ChanServ changed the topic of #zink to: official development channel for the mesa3d zink driver ||
sinanmohd has joined #zink
sinanmoh- has quit [Ping timeout: 480 seconds]
<zmike> anholt_: hm I tried your tu-marker branch and it didn't seem to be giving me any of the zink markers that I could see
<anholt_> you had MESA_GPU_TRACES=perfetto, right?
<zmike> yea
<anholt_> just that with print,perfetto was logging them for me
<zmike> hmmm ok I see them in print
<zmike> why not in the graph then
<zmike> anholt_: they should be showing up in 'GPU Queue 0' ?
<zmike> yeah I dunno what's going on, but I see them with MESA_GPU_TRACES=print,perfetto, yet they don't show up in the graph at all
<anholt_> So, my experience was that the graph would go badly once there was mismatched begin/end
<zmike> hm
<anholt_> I need to go make it dynamic, but I left off when trying to sort out the string lifetime thing.
<anholt_> (apparently the team maintaining perfetto has, unsurprisingly, been reorged/laid off)
<zmike> ugh
<zmike> seems weird that both yours and markcollins work would have the same issue of not showing up in the graph
<zmike> trying to get a graph of one of these angle traces and it's not going well
<zmike> anholt_: think I found the big perf hole
<zmike> hm this is somewhere between a 15% and 40% improvement on mobile_legends but somehow the perf is wildly unstable
<zmike> * on tu, not intel