|
00:32
AlexDani` joined
00:45
greppable6 joined
|
|||
| Geth | MoarVM: MasterDuke17++ created pull request #790: Add even more jit templates |
01:57 | |
|
02:58
ilbot3 joined
03:36
Voldenet joined
06:42
brrt joined
|
|||
| brrt | MasterDuke++ | 06:45 | |
| brrt notes that i have to implement some tiles before we can merge your patch | |||
| tile implementation has been a 'lazy' effort so far | |||
| Geth | MoarVM: 3055debefd | (Bart Wiegmans)++ | 3 files Add tiles for binary XOR and NOT We had them as nodes, but never as tiles, so any attempt to compile these would probably fail. |
07:02 | |
|
07:15
domidumont joined
07:18
travis-ci joined
|
|||
| travis-ci | MoarVM build errored. Bart Wiegmans 'Add tiles for binary XOR and NOT | 07:18 | |
| travis-ci.org/MoarVM/MoarVM/builds/335031188 github.com/MoarVM/MoarVM/compare/e...55debefd40 | |||
|
07:18
travis-ci left
07:26
brrt joined
08:02
brrt joined
08:14
domidumont joined
|
|||
| timotimo | how can these tiles be missing and still the branch passes spectest i wonder | 08:22 | |
| brrt | these templates are not ever compiled, then | ||
| timotimo | thought so :\ | 08:23 | |
| hey while you're here, can we remove PHI from the "made tree out of [a, b, c, d]" output? | |||
| nwc10 | NQP build fails for me | ||
| brrt | where? | 08:24 | |
| nwc10 | make: *** [gen/moar/stage2/nqpmo.moarvm] Error 1 | ||
| brrt | timotimo: can, i might miss them at some point, but whatever | ||
| nwc10 | backtrace started: | ||
| brrt | :-) | ||
| nwc10 | Expected native int argument, but got num | ||
| at gen/moar/stage1/QRegex.nqp:1624 (gen/moar/stage1/QRegex.moarvm:!cursor_pa | |||
| ss) | |||
| brrt | hmmm | 08:25 | |
| timotimo | brrt: but does the exprjit even do anything with them? | ||
| nwc10: i've seen that already the last few days, hum. | |||
| nwc10: can you try the different env vars we have to see what it's sensitive to? | 08:26 | ||
| also, can we build a script or something that does the env var testing for us? | |||
| nwc10 | I was just testing that | ||
| having just wondered about it | |||
| timotimo | thanks :) | ||
| brrt: i'll remove it :) | |||
| brrt | timotimo: sure, go ahead | 08:27 | |
| it doesn't do anything with them yet, but i plan to use them for the multiple-block building machinery i'm thinking of | 08:28 | ||
| Geth | MoarVM/master: 15 commits pushed by MasterDuke17++, (Bart Wiegmans)++ review: github.com/MoarVM/MoarVM/compare/3...c437b79585 |
||
| brrt | (i'm merging the commit, btw) | ||
| timotimo | ah | 08:29 | |
| so maybe only skip them at the very beginning? | |||
| brrt | they should always be at the beginning :-) | 08:30 | |
| timotimo | not if you merge multiple blocks? | ||
| brrt | but they're not useful now, so yeah, drop'm | ||
| oh, that's right, yes, but then this format is not going to work anyway, will it | 08:31 | ||
| Geth | MoarVM: 8f194be2bc | (Timo Paulssen)++ | 2 files better error in boxed ref for repr_id mismatch |
08:32 | |
| MoarVM: 783a4f07c7 | (Timo Paulssen)++ | src/jit/expr.c don't output PHI in expr jit log |
|||
| nwc10 | timotimo: yep, export MVM_SPESH_NODELAY=1 | 08:34 | |
| NQP build fails for me with that, passes without | |||
| but that was with MoarVM at 3055debefd4019f3d74f63e0e1ee8bd1d93012d8 | |||
| will try with new master | |||
| timotimo | i'll be mostly afk for the rest of the day :| | 08:35 | |
| nwc10 | have fun. | ||
| (I hope) | |||
| timotimo | probably not ;( | 08:37 | |
| nwc10 | oh :-( | ||
| still goes boom on new master | |||
| will try to bisect what caused it | |||
| but I might have a meeting soon... | |||
| brrt | is that on linux? | 08:38 | |
| nwc10 | yes | 08:42 | |
| brrt | hmm, i'll have to spin up a VM then | 08:43 | |
|
08:44
travis-ci joined
|
|||
| travis-ci | MoarVM build errored. Bart Wiegmans 'Merge pull request #790 from MasterDuke17/add_even_more_jit_templates | 08:44 | |
| travis-ci.org/MoarVM/MoarVM/builds/335051013 github.com/MoarVM/MoarVM/compare/3...37b7958559 | |||
|
08:44
travis-ci left
|
|||
| timotimo | t/nqp/049-regex-interpolation.t ........ 1/37 Expected native int argument, but got num | 08:44 | |
| so i can reproduce | |||
| but afk etc | 08:47 | ||
|
08:57
travis-ci joined
|
|||
| travis-ci | MoarVM build failed. Timo Paulssen 'don't output PHI in expr jit log' | 08:57 | |
| travis-ci.org/MoarVM/MoarVM/builds/335052193 github.com/MoarVM/MoarVM/compare/c...3a4f07c7dd | |||
|
08:57
travis-ci left
|
|||
| nwc10 | bisect says: 8210fb733ce14149c95f82a1f2a5e6c525149497 is the first bad commit | 09:17 | |
| Add JIT template for box_i | |||
| brrt | hmmm | ||
| looks decently normal to me | 09:18 | ||
| nwc10 | it's very strange how it also needs MVM_SPESH_NODELAY=1 | 09:24 | |
| or maybe that's not strange | |||
| brrt | what about MVM_SPESH_BLOCKING | 09:29 | |
| can you make it happen with blocking and without nodelay | |||
| nwc10 | will do that next | 09:31 | |
| without any env vars, ./nqp-m t/nqp/049-regex-interpolation.t | |||
| Expected native int argument, but got num | |||
| so it's a bug :-) | |||
|
09:33
Geth joined
|
|||
| brrt | hmmm | 09:34 | |
| it sucks, but that works for me | 09:35 | ||
| nwc10 | nqp builds and tests with MVM_SPESH_BLOCKING=1 | 09:49 | |
| brrt | :-( | 09:50 | |
| and it works for me with MVM_SPESH_NODELAY=1 even | 09:51 | ||
| jnthn | MVM_SPESH_NODELAY=1 is most usefully used with MVM_SPESH_BLOCKING=1 | 10:06 | |
| brrt | still doesn't fail on my machine :-( | 10:10 | |
| AlexDaniel | squashable6: next | 10:22 | |
| squashable6 | AlexDaniel, ⚠🍕 Next SQUASHathon in 2 days and ≈23 hours (2018-02-03 UTC-12⌁UTC+14). See github.com/rakudo/rakudo/wiki/Mont...Squash-Day | ||
|
10:37
zakharyas joined
10:38
stmuk joined
10:56
SourceBaby joined
11:16
unicodable6 joined,
reportable6 joined,
bisectable6 joined
11:21
dogbert17 joined
|
|||
| MasterDuke | hm. box_i had problems at first, but when i rebased to master (after nine++'s most recent JIT related commits) it was fine | 11:38 | |
| brrt | well, turns out some are still seeing errors, but not consisently unfortunately | 11:39 | |
| MasterDuke | ugh. i'll see if i can force and errors here | 11:40 | |
| dogbert17 stumbles upon a SEGV when running 'MVM_SPESH_NODELAY=1 MVM_SPESH_BLOCKING=1 ./perl6 t/spec/S02-types/num.rakudo.moar' | 12:06 | ||
| #0 0x00007ffff76da6f5 in MVM_multi_cache_find_spesh (tc=0x6760c0, cache_obj=0x42ccb50, arg_info=0x7ffff6372330, type_tuple=0x7ffff0e2b9a0) at src/6model/reprs/MVMMultiCache.c:500 | 12:09 | ||
| #1 0x00007ffff771fc1b in optimize_call (tc=0x6760c0, g=0x7ffff0950130, bb=0x7ffff0e0b3e0, ins=0x7ffff1087c68, p=0x7ffff0d04f90, callee_idx=1, arg_info=0x7ffff6372330) at src/spesh/optimize.c:1513 | |||
| #2 0x00007ffff772169c in optimize_bb_switch (tc=0x6760c0, g=0x7ffff0950130, bb=0x7ffff0e0b3e0, p=0x7ffff0d04f90) at src/spesh/optimize.c:2038 | |||
| gist for any interested parties: gist.github.com/dogbert17/1f127e24...abc39b65e8 | 12:18 | ||
| jnthn: I'll report the SEGV above unless it's a known problem | 14:24 | ||
|
14:26
evalable6 joined
|
|||
| jnthn | dogbert17: Please do | 14:26 | |
|
14:26
committable6 joined,
bisectable6 joined
|
|||
| dogbert17 | will do, one advantage, at least on my system, is that it is 100% reproducible | 14:30 | |
|
14:32
brrt joined
|
|||
| brrt | damnit that still doesn't break for me | 14:56 | |
| MasterDuke | t/nqp/049-regex-interpolation.t ? haven't gotten it to error for me either | 14:57 | |
|
15:05
zakharyas joined
|
|||
| dogbert17 | added github.com/MoarVM/MoarVM/issues/791 | 15:10 | |
|
15:24
bisectable6 joined
|
|||
| AlexDaniel | dogbert17: it's the first time I use bots to bisect something like this, so maybe you should double-check… | 15:37 | |
| also, if that's the one, then you can probably bisect moar with that | |||
| although maybe that won't tell much | 15:39 | ||
| MasterDuke | can the ops in rakudo (e.g., p6bool) be jitted? | 15:42 | |
|
15:46
Kaiepi joined
15:48
bisectable6 joined
16:05
brrt joined
|
|||
| brrt | some of them, yes, but probably not by the expr JIT | 16:06 | |
| not yet at any rate | |||
|
16:07
zakharyas joined
|
|||
| MasterDuke | ok, thanks | 16:07 | |
|
16:11
zakharyas joined
|
|||
| dogbert17 | AlexDaniel: cool, I'll take a look at those commits | 16:14 | |
| AlexDaniel: not that it's any proof but one of the listed commits (388a769) matches the row where the crash occured | 16:40 | ||
| AlexDaniel | :) | 16:41 | |
| well, I just got bisectable to point to the same rakudo commit | |||
| so it's definitely that bump | |||
| dogbert17 | do you want to add that info to the issue? | 16:42 | |
| AlexDaniel | sorry, what info exactly? :) | ||
| your comment is probably useful, yes | 16:43 | ||
| dogbert17 | I added a note containing some pure speculation | 16:48 | |
| I'm unable to repro the problem that nwc10 reported though :-( | 16:50 | ||
| AlexDaniel | which problem is that? | 16:51 | |
| “NQP build fails for me” | 16:52 | ||
| yea, that's going to be hard to reproduce :D | |||
|
17:13
domidumont joined
17:24
zakharyas1 joined
18:26
nine joined
18:27
camelia joined
19:24
stmuk_ joined
|
|||
| MasterDuke | hm. `(template: box_s! (callv (^func &MVM_box_str) (arglist (carg (tc) ptr) (carg $1 ptr) (carg $2 ptr) (carg $0 ptr))))` breaks the rakudo build | 19:43 | |
| Cannot look up attributes in a VMNull type object at src/Perl6/Grammar.nqp:4016 (blib/Perl6/Grammar.moarvm:termish) | 19:44 | ||
| #0 MVM_exception_throw_adhoc (tc=0x555555758c40, messageFormat=0x7ffff776b160 "Cannot look up attributes in a %s type object") at src/core/exceptions.c:861 | 19:45 | ||
| #1 0x00007ffff7664120 in MVM_interp_run (tc=tc@entry=0x555555758c40, initial_invoke=0x0, invoke_data=0x5555557c59a0) at src/core/interp.c:2064 | |||
| Geth | MoarVM: MasterDuke17++ created pull request #792: Yet more jit templates |
20:00 | |
|
20:17
bart__ joined
|
|||
| bart__ notes that both box_i and box_s allocate memory | 20:18 | ||
| brrt | and hence, can cause GC | 20:19 | |
| MasterDuke | should they be :invokish? | 20:24 | |
|
20:27
brrt joined
20:29
zakharyas joined
20:34
zakharyas joined
20:38
brrt joined
|
|||
| brrt | no. but there's a mechanism in there that should stop this from being a problem, and it might not be working | 20:38 | |
| MasterDuke | ah | 20:46 | |
|
21:04
Kaiepi joined
21:15
zakharyas joined
22:14
lizmat joined
23:53
lizmat joined
|
|||