RFR: 8265349: vmTestbase/../stress/compiler/deoptimize/Test.java fails with OOME due to CodeCache exhaustion. [v2]
iignatyev at openjdk.java.net
Wed Apr 28 20:52:56 UTC 2021
On Wed, 28 Apr 2021 19:07:10 GMT, Evgeny Nikitin <enikitin at openjdk.org> wrote:
>> The bug: https://bugs.openjdk.java.net/browse/JDK-8265349
>> A repetition of the JDK-8058176 (mlvm tests cause code cache exhaustion), this time with -Xcomp. My measurements show up max code cache consumption of 400-500 kb per test thread and tree (depends on random tree and other factors, of course, but still). For the whole test which usually doesn't exceed 10 threads, I've got max. value of 6.1M for one sequence/tree build (between checks for remaining free space).
>> So I suggest to raise the allowances to 10M, just to be safe. Compared to the 103 Mb of required code cache space that test requests via the 'run' parameter, it doesn't look that much.
>> Thanks in advance,
>> // Evgeny Nikitin.
> Evgeny Nikitin has updated the pull request incrementally with one additional commit since the last revision:
> Make code cache allowance Xcomp-dependent
I don't like that we kinda guessing here, but on the other hand, I don't think we have much of the choice.
could you please add the comment which roughly explains why it's 10kk and 5kk? so the next person who reads/modifies this piece of code will have an idea of how/why we get the current numbers.
Changes requested by iignatyev (Reviewer).
More information about the hotspot-compiler-dev