Save producerIndex load while polling a pooled chunk on mpmc xadd q

[franz1981]

The change on poll introduced by fc29784 has caused a performance regression due to an increased number of producerIndex loads.
This PR is using consumerBuffer::index to save some producerIndex loads, while preserving correctness of fast fail (ie return null) during a rotation if there are no other elements but the first one in the next chunk.

Some results by running

$ java -jar microbenchmarks.jar org.jctools.jmh.throughput.QueueThroughputBackoffNone -f 10 -p qType=MpmcUnboundedXaddArrayQueue,MpmcArrayQueue -p qCapacity=131072 -tg 5,5

master:

Benchmark                                    (qCapacity)                      (qType)   Mode  Cnt   Score   Error   Units
QueueThroughputBackoffNone.tpt                    131072  MpmcUnboundedXaddArrayQueue  thrpt  100  18.042 ± 0.290  ops/us
QueueThroughputBackoffNone.tpt:offer              131072  MpmcUnboundedXaddArrayQueue  thrpt  100  14.393 ± 0.279  ops/us
QueueThroughputBackoffNone.tpt:offersFailed       131072  MpmcUnboundedXaddArrayQueue  thrpt  100     ≈ 0          ops/us
QueueThroughputBackoffNone.tpt:offersMade         131072  MpmcUnboundedXaddArrayQueue  thrpt  100  14.410 ± 0.280  ops/us
QueueThroughputBackoffNone.tpt:poll               131072  MpmcUnboundedXaddArrayQueue  thrpt  100   3.649 ± 0.048  ops/us
QueueThroughputBackoffNone.tpt:pollsFailed        131072  MpmcUnboundedXaddArrayQueue  thrpt  100   0.001 ± 0.001  ops/us
QueueThroughputBackoffNone.tpt:pollsMade          131072  MpmcUnboundedXaddArrayQueue  thrpt  100   3.654 ± 0.048  ops/us
QueueThroughputBackoffNone.tpt                    131072               MpmcArrayQueue  thrpt  100   8.911 ± 0.270  ops/us
QueueThroughputBackoffNone.tpt:offer              131072               MpmcArrayQueue  thrpt  100   4.460 ± 0.141  ops/us
QueueThroughputBackoffNone.tpt:offersFailed       131072               MpmcArrayQueue  thrpt  100   0.029 ± 0.031  ops/us
QueueThroughputBackoffNone.tpt:offersMade         131072               MpmcArrayQueue  thrpt  100   4.435 ± 0.145  ops/us
QueueThroughputBackoffNone.tpt:poll               131072               MpmcArrayQueue  thrpt  100   4.451 ± 0.134  ops/us
QueueThroughputBackoffNone.tpt:pollsFailed        131072               MpmcArrayQueue  thrpt  100   0.045 ± 0.020  ops/us
QueueThroughputBackoffNone.tpt:pollsMade          131072               MpmcArrayQueue  thrpt  100   4.410 ± 0.144  ops/us

this PR:

Benchmark                                    (qCapacity)                      (qType)   Mode  Cnt   Score   Error   Units
QueueThroughputBackoffNone.tpt                    131072  MpmcUnboundedXaddArrayQueue  thrpt  100  24.164 ± 0.660  ops/us
QueueThroughputBackoffNone.tpt:offer              131072  MpmcUnboundedXaddArrayQueue  thrpt  100  19.203 ± 0.559  ops/us
QueueThroughputBackoffNone.tpt:offersFailed       131072  MpmcUnboundedXaddArrayQueue  thrpt  100     ≈ 0          ops/us
QueueThroughputBackoffNone.tpt:offersMade         131072  MpmcUnboundedXaddArrayQueue  thrpt  100  19.223 ± 0.559  ops/us
QueueThroughputBackoffNone.tpt:poll               131072  MpmcUnboundedXaddArrayQueue  thrpt  100   4.961 ± 0.124  ops/us
QueueThroughputBackoffNone.tpt:pollsFailed        131072  MpmcUnboundedXaddArrayQueue  thrpt  100   0.001 ± 0.001  ops/us
QueueThroughputBackoffNone.tpt:pollsMade          131072  MpmcUnboundedXaddArrayQueue  thrpt  100   4.968 ± 0.124  ops/us
QueueThroughputBackoffNone.tpt                    131072               MpmcArrayQueue  thrpt  100   9.535 ± 0.251  ops/us
QueueThroughputBackoffNone.tpt:offer              131072               MpmcArrayQueue  thrpt  100   4.764 ± 0.130  ops/us
QueueThroughputBackoffNone.tpt:offersFailed       131072               MpmcArrayQueue  thrpt  100   0.001 ± 0.002  ops/us
QueueThroughputBackoffNone.tpt:offersMade         131072               MpmcArrayQueue  thrpt  100   4.766 ± 0.131  ops/us
QueueThroughputBackoffNone.tpt:poll               131072               MpmcArrayQueue  thrpt  100   4.771 ± 0.121  ops/us
QueueThroughputBackoffNone.tpt:pollsFailed        131072               MpmcArrayQueue  thrpt  100   0.018 ± 0.008  ops/us
QueueThroughputBackoffNone.tpt:pollsMade          131072               MpmcArrayQueue  thrpt  100   4.756 ± 0.128  ops/us

These results show that saving unecessary producerIndex loads on rotation can make a lot of difference, hence #265 could be very beneficial to be implemented for a future release.

[coveralls]

Pull Request Test Coverage Report for Build 580

• 15 of 18 (83.33%) changed or added relevant lines in 1 file are covered.
• 19 unchanged lines in 3 files lost coverage.
• Overall coverage increased (+0.02%) to 85.421%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
jctools-core/src/main/java/org/jctools/queues/MpmcUnboundedXaddArrayQueue.java	15	18	83.33%

Files with Coverage Reduction	New Missed Lines	%
jctools-core/src/main/java/org/jctools/maps/NonBlockingHashMap.java	1	79.65%
jctools-core/src/main/java/org/jctools/maps/NonBlockingIdentityHashMap.java	5	76.31%
jctools-core/src/main/java/org/jctools/maps/NonBlockingHashMapLong.java	13	78.89%

Totals
Change from base Build 577:	0.02%
Covered Lines:	4781
Relevant Lines:	5597

---

💛 - Coveralls

[lgtm-com]

This pull request introduces 1 alert when merging 687a3fb into 671c0a8 - view on LGTM.com

new alerts:

• 1 for Useless comparison test

[lgtm-com]

This pull request introduces 1 alert when merging 75793bb into 671c0a8 - view on LGTM.com

new alerts:

• 1 for Useless comparison test

[franz1981]

@nitsanw this is ready to be reviewed/merged
Sorry for this sudden change, but I haven't had available boxes until yesterday to perform some reliable regression tests

[franz1981]

@nitsanw the travis build is pending, but I see on the CI that d30b4d1 (validated) has the same content of 3322e9f (this PR), any idea what's happening?

[franz1981]

@nitsanw Wait a sec: I see this

INFO] Running org.jctools.queues.MpqSanityTestMpscUnboundedXadd
Exception in thread "Thread-274" java.lang.AssertionError
	at org.jctools.queues.MpscUnboundedXaddArrayQueue.relaxedPeek(MpscUnboundedXaddArrayQueue.java:590)
	at org.jctools.queues.MpqSanityTest$6.run(MpqSanityTest.java:488)
	at java.lang.Thread.run(Thread.java:748)

This hasn't happened before, but seems present from the beginning, I've alread sent #270 to fix it

[franz1981]

@nitsanw Despite what CI says (see comment above), this is ready to be reviewed/merged