| MasterDuke | timotimo: you around? | 00:42 | |
| Geth_ | MoarVM: ed4201e92a | (Timo Paulssen)++ | src/debug/debugserver.c debugserver: cut module names out of filenames otherwise you'd have to supply not just the filename where your breakpoint shall reside, but also the module name in parenthesis afterwards, but not always, only if the file was included from a CompUnitRepo or something like that. |
||
| timotimo | MasterDuke: is that an answer? :) | 00:44 | |
| MasterDuke | i think it'll suffice | ||
| i'm trying to write a template for floor_n | 00:45 | ||
| the case in interp.c just calls floor() | |||
| but doing `(template: floor_n (call (^func &floor) (arglist (carg $1 ptr)) num_sz))` errors with `src/jit/core_templates.h:722:50: error: ‘floor’ undeclared here (not in a function); did you mean ‘flock’?` | 00:47 | ||
| timotimo | yup, i stumbled over that before | ||
| i think maybe you have to trick the compiler into making that function a reality | 00:49 | ||
| try putting it into the op_to_func program in graph.c and see if that changes things | |||
| though maybe that's too far away for this | |||
| MasterDuke | you mean use the lego jit, not create a template? | 00:53 | |
| timotimo | yes | 00:55 | |
| somewhere it'd be available | |||
| just a random spot | |||
|
00:59
Ven`` joined
|
|||
| MasterDuke | hm. a profile says it's jitted now, but it's not any faster | 01:13 | |
| timotimo | yeah, that can be | 01:14 | |
| ideally we'd compile floor_n to actual assembly that floors a double | |||
| roundsd xmm0, xmm0, 9 | 01:15 | ||
| cvttsd2si eax, xmm0 | |||
| ret | |||
| gcc.godbolt.org/#compilers:!((comp...,version:3 - use this fantastic tool | |||
| oh lord that url | |||
| godbolt.org/g/awTmPH | 01:16 | ||
| MasterDuke | that would be done in emit.dasc ? | ||
| timotimo | yeah, i think so | ||
| wait what the heck | 01:17 | ||
| the floor_double(double) function just became much much longer | |||
| you may have to switch on the "intel" view to have the registers in the right order | 01:18 | ||
| godbolt.org/g/2axcS2 | 01:19 | ||
| this is the one that uses only two ops | |||
| oh, possibly -ffast-math is responsible for the short one | 01:21 | ||
| MasterDuke | i haven't written assembly in 15 years, and then it was mips | 01:24 | |
| it's probably best if i leave this to someone else | 01:25 | ||
| timotimo | hah | ||
| *nice* the compiler explorer lets you right-click on an asm instruction and view asm help | |||
| MasterDuke | that is a pretty awesome tool | 01:27 | |
| timotimo | gotta go to bed, seeya! | 01:28 | |
| MasterDuke | later... | ||
|
01:56
ilbot3 joined
02:31
Kaiepi joined
03:06
robertle joined
04:15
Kaiepi joined
04:17
Kaiepi joined
06:01
robertle_ joined
06:56
robertle_ joined
07:12
domidumont joined
07:15
AlexDaniel joined
07:18
domidumont joined
07:39
FROGGS joined
08:00
zakharyas joined
08:22
Geth joined
11:07
zakharyas joined
12:12
Geth joined,
synopsebot joined
|
|||
| timotimo | .ask brrt so how do i go about finding the right line number when printing a backtrace when there's jit involved? | 12:48 | |
| yoleaux | timotimo: I'll pass your message to brrt. | ||
| MasterDuke | timotimo: think i should create MoarVM issues for floor/pow/ceil/etc? about jitting them or handcrafting some assembly | 12:53 | |
| timotimo | yeah, maybe :) | ||
|
13:39
zakharyas joined
|
|||
| Geth | MoarVM/dont_gc_in_spesh: 4ec52aed01 | (Timo Paulssen)++ | src/spesh/worker.c init a field we access right away |
13:47 | |
|
14:00
FROGGS joined
|
|||
| Geth | MoarVM: b5609f8372 | (Timo Paulssen)++ | 8 files spesh worker: when wval fix fails, retry a few times |
14:24 | |
| MoarVM: 4ec52aed01 | (Timo Paulssen)++ | src/spesh/worker.c init a field we access right away |
|||
| MoarVM: a925baeea5 | (Timo Paulssen)++ | 8 files Merge branch 'dont_gc_in_spesh' Running GC inside the spesh worker is forbidden, but when fixing up wval instructions during inlining, we sometimes caused objects to be deserialized for the first time. Now we check if an object we're refering to has not yet ... (5 more lines) |
|||
|
14:25
releasable6 joined
14:57
zakharyas joined
15:05
zakharyas joined
16:10
zakharyas joined,
AlexDaniel joined
16:40
Kaiepi joined
17:02
geospeck joined,
geospeck left
17:17
TimToady joined
17:25
Kaiepi joined
|
|||
| lizmat | and another Perl 6 Weekly hits the Net: p6weekly.wordpress.com/2018/04/02/...rl-6-idea/ | 17:41 | |
|
18:19
Kaiepi joined
18:26
Kaiepi joined
18:43
domidumont joined
19:47
FROGGS joined
20:48
lizmat joined
|
|||
| lizmat | someone here with access to modules.perl6.org ?? it seems to be giving 500's :-( | 21:54 | |
| JITting MoarVM is going to get interesting on MacOS in 2020: www.bloomberg.com/news/articles/20...-from-2020 | 22:12 | ||
| timotimo | oh, fascinating | 22:19 | |
| jnthn | Interesting indeed | 22:27 | |
| 'night | 22:28 | ||
| lizmat | night& | 22:29 | |
|
23:03
AlexDaniel joined
23:17
Kaiepi joined
23:28
MasterDuke joined
23:41
Kaiepi joined
|
|||