
return offsetInNew;
}
@Override
public final int size() {
// NOTE: because indices are on even numbers we cannot use the size util.
/*
* It is possible for a thread to be interrupted or reschedule between the read of the producer and
* consumer indices, therefore protection is required to ensure size is within valid range. In the
* event of concurrent polls/offers to this method the size is OVER estimated as we read consumer
* index BEFORE the producer index.
*/
long after = lvConsumerIndex();
long size;
while (true) {
final long before = after;
final long currentProducerIndex = lvProducerIndex();
after = lvConsumerIndex();
if (before == after) {
size = ((currentProducerIndex - after) >> 1);
break;
}
}
// Long overflow is impossible, so size is always positive. Integer overflow is possible for the unbounded
// indexed queues.
if (size > Integer.MAX_VALUE) {
return Integer.MAX_VALUE;
}
else {
return (int) size;
}
}
@Override
public final boolean isEmpty() {
// Order matters!
// Loading consumer before producer allows for producer increments after consumer index is read.
// This ensures this method is conservative in it's estimate. Note that as this is an MPMC there is
// nothing we can do to make this an exact method.
return (this.lvConsumerIndex() == this.lvProducerIndex());
}
private long lvProducerIndex() {
return producerIndex.get();
}
private long lvConsumerIndex() {
return consumerIndex.get();
}
private void soProducerIndex(long v) {
producerIndex.set(v);
}
private boolean casProducerIndex(long expect, long newValue) {
return producerIndex.compareAndSet(expect, newValue);
}
private void soConsumerIndex(long v) {
consumerIndex.set(v);
}
private long lvProducerLimit() {
return producerLimit.get();
}
private boolean casProducerLimit(long expect, long newValue) {
return producerLimit.compareAndSet(expect, newValue);
}
private void soProducerLimit(long v) {
producerLimit.set(v);
}
@Override
public long currentProducerIndex() {
return lvProducerIndex() / 2;
}
@Override
public long currentConsumerIndex() {
return lvConsumerIndex() / 2;
}
@Override
public abstract int capacity();
@Override
public boolean relaxedOffer(E e) {
return offer(e);
}
@SuppressWarnings("unchecked")
@Override
public E relaxedPoll() {
final AtomicReferenceArray<E> buffer = consumerBuffer;
final long index = consumerIndex.get();
final long mask = consumerMask;
final int offset = modifiedCalcElementOffset(index, mask);
Object e = buffer.get(offset);
if (e == null) {
return null;
}
if (e == JUMP) {
final AtomicReferenceArray<E> nextBuffer = getNextBuffer(buffer, mask);
return newBufferPoll(nextBuffer, index);
}
buffer.lazySet(offset, null);
soConsumerIndex(index + 2);
return (E) e;
}
@SuppressWarnings("unchecked")
@Override
public E relaxedPeek() {
final AtomicReferenceArray<E> buffer = consumerBuffer;
final long index = consumerIndex.get();
final long mask = consumerMask;
final int offset = modifiedCalcElementOffset(index, mask);
Object e = buffer.get(offset);
if (e == JUMP) {
return newBufferPeek(getNextBuffer(buffer, mask), index);
}
return (E) e;
}
@Override
public int fill(Supplier<E> s, int batchSize) {
long mask;
AtomicReferenceArray<E> buffer;
long pIndex;
int claimedSlots;
while (true) {
long producerLimit = lvProducerLimit();
pIndex = lvProducerIndex();
// lower bit is indicative of resize, if we see it we spin until it's cleared
if ((pIndex & 1) == 1) {
continue;
}
// pIndex is even (lower bit is 0) -> actual index is (pIndex >> 1)
// NOTE: mask/buffer may get changed by resizing -> only use for array access after successful CAS.
// Only by virtue ofloading them between the lvProcducerIndex and a successful casProducerIndex are they
// safe to use.
mask = this.producerMask;
buffer = this.producerBuffer;
// a successful CAS ties the ordering, lv(pIndex)->[mask/buffer]->cas(pIndex)
// we want 'limit' slots, but will settle for whatever is visible to 'producerLimit'
long batchIndex = Math.min(producerLimit, pIndex + 2 * batchSize);
if (pIndex == producerLimit || producerLimit < batchIndex) {
int result = offerSlowPath(mask, pIndex, producerLimit);
switch (result) {
case 1:
continue;
case 2:
return 0;
case 3:
resize(mask, buffer, pIndex, s.get());
return 1;
}
}
// claim limit slots at once
if (casProducerIndex(pIndex, batchIndex)) {
claimedSlots = (int) ((batchIndex - pIndex) / 2);
break;
}
}
int i = 0;
for (i = 0; i < claimedSlots; i++) {
final int offset = modifiedCalcElementOffset(pIndex + 2 * i, mask);
buffer.lazySet(offset, s.get());
}
return claimedSlots;
}
private void resize(long oldMask, AtomicReferenceArray<E> oldBuffer, long pIndex, final E e) {
int newBufferLength = getNextBufferSize(oldBuffer);
AtomicReferenceArray<E> newBuffer = new AtomicReferenceArray<E>(newBufferLength);
producerBuffer = newBuffer;
final int newMask = (newBufferLength - 2) << 1;
producerMask = newMask;
final int offsetInOld = modifiedCalcElementOffset(pIndex, oldMask);
final int offsetInNew = modifiedCalcElementOffset(pIndex, newMask);
newBuffer.lazySet(offsetInNew, e);
// todo fix
oldBuffer.set(nextArrayOffset(oldMask), (E) newBuffer);
// ASSERT code
final long cIndex = lvConsumerIndex();
final long availableInQueue = availableInQueue(pIndex, cIndex);
RangeUtil.checkPositive(availableInQueue, "availableInQueue");
// Invalidate racing CASs
// We never set the limit beyond the bounds of a buffer
soProducerLimit(pIndex + Math.min(newMask, availableInQueue));
// make resize visible to the other producers
soProducerIndex(pIndex + 2);
// INDEX visible before ELEMENT, consistent with consumer expectation
// make resize visible to consumer
oldBuffer.lazySet(offsetInOld, (E) JUMP);
}
/**
* @return next buffer size(inclusive of next array pointer)
*/
protected abstract int getNextBufferSize(AtomicReferenceArray<E> buffer);
/**
* @return current buffer capacity for elements (excluding next pointer and jump entry) * 2
*/
protected abstract long getCurrentBufferCapacity(long mask);
@Override
public int fill(Supplier<E> s) {
long result = 0;// result is a long because we want to have a safepoint check at regular intervals
final int capacity = capacity();
do {
final int filled = fill(s, RECOMMENDED_OFFER_BATCH);
if (filled == 0) {
return (int) result;
}
result += filled;
} while (result <= capacity);
return (int) result;
}
@Override
public void fill(Supplier<E> s,
WaitStrategy w,
ExitCondition exit) {
while (exit.keepRunning()) {
while (fill(s, RECOMMENDED_OFFER_BATCH) != 0 && exit.keepRunning()) {
continue;
}
int idleCounter = 0;
while (exit.keepRunning() && fill(s, RECOMMENDED_OFFER_BATCH) == 0) {
idleCounter = w.idle(idleCounter);
}
}
}
@Override
public void drain(Consumer<E> c, WaitStrategy w, ExitCondition exit) {
int idleCounter = 0;
while (exit.keepRunning()) {
E e = relaxedPoll();
if (e == null) {
idleCounter = w.idle(idleCounter);
continue;
}
idleCounter = 0;
c.accept(e);
}
}
@Override
public int drain(Consumer<E> c) {
return drain(c, capacity());
}
@Override
public int drain(final Consumer<E> c, final int limit) {
/**
* Impl note: there are potentially some small gains to be had by manually inlining relaxedPoll() and hoisting
* reused fields out to reduce redundant reads.
*/
int i = 0;
E m;
for (; i < limit && (m = relaxedPoll()) != null; i++) {
c.accept(m);
}
return i;
}
}