RFR (S): 8029888: PPC64: (part 219): smaller changes needed to build c2

Vladimir Kozlov vladimir.kozlov at oracle.com
Tue Dec 10 12:36:40 PST 2013

On 12/10/13 11:47 AM, Lindenmaier, Goetz wrote:
> Hi Vladimir,
> I just came up with a better idea for this, it even does not need
> shared changes.
> I figured loadConNKlass is a real constant, i.e., it's not moved or the
> like, right? So I can just check the concrete value whether it's
> 31 bit only:
> Assembler::is_uimm((jlong)Klass::encode_klass((Klass *)$src$$constant), 31)

Sounds good. You can also define special immN31Klass operand with predicate if you need it in several mach instructions.


> I'm just running a test and will remove the respective coding from the
> change.
> About the public: I added that before and now remove it again,
> as with the fix currently in the webrev it's not needed.
> Best regards,
>    Goetz.
> -----Original Message-----
> From: Vladimir Kozlov [mailto:vladimir.kozlov at oracle.com]
> Sent: Tuesday, December 10, 2013 5:53 PM
> To: Lindenmaier, Goetz; 'hotspot-dev at openjdk.java.net'; 'ppc-aix-port-dev at openjdk.java.net'
> Subject: Re: RFR (S): 8029888: PPC64: (part 219): smaller changes needed to build c2
> Goetz,
> Ther is initialization of compressed class structure:
> NarrowPtrStruct is useed for both compressed class ptrs and oops. Can you use separate Universe::_narrow_klass_max_val
> static field instead of including it into NarrowPtrStruct? And you missed initialization.
> Do you need changes in metaspace.hpp for some ppc64 code? I don't see "public" usage in current code.
> Thanks,
> Vladimir
> On 12/10/13 6:08 AM, Lindenmaier, Goetz wrote:
>> Hi,
>> Recent changes make two other smaller shared adaptions necessary
>> for the ppc port:
>> http://cr.openjdk.java.net/~goetz/webrevs/8029888-0-basic2/
>> First, introduce replacement variable $$CondRegister in adl.
>> Further, the latest changes to compressed klasses broke an
>> optimization on ppc.  This change contains a better way
>> to check for the max size of a compressed klass.  I need not mask
>> sign extension bits if compressed klasses are at most 31 bits.
>> Please review and test this change.
>> Best regards,
>>     Goetz.

More information about the hotspot-dev mailing list