[Bug middle-end/77499] [13/14/15/16 Regression] Regression after code-hoisting, due to combine pass failing to evaluate known value range

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77499

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu.org
          Component|rtl-optimization            |middle-end
             Target|arm-none-eabi               |arm-none-eabi aarch64

--- Comment #28 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
For aarch64, the problem is expansion:

(insn 31 30 32 5 (set (reg:SI 123)
        (const_int -20191 [0xffffffffffffb121])) "/app/example.cpp":10:5 -1
     (nil))
(insn 32 31 33 5 (set (reg:SI 122)
        (xor:SI (reg/v:SI 103 [ x ])
            (reg:SI 123))) "/app/example.cpp":10:5 -1
     (nil))
(insn 33 32 34 5 (set (reg/v:SI 103 [ x ])
        (zero_extend:SI (subreg:HI (reg:SI 122) 0))) "/app/example.cpp":10:5 -1
     (nil))


It is also kinda of strange for arm too:
(insn 30 29 31 5 (set (reg:SI 114)
        (const_int 45345 [0xb121])) "/app/example.cpp":10:5 -1
     (nil))
(insn 31 30 32 5 (set (reg:HI 113)
        (subreg:HI (reg:SI 114) 0)) "/app/example.cpp":10:5 -1
     (expr_list:REG_EQUAL (const_int -20191 [0xffffffffffffb121])
        (nil)))
(insn 32 31 33 5 (set (reg:SI 115)
        (xor:SI (reg/v:SI 96 [ x ])
            (subreg:SI (reg:HI 113) 0))) "/app/example.cpp":10:5 -1
     (nil))
(insn 33 32 34 5 (set (reg/v:SI 96 [ x ])
        (zero_extend:SI (subreg:HI (reg:SI 115) 0))) "/app/example.cpp":10:5 -1
     (nil))


I wonder if we could do better at the expansion for the xor and that would get
the right thing.