RFR: More efficient next bitmap resetting for traversal

Zhengyu Gu zgu at redhat.com
Mon Jul 16 15:44:53 UTC 2018


Hi,

Looks wrong ...

1) You passed in traversalSet iter, then re-creates traversalSet itr for 
each worker, so you reset the regions n_workers times.

2) Should grab heapLock in reset_next_mark_bitmap_traversal()


Thanks,

-Zhengyu


On 07/16/2018 11:33 AM, Roman Kennke wrote:
> The resetting of next bitmap for traversal is a bit inefficient:
> - It's iterating *all* regions, but then only picks traversal-set
> regions. We can iterate traversal-set to begin with instead.
> - It grabs the heap-lock for each region. Which basically serializes the
> whole process for no good reason.
> 
> http://cr.openjdk.java.net/~rkennke/better-traversal-reset-bitmaps/webrev.00/
> 
> Testing: tier3_gc_shenandoah
> 
> Ok?
> 
> Roman
> 


More information about the shenandoah-dev mailing list