[igt-dev] [PATCH i-g-t 1/1] intel-gpu-top: Support for client stats

Tvrtko Ursulin tvrtko.ursulin at linux.intel.com
Mon Dec 16 14:11:53 UTC 2019


On 16/12/2019 13:26, Chris Wilson wrote:
> Quoting Tvrtko Ursulin (2019-12-16 12:09:23)
>> +static void __restore_stats(void)
>> +{
>> +       int ret;
>> +
>> +       if (__stats_enabled)
>> +               return;
>> +
>> +       ret = __set_stats(false);
> 
> Therein lies the problem with a global toggle. :(

Inverted check bug, okay. :) But yes, global toggle is a bit of a silly 
thing.

> Hmm, if our stats toggling is always in process_csb() responding to
> events, they shouldn't affect execution latency... I think you have made
> them cheap enough to consider always enabling. Let's see.

I think they will be. Didn't spot anything in casual testing but can do 
more pathological cases if you want.

Regards,

Tvrtko


More information about the igt-dev mailing list