github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
brrt ohai #moarvm 16:36
Have I said that register allocation is not my favorite problem? 16:48
samcv i believe so :) 16:49
brrt well..
it still isn't :-)
brrt and I'm not convinced that there's much I can actually learn from luajit's allocator 16:58
maybe register hints
the register allocator and machine code emitter are totally fused 17:01
and the luajit IR appears to be designed to be emitted efficiently 17:02
masak brrt: did you see the Wikipedia section saying that linear scanning does a decent job? 17:26
(my guess is you did, and you're already way beyond that point)
timotimo i think i heard him point that out in the past 17:27
masak maybe "decent" is the wrong word. more like "interesting speed/quality ratio"
timotimo i think right now this is less the general problem and more making sure the x86 processor is also sufficiently satisfied with what you do 17:27
since brrt's currently working on getting floating point stuff into the jit 17:28
masak oh! 17:29
well, first off, brrt++ 17:30
brrt ohai masak. Yes, floating point support is work in progress. (And somewhat works, but there's corner cases in which it doesn't). 18:52
what I'm trying to figure out is if there's a way in which I can do the same, but with less corner cases
my preliminary findings are that if that's possible, luajit doesn't have it, but it does have some other ideas