RFR (S) 8247296: Optimize JVM_GetDeclaringClass

[David Holmes]

Hi Yumin,

On 24/07/2020 3:54 pm, Yumin Qi wrote:
> Hi, David
> 
>    Looks good to me. 

Thanks for taking a look at this.

> I have done a quick search and found (may not cover all):

I must stress I am sponsoring Christoph's change and only extended it 
within the current file. :) I'm sure there are many, many more 
opportunities for similar optimisations. Though you have to be careful 
to ensure you don't expose an unhandled oop.

> 1) 
> https://github.com/openjdk/jdk/blob/master/src/hotspot/share/compiler/compileBroker.cpp#L834 
> 
>     Where thread_handle resolved multiple times.

True that is unnecessary, but this is a compiler code and I'd need to 
extend the review so ... I'll pass on this one.

> 2) 
> https://github.com/openjdk/jdk/blob/master/src/hotspot/share/prims/jni.cpp#L1174 
> 
>      Where clazz resolved twice.

Fixed this as it is the same pattern in a core runtime file. Webrev 
updated in place.

Thanks,
David



>      Do you want to include those two files in your list?
> 
> Thanks
> 
> Yumin
> 
> On 7/23/20 8:49 PM, David Holmes wrote:
>> Bug: https://bugs.openjdk.java.net/browse/JDK-8247296
>> webrev: http://cr.openjdk.java.net/~dholmes/8247296/webrev/
>>
>> Please review this simple optimization contributed by Christoph Dreis 
>> in its initial form and then expanded by me to cover other cases in 
>> jvm.cpp.
>>
>> http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/2020-June/040025.html 
>>
>>
>> There is a common pattern of code of the form:
>>
>> if 
>> (java_lang_Class::is_primitive(JNIHandles::resolve_non_null(ofClass)) ||
>>     ! 
>> java_lang_Class::as_Klass(JNIHandles::resolve_non_null(ofClass))->is_instance_klass()) 
>> {
>>
>> which resolves cls twice. There are also duplicate calls to as_Klass 
>> that can be removed in a couple of cases.
>>
>> Testing: tiers 1 - 3
>>
>> Thanks,
>> David