github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
jnthn Aww, no dalek 09:59
Just pushed a leak fix, and got a proposed fix for the `make test` regression at HEAD also 10:00
timotimo oh, we've just been nulling out earlier entries of the eventloop work item list? 10:07
jnthn Yup 10:09
Geth MoarVM: 0c014bfb98 | (Jonathan Worthington)++ | src/core/exceptions.c
Remove premature opt to unbreak backtraces

It seems that this doesn't work out too well in combination with continuations; will need to look into why, but in the meantime this fixes the regression in Rakudo `make test` and preserves the leak fix introduced in 25b486dee8.
10:12
jnthn ah, it's geth that reports these, fo course :)
That resolves the regression in Rakudo `make test`, so we can bump it
timotimo i gave hack a little kick 10:17
Guest16965 oh, jnthn++ is on a bug fixing streak 10:59
Geth MoarVM: 64679b9993 | (Jonathan Worthington)++ | 5 files
Allow selection of max inline size per HLL

It's clear from measurements that the number that makes sense for Rakudo is a bit on the high side for NQP. Make it possible to set a maximum inline size limit per HLL.
11:11
MoarVM: 9cd8a06df3 | (Jonathan Worthington)++ | src/spesh/inline.h
Set the default inline max size to 192 bytes

Rakudo will set the 384 from before in its HLL config. Meanwhile, some measurements show that this number seems to come out best for NQP code, or at least for the kind of NQP code that's most common (compiling Perl 6 source). On My Machine, I observe:
... (10 more lines)
jnthn And an opt one too, ;) 11:14
Lunch time now, then got some other tasks this afternoon :) 11:15
lizmat and another Perl 6 Weekly hits the Net: p6weekly.wordpress.com/2019/04/08/...aking-off/ 13:16
jnthn Hm, curious. Put in some more accurate tracking of when a return handler is needed, so we can generate less of them, which'll lead to smaller bytecode and maybe less PHIs and thus perhaps less guards. Seems to help. 17:35
But...it tickles some kind of spesh inline bug where the handler goes missing :/
I guess somehow before, there was some other handler in place that hid the bug
jnthn gives up for today 18:15
brrt ohai #moarvm 18:55
yoleaux 6 Apr 2019 09:34Z <patrickb> brrt: Have you seen the gsoc linking proposal?
6 Apr 2019 09:35Z <patrickb> brrt: Also did you notice that in rakudo/master the perl6 executable already is no longer a shell script but instead a binary?
brrt .tell patrickb yes, I have seen it, I couldn't really do anything with it
yoleaux brrt: I'll pass your message to patrickb.
brrt .tell patrickb actually I had not
yoleaux brrt: I'll pass your message to patrickb.
brrt turns out I don't have a clue how SSE instructions really work 20:05
timotimo how they *really* work, eh? me kind of neither tbh ... 20:37
that shouldn't come as a surprise i suppose