RFR: 8163589: Add back intrinsic methods for event based tracing

[Markus Gronlund]

Hi Robbin,

I think this looks good, thanks for doing this.

Thanks
Markus

-----Original Message-----
From: Robbin Ehn 
Sent: den 11 augusti 2016 09:42
To: Aleksey Shipilev; hotspot-compiler-dev at openjdk.java.net
Subject: Re: RFR: 8163589: Add back intrinsic methods for event based tracing

Hi Aleksey,

Thanks for looking at this!

On 08/10/2016 10:37 PM, Aleksey Shipilev wrote:
> On 08/10/2016 05:05 PM, Robbin Ehn wrote:
>> Hi all,
>>
>> In 8 we had this intrinsic, I'm adding it back Used for event based 
>> tracing.
>>
>> Webrev: http://cr.openjdk.java.net/~rehn/8163589/hotspot.01/webrev/
>
> Looks okay.
>
> *) c1_LIRGenerator:
>
> 3097   __ move(new LIR_Address(arg.result(),
> java_lang_Class::klass_offset_in_bytes(), klass_pointer_type), klass, 
> info);
>
> Should klass_pointer_type be T_ADDRESS? It is an injected field of 
> intptr_t type, so current code seems fine, but T_ADDRESS matches the 
> existing uses in C1 better.

Thanks, fixed.

>
> *) Are these different? Do we want to use one of these consistently?
>
> C1:
> 3099   ByteSize offset = TRACE_KLASS_TRACE_ID_OFFSET;
>
> C2:
> 3169   ByteSize offset = InstanceKlass::trace_id_offset();
>

Yes, thanks, fixed.

>
> *) Is the new intrinsic different from other catered by 
> LibraryCallKit::inline_native_Class_query?
>

We could cater this also there, but this is not as generic as those other intrinsic.
Personally I would keep this separate. (we don't need to clobber inline_native_Class_query with ifdef, etc...) So if there is no strong opinions, I'll leave that as is.

Incremental here:
http://cr.openjdk.java.net/~rehn/8163589/hotspot.01-02/webrev/
Full here:
http://cr.openjdk.java.net/~rehn/8163589/hotspot.02/webrev/

Thanks!

(tested with JMH)

/Robbin

>
> Thanks,
> -Aleksey
>
>
>