On Sat, Feb 10, 2018 at 08:39:08AM +0000, Chris Wilson wrote:
> Quoting Oscar Mateo (2018-02-09 23:46:31)
> > 
> > On 02/09/2018 02:35 PM, Chris Wilson wrote:
> > > In order to allow the compiler to use a known constant number of
> > > available engines, disable modification of intel_device_static_info
> > > during engine bring up. Instead of trying to gracefully hide the broken
> > > setup, error out -- in theory, this should be caught during power on.
> > 
> > We are about to have a case for dynamic number of available engines. 
> > It's one of the ICL patches:
> > 
> > drm/i915/icl: Check for fused-off VDBOX and VEBOX instances
> > 
> > intel_device_runtime_info as well?
> Hmm, ring_mask is more widely used than I was expecting. I think we want
> both, static_info if we ever think we can benefit from single-platform
> LTO of the engines, but whether to use runtime_info or i915->gt.engine_mask
> (and move the engine maps to i915->gt as well).
> Advantage of runtime_info, centralised place for debugging.
> Disadvantage of runtime_info, centralised place far from code.
> Maybe we don't need to say everything is inside runtime_info (just
> anything that doesn't fit elsewhere?), but use the hooks for debugging?
> Maybe having a central runtime_info is simply a bad idea?

Yeah, I don't like the "far from the code" aspect of runtime_info
(or even the static info in some cases).

Another counter argument is perhaps that people are more likely to
update the info for new platforms if it's in a central location, as
opposed having to trawl the entire codebase.

Ville Syrjälä
Intel OTC
Intel-gfx mailing list

Reply via email to