Re: [PATCH v2 1/1] app/testpmd: add GPU memory option in iofwd engine

[Bruce Richardson <bruce.richardson@intel.com>]

On Wed, Nov 17, 2021 at 12:51:13AM +0530, Jerin Jacob wrote:
> On Wed, Nov 17, 2021 at 12:44 AM Elena Agostini <eagostini@nvidia.com> wrote:
> >
> > > From: Jerin Jacob <jerinjacobk@gmail.com>
> >
> > > Date: Tuesday, 16 November 2021 at 20:09
> >
> > > To: Ferruh Yigit <ferruh.yigit@intel.com>
> >
> > > Cc: Elena Agostini <eagostini@nvidia.com>, dev@dpdk.org <dev@dpdk.org>, Slava Ovsiienko <viacheslavo@nvidia.com>
> >
> > > Subject: Re: [PATCH v2 1/1] app/testpmd: add GPU memory option in iofwd engine
> >
> > > External email: Use caution opening links or attachments
> >
> > >
> >
> > >
> >
> > > On Tue, Nov 16, 2021 at 11:42 PM Ferruh Yigit <ferruh.yigit@intel.com> wrote:
> >
> > > >
> >
> > > > On 11/16/2021 6:06 PM, Elena Agostini wrote:
> >
> > > > >  > From: Ferruh Yigit <ferruh.yigit@intel.com>
> >
> > > > >
> >
> > > > >  > Date: Tuesday, 16 November 2021 at 19:00
> >
> > > > >
> >
> > > > >  > To: Elena Agostini <eagostini@nvidia.com>
> >
> > > > >
> >
> > > > >  > Cc: dev@dpdk.org <dev@dpdk.org>, Slava Ovsiienko <viacheslavo@nvidia.com>
> >
> > > > >
> >
> > > > >  > Subject: Re: [PATCH v2 1/1] app/testpmd: add GPU memory option in iofwd engine
> >
> > > > >
> >
> > > > >  > External email: Use caution opening links or attachments>
> >
> > > > >
> >
> > > > >  >
> >
> > > > >
> >
> > > > >  > On 11/11/2021 9:41 PM, eagostini@nvidia.com wrote:
> >
> > > > >
> >
> > > > >  > > --- a/app/test-pmd/meson.build
> >
> > > > >
> >
> > > > >  > > +++ b/app/test-pmd/meson.build
> >
> > > > >
> >
> > > > >  > > @@ -32,7 +32,7 @@ if dpdk_conf.has('RTE_HAS_JANSSON')
> >
> > > > >
> >
> > > > >  > >       ext_deps += jansson_dep
> >
> > > > >
> >
> > > > >  > >   endif
> >
> > > > >
> >
> > > > >  > >
> >
> > > > >
> >
> > > > >  > > -deps += ['ethdev', 'gro', 'gso', 'cmdline', 'metrics', 'bus_pci']
> >
> > > > >
> >
> > > > >  > > +deps += ['ethdev', 'gro', 'gso', 'cmdline', 'metrics', 'bus_pci', 'gpudev']>
> >
> > > > >
> >
> > > > >  > I didn't review the set, but in a very high level do we want to add
> >
> > > > >
> >
> > > > >  > 'gpudev' as dependency? Isn't this like adding 'rawdev' as dependency.
> >
> > > > >
> >
> > > > > gpudev is a library that can be built without a gpu driver as all the other libraries
> >
> > > > >
> >
> > > > > and itis actually used only in case of GPU memory mempool.
> >
> > > > >
> >
> > > > > Reasons for this patch are:
> >
> > > > >
> >
> > > > > - Have an upstreamed benchmark tool to measure network metrics using GPU memory
> >
> > > > >
> >
> > > > > - Test some DPDK features not really tested anywhere like the external memory mempool feature
> >
> > > > >
> >
> > > >
> >
> > > > I can see the reason, that is obvious, yet again why we are not adding rawdev
> >
> > > > testing to the testpmd? But adding gpudev.
> >
> > > > It is easier to add it to the testpmd, and for some testing perspective it
> >
> > > > makes sense, but still I am not quite sure about this new dependency, I would
> >
> > > > like to get more feedback.
> >
> > >
> >
> > > I had the similar concern earlier. IMO, It is better to have a
> >
> > > separate test application for gpudev like
> >
> > > other device classes. For eventdev cases when it needs to work with
> >
> > > ethdev for Rx adapter cases,
> >
> > > We have enabled such code in app/test-eventdev to make testpmd focus on ethdev.
> >
> >
> >
> > gpudev already has a test app in app/test-gpudev.
> >
> >
> >
> > gpudev needs to be also test with network card and today another application
> >
> > decidated to test gpudev over the network would be very similar to testpmd io.
> >
> >
> >
> > At this stage, there is no point in reinventing the wheel
> 
> 
> I think, it is not about not reinventing the wheel, It is about
> maintenance of testpmd,
> currently, the feature are specific to ethdev. Adding more
> cross-device-specific features
> will populate the testpmd. I had a similar case when it network cases
> need to be integrated to eventdev,
> I choose to have it test-eventdev so that testpmd focus remains for ethdev.

Since there is an increasing push for libraries to be able to be disabled
in DPDK, at minimum I would look to have gpudev as an optional dependency
here, so that disabling it would not completely disable building testpmd.

Regards,
/Bruce