00:07 TimToady joined 01:59 pyrimidine joined 02:48 ilbot3 joined 03:21 pyrimidine joined 04:51 TimToady joined 04:57 pyrimidine joined 05:30 pyrimidine joined 06:08 pyrimidine joined 06:11 pyrimidine joined 06:19 domidumont joined 06:26 domidumont joined 07:05 domidumont joined 07:30 pyrimidine joined 07:37 pyrimidine joined 07:51 pyrimidine joined 08:25 zakharyas joined 08:45 pyrimidine joined 08:57 pyrimidine joined 09:55 pyrimidine joined 11:16 pyrimidine joined 11:30 pyrimidine joined 11:35 yoleaux2 joined 11:37 yoleaux2 joined 11:46 brrt joined
brrt \o 11:52
lizmat brrt o/ 11:57
brrt \o lizmat 11:58
I'm preparing my blog post for the 11th...
it's about how the new JIT will allow (relatively) easy extension points 11:59
but, i'm not sure about the 'level' I should keep it?
lizmat brrt: good question, 12:00
perhaps you could relate it to the original application? news.perlfoundation.org/2015/04/per...ation.html 12:01
brrt i would be terribly embarassed by the original schedule, though 12:02
but, yes
nine brrt: speaking of... do we approach the time when others can join the JIT fun? 12:05
brrt hmm... let me be clear about the deliverables before that
it needs the integration of the linear scan register allocator 12:06
which needs the implementation of a generalised register specification per tile, and the remoal of tile templates
those are... comparatively simple things to do, but they have the potential to break things
lizmat perhaps people can help by writing tests ? 12:07
brrt it also needs the implementation of function-calls in the new register allocator (given that most arguments to most functions are stashed in registers)
actually... that is something that's been on my mind as well. 12:08
I'd like to have a script that takes a given expression tree, compiles the code, executes it, and compares the result against a given value
this is not conceptually hard
it is a bit of hacking, though 12:09
but if we had that, the whole JIT could be tested much more easily and thoroughly than can be done now 12:10
once the new register allocator is merged, I can start working on a way to have REPRs insert specialized templates. none of that is conceptually difficult 12:11
in fact, I could start on that now, but it doesn't make much sense to do so, as the subset of things we can compile correctly is still quite small
nine It appears to be that even-moar-jit is a bit outdated with regards to master and needs a bit of a rebase. 12:19
brrt yes, it is 12:20
well, I typically just merge master explicitly
dalek arVM/even-moar-jit: b205d51 | brrt++ | / (3 files):
Register assignment should also cover synthetics

Synthetics ought to follow the convention of values, in that value 0 is always a definition, and value 1 is always a use.		 12:30
Heuristic branch merge: pushed 59 commits to MoarVM/even-moar-jit by bdw 12:34
brrt and, there we go :-) updated
nine Generating src/gen/config.h ............................ FAIL unknown configuration key 'jit_arch' 12:39
brrt hey, hmmm 12:40
13:18 pyrimidine joined
dalek arVM/even-moar-jit: 78d7088 | brrt++ | Configure.pl:
Fix broken merge on Configure.pl

We need some extra definitions in the even-moar-jit branch, so let's make sure they are added.		 13:57
brrt nine++ for seeing it :-)
nine brrt++ for the fix! 13:59
The branch still doesn't seem to be up to date however? Noticable through the outdated VERSION
brrt really? hmmm 14:00
that's weird 14:01
i thought VERSION was autogenerated
nine git log VERSION suggests otherwise
brrt aha
damnit, I know what I've done wrong, in all probability 14:02
nine Also when I git merge master, it does do quite a lot and generates conflicts
brrt unuh
uhuh
should've merged with origin/master
nine Oh your local master was out of date? 14:04
Seems like I can fix those merge conflicts
brrt no worries, already busy
yeah 14:05
nine Correct answer would have been: "ok, cool, have fun! I'll continue with the hard bits then" 14:06
brrt btw, even-moar-jit really requires its' own dynasm, though
hehehe
btw, magit <3 14:12
dalek Heuristic branch merge: pushed 200 commits to MoarVM/even-moar-jit by bdw
nine src/jit/graph.c:1723:31: error: ‘jgb’ undeclared (first use in this function) 14:13
That's pretty much exactly as far as I've gotten, too ;)
I take it jgb is now just jg?
brrt yes 14:18
used to have a builder there, but stopped making sense
the 'b' is for builder :-)
dalek arVM/even-moar-jit: 3e31ed3 | brrt++ | src/jit/graph.c:
Merge: JGB to JG refactor
nine Perfect! Builds now and NQP is happy with the version 14:21
brrt :-) 14:23
nine I guess a segfault during rakudo SETTING compilation is to be expected? 14:24
Lovely 58239 lines backtrace 14:26
jnthn Of...C frames or Moar frames? o.O 14:27
nine C frames 14:28
Endless recursion of MVM_frame_find_invokee/MVM_exception_throw_adhoc/MVM_exception_throw_adhoc_free_va/run_handler 14:29
Origin is probably in JITed code as there are a bunch of frames without source information. And it's the even-moar-jit branch, so blaming the JIT seems obvious ;)
brrt ah damn 14:33
:-( 14:34
but, yes, that is probably the issue
sensitive to MVM_JIT_EXPR_DISABLE=1?
jnthn: interested in merging the extract-spesh-alloc branch, btw? 14:37
jnthn brrt: I'll try and remember to take a look at that tomorrow 14:38
Well this is weird 14:39
nine brrt: yes, it is
jnthn I have a program that does either run or Proc::Async and doesn't set any handles up, so they are inherited
Which seems to work just fine in a golfed example 14:40
But in my application, it goes and loses half (or more) of the keystrokes I type on stdin o.O
oh, lol, I was accidentally running another process at the same time that also inherited stdin, just carelessly rather than intentionally 14:43
I guess they were competing for the keystrokes or something :P 14:44
nine With disabled jit, rakudo's build is successful 14:45
timotimo disabled jit or disabled expr jit?
yoleaux2 11:38Z <babydrop> timotimo: I think it's just zoffix (RE: wordpress account for weekly perms)
nine timotimo: expr jit 14:46
brrt darn 14:51
well, i'll figure out why that happens eventually
i have the jit bisecter now 14:52
15:11 FROGGS joined 15:52 pyrimidine joined
dalek arVM/even-moar-jit: 81375ef | brrt++ | / (2 files):
Enable running jit-bisect with a single command

Thus allowing me to debug the breakage of CORE.setting		 16:10
16:36 pyrimidine joined 16:45 lizmat joined 17:15 pyrimidine joined 17:33 domidumont joined 18:01 lizmat joined 18:05 pyrimidine joined 18:30 pyrimidine joined 18:34 pyrimidine joined 18:57 pyrimidine joined 19:35 nine joined 19:36 avar joined 19:37 domidumont joined 19:40 edehont joined 19:48 domidumont joined 20:13 pyrimidine joined 20:32 Ven joined 20:34 Ven_ joined 20:35 Ven_ joined 20:47 pyrimidine joined 21:29 pyrimidine joined
jnthn Report on an interesting interaction between GCs and containerization: engineering.linkedin.com/blog/2016...rol-groups 21:43
22:38 pyrimidine joined 23:17 pyrimidine joined 23:21 pyrimidine joined

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!