[11] RFR(S) 8193126: Incorrect setting of MetaspaceSize and NewSizeThreadIncrease when using JVMCI compiler

[Vladimir Kozlov]

Only ReservedCodeCacheSize is changed after that for Tiered compilation which increases 
ReservedCodeCacheSize size x5:

http://hg.openjdk.java.net/jdk/jdk/file/ec3221b8a109/src/hotspot/share/compiler/compilerDefinitions.cpp#l376
http://hg.openjdk.java.net/jdk/jdk/file/ec3221b8a109/src/hotspot/share/compiler/compilerDefinitions.cpp#l198

Which is fine I think because I want to adjust JVMCI default setting and not ergonomic.

Thanks,
Vladimir

On 7/6/18 12:56 AM, Doug Simon wrote:
> Are these settings applied after all other ergonomics that may also change the values?
> 
> -Doug
> 
>> On 6 Jul 2018, at 01:17, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
>>
>> http://cr.openjdk.java.net/~kvn/8193126/webrev.00/
>> https://bugs.openjdk.java.net/browse/JDK-8193126
>>
>> JVMCI reset some flags values to improve Graal code generation. But some settings were copied from C1 setting which cause change of Server VM behavior with Graal JIT vs when C2 JIT is used. For example, the default MetaspaceSize value with C2 is about 21Mb but it was reset to 12Mb for JVMCI which causes addition metaspace collections and Java heap collections too (for class unloading).
>>
>> It also cause some unexpected failures in testing. Like JDK-8203174.
>>
>> I propose to select max values for some sizes.
>>
>> Tested tier1, tier2, graal-tier3, graal-tier4
>>
>> -- 
>> Thanks,
>> Vladimir
>