http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60763
Michael Meissner <meissner at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |meissner at gcc dot gnu.org
--- Comment #8 from Michael Meissner <meissner at gcc dot gnu.org> ---
Richard's analysis agrees with mine. The problem is with adding the check for
REG_CANNOT_CHANGE_MODE_P, it breaks the direct moves between GPRs and VSX
registers for SF. Because internally within the PowerPC, SFmode is stored as
DFmode, CANNOT_CHANGE_MODE_CLASS returns true. The original direct mvoe code
should not have generated a (SUBREG:DI (REG:SF ...)) in this case, since that
is illegal.
