RFR(S): 8240676: Meet not symmetric failure when running lucene on jdk8
Roland Westrelin
rwestrel at redhat.com
Wed Mar 18 12:58:48 UTC 2020
> I've submitted testing and I'm seeing timeouts on all platforms with
> compiler/types/TestMeetIncompatibleInterfaceArrays.java
That one is caused by redundant computations introduced by the
patch. For instance, in the case of an array, computing the meet causes
the computation of the meet of the elements which in the verification
code causes the computation of the meet of the dual of the
elements. Then verification code computes the meet of the dual of the
arrays which causes the meet of the dual of the elements to be computed
which causes the meet of the dual of the dual of the elements that is
the meet of the elements which was already computed.
So the patch now tries to remove redundant computations such as for
instance on the element of an array. This is done by passing a flag
which makes the change much bigger unfortunately.
Execution time on the test above is still affected by the patch: from
~1s without the patch to ~2s with the improved patch. Compilation time
for CTW on java.base is unaffected AFAICT.
http://cr.openjdk.java.net/~roland/8240676/webrev.01/
Roland.
More information about the hotspot-compiler-dev
mailing list