[dpdk-dev] [PATCH 4/4] lib/librte_vhost: improve vhost perf using rte_memset

[Yang, Zhiyong]

Hi, Thomas:

> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> Sent: Friday, December 2, 2016 5:46 PM
> To: Yang, Zhiyong <zhiyong.yang at intel.com>
> Cc: dev at dpdk.org; yuanhan.liu at linux.intel.com; Richardson, Bruce
> <bruce.richardson at intel.com>; Ananyev, Konstantin
> <konstantin.ananyev at intel.com>
> Subject: Re: [dpdk-dev] [PATCH 4/4] lib/librte_vhost: improve vhost perf
> using rte_memset
> 
> 2016-12-05 16:26, Zhiyong Yang:
> > +* **Introduced rte_memset and related test on IA platform.**
> > +
> > +  Performance drop had been caused in some cases on Ivybridge when
> > + DPDK code calls glibc  function memset. It was necessary to introduce
> more high efficient function to fix it.
> > +  The function rte_memset supported three types of instruction sets
> > + including sse & avx(128 bits),
> > +  avx2(256 bits) and avx512(512bits).
> > +
> > +  * Added rte_memset support on IA platform.
> > +  * Added functional autotest support for rte_memset.
> > +  * Added performance autotest support for rte_memset.
> 
> No need to reference autotests in the release notes.

Ok.
I will remove the two lines.
> 
> > +  * Improved performance to use rte_memset instead of
> copy_virtio_net_hdr in lib/librte_vhost.
> 
> Please describe this change at a higher level. Which case it is improving?

Ok, good comments.

* Improved performance to get 3% or so perf improvement
on IA platform by using rte_memset when running virtio/vhost non-mergeable
loopback test without NIC.

Thanks
Zhiyong