samcv what was that ENV variable that when set makes benchmarking more reproducible? 01:00
MasterDuke MVM_SPESH_BLOCKING 01:14
samcv: ^^^ 01:16
MasterDuke weird. i can perf record a regular rakudo CORE.setting compile, but if i change it to --target=parse, it pretty much goes on forever, creating a bigger and bigger data file 02:54
in the regular compile, the top function is MVM_interp_run at only 7.25%. however, in the target=parse case, the top two functions are MVM_gc_root_add_frame_registers_to_worklist and MVM_gc_root_add_frame_roots_to_worklist which are 38% and 12.3% respectively 03:00
samcv maybe i could hide it behind some define that we set if we're ok with unaligned memory accesses 03:22
and then i can hide the 32 bit aligned 64 bit integer access i use on 64 bit platforms behind that as well
nine samcv: yes, making it optional seems like the way to go 08:27
Geth MoarVM/master: 5 commits pushed by (Stefan Seifert)++ 15:04
brrt .tell nine i'm suspicious of encoderepconf. I know that's what interp.c does, but it is not what oplist says 20:24
yoleaux brrt: I'll pass your message to nine.
brrt oplist says "encoderepconf w(obj) r(str) r(str) r(str) r(obj) r(int64)" - which means that the zeroth argument should be the one written to 20:28
also, because of this, the expr template will do the wrong thing; it will try to assign to the object address, not to the variable slot
$4 will be the pointer, not the address-of-the-work address 20:29
anyway, that's all i have. see you around 20:30

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!