[9] RFR(XS): 8172010: [AOT] RecompilationTest.java fails with "expected compilation level after compilation to be no less than 1"

Igor Veresov igor.veresov at oracle.com
Wed Jan 4 07:54:27 UTC 2017


Looks good.

igor

> On Jan 3, 2017, at 11:52 PM, Tobias Hartmann <tobias.hartmann at oracle.com> wrote:
> 
> Hi,
> 
> please review the following patch:
> https://bugs.openjdk.java.net/browse/JDK-8172010
> http://cr.openjdk.java.net/~thartmann/8172010/webrev.00/
> 
> The test fails with tiered AOT because an AOT compiled method is not recompiled after reaching the invocation threshold. This behavior is caused by probabilistic profiling of using the rdtsc (read timestamp counter) instruction:
> 
> 30e6: rdtsc  
> 30e8: test $0xff, %eax
> 30ee: je   3118 <compiler.whitebox.SimpleTestCaseHelper.<init>(I)V+0x78>
> 
> If we are very unlucky, we don't branch while executing the method in the CompilerWhiteBoxTest::compile() loop and therefore never notify the tiered compilation system to recompile the method. Turning probabilistic profiling off (or setting -Dgraal.TierAInvokeProfileProbabilityLog=0), fixes this testbug.
> 
> Tested on failing platform and with RBT.
> 
> Thanks,
> Tobias



More information about the hotspot-compiler-dev mailing list