RFR(T)  : 8218178 vmTestbase/vm/mlvm/mixed/stress/regression/b6969574/INDIFY_Test.java fails with -Xcomp
igor.ignatyev at oracle.com
Fri Feb 1 21:34:11 UTC 2019
Thanks Dean, pushed.
> On Feb 1, 2019, at 1:28 PM, dean.long at oracle.com wrote:
> Your change looks OK for 12, allowing us to continue running with -Xcomp.
> On 1/31/19 6:20 PM, Igor Ignatyev wrote:
>> Hi Dean,
>> we observe the exact opposite: w/ Xcomp MH::invokeExact becomes faster than Method::invoke and this is the reason the test failed.
>> -- Igor
>>> On Jan 31, 2019, at 5:54 PM, dean.long at oracle.com wrote:
>>> I don't think it makes sense to run this test with -Xcomp, which forces an early compile that might not give maximum performance. How about if we just skip this test with -Xcomp is used?
>>> On 1/31/19 5:33 PM, Igor Ignatyev wrote:
>>>>> 1 line changed: 0 ins; 0 del; 1 mod;
>>>> Hi all,
>>>> could you please review this one-liner fix for b6969574 test?
>>>> the test was unproblem-listed by 8067250[1,2], and started to fail in runs w/ Xcomp. 8067250 fixed only the problem described in the bug (unstable results from MH.invokeWithArguments), the test however appears to have another problem which manifests only in certain configurations. the test expects Method.invoke() <= MH.invokeExact(), and in some cases, most notably -Xcomp, we might get MH.invokeExact < Method.invoke. the test is a regression test for 6969574, where it was reported that indy is slower than reflection, therefore comparing MH::invokeExact w/ Method::invoke should be irrelevant for the test and using assumptions about their relation is wrong. the fix replaces comparison of Method.invoke vs MH.invokeExact by comparison of Method.invoke vs indy.
>>>> webrev: http://cr.openjdk.java.net/~iignatyev//8218178/webrev.00/index.html
>>>> JBS: https://bugs.openjdk.java.net/browse/JDK-8218178
>>>> - the test on product and fastdebug w/o any extra flags
>>>> - the test w/ one configuration it was seen failed (-Xcomp -XX:+CreateCoredumpOnCrash -ea -esa -XX:CompileThreshold=100 -XX:+UnlockExperimentalVMOptions -server -XX:-TieredCompilation)
>>>> - :vmTestbase_vm_mlvm in all configurations used in tier6, the first tier the test fails
>>>>  https://bugs.openjdk.java.net/browse/JDK-8067250
>>>>  http://hg.openjdk.java.net/jdk/jdk12/rev/207d3bb450e5
>>>>  https://bugs.openjdk.java.net/browse/JDK-6969574
>>>> -- Igor
More information about the hotspot-compiler-dev