Request for reviews (S): 6658428: C2 doesn't inline java method if corresponding intrinsic failed to inline

Nils Eliasson nils.eliasson at oracle.com
Wed Feb 22 05:27:51 PST 2012


Thanks Vladimir and Tom for the feedback,

I agree with removing the out-of-line part.

Removing the UsePopCountInstruction-guard in library_call also makes it 
nicer and cleaner, but noting that we in a few rare cases create call 
generators for intrinsics that always return false.

http://cr.openjdk.java.net/~neliasso/6658428.01/webrev.01/ 
<http://cr.openjdk.java.net/%7Eneliasso/6658428.01/webrev.01/>

Skiping reviewboard for now, doesn't seem to work well with mq and cygwin.

//N

Tom Rodriguez skrev 2012-02-21 20:05:
> On Feb 21, 2012, at 2:26 AM, Nils Eliasson wrote:
>
>> http://cr.openjdk.java.net/~neliasso/6658428.01/webrev.00/
>>
>> 6658428: C2 doesn't inline java method if corresponding intrinsic failed to inline.
>>
>> Allow intrinsic inline to fallback to plain java inline case. Changing some intrinsics to not be created when they cant be used and adding some print inline changes.
> I don't think you need to try inline and out of line separately.  I think the old code forced out of line because that was the only way to suppress the use of the intrinsic.  So remove the try_inline = false line and just add the extra false argument to the existing call.  It should use inline if it's appropriate or out of line if not.
>
> In library_call.cpp I think it would be better to stop using !UsePopCountInstruction to guard those intrinsics and use the appropriate Matcher::has_match_rule for each one.  That should take into account the platform dependent meaning of UsePopCountIntstruction.
>
> tom
>
>> And I will need someone to submit it.
>>
>> Thanks,
>> Nils E.
>>


-- 
Oracle <http://www.oracle.com>
Nils Eliasson | Senior Member of Technical Staff
Oracle Java Platform Group, JVM Engineering
ORACLE Sweden

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20120222/5bfb71a3/attachment.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: oracle_sig_logo.gif
Type: image/gif
Size: 658 bytes
Desc: not available
Url : http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20120222/5bfb71a3/oracle_sig_logo.gif 


More information about the hotspot-compiler-dev mailing list