On Thu, Aug 30, 2007 at 12:49:15AM +0100, Andrew Doran wrote:

> I just merged a bunch of x86 changes into -current from the vmlocking
> branch. I have been running these changes for quite some time and believe
> that they're stable, but I have a limited selection of hardware and apps to
> test with. If stability is a priority then please hold off updating for a
> little while.
> 
> By and large the merge is to reduce diffs but some optimizations have been
> brought over. On a dual core system I'm seeing real time for builds down by
> between 3 to 4% after these changes. With more CPUs I think the impact will
> be greater, especially on amd64 systems.

I have heard that with these changes, real time for a build of NetBSD has
gone down by about 12% on a quad core Opteron. For single-CPU machines that
have SSE2 I think there may be a speed-up too - using the SSE2 instructions
to zero and copy pages was a large chunk of the improvement on the dual core
machine that I have. I haven't had a chance to verify that yet.

Thanks,
Andrew