|
github.com/moarvm/moarvm | IRC logs at irclog.perlgeek.de/moarvm/today Set by moderator on 4 May 2018. |
|||
|
00:27
dalek joined,
p6lert joined,
Geth joined,
synopsebot joined
01:57
ilbot3 joined
|
|||
| moderator | github.com/moarvm/moarvm | IRC logs at irclog.perlgeek.de/moarvm/today | ||
| samcv | timotimo: odd... can you choose my name by chance? | 05:39 | |
|
10:25
Kaypie joined
10:26
Kaypie joined
11:32
Ven`` joined
11:33
Ven` joined
|
|||
| timotimo | i'll try and see what happens | 11:56 | |
| i chose samcv but it looks no different from when i chose timo | |||
| oh | 11:57 | ||
| my "projects" list now has a bunch of samcv/foo in them instead of mine, but the top right still says "timo" | |||
| seems like it allows me to change settings on your projects? | |||
| Geth | MoarVM: bdw++ created pull request #854: We can haz moar JIT-compiled opcodes |
12:21 | |
|
12:21
brrt joined
12:38
domidumont joined
12:46
domidumont joined
|
|||
| timotimo | i just figured out how best to fix the "profiler leaves frames skipping over inlines and doesn't realize it" issue | 13:18 | |
|
13:18
brrt joined
|
|||
| timotimo | we basically have a representation of the stack in the node structure of the profiler data | 13:19 | |
| so if i store a little flag in the nodes when something is entered, i can pop the right amount of nodes off if i know whether we're doing an inlined-frame-leave or a regularly-entered-leave, plus a bit of cleverness regarding OSR | 13:21 | ||
| MasterDuke | timotimo: nice | 13:27 | |
| timotimo | nedroid.com/2018/05/driven-to-greatness/ | 13:28 | |
| MasterDuke | ha | 13:30 | |
| timotimo | ok, looking at the code actually, we do already record the enter mode | 14:03 | |
| so i'll just have to use it when leaving i suppose | |||
| dogbert17 | o/ | 14:07 | |
| MasterDuke | timotimo: will this fix crashes, or make the profiler more accurate? | 14:10 | |
| timotimo | this ought to properly fix the case where call trees get nested far too deep | ||
| i.e. loops turning into deep graphs | 14:11 | ||
| MasterDuke | ah, cool | 14:12 | |
| timotimo | hmm. now how do i dig out a piece of code that triggered the problem ... | 14:19 | |
| well, that plan failed :) | 14:21 | ||
| yeah, that wasn't correct :\\ | 14:22 | ||
| Geth | MoarVM/pluggable-spesh: a6e45a7cc8 | (Jonathan Worthington)++ | src/spesh/facts.c Correct inverted condition |
14:39 | |
| dogbert17 notes people working hard on a saturday :) | 14:41 | ||
|
14:48
Kaiepi joined
|
|||
| Geth | MoarVM/pluggable-spesh: 13a5955e30 | (Jonathan Worthington)++ | 3 files Transform speshresolve into spesh slot lookup So long as there's a stable reuslt. Currently does not emit any guards, so will fail to deopt when it should. |
15:04 | |
| jnthn | Now I need to write some tests that actually need the deopt :) | 15:09 | |
| Well, next. Not right now. It is Saturday. :-) | 15:10 | ||
| dogbert17 suggests some kind of cooling liquid if the temps are high | 15:20 | ||
| jnthn | 21C, which isn't too bad :) | ||
| Tsssk, the deopt will be a bit more complicated | 15:21 | ||
| dogbert17 | wel then :) | ||
| *well | |||
| jnthn | Since we need to deopt to the prepargs | 15:22 | |
| timotimo | we can't just jump to before the prepargs and do it again? | ||
| jnthn | Yes | ||
| Actually, I was reading some paper recently where they pointed out that you can actually always deopt to the last non-side-effecting instruction | 15:23 | ||
| timotimo | that makes sense | ||
| jnthn | Which might let us avoid more instruction retention for deopt | ||
| timotimo | oh, right! | ||
| that could be worth a lot | 15:24 | ||
| jnthn | Indeed | ||
| Of course, it's deopt, so it'll be a mild headache to actually do it :-) | |||
| timotimo | sure | ||
| jnthn | Hmm, I wonder how we can efficiently do it | 15:25 | |
| I guess we can check instruction purity at CFG construction time, and maybe when we see a deopting instruction record the last non-pure or something | 15:26 | ||
|
15:44
brrt joined
|
|||
| jnthn | Anyway, ended up writing the failing test that needs the deopt :) | 15:44 | |
| Now I need to either deal with its unusual situation with regards to wanting to be at the start of the prep args, or maybe try and tackle the more general "last pure point" challenge. :) | 15:45 | ||
| jnthn goes to marinate chicken for tonight's curry and ponder that a bit :) | 15:49 | ||
|
15:51
zakharyas joined
|
|||
| timotimo | mhhh | 15:53 | |
|
17:11
SourceBaby joined
18:17
brrt joined
20:04
squashable6 joined
20:17
brrt joined
|
|||