#
✕ nick:chadmed
#asahi · 2024-04-29 08:38
chadmedyeah its not going to be network lol
#asahi · 2024-04-25 12:10
chadmedthe chip behind it is a variant of one that was already mostly supported upstream
#asahi · 2024-04-25 12:10
chadmedthe headset jack works for input and output
#asahi · 2024-04-21 08:06
chadmedmight get you to test some audio related stuff when we get around to it
#asahi · 2024-04-21 08:05
chadmedno wokkas
#asahi · 2024-04-21 08:04
chadmedyour issues _should_ go away once lina and alyssa get around to implementing framebuffer compression
#asahi · 2024-04-21 08:03
chadmedvveapon: yes if you overuse uclamp or try to hack everything to clamp at 1024 then youre preventing the scheduler from doing its job properly
#asahi · 2024-04-21 08:03
chadmedthe reason we do not simply peg things to the pcores is that it is sometimes the case that an ecore could provide better performance, efficiency and availability for a given task, even if it is "slower". only the scheduler can know if this is the case though, so we just tell the scheduler about the task's perf requirement
#asahi · 2024-04-21 08:02
chadmedvveapon, ml00: it doesnt explicitly make the task prefer pcores, that would be taskset. what it does is tells the scheduler that the task _always_ requires max performance, even when it doesnt. this causes the scheduler to _always_ schedule it more aggressively
#asahi · 2024-04-21 05:43
chadmedvveapon: i cant see anything dodgy in there
#asahi · 2024-04-21 05:19
chadmedcan you paste the output of asahi-diagnose somewhere like tpaste and link to it
#asahi · 2024-04-21 05:18
chadmedwhat does plasma settings say your gpu is
#asahi · 2024-04-21 05:18
chadmedahhhhh
#asahi · 2024-04-21 05:18
chadmedtheres probably something else wrong then because that forces the cores into the maximum OPP when firefox is running, and also forces it to migrate to the pcores
#asahi · 2024-04-21 05:17
chadmedimacs are the least-tested/least-supported machine round these parts so it is possible that there is something innate to them that causes more issues than we're aware about
#asahi · 2024-04-21 05:15
chadmedthere are only a few edge cases where schedutil has a noticeable interactivity penalty
#asahi · 2024-04-21 05:15
chadmedno that is not expected at all
#asahi · 2024-04-21 05:14
chadmedvveapon: there are no user-controllable tunables for the scheduler however we recently got UCLAMP support enabled in the fedora kernel. you may try using uclampset to clamp firefox's util_max to 1024 and see if that helps
#asahi · 2024-04-20 23:05
chadmedi still have a patch in /etc/portage/ patches/ that modifies the timing to fix it, but not sure if we can or even should gun for that to be included in f40
#asahi · 2024-04-20 23:04
chadmedjannau, fooxi: ISTR some discussion upstream about this in kwin when we originally ran into this. there was talk of reverting the swap timing changes, but then it was decided that this neded to be fixed "properly", whatever that means///
#asahi · 2024-03-23 13:13
chadmedits a bit jank still, like panels spontaneously turning black etc but that might also be the gpu driver
#asahi · 2024-03-23 13:12
chadmedyep, i have a display p3 icc profile applied right now
#asahi · 2024-03-23 13:11
chadmedso we dont need to care about gamma_lut for colour management
#asahi · 2024-03-23 13:11
chadmedyeah, but point is its applied to the pixel data before it even gets to the ctm
#asahi · 2024-03-23 13:09
chadmedgamma_lut is not neccessary, and even if we did use it all it does is encode a vector of linear coefficients expressed in hex
#asahi · 2024-03-23 13:08
chadmedit works fine for ICC profiles too
#asahi · 2024-03-23 13:08
chadmedwith the DisplayP3 profile on my monitors kwin night colour works absolutely perfectly
#asahi · 2024-03-23 13:07
chadmedyes, hence why CTM support is important
#asahi · 2024-03-23 13:07
chadmeduserspace is easier to change than taping out a new display controller
#asahi · 2024-03-23 13:07
chadmedthere is a whole mess of hardware that doesnt support gamma_lut at all
#asahi · 2024-03-23 13:06
chadmedthat kind of legacy xorg rubbish needs to go away
#asahi · 2024-03-23 13:06
chadmedparting words before i hit the hay, everything should simply use the CTM property correctly and stop relying on gamma_lut to behave like a CTM
#asahi · 2024-03-23 12:47
chadmedPaulFerster: the asoc driver determins the control names not the dt. the kernel is probably just really old
#asahi · 2024-03-20 06:48
chadmedand copy the parameters exactly
#asahi · 2024-03-20 06:48
chadmedRoundDuckKira: have a look at the openrc script in my gentoo overlay
#asahi · 2024-03-19 06:47
chadmedif we want to keep our somewhat-open platform, never :p
#asahi · 2024-03-18 11:17
chadmedjust ask the question you dont need to prompt us to ask you what the issue is
#asahi · 2024-03-02 07:12
chadmedu154ss: NF_TABLES_IPV4 is a bool not a tristate. you cant build it as a module
#asahi · 2024-02-28 14:19
chadmedif your host kernel is new then slp's mesa wont work at all because it has an outdated uapi implementation
#asahi · 2024-02-28 13:32
chadmedknedlik: as in you built mesa in the microvm? the version in the microvm needs to match the host kernel
#asahi · 2024-02-27 12:34
chadmedsince it is applied between the CTM and the linearised data going to the crtc
#asahi · 2024-02-27 12:34
chadmedGAMMA_LUT would break that assumption, no?
#asahi · 2024-02-27 11:29
chadmedwell yeah its just a 3x3 matrix of coefficients
#asahi · 2024-02-27 11:25
chadmed its got nothing to do with ICC profiles though
#asahi · 2024-02-27 11:25
chadmedim literally using night colour right now
#asahi · 2024-02-27 11:25
chadmedyep
#asahi · 2024-02-27 11:24
chadmeddue to the colour space mismatch though (these monitors are not sRGB but DCP reports them as such) it looks kinda messed
#asahi · 2024-02-27 11:24
chadmedyeah on plasma which supports doing so via CTM
#asahi · 2024-02-27 11:22
chadmedcolour management is not going to work properly yet. there is no point. even proper ICC profiles will not do anything correctly yet
#asahi · 2024-02-25 12:19
chadmed^^

plus ›