On 13/11/2018 11:51, Chris Wilson wrote:
Quoting Jani Nikula (2018-11-13 11:45:02)
On Mon, 12 Nov 2018, Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxxxxxxx> wrote:
From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
Now that we are down to one caller, which does not even modify copied
device info, we can remove the mkwrite_device_info helper and convert the
device info pointer itself to be a pointer to static table instead of a
copy.
Only unfortnate thing is that we need to convert all callsites which were
referencing the device info directly to using the INTEL_INFO helper.
I'm not sure if that's all that bad. When I was toying around with
mkwrite_device_info removal, I actually started off with changing all
device info references to INTEL_INFO. It's a big patch, but it nicely
centralizes many of the other changes instead of splattering all over
the place.
Fwiw, replacing all the static i915->info accesses with INTEL_INFO (or
DEVICE_INFO since STATIC_INFO I think is too confusing with C, and
INTEL_INFO is not distinct enough from RUNTIME_INFO) is perhaps a
You propose DEVICE_INFO for the static part and RUNTIME_INFO for
dynamic, all with INTEL_ prefix?
pre-requisite for single platform DCE. Along the lines of
#define INTEL_INFO(i915) (&skylake_gt3_info)
Definitely, and I think this patch changed them all. There weren't that
many.
We also need to minimize the runtime portion for best DCE results. But
AFAIR it was quite good already with just IS_GEN and IS_PLATFORM changes
and not even LTO.
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
