20:58dreamingcat: karolherbst hi
21:02karolherbst: dreamingcat: ahh hi.. I checked it out today and I only saw the tearing with dri 3 disabled... and with dri 3 everything was fine :/
21:02karolherbst: it was on 5.15.1 though
21:03karolherbst: 5.15.1 doesn't contain any relevant fixes though afaik
21:03dreamingcat: karolherbst: really everything? do you have plasma?
21:04dreamingcat: and settings center didn't flicker on any panels?
21:04karolherbst: tried firefox and doing random stuff in systemsettings
21:04dreamingcat: well at least I know it's not only my problem :)
21:05karolherbst: ahh, so there are other reports?
21:05karolherbst: although I can also see this being some mesa issue
21:05karolherbst: dreamingcat: one thing I forgot to ask.. what CFLAGS was mesa compiled with?
21:05dreamingcat: no, I mean you've managed to replicate it
21:05karolherbst: ahh no, I didn't
21:06karolherbst: just the tearing, but that's expected
21:06karolherbst: can't fix tearing without dri 3
21:06dreamingcat: it wasnt Ofast if that's what you mean
21:06dreamingcat: I read the logs
21:06karolherbst: just checking
21:06karolherbst: gentoo users... tend to do such things
21:07dreamingcat: we definitely do.
21:07dreamingcat: "-O2 -pipe -march=znver1 -mtune=znver1 -mmmx -msse -msse2 -msse3 -mssse3 -msse4a -mcx16 -msahf -mmovbe -maes -msha -mpclmul -mpopcnt -mabm -mfma -mbmi -mbmi2 -mavx -mavx2 -msse4.2 -msse4.1 -mlzcnt -mrdrnd -mf16c -mfsgsbase -mrdseed -mprfchw -madx -mfxsr -mxsave -mxsaveopt -mclflushopt -mxsavec -mxsaves -mmwaitx -mclzero --param l1-cache-size=32 --param l1-cache-line-size=64 --param l2-cache-size=512"
21:07karolherbst: dreamingcat: I used to use gentoo... with USE=-* and shit :D
21:07dreamingcat: karolherbst: me too. but then I discovered profiles
21:08karolherbst: I knew about profiles from the start :D
21:08karolherbst: I was still doing it
21:08karolherbst: it was kind of fun, but then again...
21:08dreamingcat: karolherbst: I actually expected you to cringe over my CFLAGS :)
21:11karolherbst: yeah no clue.. I knew that gentoo added a bunch of extension flags and random stuff, but...
21:11karolherbst: no clue if any of those actually do any harm
21:11karolherbst: at least I doubt they do...
21:11dreamingcat: errr... those are gcc params
21:12karolherbst: yeah.. but still mostly CPU extension thingies
21:14dreamingcat: Another proof that english-speaking gentoo community is far more polite and forthcoming than those of my native languages... Had I shown such CFLAGS to them I'd be chastised for specifying instruction set implicitly instead of trusting march/mtune to enable them
21:15karolherbst: :D oh well
21:15dreamingcat: anyway, I see 5.15.1 in portage, going to try it now
21:15imirkin: dreamingcat: fyi there aresome shortcuts like -march=skylake or whatever
21:15imirkin: so you don't have to enable stuff 1-by-1
21:15dreamingcat: see... that's exactly what I mean :)
21:16imirkin: that's ... not chastising.
21:16imirkin: oh. and i see you have the -march in there anyways
21:16imirkin: i somehow glazed over it.
21:17dreamingcat: yes, that's not chastising. I tend to exaggerate things for dramatic purpose. But you've got the point
21:24karolherbst: sometimes gcc can also be buggy _but_ sometimes extensions get disabled
21:24karolherbst: like mmx
21:24karolherbst: there is hardly any point in using mmx afaik
21:24imirkin: except if you're writing code you want to be compatible across lots of CPUs
21:24imirkin: i keep meaning to push out some ddx changes which optimize a few things for mmx
21:24karolherbst: ohh sure, but not if you compile for your own system
21:25imirkin: well, the code will have mmx ops
21:25imirkin: for all systems
21:25imirkin: (that support mmx)
21:25karolherbst: I think it would be better to just do sse2 stuff
21:25imirkin: just ... more code
21:25imirkin: and more things to test
21:25imirkin: iirc my latest iteration does do that though ;)
21:26karolherbst: well we could also rely on autovec to not mess up....
21:26imirkin: that was a good laugh
21:26karolherbst: yeah.. autovec is terrible
21:26imirkin: esp for optimizing this kind of stuff.
21:26imirkin: weird bit-packed math
21:26karolherbst: but you can change code in a way that compilers are more likely to autovec
21:26imirkin: but i get to use "pavg" so that's cool
21:27karolherbst: ahh yeah...
21:27karolherbst: I guess more then doing weird loop opts to operate on 256 bit instead of 32 is the only reliable way here
21:27imirkin: this site is very useful btw: https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html
21:27karolherbst: s/the only reliable way here/the only thing one can do/
21:28karolherbst: yeah, the intrinsics are nice
21:31dreamingcat: karolherbst: 5.15.1 is the same way as 5.15.0 here
21:32karolherbst: yeah.. strange
21:32karolherbst: I guess there is something different...
21:32karolherbst: might also be just mesa
21:32karolherbst: what version is mesa on your system?
21:33karolherbst: I tested with.. 21.2.5 here I think
21:34karolherbst: imirkin: did we fixed weird stuff recently in mesa? I highly doubt it, but...
21:34imirkin: at least not that i'm aware
21:35dreamingcat: ok I'm going to test ~ mesa, which is 21.3.0_rc3
21:35karolherbst: dreamingcat: there are also no random messages inside dmesg or something, right?
21:37dreamingcat: karolherbst: no
21:40dreamingcat: Though the latest mesa has (-dri3%*) in useflags
21:44karolherbst: why though?
21:44karolherbst: enabled by default
21:44karolherbst: not by default
21:44karolherbst: but always
21:45imirkin: the use flag got removed
21:45imirkin: this is what (-dri3%*) indicates
21:45imirkin: i.e. it was previously enabled, but in the new ebuild it's gone entirely
21:45karolherbst: imirkin: how do you see if it was enabled though.. the *?
21:52dreamingcat: same luck with fresh mesa :(
21:52karolherbst: dreamingcat: yeah sooo... no idea what's up honestly :/
21:53karolherbst: could really be just random issue we never actually figured out yet...
21:53imirkin: karolherbst: yes, the *