Ingo Weinhold <ingo_weinhold@xxxxxx> wrote: > On 2009-10-19 at 08:16:54 [+0800], axeld@xxxxxxxxxxxxxxxx wrote: > > Author: axeld > > Date: 2009-10-19 02:16:54 +0200 (Mon, 19 Oct 2009) > > New Revision: 33647 > > Changeset: http://dev.haiku-os.org/changeset/33647/haiku > > > > Modified: > > haiku/trunk/headers/private/kernel/lock.h > > haiku/trunk/src/system/kernel/lock.cpp > > Log: > > * Reverted r33643 - while it doubled the performance for my test > > case (with > > high contention of the read lock (I experimented with the VM page > > mapping > > lock)), it actually hurt the compile performance pretty > > obviously. > I hope you're testing with KDEBUG=0, since otherwise the results are > not > really saying much. Though it's obviously a good thing, when things > improve > with KDEBUG=1 as well. Good thinking! I completely forgot about that, and I'm actually always testing with KDEBUG=1 (and tracing, for that matter). Will fix that, and try again. Bye, Axel.