[dpdk-dev] [RFC PATCH v3 3/6] build: automatic NUMA and cpu counts detection

[Bruce Richardson]

On Fri, Oct 23, 2020 at 10:07:17AM +0000, Juraj Linkeš wrote:
> 
> 
> > -----Original Message-----
> > From: Bruce Richardson <bruce.richardson at intel.com>
> > Sent: Wednesday, October 21, 2020 4:28 PM
> > To: Juraj Linkeš <juraj.linkes at pantheon.tech>
> > Cc: Ruifeng.Wang at arm.com; Honnappa.Nagarahalli at arm.com;
> > Phil.Yang at arm.com; vcchunga at amazon.com; Dharmik.Thakkar at arm.com;
> > jerinjacobk at gmail.com; hemant.agrawal at nxp.com; dev at dpdk.org
> > Subject: Re: [dpdk-dev] [RFC PATCH v3 3/6] build: automatic NUMA and cpu
> > counts detection
> > 
<snip>
> > What could possibly work is to set the defaults for these to "0" as done in your
> > patch, but thereafter have the resulting defaults set per-architecture, rather
> > than globally. That would allow x86 to tune things more for native-style builds,
> > while in all cases allowing user to override.
> > 
> > /Bruce
> 
> I'll change it so that per-arch (and for arm, per processor type) defaults are the defaults that will be used (when set to 0). I'd like to give users the option to use the discovery mechanism - I think we'll have to use another value with specific meaning and we can't use positive integers so I guess the best we can do is use -1 for this.
> 
Well, if you want, if the defaults are set per-architecture, is you could
use detection for the default values for the ARM builds if the build-type
is set to "native" and the max_cores etc. are set to zero. For x86 that is
almost certainly not what we want, so we'd use hard-coded values for those
but no reason other architectures all need to do the same.

/Bruce