github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
timotimo pamplemousse: github.com/timo/moarperf/blob/mast...o_stuff.sh is rakudo + application 00:19
github.com/timo/rakudo-appimage - this is rakudo only
01:28 leont left 01:41 Altai-man joined 01:44 sena_kun left 01:50 pamplemousse left
nwc10 good *, #moarvm 05:17
05:42 sena_kun joined 05:43 Altai-man left 07:25 eater joined 07:36 domidumont joined
MasterDuke oh, if zoffix++ is back maybe he can help me adjust MVM_bigint_div_num for gmp, he did the last big changes to it 07:37
07:53 brrt joined 07:57 zakharyas joined
brrt \o 07:59
nwc10 o/
brrt .tell pamplemousse no jit news today
tellable6 brrt, I'll pass your message to pamplemousse
brrt sadly
MasterDuke but if you get an arm laptop...jit on arm in the future? 08:00
nwc10 I guess that it depends how fun it is 08:01
I suspect that there are more ARM chips in this room than anything else, but I don't seem to think that writing a JIT is fun 08:02
actually, I'll be more confident onthis - there are almost certainly more 64 bit ARM chips int his room than 64 bit anything else, as the printer won't have a 64 bit CPU 08:03
although I have no idea whether the SSD control is actually something ARM, so ARM might still win
brrt hmm, I think I'm personally still surrounded by arm32
and getting an arm64 (apple arm) laptop is rather a big 'if' :-) 08:04
nwc10 8Gb Pis are still cheaper
but "only" 8Gb
There was a lovely statement (but maybe a bit too much context needed for SLS): If SpaceX gets Starship to orbit prior to SLS launching, well.. let's just say I am starting to think "Elon Time" is not so much a joke as it is a temporal distortion field. Everything is behind schedule until suddenly it isn't and we're all staring slack-jawed at how quickly they're moving. 08:05
I'm wondering if ARM in datacenters will be something like this
brrt that actually makes sense 08:06
engineering is mostly removing faults
you start with a general idea, you find 100 faults
you can sort of parallelize, but if you set 1000 people on it, you'll only get the first 5 fixes slightly faster 08:07
then the next 5 fixes. and then the next, and then the next
nwc10 I wonder if there will be a ("shock") tipping point, where it turns out that there are more AWS instances running on ARM than x86, because it's cheaper
some point in the next few years
brrt we might see that
nwc10 or Google announce something similar about their own datacentres 08:08
or Facebook
brrt at the start, only a select few will believe that something might ever work (with 100 failures); at 50 failures, more people will believe it, and more people will work on it;
probably at 5 or 10 remaining faults it is shipped 08:09
nwc10 heck, actually Apple. We *know* that Apple have seriously fast ARM CPUs
and Apple are the most secretive
brrt Hmmm.
There's two opposing dynamics though
nwc10 "a lot of existing machines" ? 08:10
MasterDuke do MS's new surface devices run on arm?
brrt On one hand, apple fans will probably swiftly adopt the new platform
nwc10 not everything ports nicely
brrt yes, but there are also a lot of existing nokia's still
so that's not the point.
the point is that apple consumers tend to hold fast to the low depreciation rates of apple computers
i.e. everybody who's bought an apple thinks they can sell it for 60% of new value a few years down the road 08:11
and I'm not sure whether that works together well. 08:12
nwc10 brrt: I was really meaning servers in datacentres. Not consumer devices
brrt no, datacentres I agree 08:14
probably because for datacentres opex is on a similar scale to capex, and for consumer devices that's not true
on the other hand, 'will apple consumers take a financial hit in order to get the newest gadget' - I think the answer to that is evident :-D 08:19
08:49 squashable6 left 08:52 squashable6 joined 09:41 Altai-man joined 09:43 sena_kun left 09:48 MasterDuke left 10:00 MasterDuke joined 10:13 MasterDuke left 10:35 leont joined
lizmat anybody has a problem with thundergnat's bump of Unicode ? 10:40
11:03 zakharyas left, brrt left 11:59 brrt joined 12:05 thundergnat joined
thundergnat lizmat: I would feel better if samcv had a chance to look it over to make sure I didn't do something completely boneheaded. It compiles, Rakudu built on top complies, passes sanity and spec tests, but I really don't know if it is 100% correct. 12:08
lizmat understood...
thundergnat It is almost certainly going to need a closer look at the grapheme break rules. 12:09
12:10 pamplemousse joined
pamplemousse o/ 12:12
tellable6 2020-10-02T07:59:49Z #moarvm <brrt> pamplemousse no jit news today
brrt \o
I have an oldish JIT devirt branch that I could resurrect somewhat safely 12:13
12:13 thundergnat left
pamplemousse Somewhat safely? 12:27
12:32 brrt left 12:50 MasterDuke joined 12:55 pamplemousse left 13:42 sena_kun joined 13:43 Altai-man left 13:51 zakharyas joined 14:48 zakharyas1 joined 14:51 zakharyas left 14:52 zakharyas1 left 14:55 zakharyas1 joined
MasterDuke ugh. thought i had a fix for MVM_mp_get_double_shift, but it doesn't work for everything. one version has 13 fails in t/spec/S32-num/rat.t, the other has 2 14:57
and differing number of fails in t/spec/S02-types/num.rakudo.moar 14:59
15:19 pamplemousse joined 15:24 zakharyas1 left
MasterDuke github.com/MoarVM/MoarVM/compare/m...234758e807 if anyone feels like looking at what i have now 15:27
15:39 domidumont left 15:44 zakharyas joined 16:36 pamplemousse left, pamplemousse joined 16:53 pamplemousse left 17:08 pamplemousse joined 17:41 Altai-man joined 17:44 sena_kun left 18:46 rypervenche left
MasterDuke or maybe i switch back to using gmp floats like i first tried, but with massively increased precision... 18:47
18:54 rypervenche joined
timotimo MasterDuke: do you have outputs of the numbers involved? like all the limbs and the resulting double at every step? 18:54
MasterDuke mostly. not in a very organized way 18:59
fyi, requesting 633 bits of precision for the gmp floats didn't improve anything 19:00
timotimo improve in what way?
MasterDuke still fails, i think the doubles generated with the gmp floats was the same as when i just used their default precision
timotimo wanna independently verify the results with per-hand math? :\ 19:03
MasterDuke not really
timotimo yeah
MasterDuke these are large numbers 19:04
timotimo maybe with like wolfram alpha or maxima 19:05
neil cicierega's newest album is already incredible, having only listened to three tracks of it 19:07
MasterDuke gist.github.com/MasterDuke17/b2258...273017ca3d ther's an example 19:12
gist updated 19:20
19:54 raku-bridge1 joined, raku-bridge1 left, raku-bridge1 joined 19:56 Kaiepi joined, raku-bridge left 19:57 raku-bridge1 is now known as raku-bridge 20:27 zakharyas left
timotimo m: say 1152921504606846976.base(16) 20:36
camelia 1000000000000000
MasterDuke gist updated 20:39
21:02 Altai-man left 21:08 pamplemousse left 21:21 Altai-man joined 21:25 Altai-man left 21:38 MasterDuke left
moon-child got this from a long-running raku process 'MoarVM panic: Internal error: invalid thread ID 25 in GC work pass' 21:45
can't repro reliably, though, and no corefile. Any setup I can do to get a useful dump if it happens again?
22:51 pamplemousse joined 23:11 pamplemousse left