Re: [PATCH] drm/xe/uapi: Expose EU width via topology query

["Souza, Jose" <jose.souza@intel.com>]

On Wed, 2024-07-10 at 09:59 -0500, Lucas De Marchi wrote:
> On Wed, Jul 10, 2024 at 07:52:38AM GMT, Matt Roper wrote:
> > On Tue, Jul 09, 2024 at 10:53:54PM -0700, Lucas De Marchi wrote:
> > > PVC, Xe2 and later platforms have a 16 wide EU. We were implicitly
> > > reporting for PVC the number of 16-wide EUs without giving userspace any
> > > hint that they were different than for other platforms. Xe2 and later
> > > also have 16-wide, but in those case the reported number would
> > > correspond to the 8-wide count.
> > > 
> > > Add a new item to the topology that aims to clarify what the EU_PER_DSS
> > > mask means. This new item uses mask[] as a single u8 value. Xe2 and
> > > later platforms start returning the number of SIMD16 EUs.
> > 
> > So it looks like your patch is adding a new standalone "EU type" item to
> > the list of masks returned by the query.  I.e.,
> > 
> >        - DRM_XE_TOPO_DSS_GEOMETRY = ...
> >        - DRM_XE_TOPO_DSS_COMPUTE = ...
> >        - DRM_XE_TOPO_DSS_EU_PER_DSS = ...
> >        - DRM_XE_TOPO_EU_WIDTH = {0, 1}
> > 
> > I think it would be more natural if we just didn't include
> > DRM_XE_TOPO_DSS_EU_PER_DSS in the list of masks returned on new
> > platforms, but instead added a new type of mask that would be returned
> > on the newer platforms.  I.e.,
> > 
> >   Old Xe1 platforms receive from query:
> >        - DRM_XE_TOPO_DSS_GEOMETRY = ...
> >        - DRM_XE_TOPO_DSS_COMPUTE = ...
> >        - DRM_XE_TOPO_DSS_EU_PER_DSS = ...
> > 
> >   New Xe2 platforms receive from query:
> >        - DRM_XE_TOPO_DSS_GEOMETRY = ...
> >        - DRM_XE_TOPO_DSS_COMPUTE = ...
> >        - DRM_XE_TOPO_DSS_SIMD16_EU_PER_DSS = ...
> 
> yeah, I like this and wouldn't be an abuse of the field "mask[]" like
> I'm doing here.

Ack on this approach

> 
> Should we document that the query may not have all the items?
> 
> thanks
> Lucas De Marchi