https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126299
pkoning at gcc dot gnu.org changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2026-07-20
Status|UNCONFIRMED |ASSIGNED
Ever confirmed|0 |1
CC| |pkoning at gcc dot gnu.org
--- Comment #3 from pkoning at gcc dot gnu.org ---
Thanks Mikael. But that "better looking" output only considers the upper half
of the SImode value, which is clearly wrong. If that is what GCC does as the
fallback when no mode-specific compare operators are defined, that's a
different bug in a different area.
What actually needs to be done is to use signed branch on the compare result of
the upper bits, and unsigned branch on the remaining pieces. That's clearly
doable, I'm just trying to figure out the cleanest and most efficient way to do
that.
