Review Request (S): 8002364 - compile_id is not unique when CICountOSR is set

[Vladimir Kozlov]

On Nov 8, 2012, at 8:54 AM, Vladimir Kozlov wrote:

> 
> On Nov 8, 2012, at 7:15 AM, Nils Eliasson wrote:
> 
>> When developing the compiler events for JFR i noticed that the compile IDs where not always unique. And that guarantee is needed. The CICountOSR flag (default in develop) makes the compile_id be used as an osr_id instead. This can cause confusion when used with JFR. And it is never wrong to have a dependable unique id.
> 
> It is not true. Vladimir I. changed the flag default value to false in recent changes for 7177003.
> Also in our developing environment we will not use JFR but we want a functionality which help us debug problems. So I would like to keep the flag and have a separate number in output (tty and LogCompilation log) for osr_id if the flag is on. So I agree with your general changes but you need to add more code.

After thinking about this more, maybe we should always print osr_id as an additional number then we don't need CICountOSR flag. In this case you only need to add code to print this number in PrintCompilation and LogCompilation.

Vladimir

> 
> To have separate osr_id is useful because usually it is more consistent between different runs than compile_id (OSR compilations are not frequent).
> 
> Thanks,
> Vladimir
> 
>> 
>> I suggest this change that makes us always have unique compile ID while keeping the osr_id separately.
>> 
>> http://cr.openjdk.java.net/~neliasso/8002364/webrev.03/ <http://cr.openjdk.java.net/%7Eneliasso/8002364/webrev.03/>
>> 
>> This has the consequences:
>> Flag CICountOSR is removed - the functionality is there always anyway.
>> Flag CIBreakAtOSR,  CIStartOSR and CIStopOSR is still in use but the osr ID isn't printed in the compilation log, so it may be a little harder to use.
>> 
>> I considered removing the CIBreakAtOSR flag and then there was no need for saving the osr_id after the CIStart/StopOSR check. That would make the code a bit nicer (and save a field).
>> I have kept the osr_id increment behavior making the code slightly more complex.
>> 
>> //Nils
>