On 10/12/20 1:13 PM, Magnus Karlsson wrote:
[...]
Nope, that was a bad idea. After measuring, this one produces worse
performance than the original suggestion with padding in between all
members. Cannot explain why at the moment, but the numbers are
convincing and above noise level for sure. So l
On Mon, Oct 12, 2020 at 11:37 AM Magnus Karlsson
wrote:
>
> On Mon, Oct 12, 2020 at 10:37 AM Magnus Karlsson
> wrote:
> >
> > On Fri, Oct 9, 2020 at 5:03 PM Daniel Borkmann wrote:
> > >
> > > On 10/8/20 4:12 PM, Magnus Karlsson wrote:
> > > > From: Magnus Karlsson
> > > >
> > > > Introduce one
On Mon, Oct 12, 2020 at 10:37 AM Magnus Karlsson
wrote:
>
> On Fri, Oct 9, 2020 at 5:03 PM Daniel Borkmann wrote:
> >
> > On 10/8/20 4:12 PM, Magnus Karlsson wrote:
> > > From: Magnus Karlsson
> > >
> > > Introduce one cache line worth of padding between the producer and
> > > consumer pointers
On Fri, Oct 9, 2020 at 5:03 PM Daniel Borkmann wrote:
>
> On 10/8/20 4:12 PM, Magnus Karlsson wrote:
> > From: Magnus Karlsson
> >
> > Introduce one cache line worth of padding between the producer and
> > consumer pointers in all the lockless rings. This so that the HW
> > adjacency prefetcher w
On 10/8/20 4:12 PM, Magnus Karlsson wrote:
From: Magnus Karlsson
Introduce one cache line worth of padding between the producer and
consumer pointers in all the lockless rings. This so that the HW
adjacency prefetcher will not prefetch the consumer pointer when the
producer pointer is used and
From: Magnus Karlsson
Introduce one cache line worth of padding between the producer and
consumer pointers in all the lockless rings. This so that the HW
adjacency prefetcher will not prefetch the consumer pointer when the
producer pointer is used and vice versa. This improves throughput
performa
