nwc10 jnthn: paste.scsys.co.uk/506958 -- fails a test‽ :-) origin/lazy-strings doesn't even *build* under ASAN (just a NULL pointer) 08:59
jnthn nwc10: When I get a test fail in a ready-built NQP, I tend to assume the chances of NQP building aren't great :) 14:29
(So didn't try it...)
nwc10 ah right 14:46
jnthn It may be the JIT thing that I just didn't get to yet 14:47
nwc10 seesm to be. just passed all NQP tests after building/testing with MVM_SPESH_DISABLE=1 14:55
jnthn What about MVM_JIT_DISABLE=1?
nwc10 er, didn't find that first in shell history 14:57
and also should I stop the rakudo build? 14:58
and also might be abducted by "Being socialable (TM)"
jnthn Yeah, I need to get to the kitchen shortly too :) 15:01
No worries, I don't think I'll have any problems reproducing the issue here :) 15:02
nwc10 with MVM_JIT_DISABLE=1 NQP builds and tests 15:36
and Rakudo gets to the end of the spectest with nothing strange 16:05
nwc10 paste.scsys.co.uk/507024 -- valgrind output makes it look like a plain old NULL pointer dereference 16:09
it's in MVM code, called from JIT generated code 16:10
which I think is pretty much what you thought
timotimo hmm "Being So Scalable" 16:39
timotimo hm 18:46
so if i have a one-line patch for moarvm that makes nqp build and test on lazy-strings ... 18:47
should i push that? :)
dalek arVM/lazy-strings: a501a41 | timotimo++ | src/jit/emit_x64.dasc:
force sp_findmeth to decode strings in the CU

makes nqp build&test and rakudo build.
18:49
timotimo we may want to actually use the return value of cu_string there, though 18:50
but this makes things work already
hm. is Stage parse : 48.131 18:51
kind of much faster than usually?
nope, not much faster 18:56
timotimo i made a particle system for sdl2. it ends up taking a gigantic amount of IntLexRef and NumPosRef :( 19:52
with the maximum particle count set to 1000 i get only about 30 fps average :| 19:53
hm, but there's also lots of Num objects being allocated by Bridge
and * and +
i can potentially fix those
timotimo cool. i'm up to 51 median, 35 in the 5th percentile and 257 in the 95th percentile 19:58
i guess the very high frame speeds come from having only few particles alive 20:00
nwc10 jnthn: works on "my" machine (to spectest) 20:15
timotimo nwc10: what does? the whole thing with jit disabled? how about with my latest commit and jit & spesh re-enabled? 21:11
nwc10 oh, yes, it's your commit on what started as his branch 21:12
oarVM/lazy-strings
^M
ie MoarVM commit a501a41e37e4 21:13
timotimo cool
timotimo huh, somehow i got find_best_dispatchee back into my code 23:35