|
01:02
live_the_dream joined
01:59
ilbot3 joined
02:39
releasable6 joined
03:43
Kaiepi joined
03:45
Kaiepi joined,
Kaypie joined
06:42
domidumont joined
06:49
domidumont joined
07:24
lizmat joined
|
|||
| Geth | MoarVM: fa167464ac | (Zoffix Znet)++ | src/math/bigintops.c Define INFINITY for MVC Should unbreak build on Windows |
07:58 | |
|
08:41
brrt joined
|
|||
| brrt | good * | 08:41 | |
| yoleaux | 6 Apr 2018 22:05Z <timotimo> brrt: don't worry about the "potential version skew detected" bug, it's entirely my fault and there'll be a fantastic patch that'll make everything nicer tomorrow or so | ||
| brrt | .ask timotimo i'm far too curiuos aa to what the issue might be | 08:42 | |
| yoleaux | brrt: I'll pass your message to timotimo. | ||
| brrt | have been | ||
| timotimo | oh | 09:04 | |
| yoleaux | 08:42Z <brrt> timotimo: i'm far too curiuos aa to what the issue might be | ||
| timotimo | so you'll debug it even if i rip it out completely? :) | 09:05 | |
| looks like currently we have an MVM_oops if fix_coderef ever gets called | 09:14 | ||
| so i might not have to implement that part yet, if it's never used | |||
|
09:23
domidumont joined
|
|||
| timotimo | jnthn: my spesh inlining work leads me into deopt.c =o | 09:53 | |
| jnthn | Congrats? :) | 09:55 | |
| timotimo | hah, some guardconc instruction has a want of "0x1234" | 09:57 | |
| no, 011248 | 09:58 | ||
| 0x1248 | |||
| jnthn | That...doesn't look like a pointer | ||
| nwc10 | dear hackathon, please don't forget lunhc | 10:27 | |
| ah, oops, that won't highlght | |||
| lunch | |||
| lizmat | LUNCH! ;-) | ||
| BRÖTCHEN! | |||
| nwc10 | BIER | ||
| (Pivo) | |||
| lizmat | .oO( Pivotal for a good hackathon ) |
10:28 | |
| nwc10 | lizmat: I believe that ilmari has now configured "lunch" as an "alternative nick" in his IRC client | ||
| TimToady | here they only have Piwo... | ||
| nwc10 | (not that he's at the hackathon) | ||
| and, I guess, Wódka | 10:29 | ||
| good *, * | 10:30 | ||
| timotimo | kuchen | ||
|
10:40
FROGGS joined
|
|||
| timotimo | jnthn: turns out, since the fixups are only ever invoked if we inline, turning inline off completely eliminates all code i have, and so finding the bug is completely impossible :D | 10:46 | |
| jnthn | oh, of course! | 10:49 | |
| duh :) | |||
| Shows how smart I am today :) | |||
| FROGGS | haha | ||
| timotimo | heyo FROGGS | 10:51 | |
| FROGGS | o/ | ||
|
11:02
lizmat joined
|
|||
| timotimo | the "much more elegant" code is rather a bit explosive | 12:01 | |
| dogbert17 | elegantly explosive | 12:06 | |
| timotimo | i'm sitting in the splash zone and i can tell you it's not actually elegant | 12:08 | |
| Geth | MoarVM/lazier_inline_fixups: b3ff8b78d8 | (Timo Paulssen)++ | 7 files try to make fixups on inline less bad very crashy |
12:15 | |
| timotimo | ==31440== Thread 2: | 12:37 | |
| ==31440== Invalid write of size 8 | |||
| ==31440== at 0x512962D: MVM_spesh_add_spesh_slot (optimize.c:79) | |||
| ^- smoking is bad for a gun's health! | |||
| fix_callsite and fix_str can have their inlinee changed to inliner and it won't do an invalid write any more, but it'll MoarVM panic: Deopt: did not find code object when uninlining | 12:45 | ||
| so it's finding garbage in the code_ref_reg, but wouldnt that be set before entering the inline and thus not influenced by the fixed-up inline code at all? | 12:49 | ||
| jnthn | greppable6: APPEND_TO_ORIG | 13:27 | |
| greppable6 | jnthn, Found nothing! | ||
| jnthn | timotimo: Odd indeed, I thought that was just set up by the calling code | 13:28 | |
| timotimo | could be that the inlinee writes some garbage to that register for some reason | 13:30 | |
| jnthn | But inlines should only ever touch their own registers | ||
| FROGGS | damn, it doesnt help to be absent for a few years... inlined fixed sized CArrays arn't there still :S | 14:03 | |
| example: class Foo is repr<CStruct> { has uint8 @.bar[32] is CArray; } | 14:04 | ||
| first I need to know what happens to [32] I guess | |||
| IIRC it became part of the attribute... | 14:14 | ||
| timotimo | jnthn: we're adding the STable to the spesh slots list when logging the facts of an invoke_o op, it comes from agg_type->st | 14:27 | |
| while optimizing EXPR | 14:28 | ||
| the annotation that belongs to it has a bytecode offset of 322, but in the spesh log i only see 264 and 344 | 14:33 | ||
| jnthn | timotimo: Hmm...I think we started to add facts for inlined code (nine++ did it iirc), is it happening during that phase? | ||
| timotimo | no, looks not to be the case | 14:35 | |
| #8 0x00007f3499f28fac in MVM_spesh_facts_discover (tc=0x23eb2f0, g=0x7f34900de110, p=0x7f34900cff50) | |||
| at src/spesh/facts.c:659 | |||
| #9 0x00007f3499f2118f in MVM_spesh_candidate_add (tc=0x23eb2f0, p=0x7f34900cff50) at src/spesh/candidate.c:63 | |||
| #10 0x00007f3499f38d52 in worker (tc=0x23eb2f0, callsite=0x7f349a673360 <null_args_callsite>, args=0x0) | |||
| jnthn | Hm, ok | ||
| timotimo | maybe i'll drive home soon-ish | ||
| jnthn | Yeah, I'm a bit tired to look at it more for now also | 14:38 | |
| timotimo | i was looking at a different candidate, actually | 14:41 | |
| now i see the logged at bytecode part that matches better | |||
| 11 x type NQP::Grammar (Conc) | |||
| 20 x static frame '!cursor_start_cur' (84) (caller is outer: 0, multi 0) | |||
| 11 x type tuple: | |||
| Type 0: NQP::Grammar (Conc) | |||
|
14:43
zakharyas joined
14:48
zakharyas joined
14:49
zakharyas joined
14:50
AlexDaniel joined
15:08
domidumont joined
15:44
zakharyas joined
17:16
dogbert17 joined
18:35
zakharyas joined
18:37
zakharyas joined
19:11
zakharyas joined
19:36
zakharyas joined
19:51
lizmat joined
|
|||
| Geth | MoarVM/inlined-shaped-carray: 2ed75977a1 | (Tobias Leich)++ | 2 files Implement one dimensional inlined shaped CArrays A C(PP)Struct can now hold an inlined array with a fixed sized shape. There is one issue though. We always assumed that CArray attributes of CStructs will always hold pointers to other things than numeric types. I've changed the default for inlined CArray attributes which have a shape. The memory of the CArray slots will be inlined by default as well. If someone wants the opposite, (s)he can declare a CArray of pointers. |
22:07 | |
| FROGGS | woohoo | ||
| jnthn | wow! | 22:08 | |
| FROGGS++ | |||
| FROGGS | :D | ||
| hope it gets your approval | |||
| jnthn | Will take a look when I'm not about to fall asleep :) | 22:09 | |
| Nice long train journey tomorrow... :) | |||
| FROGGS | sure :o) | ||
| timotimo | holy wow | 22:11 | |
| that's a thing we've been wanting for so long! | |||
| FROGGS | I need that for my Box2D port, so I can finally continue to hack on my game | 22:12 | |
| timotimo | i like to hear that! | 22:13 | |
| FROGGS | jnthn: I hope the internet connection does not suck too much in germany | 22:14 | |
| (while being on a train, that is) | |||
| jnthn | It was pretty alright on the ICE from Berlin - Hagen | ||
| Will try it out on the Cologne - Munich one tomorrow :) | |||
| jnthn goes to rest ahead of that trip | 22:16 | ||
| 'night | |||
| FROGGS | gnight jnthn | 22:17 | |