|
01:33
colomon joined
01:47
ilbot3 joined
02:54
agentzh_ joined
04:51
agentzh_ joined
05:19
jnthn_ joined
05:49
FROGGS joined
|
|||
| dalek | arVM/cunion: 9edfc5f | FROGGS++ | src/6model/reprs/CUnion. (2 files): allow to inline CStructs in CUnions |
06:53 | |
|
07:12
Ven joined
07:45
zakharyas joined
08:09
Ven joined
08:12
brrt joined
09:05
FROGGS joined
10:24
d4l3k_ joined
10:37
ilbot3 joined
10:38
Ven joined
11:10
Ven joined
11:48
rurban joined
|
|||
| brrt | \o | 11:56 | |
| rurban, now that you're here; what exactly did you mean with '100% performance improvement' for a JIT | 11:57 | ||
| i liked your slides, mostly, though :-) | |||
| masak | are the slides online somewhere? | 12:01 | |
| found them: dresden-pm.github.io/gpw2015/html/...tus.html#1 | 12:02 | ||
| wow, those are some bitter slides. :) | 12:08 | ||
| jnthn | .oO( "Dammit, Jonathan, stop getting stuff right!" ) |
12:43 | |
| masak | :P | 12:47 | |
| masak .oO( jnthn got one moar thing right ) | |||
| brrt | some of those are really bitter, yes | 12:58 | |
| [Coke] | To be fair, it's really hard not to get bitter after working on this project on and off for a decade. | 13:02 | |
| rurban | I wanted to give a fair parrot post-portem, with the related and alarming p5p problems. But it's not that bitter, python is still much worse, and there's a lot of light at the end of the tunnel | 13:04 | |
| brrt: I estimated the improved parrot perf by 2x faster by re-instating our old jit. (and fix 64bit mode). verbally I said that yours is still much better as it does much better optimizations | 13:06 | ||
| [Coke] | rurban: fyi, broken image on dresden-pm.github.io/gpw2015/html/...us.html#27 seems to have a broken image. | 13:07 | |
| rurban | I'm not sure if I mentioned the possible trace-based optimizations, which are stilll missing. | ||
| brrt | i'd argue that these optimisations are mostly jnthns and timotimos' work, though :-) | ||
| [Coke] | rurban - moarvm is installable via macports, btw. I got that going about 2 months ago. | 13:08 | |
| rurban | [Coke] thanks, I'll fix the link. It's online | ||
| brrt | but fair enough. i was just stuck on the word '100% perfomance' rather than '2x' performance' | ||
| rurban | 50% perf win is for me 1.5x faster | ||
| 100% 2x | |||
| [Coke] I tried macports with gcc-mp-5 last week, and didn't work. I fixed dyncall before. | 13:09 | ||
| It was just an example of the limited architectural focus, which is a fair point to make for moar. | 13:10 | ||
| [Coke] | I basically relied on their buildbots to tell me if there was an issue. | ||
| rurban | if we would get a proper moar macports package that would be super! | 13:11 | |
| [Coke] | FYI, the simple portfile is here: github.com/MoarVM/MoarVM/tree/mast...s/macports - if we can do a straight build on whatever architecture, should be OK here - when we did the initial rollout there were some macport-used env vars we had to respect on the build; might be missing a few more. | 13:15 | |
| rurban | fixed the image | 13:16 | |
| [Coke] | hoping to get back to this so I can make an nqp macport that has both moar & jvm support. (and if I can get the parrot one updated, maybe that a well - although nqp-parrot currently has a few test failures) | ||
| rurban | I never got around with tcl, thanks for doing that | 13:17 | |
| dalek | arVM: d2e5c76 | hoelzro++ | build/setup.pm: Add lib install dir to rpath This will allow Moar to find shared objects used by NativeCall libraries without hacks to set LD_LIBRARY_PATH for local installs |
13:32 | |
| tadzik | aww yiss | 13:35 | |
| hoelzro++ | |||
| brrt | uh, what is rpath :-) | 13:40 | |
| ah, i see | 13:41 | ||
| nice hoelzro++ | |||
|
13:56
Ven_ joined
14:30
FROGGS[mobile] joined
14:57
Ven_ joined
15:20
FROGGS joined
|
|||
| japhb | Are the .so's being saved under unique names, so that multiple versions of the | 18:11 | |
| "same" .so can be installed at once? | |||
| Or is that not yet addressed? | |||
| (I guess this is partly a Panda/CUR/S11 question, too.) | |||
| tadzik | yeah, depends on what CUR does | 18:16 | |
| timotimo | i'm considering inspecting bb's we inline for any deopt annotations and if none exist, just paste the bb's contents into the inlining block | 18:52 | |
| since we don't re-use inlined bb's anyway | |||
|
19:53
rurban joined
20:34
colomon joined
20:41
lizmat joined
21:05
moritz_ joined,
betterworld joined
21:26
Peter_R joined
21:35
colomon joined
|
|||