github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
MasterDuke does it make sense to jit the speshguard* ops? 08:08
nwc10 good *, #moarvm 09:52
nine MasterDuke: I think spesh plugins are going to be replaced by the new dispatch architecture. So probably not... 09:54
tellable6 nine, I'll pass your message to MasterDuke 09:55
nwc10 good *, MasterDuke 10:13
MasterDuke ahoy 10:14
timotimo ahoihoi 17:26
Geth MoarVM: MasterDuke17++ created pull request #1320:
Add JIT templates for return_(i|n|s)
19:34
timotimo MasterDuke: are the BBs that have return_* in them usually big or small? 19:43
MasterDuke no idea. can take a look
timotimo i have one example speshlog right now that only has return_* in some parts of qast 19:44
those are all single-BB functions, though, which is cool
they are now all exprjitted from front to back
MasterDuke a couple are small (1-4), a couple are bigger (40-80) 19:45
timotimo what measurement unit is that?
MasterDuke well, just number of BBs
timotimo oh, i meant to ask how big the BB is that has the return_* in it 19:49
in theory, the bigger the BB the more there is to gain from exprjit
MasterDuke between 1 and 16 instructions i've seen so far 19:51
most are about 4 19:52
huh. why does the overview say 636442 deopts, but the deopts tab shows far fewer (around 200k)? 19:58
fwiw, it shows the most at github.com/rakudo/rakudo/blob/mast...st.pm6#L99 19:59