Re: Use gcc built-in atomic inc/dec in lock.c

----- Цитат от Mikko Tiihonen (mikko.tiihonen@nitorcreations.com), на 14.12.2012 в 17:33 -----<br /><br />> On
12/13/201212:19 AM, Peter Geoghegan wrote:<br />>> On 12 December 2012 22:11, Mikko Tiihonen<br />>>
wrote:<br/>>>> noticed a "XXX: It might be worth considering using an atomic fetch-and-add<br />>>>
instructionhere, on architectures where that is supported." in lock.c<br />>>><br />>>> Here is my
firsttry at using it.<br />>><br />>> That's interesting, but I have to wonder if there is any evidence
that<br/>>> this *is* actually helpful to performance.<br />> <br />> One of my open questions listed in
theoriginal email was request for help on<br />> creating a test case that exercise the code path enough so that it
any<br/>> improvements can be measured.<br />> <br /><br />Running pgbench on 16+ cores/threads could stress
lockingprimitives. From my experience even benchmarks run on 8 core systems should tell the difference.<br /><br
/>--<br/>Luben Karavelov