Loading...

XML

Word

Printable

Issue	Fix Version	Assignee	Priority	Status	Resolution	Resolved In Build
JDK-8135635	emb-9	Aleksey Shipilev	P4	Resolved	Fixed	team
JDK-8149241	8u101	Aleksey Shipilev	P4	Resolved	Fixed	b01
JDK-8145831	8u92	Aleksey Shipilev	P4	Resolved	Fixed	b02
JDK-8155337	emb-8u101	Aleksey Shipilev	P4	Resolved	Fixed	b01

This is observed in VarHandles work for ~~JDK-8130923~~.

The reason is that C1 canonicalizer seems to assume a particular constant shape (InstanceConstant), while the Class instances coming through @Stable/static-final seem to be ObjectConstants. Accepting ObjectConstants as well fixes it:
  http://cr.openjdk.java.net/~shade/8131782/webrev.01/

Simple benchmark articulates it clearly:
  http://cr.openjdk.java.net/~shade/8131782/ClassCastConstants.java

(Runnable JAR: http://cr.openjdk.java.net/~shade/8131782/benchmarks.jar)

Baseline result calls into Runtime::isInstance in "static final" case, experiencing the 2x hit:
  http://cr.openjdk.java.net/~shade/8131782/baseline.perfasm

Patched VM behaves perfectly well in both cases:
  http://cr.openjdk.java.net/~shade/8131782/patched.perfasm

backported by

JDK-8135635 C1 Class.cast optimization breaks when Class is loaded from static final

JDK-8145831 C1 Class.cast optimization breaks when Class is loaded from static final

JDK-8149241 C1 Class.cast optimization breaks when Class is loaded from static final

JDK-8155337 C1 Class.cast optimization breaks when Class is loaded from static final

blocks

JDK-8130923 C1 trouble in VarHandle.getObject tests

relates to

JDK-8130923 C1 trouble in VarHandle.getObject tests

(1 relates to)