RFR 8038418 New tests development for type profiling and speculation

Pavel Punegov pavel.punegov at oracle.com
Tue Apr 8 22:05:32 UTC 2014


Hi Roland,

Yes, it was a mistake. They should be -XX:TypeProfileLevel=222 -XX:
+UseTypeSpeculation

New version has added a WhiteBox.clearMethodState usage to clear collected 
data about methods invoked in each scenario.
We use it and ClassLoaders to make every test case to have a clean profile 
data.
Also new webrev has small fixes in javadocs and results printing.

Webrev: http://cr.openjdk.java.net/~iignatyev/ppunegov/8038418/webrev.01/

-- 
Thanks,
Pavel Punegov


On Tue, 08 of Apr 2014 20:19:35 Igor Ignatyev wrote:
> Hi Roland,
> 
> it was a mistake.
> 
> there's a new webrev:
> http://cr.openjdk.java.net/~iignatyev/ppunegov/8038418/webrev.01/
> 
> No we use new class loader for each execution and clear state of all
> 
> methods in scenario through WhiteBox::clearMethodState():
>>   MethodData* mdo = mh->method_data();
>>   
>>   if (mdo != NULL) {
>>   
>>     mdo->init();
>>     ResourceMark rm;
>>     int arg_count = mdo->method()->size_of_parameters();
>>     for (int i = 0; i < arg_count; i++) {
>>     
>>       mdo->set_arg_modified(i, 0);
>>     
>>     }
>>   
>>   }
> 
> I think it should also clear type profiling data. Am I right?
> 
> Pavel should come w/ more comments about what was done in new version.
> 
> Igor
> 
> On 04/08/2014 07:37 PM, Roland Westrelin wrote:
>> Hi Pavel,
>> 
>>> webrev: http://cr.openjdk.java.net/~iignatyev/ppunegov/8038418/webrev.00/
>> 
>> + *                   -XX:TypeProfileLevel=0 -XX:-UseTypeSpeculation
>> 
>> Are they only run with profiling off and type speculation off?
>> 
>> Roland.


More information about the hotspot-compiler-dev mailing list