github.com/moarvm/moarvm | IRC logs at irclog.perlgeek.de/moarvm/today
Set by moderator on 7 June 2013.
00:38 benabik joined 04:26 eternaleye joined 04:54 birdwindupbird joined 06:08 FROGGS joined 07:44 FROGGS joined 11:19 JimmyZ joined 12:07 cognominal joined 12:16 lizmat joined 12:37 FROGGS joined 13:25 tgt joined 13:48 colomon joined 14:06 tgt joined 14:39 tgt joined 15:03 tgt joined 15:30 kbenson1 joined 15:33 FROGGS joined 15:40 tgt joined 16:25 benabik joined
diakopter . 16:48
.tell JimmyZ I keep missing you by less than an hour :/
yoleaux diakopter: I'll pass your message to JimmyZ.
17:07 colomon joined 17:43 yoleaux joined 17:46 yoleaux joined
diakopter jnthn: did you see what I mentioned yesterday on #perl6 about a mmap-able disk cache of a loaded set of SCs/compunits? 17:46
jnthn no 17:47
we already mmap the bytecode
diakopter ok :) well there wasn't anything more to it than that (how I just summarized it)
I was thinking of ways to cut down even the deserialization time 17:48
might be worth doing
(might be truly Awesome; might not be worth it at all) 17:49
(afaik)
jnthn: any comments? 17:50
I'm anxious to regain momentum on moarvm 17:53
hopefully this weekend.
masak \\o/ 17:55
jnthn diakopter: It's a hard problem as it's an object graph which would need relocating in memory which means it can't just be mmap'd really, unless we stop using pointers or something... 17:58
diakopter jnthn: I was thinking having a loading mode where all allocation happens into one large region.. and then all the offsets of pointers are added to a list 18:12
jnthn: so loading it is to mmap and then fixup the pointers 18:13
so the whole region would be made of roots, effectively
even allocation of non-"managed" memory, I mean 18:14
jnthn: seems doable to me; what am I missing? :) 18:16
jnthn If you're going to change it to fix up pointers, you can't really map it in... 18:17
But moreover, a bunch of the pointers it needs are to outside the area. 18:18
Also, endianness...
diakopter oh, it's machine-specific
jnthn Pointer size
etc.
I don't want to go there for the time being.
diakopter forgot to mention; said that yesterday
jnthn I think the deserialization can be fast enough.
diakopter don't want to go where? machine-specific disk cache?
jnthn Especially if we get rid of the base-64
It's just the wrong prioritization to be looking at this right now. 18:19
We don't know how much of a problem startup time is yet...
diakopter not sure what you mean by "looking at this right now"
jnthn Already all the allocations go straight into gen2.
diakopter jnthn: are you saying it's stupid of me to bring it up? it's not worth a few minutes discussion? or not worth someone coding/investigating it? 18:20
jnthn diakopter: It's not a stupid idea, just one that I can see being difficult to realize and that we don't know how badly we need yet. 18:21
diakopter I didn't ask if you thought it was a stupid idea; I just didn't understand if I was being chewed out for bringing it up, asking you about it, or if you thought I was proposing someone [possibly me] coding it right now or what 18:22
jnthn I was trying to discourage coding it right now. I don't mind it being suggested. 18:23
(where "right now" means "before we get the hard blockers to run Rakudo out of the way")
diakopter ok. I saw no way of getting that from "I don't want to go there for the time being" - usually people say that to mean they don't want to discuss something right now, in my experience. Also "looking at this right now" - I was confused because that could definitely include talking about it here and not starting to code it, which I definitely hadn't suggested nor meant to come anywhere close to suggesting. 18:26
jnthn OK. Sorry for overreacting.
diakopter jnthn: I guess my thought was that deserialization time is far from insignificant (afaiui) on parrot and jvm, and I didn't see a reason it would naturally be that much better on moarvm 18:28
19:20 benabik joined 20:39 FROGGS joined 20:40 tgt joined 20:54 colomon joined 21:18 dalek joined, colomon_ joined 21:45 tomyan joined 21:51 FROGGS joined 22:30 benabik joined 23:05 benabik joined 23:27 kbenson1 left