Welcome to the main channel on the development of MoarVM, a virtual machine for NQP and Rakudo (moarvm.org). This channel is being logged for historical purposes.
Set by lizmat on 24 May 2021.
MasterDuke huh. why would something take longer with just MVM_SPESH_OSR_DISABLE=1 than with MVM_SPESH_DISABLE=1 ? 00:53
timo1: i did try adding `&& arg_type != MVM_CALLSITE_ARG_INT` to the conditional, but that just caused a segfault, and i didn't investigate that further 00:56
i sort of hope there's a solution that gets us to specializing for uint, not int 00:57
:q
ww
something about the facts gathered perhaps? 01:01
MasterDuke ugh. i looked at pretty much everywhere a `*_i` op is used and added something for `*_u` if it wasn't there already, but still no change 01:19
in src/spesh/* 01:20
timo1 "no change" from "causes a segfault"? 01:43
MasterDuke "no change" eq "no segfault, but not any faster" 02:38
gist.github.com/7c8c1862d593a8ea09...51c6d2f435 should be my current diff 02:40
updated with some jit addition 02:53
timo1 does MVM_op_get_mark update correctly when update_ops runs? 10:00
in jit_emit_primitive, do we have to use a different mov for p6òget_u32 to get the right sign extend behaviour? 10:03
timo1 if you post a bit of spesh log maybe with a function that used to bail arg spesh because of uint and now succeeds, and maybe if that same frame ever gets mentioned as inlined? i can take a look later 13:55