Messages in this thread | | | Date | Tue, 19 Jun 2018 23:06:39 +0100 | From | Matt Fleming <> | Subject | Re: [RFC 00/11] select_idle_sibling rework |
| |
On Wed, 30 May, at 04:22:36PM, Peter Zijlstra wrote: > Hi all, > > This is all still very preliminary and could all still go up in flames (it has > only seen hackbench so far). This is mostly the same code I posted yesterday, > but hopefully in a more readable form. > > This fixes the SIS_PROP as per the outline here: > > https://lkml.kernel.org/r/20180425153600.GA4043@hirez.programming.kicks-ass.net > > and Rohit's suggestion of folding the iteration loops. > > For testing I would suggest to ignore the last 3 patches, those are purely > cleanups once the first lot is found to actually work as advertised.
This series looks pretty good from my testing. I see double-digit improvements to hackbench results and only one case of a clear regression (easily offset by all the wins).
Are you aware of any regressions for particular benchmarks I should take a look at?
| |