[Bug middle-end/42919] [4.3/4.4/4.5 Regression] Argument unnecessarily spilled

[matz at gcc dot gnu dot org]

------- Comment #3 from matz at gcc dot gnu dot org  2010-02-01 00:21 -------
3.4 had this as initial RTL:

(insn 3 2 4 (set (reg:DI 60)
        (mem/f:DI (reg/f:SI 53 virtual-incoming-args) [2 a+0 S8 A32])) -1 (nil)
    (expr_list:REG_EQUIV (mem/f:DI (reg/f:SI 53 virtual-incoming-args)
     [2 a+0 S8 A32])
        (nil)))

(insn 4 3 5 (set (mem/f:DI (addressof:SI (reg/v:DI 59) 58 0xf7c9a4a4)
                 [2 a+0 S8 A64])
        (reg:DI 60)) -1 (nil)
    (nil))

(insn 15 7 16 0 (parallel [
            (set (reg/f:SI 7 sp)
                (plus:SI (reg/f:SI 7 sp)
                    (const_int -12 [0xfffffff4])))
            (clobber (reg:CC 17 flags))
        ]) -1 (nil)
    (nil))

(insn 16 15 17 0 (set (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [0 S4 A32])
        (addressof:SI (reg/v:DI 59) 58 0xf7c9a4a4)) -1 (nil)
    (nil))

(call_insn 17 16 19 0 (call (mem:QI (symbol_ref:SI ("foo") [flags 0x41]
 <function_decl 0xf7c9a438 foo>) [0 S1 A8])
        (const_int 16 [0x10])) -1 (nil)
    (nil) (nil))

addressof RTL and handling was removed in 2004, but was our way of making this
optimization on RTL.  So this would now need to be implemented in expand
or something :-/  Well, or by implementing some more memory optimizations
on RTL (recognizing that mem1 has same content as mem2, and using &mem1 where
&mem2 was used and the former somehow is "better").  Not committing to MEM
to start with would be better, obviously.


-- 

matz at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |matz at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42919