Perl 6 language and compiler development | Logs at colabti.org/irclogger/irclogger_log/perl6-dev | For toolchain/installation stuff see #perl6-toolchain | For MoarVM see #moarvm
Set by Zoffix on 27 July 2018.
timotimo jnthn: got an idea how we should handle an attribute being used before it's declared in the class when it'll be declared to be native, but the methods using it already code-genned a regular old object access? 00:02
00:16 samcv left 00:18 samcv joined
jnthn timotimo: Would have to keep track of the attribute access QAST::Var nodes and then set the .returns on them after composition time 00:26
timotimo oh, it's that easy? cool. 00:27
jnthn I think so :)
timotimo i think this is the fourth time i've stumbled upon seeing the null op emitted by takedispatcher being spesh'd in inline not being used to give the isnull after it a constant value
i seem to recall we wanted to postpone that until we can just run the regular optimizer post-inline or something?
i think that point is coming ever closer? 00:28
jnthn Yes, and in theory I've done that work
timotimo just have to sort out some explosions from the real world? :)
jnthn I did switch on more DIE, which went well. Then tried to enable some further opts and got some explosions.
But that was before the deopt information fix I did this last week 00:29
Which might well have caused such troubles
timotimo ah, yes
jnthn But yeah, in theory we'll be able to make the normal optimize_bb run over the blocks that we inlined 00:32
Which should take out the very thing you'd like to see gone :)
timotimo i don't see an optimize_null anywhere :P 00:33
jnthn Well, modulo that :P
timotimo :)
jnthn But that's the easy bit 00:34
timotimo hm, currently we're building a low-level list object every time we want to execute a native call, to hold all the arguments 00:38
that also forces boxing to happen, of course
i wonder how we could handle that better
have a tc-registered MVMRegister* that can be used to hold stuff? though that doesn't account for rw natives and such 00:39
jnthn That's a bit of a tricky one; I've wondered if we might manage to do something vaguely clever with a spesh plugin there 00:42
Though never figured it out all the way (didn't spend that much time thinking about it yet)
timotimo mhm 00:44
well, at least the jit devirtualizes every push as well as the setelemspos 00:59
though i could imagine a series of create, set size, reset size to 0, push a bunch, could actually be special-cased in the jit
since it's how we emit an nqp::list(1, 2, 3, 4) 01:00
01:00 lizmat joined
timotimo though of course i'd first want to have a measurement of how often that actually happens in "boring" code 01:01
create_BUILDPLAN has it; not exactly hot in regular user code, of course 01:03
01:29 ExtraCrispy left 01:32 Tux__ left 01:33 |Tux| joined 01:34 lizmat left 01:42 ExtraCrispy joined
vrurg Wonder if the list of concretizations would always match the list of roles? 02:04
timotimo hm, only for roles with arguments i guess? 02:07
vrurg No as any role gets specialized at composition time. 02:08
timotimo OK 02:13
vrurg Too bad I missed jnthn 02:15
Geth rakudo: ramillies++ created pull request #2689:
Fixed the bug #2688 (wrong values for asinh of negative argument)		 02:28
synopsebot RAKUDO#2688 [open]: github.com/rakudo/rakudo/issues/2688 asinh() is not an odd function
releasable6 Next release in ≈6 days and ≈15 hours. 10 blockers. Please log your changes in the ChangeLog: github.com/rakudo/rakudo/wiki/ChangeLog-Draft 03:00
06:09 Kaiepi joined 08:02 ufobat___ joined
nine timotimo: we only do the list object thing for native calls that cannot be JIT compiled. If you want to get rid of lists, just teach MoarVM to JIT compile more argument types 09:05
timotimo oh, interesting 09:24
nine See NativeCall's !create-jit-compiled-function-body 09:26
timotimo i see it, but i don't see anything like checks for what we support 09:32
or is that inside of moar? 09:34
nine github.com/MoarVM/MoarVM/blob/mast...all.c#L435 09:35
Look for goto fail;
timotimo haha, that's a name i remember fondly
the type that's bailing out my native calls is CSTRUCT, and i dunno how this part of the code has to look in order to happen 09:51
09:58 lizmat joined
nine timotimo: would that be passed as an actual struct or as struct*? 09:58
timotimo i'd expect struct* 09:59
last night i was trying to build sp_p6oget_i32 and sp_p6obind_i32, but i was getting super strange segfaults .. today i take a look being a little more awake and see i had the exact same signature for both ops, even though bind obviously doesn't have a write register as its first arg 10:01
well, that mystery is solved 10:02
nine: i think your new compiler code was the first to require those ops for optimization :) 10:03
nine nice :) 10:04
timotimo i wouldn't expect it to make much off a difference, but of course every little bit helps :) 10:06
nine Would it be possible to move the void *cstruct; to the top of MVMCStructBody? Then supporting CStruct in NativeCall JIT would be as simple as adding it to this case: github.com/MoarVM/MoarVM/blob/mast...all.c#L522
timotimo i could imagine moving that to be absolutely problem-free 10:07
MoarVM::Handler and MAST::Frame have accessors for 32bit attributes that now get an sp_p6o[bind|get]_i32 for them 10:11
how often are they called?
nine I don't think it's terribly often 10:12
timotimo darn 10:13
i was so excited! :P
nine Sorry about that... You could try that CStruct thing. As you brought it up, getting that to JIT compile could speed up your code a lot 10:14
timotimo working on it 10:16
now i have one more bail for the double arg type 10:19
that's of course more fun because of float registers, right?
nine probably, yes 10:27
timotimo that's the function i'm most excited about, argh :) 10:28
it's SDL_RenderCopyEx; it has an "angle" argument that lets me rotate stuff
hold on, we just use the jit's c call nodes 10:33
they know about float arguments already
shouldn't this be trivial?
10:35 lizmat left
timotimo oh 10:37
it might be doable with not too much work 10:39
looks like it's now always passing 0 after compilation has happened 10:59
11:02 lizmat joined
timotimo m: my buf8 $foo .= new; buf8.write-double(0, 3.7530748180290279e-316); say $foo 11:06
camelia No such method 'write-double' for invocant of type 'Buf[uint8]'
in block <unit> at <tmp> line 1
timotimo m: my buf8 $foo .= new; buf8.write-num64(0, 3.7530748180290279e-316); say $foo
camelia Invocant of method 'write-num64' must be an object instance of type 'Buf', not a type object of type 'Buf[uint8]'. Did you forget a '.new'?
in block <unit> at <tmp> line 1
timotimo m: my buf8 $foo .= new; $foo.write-num64(0, 3.7530748180290279e-316); say $foo
camelia Buf[uint8]:0x<C8 1A 87 04 00 00 00 00>
timotimo m: my buf8 $foo .= new; $foo.write-num64(0, 3.7530748180290279e-316); say $foo.read-uint64(0).base(16) 11:07
camelia 4871AC8
timotimo m: my buf8 $foo .= new; $foo.write-num64(0, 3.7530748180290279e-316); say $foo.read-uint64(0).base(2) 11:08
camelia 100100001110001101011001000
[Tux] Rakudo version 2018.12-293-gd904b7048 - MoarVM version 2018.12-95-gf55ecdecd
csv-ip5xs0.748 - 0.765
csv-ip5xs-206.311 - 6.511
csv-parser22.241 - 22.484
csv-test-xs-200.437 - 0.443
test7.480 - 7.514
test-t1.861 - 1.889
test-t --race0.851 - 0.856
test-t-2031.541 - 31.882
test-t-20 --race10.522 - 10.649
 11:17
12:57 lucasb joined, dogbert2_ left 13:45 dct joined 13:52 dct left, dct joined 14:10 dct left 14:36 dct joined 15:11 dct left 15:16 lucasb left 15:46 Ven`` joined 16:00 lizmat left 16:30 tobs joined 16:40 ufobat_ joined 16:44 ufobat___ left 16:52 patrickb joined 17:00 cygx joined
cygx m: sub foo($a) { my class Foo { method a {$a} } } 17:00
yoleaux 3 Feb 2019 19:10Z <timotimo> cygx: i want to put something in the ecosystem to bridge Image::RGBA::Text and SDL2::Raw, how does that sound to you?
camelia ( no output )
yoleaux 5 Feb 2019 18:44Z <timotimo> cygx: i look forward to you having a look at my pull requests and issues for Image::RGBA::Text :) :)
timotimo oh hey! 17:01
cygx o/
m: sub foo($a) { my class Foo { method a {$a} } }; my $f1 = foo(1); my $f2 = foo(2); my $f3 = foo(3); say $f1.a, $f2.a, $f3.a
camelia 333
cygx if classes do not form closures, should such code be disallowed?
timotimo cygx: i'm pretty glad your module sparked my imagination like that
cygx glad to hear it 17:02
I've done more work on github.com/cygx/p6-image-rgba , but I think I'm done for now
timotimo i'm wondering if it'd be interesting to make operators from that Color module available in there
like, you could have a palette of a, a-but-10%-brighter, a-but-20%-brighter etc 17:03
cygx bit blit would also be an obvious candidate 17:05
pull requests welcome ;) 17:06
timotimo you mean like "this rectangular part actually belongs at these coordinates in the image"? 17:07
i've thought of that too, but i thought that might be stretching it 17:08
do you think it should be fine to put in more pixel data than you've declared the image size to be?
btw wanna see my crappy art? :D 17:09
hack.p6c.org/~timo/spaceship_and_coins.mp4 17:14
17:14 lizmat joined
timotimo cygx: ^ 17:15
got an idea what the module should be named that uploads an Image::RGBA into an SDL2 Texture? 17:16
cygx that's not crappy, that's retro 17:17
the point of factoring out an Image::RGBA class was that you have something to hang off such transformations
re naming, Image::RGBA::SDL2 or SDL2::RGBA come to mind... 17:18
timotimo SDL2::Raw + SDL2::RgbAW 17:20
17:29 ufobat_ left
timotimo fwiw, retro can look a lot better than what i made there :D 17:31
17:32 dct joined 17:37 ufobat_ joined, tyil left 17:45 Ven`` left 17:51 tyil joined 18:01 dct left 18:37 dct joined 18:48 dct left
AlexDaniel there are a bunch of blockers, so please take a look 20:24
also, I'll be away this weekend (Friday…Sunday) 20:25
leont Is there a rakudo ticket for the App::Prove6 issue? 20:55
Geth rakudo: Kaiepi++ created pull request #2691:
Fix segfault in Blob.new		 21:02
AlexDaniel leont: sorry, are we talking about this? github.com/perl6/ecosystem-unbitrot/issues/554 21:19
leont: I don't think there's a ticket for that
leont Yeah 21:20
21:31 cygx left 21:47 patrickb left
leont I haven't been able to reproduce it though 21:51
AlexDaniel leont: `Flapper`, so you likely need to run it in a loop… and perhaps you need to give your cpu some extra work to make it happen 21:56
22:00 dct joined
japhb timotimo: Run it through hq3x or hq4x (en.wikipedia.org/wiki/Hqx) or one of the other pixel art scaling algorithms? (en.wikipedia.org/wiki/Pixel-art_sc...lgorithms) I bet one of them would make your art look pretty nice. 22:49
Come to think of it, with my limited patience for placing individual pixels, that's probably how I'd do an 8-bit look at a scale that actual can be seen on a modern display. 22:50
timotimo japhb: i wanted to, but i didn't have the patience to implement one in perl6 :) 22:58
i used piskel to draw the thing and actually used vim to transfer the image over because the lazyness 22:59
releasable6 Next release in ≈5 days and ≈19 hours. 11 blockers. Please log your changes in the ChangeLog: github.com/rakudo/rakudo/wiki/ChangeLog-Draft 23:00
23:30 dct left 23:51 AlexDaniel left

Powered by Cro and the Raku® Programming Language.

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

Thank you!