[igt-dev] [PATCH i-g-t 3/3] benchmark: add gem_exec_delay

Lionel Landwerlin lionel.g.landwerlin at intel.com
Wed Apr 25 12:14:05 UTC 2018


On 25/04/18 13:01, Chris Wilson wrote:
> Quoting Lionel Landwerlin (2018-04-25 12:48:28)
>> This benchmark's goal is to measure the time it takes between a
>> request submitted to the hardware (either from i915 or guc) and when
>> the first RCS command stream instruction from the user batchbuffer is
>> executed. The measurement includes a number of things that need to
>> happen before the first command streamer instruction can be executed
>> like context save/restore, some hardware reconfiguration like
>> powergating, per context workaround batchbuffers...
> So what's OA doing here? RCS timetamp from submit to exec, i.e. what is
> the improvement over the existing benchmarks?
> -Chris
>

I didn't find an existing benchmark that would be able to give me that 
kind measurement in the GPU time domain.
Which one are you thinking about?

-
Lionel


More information about the igt-dev mailing list