[Bug rtl-optimization/25381] New: local alloc is not really that good

["pinskia at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

Take this little (stupid) C example:
int f1(int a)
{
  int b = a*2;
  return b+a;
}
---
Currently we produce:
_f1:
        mr r0,r3
        slwi r3,r3,1
        add r3,r3,r0
        blr

-----

We should be able to produce:
_f1:
        slwi r0,r3,1
        add r3,r3,r0
        blr

-------
If we look at the RTL which used:
(insn:HI 6 17 12 0 (set (reg/v:SI 121 [ a ])
        (reg:SI 3 r3 [ a ])) 326 {*movsi_internal1} (nil)
    (expr_list:REG_DEAD (reg:SI 3 r3 [ a ])
        (expr_list:REG_EQUIV (mem/c/i:SI (plus:SI (reg/f:SI 67 ap)
                    (const_int 24 [0x18])) [2 a+0 S4 A32])
            (nil))))

(insn:HI 12 6 20 0 (set (reg/v:SI 119 [ b ])
        (ashift:SI (reg/v:SI 121 [ a ])
            (const_int 1 [0x1]))) 188 {ashlsi3_no_power}
(insn_list:REG_DEP_TRUE 6 (nil))
    (nil))

(insn:HI 20 12 26 0 (set (reg/i:SI 3 r3 [ <result> ])
        (plus:SI (reg/v:SI 119 [ b ])
            (reg/v:SI 121 [ a ]))) 94 {*addsi3_internal1}
(insn_list:REG_DEP_TRUE 6 (insn_list:REG_DEP_TRUE 12 (nil)))
    (expr_list:REG_DEAD (reg/v:SI 119 [ b ])
        (expr_list:REG_DEAD (reg/v:SI 121 [ a ])
            (nil))))

------
And then look at what local alloc does:
Pass 0

  Register 119 costs: BASE_REGS:0 GENERAL_REGS:0 NON_SPECIAL_REGS:4000
LINK_REGS:4000 CTR_REGS:4000 LINK_OR_CTR_REGS:4000 SPECIAL_REGS:4000
SPEC_OR_GEN_REGS:4000 NON_FLOAT_REGS:4000 ALL_REGS:4000 MEM:8000
  Register 121 costs: BASE_REGS:0 GENERAL_REGS:0 NON_SPECIAL_REGS:6000
LINK_REGS:6000 CTR_REGS:6000 LINK_OR_CTR_REGS:6000 SPECIAL_REGS:6000
SPEC_OR_GEN_REGS:6000 NON_FLOAT_REGS:6000 ALL_REGS:6000 MEM:12000

  Register 119 pref GENERAL_REGS
  Register 121 pref GENERAL_REGS


Pass 1

  Register 119 costs: BASE_REGS:0 GENERAL_REGS:0 NON_SPECIAL_REGS:4000
LINK_REGS:4000 CTR_REGS:4000 LINK_OR_CTR_REGS:4000 SPECIAL_REGS:4000
SPEC_OR_GEN_REGS:4000 NON_FLOAT_REGS:4000 ALL_REGS:4000 MEM:8000
  Register 121 costs: BASE_REGS:0 GENERAL_REGS:0 NON_SPECIAL_REGS:6000
LINK_REGS:6000 CTR_REGS:6000 LINK_OR_CTR_REGS:6000 SPECIAL_REGS:6000
SPEC_OR_GEN_REGS:6000 NON_FLOAT_REGS:6000 ALL_REGS:6000 MEM:12000

123 registers.

Register 119 used 2 times across 2 insns in block 0; set 1 time; user var.

Register 121 used 3 times across 6 insns in block 0; set 1 time; user var.

1 basic blocks, 2 edges.

Basic block 0 prev -1, next -2, loop_depth 0, count 0, freq 10000, maybe hot.
Predecessors:  ENTRY [100.0%]  (fallthru)
Successors:  EXIT [100.0%]  (fallthru)
Registers live at start: 1 [r1] 3 [r3] 30 [r30] 67 [ap] 113 [sfp]
Registers live at end: 1 [r1] 3 [r3] 30 [r30] 67 [ap] 113 [sfp]

;; Register 119 in 3.
;; Register 121 in 0.
------
it puts 119 in r3 and 121 in r0, when the opposite is better.


-- 
           Summary: local alloc is not really that good
           Product: gcc
           Version: 4.2.0
            Status: UNCONFIRMED
          Keywords: missed-optimization, ra
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org
 GCC build triplet: powerpc-darwin


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25381