RFR: 8219713: Reduce work in DefaultMethods::generate_default_methods

[Claes Redestad]

Hi Karen,

On 2019-03-05 23:48, Karen Kinnear wrote:
> Claes,
> 
> Lois and I met today to discuss these changes.
> First - we are very grateful you dove in, identified performance issues and are proposing
> solutions.
> 
> Second - could you please check in three out of the four kinds of changes:
> 1) remove extra walking of Object
> 2) remove unused code
> 3) change the order of the conditionals so we reduce time spent in Lookup.

Ok, basically the entire patch, plus this to revert the changes on line
606:

diff -r 9ee9ac93e39b src/hotspot/share/classfile/defaultMethods.cpp
--- a/src/hotspot/share/classfile/defaultMethods.cpp	Wed Mar 06 21:28:03 
2019 +0100
+++ b/src/hotspot/share/classfile/defaultMethods.cpp	Wed Mar 06 21:28:33 
2019 +0100
@@ -603,7 +603,7 @@
    while (super != NULL) {
      for (int i = 0; i < super->methods()->length(); ++i) {
        Method* m = super->methods()->at(i);
-      if (m->is_overpass() || (m->is_static() && 
!SystemDictionary::is_nonpublic_Object_method(m))) {
+      if (m->is_overpass() || m->is_static()) {
          // m is a method that would have been a miranda if not for the
          // default method processing that occurred on behalf of our 
superclass,
          // so it's a method we want to re-examine in this new context. 
  That is,

I'll run some sanity testing and push.

> 
> And - could you please file a follow-up RFE to investigate improvements in the
> default method processing for static methods, with notes about the performance benefits
> of skipping processing for RegisterNatives and <clinit>?

Sure, I'll do that.

Thanks!

/Claes