|
github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm Set by AlexDaniel on 12 June 2018. |
|||
|
00:42
Ven`` joined
00:43
p6bannerbot sets mode: +v Ven``
02:34
Ven`` left
03:23
leont left
04:54
MasterDuke joined,
p6bannerbot sets mode: +v MasterDuke,
MasterDuke left,
MasterDuke joined,
herbert.freenode.net sets mode: +v MasterDuke,
p6bannerbot sets mode: +v MasterDuke,
Ven`` joined
04:55
p6bannerbot sets mode: +v Ven``
05:06
Ven`` left
06:20
lizmat left
06:23
MasterDuke left
06:47
lizmat joined,
p6bannerbot sets mode: +v lizmat
08:48
brrt joined
08:49
p6bannerbot sets mode: +v brrt
09:13
brrt left
09:17
zakharyas joined
09:18
p6bannerbot sets mode: +v zakharyas
09:32
domidumont joined
09:33
p6bannerbot sets mode: +v domidumont
10:02
brrt joined,
p6bannerbot sets mode: +v brrt
10:40
AlexDaniel left
10:41
AlexDaniel joined,
p6bannerbot sets mode: +v AlexDaniel
|
|||
| brrt | \o | 11:06 | |
| jnthn | o/ brrt | 11:11 | |
| Happy New Year :) | |||
| lizmat | jnthn brrt o/ | 11:12 | |
| timotimo | o/ | 11:15 | |
| the muldis object notation grammar may be a good benchmark for building big NFAs | 11:16 | ||
| as i just posted on #perl6, it spends a lot of time in mergesubrule, mergesubstates, and optimize | 11:17 | ||
| that said, the subs in question aren't necessarily very slow; about 9.6 us per call, it's just that they get called 800k times in this particular case | 11:19 | ||
| addedge is called 1m6k times | |||
| optimize on the other hand only gets called 77 times, but spends 40ms per entry | |||
| from the looks of the progress bar thingie, optimize is like a quarter or a fifth of total program run time | 11:20 | ||
| (this is not the qast optimizer, it's the nfa optimizer) | |||
| oh, one thing to point out is that it's at 30% time spent in GC | 11:24 | ||
| nine | addedge doesn't use any native types at all | 11:25 | |
| Happy new year btw. :) | 11:26 | ||
| timotimo | happy new year nine :) | ||
| or as people at the XXc3 tend to say: merry crisis and a happy new fear | |||
| allocations-wise, BOOTInt is the second place, with the top entries being: optimize (6.5m), addedge (1.6m), mergesubrule (0.8m), dentout (!) (0.8m), insert_bytecode (11k) | 11:27 | ||
| the number 1 allocated type is NQPArray, though: mergesubstates (4.5m), optimize (1.8m), mergesubrule (0.8m), addstate (0.8m), op (239k) | 11:28 | ||
| oops | |||
| that last one was supposed to be 29k | |||
| the cat wants me on the sofa :D | |||
| bbl | |||
| brrt | happy new year jnthn, nine, timotimo | 11:31 | |
| and lizmat of course :-) | |||
| lizmat | :-) | ||
| timotimo | happy new year brrt, jnthn, lizmat :) | 11:48 | |
| lizmat | timotimo o/ | 11:49 | |
| timotimo | oh, i neglected to continue looking at the allocations list | ||
| the next one in the list is BOOTNum with 6.2m total allocations. compare that to BOOTInt with 9.9m and NQPArray with 8.1m allocations total | 11:50 | ||
| it stands to reason that we'd want to eradicate almost all BOOTNum in compilation | |||
| allocators of BOOTNum are: mergesubstates (4.5m), optimize (1.6m), isa (23k), sanity_check (3k), capnames (1k), rxjump (980), unique (560), visit_children (360) | 11:52 | ||
| and a bunch more that are smaller | |||
| the GC tab shows a bunch of distinct "phases" when looking at percentages promoted: first a couple of runs with 15% promoted, then about 100 runs with 30% promoted (very stable), then 20 runs with 100% promoted (100% stable), 15 with ~50% promoted, a few noisy ones, then another 100 runs of 30% promoted, then 100% again, then ~50% again, the rest is noisier | 11:56 | ||
| oh wow | 12:00 | ||
| i didn't see that until i got the flame graph, but | |||
| mergesubstates and mergesubrule recurse *pretty* deep | |||
| about 50 frames between the two of them | 12:01 | ||
| nine: are you already looking at changes to addedge's code? | 12:03 | ||
| nine | timotimo: no, have to earn some money instead :) | 12:08 | |
| timotimo | damn :) | 12:09 | |
| jnthn | Lunch time for me :) | 12:11 | |
| timotimo sprinkled some native types all throughout NFA.nqp, hoping they are all correct %) | 12:50 | ||
| lizmat | timotimo: any idea why that wasn't done before ? | 12:52 | |
| was it because they didn't exist yet when that part of NQP was written ? | |||
| just OOC | 12:54 | ||
|
13:02
leont joined,
zakharyas left
13:03
p6bannerbot sets mode: +v leont
|
|||
| timotimo | not sure; that code is probably very old and not touched very often | 13:04 | |
| i wish it weren't as hard to see where nqp floats and where it ints | 13:05 | ||
|
13:07
lucasb joined,
p6bannerbot sets mode: +v lucasb
|
|||
| jnthn back for a bit | 13:16 | ||
|
13:45
MasterDuke joined,
p6bannerbot sets mode: +v MasterDuke,
MasterDuke left,
MasterDuke joined,
herbert.freenode.net sets mode: +v MasterDuke,
p6bannerbot sets mode: +v MasterDuke
14:13
zakharyas joined
14:14
p6bannerbot sets mode: +v zakharyas
14:33
brrt left
14:54
Ven`` joined
14:55
p6bannerbot sets mode: +v Ven``
15:39
Ven` joined,
p6bannerbot sets mode: +v Ven`
15:40
Ven` left
15:45
brrt joined,
p6bannerbot sets mode: +v brrt
15:49
brrt left
17:10
Ven`` left
17:26
lucasb left
17:32
leont left
17:37
brrt joined
17:38
p6bannerbot sets mode: +v brrt
17:50
domidumont left
17:51
domidumont joined
17:52
p6bannerbot sets mode: +v domidumont
17:59
domidumont left
18:10
leont joined
18:11
p6bannerbot sets mode: +v leont
18:35
zakharyas left
19:40
zakharyas joined
19:41
p6bannerbot sets mode: +v zakharyas
19:51
brrt left
19:59
domidumont joined
20:00
p6bannerbot sets mode: +v domidumont
20:04
MasterDuke left
20:07
kiwi_60 joined,
p6bannerbot sets mode: +v kiwi_60
20:22
domidumont left
|
|||
| dogbert11 | hah, so I've managed to repro the 'continuationinvoke expects an MVMContinuation' bug on my Linux system as well. So it's not only a RPi problem. | 21:07 | |
| the bad part is that it only shows up if the system is heavily loaded | 21:08 | ||
|
21:14
Voldenet joined,
Voldenet left,
Voldenet joined,
p6bannerbot sets mode: +v Voldenet
21:15
kiwi_60 left
|
|||
| dogbert11 | Ok, M#1029 | 21:20 | |
| synopsebot | M#1029 [open]: github.com/MoarVM/MoarVM/issues/1029 Flapping test t/spec/MISC/bug-coverage-stress-6.d.t | ||
|
21:20
zakharyas left
21:22
MasterDuke joined,
p6bannerbot sets mode: +v MasterDuke,
MasterDuke left,
MasterDuke joined,
herbert.freenode.net sets mode: +v MasterDuke,
p6bannerbot sets mode: +v MasterDuke
|
|||
| timotimo | sounds like you may want to try something that fiddles with thread scheduling | 21:25 | |
| rr has something for that, i believe | |||
| dogbert11 | the sad thing is that Ryzen CPU's and rr don't mix very well :( | ||
| timotimo | is that still an issue? :< | 21:26 | |
| does rr work on hack? on the whateverable server? | |||
| dogbert11 | dunno | 21:27 | |
| turning off spesh does not help | 21:29 | ||
| MasterDuke | pretty sure it does on the whateverable server, think i've used it there before | ||
| timotimo | does it actually have multiple threads? | 21:31 | |
| i think it's a crash related to the thread pool scheduler? | |||
| dogbert11 | what has multiple threads, the test? | 21:37 | |
|
21:37
leont left
|
|||
| timotimo | yeah, whatever process crashes | 21:38 | |
| dogbert11 | at least we can't blame this bug on jnthn's latest merge :) | 21:45 | |
| asan is also silent | 21:49 | ||
|
22:04
leont joined
22:05
p6bannerbot sets mode: +v leont
22:29
robertle left
|
|||
| jnthn | For once, I'm innocent :P | 22:41 | |
|
22:54
Ven`` joined,
p6bannerbot sets mode: +v Ven``
|
|||