github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
nwc10 good *, #moarvm 06:16
nine good *, nwc10 06:52
nwc10 hopes that jnthn and the coffee machine are better aligned this morning than yesterday morning. 08:07
jnthn nwc10: Yes, I seem to have a quite normal cup of coffee today 09:24
nwc10 excellent 09:33
brrt \o 12:25
hmm, I've been using xorg for a bit, and it is markedly slower than wayland 12:27
nwc10 o/ 12:33
[Coke] needs to stop using the k-cup brewer. 13:43
brrt good * #moarvm 20:05
timotimo heyo brrt :) 20:08
brrt ohai timotimo 20:12
timotimo how are things? 20:16
is the dispatch chains mechanism making you excited? :)
timotimo i was thinking some time ago that it could be a natural mechanism to handle dispatching to the right devirtualized candidate for, say, a VMArray's at_pos for all the variants of slot kinds 20:22
timotimo currently that's hard-coded, which isn't great, but the results are good enough 20:22
and when reprops will all go via the dispatch op so that we can elide checks of all kinds at spesh time it'll be sort-of natural already 20:24
brrt I'm sorry I haven't even followed the dispatch chain thing :-S 21:02
what is it about? 21:03
timotimo it's an implementation of a paper called something with like "zero-overhead metaprogramming" in the title 21:04
brrt heh, that sounds like it should be interesting to us 21:06
timotimo yup 21:12
it's already replaced every spesh plugin in rakudo if i remember correctly 21:13
but i haven't implemented "turning a dispatch recording to spesh graph of guards" for all ops that can occur
every dispatch chain sits a boot-dispatcher which we can implement rather directly in the jit or even spesh 21:20
some invoke a code eobject, which spesh can inline
some invvoke a C function, which the jit can turn into a low-level call
and of course for some C functions we can have simpler implementations in the jit itself 21:26