00:02 KungFuJesus: I guess to vectorize that loop I'd have to figure out the recurrence relation to do 4 at a time similar to what OpenZFS does for fletcher4
00:04 KungFuJesus: is XOR associative?
00:05 HdkR: yes
00:05 HdkR: commutative and associative
00:07 KungFuJesus: oh yeah then this shouldn't be super difficult
00:08 KungFuJesus: of course, it's only reporting about 8% of the CPU cycles in that function and not much anywhere else, really
00:10 KungFuJesus: that table lookup is definitely not cache friendly, that's for sure
03:51 imirkin: emersion: weird, i can't figure out why present doesn't seem to work with wlr
03:51 imirkin: i've tried messing with linear/not-linear, gart/vram, but it's all the same result
03:53 imirkin: i see the imported buffer being set as the rendertarget
03:54 imirkin: i see XCB_PRESENT_COMPLETE_NOTIFY events
03:54 imirkin: i just don't see anything other than the gray background on screen
07:13 fling: How to disable hardware accel?
07:13 fling: I can export llvmpipe in bashrc but what is the proper way?
07:21 imirkin: LIBGL_ALWAYS_SOFTWARE=1
07:21 imirkin: you can also boot with nouveau.noaccel=1
07:21 imirkin: although that will also disable e.g. xf86-video-nouveau's acceleration
07:21 imirkin: which may or may not be desired
07:22 imirkin: hrmph. looks like at least on pascal, we sometimes get GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN wrong. got some random-looking test failures in dEQP-GLES3 =/
07:24 imirkin: that's annoying.
08:31 fling: ok
08:33 fling: imirkin: I tried exporting GALLIUM_DRIVER=llvmpipe and LIBGL_ALWAYS_SOFTWARE=true via .xinitrc but this crashes my box after fex and somewhere in exec awesome
08:33 fling: imirkin: exporting via .bashrc works fine though
08:34 fling: if I run superslicer which is an opengl app without these exports ^ it crashes after a while
08:34 fling: But superslicer is too slow with software render haha
08:35 imirkin: moral of the story: can't win, don't try
08:36 fling: What to buy instead of this gt215?
08:36 fling: I want it to be low power and tiny
08:36 imirkin: not-nvidia
08:36 imirkin: i guess qcom
08:36 fling: I want it to work with linux libre :>
08:36 fling: oh
08:36 fling:checking
08:36 imirkin: that won't work with linux-libre
08:37 imirkin: if you want it to work with linux-libre, then ... you're in for a bad time.
08:37 imirkin: that distro makes completely arbitrary choices which i see zero foundation for in "open" philosophy, which causes nvidia to basically be the only supported hardware. which is also amusing.
08:38 fling: fsf approved! :D
08:38 imirkin: woohoo?
08:38 imirkin: i guess intel might work ok too? not 100% sure.
08:39 fling: yeah intel works but this box has ast builtin
08:39 fling: ast driver has no direct rendering yet
08:39 imirkin: not gonna get too much accel out of ast ;)
08:39 imirkin: if you're going for power, GTX 780 Ti is the best nouveau supports with reclocking
08:39 imirkin: low power and tiny -- i mean there's like GT 710
08:40 imirkin: but it's crap
08:40 fling: I have 760 but I had crashes with it too
08:40 imirkin: but it's low power and tiny ;)
08:40 fling: and it is not tiny
08:40 imirkin: it's nouveau. you're gonna have crashes.
08:40 fling: ah
08:40 imirkin: you should be amazed that there are short periods of time where you *don't* get crashes
08:40 fling: I had no crashes for a month recently!
08:40 fling: now crashing every day
08:40 imirkin: in fairness, i tend to get pretty high uptimes with nouveau
08:41 imirkin: but i know what to do and not do
08:41 imirkin: i think i've had like 6mo before
08:41 fling: yeah
08:41 fling: you type a command and ^W
08:41 fling: crash avoided
08:41 imirkin: ;)
08:41 imirkin: this box has been up 18 days, and i've been running test suites on it
08:41 fling: maybe I will try 760 again
08:41 imirkin: so it's not completely horrid
08:42 imirkin: nouveau development tends to have negative effects on one's uptime
08:43 ccr: my nouveau usage is only on machines that get shut down / booted at end of use, so haven't really had much of problems. mainly some odd "frame dropping" with video playing on youtube with a NV50
08:43 fling: Should I try with GALLIUM_DRIVER=llvmpipe or without it btw?
08:44 fling: imirkin: will it crash with nv50? ^
08:44 fling: I used some old gtx-8600-super for long time
08:46 fling: I have nv40 card somewhere, going to try it too
08:48 imirkin: fling: lol, nv40 is going to be horrid
08:48 imirkin: you're already using a nv50 i thought
08:48 fling: I played sc2 on it without issues
08:49 fling: using this now -> VGA compatible controller: NVIDIA Corporation GT215 [GeForce GT 240] (rev a2)
08:49 imirkin: huh. dEQP-GLES31.functional.shaders.multisample_interpolation.interpolate_at_sample.non_multisample_buffer.sample_n_default_framebuffer fails on gp108. probably missing some stupid weird case. gr.
08:49 imirkin: fling: yeah, that's nv50 (family)
08:49 fling: wow ok
08:49 imirkin: fling: wait, do you have GDDR5 on that board?
08:50 fling: probably not, I seen ddr3 somewhere in the log sometime
08:50 imirkin: ok. GT215 + GDDR5 is extra-bad
08:50 fling: right
08:50 ccr: interesting .. how does memory type affect badness?
08:50 imirkin: the generation wasn't designed for GDDR5
08:51 imirkin: but then time happened
08:51 imirkin: and they wanted to put out a high-end board
08:51 imirkin: so they did, and hacked it on
08:51 fling: sneaky
08:51 imirkin: but i don't think they got everything QUITE right in the bios setup
08:51 imirkin: and ended up relying on the driver configuring the board correctly
08:51 imirkin: now take a wild guess whether nouveau does this :)
08:51 ccr: I see
08:52 ccr:rolls a die
08:52 imirkin: it used to hang on load
08:52 imirkin: now it at least limps along
08:52 imirkin: it's not supported by the existing reclocking logic
08:52 imirkin: a couple of people have one, myself included, but the motivation to work on it is ... limited
08:53 ccr:nods
09:29 emersion: imirkin: hm, mesa should use Present as well, i assume mesa clients still work? i wonder what we're doing differently
09:45 imirkin: emersion: yeah, i looked at loader_dri3_helper.c -- looks quite similar
10:20 emersion: Lyude: yeah, i'd like to fix a few display bugs at some point
10:21 emersion: also your patches fix some bugs for my users, so that's why i'm interested in these
10:25 emersion: (we refuse to officially support any proprietary driver, so i'm all for making the open-source one work as smoothly as possible)
15:07 KungFuJesus: How do people lurk in IRC with screen sessions for days when freenode periodically forces you to re-identify?
15:09 emersion: if you use a bouncer, the bouncer will re-auth you automatically
15:12 KungFuJesus: is this some script magic that irssi can leverage?
15:13 KungFuJesus: oh, sounds like it's a middle man that connects on your behalf to hide your IP
15:13 emersion: a bouncer is something that sits between you and the IRC servers, to maintain the conenction and collect logs
15:15 ccr: basically kind of a proxy
15:17 ccr: I've not noticed any periodical forced re-identification tho? only when disconnected or so.
15:27 KungFuJesus: evidently my connection timed out for whatever reason
15:27 KungFuJesus: perhaps the freenode mirror I was hitting went down
15:29 KungFuJesus: was connected to orwell
15:29 KungFuJesus: imirkin: I think I have a pretty clever solution to portage spending forever in xz decompression. Looking at XZ's code, it respects the XZ_OPT environment variable. There's an argument to ignore the checksum
15:30 KungFuJesus: and given that portage has a manifest for all the source archives...
15:34 karolherbst: KungFuJesus: also most clients have support for auto auth
15:58 KungFuJesus: Yeah I figured there'd be a way to make irssi re-identify on reconnection
15:59 KungFuJesus: I guess at that point you're just racing the freenode's identify police for channels that require it
16:00 HdkR: https://irssi.org/documentation/startup/ grep for nickserv there
16:01 emersion: KungFuJesus: that's why freenode supports SASL authentication
16:01 emersion: ie. authentication that happens early in the IRC protocol
16:02 emersion: sounds like SASL is covered in the docs linked by HdkR
16:03 KungFuJesus: hmm, might look into that
16:04 KungFuJesus: irssi's auto might just work for this too, without SSL
16:15 imirkin: KungFuJesus: i use SASL auth
21:51 KungFuJesus: Do all those recent commits from collabora to mesa mean I might finally get a working 3d accelerated X on my Odroid N2?
21:52 KungFuJesus: (Mali G52)
21:54 KungFuJesus: oh hell yes, that's promising: https://www.collabora.com/news-and-blog/blog/2021/01/13/desktop-opengl-3-1-on-mali-gpus-with-panfrost/
21:56 HdkR: Which that's a bit unrelated to this particular channel, but yes.
21:57 RSpliet: well, considering we came from xz, we're getting closer...
21:58 KungFuJesus: Sorry, I know #mesa's probably more appropriate, but I figured there's a fair bit of overlap in here
22:02 imirkin: probably #panfrost for bifrost/midgard gpu discussions
22:02 imirkin: and yes, there's a lot of overlap
22:07 KungFuJesus: the chatter for xz was just frustration patching an compiling fixes for endianness bugs in firefox with Gentoo's portage - which in turn was (loosely) carry over from endianness bug discussion in nv42's code
22:25 RSpliet: KungFuJesus: don't worry. If people are genuinely bothered they should've let you know ;-)
23:42 Lyude: emersion: looking right now to see which of my fixes are ready to be sent to the ML btw
23:48 emersion: :)