02:46 colomon joined 03:10 colomon joined 03:21 colomon joined 05:51 Ven joined 06:50 geekosaur joined 06:51 geekosaur joined 08:09 Ven joined 08:32 konobi joined
konobi was just reading some papers and came across this... www.cs.utexas.edu/users/mckinley/p...a-2004.pdf 08:32
09:08 vendethiel joined 09:16 Ven joined 10:57 leont joined
jnthn konobi: Hm, interesting one. Thanks :) 11:42
jnthn hadn't seen it before
12:00 masak_ joined 12:01 [Coke]_ joined 12:12 JimmyZ joined 12:48 colomon joined 13:15 colomon joined 13:18 Ven joined 13:20 colomon joined
timotimo this age-based heuristic would have made the split "benchmark" do much less nursery collections, i'd think 13:31
hm, except that's probably something that ran 100% inside the C code, so we would have had to annotate the C code with the coalloc thing manually 13:43
14:28 domidumont joined 14:33 domidumont joined 14:35 brrt joined
jnthn For those who don't pick it up elsewhere: 6guts.wordpress.com/2016/04/30/ref...d-torture/ is about MoarVM :) 15:23
lizmat jnthn++ 15:41
jnthn: btw, why 13K, why not 4K for instance ? 16:00
jnthn lizmat: A prime to try and bust missing things that happen in patterns. 16:17
lizmat: Lots of stuff happens in powers of 2, so triggering GC at different points in such cycles is good.
lizmat ok, the blog post didn't make that clear (to me, at least :-) 16:18
jnthn lizmat: Besides that, it's mostly a trade-off. The lower you go, the longer it takes. But also if you go really low then you miss out on the nice "is it in fromspace" heuristics.
Because stuff ends up in gen2
mst I sort of went "ooh, a non-regular value, that was clever" 16:19
lizmat: the fact it was a prime was what clued me in, but then again I used to be rather fond of number theory, so I'm not expecting that to've helped everybody else
jnthn lizmat: Yeah, I didn't really go into why that number 16:20
lizmat eh, but 13K is not a prime :-)
jnthn Hah, true that :P
13 is though :)
lizmat m: say 13001.is-prime 16:22
camelia rakudo-moar d7698f: OUTPUT«True␤»
lizmat :-)
close :-)
jnthn ;)
I've got an 11500 byte nursery one running now, and I set it off 3 hours ago and it's still working on CORE.setting :)
If I automate this I'm inclined to have it pick numbers differently each time. 16:23
And note which ones it picked to allow reproduction.
lizmat yeah, sounds like a plan
specially like the comment: (In the future, we'll make this adaptive rather than a constant.)
jnthn :) 16:25
arnsholt Whee, new 6guts post! \o/
jnthn Seems we're getting hungry here...guess I'll make some dinner :) 16:26
lizmat same here, afk&
16:34 geekosaur joined 17:23 Ven joined 18:16 brrt joined
dalek MoarVM/reframe-jit: e1bcdc4 | brrt++ | src/core/ (4 files): 18:20
MoarVM/reframe-jit: Make frames identifiable using a sequence number
MoarVM/reframe-jit: Because frames can now be garbage-collected, they can be moved, and
MoarVM/reframe-jit: the identity check used in the JIT will no longer work. Thus, we'd
brrt freenode is not so lenient to dalek
but ehm
the last commit kind of fixes jit-on-reframe
18:21 dalek joined
brrt sorry dalek 18:21
timotimo "kind of" :) 18:23
brrt well, ok, it fixes it, as far as i know :-)
is ther e a rakudo branch for reframe? 18:24
timotimo i think there's one, yeah
brrt cool
aaargh :-( cannot invoke this object 18:25
possibly a JIT bug 18:26
hmmm 18:27
be back later 18:28
timotimo hm, that sounds like a place where a debug_name could help
timotimo greps for stuff
yup 18:29
dalek arVM: cbb6af9 | timotimo++ | src/6model/6model.c:
Default invoke handler shall output the debug name
arVM: 2f6c8c7 | timotimo++ | src/core/frame.c:
invalid register kind in getdynlex shall output the kind number
arVM: 9af346f | timotimo++ | src/6model/6model.c:
cannot find method on what object? debug_name knows!
arVM: 1ba147e | timotimo++ | src/core/frame.c:
cannot invoke this object will tell debug_name now.
arnsholt timotimo++ # debugging help! 18:48
timotimo i'm glad travis doesn't complain :P 19:13
19:28 Ven joined 21:14 zakharyas joined