moritz \o 11:15
FROGGS morning 11:30
timotimo brrt: i'm glad to hear you found that nasty bug in the register allocator (or where exactly it was) 14:14
brrt :-) 14:15
wasn't a register allocator bug after all
don't do drugs and don't take pointers while you may realloc
timotimo yeah 14:16
i've stumbled over that a few times myself
i think the last time was when i implemented "how many percent of these allocated things survive the first collection?"
so, now that nqp builds again, you'll be working on what? tiler serialization?
i forget where exactly you were hanging 14:17
brrt yeah, this was tiler serialization
tiler linearisation actually
but that is done now
timotimo er, yeah 14:20
it's done? nice!
so, hmm, tiler serialization is done. so the next step is to make the register allocator work off-line? 14:24
brrt yeah 14:26
basically, the offline register allocator is needed to deal correctly with multiple-basic-blocks 14:27
timotimo ah, yes
that'll be something we really want to have :)
brrt aye, because we can then deal with conditionals 14:28
i'm currently thinking of a many-small-temporary-structures design 14:29
timotimo i hear conditionals are all the rage in programming nowadays
we should support them in our project, too
brrt dunno, i'm thinking they're a bit of ahype 14:31
timotimo without conditionals i can't decide whether or not they are 14:32
brrt are you going to GPW btw? 14:35
timotimo i'm hoping to! but i've been procrastinating finding a hotel room
and i'm still supposed to come up with an idea for a talk, too
there was a stretch of time where hotel rooms were discounted and i missed that >_<
brrt ah, that happens 14:36
you can also pick a shorter talk slot if you can't find something long to talk about
timotimo i was kind of considering a talk along the lines of "from 1.5 fps to 160" about the white noise thing in SDL2 14:39
thing is: before that i really should fix the multiple-inline stuff i've found while doing that 14:40
brrt no, you should totally do a talk like that 14:59
that is a great idea
even if it's still WIP, that is very interesting
timotimo well, it was just a basic inner loop that i could get gc-free and invocation-free
the gc wouldn't have been so bad if it wasn't full-blown Int objects, which require a malloc and a free 15:00
and the invocation-free-ness won't be so interesting any more once jnthn makes invocation faster this year
brrt still interesting today
timotimo in that case i could turn this topic into "texture generation" in general
brrt okay, the interesting bit is not necessarily 'i did something genius and you should do it all in the same way 15:01
timotimo because that's kind of what i was doing; stream data into a big buf and push that into video memory, then just blit it to the display surface with opengl
brrt the interesting bit is that there are a lot of vague ideas floating around regarding perl6 performance
'isn't it still super slow' vs 'we have a jit so we can while $i++ faster'
timotimo hehe.
brrt and you have a really practical example of how you a): hit perl6 performance constraints, b): why these constraints happen, c): how to detect what's going on, d): how you can improve, e): what is being done to improve this on the backend 15:03
timotimo that's ... totally true!
brrt basically, you can make a vague, 'emotional' topic (performance) very specific, and that is interesting 15:04
timotimo i bet i'm too late to submit the talk :)
brrt ah well
still a good idea for a talk i'd think :-) 15:05
timotimo worst case i'll just build a youtube video screencast-style
timotimo GPW is a bit earlier than i had remembered :o 15:12
nine_ still has to prepare two talks for the GPW 15:35
brrt is preparing his thesis defense presentation 15:36
nine_ And one of them on an area that I currently rebuild leaving little untouched...
brrt first thesis of two, before y'all might think i'm nearly finished 15:37
what area?
nine_ I'm trying to fix all precomp/module management issues at once ;) 15:39
nine_ The sha in file names issue, the "we can't use precomp files generated during installation" issue, the "we can't package modules for distros" issue... 15:39
brrt oh, everything all at once 15:45
great idea :-)
good luck though
brrt is having sunday now 15:46
see you!
(i'm serious about the good luck :-))
timotimo .o( thesis dance presentation ) 15:55
japhb_ timotimo: I think I would pay to see that. People could make their first payment on student loans from the proceeds of ticket sales for that. 17:00
timotimo :D 17:03

Powered by Cro and the Raku® Programming Language.

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

Thank you!