|
github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm Set by AlexDaniel on 12 June 2018. |
|||
|
00:00
nebuchadnezzar left
00:08
nebuchadnezzar joined
06:12
robertle left
06:37
domidumont joined
07:19
domidumont left
|
|||
| nwc10 | good *, #moarvm | 07:26 | |
|
08:00
domidumont joined
08:29
zakharyas joined
11:18
brrt joined
|
|||
| brrt | \o #moarvm | 11:18 | |
| nwc10 | o/ | 11:19 | |
| brrt | ohai nwc10 | 11:24 | |
| brrt grumbles about how register allocation is a hateful, messy problem | 11:36 | ||
| If I had only sufficient time, I'd rip out the whole thing *again* | 11:37 | ||
| I mean, I only now find a potential bug... | 11:47 | ||
|
12:26
zakharyas left
12:52
lucasb joined
13:01
brrt left
14:10
zakharyas joined
16:01
lucasb left
16:33
domidumont left
16:34
Kaiepi joined
16:43
zakharyas left
17:33
robertle joined
17:35
domidumont joined
|
|||
| timotimo | oh damn today's the soft deadline for gsoc projects | 17:36 | |
| can i really mentor a project? | |||
|
17:56
Kaiepi left
|
|||
| MasterDuke | timotimo: i don't actually know what is required, but you've been very helpful with all my questions | 17:56 | |
|
17:56
Kaiepi joined
|
|||
| timotimo | but if i suggest a project, i'll be on the hook for whether it makes sense, is the proper size, is doable for a beginner, is useful to have ... | 17:58 | |
| MasterDuke | heh, i got around that by stating up front i wouldn't be able to mentor | 17:59 | |
|
18:02
brrt joined
18:12
brrt left
18:22
brrt joined
18:27
patrickb joined
|
|||
| Kaiepi | same MasterDuke | 18:33 | |
| timotimo, what about support for wide strings? | |||
| i tried implementing that a while ago but couldn't manage to work out how to fully do it | |||
| brrt | timotimo: what project would you like to mentor? | 18:34 | |
| luajit code is dense... | 18:42 | ||
| yay, I can compile nqp with floating point register support | 18:44 | ||
| timotimo | brrt: maybe integrating rakudo-moar with godot, maybe an interactive online tutorial a la go playground with things like generating images and showing them in the browser directly | 18:47 | |
|
18:48
patrickb left
|
|||
| brrt | hmm.... I think integrating with godot is kind of interesting | 18:50 | |
| anyway, you can always suggest it | |||
| Geth | MoarVM/jit-expr-float: 068130a68b | (Bart Wiegmans)++ | 10 files [JIT] Floating point support in the register allocator This is still fairly brittle and will not work with generic register requirements. But it seems to work decently so far, with NQP compiling succesfully. |
18:54 | |
| MoarVM/jit-expr-float: 15 commits pushed by (Bart Wiegmans)++ review: github.com/MoarVM/MoarVM/compare/0...f4092d945a |
|||
| brrt | brittle as hell, but it works, I think | ||
| jnthn | Fist make it work, I guess... :) | 18:56 | |
| Um, first :) | |||
| timotimo | ohai jnthn :) | 18:58 | |
| jnthn | o/ | 18:59 | |
| jnthn is sort of back | |||
| brrt | ohai jnthn | ||
| welcome back | |||
| jnthn | Rather tired, but that's nothing a good night's sleep won't fix :) | ||
| brrt | where'd you go? | ||
| jnthn | Jesenik and Ostrava, for a couple of nights each :) | 19:01 | |
| There was snow, which made me happy. And some nice food too. :) | 19:02 | ||
| brrt | :-) | 19:03 | |
| timotimo | \o/ | 19:04 | |
| you can also eat snow! | |||
|
19:08
brrt left
|
|||
| timotimo writes documentation for someone else's module | 19:10 | ||
|
19:11
patrickb joined
|
|||
| MasterDuke | hm, it goes into github.com/perl6/nqp/blob/master/s...r.nqp#L664 for postcircumfix again (and again...) | 19:15 | |
| is the self not the same maybe... | |||
| timotimo | oh | 19:20 | |
| i can imagine the cursor would be a different one all the time | |||
| but i don't know how the HOW implements cache_get | |||
| hopefully via the parse_shared | |||
| MasterDuke | github.com/perl6/nqp/blob/master/s...#L709-L712 | 19:24 | |
| timotimo: i don't see a parse_shared anywhere | 19:25 | ||
| timotimo | ah, it's ParseShared | 19:27 | |
| MasterDuke | but ClassHOW wouldn't know about ParseShared, right? | 19:30 | |
| timotimo | i'd expect GrammarHOW or so to be involved | 19:31 | |
| hm | |||
| doesn't exist, eh? | |||
| MasterDuke | doesn't look like it | 19:32 | |
| timotimo | aha | ||
| NQPClassHOW has it | |||
| MasterDuke | has what? | 19:33 | |
| timotimo | cache_get and cache_add | 19:35 | |
| MasterDuke | right | ||
| hm, so what happens if i instead use $!shared, which is a ParseShared as the cache? | 19:47 | ||
| timotimo | well, with cache_get, the self shouldn't matter, right? | 19:48 | |
| oh | 19:49 | ||
| maybe mixing in the operator actually causes the type to change, because mixins, you know? | |||
| that would mean that the cache actually doesn't get carried over | |||
| MasterDuke | oh, you're right, since it's self.HOW. but yes, invalidated because of the mixin | ||
| hm, create a new cache? or is there a way to transfer/copy the cache on mixin? | 19:50 | ||
| timotimo: heh, github.com/perl6/nqp/blob/master/s...#L747-L748 | 19:53 | ||
| timotimo | i'm not sure what the proper fix is :( | ||
| MasterDuke | i wonder if there's a way to tell what NFA don't need to be invalided and those could be copied to the new cache right away | 19:58 | |
|
20:05
domidumont left
20:06
patrickb left
|
|||
| MasterDuke | jnthn: are you back enough for a question about cached NFA invalidation because of type mixins? | 20:27 | |
| after all, isn't cache invalidation famously one of the easy problems of CS? | |||
| timotimo | totally | 20:28 | |
| MasterDuke | timotimo: btw, speaking of caching, didn't you have something in the works to make nqp::getenvhash() cache its rv? | 20:32 | |
| timotimo | it already does i believe | 20:33 | |
| what i had was constant folding values from the envhash at spesh time | 20:34 | ||
| MasterDuke | ah. becuase i added a bunch of `<...> if nqp::existskey(nqp::getenvhash(),'<my made up env var>')` for debugging in QRegex/Cursor.nqp, but it does noticeably slow down building rakudo | 20:36 | |
| timotimo | mhhh | ||
| the branch should still exist but i think it needs to be adapted to the new definition-usage API spesh got now | 20:37 | ||
| and it was a little fiddly anyway | |||
| MasterDuke | oh, the type name `Perl6::Grammar+{Postcircumfix}` is the same for all those cache misses | 20:38 | |
| nm, it does seem to change. now it's `Perl6::Grammar+{Postcircumfix}+{stop}` | |||
| and now `Perl6::Grammar+{Postcircumfix}+{stop}+{stop}` | 20:39 | ||
| but it does look like it tries to look up the same name once and fail | 20:40 | ||
| i.e., i see two failed attempts for the same name, then to failed attempts for the next version of the same name | |||
| that i can't explain | 20:41 | ||
|
20:50
brrt joined
|
|||
| jnthn | MasterDuke: Maybe sort of... :) | 20:55 | |
| MasterDuke: I do know that the invalidation there is extremely coarse | |||
| MasterDuke | i'm not sure where the problem is best solved, timotimo++ and i have sort of been working our way up/back, and that cache invalidation seems like it might be where we've ended up so far | 20:57 | |
| the backstory is that i was looking at rt.perl.org/Public/Bug/Display.html?id=128760 again | |||
| so went and read colabti.org/irclogger/irclogger_lo...07-23#l106 | |||
| and now i'm not sure if i've slipped into the second issue mentioned | 20:58 | ||
| or am still working on the first | 20:59 | ||
| jnthn | Solving the second needs assembling a directed graph of what uses what | ||
| So that when we produce a new token, we can then use the graph to work out what other things will be invalidated | 21:00 | ||
| The first problem is in theory easier | |||
| github.com/rakudo/rakudo/blob/mast....nqp#L4427 | 21:01 | ||
| I'm guessing this condition fails bogusly? | 21:02 | ||
| MasterDuke | i think so, we were looking there a day or two ago, but i don't remember exactly what we found | ||
| colabti.org/irclogger/irclogger_lo...-02-03#l90 | 21:03 | ||
| "MasterDuke ah. add_categorical is being called four times, but after the first it doesn't get past this: if nqp::can(self, $canname) { return 1; }" | 21:04 | ||
| jnthn | Hm, but if it doesn't get past that, do we still have the first problem? | ||
| MasterDuke | that sounds like it isn't failing | ||
| well, the example still behaves exactly the same as in the ticket | 21:05 | ||
| jnthn | Hmm | 21:12 | |
| As in, it takes a long time for each operator? | |||
| MasterDuke | yep | 21:13 | |
| about 2s per | 21:14 | ||
| jnthn | Odd. I wonder what it is doing per operator to cause that... | 21:17 | |
| MasterDuke | well, it runs this line github.com/perl6/nqp/blob/master/s...r.nqp#L665 29 times for `postcircumfix:sym<⦋ ⦌>` | 21:21 | |
| to be exact, !protoregex_nfa() is run 29 times for that (not 100% certain they're all from that line) | 21:22 | ||
| it doesn't seem to take any longer per operator the more there are, so i don't think it's redoing everything with each new one | 21:51 | ||
|
21:54
robertle left
|
|||
| jnthn | MasterDuke: Hm, I'm confused; you said "about 2s per" operator, but then "it doesn't seem to take any longer per operator" :) | 21:57 | |
| (I'm also tired and probably missing something) | |||
| bbiab | 21:58 | ||
| MasterDuke | any *extra* longer per operator. i.e., not .5s for the first, 1s for the second, 2s for the third, 4s for the fourth | 22:04 | |
| it doesn't *seem* like factorial or exponential growth in processing time with each added operator, just linear growth | 22:09 | ||
|
22:23
brrt left
|
|||
| Geth | MoarVM: ffontaine++ created pull request #1045: Configure.pl: fix cross-compilation |
22:26 | |
| jnthn | MasterDuke: Ah, I see | 22:44 | |
| MasterDuke: And is it per operator or per instance of the same operator with different multi candidates? | 22:45 | ||
| timotimo | our testing so far has been for &postcircumfix:<[ ]> multiple times in a row i believe | 22:50 | |
| MasterDuke | jnthn: same with a mix of type of operators and different terms (e.g., `multi postcircumfix:« ⦋ ⦌ »`, `multi postcircumfix:« | | »`, `multi infix:«qwerty»`, and `multi postfix:«asdf»`) | 23:09 | |
| also doesn't matter their their multi or not | 23:35 | ||
| jnthn | Yeah, in theory the nqp::can check should prevent the multi case repeatedly triggering the operator generation... | 23:38 | |
| MasterDuke | *if they're | 23:40 | |
| jnthn: any thoughts on how to figure out why that isn't happening? | 23:43 | ||
| jnthn | MasterDuke: Dunno, could we instrument where the mixin stuff takes place and dump a stack trace there? | 23:46 | |
| MasterDuke | jnthn: here? github.com/perl6/nqp/blob/master/s...W.nqp#L731 | ||
| jnthn | MasterDuke: Yeah | 23:48 | |
| There'll be loads of hits on named parameters too I guess, but maybe you could filter on the name of the type being mixed in to | |||
| MasterDuke | building rakudo now... | 23:52 | |
| huh, my breakpoint only hit once | 23:56 | ||
| `nqp::sech_n(1.0) if nqp::existskey(nqp::getenvhash(),'NQP_OPT_DEB') && self.name($obj) eq 'Perl6::Grammar';` | |||
| and gdb complained when trying to print a backtrace | |||
| jnthn | Missing or optimize-busted debug symbols? | 23:57 | |
| (You can recompile MoarVM with --debug=3 --optimize=0 if so without having to build Rakudo again) | 23:58 | ||
| MasterDuke | gist.github.com/MasterDuke17/07de1...5eea18983e | ||
| i pretty much always have moar built with --debug=3, but i don't touch --optimize | 23:59 | ||
| maybe cosh isn't a great thing to try and break on? | |||