From: Eric Wong <normalperson@...>
Date: 2014-03-14T06:20:44+00:00
Subject: [ruby-core:61488] Re: [REJECT?] xmalloc/xfree: reduce atomic ops w/ thread-locals

Eric Wong <normalperson@yhbt.net> wrote:
> How about only using thread local and remove the process-wide globals?

  git://80x24.org/ruby.git gc-tlscount

http://bogomips.org/ruby.git/patch?id=ee41b3938535a

Almost no change for bm_so_count_words, but (normal-sized) array and
bighash creation got speedups :o   I prefer this one, but this may
allow too much malloc for some MT users(*).


(*) I plan to cut overall memory use with things like Feature #9614.
    Even if matz requires ordering, we'll still save malloc using
    the linked-list from Feature #9632.