https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120479
Jeffrey A. Law <law at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Last reconfirmed| |2025-06-04
--- Comment #4 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Unfortunately I'm not seeing any pattern we can really target. We're looking
at a 3 insn combination here. We need both source arguments to be substituted
into the AND so that we know we're only dealing with 0/1.
combine does the substitution for the equality test, but not the output of the
slt. So the closest we get to a useful combine step is:
Failed to match this instruction:
(set (reg:SI 181)
(and:SI (eq:SI (reg:SI 175)
(const_int -1 [0xffffffffffffffff]))
(reg:SI 135 [ _2 ])))
Maybe we could do something in match.pd. I haven't really looked to see if
that's a viable path or whether or not the result meets the gimple costing
conditions.
At least in gimple we know the types are single bit which simplifies the
problem a bit.
The other thought would be look at the expansion code itself. But I'm hesitant
to put any special code in there.
Sigh. At least right now I don't see a direct path forward.
