[Issue 5650] A RedBlackTree performance problem

[d-bugmail at puremagic.com]

http://d.puremagic.com/issues/show_bug.cgi?id=5650



--- Comment #5 from David Simcha <dsimcha at yahoo.com> 2011-02-24 14:40:27 PST ---
(In reply to comment #2)
> This looks like a GC problem, I wonder if David's recent changes help...
> 
> I'm going to leave this bug open.  We'll see how the thing works on the next
> version with David's GC improvements.
> 

I'd doubt that the changes I've posted to Bugzilla so far would help this
benchmark, because those changes focus entirely on large (>2KB) allocations. 
However, in rebuilding my mental model of the GC, I've come up with several 
smaller (i.e. a few percent each) ideas for optimizing the GC.  These are
currently half-implemented.  I'm going to fork druntime on GitHub and commit
them all there, and show the progressive performance improvements on a few
different benchmarks with each one on the GitHub wiki page for my fork.

This, I guess, will be added to my list of GC benchmarks for small allocations,
though I'm not sure if it will tell us anything that Bearophile's other recent
tree-building GC benchmark didn't.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------