03:28
lue joined
04:45
ggoebel11111113 joined
06:42
FROGGS joined
06:44
flaviusb joined
06:45
camelia joined
07:26
camelia joined
08:08
vendethiel- joined
09:35
kjs_ joined
10:07
cognome joined
12:02
cognome joined
|
|||
timotimo | the decont_all that's put into the nativecall MAST Op ... it can't get jitted because param_sp_o | 13:01 | |
can we get that to work better? | |||
i don't even know exactly why param_sp_o gets emitted sometimes | |||
timotimo tries $args instead of @args | 13:02 | ||
13:05
colomon joined
|
|||
timotimo | seems like that gives the same op | 13:07 | |
er | 13:10 | ||
i meant to write param_rp_o | |||
"required positional object", right? how come we can't jit that :S | |||
we don't jit named parameters? or something like that? | 13:11 | ||
i differentiate between relative and absolute move_to and line_to with a :relative at the end | 13:12 | ||
seems like the jit doesn't like that so much | |||
oh interesting | 13:16 | ||
10 bails on "multicachefind" | |||
that should be easy | |||
moritz | does removing reasons to bail actually improve performance measurably? | 13:19 | |
timotimo | depends entirely! | 13:23 | |
it's just more stuff i can do | |||
i'd think moving between jit and interpeter has a tiny amount of cost | |||
otherwise we can inline and stuff | |||
hm. invokewithcapture follows all instances of multicachefind | 13:24 | ||
invokey ops are a bit harder | |||
dalek | arVM/multicachefind: 7fd90e7 | (Timo Paulssen)++ | src/jit/graph.c: multicachefind for the jit needs invokewithcapture in order to actually work. |
13:25 | |
timotimo | brrt will have to do something about this : | ||
:P | |||
the three ops that bail the most in my game are: param_sn and param_rp_o with 45 frames each | 13:26 | ||
26 frames with param_sp | |||
10 with multicachefind (and after that invokewithcapture) | |||
9 argconst_s, 7 scgetobjidx | |||
exception, setobjsc, setcodeobj, captureposelems, capturehasnameds, reprname, param_on2_o | 13:28 | ||
newexception, isinvokable, getcomp, ctx, cmp_n, | |||
all after that have 2 or 1 bails | |||
dalek | arVM: 41216a0 | (Timo Paulssen)++ | src/jit/graph.c: jit pow_n and sqrt_n |
13:34 | |
13:38
cognome joined
|
|||
timotimo | there's really not terribly much to add ... | 13:55 | |
14:10
_sri joined
|
|||
timotimo | but it's still too slow! oh no! | 14:13 | |
14:16
colomon joined
14:40
kjs_ joined
14:51
zakharyas joined
16:35
cognome joined
16:38
cognome_ joined
17:04
zakharyas joined
17:50
cognome joined
19:19
Ven joined
22:27
kjs_ joined
22:49
flussence joined
22:58
cognome joined
23:49
kjs_ joined
23:55
colomon joined
|