Tuesday, 2022-10-04

lkclprogrammerjake, rfp confirmed do update bugreports00:25
programmerjakeoh, yeah...sorry I forgot00:25
programmerjakedone. bugzilla email is slow...00:29
lkclyes, there's a massive DDOS attack underway out of brazil01:25
programmerjakeah, ok. imho the mail server needs to be set to allow localhost to bypass any rate limits, iirc that was a problem before01:27
lkclit's currently under attack01:28
lkclso that's a no01:28
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC01:44
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc01:58
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC04:24
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc04:25
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc04:26
*** doppo <doppo!~doppo@2604:180::e0fc:a07f> has quit IRC05:50
*** doppo <doppo!~doppo@2604:180::e0fc:a07f> has joined #libre-soc05:52
*** ghostmansd <ghostmansd!~ghostmans@broadband-188-32-220-156.ip.moscow.rt.ru> has joined #libre-soc07:54
*** ghostmansd <ghostmansd!~ghostmans@broadband-188-32-220-156.ip.moscow.rt.ru> has quit IRC08:34
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@broadband-188-32-220-156.ip.moscow.rt.ru> has quit IRC09:41
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc09:43
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC12:54
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc12:55
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC12:59
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc13:00
*** tplaten <tplaten!~isengaara@55d40ccb.access.ecotel.net> has joined #libre-soc13:02
*** octavius <octavius!~octavius@> has joined #libre-soc13:05
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC13:32
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@broadband-188-32-220-156.ip.moscow.rt.ru> has joined #libre-soc13:33
tplatenXDC 2022 - Day 1 - October 4th, 2022 is starting, tomorrow I will do my talk at FOSSXR: Using the Libre-SOC for building a mobile VR headset14:34
*** ibot <ibot!~supybot@libre-soc.org> has joined #libre-soc15:28
*** ibot` <ibot`!~supybot@libre-soc.org> has quit IRC15:28
*** ghostmansd <ghostmansd!~ghostmans@broadband-188-32-220-156.ip.moscow.rt.ru> has joined #libre-soc15:40
*** ghostmansd <ghostmansd!~ghostmans@broadband-188-32-220-156.ip.moscow.rt.ru> has quit IRC15:56
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@broadband-188-32-220-156.ip.moscow.rt.ru> has quit IRC16:19
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc16:21
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC16:30
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@broadband-188-32-220-156.ip.moscow.rt.ru> has joined #libre-soc16:38
lkcllet us know where it's accessible?16:45
octaviuslkcl, currently livestreamed: https://www.youtube.com/c/XOrgFoundation16:46
lkclok brilliant16:52
tplaten"How to write a Vulkan driver in 2022" is really interesting.16:59
lkclprogrammerjake, just have to go out to the shops, got the single-review, back in 1/2 hour17:14
programmerjakelkcl: also, can you check if it shows up in the archives *when you get back*, libre-soc-dev seems to be dropping all my messages starting some time before Mon Oct 3 19:46 PST...17:17
*** awygle <awygle!~quassel@2604:a880:2:d0::5380:3001> has quit IRC17:21
*** sauce <sauce!~sauce@omae.wa.mou.shindei.ru> has quit IRC17:21
*** sauce <sauce!~sauce@omae.wa.mou.shindei.ru> has joined #libre-soc17:22
*** awygle <awygle!~quassel@2604:a880:2:d0::5380:3001> has joined #libre-soc17:22
programmerjakelast message i sent that actually made it to the archives was 47hr ago, everything i sent after that isn't on there17:25
programmerjakethe archives seem to be dropping all of *everyone's* messages in the last 26hr17:51
lkclthat's when the attacks started and i am not going to bother investigating.  there is absolutely no point whatsoever.17:55
lkclthere is absolutely no point whatsoever in investigating or wasting time investigating or attempting to mitigate a situation affecting world-wide internet traffic that was already up to a MILLION attacks on hosts world-wide in under 3 hours.17:57
lkclat that point it was literally one of the world's largest DDOS attacks that's ever been recorded and it's still not finished, 18 hours later.17:58
lkclmy other server is in the middle of an area in iomart that has been completely cut off, i can't gain any kind of access to it.18:05
programmerjakehmm, i got just around 1000 http requests on build.libre-soc.programmerjake.tk in the last 2 days...so the ddos isn't everywhere or they're being blocked before they get to my server18:07
programmerjakethe server is in germany iirc18:07
markossigh, did it have to be dav1d? if it was libaom, I would be done by now, wasted so many days just trying to figure out how their testing system works, full of macros and function creators that load functions on the fly to test them, how much more complex can you make a unit test?18:25
markosofc the problem was isolating the actual unit test18:25
markosI could just commit the whole dav1d library and be done with it, but that's not acceptable18:25
markosif I don't do this step today, I'll abandon this task and move to mp318:26
markosthere just is not enough time18:26
programmerjakehttps://xkcd.com/2680/ XD18:28
programmerjakelkcl: if libre-soc-dev is ignoring all incoming email, then we should have libre-soc conversations somewhere that is public & archived, otherwise the latest conversations on svp64 vl=1 overrides and on my plans for the bigint presentation are effectively private and invisible to everyone not cc-ed18:41
programmerjakebugzilla seems to still work, so i'll put the bigint presentation stuff there18:42
lkclprogrammerjake, if it's a large file (binary format) please don't for goodness sake put it into the bugtracker18:51
lkcli really don't want to have to even remotely attempt to have to roll back the postgresql database or attempt manual SQL delete queries18:51
lkcl*do not* under *any circumstances* use the bugtracker as a file store18:51
lkclscp or rsync the file up to your home directory on libre-soc.org or talos1.libre-soc.org and i will upload it to the ftp site from there18:52
programmerjakeyeah, it's email text, not giant binary files18:52
lkclok whew19:00
lkcljacob please *don't* mention timing *anywhere*.  at all19:06
lkclevery time you do so i am forced into a position of explicitly stating publicly that it is out of scope19:07
lkclthe moment you start making claims of meeting a certain timing you place the entire project in jeapordy19:07
lkclsomeone could come along and say, "but you made a claim that this was guaranteed timing and therefore guaranteed secure, i'm going to sue you for false and misleading statements"19:08
lkclit's really serious19:08
lkclwe need literally TEN times more money than we're being funded to do the analysis properly at every single stage of the entire project19:08
lkcl"reduces number of instructions" is perfectly fine19:09
programmerjakewould it be fine if i add an explicit disclaimer that this is example timing and the final cpu will almost certainly be different?19:10
programmerjakebecause imho removing timing from the demo removes like 50% of the impact...19:11
markosI don't think anyone could make *any* claims about timing at this point, it's all hardware dependent19:21
programmerjaketrue, but mostly reasonable example hardware showing a 4x speedup is impressive, which is most of the point19:23
markosit might create expectations which could or could not be true, eg. if a hw vendor decided supporting the instructions but avoiding the actual recommended hardware implementation -which is entirely possible19:25
markosI've seen it happen multiple times on Arm19:25
markosadopting the ISA does not necessarily mean adopting the micro-architecture19:25
programmerjakehence the disclaimer19:26
markosI can't decide about that, I do think it doesn't do any harm to demonstrate that it's possible19:26
markosI mean getting sued for something that is open source and a proof of concept, which might or might not get in this particular form in the final ISA is a long stretch19:27
*** choozy <choozy!~choozy@75-63-174-82.ftth.glasoperator.nl> has joined #libre-soc19:28
markosfor the same reason not much can be said about timing performance on the vp8/vp9 svp64 loops, they are certainly simpler to write, but are they faster in comparison to the same VSX/NEON algorithm? I don't know, perhaps, if the hardware SV loop has been optimized and can issue multiple instructions per iteration and loads can run in parallel with adds/muls, etc19:31
markosbut if not, it might be also slower19:31
markosbeing able to handle tons of registers helps ofc but otoh, AVX512 could do many elements in a single cycle with the right instructions, which we cannot19:32
markosI don't think it's bad to mention it, but I wouldn't put a stretch to timing claims19:33
programmerjakeok. i still think it would make an awesome demo to watch a simple bigint mul algorithm go through the simd alus...19:36
markosindeed, that would be something that I would put on the fpga19:36
markoswhere you would at least have an actual reference implementation19:37
markosand from that you can *actually* extrapolate performance numbers19:37
markosnot very accurate, but a good indication19:38
programmerjakebut we don't have simd alus with the required issue/merging hardware, and i doubt we'll have it before the presentation date ... also running it on a fpga isn't a good animation of what's happening inside so people can understand how it works19:41
markosbut it can tell you how many cpu cycles the reference implementation will take19:43
markosah wait19:43
markospresentation date, you mean for the OPF summit?19:43
programmerjakeyes, which is useful for perf metrics but not for pretty animations of how data moves in the cpu19:44
markosright, yeah I don't know19:44
markosI personally don't think people will take presentation material as written in stone19:44
programmerjakeyes, opf summit19:44
markosmany people present claims in presentations19:45
markosnot necessarily lying, but until things finalize many things change19:45
markosyou can't really hold someone accountable because a claim they made 2-3+ years ago is not living up to the expectations19:45
markosimho -and it's just that, a humble opinion- I think people will understand that you can't possible claim performance timings at this stage19:46
tplatennow I listen to "Tasting the Forbidden Apple", for me it is the most interesting GPU talk.20:06
*** tplaten <tplaten!~isengaara@55d40ccb.access.ecotel.net> has quit IRC21:09
markosguys, I'll skip tonight's meeting, I'm very tired, I'll join tomorrow however, fwiw, I've just finished integrating mp3 benchmark with pypowersim, took less than an hour, dav1d has been bugging me for days...21:20
markosI expect mp3 to take a couple of days -fingers crossed21:20
markosanyway, ttyt21:20
lkcli'll be about 10 mins late this evening21:42
programmerjakemeeting in 16min21:44
octaviuslkcl, cesar, meeting22:09
*** choozy <choozy!~choozy@75-63-174-82.ftth.glasoperator.nl> has quit IRC22:29
programmerjakelkcl, i'll respond here since jitsi is acting up again, the vcd will likely not contain the required info such as simd alu sizes and wiring diagrams22:45
programmerjakelkcl: https://github.com/wavedrom iirc it has vcd input22:48
programmerjakebut that's not the style i'm planning on for the bigint demo, instead it'll be more like circuitjs's animation22:49
programmerjakewhere sim time is time in the video, not the x axis in some graph22:50
*** octavius_ <octavius_!~octavius@> has joined #libre-soc23:31
*** octavius <octavius!~octavius@> has quit IRC23:34

Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!