04:22 <tomeu> alyssa: because we didn't start to get any logs, I guess no kevin was available for a long time

04:23 <tomeu> guess we could increase the max time allowed before we timeout (right now it's 1h), but I just retry those

04:23 <tomeu> ah, both kevins are out: https://lava.collabora.co.uk/scheduler/device_type/rk3399-gru-kevin

04:24 <alyssa> tomeu: ...of the closet?

04:24 <tomeu> was thinking for the weekend :p

04:25 <alyssa> :P

04:25 <tomeu> alyssa: btw, has the CI been useful so far?

04:36 <alyssa> tomeu: Yes and no

04:36 <alyssa> Yes, in the sense that

04:36 <alyssa> 's it's caught real bugs before I've pushed out bad code, so that's helpful.

04:36 <alyssa> No, in the sense that it's not returning consistently useful results since there are, well, consistently flip-flops. That's just time and test fixing, alas.

04:39 <tomeu> yeah, it has to do with the maturity of the codebase

04:40 <tomeu> but on friday I wasn't seeing any flip-flops left, could it be that new code made some tests to start flip-flopping?

04:41 <alyssa> I dunno

04:47 <tomeu> alyssa: btw, this could be due to the severity of the leaks having worsened: https://gitlab.freedesktop.org/tomeu/mesa/-/jobs/294080

04:48 <tomeu> will give a look to blend-v3

04:52 <alyssa> Not sure why the leaks would -worsen-

04:52 <alyssa> I haven't fixed anything but I don't know that I've made anything worse..?

04:52 <tomeu> ok, I may lower the batch size just in case

04:53 <alyssa> (I mean, I totally might have. I suppose blend shaders probably introduced leaks..)

04:54 <tomeu> ok, but I guess no possible leaks of full BOs?

04:55 <alyssa> I hope not~

04:55 <alyssa> Certainly possible

05:00 <tomeu> ok, first I will look at the apparent regressions in blend-v3

05:00 <alyssa> tomeu: Keep in mind blend-v3 includes a lot of bizarre new code so a lot of the regressions are probably actually regressions, hence why I haven't pushed it (CI in action :P)

05:00 <alyssa> Same with move-typing for that matter

05:00 <tomeu> alyssa: ah, so which are the flip-flops that you found worrying?

06:01 <tomeu> dEQP-GLES2.functional.shaders.linkage.varying_7 seems to have been genuinely fixed in master since friday

06:03 <tomeu> dEQP-GLES2.functional.shaders.functions.control_flow.return_in_nested_loop_fragment as well

14:42 <alyssa> tomeu:Huh, not sure why those would be genuinely fixed but I won't complain :)

15:21 <bbrezillon> alyssa: is that okay if I dump the counters the way they were dumped with the mali_kbase backend?

15:32 <bbrezillon> s/mali_kbase/pan_nondrm/

18:10 <bbrezillon> alyssa: https://github.com/bbrezillon/linux/commits/panfrost/perfcnt-v2 and https://github.com/bbrezillon/mesa/commits/panfrost/perfcnt-v2

18:10 <bbrezillon> just in case you need it before the end of the week

18:11 <bbrezillon> should allow to dump the counters the way you did with the nondrm driver