https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89435
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
In particular, the wrong combination is I think:
Trying 77 -> 78:
77: r164:SI=r125:QI#0&0xff
REG_DEAD r125:QI
78: r124:SI=r164:SI+r162:SI
REG_DEAD r164:SI
REG_DEAD r162:SI
Successfully matched this instruction:
(set (reg:SI 124 [ _15 ])
(plus:SI (reg:SI 162)
(subreg:SI (reg:QI 125 [ _18 ]) 0)))
That is simply not equivalent, even if we know that the value assigned to
pseudo 125 is constant 0xff and we loaded it into SImode register somewhere.
