IRCloggy #perl6-dev 2019-02-14

Logs Search ←Prev date Next date→ Channels Documentation

Provider of IRC logs since 2005.
WARNING: As Freenode became unjoinable and lost all warnings in topics, we cannot log channels on Freenode anymore.

2019-02-14

tyil left00:33
tyil joined00:39
ExtraCrispy left01:40
ExtraCrispy joined01:43
MasterDuke left01:56
MasterDuke joined01:56
MasterDuke left01:57
MasterDuke joined01:57
ExtraCrispy left02:45
vendethiel- joined04:12
vendethiel- left04:21
robertle left05:45
ufobat_ joined06:14
releasable6 Next release in ≈2 days and ≈11 hours. 4 blockers. Please log your changes in the ChangeLog: https://github.com/rakudo/rakudo/wiki/ChangeLog-Draft07:00
Tux__ joined07:11
|Tux| left07:11
tyil left08:26
tyilanmenyn joined08:26
tyilanmenyntyil08:39
llfourn_ joined09:27
llfourn left09:29
AlexDaniel left13:01
Ven`` left13:08
timotimo adding native subs to the file where i'm trying to "enhance" my enums makes the precomp explode13:23
i should try to golf it13:24
lucasb joined13:53
ufobat_ left15:45
robertle joined17:13
AlexDaniel joined18:18
japhb Quiet day ...18:34
jmerelo joined18:34
jmerelo hi18:34
In a couple of weeks we have a bug squash day, and it will be devoted to Rakudo/MoarVM/NQP this time around: https://github.com/rakudo/rakudo/wiki/Monthly-Bug-Squash-Day18:35
That implies there will be some flying PRs. Also this has not been done for some time, and the guide could be a bit obsolete: https://github.com/rakudo/rakudo/wiki/Rakudo-SQUASHathon-Guide18:36
If you feel it's worth the while to go ahead with this, there's going to be enough people around to help, and someone can take care of updating the guide18:36
vrurg AlexDaniel`: as we spoke last time, R#2694. Hope jnthn would give his word too.18:52
synopsebot R#2694 [open] : https://github.com/rakudo/rakudo/issues/2694 Role consumption by multiple roles 18:52
lucasb left19:02
jmerelo Er, hi again.19:27
Any reactions on the Squashathon I referred to above?19:27
[TuxCM] Rakudo version 2018.12-302-g121ca5fda - MoarVM version 2018.12-105-gc0435e55a19:38
csv-test-xs-20 0.430 - 0.43119:38
csv-ip5xs 0.744 - 0.76119:38
test-t --race 0.880 - 0.89619:38
test-t 1.945 - 1.99819:38
csv-ip5xs-20 6.243 - 6.35319:38
test 7.245 - 7.92519:38
test-t-20 --race 10.429 - 10.48319:38
csv-parser 22.441 - 23.14019:38
test-t-20 31.732 - 32.84019:38
vrurg jmerelo: I would participate, but I too much of a "new in town". Sorry.19:43
jmerelo left20:07
dct joined20:08
AlexDaniel left20:26
dct left20:28
Ven`` joined21:09
dogbert17 joined21:14
dogbert17 rather silent here today21:16
have a short program which has regressed quite a bit speedwise lately21:17
with 2018.10 I have according to time a runtime of real 0m4.696s but on head it's real 0m9.144s21:19
timotimo could it per chance be that it gets that new optimization i made but that disabled the jit for an important frame?21:21
dogbert17 using the profiler I see, among other things: 2018.10 (#gc's 63) while on head the number is 37621:21
on 2018.10 2_000_000 scalars are allocated, on head the number is 29_000_00021:22
timotimo: how do we check that?21:23
timotimo: here's the code, do you see anything fishy? https://gist.github.com/dogbert17/7423640a8bc4d5b9dd7c960295154ee921:25
timotimo looking at the spesh log and jit bails would give that answer21:25
but the number of gc runs is interesting. could be it's doing more boxing or lexicalrefs and such21:26
btw, are you on head, or do you have postrelease-opts?21:26
dogbert17 lemme check21:26
master, last change by you - CStruct bindattr_i Opt Shall Release Its Temp Register21:27
timotimo oh btw try use v6.d21:27
or is that already active because of v6?21:27
dogbert17 good question, anyway there's no difference21:29
looking at scalar allocs on 2018.10 almost all are allocated in SETTING::src/core/SlippyIterator.pm6:4921:29
on head the majority 25_000_000 are allocated by SETTING::src/core/native_array.pm6:616, i.e. ASSIGN-POS21:30
ASSIGN-POS doesn't even show up in the 2018.10 profile21:31
very strange, guess I should bisect it21:33
timotimo is scalar the highest allocated type?21:33
dogbert17 yes21:34
number 2, in both cases, is IntPosRef with 2_000_001 allocs21:35
sorting routines on exclusive time also show differences21:37
i.e. there are different routines on top21:38
timotimo aye21:38
when do i put in the feature that generates a nice copy-paste-able summary for all the tables ...21:38
dogbert17 :)21:39
I see that on head push_all is not jitted but it is on 2018.1021:40
timotimo oh, interesting!21:40
wanna check it out in the spesh log?21:40
dogbert17 i.e. SETTING::src/core/Any-iterable-methods.pm6:40521:40
sure21:40
the spesh log file is humungous21:42
what should I look for, BAILs?21:42
timotimo they no longer look quite like that21:49
i think "completely bailed"?21:49
dogbert17 there's one, both in 2018.10 and head - # JIT: bailed completely because of <ctxouter>21:50
I'll try some bisecting22:06
timotimo: seems to be https://github.com/rakudo/rakudo/commit/ee2238f731f548861e26d745ea1be2ba22d24f4e22:28
Ven`` left23:19
dct joined23:55

Logs Search ←Prev date Next date→ Channels Documentation