https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121947
--- Comment #6 from Hongtao Liu <liuhongt at gcc dot gnu.org> ---
(In reply to H.J. Lu from comment #5)
> (In reply to Hongtao Liu from comment #4)
> > (In reply to H.J. Lu from comment #3)
> > > Created attachment 62385 [details]
> > > A patch
> > >
> > > This is a test patch:
> > >
> > > 1. Move pass_x86_cse after pass_split_all_insns.
> > > 2. Split 3 UNSPEC_INSN_FALSE_DEP patterns before RA.
> > >
> > > Can we try if it works?
> >
> > I'd rather keep the original implementation for below 2 reasons.
> >
> > 1) It needs quite a lot of work to do it as splitter way, and more efforts
> > for maintenance.
> >
> > 2) The tune is only for SPR(GLC), and no longer needed for the
> > successors(P-core) or any E-Core processors.
>
> I noticed this with generic tuning. Can
> we remove it from generic?
It shouldn't be in the generic tuning.
/* X86_TUNE_DEST_FALSE_DEP_FOR_GLC: This knob inserts zero-idiom before
several insns to break false dependency on the dest register for GLC
micro-architecture. */
DEF_TUNE (X86_TUNE_DEST_FALSE_DEP_FOR_GLC,
"dest_false_dep_for_glc", m_SAPPHIRERAPIDS | m_ALDERLAKE)
