This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Fix DCE REG_LIBCALL note moving from noop move insns (PR rtl-optimization/33644)
> If you don't see the practical advantage of being able to determine
> whether an insn is part of a libcall without having to walk the entire
> function to figure out whether you are allowed to touch an insn,
> please just remove it.
I've posted a patch to that effect. It restores the correct interpretation of
libcall notes as optimization helpers rather than optimization barriers as
far as DCE is concerned and, thus, eliminates the need for being able to
determine whether an insn is part of a libcall in dce.c.
But I won't install it myself, given how controversial this issue has become.
I therefore propose that a 3rd party decides here, and I'd suggest Jakub.
Yes, I see theorical advantages over being able to determine whether an insn
is part of a libcall without having to walk the entire function.
No, I don't see any practical advantages given how the RTL passes are written,
including the new DCE pass which starts with a linear walk over the insns.
> God forbid we ever start somewhere and work towards a goal.
> I'm completely done working on our backend or trying to improve it in any
> way. I'm removing my name from the dataflow maintainer list.
Everything I've written above doesn't change in any way my judgement on the
great work you have done with Kenneth and others on the dataflow branch: