github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
brrt \o #moarvm 08:19
a different register allocation algorithm is starting to sound attractive
nwc10 o/ 08:20
based on bogosort? Or it's not that bad, *yet*? :-)
brrt not that bad yet 08:25
want the gory details?
or the short summary? :-P 08:26
nwc10 it's all, um, I'm all teddy-bear to it
so, I think that the short summary is more useful
brrt the short summary is, linear scan, as a greedy algorithm, is great at just handing out registers fast. But it is considerably less great when there are constraints on what registers to hand out 08:27
now maybe there is a way out of there that I don't know yet
a hint bitmap that we collect... hmmm
but the thing is, x86-64 ISA is full of constraints like those 08:28
maybe what I really want is linear scan for something more regular (ARM perhaps), and something more complex for x86-64
nwc10 but right now you're working on x64-64, and "ARM JIT" is just an optimistic sentance? Hence solving the problem for x86-64 matters? 08:30
brrt correct
nwc10 and so by the time x86-64 is solved, it might still be just as easy to keep using the same approach for ARM (or arm64, which might be the only thing that matters by then) 08:30
er, I forget, they call it aarch64? 08:31
brrt yeah, it's weird
brrt \o 22:36