Wednesday, 2023-03-15

*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc00:03
*** gnucode <gnucode!~gnucode@user/jab> has quit IRC01:13
*** gnucode <gnucode!~gnucode@user/jab> has joined #libre-soc01:14
*** gnucode <gnucode!~gnucode@user/jab> has quit IRC01:51
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC02:19
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc02:54
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC03:05
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc03:32
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC03:58
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc04:00
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC04:11
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc04:14
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC04:26
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc05:00
lkclprogrammerjake, i've replied *very bluntly* to your comments on #1017.06:54
lkclplease refrain from making premature judgement calls06:54
lkclyou've now forced me into a position of having to back-track and undo damage that you're doing, under pressure, because you've jumped to a premature and mis-informed conclusion and stated that as fact06:55
lkclyou *have* to stop doing that06:56
lkclprefixing things with "imho" does not help if your "opinion" is not properly technically informed06:56
lkcli have said it many times and i am getting fed up of saying it: if i say NO you HAVE to ask yourself "why has he said NO"06:57
programmerjakewell, sorry, but you're likely wrong about power usage. we would need to measure to be sure.06:57
lkcli referred you to comment 5 and i expect you to LISTEN06:57
lkclthen YOU need to work out WHY i have said that06:57
lkclnot go "well you're just plain wrong therefore i can just blithely smash through everything and jump to a conclusion"06:58
programmerjaketo be blunt I think you said no because you don't want to make any changes now because it might slow down submission and you need money so want to finish tasks faster.06:58
lkclno, there are plenty of other RFCs and i do not wish to damage the chances of getting RED Semiconductor its funding06:58
programmerjakealso, I got tested and don't actually have Aspergers.06:58
lkclif i were to do anything that fucking stupid it would be picked up by VCs during their Due Diligence and we would not receive USD 25 to 40 million being sought06:59
programmerjakeand I am actually more technically informed than you apparently percieve me to be07:00
lkclcan you please watch Tim Forsyth's Larrabee talk, paying particular attention when he mentions about when they had to add predicate mask instructions07:02
programmerjakeI do think we need to get our priorities straight and realize binlog has few use cases so far so we need to work on crternlogi/ternlogi which are actually obviously useful in lots of places and not bother with arguing over how to implement crbinlog for now.07:05
programmerjakeyes, I'll rewatch it probably tomorrow07:05
lkcllet me think about it (but yes, i'm tending to agree)07:07
lkclre-work / mistakes-detected like in tends to suggest they're just not ready07:08
lkcland would delay the ternlog ones07:08
lkclbut crternlogi still needs adjustment down to 3in-1out (an overwrite)07:09
programmerjakeif we mush msk into the holes left by BF/BFA being 3-bits, we can mostly use TLI-form unmodified -- iirc you basically proposed that07:09
lkclbut also had to move 3 bits of TLI to other locations because the top-end needs to be dedicated to XO07:10
programmerjakeand then skooch the 8-bit immediate over to fill the other holes to extend XO07:10
programmerjakeoh, we typed the same thing at the same time :)07:10
lkclapologies i need to go afk.07:11
programmerjakewell, I think we just resolved all our recent arguments :)07:11
lkclthank you for listening jacob, sorry i'm... meh07:11
programmerjakety too07:12
programmerjaketurns out I forgot OE=1 for fcvttg when constructing the reduced instruction list so I added fcvttgo -- 5 instructions is still plenty small08:02
programmerjakefcvttg's pseudo-code is crazy long...08:41
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC09:24
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc09:25
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC09:55
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc11:50
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC12:32
sadoon[m]This is qtwebengine actually working almost flawlessly on debian 12 power913:10
sadoon[m]The only missing dependency for KDE apps :D13:11
sadoon[m]Only missing features again, as with gentoo: sandboxing, JIT, and with debian specifically ffmpeg might be broken for some formats because qtwebengine uses ancient chromium builds13:16
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc13:20
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC13:30
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc13:31
*** openpowerbot_ <openpowerbot_!> has quit IRC13:39
*** openpowerbot_ <openpowerbot_!> has joined #libre-soc13:40
*** openpowerbot_ <openpowerbot_!> has quit IRC13:50
*** openpowerbot_ <openpowerbot_!> has joined #libre-soc13:51
lkclsadoon[m], wha-hey!14:03
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC14:03
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc14:04
lkclprogrammerjake, it's perfectly fine to call it as a separate function that's proposed to be added to... err... Appendix A Book I, i think it is. joining the SINGLE_TO_DOUBLE() functions and other rounding14:04
lkcli think the "excessive" length is probably why none of the other FP-conversion-etc functions are listed in the actual pseudocode14:05
lkclcertainly there is a desire to keep the listing of instructions within a given section "compact" (i cannot say more on that)14:05
markossadoon[m], nice!14:07
sadoon[m]I'll create the relevant patches when I'm free and host it on my server14:14
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC14:33
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc14:34
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC14:42
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc14:43
lkcltoshywoshy, i'm putting together the "ongoing" grant, which is where implementation (and unit tests) occur, prior to OPF ISA WG submission.16:49
lkclwould you be interested in implementing the pseudocode for the CRweird set?
lkclit involves writing one of the markdown files;a=tree;f=openpower/isa;hb=HEAD16:51
lkcladding to the decoder csv files and the machine-readable version of Power ISA Section 1.6, and generally understanding what the hell is going on :)16:51
markoslkcl, on a related note, I'd be interested in the int butterfly instructions one, as I have exactly the code I would test it on16:54
lkclit's been done enough times that there's sort-of a guide - by following relevant diffs for many *many* instructions already previously added16:55
lkclmarkos, sure!16:55
markosI'd probably need your help in that though so we could do it together if you'd have time16:55
lkclyes of course16:55
lkclalthough it really should be blindingly obvious from so many patches, ah you already did fmvis, remember?16:56
lkcland i insisted you put the diff-patch commit-links for exactly this reason: to be able to refer to it again and see what was done16:56
lkclbut yes (toshywoshy as well) i'm happy to walk you through the process16:57
lkclsomeone really needs to finish the documentation as well
lkclwhich yes, all "paid"16:57
lkcllet me just put that on the list16:57
markosno, I mean what needs to be done in that regard, I remember I did fmvis, unless the process changed I don't expect many difficulties with the process itself :)16:59
markosis there a ticket for int butterfly instructions?16:59
lkclno - any change would be so absolutely massive and disruptive it's highly unlikely to even take place - ever17:01
lkcldon't know yet - looking for it17:01
lkclyes -  96217:01
lkclhm that's the *design* - not the implementation17:01
*** adi_ <adi_!> has quit IRC17:17
sadoon[m]markos you might be interested since you are running deb12
markosI was a while ago, now my talos is running headless17:27
toshywoshylkcl: yes, I can work on that17:33
toshywoshygiven that markos already has that, it might be good as a first one, for me to do it with markos and have you assist us17:34
markosthe int butterfly you mean? I agree it would be a great way for both of us to get acquainted with the internals17:35
toshywoshyyes, or any one instruction, this way we get to know more while it does not drain luke as much17:43
toshywoshyI also agree with your previous idea, that we should have internal get acquinten moments, in additional to the actual documentation17:43
markosoh you also meant the crweird as well? I have no objections, though I have no idea what this instruction does17:46
markosreg. the knowledge sharing I suggest we pick a convenient day/time and set up recurring events in the calendar17:46
toshywoshywell, we can see, I would like to start with one, where you and luke are working on, just to make it easier to start17:46
toshywoshyyes, that was the word I was looking for, correct17:47
markosI could start with showing and explaining the simpler algorithms first (eg libvpx variance) and moving on to more complicated (chacha20)17:53
toshywoshyyes, lets do that first18:05
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC19:20
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc19:21
lkclCR ops are not powerful enough.  they are one-bit-at-a-time.19:31
lkclCRweird ops do up to *four* bits at once. [in the scalar versions]19:32
lkcllike, "copy this CR *field* to this other CR *field*"19:32
lkcl"take a selection as specified by an immediate mask of any four of the CR *field* bits, then AND (or OR) them all together, and put the result into *this* bit of the CR"19:33
lkclwhich when combined with SVP64 Predication is stunningly powerful19:33
lkcltoshywoshy, btw what's up with the domain named ""?19:33
lkclit appears to have gone walkies19:34
lkclcrops - crand cror crxor - are just not adequate. doing one bit at a time transfer between GPR and CR fields?? meh19:36
markostoshywoshy, do you remember what's the type of RAM the tyan board takes?20:07
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC20:08
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc20:08
toshywoshymarkos: RDIMM-DDR3L 133321:03
toshywoshyit has 32 dimm slots, of which a minimum of 8 must be populated to start the machine21:04
markosmax capacity?21:05
markosI mean per dimm21:05
toshywoshyor 512GB per cpu, so if you have a dual socket 1TB21:06
markosit has 32 slots21:06
markosECC I guess21:07
toshywoshyyes yes, standard ecc registered memory21:08
markosI find different types 10600R/10600U/10600E21:09
toshywoshywell, if you get devel boards, they might only have 16 slots and if you get the 'early developer board' those only had 8 slots with 1 cpu21:09
toshywoshy10600R, the R stands for registered21:10
toshywoshywhile the U standard for unbuffered and the E, I think, stands for enhanced, it has an extra buffer21:10
markoswould these work: ?21:11
markosmaybe I would buy 2 sets for 128GB21:11
markosprice is cheap21:11

Generated by 2.17.1 by Marius Gedminas - find it at!