https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84826
sudi at gcc dot gnu.org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |ebotcazou at gcc dot gnu.org
--- Comment #3 from sudi at gcc dot gnu.org ---
So I looked into this. Turns out the actual issue is that USE_RETURN_INSN
(FALSE) changes its value and becomes false after pass ce3.
According to what I can see, arm_r3_live_at_start_p() starts to return true
after ce3.
+static bool
+arm_r3_live_at_start_p (void)
+{
+ /* Just look at cfg info, which is still close enough to correct at this
+ point. This gives false positives for broken functions that might use
+ uninitialized data that happens to be allocated in r3, but who cares? */
This particular test cares :P
+ return REGNO_REG_SET_P (df_get_live_out (ENTRY_BLOCK_PTR_FOR_FN (cfun)), 3);
+}
r3 in this test case is being allocated for b and after ifcvt, it becomes
partially defined and thus the liveness does not get killed. (Look below for
excerpt from ce3)
;; lr in 0 [r0] 3 [r3] 12 [ip] 13 [sp] 14 [lr]
;; lr use 0 [r0] 3 [r3] 12 [ip] 13 [sp]
;; lr def 100 [cc]
;; live in 0 [r0] 3 [r3] 12 [ip] 13 [sp] 14 [lr]
;; live gen 3 [r3] 100 [cc]
;; live kill
(note 5 1 47 2 [bb 2] NOTE_INSN_BASIC_BLOCK)
(note 47 5 2 2 NOTE_INSN_PROLOGUE_END)
(note 2 47 4 2 NOTE_INSN_DELETED)
(note 4 2 7 2 NOTE_INSN_FUNCTION_BEG)
(insn 7 4 10 2 (set (reg:CC 100 cc)
(compare:CC (reg:SI 0 r0 [ cD.5556 ])
(const_int 0 [0]))) "ice.i":6 193 {*arm_cmpsi_insn}
(expr_list:REG_DEAD (reg:SI 0 r0 [ cD.5556 ])
(nil)))
(insn 10 7 11 2 (cond_exec (ne (reg:CC 100 cc)
(const_int 0 [0]))
(set (reg:SI 3 r3 [orig:115 CHAIN.1_5(D)->bD.5567 ] [115])
(mem/j:SI (reg/f:SI 12 ip [orig:113 CHAIN.1D.5566 ] [113]) [1
CHAIN.1_5(D)->bD.5567+0 S4 A32]))) "ice.i":7 4068 {*p *arm_movsi_vfp}
(expr_list:REG_EQUIV (mem/j:SI (reg/f:SI 12 ip [orig:113 CHAIN.1D.5566 ]
[113]) [1 CHAIN.1_5(D)->bD.5567+0 S4 A32])
(nil)))
My question is:
1) Is there any easy way to avoid the false positives from
arm_r3_live_at_start_p()
2) Why is r3 still live at IN of BB2 when there is no reaching definition of
it? I mean also there will never be any reaching definition at IN of BB2. So
shouldn't all the liveness (barring the artificially create ones for the
prologues/stack requirements) be killed there? I know I may be over simplifying
all the intricate details of liveness analysis when I ask this question, but I
am looking for some help here!
