RFR: 8233497: Optimize default method generation by data structure reuse

[Lois Foltan]

On 11/8/2019 6:57 AM, Claes Redestad wrote:
> Hi,
>
> when loading classes with complex hierarchies and many default methods,
> we can end up spending significant time in
> DefaultMethods::generate_default_methods
>
> This optimization reduces work done and memory requirements by reusing
> allocated data structures. For example by maintaining free lists of
> allocated Node objects.
>
> Bug:    https://bugs.openjdk.java.net/browse/JDK-8233497
> Webrev: http://cr.openjdk.java.net/~redestad/8233497/open.00/
>
> Testing: Tier1-3, will make sure tier4-7 pass before push
>
> Performance notes: On one of our more complex startup tests we see a 3%
> improvement on the execution time total. Much less on simpler
> applications.
>
> I've not done a formal complexity analysis, but I think the memory
> complexity is now down from O(N*M) to O(N+M) where N is the number of
> classes and interfaces in the hierarchy and M the number of methods of
> interest in that hierarchy. Algorithmic complexity is probably O(N*M)
> still, but with much better constants.
>
> Special thanks to Lois for patience and persistence over several rounds
> of pre-review!
>
> Thanks!
>
> /Claes

Hi Claes,

I have reviewed your final webrev and I think this looks good.  It 
certainly is an area that can benefit from performance improvements so 
thank you for putting the time in on this.  A couple of final comments:

- line #79: Can you update the comment, it indicates that new_node_data 
takes an InstanceKlass* parameter which has been removed

- line #716: I would like an assert as part of the if statement in the 
situation where _free_scopes is not empty and the StateRestoreScope node 
is being popped instead of newly allocated. The assert would then check 
if that StateRestorerScope's _marks GrowableArray is empty, which it 
should be.  That would make me more comfortable with the idea that a 
previous StateRestoreScope's _marks array's data is not getting mixed 
with new data being established for a new Node in the hierarchy.

Thanks,
Lois