github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
brrt good * #moarvm 07:56
nwc10 good *, brrt 07:59
brrt ohai nwc10 08:00
lizmat and another Perl 6 Weekly hits the Net: p6weekly.wordpress.com/2018/06/18/...lake-city/ 19:57
timotimo heh, commemnts 20:13
Zoffix New blog post: "Perl 6 Colonpairoscopy": rakudo.party/post/Perl-6-Colonpairoscopy 20:39
timotimo i liked that post, zoffix++ 21:49
jnthn The title's a bit of a pain in the butt, though :P 22:25
timotimo it's a pun in the butt 22:26
lizmat depends how you look at it
jnthn I'm not sure I want to look there :P 22:28
timotimo leave that to your progtologist
MasterDuke has anyone tried building MoarVM with LTO? 23:14
jnthn Link Time Opt? Ain't that on by default these days? 23:15
MasterDuke oh, you don't need -flto anymore?
jnthn I'm sure all my MSVC builds were doing that, anyway...I kinda assumed it was happening on gcc/clang too
Would be worth checking, but the linking step seems to take long enough that I hope it's doing something useful there :-) 23:16
MasterDuke hm: "Link-time optimization does not work well with generation of debugging information. Combining -flto with -g is currently experimental and expected to produce unexpected results." 23:17
that's from the 7.3 man page
8.1 isn't so scary: "Link-time optimization does not work well with generation of debugging information on systems other than those using a combination of ELF and DWARF." 23:18
oh! adding '-flto' to the CFLAGS makes the build take much longer (all during the "linking libmoar.so" stage). from 30s to 70s 23:24
geekosaur well, yes 23:29
MasterDuke hm. one random benchmark was slower using a moar built with -flto 23:33
only very slightly. stage parse of building rakudo seemed exactly the same 23:43
geekosaur I wouldn't actually expect LTO to gain that much for something like moarvm 23:44
which is not that there's nothing to gain, but for something like moarvm you're going to need to do better than automatics. 23:48
MasterDuke you mean profile-guilded instead? 23:50
*guided
geekosaur I mean very careful design, and profiling is only part of it. It's an interpreter; you need to put thought into what and how it interprets 23:51
and it will be much more sensitive to memory layout than most code 23:52
but a decent amount of *that* might actually have to be at nqp level or higher to be effective
...or where the effort is actually being put: JIT 23:53

Powered by Cro and the Raku® Programming Language.

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

Thank you!