RFR: 8233159: Method::result_type should use calculated value in constMethod

coleen.phillimore at oracle.com coleen.phillimore at oracle.com
Wed Oct 30 13:09:58 UTC 2019


This looks good.  There are a couple other places where we iterate the 
signature with ResultTypeFinder and don't need to.  I thought you were 
going to remove these also?
thanks,
Coleen

On 10/29/19 5:39 PM, Claes Redestad wrote:
> Hi,
>
> a simple, independent optimization extracted from the changes proposed
> in JDK-8230199[1]:
>
> Webrev: http://cr.openjdk.java.net/~redestad/8233159/open.00/
> Bug:    https://bugs.openjdk.java.net/browse/JDK-8233159
>
> Method::result_type is showing up in various places in profiles, and
> while non of those places are very hot, the optimization is trivial and
> effecient (turns calls that retire 2-400 instructions into an inlined
> memory read).
>
> The patch also removes a redundant iterate() call in the related
> Bytecode_member_ref::result_type() method (rts.type() will trigger
> a scan "lazily").
>
> Testing: tier1-3
>
> Thanks!
>
> /Claes
>
> [1] https://bugs.openjdk.java.net/browse/JDK-8230199



More information about the hotspot-runtime-dev mailing list